Labour Market Policy in Switzerland - Universität St.Gallen

6 downloads 6 Views 3MB Size Report
Während der Erstellung dieser Dissertation war ich am Schweizerischen Institut für ... politik sehr viel lernen und möchte mich herzlich bedanken dafür, dass er mir immer ...... b benefits paid in canton i, and B total benefits paid in Switzerland.

Labour Market Policy in Switzerland: Institutions, Design, Effects

DISSERTATION of the University of St. Gallen, Graduate School of Business Administration, Economics, Law and Social Sciences (HSG) to obtain the title of Doctor of Philosophy in Economics and Finance

submitted by

Heidi M. Steiger from Altstätten (St. Gallen)

Approved on the application of

Prof. Dr. Michael Lechner and Prof. Dr. Josef Zweimüller

Dissertation no. 3256

Gutenberg Schaan, 2007

The University of St. Gallen, Graduate School of Business Administration, Economics, Law and Social Sciences (HSG) hereby consents to the printing of the present dissertation, without hereby expressing any opinion on the views herein expressed.

St. Gallen, November 22, 2006

The President:

Prof. Ernst Mohr, PhD

Danksagung

Während der Erstellung dieser Dissertation war ich am Schweizerischen Institut für Aussenwirtschaft und Angewandte Wirtschaftsforschung (SIAW) der Universität St. Gallen beschäftigt. Dort konnte ich sehr vom professionellen, ambitionierten und kompetenten Umfeld profitieren und lernen. Professor Michael Lechner schulde ich grossen Dank für seine Unterstützung und wertvolle Zusammenarbeit. Von Markus Frölich konnte ich in gemeinsamen Projekten über die Schweizerische Arbeitsmarktpolitik sehr viel lernen und möchte mich herzlich bedanken dafür, dass er mir immer wieder seine wertvolle Zeit geschenkt, mit mir verschiedene Aspekte der Arbeit diskutiert und mir gute Ideen vermittelt hat. Stefanie Behncke, Dragana Djurdjevic, Justina Fischer, Michael Huber, Sangeeta Khorana, Blaise Melly, Ruth Miquel, Patrick Puhani, Rosalia Vazquez-Alvarez, Stephan Wiehler, Conny Wunsch und Sacha WunschVincent waren geschätzte Kolleginnen und Kollegen am SIAW, die sich viel Zeit nahmen, an Seminaren und Vorträgen meine Arbeit kritisch zu hinterblicken, sich gegenseitig auszutauschen oder auch einfach eine gute Kameradschaft zu pflegen. Ich arbeitete in verschiedenen Projekten für das Schweizerische Staatssekretariat für Wirtschaft (seco) mit. Das seco hat es freundlicherweise erlaubt, die hervorragenden und einmaligen Daten auch für die Dissertation weiterzuverwenden, und insbesondere Jonathan Gast verdanke ich viele wertvolle Informationen rund um die Daten, aber auch allgemein zum Thema Arbeitslosenversicherung. Ganz herzlich bedanken möchte ich mich auch bei der Schweizerischen Nationalbank, die es mir ermöglicht hat, in ihrem international renommierten Doktorandenprogramm am Studienzentrum Gerzensee teilzunehmen, und die auch hinter der Finanzierung von Vortragsprämien des Vereins für Socialpolitik für die Teilnahme von Nachwuchsforschenden an renommierten internationalen Konferenzen steht. Vom Nationalfonds wurde ich verdankenswerterweise während dreier Jahre im Rahmen eines Nationalen Forschungsprogrammes unterstützt. Mein allergrösster Dank geht an meine ganze Familie. Parikshit Sharma hat mich liebevoll und selbstlos nach all seinen Kräften im Hintergrund unterstützt und mir immer wieder mit passenden Formulierungen in Englisch ausgeholfen. Während der Zeit, die ich am Lehrstuhl und an der Dissertation beschäftigt war, durften wir auch unsere beiden Kinder Manu und Rhea in unser Leben einschliessen. Sie waren ein grosser Ansporn für diese Arbeit, mussten manchmal aber auch etwas auf mich verzichten. Es

gab sehr schwierige Zeiten, in denen es unmöglich war, sämtliche verschiedenen Ansprüche unter einen Hut zu bringen, was oftmals zu Entbehrungen auf allen Seiten führte. Meine Eltern haben mir früh die richtigen Werte mit auf den Weg gegeben, mich immer gefördert und in jeglicher Hinsicht sehr grosszügig unterstützt, wofür ich ihnen sehr dankbar bin. Ohne den Einsatz meiner ganzen Familie wäre diese Arbeit niemals möglich gewesen, weshalb sie ihr gewidmet sein soll.

St. Gallen, November 22, 2006

Heidi Steiger

I

Table of Contents

Introduction..................................................................................................................... 1 Methodological aspects................................................................................................... 6 Chapter Synopsis ............................................................................................................ 8

Part I: Swiss unemployment insurance system ............................................... 9 1

Introduction........................................................................................................ 10

2

Historical development ...................................................................................... 11

2.1

Development of unemployment......................................................................... 12

2.2

Unemployment insurance .................................................................................. 13

2.3

Active labour market programmes .................................................................... 15

2.4

Specific regulations............................................................................................ 17

2.4.1 2.4.2 2.4.3

Entitlement period and benefit exhaustion..................................................... 17 Contribution rates........................................................................................... 18 Benefit level (replacement rates) ................................................................... 20

2.4.4 2.4.5 2.4.6 2.4.7 2.4.8 2.4.9 2.4.10 2.4.11 2.4.12

Duration of benefit entitlement ...................................................................... 23 Sanctions ........................................................................................................ 26 Waiting period................................................................................................ 26 Acceptability conditions of employment ....................................................... 27 Monitoring and regional employment offices................................................ 28 Interim job scheme ......................................................................................... 30 Employment programmes .............................................................................. 32 Minimum programme slots ............................................................................ 33 Financial incentive system and cost sharing .................................................. 34

3

Studies about effectiveness of Swiss labour market policy............................... 37

4

Micro data resources .......................................................................................... 39

4.1

Introduction........................................................................................................ 40

4.2

General Overview .............................................................................................. 41

4.3

Access to data .................................................................................................... 43

4.4

Some empirical studies based on these data ...................................................... 44

4.5

Outcomes in data sources .................................................................................. 44

4.5.1

Definition ....................................................................................................... 44

II

4.5.2

Data and descriptive statistics ........................................................................ 47

4.5.3 4.5.4

Congruency and accuracy of data .................................................................. 50 Delayed entries in pension system data ......................................................... 52

4.5.5

Summary and conclusions ............................................................................. 55

5

Unemployment profiles ..................................................................................... 55

5.1

Data .................................................................................................................... 56

5.2

Subgroup profiles............................................................................................... 58

5.3

Active labour market programmes .................................................................... 65

5.4

Sanctions ............................................................................................................ 68

5.5

Interim jobs ........................................................................................................ 69

6

Summary and conclusions ................................................................................. 70

Appendix....................................................................................................................... 73

Part II: Effectiveness of Active Labour Market Programmes .................... 77 1

Introduction........................................................................................................ 78

2

Swiss active labour market policy and unemployment insurance ..................... 80

2.1

Design ................................................................................................................ 80

2.2

Past findings about the effectiveness of Swiss active labour market policy ..... 81

2.3

Assignment to programmes and nonparticipation ............................................. 82

3

Econometrics ..................................................................................................... 83

3.1

Identification and estimation of treatment effects ............................................. 83

3.2

Static method ..................................................................................................... 86

3.3

Dynamic assignment of first treatment.............................................................. 86

3.4

Matching and estimation procedure................................................................... 88

3.5

Comparison of estimators .................................................................................. 91

4

Data and sampling ............................................................................................. 93

4.1

Data .................................................................................................................... 93

4.2

Sample selection ................................................................................................ 94

4.3

Outcomes ........................................................................................................... 94

4.4

Programmes ....................................................................................................... 96

4.5

Definition of nonparticipants and programme starting dates ............................ 97

4.6

Descriptive statistics .......................................................................................... 98

5

Results.............................................................................................................. 100

III

5.1

Propensity Scores............................................................................................. 100

5.2

Common Support ............................................................................................. 103

5.3

Who is matched to whom?............................................................................... 103

5.4

Matching results............................................................................................... 103

5.4.1 5.4.2

Personality programmes............................................................................... 104 Basic computer courses................................................................................ 105

5.4.3 5.4.4

Language courses ......................................................................................... 106 Vocational training....................................................................................... 106

5.4.5 5.4.6 5.4.7

Other courses................................................................................................ 108 Employment programmes ............................................................................ 108 Summary of findings.................................................................................... 109

6

Conclusion ....................................................................................................... 109

Appendix..................................................................................................................... 111

Part III: The effect of maximum unemployment benefit duration ........... 123 1

Introduction...................................................................................................... 124

2

Unemployment insurance system and policy change...................................... 127

2.1

Unemployment insurance in Switzerland ........................................................ 127

2.2

Policy change 2003 .......................................................................................... 128

2.3

Maximum duration of unemployment benefits ............................................... 130

3

Identification and estimation strategy.............................................................. 131

3.1

Year-to-year comparison ................................................................................. 133

3.1.1 3.1.2 3.2 3.2.1 3.2.2

Identification ................................................................................................ 134 Estimation .................................................................................................... 135 Regression-discontinuity approach.................................................................. 138 Identification ................................................................................................ 138 Estimation .................................................................................................... 139

4

Data .................................................................................................................. 140

4.1

Data source and sampling ................................................................................ 140

4.2

Outcome variables ........................................................................................... 143

4.3

Accuracy of outcome ....................................................................................... 146

5

Results.............................................................................................................. 149

5.1

Number of benefits and consequent entitlement periods................................. 149

IV

5.2 5.2.1 5.2.2 5.3

Year-to-year comparison ................................................................................. 150 Matching results ........................................................................................... 150 Estimated Effects ......................................................................................... 155 Regression-discontinuity approach.................................................................. 157

5.3.1 5.3.2

Effect on job finding .................................................................................... 158 Leaving unemployment without finding a job............................................. 160

5.3.3

Interim jobs .................................................................................................. 160

6

Conclusions...................................................................................................... 161

Appendix..................................................................................................................... 164

Summary and outlook..................................................................................... 169 References................................................................................................................... 173 Curriculum Vitae ........................................................................................................ 181

V

Abbreviations

AHV

Alters- und Hinterlassenenversicherung

ALMP

Active labour market programmes

AS

Amtliche Sammlung

ASAL

Auszahlungssystem der Arbeitslosenversicherung

AVAM

Arbeitsvermittlungs- und Arbeitsmarktstatistik

AVIG

Arbeitslosenversicherungs- und Insolvenzentschädigungsgesetz

AVIV

Arbeitslosenversicherungs- und Insolvenzentschädigungsverordnung

BBl

Bundesblatt

LMP

Labour market policies

PS

Pension system

UI

Unemployment insurance

UIA

Unemployment Insurance Act (AVIG)

1

Introduction Labour market policies Labour market policies (LMP) "mediate between supply (jobseekers) and demand (jobs offered) in the labour market" (ILO, 2003). LMP are a subset of wider employment policies that also include wage policies, collective bargaining guidelines, labour laws, child care provisions and other policies. The objectives of, and justification for LMP are economic, social, and political. Social welfare aspects of LMP include poverty prevention, equity, psychological and social stabilisation.1 Examples of political goals are preferential treatment of disadvantaged groups (as a positive feature) and systematic favouritism (as a negative aspect).2 Economic goals of LMP are improved matching efficiency3, stabilisation of the business cycle through consumption smoothing4, as well as easier adaptation to and acceptance of technological change.5 LMP can be divided into active and passive measures. The latter measures are usually associated with income support in the form of unemployment benefits or pensions, whereas active measures concern labour market integration and job matching directly. Passive policies provide income support for periods of jobsearch (unemployment benefits) or in situations where persons have substantial difficulties finding a suitable job (such as disability pensions, early retirement schemes). They allow the job-seeker to maintain his living standard while looking for a suitable job, in order to improve the efficiency of the match. The specific functions and types of active labour market policies are6:

1

E.g. Clark and Oswald (1994) quantify the psychological costs of unemployment, and Schmid and Rosenbaum (1995) summarise the role of social stabilisation.

2

E.g. Saint-Paul (1995).

3

E.g. Burdett (1979).

4

E.g. Acemoglu and Shimer (1999).

5

For a more comprehensive overview of the goals of unemployment insurance and labour market policy, see Schulze (2004), p. 29-36.

6

The OECD divides active labour market policies into five categories (OECD 1993): (1) employment services and counselling, (2) labour market training, (3) special youth measures, (4) subsidised employment, job creation, self-employment assistance, (5) measures for the disabled.

2

Contributing directly to the matching process of workers and jobs and general labour market functioning (e.g., placement services, job-search assistance, providing labour market and vacancies information). Enhancing the skill levels of job-seekers (e.g., training, education loans). Reducing structural imbalances (e.g., mobility assistance, re-training). Reducing or increasing labour supply (e.g., retirement policies). Creating jobs directly (e.g., wage/employment subsidies, public works, selfemployment assistance). Providing financial support for job-seekers in specific situations (e.g., commuting allowances, bad weather compensations). Labour market policies in Switzerland In Switzerland, unemployment insurance is the most important institution regarding passive and active labour market policy. The following characterise current Swiss labour market policy: Unemployment insurance is public and obligatory for every employee. Unemployment insurance is self-financed, although administrative costs and programmes are partly borne by public funds. Contributions are paid by employees and employers to an equal share, as a fixed percentage of the salary, up to the maximum insurable amount. The minimum contribution time for a benefit claim is 12 months. Benefit entitlement is longer than minimum contribution time. Unemployment benefits are paid for a limited time period (18.5 or 24 months, depending on age). This period is supposed to be sufficient to provide bridgeover finances until a new job is found. There is an initial waiting period of one week before benefits are paid out. The unemployment benefit is a certain percentage of the former salary, up to a given maximum. Persons with dependants or who receive lower than a benchmark income receive a higher percentage. Each person is expected to strive to his maximum to retain or regain employment. In the case of job loss, this insurance facilitates training and provides financial subsidies that increase employability.

The basic characteristics of the unemployment insurance system have remained the same since its introduction 21 years ago, though the parameters and rates are occasion-

3

ally adjusted (particularly the time profiles of the replacement rate and the adequacy of contribution time and entitlement duration). According to the OECD (2002), the Swiss system is one of the most generous systems in the world, providing a very high level of insurance. Switzerland maintains an internationally competitive low unemployment rate.7 The first part of this thesis describes the evolution of the unemployment insurance system since its introduction in 1984. Before the 1990s the Swiss unemployment rate was low, and thus this issue failed to attract the attention of members of the government, media or population at large. Similarly, there was little focus on the Swiss unemployment insurance system. Flückiger (1998) provides various reasons for Switzerland's low unemployment rate in general, including flexibility of the labour force due to very restrictive immigration laws, small size and openness (absence of heavy industries, export-orientation), good education system with prominence given to vocational education, absence of labour conflicts, low unionism, absence of government interference in wage settlements, and general wage flexibility. Switzerland also profited from its absence from the Second World War, and from a strong protestant work ethic. During the 1990s Switzerland suffered a steep rise in unemployment, from 0.5% in 1990 to 5.2% in 1997. This increase in unemployment has been attributed to increased frictional and structural unemployment, changing immigration laws, which made the labour force less flexible, the increased female labour supply, and to some extent the negative incentive effects of the unemployment insurance itself (Flückiger, 1998). Other explanations for this rise that have been put forward are firms' under-investment in training (Wolter and Weber, 1998) and related wage rigidity (Puhani, 2003, and Fehr und Götte, 2005). The sharp rise in unemployment during the 1990s initiated substantial activity aimed at improving the system to better suit the economic and social climate. Entitlement periods were extended because higher frictional unemployment and associated longer search unemployment were assumed. Additionally, much more emphasis was put on active labour market policies in order to prevent long-term unemployment. This activation was inflated to the extent that entitlement to benefits became temporarily conditional on participation in programmes. At this time almost everybody was thought to

7

Standardised OECD unemployment rate of 4.4 percent in 2004 compared to 5.5 in the US, 4.7 in Japan, 4.8 in Austria, 9.5 in Germany, 9.7 in France, 8.0 in Italy, 6.4 in Sweden, 4.4 in Norway, and 4.7 in the UK (source: OECD).

4

require either subsidised employment programmes or vocational training, and minimal quotas for programme participation were imposed. However, the subsequent recovery period in 1998 and 1999 put this principle somewhat into perspective, and expenditure for active labour market programmes was reduced. Evaluation of the effectiveness of active labour market programmes in 2000 and 2001 is the subject of the second part of the thesis. Concurrent with attempts to reduce unemployment during the 1990s, administrative procedures were reorganised. Regional placement offices were introduced in 1996 and 1997, replacing communal offices. Specialised unemployment officers were placed in charge of administrating, controlling and counselling the unemployed. Not only were active labour market programmes reduced in the course of the recovery period, but also – at a later stage – passive maintenance. The maximum duration of unemployment benefits was increased in 1996, but was then reduced by almost a quarter in 2003. This reduction was the first in the history of Swiss unemployment insurance. The third part of this thesis will deal with this reduction in unemployment benefits and evaluate its impacts on job finding rates.

Effectiveness of active labour market policies The OECD launched Jobs Strategy in 1994, with a particular emphasis on active labour market policies, in order to tackle labour market problems. With the availability of microeconomic data from information databases, evaluation of these programmes has become a busy and also lucrative research field. Betcherman, Olivas and Dar (2004) provide a comprehensive international survey of recent findings.8 These authors suggest that employment services are very effective in terms of higher employment and earnings. Participants in training courses benefit in terms of employment rates over the long-term. In contrast, retraining, youth measures and public works were found to be largely unsuccessful. In terms of wage and employment subsidies, these authors conclude these measures have little positive impact, but carry substantial deadweight and substitution costs. Relative to Switzerland in particular, Gerfin and Lechner (2002) and Gerfin, Lechner and Steiger (2005) report a positive effect on employment for the interim job scheme, but a negative effect for employment programmes. Vocational training and computer

8

Earlier surveys can be found in Martin and Grubb (2001), Dar and Tzannatos (1999) and Fay (1996). Heckman, LaLonde and Smith (1999) provide an empirical and theoretical overview.

5

courses had a non-negative effect on employment, whereas negative effects were found for language and basic courses. Lalive and Zweimüller (2001) report negative impacts on total unemployment duration for all programmes except for interim jobs. On an aggregate programme level, Curti (2002) and Frölich and Lechner (2004) report positive effects, whereas Vassiliev (2003) reveals negative effects on unemployment duration for the canton of Geneva. Falk, Lalive and Zweimüller (2005) report that computer programmes had a negative but non-significant effect on invitations to job interviews. Note that all of these studies examined data from the late 1990s, and therefore this research is still influenced by the transition process from the old passive regime to the new emphasis on activation. Additionally, the administrative changes caused by the introduction of regional placement offices and caseworkers were associated with some uncertainty and lack of experience, and thus may have led to a lack of efficiency during this time. For these reasons, this period might not be representative of the present unemployment system. Comprehensive evaluation of the effectiveness of labour market programmes after 2000 has not yet been undertaken and thus is the subject of the second part of this thesis.

Effectiveness of passive labour market policies Passive benefits are intended to provide income support during a time of intense job search. One goal of this support is the facilitation of better matches; another goal is the stabilisation of consumption. While from an economic point of view, insurance for risk-averse employees in periods of involuntary joblessness is efficient, these policies can also be a problem such that they may provide inappropriate incentives.9 These incentives can relate to the labour market as a whole and the type of jobs available, but also can have an effect on the behaviour of the unemployed in their job search. The theoretical literature about optimal unemployment insurance focuses on the level and duration of insurance, time profiles, how this insurance should be financed, and what the requirements on the unemployed should be. Monitoring and sanctions, as well as public work requirements, have been cited as possible means of improving incentives for the unemployed, and hence a way of providing a more generous unemployment insurance without the negative side effects. A large empirical literature has used micro data on individual periods of unemployment to investigate the influence of these un-

9

The seminal paper establishing the theoretical framework for this argument is by Mortensen (1977).

6

employment insurance system parameters in practice.10 Generally, the negative impacts of more generous unemployment insurance programs have been confirmed by research, but not always to the extent suggested by theory. Little empirical work has examined the incentives of unemployment benefits in Switzerland in particular. There is, however, evidence which suggests that benefit sanctions in Switzerland are effective in bringing people back to work (Lalive, van Ours, Zweimüller, 2002). The lack of empirical data investigating the effect of replacement rates and maximum benefit duration may be due to the fact that these parameters have remained largely stable in recent years as unemployment has became a problem. Finally, it is important to note that the scope of this thesis is economic. That is, the current research focuses primarily on labour market status such as unemployment, employment, inactivity and income. It has been shown that unemployment itself has a negative effect on economic factors, for instance future employment quality, career paths and wages.11 However, negative outcomes might also take the form of impacting social and psychological well-being, which is not directly addressed here. Unemployment and its consequences constitute a major life event for many people, sometimes leading to severe hardship. Even though individual persons may appear only as one of many "observations" presented in this thesis, it is vital to recognise the individuals and their personal fate behind these figures.12

Methodological aspects This thesis provides empirical analyses of the unemployment insurance system in Switzerland. Broadly divided, part of the thesis is descriptive only, while another is evaluative, providing the basis for inferences about causal relationships, specifically the effects of participation in active labour market programmes and a reduction in unemployment benefit duration on a variety of outcomes. "Causality" as used here is defined according to Rubin (1974) and Holland (1986). The cause (in the case of pro-

10

A recent survey with an emphasis on Europe is provided in Lalive, van Ours and Zweimüller (2005).

11

Arumpalam (2000) reports that unemployment has a large effect on future wages for persons in the UK. Oberholzer-Gee (2005) also confirms the presence of a stigma of non-employment in Austria and Switzerland using an experimental design.

12

“Surprise”, a magazine sold on the street by long-term unemployed, provides interesting real-life insights.

7

grammes there are usually multiple causes) is the basis of the analysis, and inferences are made about the effect of the cause, rather than about different causes for a given effect. Individuals can be exposed to the cause or not, and their individual future depends on this exposure in a certain way. According to the type of problem, different untestable, but arguable assumptions for the identification of the causal effects are made. The stable-unit-treatment-value assumption (SUTVA) introduced by Rubin (1980) is assumed during this research. This assumption states that, regardless of the assignment procedure for the corresponding programme or policy change, individual outcomes related to a certain programme or policy change are the same and have the same effect. The SUTVA rules out large-scale and macro effects. Possible violations of the SUTVA are manifold, for instance active labour market policies might lead to distortions and crowding out on the labour market, possibly distorting wages, the hiring of subsidised persons only to replace unsubsidised workers, firms/workers imposing costs of otherwise self-funded training to the public, and increasing contribution rates and taxes. Similarly, the duration, availability and size of passive maintenance might significantly alter people’s behaviour on and off the job, for instance in terms of higher or lower reservation wages and more or less intense job search intensity. The SUTVA is assumed to be satisfied for the evaluation of both active and passive policies. The key argument put forward for active policies is that programmes are implemented on a comparatively small scale. Additionally, caseworkers are often recruited locally and have quite good knowledge of the industrial structures and employers, such that they can observe "abuses" to a certain extent. The limited time horizon argument can also be put forward, which holds for the passive maintenance as well. Throughout the thesis appropriate statistical and econometric methods are applied to every problem. Matching estimation in particular is used. The relevant formal models are introduced, explained and motivated in the corresponding sections. Note that the large micro datasets from the unemployment insurance and pension system that are used in this thesis cannot be handled without the use of statistical software tools. SPSS was the main software used to clean, adjust and process the data. All estimations as well as graphs were made using STATA. For Mahalanobis matching estimation, Leuven and Sianesi's (2003) PSMATCH2 code for STATA provided a useful subroutine.

8

Chapter Synopsis This thesis consists of three parts. The first part is essentially institutional and serves as a guide and handbook to the Swiss environment. It unfurls the history of the Swiss unemployment insurance and describes the development of specific important regulations up to the present system. A description of the data for microeconometric evaluation of the unemployment insurance system is provided, together with an evaluation of its internal consistency. Finally, a descriptive analysis of unemployment profiles is provided. Part two of the thesis addresses the effectiveness of active labour market programmes. Programmes of persons registering in 2000 and in the first half of 2001 are evaluated. The results of two matching estimators are compared, and a variety of outcomes are taken into account. The third and last part of the thesis deals with passive labour market policy; it evaluates the effect of a reduction in maximum duration for unemployment benefits that took place in Switzerland in 2003. The findings are summarised in a concluding section. All three parts of the thesis are self-contained and may be read separately. Information about institutions and data are thus occasionally provided repeatedly, but in an appropriate short form. Appendices are positioned at the end of the individual parts, and the references are listed at the end of the thesis.

9

Part I: Swiss unemployment insurance system

10

1 Introduction This part of the thesis describes the Swiss unemployment insurance system and labour market policy. First, an analysis of the development of unemployment insurance and active labour market policy in Switzerland is presented, beginning with the introduction of the current unemployment insurance act (UIA hereafter), which was introduced in 1983 and 1984. The focus in this section is on individual services in the form of passive benefit payments, active labour market programmes, and their interdependencies.1 Second, unemployment insurance and pension system data is introduced as a rich and preferential data source for empirical research, together with an assessment of its internal consistency and accuracy. Finally, a descriptive analysis of time profiles based on this data is provided, specifically examining labour market programmes, sanctions, interim jobs, and evolvement of labour market status for sub-groups defined on the basis of individual characteristics. This section should be conceptualised as a handbook for those people interested in the design of the unemployment insurance system in Switzerland, whether as a researcher or policy maker. The main focus is on the individual services of unemployment insurance in the form of passive benefit payments and active labour market programmes, as well as their interdependencies. The Swiss confederacy has the constitutional legislative competence providing unemployment insurance. This paper deals mainly with federal legislation, and the data used in this analysis are also at the federal level. However, the law is executed by the cantons, which have (or take) a considerable amount of freedom in interpretation and implementation. Curti und Meins (1999), as well as Battaglini and Giraud (2003), deal with cantonal differences in the execution of the unemployment insurance act. Filippini (1998) also examines regional unemployment disparities in Switzerland. However, that issue is not addressed here. Furthermore, the current research deals only with the minimum standards of federal unemployment insurance, and not with auxiliary services provided by cantons (e.g., unemployment aid, public provision of work).

1

Besides these services the service catalogue of the UI incorporates insolvency, bad weather allowances, and reduced hours compensation. Those services are not taken into account here, because they are on a collective or firm level rather than on an individual level.

11

Legal sources for this work are the official collection of federal laws (Amtliche Sammlung, AS), The Federal Journal (Bundesblatt, BBl), the Unemployment Insurance Act (Arbeitslosenversicherungsgesetz, AVIG), and the Unemployment Insurance Ordinance (Arbeitslosenversicherungsverordnung, AVIV). Furthermore, extracts from protocols of sessions of the federal assembly were examined and incorporated where relevant.

2 Historical development This chapter gives an overview of unemployment insurance in Switzerland.2 It begins by providing unemployment figures, then presents a short introduction into the history of Swiss unemployment insurance, and finally gives an overview of the financial development of unemployment insurance in Switzerland. Unemployed and unemployment rate since 1936.

0

0

1

50

2

%

in 1000 100

3

4

150

5

200

Figure 1.

1940

1950

1960

1970

Unemployed in 1000

1980

1990

2000

Unemployment rate

Yearly unemployment; source: seco

Until 1983 only full-time unemployed; since 1984 part-time unemployed also included.

2

See also Schmid (1998) and SECO (2002).

12

2.1 Development of unemployment This section first examines the recorded history of the number of unemployed persons in Switzerland. Figure 1 displays unemployment in Switzerland since 1936. After a time of high unemployment in the years between the two World Wars unemployment fell very quickly. In the boom years of the 1950s and 1960s unemployment was virtually non-existent. Flückiger (1998) summarises various reasons for the low unemployment rate of Switzerland in general, citing flexibility of the labour force due to very restrictive immigration laws, small size and openness (absence of heavy industries, export-orientation), a good education system with prominence given to vocational education, absence of labour conflicts, low unionism, absence of government interference in wage settlements, and general wage flexibility. Switzerland also profited from its absence from the Second World War, and from the presence of a strong protestant work ethic. In the 1970s the Oil Crisis was associated with a significant rise in the unemployment rate, from 0.003% in 1973 (which means that only 81 persons were unemployed!) to 0.339% 1975 and 0.691% in 1976, a 20-fold jump in unemployment in 3 years. 1976 saw the peak of unemployment during the 1970s. After that unemployment decreased, only to rise again at the beginning of the 1980s. Switzerland then suffered a steep increase in unemployment, from 0.5% in 1990 to 5.2% in February 1997. Reasons put forward for this rise include increased frictional and structural unemployment, changing immigration laws, which made the labour force less flexible, as well as the increased female labour supply, and to some extent the effect of unemployment insurance itself (Flückiger, 1998). Other explanations include firms' under-investment in training (Wolter and Weber, 1998) and related wage rigidity (Puhani 2003, Fehr and Götte, 2005). Thus, during this period Switzerland experienced unemployment which exceeded even the historical peaks of pre-Second World War time. In February 1997 a maximum (to date) unemployment rate of 5.7% (206,291 unemployed) was reached. This period was followed by a recovery in 1998 and 1999. In April 2000 the unemployed rate came down below the 2% mark, but has begun rising again.

13

2.2 Unemployment insurance The history of unemployment insurance (UI) in Switzerland dates back to the 19th century.3 The first insurance funds emerged in the cities of Berne and St. Gallen, including private as well as public funds. In 1905 the first private fund, with contributions from both employers and employees, was constructed during the embroidery crises. In 1924 federal parliament passed a law impacting federal subsidies for unemployment insurance funds. The funds remained autonomous, but they had to satisfy certain minimal criteria. For the first time a few cantons declared unemployment insurance obligatory for persons with low incomes. By 1942 the cantons were committed to paying equal contributions. A revision of a paragraph in the federal constitution in 1947 formed the basis for common unemployment insurance in Switzerland. The respective law was passed in 1951, but did not declare UI obligatory for all employees. Funds already existing on a cantonal, communal or union level constituted the body of Swiss insurance. On the 1st of April, 1977, unemployment insurance was declared obligatory for all persons, based on a constitutional change in 1976. In 1982 a new unemployment insurance act, based on the same constitutional mandate, was passed. It was implemented in 1984. Active labour market programmes – denoted as "preventive measures" – were implemented directly in the law for novelty sake. Three major revisions and other measures have since been made to the law relevant to UI. 1989 saw the first major revision initiated by the federal ministry. The focus of the revision was the compensation regulations for reduced-hours work and bad weather. In October 1990 the law was passed. In 1993 an urgent resolution was adopted in the context of rising unemployment rates, extending the benefit entitlement period but shortening the replacement ratio for single persons with high incomes. In the same year a second revision was initiated. It emphasised activation measures by making benefit entitlement dependent on participation in these measures. It also changed the organizational structures of the UI system by introducing regional employment offices and cantonal centres for the logistics of active labour market policies. The act finally passed in 1995 and was introduced in 1996. Also in 1996 an urgent resolution attempted to lower the replacement ratio and therefore unemployment benefits. A referendum was taken, and voters overruled the resolution. An adapted version of this legislative change finally was adopted 1999. The third revision of the unemployment in-

3

An overview of the history of the Swiss unemployment insurance system can be found in Gerhards (1987), p. 30.

14

surance act passed in 2002. It cut maximum unemployment benefit duration by around one third and doubled minimum contribution time. This legislation demonstrates that many of these changes were made as a reaction to, or in expectation of labour market conditions. Most notably, the extensions of benefit entitlement duration were a consequence of the increasing observed unemployment duration. Increases in contribution rates were due to the negative balance of the funds. Figure 2 reveals clearly the activity in legislative changes since the increase of the unemployment rate in the beginning and mid-1990s. In contrast, not a single legislative change was made in the 1980s. Figure 2.

Monthly unemployment rate 1980-2003 and changes in legislation.

6.00

resolution about measures in the UI

5.00

1st revision passed

UIA passed

3rd revision implemented

resolution financing resolution financing resolution reorganisation measures

4.00

UIA implemented

3.00

nd 2nd revision 2 revision implemented passed

reorganisation programme

2.00

1st revision implemented

1.00

"technical" revision

3rd revision passedn

1/2005

1/2004

1/2003

1/2002

1/2001

1/2000

1/1999

1/1998

1/1997

1/1996

1/1995

1/1994

1/1993

1/1992

1/1991

1/1990

1/1989

1/1988

1/1987

1/1986

1/1985

1/1984

1/1983

1/1982

1/1981

1/1980

0.00

Source for unemployment rate: monthly unemployment rate, not seasonally adjusted, www.snb.ch.

The development in the finances of the UI system since 1975 is shown in Figure 3. In the 70s and 80s the account of the UI system was never negative, and there was rarely a deficit. In 1992 only, after two years of negative balance, the account was negative for the first time. Since 1992 expenditures have become much higher than in earlier years. They reflect more or less the unemployment rate. They are tied to the number of unemployed in a very high degree and can be reduced only by means of reduction in either benefit level or entitlement duration. However, this would mean that persons would demand welfare institutions which are mainly on a communal level in Switzer-

15

land. Unemployment insurance therefore attempted to cover its deficits by means of additional revenues in form of higher contribution rates. In this way the UI system account became positive in 2001. Finances of the UI.

0

-10000

-5000

1

%

Mio CHF 0

2

5000

3

10000

Figure 3.

1975

1980

1985

Total revenues Total account

1990

1995

2000

Balance Contribution rate (right scale)

Source: Swiss statistics of social insurance (Schweizerische Sozialversicherungsstatistik)

Figures listed in Appendix.

2.3 Active labour market programmes The unemployment insurance act (UIA) of 1982 included active labour market programmes in the services offered by unemployment insurance as a new feature. However, these were not truly a novelty. During the world economic crisis in the 1920s these measures had been offered on a grand scale.4 Before the introduction of the new UIA, active labour market programmes were regulated in the placement services law and in cantonal laws. In the UIA they were denoted as, "services for measures to pre-

4

More extensive historical overviews can be found in Hug (1986) and Freiburghaus (1987). Freiburghaus (1987) also provides deeper insight into the discussions that were taking place when the UIA was designed.

16

vent and fight unemployment (preventive measures)". 5 It was emphasised that these programs are not structural or countercyclical instruments on a macro level, but on an individual level. As such they were required to improve the individual employability of a job seeker. Another important point was that these measures should not be a replacement for conventional primary or further education. Rather, they are intended to provide retraining in direct relation to employability.6 A change of the UIA in that other direction has never been intended. The UIA mainly refers to services for those people who are eligible for benefits, and thus measures for people who are not entitled are regulated in placement services law. Furthermore, the UIA references only the services and contributions of the UI system, but rules regarding execution and supply of measures are cantonal. The payments of UI to those eligible concerning courses and employment programmes consist of direct costs and benefits during the time of participation. Expenditures for active labour market programmes.

0

0

200

5

%

Mio CHF 400

600

10

800

Figure 4.

1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Tot. expenditures ALMP

as % of total exp.

Source: Swiss statistics of social insurance (Schweizerische Sozialversicherungsstatistik)

Figures listed in Appendix.

5

Chapter six, Art. 59 et sqq. AVIG., AS 1982 2184.

6

Explanations of the federal council from the 2nd of July 1980, BBl 1980 489, 535 et sqq., 610 et sqq.

17

In terms of active programmes, a significant change was made in 1996 and 1997 when the second revision of the UIA was implemented. At the same time the employment offices were completely reorganised. While previously administrative and monitoring tasks were performed by communal employment offices, they were now assigned to regional placement offices, with individual counselling for unemployed persons. At the same time an activation principle was introduced, namely that active labour market programmes should be used widely to improve the chance of reintegration to the workforce. In order to achieve this goal, a part of the benefit entitlement was dependent on participation in programmes. Additionally, cantonal centres dealing with the logistics of these programmes were introduced. Since then, active measures have gained importance. This is reflected in expenditures for active labour market programmes (Figure 4), which have increased enormously since 1993. This increase is large not only in absolute figures, but also in relation to total expenditure.7

2.4 Specific regulations 2.4.1 Entitlement period and benefit exhaustion The UIA defines two fixed periods, one determining benefit reception, and one for contribution. Both periods are two years.8 For two years before registration the necessary contribution time has to be fulfilled, and two years from registration reception of benefits is first possible. The waiting period for receiving benefits should not be mistaken for the maximum entitlement period for benefits. The entitlement period can be shorter than the entitlement period, and the reverse situation is also possible (i.e., if at the end of the period not all daily allowances (see section 2.4.3) are exhausted and still there is no claim for the remaining benefits). A new entitlement period can be awarded only if the last period has expired and if the unemployed person again satisfies the minimal contribution period (i.e., a certain minimum time of gainful employment can be shown). Interim jobs during unemployment are credited for further benefit claims, and until 1996 employment programmes also had the same effect.9

7

These expenditures include project costs and further expenses, but not benefits paid to the unemployed during participation.

8

Art. 9 AVIG.

9

Art. 13 para. 2quater AVIG, AS 1996 273.

18

In 1996 restrictions for renewed entitlement periods were introduced by increasing the necessary minimal contribution period for persons who had been unemployed recently.10 However, this differentiation was cancelled 2003, such that repeated and firsttime unemployment are equally dealt with in the current system.11 This entitlement period system can lead to situations that appear somewhat unfair or even arbitrary. For instance, assume a person who was unemployed for 13 months then found a job, but lost it again after 11 months. The entitlement period of two years has thus expired, but since the minimum contribution period of 12 months is not fulfilled, this person does not receive benefits even though they have not exhausted the maximum number of benefits. In this situation adequacy of contribution and benefit duration is not fulfilled. The term "benefit exhaustion" therefore covers two situations, one in which the entitlement period is over (and a second entitlement period cannot be claimed) and the second in which the maximum number of daily allowances is exhausted.12 Certain cantons have a system of unemployment aid for people who have exhausted their benefits from unemployment insurance. In absence of such an aid, these persons do not get any further assistance, and in the worst case they have to claim public welfare payments. 2.4.2 Contribution rates When the UIA of 1982 was introduced in 198313, the contribution rate was fixed at 0.6% of the wage, paid in equal share by employers and employees. The contribution is levied only up to a maximum wage, as defined by the maximum insurable earnings according to the compulsory accident insurance. This maximal amount was 69,000 francs at the time of the final implementation of the law in 1984. In 1987 it was increased to 81,600 francs, in 1991 to 97,200, and since 2000 it has been set at a salary of 106,800 francs per annum.14

10

Art. 13 para. 1 AVIG, AS 1996 273.

11

AS 2003 1728.

12

More detailed studies examining the exhaustion of benefits by individuals can be found in Aeppli et al. (1996) and Aeppli, Hoffmann and Theiss (1998).

13 14

The insolvency insurance was implemented in 1983, all other parts of the law in 1984 (SR 837.01).

This corresponds to 5800/6800/8100/8900 francs monthly earnings. Sources: Law and Statistical Yearbook, different volumes.

19

Concurrently, the Federal Council readapted contribution rates several times within the narrow borders of the law (maximum 2%). It was reduced to 0.4% in 1990.15 In 1993, when unemployment insurance funds were close to exhaustion in the course of a recession, the Federal Council had to raise the contribution rate to 1.5%.16 However, before this change could be implemented it had to be readjusted upwards to 2%. The ongoing recession and high expenditures of the UI system made a further raise to 3% necessary in 1995.17 The contribution rate remained at 3% until 2002, though a part of this payment served for the amortisation of the UI system debt accumulated during the recession years. The 3% rate was originally to be kept until the end of 2003,18 but was reduced by the beginning of 2003 to 2.5%.19 Since 2004 the rate has remained at 2%.20 Unemployment insurance contribution rates.

Contribution rate up to insurable salary

2004

2002

2000

1998

1996

1994

1992

1990

1988

1986

1984

1982

1980

1978

.5

1

% 1.5

2

2.5

3

Figure 5.

Solidarity rate

Source: Law. Figures for 1977-1983 Statistical Yearbook, different volumes. Figures are listed in Appendix.

15

Ordinance concerning contribution rate in UI, AS 1989 1243.

16

Ordinance concerning contribution rate in UI, AS 1992 1371.

17

Resolution about measures in UI, AS 1994 3098.

18

Art. 4a, resolution about financing of UI, AS 1999 1298.

19

Ordinance from 29 November 2002, AS 2002 4288.

20

Art. 3 para. 2 AVIG.

20

As an extraordinary measure for amortisation of the accumulated debts, a so-called solidarity contribution was introduced. From the 1st of January 1995 an additional percent had to be paid on that part of the wage exceeding the maximum insurable wage by up to 2.5 times.21 The maximum insurable earnings were increased to 106,800 francs for 2000, and the solidarity contribution rate was increased to 2%. In 2003 the solidarity rate was again lowered to 1%, and in 2004 was cancelled altogether. 2.4.3 Benefit level (replacement rates) Unemployment insurance compensation is paid in daily allowances22, where five daily allowances are paid a week, 21.7 on average per month.23 The level of benefits is a replacement rate of the former insured earnings. When the UIA was introduced, this replacement rate was contingent on the age, family status (married and/or with dependants) and the number of already received allowances during the actual entitlement period. The replacement rate for a married person or for somebody with dependants was 80%, a single person without dependants received 70%. For a person below 55 years of age the replacement rate was reduced by 5% after consumption of 85 daily allowances (around four months), and by another 5% after 170 benefits.24 People living in certain economically disadvantaged regions were partly excluded from this reduction,25 and similarly exclusions were in place for persons with daily allowances below 90 francs and participants in active labour market programmes. As mentioned above, daily allowances are bound by the corresponding maximum insurable earnings (see section 2.4.2). Besides this maximum there is a minimum level of insured earnings required in order to be eligible for benefits. However, this minimum is so low (500 francs at the introduction of the UIA), that it is not addressed here.

21

The maximum insurable earnings were set at 97,200 CHF. Up to this amount 3% had to be paid to UI. For wage elements between 97,200 and 243,000 1% was levied, and the 243,000 francs exceeding part was still contribution-free.

22

The terms benefits, allowances and instalments are here used interchangeably, with allowances and instalments emphasising payments for single days.

23

Art. 21 AVIG, AS 1982 2184. The total benefits vary among months, according to the number of days and weekends. They are based on daily income (monthly salary divided by 21.7). Art. 40a, inserted in AS 1985 648.

24 25

Art. 22 AVIG, AS 1982 2184.

The definitions and applications of these regions varied over time (AS 1984 460, AS 1986 1064, AS 1987 566, AS 1989 1756, AS 1991 2242).

21

The first revision of the UIA, valid from 1992, generally increased replacement rates to 80%, in order to correct the differential treatment of married and unmarried persons. However, the duration based reduction was maintained. At this time the age limit for the reduction was lowered to 45 years instead of 55,26 and the special status of economically disadvantaged persons was cancelled.27 Urgent measures for the UI system reduced the replacement rate for persons without dependent children and with daily allowances of more than 130 francs again to 70% in 1993.28 At the same time the duration based reduction of benefits was cancelled.29 This legislation is still valid, with a small change of the threshold benefit from 130 to 140 francs in July 2003. An urgent resolution in December 1996 reduced benefits by 3% for persons without dependants and daily allowances of more than 130 francs and by 1% for all others.30 A referendum was called against this resolution,31 and the plebiscite overruled the resolution. Hence, this regulation was in force only for 11 months (January until November 1997). Figure 6 shows the replacement rates for different benefit groups by the number of already received allowances, which roughly corresponds to the duration of unemployment. The first graph shows the rates for those who received up to 85 allowances, the second graph for those who received 85 to 170, and the third graph for more than 170 already received allowances. Persons with dependants always received the maximum number, except for the original duration dependent reduction after 85 and 170 daily allowances. Older persons with dependants were also excluded from this reduction and received the maximum replacement rate of 80% throughout. A person with a lower income receives the maximum replacement rate since 1990. Before that they received 70% without reduction by duration. Persons without dependants and not fitting into the low income category always received the minimum rate. Elderly persons without de-

26

Explanations of the Federal Council regarding a part revision of the AVIG, BBl. 1989 III 377, and corresponding law change (Art. 22 para. 1 and 4 AVIG, AS 1991 2125).

27

Corresponding ordinance, AS 1992 655.

28

Art. 22 para. 1bis AVIG, changed in AS 1993 1066.

29

Art. 3 ordinance to the resolution, AS 1993 1268.

30

AS 1996 3459.

31

BBl 1997 117.

22

pendants were better off than the younger ones in the sense that they did not have to accept duration dependent reductions earlier. A particular technical detail is worth mentioning here. The transition from a 70% to 80% replacement rate is smooth within a certain bandwidth. If daily allowances for a person without dependants are below 140 francs computed with a replacement rate of 70%, but above the limit with 80%, benefits of exactly 140 francs are paid. Therefore, the effective replacement rate is between 70% and 80%.32 Figure 6.

Benefit level as replacement rate of former earnings.

2004

2003

2002

2001

2000

1999

Dec 1997

1996

No dependants Dependants

Jan 1997

1995

1994

1992

April 1993

1991

1990

1989

1988

1987

1986

1985

1984

1983

% of former income 68 70 72 74 76 78

80

Less than 85 benefits consumed

Low income

=18 months contrib. 55+ years, 1 year) relative to the total population of unemployed

Prevention of benefit exhaustion (0.2)

Number of benefit exhaustions relative to the total population of unemployed

Prevention of re-registration (0.1)

Number of re-registrations within 4 months after de-registration, relative to the total population of unemployed

An econometric model was used to correct these indicators for the effects of exogenous factors. They were put in a relative order, weighted with 0.5 (indicator 1), 0.2 (indicator 2), 0.2 (indicator 3) and 0.1 (indicator 4), and summed, such that they resulted in a cantonal efficiency index. The average of this index is set to 100, and is intended to represent the relative efficiency of the cantons compared to the average. This index provided the basis for compensation of the cantons' relevant costs for their regional employment offices and LAM. For 2000, the rate for efficient cantons was increased to 103%, but not lowered for inefficient cantons. 2001 saw the rate lowered down to 97% for cantons with an index lower than 98. In 2002 the corresponding bandwidth between the rates was widened to 105 and 95% respectively.

limit for periods with low unemployed (lower than 3%) one counsellor per 60 unemployed was determined (AS 1997 495). 78

Ordinance about AVIG execution cost sharing from the 29th of June 2001, AS 2001 2271.

36

When the contract had to be renewed for another 3 years this bonus malus system was rejected, because the underlying econometric model faced many opponents. Though the efficiency indicators are still computed, they are only used for a "yearly evaluation" of efficiency. The publication of the numbers is intended to induce "yardstick competition" and arouse interest in other countries.79 80

Placement services and active labour market programmes The costs for vocational re-training and education were paid by unemployment insurance from the beginning, if the cantonal administration consented to participation in the system. A difference between individual and collective courses was made. For individual courses the full costs were paid, but for collective courses only a part of the cost was covered. This rule resulted in courses being defined as individual, in order to be compensated fully for them. This flaw in the legislation was eliminated in the first UIA revision 1992, where collective and individual courses were put at par with each other. Employment programmes were partly co-financed (maximum 50%) only if the cantons also contributed adequately.81 The rest of the costs were imposed on the organiser, which is either the canton or the community. Employment programmes were assigned mainly to persons who had already exhausted their benefits. The unemployment insurance system wanted to encourage early intervention and increased the subsidy for employment programmes to 85% (in exceptional cases up to 100%) for those persons who have not yet exhausted their benefits.82 The second revision of the UIA standardised the cost share of the cantons in active labour market policies. The cantons had to pay a fixed amount of 3000 francs per yearly slot in a programme, up to the minimum level of places.83 If a canton provided more than the required minimum, no contribution had to be paid. If a canton was below the average costs for programmes, its share was also reduced. If cantons did not comply with the requirements, this was associated with a cut in contributions. For 79

See Imboden et al. (1999) and Lenz, Egger and Zürcher (2001) for further information about the incentive system.

80

Hilbert (2004) provides an overview of the Swiss incentive system and a thought experiment for Germany.

81

Art. 96 para. 1 AVIG, AS 1991 2132.

82

Explanations of the Federal Council, BBl 1993 I 677, and resolution AS 1993 1066.

83

Art. 60 para. 5 AVIG, AS 1996 273.

37

placement services the canton's share in costs was set between 20 and 40% of the corresponding costs. If they did not fulfil the minimum quota of places, they were asked to pay 20% of the substitutional special benefits that are paid instead of programme participation, if a programme is not available. This scheme was thought to give cantons adequate incentive to provide enough places at low cost, and to assign the unemployed to programmes. However, system actually triggered an over-supply of measures by the cantons. A technical revision, implemented in 2001, cancelled the minimum places requirement and fixed the cantonal contribution for programmes at 10%, instead of keeping it constant per place as before.84 A canton i's financial contribution fi to the overall costs C of active programmes was calculated according to the following formula:

fi = 0.1 C

bi , B

with bi benefits paid in canton i, and B total benefits paid in Switzerland. The third revision, implemented in July 2003, standardised all costs of placement and programmes together. The total contribution of the cantons is fixed at 10% of the overall costs. A specific canton's share is determined by the following formula:85

fi = e0.0045 IFCi di F where IFCi is an index for the financial strength of a canton in the corresponding fiscal year, di days of unemployment in respective canton i in corresponding year, F = 0.1 C total contribution of cantons (10% of total expenditures), and a constant, fixed such that sum of all contributions equals the total cantonal contributions

F=

fi . i

3 Studies about effectiveness of Swiss labour market policy The second revision of the Swiss unemployment insurance act in 1997 heralded a new era by emphasising the active role of jobseekers, instead of passive maintenance. This

84

Art. 72c AVIG, AS 2000 3093.

85

Art. 9 AVFV, SR 837.141.

38

change in thinking was associated with a substantial increase in spending on active labour market policy. The Swiss State Secretariat for Economic Affairs (seco) mandated several researchers, institutions and consultancies with the evaluation of Swiss ALMP and placement services. The research groups used different methodologies and data. Evaluating ALMP in Switzerland thus became a major research topic. About half of the studies were conducted used self-collected data. Martinovits-Wiesendanger and Ganzaroli (2000) evaluated two very different courses, namely Winword and the hotel/restaurant industry, and employment programmes in different areas of Switzerland. They used a telephone survey of former participants and also control groups. These researchers found that the courses had a positive effect on people’s chances of subsequently finding stable employment. Hunold (1998) evaluated the quality of the regional placement offices (RAV) by conducting a telephone survey with jobseekers and firms, with results supporting the effectiveness of the RAV. Another report by ATAG (1999) dealt with the effectiveness of the regional employment offices subject to active labour market programmes. On the basis of this report the above-mentioned agreement between the federal department and the cantons was built. The econometric methods used in all of these studies consist mainly of descriptive analyses and basic OLS regressions. Falk, Lalive and Zweimüller (2005) applied an experimental approach, where the CVs of computer course participants were sent to different employers before and after participation (and therefore either with or without course completion). The effect of the computer course was negative but was not significant. Bauer, Baumann and Künzli (1999) used self-collected data from interviews with case workers and jobseekers in combination with data extracts from the Swiss unemployment insurance system (AVAM and ASAL). They investigated the effects of subsidised interim jobs, finding that they were generally rated as favourable. All other studies primarily used these AVAM/ASAL administrative records provided by the State Secretariat of Economic Affairs (seco). Sheldon (2000) investigated the productivity of the public employment service in Switzerland by comparing RAV in terms of different efficiency measures, controlling for regional circumstances. He found that the introduction of regional placement offices had increased the efficiency of labour market policy. Ferro-Luzzi et al., (2001) also investigated the effectiveness of regional employment offices. Prey (2000) evaluated ALMP measures in the canton of St. Gallen using propensity score matching. Vassiliev (2003) evaluated active labour market programmes for the canton of Geneva, concluding that they prolong unemployment duration. In contrast, Curti (2002) reports the programmes have positive effects. A study by Lalive, van Ours and Zweimüller (2002) revealed that benefit sanctions

39

reduce the duration of unemployment. Lalive (2001) also provides empirical studies of unemployment duration in Switzerland. Two groups used the AVAM and ASAL combined with social security (AHV) data. Gerfin and Lechner (2002) applied nonparametric matching to the propensity score approach in order to estimate the effects of different labour market programmes. Lalive, van Ours and Zweimüller (2001) chose a parametric duration model approach. Lechner and Smith (2003) studied the effectiveness of the assignment of programmes and find no differences between a random assignment of programmes and the allotment done by caseworkers. They conclude that the assignment process can be significantly improved. Frölich and Lechner (2004) estimated the effects of programmes with an instrument variable technique, using differences in regional treatment intensity as an instrument. They report positive, but non-significant effects for the programmes. Aeppli (2004) found positive effects for employment programmes on persons who had exhausted their benefits. A new "round" of evaluation of ALMP is now being undertaken (2004 to 2006).

4 Micro data resources Different data resources for research about unemployment insurance in Switzerland are available at a micro level. The Swiss Household Panel86 and the Swiss labour force survey are both surveys representative of the total population in Switzerland. They are publicly available and have detailed information about employment of individuals. However, for more extensive studies the number of unemployed in corresponding datasets might be too small. For this specific purpose, unemployment insurance records have been made available by the Swiss State Secretariat for Economic Affairs (seco). These records contain rich information about the unemployed and job seekers. Moreover, they can be linked to pension system records by unique personal identification numbers. These two sources have been made available for the specific purpose of research, and together they provide a database which contains uniquely detailed information about job-seekers. This chapter describes this data, together with the relevant administrative procedures which generate it.87

86

For further information see www.swisspanel.ch.

87

Part of this chapter has been published (Gast, Lechner and Steiger, 2004).

40

4.1 Introduction The information system for placement and labour market statistics (Arbeitsvermittlung und Arbeitsmarktstatistik, AVAM) and the unemployment offices payment system (Auszahlungssystem der Arbeitslosenkassen, ASAL), as well as the social security system (Alters- und Hinterbliebenenversicherung, AHV), are information systems used for administrative purposes. The AVAM system is an online information database for regional placement offices. All information about job openings, characteristics of the individual job seeker, services provided, and interactions between caseworkers and job seekers is contained in this system. A person remains in the information system of AVAM as long as he or she makes use of the services of the UI system, regardless of employment status. The AVAM system is not only used for internal administration, but certain information about job seekers can be accessed by employers for recruitment purposes. The part of the database with open positions can be publicly accessed via the internet. The ASAL system is the database of the unemployment insurance fund. It contains all information about payments and benefit entitlements. Although AVAM and ASAL are independent systems, they are linked in the sense that certain information is exchanged. At the end of each month all existing entries in AVAM and ASAL are saved and archived as flat files for statistical needs. The data cover all persons who are registered at any time at a regional placement office since 1993. Until now the statistical section of seco has supplied recipients with data, tables and analyses by submitting host programs. Computer technology has since progressed, and a data warehouse is currently under construction. It will contain the (statistical) AVAM and ASAL data, already connected in an entity relationship model. Data will be loaded monthly, and perhaps at a later point even daily. Users will have direct access via the internet and will be able to access existing tables or even exploit the data themselves according to their specific rights. This information will no longer be stored in monthly records of stocks and flows. Instead, unemployment spells will be constructed with the AVAM data. Contributions to unemployment insurance are processed together with the obligatory contributions to the pension system. Social security numbers serve as the primary identification for each person. Therefore it is possible to link entries in the two different information systems by that number.

41

This data, as obtained from the registrar, is not immediately suitable for scientific analysis and research, rather it needs to be processed carefully. This data may contain incorrect or inconsistent entries, as well as corrections to some entries. Thus in order to evaluate the information extensive and detailed knowledge about the exact data generating process is needed. However, once this time consuming and demanding task is mastered, a very rich and reliable database is at the researcher’s disposal.

4.2 General Overview Taken together, AVAM, ASAL and AHV form a broad information base about the unemployed in Switzerland. The disadvantage of the AVAM and ASAL data is mainly their limitation to the unemployment and job seeking status of a person, as there is no compulsory follow-up when a person has exhausted their UI benefits. However, this shortcoming can be partly compensated by the addition of the AHV records, which are available for employed as well as for unemployed individuals.

AVAM AVAM contains information about job seekers, including interactions with their caseworker, services, job openings, and employers. Data is organised in monthly records based on the last working day of each month. In addition, records of entrants and exits from the register are archived in separate files. Table 2.

Variables available in AVAM.

Personal characteristics

date of birth, sex, marital status, number of persons in same household, canton and place of residence, nationality, type of work permit, mother tongue, foreign languages, education1 Labour market related characteris- mobility, qualification, caseworker’s rating of chances to find a job, position tics in last job, occupation and industry of last job, size of town where worked before, looking for part-time/full-time job, occupation and industry of desired job Information related to current date of registry and de-registration, reason for de-registration, unemployunemployment spell ment status Services obtained type of programme or subsidy, start and end date, date of caseworker’s decision about programme 1

Variable was introduced in 2001; in January 2003 still missing for 20% of all registered persons.

42

Not every caseworker enters information with the same diligence. While certain entries are updated regularly and carefully, others are not, depending on the person responsible for the respective client. Therefore the quality, timeliness and reliability of information vary widely. Table 2 provides an overview of variables that are of interest for labour market research. Note that until 2001 no direct information about education or schooling was available.

ASAL ASAL is the payment information system of the unemployment insurance system. Data is provided in monthly records saved the last working day each month. Information is directly relevant for payments to benefit recipients and therefore can be rated as being much more reliable than AVAM. It contains first and foremost all information about an individual job seeker necessary to determine benefits. An ASAL record is available only if a person actually receives benefits. Once the job seeker has exhausted their UI benefit entitlement period there is no trace of them in this database, though they can still be registered as unemployed at the regional employment office (and thus has an AVAM record). The relevant information is listed in Table 3. Table 3.

Variables available in ASAL.

Personal information Monthly payments

earnings in last job, relevant duration of contribution to unemployment insurance benefits, begin and end of entitlement period, disability, number of dependent persons gross and net benefits, children allowances, sanction days, sickness and accident benefits, subsidies for interim jobs, additional payments for programme participation

AHV In the social security records earnings from all sources that are subject to contributions to the pension system are archived (i.e., from employment, self-employment, and unemployment). The source of income can be identified in most cases. The data consists of yearly files with records indicating the months of the beginning and the end of each earnings period. The nationality of the person and the branch of the pension insurance administration in charge for this record are additional variables provided by this data (refer to Table 4).

43

One problem with these records is that they are entered when earnings are reported and contributions are paid. There can be a significant time lag between the declaration of the payment and the period the declaration is related to. Corrections for a period are sometimes made years later, and some entries are delayed entirely for several years. Therefore, missing information in very recent data can be due to actually being out of labour force (and thus not subject to contributions), receipt of welfare benefits or just delayed declaration of earnings. It is thus best practice not to rely on the most recent portions of this data. Table 4.

Variables available in AHV.

Personal information Income

Nationality income in CHF, source of income (employment, unemployment, self employment, self contributors without employment), corresponding period in months, year of declaration, corresponding branch of social insurance system, code for employer

4.3 Access to data All data mentioned here are subject to privacy protection because they contain very sensitive personal information about individuals. Access to this data is restricted. Since the social security number codes personal information according to a certain public key, for privacy reasons the number is not given to the researchers directly. To have access to AVAM and ASAL data researchers have to sign specific contracts for data protection. It is usually required that the data be deleted one or two years after the project is completed. Preferred access to AVAM and ASAL records is given for research studies conducted on behalf of the seco itself. Second priority is assigned to projects of the Swiss National Science Foundation, and third to other Swiss administrative projects or academic studies. The federal social insurance office in charge of the AHV data has been very restrictive in providing the data for external studies that link them to other data sources. Until now, only two research groups have been given access for microeconomic evaluation studies by order of the seco.

44

4.4 Some empirical studies based on these data The unemployment insurance system has only been in place since the early 1980s and at that point did not cover all cantons. Sheldon (1989) provides an illustration of the dynamics of unemployment in Switzerland on the basis of these data and gives an assessment of the database itself. Recently, the data have been used primarily to evaluate the effectiveness of regional employment offices and active labour market policies (ALMP) in Switzerland. Sheldon (2000) investigated the productivity of the public employment service in Switzerland by comparing regional placement offices in terms of different efficiency measures. A study of Ferro-Luzzi et al. (2001) identified determinants of ineffectiveness of regional employment offices. Bauer, Baumann and Künzli (1999) combined AVAM/ASAL extracts with data from interviews with case workers and job seekers to investigate the effects of subsidised interim jobs. Curti (2003) evaluated Swiss active labour market programmes for Switzerland, Prey (2000) in the canton of St. Gallen, and Vassiliev (2004) for Geneva. Combinations of AVAM and ASAL with social security (AHV) data were used by Gerfin and Lechner (2002), Gerfin, Lechner, and Steiger (2003), and Lalive, van Ours and Zweimüller (2002b) in order to estimate the effect of different labour market programmes. A study of Lalive, van Ours and Zweimüller (2002a) investigated the effect of benefit sanctions on the duration of unemployment on the basis of AVAM and ASAL data. A study by Lechner and Smith (2006) examined the performance of caseworkers in allocating people into labour market programmes. Ongoing research in Switzerland about the effectiveness of active labour market policy is also based on these data, but no results have been available until now. Fehr and Götte (2005) use AHV data only for determining wage rigidities.

4.5 Outcomes in data sources 4.5.1 Definition Three terms usually provide the different labour force states of a person, namely unemployed, employed, and out of labour force.88 The term unemployed usually subsumes those persons who are looking for a job or waiting for their job start and who are not employed. However, there are also persons who are employed but not to the 88

The two terms "inactive" and "out of labour force" are used interchangeably for the same state.

45

extent they desire (under-employment). Furthermore, a person can be found to be registered at the unemployment office while in full-time employment, or in subsidised employment (outcome states are defined in Table 5). A person is out of labour force if they are neither employed nor looking for a job.89 The labour market states investigated in the current research were unemployment, part-time unemployment, employment and inactivity. Retrieval of information about these outcome states is described and compared between the two data sources. Table 5.

Information about labour market status in UI and PS data.

Information 1 Full employment

2 Part-time unemployment 3 Unemployed with benefit entitlement 4 Searching for a job 5 Inactivity

Unemployment insurance A person is de-registered and does not receive payments or other subsidies any more; at the same time the last de-registration code was "found a job"; additionally, we can see if a person is full-time employed while searching for a job90 A person is registered and at the same time part-time employed91

Pension system A person has a positive entry of a salary from employment and no entry from unemployment

A person has a positive entry of a salary from employment and also a positive entry from unemployment A person has a positive entry for unemployment benefits

The person is registered, is still in the corresponding entitlement period, and receives payments or is in an active labour market programme A person is registered Person is de-registered with a deNo entry in the pension system registration status announcing "no job found"

All information is on monthly basis.

While unemployment insurance data provides extensive information about people during unemployment, these databases contain little information about people after they leave unemployment. Only some forms of employment are observable during their registration (unemployment status variable), especially if a person is part-time employed or full-time employed while being registered. Information about de-registration 89

This term does not incorporate the willingness to work, but only that no time is invested into search efforts. 90 91

According to variable "type of job-seeker" (ARTST) in AVAM.

This variable is set at 1 if the "type of job-seeker" (ARTST) indicates employment, or if the person receives a compensation for interim job.

46

status is also available from the regional unemployment office (towards employment or non-employment). On the other hand, the pension system data provides information about employment and unemployment. Combining the two data sources can lead to a very rich data set. However, there are a few problems with interweaving these two sources. Firstly, the pension system data is only available with some time lag (1.5 years), while the unemployment insurance data is available almost instantly. Secondly, entries into the pension system sometimes be delayed by up to a decade, leading to an permanently incomplete dataset. Furthermore, the pension system data contains some fuzziness in the definition of beginning and end of an income spell, such that an employer might not have any reason to state the right month of beginning and end of a spell in case the spell did not cover the full calendar year. No research has yet examined the congruency of pension system and unemployment insurance system data, and identified what can be learned from the combination of the two data sources. In the pension system data, only incomes from unemployment or employment can be observed and distinguished (and therefore the person’s corresponding status). If a person has only an income from employment, it can be concluded that they were employed during that month. On the other hand, if they have a positive unemployment benefit entry, this means that they received benefits or subsidies from unemployment insurance and therefore has to be counted as unemployed. If there is no entry in the pension system, this can be interpreted such that the person has chosen to opt out of the labour force, or that they are still actively searching for a job, but have exhausted their benefit entitlement. Additionally, this last case can also be due to the problem of delayed entries. The unemployment insurance data contains much information about unemployment and receipts of benefits, but the information available after the exhaustion of benefits can be very vague. Once a person has exhausted their benefits they do not necessarily have an incentive to remain in the registry of the regional employment office, even if they are still actively searching for a job. If they leave unemployment to take up a job, this can be observed in the data, as the employment officer sets the de-registration code accordingly.92 On the other hand, if they leave the registry without actually finding a job, they cannot be followed up later unless a re-registration follows.

92

It is unknown how seriously and reliably this information is entered by the caseworkers, or if a job seeker always announces to the officer that they have found a job. See section 0 for de-registration codes.

47

4.5.2 Data and descriptive statistics During the current research data was gathered from persons registering between January and December 2000, and for whom both unemployment insurance as well as pension system data is available. The sample consists of 128,569 people. These people were followed until December 2002, and thus each was observed for at least two years after their registration. The indicators 1-5 are computed monthly for every person. Figure 9 displays employment patterns for this group. The unemployment insurance data always shows lower employment than the pension system data. 1.1% of people are employed throughout in both the datasets. Employment has a typical seasonal pattern, such that after one and two years, employment rates come down again. Also note that around 40 to 47% of persons are still employed when they register. They register one to two months before they actually stop working. The difference in employment rate between the two sources is on average 6.7%. 5.4 and 17.6% of people are never employed according to the PS and UI data respectively, and 3.9% overall. 3.3 and 5.5% of persons are always employed (PS and UI respectively), and 0.9% overall. Employment.

0

20

employed (%) 40

60

80

Figure 9.

3

6

9

12

15 18 21 24 months after registration UI data

Own calculations.

27

PS data

30

33

36

48

Unemployment.

0

20

unemployed (%) 40 60

80

100

Figure 10.

3

6

9

12

15 18 21 24 months after registration

Unemployed (UI) Unemployed (PS)

27

30

33

36

Searching (UI)

Own calculations.

Figure 10 displays the statistics for unemployment and job-search. The data from the pension system underestimates the receipt of unemployment benefits every month. Again it is clear that persons register before they experience actual job loss. According to the UI data, a maximum of 68.5% of persons receive unemployment insurance in the second month of their registration. 10.1% in the PS compared to 6.9% in the UI never have any payment from unemployment insurance. 0.5% do not have any entry in both the data sources. We can see a further drop again from month 25 to month 26, where unemployment benefits go down, most probably due to benefit exhaustions. In terms of registration, 53% de-register within the first six months. Only 32% are still registered after one year, and 24% after two years. Figure 11 displays the descriptive statistics for part-time unemployment. These are the persons who receive unemployment benefits, but are also in some employment or an interim job. The unemployment insurance data shows a higher rate of part-time unemployment than the pension system data after approximately six months. Part-time unemployment seems to be measured particularly inaccurately at the beginning of the spell, when it might be due to an ongoing work contract. In general, the share of parttime unemployment significantly drops after around two years. The two sources do differ, sometimes substantially (2.5% in month 10 at a level of 9.8 and 12.3%).

49

Part-time unemployment.

5

part-time unemployed (%) 10 15 20

25

Figure 11.

3

6

9

12

15 18 21 24 months after registration UI data

27

30

33

36

30

33

36

PS data

Own calculations.

Inactivity.

0

5

inactive (%) 10 15

20

25

Figure 12.

3

6

9

12

15 18 21 24 months after registration UI data

Own calculations.

27

PS data

50

The inactivity indicators show considerable differences, especially at the beginning, because pension system cannot "see" if a person is searching for a job or not. Hence, during the first few months the differences are those who are registered, but do not receive benefits. Not surprisingly, there is an increase at around month 25, when for most people benefit entitlement is exhausted. However, the unemployment insurance data usually overestimates corresponding inactivity rates. 4.5.3 Congruency and accuracy of data The compatibility of the data sources (the unemployment insurance and pension system) is measured by comparison of different states. Given that a person is in a particular state in one of the data sources, this can be contrasted with the state the other source indicates. The outcomes employment, unemployment part-time unemployment, and inactivity can be compared. Figure 13.

Comparison of states in unemployment insurance data. Employed

0

0

20

20

40

40

60

60

80

80

100

100

Unemployed

3

6

9

12 15 18 21 months after registration

Employed in PS Unemployed in PS

24

27

30

3

6

Inactive in PS Empl. + Unempl. in PS

9

12 15 18 21 months after registration

Employed in PS Unemployed in PS

27

30

Inactive in PS Empl. + Unempl. in PS

Part-time unemployed

0

0

20

20

40

40

60

60

80

80

100

Inactive

24

3

6

9

12 15 18 21 months after registration

Employed in PS Unemployed in PS

24

27

30

Inactive in PS Empl. + Unempl. in PS

Own calculations. otal number of observations 128,569.

3

6

9

12 15 18 21 months after registration

Employed in PS Unemployed in PS

24

27

Inactive in PS Empl. + Unempl. in PS

30

51

Figure 13 depicts the comparison for the unemployment insurance data. For example, the first graph shows that of those unemployed (according to UI), after two months around 90% information is compatible. However, roughly 35% of persons also show an income from employment at the same time, and almost 10% are employed only. During the first month of the spell, the information is the least compatible. The congruence concerning employment (de-registered towards a job) is rather high, as around 90% of persons are employed according to the PS as well, and the difference is inactive. The congruence decreases over time, which is consistent with the fact that the information in the unemployment insurance is only the de-registration status at the time of de-registration and is not altered further if the person does not re-register. PS data, on the other hand, are updated monthly and indicate a possible job loss. Inactive persons are found to be employed to between 50 and 60% (note that there are only two observations in the first month). Congruence of this type of outcome is typically quite low. Around half the persons who are de-registered without having found a job are employed in the pension system. However, it unknown if this really means that they found a job, or they rather continue a part-time job they had earlier. Those who are indicated as part-time unemployed have the same status in around half the cases. A considerable share of around 15% are employed, without receiving benefits from UI. Figure 14 displays the comparison in the opposite direction (the pension system data). The first item of note is that the congruence in terms of receiving unemployment benefits is nearly perfect throughout the observation period. Note that since states are not exclusive, the other outcomes cannot be interpreted. Entries concerning employment are not congruent at the beginning of a spell, because the possibility of transition problems from employment to unemployment. The more time passes, the more equal the systems become. After around six months, 75 to 80% of persons are employed according to the pension data as well. However, it is clear that a considerable share of persons de-registered from the UI system without announcing a job. Inactive persons are employed according to the UI system in almost 40% of the cases. Another 45% are de-registered without job in the longer run. The pension system data contains only income information and thus overstates out of labour force status, as it is clear that many persons are registered with UI and are looking for a job. Part-time unemployed people (essentially unemployed and employed at the same time) are indi-

52

cated as part-time unemployed in the UI system in around 50 to 60% of the cases. The pension system does not appear to capture this outcome well. Figure 14.

Comparison of states in pension system data. Employed

0

0

20

20

40

40

60

60

80

80

100

100

Unemployed

3

6

9

12 15 18 21 months after registration

Employed in UI Unemployed in UI Inactive in UI

24

27

30

3

6

Part-time unempl. in UI Registered in UI

9

12 15 18 21 months after registration

Employed in UI Unemployed in UI Inactive in UI

27

30

Part-time unempl. in UI Registered in UI

Part-time unemployed

0

0

20

20

40

40

60

60

80

80

100

100

Inactive

24

3

6

9

12 15 18 21 months after registration

Employed in UI Unemployed in UI Inactive in UI

24

27

30

Part-time unempl. in UI Registered in UI

3

6

9

12 15 18 21 months after registration

Employed in UI Unemployed in UI Inactive in UI

24

27

30

Part-time unempl. in UI Registered in UI

Own calculations. Total number of observations 128'569.

4.5.4 Delayed entries in pension system data Pension system data are available with some time lag only. This lag is around 1.5 years. Even if data are processed by the AHV administration and made available for research, a considerable portion of the data is missing. The board provides a rough estimate of 10% for the most recent year. Entries are delayed by up to 10 years. The delay can be due to late announcement of correct wage data by employers, corrections,

53

or due to delayed administrative exchange of data between cantons and the central federal board. Employer-related reasons are mentioned as the main causes.93 Table 6.

Delayed entries for employment spells in the pension system by year.

Delay 1991 1992 1993 0 year 12.99 9.85 9.13 1 year 81.97 85.27 85.24 2 years 2.77 2.55 3.37 3 years 0.99 1.18 0.90 4 years 0.76 0.65 0.57 5 years 0.37 0.24 0.54 6 years 0.07 0.21 0.16 7 years 0.04 0.02 0.04 8 years 0.01 0.01 0.02 9 years 0.01 0.01 0.01 10 years 0.01 0.01 0.00 N 129,435 129,787 130,211

1994 10.46 83.18 4.16 0.79 0.90 0.33 0.12 0.03 0.02 0.01 0.00 143,438

1995 14.31 79.32 2.82 2.24 0.62 0.40 0.20 0.06 0.01 0.00

1996 8.67 67.58 21.67 0.93 0.60 0.35 0.13 0.06 0.01

1997 11.68 68.87 17.39 0.89 0.59 0.39 0.17 0.03

1998 10.64 79.10 8.32 0.82 0.71 0.35 0.06

1999 7.33 84.11 6.97 0.87 0.55 0.17

2000 14.29 80.19 4.38 0.88 0.26

2001 10.58 85.27 3.60 0.55

154,682 163,255 174,772 180,668 217,970 264,411 201,654

Own calculations. Number of employment entries in 1999 only. Sample: Persons with registration in 2000. Total number of persons: 128,569.

The data contains the year of entry for every record. Table 6 displays a summary of the data by year of entry. "0" denotes that the entry was made during the same year (e.g., the entry for 1991 was made in 1991). The numbers are based on the population of all persons starting an unemployment spell in 2000 (128,569). These are the percentage of employment spells delayed by corresponding time. There is considerable heterogeneity among the years as to what extent the entries are delayed. For 1996 and 1997, a larger fraction is announced two years later. However, for the years 1991 to 1994, between 93.7 and 95% of observable entries were made within one year. For the more recent years 1999 and 2000, 91.4 and 94.5% of employment entries were made within one year.94 In 2001, the collection procedure for the contributions changed, with more

93

The research could not detect the real reason for these delays, because employers have to pay substantial interest on delayed payments.

94

A probit regression was run for the year 1999 with the dependent variable "entry is delayed by at

least 2 years" and control variables for the timing of the beginning of the period covered (month dummies), duration of the income spell, and some individual characteristics (gender, qualification, age). Furthermore, a variable is included which notes if there is a parallel income spell overlapping for at least one month. Most coefficients were highly significant, and especially the coefficient of the parallel spell at the same time is high.

54

pressure on employers for punctual payments. This data covers entries made up to the first half of 2004. Hence, data for 2002 are not listed in the table. Figure 15 displays the incoherence across the two systems in terms of employment data in calendar time. It reveals two types of inconsistencies, first if there is a positive employment entry in the pension system, but not in the unemployment insurance, and second, the opposite pattern. The first type of inconsistency seems to be unrelated to time, whereas the second type follows a time profile. For this second type, after a decline in the first 12 months (this also corresponds to pre-registration periods), the inconsistency increases over time, with distinct steps at the beginning of a new calendar year (months 12 and 24). Incoherence of employment data in UI and PS.

0

5

10

15

20

Figure 15.

0

10

20 calendar months (1=jan 2000)

PS: empl, UI: not empl.

30

40

PS: not empl., UI: empl.

Own calculations.

This finding could be the result of people losing their jobs without re-registering to the unemployment insurance system, which leads to an over-estimation of employment by UI over time. However, it is also fully consistent with information about delayed entries, specifically the obvious "steps" observed at the end of every calendar year. The inconsistency of the two sources is 3% in calendar month 12, this jumps to 3.7% in month 13, at the end of the second year it increases from 6.0 to 6.9% and reaches 7.6%

55

in month 36. Hence, between the end of 2002 and the end of 2000 4.6% precision is lost. This loss can be partly attributed to missing entries in the pension system. 4.5.5 Summary and conclusions This section can be summarised and conclusions drawn as follows: There is a very high congruency between unemployment insurance and pension system data in terms of unemployment and receiving benefits. False entries are rare in the PS data, and corresponding entries in the UI data are found in PS for approximately 90% of the cases. Being directly relevant to benefit payments, the UI data is more reliable for actual unemployment information. The UI data underestimates employment when persons de-register without indicating a job. Some persons might not announce a successful application, continue with job-search after de-registration, or continue a part-time job that they had already. On the other hand, UI overestimates employment, especially in the longer run when persons de-register for a job, as job separations cannot be monitored. Pension data underestimates unemployment as well as employment. The most recent year(s) of the pension data are affected by delayed entries and not reliable. However, the delay is not as large as official statements (10%) would suggest. Part-time unemployment might be considered as an additional outcome, as it measures different combinations between unemployment and employment.

5 Unemployment profiles This section provides a short analysis of the unemployment profiles of various subgroups, as defined by personal characteristics. This data is from all unemployed persons registering in 2000, which are followed until the end of 2003, based on information from the unemployment insurance system. All re-registrations are taken into account. Section 0 describes the data and 5.2 displays the subgroup profiles. Sections 5.3 to 5.5 summarise the typical use of labour market programmes, sanctions, and interim jobs for the corresponding sample.

56

Distribution of unemployment duration.

0

.001

Density .002 .003

.004

.005

Figure 16.

90 180

365

730 days

1095

Own calculations. Duration of first unemployment spell starting in 2000. Number of observations: 128,569. Mean: 235.9 days. Std. dev.: 255.1. Percentiles 1%: 12 5%: 28 10%: 38 25%: 70 50%: 139 75%: 294 90%: 618 95%: 778 99%: 1218.

5.1 Data The sample presented here was described above in section 4.5.95 There are two distinct exit states in the unemployment insurance: inactivity and employment. Since these individuals were followed for three years, they can have multiple periods of unemployment (in the data up to nine). The average duration of these periods is almost eight months (see Figure 16 and corresponding note), although 50% of all first spells are shorter than five months. The distribution has a mass point at around two years, when usually entitlement is over. Individuals in the sample had up to nine different consecutive spells starting until the end of the observation period (end of 2003). The distribution of the number of spells is as follows (Table 7).

95

Individuals missing important data points (1,705 observations) were excluded from this sample.

57

Table 7.

Distribution of number of spells.

Number of spells 1 spell 2 spells 3 spells 4 spells 5 spells 6 spells 7 spells 8 spells 9 spells

Share in sample 52.27% 29.97% 11.71% 4.60% 1.04% 0.24% 0.10% 0.07% 0.00%

number of obs. (67,203 obs.) (38,526) (15,057) (5,920) (1,339) (309) (126) (86) (3)

The duration between the first and the second unemployment spell is on average 395 days. 10% of second observed spells start within two months after completion of the first spell, and 50% start within ten months. Table 8.

De-registration codes.

De-registration code Missing 11 placed by communal placement office 12 placed by regional placement office 13 placed by cantonal office 14 found a job independently 15 found job through private placement office 16 other reason 17 found job through placement by private placement with regional employment office Total found job 21 did not appear for counselling 22 does not want placement services any more 23 moved away 24 not employable 25 other reason Total left without finding a job Total

Cases 2,142 327 23,276 183 63,682 1,804 2,905 1,081 95,400 4,949 7,007 1,437 3,075 14,871 31,339 126,739

% 1.7 0.3 18.4 0.1 50.3 1.4 2.3 0.9 75.3 3.9 5.5 1.1 2.4 11.7 25

Own calculations. All de-registration codes of all completed first spells starting in 2000.

For every period of unemployment there is a de-registration code. This code has 12 possible values between 11 and 25, of which values below 20 indicate that the person has found a job and values above 20 that the person has not found a job. The actual deregistration codes of the first periods of unemployment in this sample are listed in Table 8. Most individuals found a job on their own (around two-thirds of those who found a job), followed by those placed by the regional placement office (around onefourth). “Other reasons” are less important for those who find a job. For those who de-

58

register without finding a job, there is no compelling evidence why they do so. Almost half of those people leave for "other reasons".

5.2 Subgroup profiles For illustrational purposes, the "profiles" of the unemployed in the sample are displayed in terms of the states they go through after their first registration in 2000. This data follows them to the end of 2003.96 Figure 17 displays the profiles for all unemployed individuals. Focusing on the "registered" group, it is clear that there is a cyclical movement, with re-registrations every 12 months (365 days). Note that after day 1095 the number of observations decreases, because those individuals registering in late 2000 cannot be observed any longer. The share of persons registered does not disappear, but remains always above 17.5%. However, these are not always the same people. Only 1.4% of the sample is constantly registered. Profile, all unemployed.

20

40

60

80

100

Figure 17.

1

90 180

365

547 730 days after registration

Left w/o job Found job In course

1095

1460

In employment prog. Registered w/o progr.

Own calculations. N=128,569.

96

Note that all figures for corresponding subgroups are based on the total sample and are not controlled for further individual characteristics.

59

Regarding employment, a maximum of 60.2% is de-registered towards employment after 640 days. After two years, when the entitlement period is expired for most of the persons, there is an increase in the share of "inactive" persons, those leaving without job, from 17.8% right before to 21.5% three months later. No overall increase in total employment can be observed at that point of time. The share of registered persons in courses displays a maximum of 10% after around four to five months and decreases over time to 3.8% after two years. 9.3% of people still registered after around 11 months are in employment programmes, which is the maximum share. Figure 18 displays the sample as divided by gender. Women appear to be less exposed to cyclical re-registrations, most probably because they are working in industries that are less cyclical. In the long run, they have a higher job finding rate than men (60.1% after 1040 days compared to 57.6% of men). Women are more likely to be in a course (11.5% after five months compared to 8.8% of men) and are less likely to be in an employment programme (8.5% after one year compared to 9.1% of men). In general, women seem to be less attached to the labour market if they do not find a job quickly. In the longer run, they are less likely to be registered (17.5% after three years compared to 22.2% of men) and slightly more inactive. Figure 19 displays the unemployment profiles according to nationality and permit status. Swiss persons have the highest job finding rates and lowest inactivity rates. Persons with permits of type C have the lowest attachment to the labour market; they are the least likely to find a job, and they are more likely to leave the registry without finding a job. Foreigners with a B type permit are more likely to remain registered. Foreigners' joblessness is more cyclical than that of the Swiss. Persons holding a B type permit have the highest likelihood of being enrolled in a course (14.5% after five to six months), and in the second year they are more likely to be in an employment programme than Swiss and foreigners with permit C.

60

Figure 18.

Profiles for women and men. Men

20

20

40

40

60

60

80

80

100

100

Women

1

90 180

365

547 730 days after registration

Left w/o job Found job In course

1095

1460

1

90 180

365

In employment prog. Registered w/o progr.

547 730 days after registration

Left w/o job Found job In course

1460

In employment prog. Registered w/o progr.

Inactive

1

90 180

365

547 730 912 days after registration

1095

0

0

20

5

40

20

10

60

15

40

80

20

60

25

Employed

100

Registered

1095

1

90 180

365

547 730 912 days after registration

1095

90 180

365

547 730 912 days after registration

In training courses

0

0

2

5

4

6

10

8

10

15

In employment programmes

1

1

90 180

365

547 730 912 days after registration Women

1095 Men

Own calculations. 58,012 women, 70,551 men.

1460

1

90 180

365

547 730 912 days after registration Women

1095 Men

1460

1095

61

Figure 19.

Profiles for Swiss and foreigners. Foreigner with permit C

20

20

40

40

60

60

80

80

100

100

Foreigner with permit B

1

90 180

365

547 730 days after registration

1095

1460

1

90 180

365

547 730 days after registration

1095

1460

20

40

60

80

100

Swiss

1

90 180

365

547 730 days after registration

Left w/o job Found job In course

1460

In employment prog. Registered w/o progr.

Inactive

1

90 180

365

547 730 912 days after registration

0

0

20

40

20

10

60

40

20

80

60

30

Employed

100

Registered

1095

1

1095

90 180

365

547 730 912 days after registration

1095

90 180

365

547 730 912 days after registration

In training courses

0

0

2

5

4

6

10

8

10

15

In employment programmes

1

1

90 180

365

547 730 912 days after registration Swiss Permit C

1095

1460

1

Permit B

Own calculations. 75,646 Swiss, 33,593 permit C, 16,529 permit B.

90 180

365

547 730 912 days after registration Swiss Permit C

1095

Permit B

1460

1095

62

Figure 20.

Profiles for age groups. 35-45 years

20

20

40

40

60

60

80

80

100

100

Age < 25 years

1

90 180

365

547 730 days after registration

1095

1460

1

90 180

365

547 730 days after registration

1095

1460

80

100

>=55 years

In employment prog. Registered w/o progr.

20

40

60

Left w/o job Found job In course

1

90 180

365

547 730 days after registration

1095

1460

Employed

Inactive

1

90 180

365

547 730 912 days after registration

0

0

20

10

40

20

20

60

40

30

80

40

60

100

Registered

1

1095

90 180

365

547 730 912 days after registration

1095

90 180

365

547 730 912 days after registration

1095

In training courses

0

0

2

4

5

6

8

10

10

In employment programmes

1

1

90 180

365

547 730 912 days after registration =55 years

1460

1

90 180

365

547 730 912 days after registration =55 years

Own calculations. 22,702 =55 years.

Figure 20 displays the profiles according to age group. There are distinct profiles for the youngest and the oldest age groups. Generally, the older a person is, the less likely it is that they will find a job. The highest initial inactivity rates are among younger persons. The oldest group shows the highest effect at the end of the entitlement period,

63

which is reflected in a jump in inactivity rates. Regarding programme participation, younger persons are more likely to be in employment programmes (which also include motivation semesters specifically targeted to youth) at the beginning of the spell than other groups. The oldest age group is less likely to be in training courses. Figure 21.

Profiles for skill levels. Unskilled

20

20

40

40

60

60

80

80

100

100

Skilled

1

90 180

365

547 730 days after registration

Left w/o job Found job In course

1095

1460

1

90 180

365

In employment prog. Registered w/o progr.

547 730 days after registration

Inactive

1

90 180

365

547 730 912 days after registration

1095

0

0

20

20

40

10

40

60

20

60

30

80

100

Employed

80

1460

In employment prog. Registered w/o progr.

Left w/o job Found job In course

Registered

1095

1

90 180

365

547 730 912 days after registration

1095

90 180

365

547 730 912 days after registration

1095

In training courses

0

0

2

2

4

4

6

6

8

8

10

10

In employment programmes

1

1

90 180

365

547 730 912 days after registration Skilled

1095

1460

Unskilled

1

90 180

365

547 730 912 days after registration Skilled

1095

1460

Unskilled

Own calculations. 77,945 skilled, 35,316 unskilled.

Figure 21 displays time profiles according to skill levels. Skilled persons (with at least three years higher secondary education) are more likely to find a job and less likely to leave the registry without finding a job (17.9% compared to 24.3% of low-skilled individuals after two years). The jobs of skilled persons also seem to be less seasonal than those of lower skilled persons. Lower skilled people are found more often in em-

64

ployment programmes (10.2% compared to 8.8% of higher skilled people after 11 months). Figure 22.

Profiles for employability estimation. Employability poor or very poor

20

20

40

40

60

60

80

80

100

100

Employability easy or very easy

1

90 180

365

547 730 days after registration

Left w/o job Found job In course

1095

1460

1

90 180

365

In employment prog. Registered w/o progr.

Left w/o job Found job In course

40 30 20

40

10

20 547 730 912 days after registration

1095

0

0 365

1460

Inactive

60

100 80 60 40 20

90 180

1095

In employment prog. Registered w/o progr.

Employed 80

Registered

1

547 730 days after registration

1

90 180

365

547 730 912 days after registration

1095

90 180

365

547 730 912 days after registration

1095

In training courses

0

0

2

2

4

4

6

6

8

8

10

10

In employment programmes

1

1

90 180

365

547 730 912 days after registration

High employability Medium employability

1095

1460

Low employability

1

90 180

365

547 730 912 days after registration

High employability Medium employability

1095

1460

Low employability

Own calculations. 20,987 easy/very easy, 20,797 poor/very poor.

Figure 22 displays the profiles for persons who are rated to have high versus medium versus low employability (as estimated by their caseworkers).97 Those rated easy to place found a job more quickly. However, their employment is clearly more seasonal. They are less likely to leave without finding a job, whereas a comparably high share

97

This variable has five different values: 1 (very easy) 2 (easy) 3 (medium) 4 (poor) and 5 (special case). The first two and the last two

65

(38%) of those with poor or very poor employability are inactive three years after registration. At the end of the entitlement period the tendency towards inactivity is also higher in this group. More individuals with low and medium employability are found in employment programmes than those with high employability. In contrast, those judged as having high employability are more likely to be in training courses, as well as medium employable persons. The caseworker rating seems to be representative of the actual changes of jobseekers finding employment.

5.3 Active labour market programmes This section gives an overview of participation in active labour market programmes. Including only those programmes which were attended during the first unemployment period starting in 2000, the following distribution emerges (Table 9). Table 9.

Number of programmes.

Number of programmes no programme 1 programme 2 programmes 3 programmes 4 programmes 5 programmes 6 programmes 7 programmes 8 programmes 9 programmes

Share in sample 72.9% 15.0% 6.6% 3.0% 1.4% 0.6% 0.3% 0.1% 0.1% 0.0%

Number of obs. (93,715 obs.) (19,246) (8,471) (3,849) (1,837) (827) (375) (174) (61) (14)

Around 27% of the sample participated in at least one programme, and 12% participated in two or more programmes. Examining now the variety of programmes, Table 10 lists the different courses and programmes separately. The most common programmes are basic programmes, language courses, computer courses, and employment programmes. More than one-fourth of first courses are basic programmes (column "percentage of first programmes"). Second programmes are more likely to be language and computer courses or employment programmes. Note that individuals can complete several programmes of the same type. For example, there are some individuals who completed nine subsequent computer courses, or six consecutive employment programmes.

66

Figure 23 displays the distribution of the timing of the programmes. The focus is on the beginning of the unemployment spell. More than half the programmes start within the first four months after registration, particularly in the second and third month.

Percentage of first programmes1

Percentage of second programmes2

No programme Basic programmes Personality courses Basic qualification course (low level language course) Language course Basic computer course Advanced computer course Basic commercial vocational training Advanced commercial vocational training Basic technical vocational training Advanced technical vocational training Practice firm Educational internship Basic vocational education in hotel, home, or cleaning services Basic vocational education in health and social care Other courses Self-employment courses Employment programme at collective workplaces Employment programme at single workplaces Motivation semester (youth only) Internship Settling-in allowance Education allowance Commuter allowance Allowance for weekly commuter with second household at workplace

73.0 8.7 2.8

93,832 11,148 3,574

7 5

28.1 7.0

8.9 5.4

0.7 4.9 6.5 1.3 0.7 0.3 0.5 0.1 0.8 0.2

847 6,296 8,326 1,704 866 431 652 132 1001 236

4 8 7 9 5 4 5 6 4 4

2.0 12.9 17.0 3.2 1.1 0.6 1.2 0.2 1.3 0.3

1.2 16.1 20.7 4.0 1.8 1.0 1.3 0.3 2.1 0.5

0.4

554

3

1.0

0.9

0.4 0.7 0.9

512 926 1,210

4 6 4

0.8 1.7 2.3

1.2 2.1 2.5

4.8 2.8 0.1 0.5 1.3 0.1 0.2

6,217 3,621 100 632 1,636 79 273

9 6 3 3 3 6 6

9.3 5.3 0.3 1.3 2.5 0.1 0.4

16.7 8.6 0.2 0.9 2.8 0.2 0.5

0.1

104

6

0.2

0.2

Number of persons total

Total number of observations: 128,569

Maximum number per individual

Participation in labour market programmes. Percentage of persons total

Table 10.

Own calculations. Notes: 1Observations having at least one programme: 34,854; 2 Observations with at least two programmes: 15,608.

There is considerable variance in the duration of programmes (Figure 24). The average duration of the first programme is 57 days, which corresponds to almost two months. 50% of programmes are shorter than equal to one month in duration. There is a peak at

67

180 days, which is specifically for employment programmes, which are usually assigned for half a year. Timing of programme.

0

.05

% .1

.15

.2

Figure 23.

6

12

18

24 months

30

36

42

48

Own calculations. Timing of first programme. Number of obs.: 34,854. Mean: 4.9 months. Std.dev.: 4.1. Max: 47. Percentiles: 1%: 1 5%: 1 10%: 2 25%: 2 50%: 4 75%: 6 90%: 9 95%: 13 99%: 21.

Distribution of duration of first programme.

0

.005

Density .01 .015

.02

.025

Figure 24.

60

90

180 days

365

Own calculations. Duration of first programme, cut at 365. Number of obs.: 34,854. Mean: 56.6 days. Std.dev.: 58.4. Max: 1096 Percentiles: 1%: 5 5%: 9 10%: 11 25%: 17 50%: 32 75%: 81 90%: 153 95%: 184 99%: 221.

68

5.4 Sanctions A caseworker is able to rule sanctions in case of non-compliance of a job-seeker regarding efforts in job-search, participating in labour market policy, or other forms of misbehaviour. This section presents an overview of how often these measures are used at different stages during the unemployment periods. Figure 25 displays the average imposed sanctions (only if positive) and the percentage of persons affected by sanctions. Sanctions are used more often at the beginning of a spell. Around 10.8% of the sample had sanctions imposed in the second month, whereas only 1.1% had them imposed in the twelfth month. Sanctions were also heavier if imposed at the beginning of the period. Note that sanctions at the beginning of the period are mostly are due to non-compliance with the former employer, or due to self-inflicted unemployment. Sanctions.

0

6

2

8

4

%

6

sanctions 10

8

12

10

14

Figure 25.

3

6

9

12 15 18 21 24 months after registration

av. sanctions (if >0)

Own calculations.

27

30

33

36

% unempl. with sanctions

69

5.5 Interim jobs The interim job scheme (see also section 2.4.9) requires or allows persons to take up non-suitable98 jobs temporarily or permanently while remaining registered with the unemployment office. To quantify interim jobs it is necessary to define how they are measured. This data set does not include the number of days each person spent in an interim job. However, it does include how many daily allowances were received for compensation of an interim job in one month. The total number of monthly benefits in one month is also available. This allows the calculation of how many days a particular person was unemployed, but not in an interim job. The rest of the month (counting only working days in the corresponding month) thus must have been dedicated to an interim job. Intensity = 1

Interim jobs.

5

65

10

%

intensity 70

15

75

20

Figure 26.

total benefits – benefits for interim jobs number of working days in corresp. month

3

6

9

12 15 18 21 24 months after registration

av. intensity (if >0)

27

30

33

36

% unempl. with int. job

Own calculations.

98

For definition of "non-suitability" or "non-acceptability" of job see section 2.4.7.

70

Figure 26 displays the extent to which the interim job scheme is used. At the beginning of the unemployment period, around 11% of persons hold an interim job (refer to the right scale of the graph). This figure drops slightly and increases only after one year of unemployment. Only after two years a rise is again observed. Almost 20% of those who are still registered after three years are in a subsidised interim job. These persons are entitled to a new benefit claim, and this increase might be because persons with an ongoing interim job are more likely to remain in the registry. The intensity of interim jobs increases slightly within the entitlement, signalling that more people who are still registered take up a lower paid permanent job instead of shorter casual work.

6 Summary and conclusions Historical development The history of unemployment insurance in Switzerland is characterised by increasing unemployment in the early 1990s and a recovery period in the late 1990s. The generosity of the system increased with growing unemployment and was partly reduced after the recovery period. This statement applies most obviously to benefit duration and partly to replacement rates, both of which were originally staggered, such that benefit duration increased with contribution time, and replacement rate decreased with unemployment duration. Both differentiations were repealed in the 1990s. Requirements in terms of contribution time were reduced over time until they increased in 2003. Duration itself was increased consistently, and it was reduced for the first time in 2003. In contrast, acceptability conditions for jobs were degraded over time, and individuals now have to accept lower paid jobs under the interim job scheme. In the mid 1990s, Switzerland (like many other OECD countries) emphasised the importance of activating job-seekers using active labour market programmes. This strategic re-orientation was accompanied by a change from purely administrative control to a counselling system in 1995. Design of unemployment insurance today Switzerland maintains one of the most generous unemployment insurance systems in the world. Current unemployment benefits are characterised by comparatively long durations (of around one and a half years) and a net replacement rate of 70 to 80%. It is funded to an equal degree by employees and employers. Unemployed persons have to pass a waiting period of five working days before they become eligible for benefits.

71

Job seekers have monthly meetings with caseworkers, where they have to give proof of sufficient search effort. Job offers have to be accepted if they satisfy certain minimum criteria in terms of wage, distance, and other working conditions. A person is encouraged and enforced to take up "unsuitable" jobs temporarily through the interim job scheme, in order to avoid inactivity and to minimise the cost of the insurance. Active labour market programmes are imposed or granted by the caseworkers and fully financed by the unemployment insurance if they are likely to enhance labour market prospects for the unemployed. Caseworkers can impose sanctions if the job seeker does not comply. The Swiss unemployment insurance system shares several features with an optimal unemployment insurance system:99 It offers a high level of insurance in terms of duration and level of benefits, together with the possibility of sanctions and tight monitoring by caseworkers. A variety of active labour market policies are available to address specific problems of unemployed persons. The following two parts of the thesis will evaluate to what extent active labour market programmes serve their purpose of bringing persons back to work, and to what extent the negative incentives of long benefit duration are overcome by other characteristics of the system (strong monitoring and sanctions, strong work ethics). Data for scientific research In order to facilitate scientific research, the central unemployment insurance office provides administrative records about unemployed persons from its information systems. This data provides very rich information about individual characteristics, payments and labour market programmes. This data can also be combined with data from the pension system, which not only makes it possible to track people after deregistration, but also to control for past labour market experiences. Each of the two data sources has its disadvantages, and they sometimes provide conflicting information. Information from the unemployment insurance system is especially imprecise after the unemployment spell is over, whereas the pension system data cannot distinguish between periods of job search without income and out of labour force status. The pension system data also has a lag in employment information and a considerable imprecision at the beginning and end of income spells. With certain realistic assumptions these sources can be combined to provide a very useful database.

99

Refer to the introductory part of the thesis for references about optimal unemployment insurance.

72

It is difficult to attribute individuals to a specific employment state, as multiple states are possible: persons can be at the same time registered and full-time employed, fulltime employed and receiving subsidies, and part-time employed and unemployed. The interim job scheme enhances this problem in several ways, namely by allowing all kinds of different part-time, temporary, and full-time jobs, and by requiring people to take up lower paid (and hence subsidised) jobs until the end of their entitlement period. Job acceptance at the end of entitlement period therefore cannot be readily interpreted as a consequence of negative incentives, but must partly be attributed to the interim job scheme. Unemployment profiles Almost half of the persons re-registered within four years after their registration in 2000. The average duration of the first unemployment period for those who registered at some point in time in 2000 was 233 days, with a median of 140 days. Shorter spells are more common, as most people seem to find a job on their own. The main reason for leaving without a job is not known. Typically, three years after first registration, around 22% of people are inactive, and around 60 and 65% employed. The remaining persons are still (or again) unemployed. Women are slightly more likely to be employed and also slightly more likely to be inactive three years after their first registration, whereas men are more likely to be unemployed. Women are more often observed in training courses and less often in employment programmes. Foreigners have more difficulties leaving the unemployment registry and less likely to find a job. Qualified persons have better job market chances than the unskilled. The caseworker's evaluation of the employability of a jobseeker appears to reflect corresponding job market chances. More than one-fourth of all unemployed people have participated in at least one programme. The most common are basic programmes, language courses, computer courses and employment programmes. Programmes are usually assigned at the beginning of the spell, on average after five months, the median is four months. Half of the programmes last around one month or less. More than 10% of individual have sanctions imposed within their first two months. On average, around 10% of unemployed people are in an interim job every month.

73

Appendix Table A1.

Labour force according to census. 1930 1940 1950 1960 1970 1980 1990 2000

1,942,626 1,992,487 2,155,656 2,512,411 2,995,777 3,091,694 3,621,716 3,946,988

Source: seco, Labour market statistics.

Table A2. Year 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990

Contribution rates. Contribution Max. insurSolidarity rate rate able earnings 0.8 0 46,800 Fr. 0.8 0 46,800 Fr. 0.8 0 46,800 Fr. 0.5 0 46,800 Fr. 0.5 0 46,800 Fr. 0.3 0 46,800 Fr. 0.3 0 69,900 Fr. 0.6 0 69,900 Fr. 0.4 0 69,900 Fr. 0.4 0 69,900 Fr. 0.4 0 81,600 Fr. 0.4 0 81,600 Fr. 0.4 0 81,600 Fr. 0.4 0 81,600 Fr.

Year 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Contribution Max. insurSolidarity rate rate able earnings 0.4 0 97,200 Fr. 0.4 0 97,200 Fr. 2 0 97,200 Fr. 3 0 97,200 Fr. 3 0 97,200 Fr. 3 1 97,200 Fr. 3 1 97,200 Fr. 3 1 97,200 Fr. 3 1 97,200 Fr. 3 2 106,800 Fr. 3 2 106,800 Fr. 3 2 106,800 Fr. 2.5 1 106,800 Fr. 2 0 106,800 Fr.

Solidarity contribution rate is applicable from max. insurable earnings up to 2.5 times that amount. Figures for 1977 until 1983 from Statistics of the social insurances (Schweizerische Sozialversicherungsstatistik) 2003.

74

Table A3. Year 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Maximum benefit entitlement duration.

6 months contribu- 12 months contribu- 18 months contribution time tion tion 85 170 250 85 170 250 85 170 250 85 170 250 85 170 250 85 170 250 85 170 250 85 170 250 85 170 250 170 250 400 170 250 400 170 250 400 170 250 400 520 520 520 520 520 520 520 520 520 520 520 520 520 520 520 520 520 520 0 400 400 0 400 400

>55 years, 18 months contr. 250 250 250 250 250 250 250 250 300 400 400 400 400 520 520 520 520 520 520 520 520

Over 55 years, less contribution 250 250 250 250 250 250 250 250 300 400 400 400 400 520 520 520 520 520 520 400 400

From 1984 to 1996, for economically disadvantaged regions different durations for entitlement and contributions were applicable sometimes.

Table A4. Year Revenues 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984

203 612 407 599 626 474 498 365 362 682

Finances of the UI. Expendi- Balance Total Account tures (deficit, gain) 242 -39 567 586 26 548 141 266 464 209 390 855 210 416 1271 153 320 1592 155 343 1935 428 -63 1871 800 -438 1438 779 -97 1341

– table to be continued –

Year Revenues 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

976 786 866 804 3556 3680 5488 5955 5745 5876

Expendi- Balance Total Account tures (deficit, gain) 442 534 2640 502 284 2924 1340 -474 2450 3461 -2657 -207 5986 -2430 -2637 5921 -2241 -4878 5240 247 -4631 6124 -168 -4799 8028 -2283 -7082 6208 -333 -7415

75

– Table A4 continued – Year Revenues 1985 1986 1987 1988

744 782 841 906

Expendi- Balance Total Account tures (deficit, gain) 698 45 1379 617 165 1544 636 206 1749 550 357 2106

Year Revenues 1999 2000 2001 2002

6378 6646 6852 6969

Expendi- Balance Total Account tures (deficit, gain) 5056 1323 -6093 3711 2935 -3157 3415 3437 279 4966 2003 2283

Source: Statistical Yearbook and Statistics of the social insurance (Schweizerische Sozialversicherungsstatistik) 2003. 1998 is based on figures for January-November only and therefore not directly comparable with others. 2002 is provisional.

Table A5.

Expenditures for ALMP.

% of total Expenditures Year expenditures for ALMP of the UI 1984 4 0.56 1985 14 1.94 1986 17 2.75 1987 17 2.71 1988 17 3.04 1989 15 3.37 1990 17 3.34

% of total exExpenditures Year penditures of for ALMP the UI 1991 22 1.63 1992 55 1.57 1993 137 2.28 1994 322 5.44 1995 496 9.47 1996 405 6.61 1997 815 10.16

% of total Expenditures Year expenditures for ALMP of the UI 1998 618 9.96 1999 570 11.28 2000 361 9.73 2001 323 9.46 2002 424 8.54

Source: Statistical Yearbook and Statistics of the social insurance (Schweizerische Sozialversicherungsstatistik) 2003. 1998 is based on figures for January-November only and therefore not directly comparable with others. 2002 provisional.

Table A6. Year 1985 1986 1987 1988 1989 1990

Benefit exhaustions.

Benefit exhaustions 11,850 8,145 7,545 6,719 4,977 4,492

Source: seco.

Percent of unemployed 12.23 9.95 9.98 9.88 9.05 7.68

Year 1991 1992 1993 1994 1995 1996

Benefit exhaustions 9,433 22,401 22,496 42,424 36,802 31,982

Percent of unemployed 8.27 10.26 7.40 13.48 12.45 9.84

Year 1997 1998 1999 2000 2001 2002

Benefit exhaustions 23,703 42,209 34,124 17,129 13,226 14,511

Percent of unemployed 6.70 13.20 13.21 8.27 6.80 5.75

76

Table A7.

Minimum places for active labour market programmes per canton.

Kanton TOTAL Zurich Berne Lucerne Uri Schwyz Obwalden Nidwalden Glarus Zug Fribourg Solothurn Basle-Town Basle-Country Schaffhausen Appenzell Outer Rhodes Appenzell Inner-Rhodes St. Gall Grisons Argovia Thurgovia Ticino Vaud Valais Neuchâtel Geneva Jura

1997 25,000 4,258 2,947 1,000 64 342 60 76 111 283 841 743 712 774 249 117 15 1,311 369 1,629 656 1,514 2,833 1,246 715 1,875 260

1998 25,000 4,325 2,966 1,040 89 370 75 90 119 288 805 773 685 758 242 142 28 1,370 478 1,697 694 1,445 2,669 1,194 652 1,750 256

1999 25,000 4,428 2,968 1,031 88 367 71 78 114 282 794 792 682 751 245 138 31 1,392 489 1,744 698 1,438 2,571 1,177 639 1,747 245

2000 15,000 2,694 1,768 608 52 221 44 53 69 171 480 475 398 451 146 82 17 839 305 1,024 417 871 1,490 715 398 1,068 144

Source: Figures for 1997: AS 1996 3071, 1998: AS 1997 2415, 1999: AS 1998 1670, 2000: AS 1999 3614.

77

Part II: Effectiveness of Active Labour Market Programmes

78

1 Introduction This part of the thesis contributes to the evaluation of active labour market programmes in two ways. First, Swiss programmes during the years 2000 and 2001 are evaluated with regard to a variety of outcomes such as unemployment, employment, earnings, and inactivity. Before this research no other study had examined labour market programmes for that period of time. Additionally, this section covers a longer after-programme period than earlier studies, together with a variety of outcomes. Hence, the findings of the current research provide a more comprehensive look at how labour market programmes affect participants. Second, this research provides a comparison of two estimation strategies, one of them static, the other controlling for dynamic assignment of the first treatment. This examination reveals that the two estimators lead to different results for some programmes. Use of the term "effectiveness" in the evaluation of active labour market programmes requires a definition of the target variable, in other words, the outcome by which effectiveness is measured. Earlier studies by Gerfin and Lechner (2002) and Lalive, van Ours and Zweimüller (2002) used the outcome "de-registered from unemployment" as a measure of efficiency. The current research aims to avoid restricting this term to either a single indicator or some composite welfare measure, and thus instead uses a variety of separate outcomes, analysed separately as well as grouped. Outcome variables were attained using a combination of the pension system data and unemployment insurance records. This had partly been undertaken previously by Gerfin, Lechner and Steiger (2004), but the focus of that study was on employment subsidies only, and did not achieve an in-depth analysis of the other programmes. Moreover, the current study covers a more recent time period, and thus the results are not directly comparable with this earlier work. The findings of this part of the thesis will not only provide insight into how the employment status of individuals evolves within and after periods of unemployment, but also examines effects of earnings and identifies indicators for dropping out of the labour force and/or possible dependency on social assistance. Effectiveness in the present study is measured using a micro perspective, such that macro effects are ruled out.1 The focus of this research is the effect of programmes by comparing individuals who did and did not participate in that programme. First, only

1

Stable Unit Treatment Value Assumption (SUTVA), introduced by Rubin (1980).

79

effects for those who actually participated in programmes are estimated (treatment effects on the treated), and not effects for persons who did not. Second, the natural comparison state "nonparticipation" is used.2 A comparison between nonparticipation and participation is often purported to be the most interesting policy question in terms of active labour market programmes, because it directly addresses the question of whether public spending is effective at all, since all programmes are costly. Efficient spending would require that there is a return on active labour market programmes.3 Estimation will be undertaken by propensity score matching. This method was chosen because we have an unprecedentedly large sample at hand, with high quality information from two different data sources, the unemployment insurance registry and the social security system. The matching method requires a strong assumption about the comprehensiveness of data for the underlying problem, which is likely to hold in this setup. Given that, we give preference to less parameterisation as compared to other methodologies. We apply two methodologies. The first and more static method is equivalent to the one applied by Gerfin and Lechner (2002), who defined nonparticipants as those who never participated in any programme throughout the period of their unemployment. The timing of the programme is not explicitly considered, but pretreatment durations are matched by a trimming method. Recent work by Fredriksson and Johansson (2004) has raised doubts about the validity of that method; they argue that it can lead to negatively biased programme effects. These authors suggest another estimation method using a time-varying treatment indicator. This paper uses as a second estimator a very similar method to the one applied by Sianesi (2004) for evaluating active labour market programmes in Sweden.4 The results of the two methodologies (static and dynamic) are compared. This paper is structured in the following way. Section 2 gives an overview of active labour market policy in Switzerland. Section 3 explains the two estimation strategies based on different definitions of nonparticipation. Section 4 describes the data and how it is sampled for this study. Section 5 presents and discusses the results of the estimation, and section 6 provides a conclusion for this part of the thesis.

2

Gerfin and Lechner (2002) focused on mutual comparisons between programmes.

3

Measuring a "return" on programmes would incorporate gains as well as costs of programmes. However, costs are not addressed here.

4

Swedish active labour market programmes are set up very similarly to those in Switzerland.

80

2 Swiss active labour market policy and unemployment insurance 2.1 Design Traditionally, Switzerland has experienced very low unemployment, which facilitates high public spending on the unemployed, combined with a relatively low contribution rate. Passive unemployment benefits pay between 70 and 80% of an individual’s former salary, conditional on the level of earnings and dependants.5 The maximum duration of benefits in the years 2000 to 2002 was two years, provided that a person had a minimum contribution time of six months. After expiration of the benefits, cantonal unemployment aid and/or social assistance on the communal level step in.6 Counselling and placement services are available to every job seeker in Switzerland, regardless of his or her employment status. Unemployed persons are entitled to more intensive programmes such as personality, computer, and language courses, or other vocational training, in order to refresh and improve their professional skills. For work experience, different kinds of employment subsidies or specifically designed temporary jobs are available. The duration of these programmes varies substantially: fulltime courses may not last longer than two months, whereas employment programmes normally are assigned for time periods as long as six months. The administrative procedure for a job seeker is as follows. First, an unemployed person registers with the corresponding regional unemployment office. A caseworker adopts the case and meets the unemployed individual on a regular basis to evaluate his job seeking endeavours and help him to chart a roadmap to sustained employment. Meetings between the unemployed person and their caseworker take place at least once a month. Caseworkers undertake counselling and decide participation in active labour market programmes. Once decreed by the officer, a programme is compulsory, such that non-attendance by the unemployed person is sanctioned by cancellation of benefits. A participant in an active labour market programme usually has to continue

5

This is the gross replacement rate. The net replacement rates after taxes are even higher (see OECD 2002).

6

Note that social assistance payments in Switzerland are means-tested. In some cantons they are similar to a loan and might have to be paid back later. In the past, this sometimes led to dependency spirals.

81

his job search during the programme, but can be exempt from it if the programme is considered too time-consuming.

2.2 Past findings about the effectiveness of Swiss active labour market policy Earlier research into the effectiveness of active labour market policy in Switzerland was conducted in the course of an evaluation programme of the State Secretariat for Economic Affairs (seco) between 1998 and 2000. Gerfin and Lechner (2002) found only nonparticipation and subsidised interim jobs7 to be superior programmes in terms of the number of people leaving unemployment and finding a job. Gerfin, Lechner and Steiger (2004) used updated data of the same sample to have a closer look at the effects of different kinds of temporary subsidised employment in the form of employment programmes and subsidised interim jobs. They found that, with better outcomes that also accounted for earnings effects, interim jobs were superior for almost all possible subgroups of individuals. Thus the superiority of nonparticipation compared to employment programmes was challenged, although this research did not allow a clear picture to be drawn. A similar finding was reported by Vassiliev (2003) for the canton of Geneva, and also by Lalive, van Ours and Zweimüller (2002) for the whole of Switzerland. Curti (2002) found positive effects on an aggregate programme level. However, none of these studies examined programmes during or after 2000.

7

A job is taken up interim in order to avoid joblessness, though the job is not "suitable" in the sense of the unemployment insurance act (i.e., more than two hours away from home, pays less than actual unemployment benefits, has other unacceptable working conditions for the person – e.g., work on call) or temporary only. The unemployment insurance remunerates a part of the difference between insured salary and the income from an interim job. Interim jobs or temporary subsidised jobs are not part of the official active labour market programmes. In fact, they cannot readily be classified as a programme, but would be placed somewhere between a programme and an outcome. Interim jobs are an outcome in the sense that they take place in the regular labour market. They cannot be supplied by the government, but the legislation forces the unemployed to take up lower paid jobs under this scheme. It is not rare that persons take up a job with a lower salary without even knowing that they are entitled to a compensation of the income difference to their last income, as long as their entitlement period is not expired (the employment officer then usually points out this possibility). Most of these programmes are found by the unemployed themselves. Nevertheless, it can also be interpreted as treatment, because jobs under this scheme are not necessarily intended to be permanent, but more to keep persons attached to the labour market. Additionally, in some cases they are arranged by the unemployment officer. There is also a considerable amount of heterogeneity regarding temporary and permanent employment. For these reasons, interim jobs are considered neither as programmes nor as outcomes in this study, but are taken into account later in the definition of nonparticipation.

82

2.3 Assignment to programmes and nonparticipation Why do some unemployed individuals participate in programmes, but others do not? Nonparticipation in the Swiss context arises if the unemployed person has not been assigned to a programme by the caseworker. It is also possible that the unemployed person intended to participate in a programme, but his request was not granted. A person may also not participate in a programme because she or he is already in an interim job. Finally, finding a job earlier than possible programme participation also leads to nonparticipation. The policy for assignment of unemployed people into programmes is not standardised in Switzerland; cantons have a considerable amount of autonomy in executing the federal law.8 Assignment of persons into programmes also varies between employment offices and officers themselves. A person might not be assigned to a programme by the caseworker because there is no programme available (limited supply), because the caseworker does not see any need for training, or because the officer wants to give the person more time to search for a job instead of keeping them busy with programme participation. An unemployed person can themselves apply for programmes. The request has to be granted by the caseworker or – in the case of expensive programmes – by the cantonal authority. The programme by law has to increase the unemployed person's employability in order to be granted. Furthermore, participation in programmes is not directly related to a point of time within the total unemployment period. There is no rule which states that after a certain point of time a person is obliged to participate in a specific programme (or indeed the opposite, that participation is no longer possible). The federal authority gives a general recommendation to assign people into programmes early in their spell. However, this recommendation is not binding and often is not followed. Therefore, if a person has not participated in any programme, this might be more a combination of dynamic decisions at a series of meetings with the caseworker, rather than a predefined, intentional strategy. As meetings with caseworkers normally take place at least once in a month, a monthly decision making process is also an appropriate basis for evaluation purposes. 8

These differences are used as an instrumental variable for estimating treatment effects by Frölich and Lechner (2004).

83

Employment officers report that if an unemployed person finds a job, they are very unlikely to start that job within the next week or month. Typically, a job is taken up a few weeks later only, around one to two months. A person remains in the registry in the meantime, possibly looking for a temporary opportunity in form of a so-called interim job scheme. However, the data does not include the time at which a person signs a contract, but only the date when he or she de-registers. As an example, a person might not begin in a programme in May because they have already found a job starting at the beginning of July.

3 Econometrics 3.1 Identification and estimation of treatment effects The definition of an individual causal effect of a treatment9 is the difference between what a person achieves after receiving the treatment as compared with what they would have achieved in the absence of it. The average effect is thus the mean of all those individual effects if macro effects are ruled out (stable unit treatment value assumption, SUTVA10). M programmes are available. By denoting the outcome (achievement) as Y, M+1 different possible states for a person can be defined: Y 0 is the state in the absence of any treatment, Y m in the presence of treatment m, m

{1,...M } .

This concept is called the potential outcome framework.11 Since outcomes are usually measured at various timepoints after programme participation, a time index t is used. The causal treatment effect at time t after the programme begins is the difference between the two potential outcomes in the participation and nonparticipation states, respectively, Yt m Yt 0 .12 This research specifically asks how the programme affected those who actually participated in it, compared to a state where they would not have

9

Active labour market programmes are referred to as “treatments” in this literature (following the biostatistical literature). 10

SUTVA was introduced by Rubin (1980).

11

The potential outcome framework is attributed to Neyman (1921), Roy (1951) and Rubin (1974).

12

The treatment of interest is defined as the first treatment observable during that period. All future outcomes are assumed to be effects of the first treatment. Miquel and Lechner (2005) provide identification results when outcomes of sequences of treatments are of interest.

84

participated in any programme (treatment effect on the treated). The participation indicator is denoted as D. The estimand of interest therefore is: m t

= E Yt m Yt 0 D = m .

This expression can be written as: m t

= E Yt m D = m

E Yt 0 D = m

People can be observed with a specific programme or without any programme, such that the specific outcome for that individual is known after actual programme participation. The outcome at time t for participants can thus be identified ( E Yt m D = m ), but not in the corresponding counterfactual nonparticipation state ( E Yt 0 D = m ). The potential outcome after not participating at all is unknown. However, the outcome of persons who did not participate can be observed ( E Yt 0 D = 0 ). Taking the difference between outcomes of participants and nonparticipants generates: E Yt m D = m

E Yt 0 D = 0 = E Yt m D = m E Yt 0 D = m + E Yt 0 D = m E Yt 0 D = 0 1444442444443 144444244444 3 m t

selection bias

The first part of this expression is the parameter of interest, the second part the selection bias (i.e., the difference between their (hypothetical or actual) outcomes after nonparticipation). In non-experimental studies this bias is not expected to be zero, such that persons who actually participated in programme m might have a different expected outcome after nonparticipation than those who actually did not participate. In other words, people might have been selected into a programme according to their expected profits from completion. If this selection bias was zero, the average outcomes in both participant groups of programmes 0 and m could be calculated, generating the causal effect. This is – under further assumptions – the case in experimental studies. However, experimental data is not available. For the identification of treatment effects from non-experimental data, in principle two strategies are possible. The first strategy is to assume that all relevant information about the connection between programme selection and outcome is given. This is called “selection on observables” or “conditional independence assumption” (CIA). If this assumption is not credible with the data at hand, other strategies based on “selection on unobservables” have to be applied

85

(instrument variable, difference-in-difference, distributional assumptions of selection bias components, etc.).13 The CIA necessary for identification of the treatment effect on the treated states that, given all known characteristics X that jointly influence participation decision and outcome after participation, the (hypothetical) outcome after nonparticipation in a programme is independent of the actual participation D (here this assumption is only for pair wise comparisons): Yt 0 C D X = x, D {0, m}

(1)

For the matching estimator a second assumption is required, namely that everybody in a programme has a positive probability not to be in any programme: P ( D = 0 D = m, X = x ) > 0

m = 1,…,M.

With this additional assumption, the following can be identified: E Yt m D = m, X = x = E Yt 0 D = 0, X = x

By integrating over x: E Yt 0 D = 0 = E Yt 0 D = 0 , X = x f ( x D = m ) dx x

The average treatment effect of m on those who participated in m compared to those participating in 0 then is: m t

E Yt 0 D = 0, X = x f ( x D = m ) dx

= E Yt m D = m x

E Yt m D = m is estimated by the sample mean outcome of all participants in treatment

m. The second part is slightly more difficult to estimate, because X is multidimensional. One simple way to do this is to match characteristics of nonparticipants to those of participants and only consider nonparticipants who are very "similar" (in terms of all relevant X characteristics) to those who actually participated in programme m. To reduce the dimension of X, the balancing score property of the propensity score is used (Rosenbaum and Rubin, 1984), which reduces the dimension of the X to a single score, the probability for nonparticipation:

13

For an overview of strategies see Heckman, LaLonde & Smith (1999).

86

Yt 0 C D p ( D = 0 X , D {0, m}) .

"Similarity" then is measured in terms of the Mahalanobis distance14, based on the propensity score plus some further X variables (see below for details).

3.2 Static method The static method is equal to that used in Gerfin and Lechner (2002). It defines participants as those who participated in the programme as the first programme in their total unemployment period, and nonparticipants as those who did not participate in any programme. Since after-programme effects are the subject of this evaluation, an artificial starting date for nonparticipants can be constructed. It is drawn for every person without treatment throughout their unemployment period from the distribution of starting dates of all participants, given the characteristics X. Whosoever is already deregistered by that time or shortly after is deleted from the sample. Individuals are matched subject to personal characteristics as well as the starting date of the programme. The CIA (1) must hold over all time periods. Fredriksson and Johansson (2004) argue that this can lead to negatively biased effects of treatments compared to nonparticipation, because the conditional independence assumption might be violated, given that the treatment indicator itself is defined based on future outcomes. This problem arises because in Switzerland every person is, in principle, entitled to a programme, and it can take place at any time during the period of unemployment. Nonparticipation simply could mean that the person found a job before participation in a programme was offered or considered.

3.3 Dynamic assignment of first treatment Other definitions avoiding the problem in the static setup have been applied. Sianesi (2004) and Brodaty, Crépon and Fougère (2001) have designed empirical approaches which define nonparticipation more dynamically:15 for every point of time, a group of nonparticipants is defined as “waiters” (those who remain in open unemployment during a certain time period instead of entering a programme). Moreover, all of these 14

Mahalanobis distance between two points X 1 and X 2 drawn from same distribution:

(X

1

X 2 ) SX

1

(X

1

X 2 ) with S X covariance matrix of the distribution of the X.

87

"nonparticipants" have not yet participated in any programme during their relevant unemployment period. Nonparticipation is not treated as a programme, but as a “default” state with more intensive job search. In this sense, everybody is a nonparticipant until she or he enters a programme or finds a job. This approach resembles the hazard model with multiple risks, such that at any time after registration a person decides whether to participate in any programme m or to wait. All persons who haven’t participated until time s and who are still unemployed are “waiters”. The participation indicator D therefore receives a (discrete) time index. Ds denotes participation at time s. Note that persons can participate in a programme at one time only, and in this analysis only the first participation is considered, such that: M

D=

S

m =1 s =1

m 1( Ds = m )

The CIA that is required for identification of a treatment effect is: Yt 0 C D X = x, D1 = 0,...Ds 1 = 0 , still unemployed at time s

This CIA states that how the (hypothetical) outcome evolves after not participating up to time s is independent of programme participation at time s, given observed characteristics X. In other words, this analysis can observe and control for all information that influences the participation decision at a certain point of time s and the afterprogramme outcome simultaneously, given that a person has not participated before. The treatment effect of having participated in programme m at time s instead of waiting at least one period longer, for those who participated, is (note that the outcome is measured at different timepoints t after programme participation): m ,s t

= E Yt m Ds = m

E Yt 0 Ds = m .

All single-time treatment effects can be combined into a composite treatment effect by adding up: m t

=

S s =1

m,s t

P ( Ds = m D = m ) .

Note that meaning of “causality” in this context is not straightforward.

m t

is the over-

all treatment effect on those who participated in programme m at any time in their un-

15

A very similar method is "balanced risk set matching" (Li, Propert and Rosenbaum, 2001).

88

employment spell instead of waiting longer, which might include another programme participation later in their spell (including the same programme). While the effects for single time periods have a causal interpretation, the composite effect does not. Fredriksson and Johansson (2004) conclude that the effect, as measured in this way, can be used for testing if a treatment was effective or not.

3.4 Matching and estimation procedure The matching procedure can be based on one nearest-neighbour only, or on a weighted average of several similar persons. For every person in the sample of participants in treatment m one or several very similar persons in treatment 0 are found and the (weighted average) outcome is calculated. “Similarity” is conceptualised in terms of the participation probabilities as well as other variables. A comprehensive overview of matching estimators is provided by Frölich (2004). In this study, nearest-neighbour-matching is chosen, in the context of the two different definitions of nonparticipation. Participation and outcomes are measured on a monthly basis. In the dynamic method, for every programme only those months with at least 90 persons starting the programme are taken into account. This number is a compromise between not losing too many participants and obtaining reasonable numbers of observations for every month. Another solution would have been aggregation either on the basis of treatment or time. Neither of these is a satisfying strategy. Aggregating treatments would lead to a loss of information about the variety of programmes. The second strategy would require aggregation of treatments as well as outcomes over several months, which seems to be a fairly high information loss, and could suffer from the same potential problem of validity of the CIA. Another solution is matching on the Mahalanobis distance only (see Rubin, 1980). However, results of two definitions of nonparticipation are comparable only if they use the same metric. Therefore, in both methods the Mahalanobis distance is chosen, with the propensity score as one component of it. It is based on the propensity score, age, gender, calendar month of programme, and the static method duration of unemployment at beginning of programme. 1. Propensity scores M binary probits are used for estimation of:

(

)

Pˆ D( s ) = 0 X , D( s ) {0, m}

m = 1, … M

89

For the Sianesi approach, these propensity scores are computed based on participants and nonparticipants in every month (dependent variable "no participation until end of corresponding month"). 2. Common support The common support criterion is applied in order to ensure treated persons are comparable to the non-treated group.16 For every programme (and in the month-wise estimation for every month), this criterion eliminates all observations that have propensity scores smaller than the minimum of the nonparticipant group. 3. Matching procedure The matching procedure is nearest-neighbour-matching with replacement. For this procedure within a treatment group m, for every observation (person) i the most similar person j from the nonparticipants pool is chosen. The person with the minimum Mahalanobis distance to person i from the pool of participants in treatment 0 is defined as their “counterfactual”. A person in 0 might be the closest match for more than one observation in programme m. w mj counts the numbers of times a person in the nonparticipants pool is used as the closest match for a person in pool m. The mean counterfactual outcome at time t after the start of the programme for participants in treatment m is computed as: 1 Yˆt cf ,m = Nm

wmj yt , j j I0

The mean outcome for participants in treatment m is: 1 Yˆt m = Nm

yi ,t i Im

The treatment effect is calculated as: m t

= Yˆt m Yˆt cf ,m

Simple standard errors of the ATET, ignoring the variance of the weights, are computed following Lechner (2002a) as the square root of:

( )

1 1 ˆ ˆtm = ˆ Yt + 2 Var Var 2 i I m Nm Nm

16

( w ) Varˆ Y . m 2 j

j I0

j I0

t

Some discussion of the common support criterion and how it affects estimates is provided by Lechner (2001).

90

This matching procedure in principle is independent of the outcome measured. However, in this framework there is a problem. For persons with an early programme start date (e.g., January 2000), this outcome is observed for a very long time. For persons with either a late programme start or a late registration date only a short time interval after programme start is observed, and later outcomes are missing. The following solution is thus chosen, namely that persons here are matched only once. For later time periods the counterfactual outcome exists only if the counterfactual person also still has an outcome. The variance takes into account only the observations with positive weights. The estimator %tm for "waiting" is the following: %m = t

S s =1

ˆ m , s Pˆ ( D = m D = m ) t s

with Pˆ ( Ds = m D = m ) =

N i =1

1( Ds = m )

N i =1

1( D = m ) and ˆtm ,s the matching estimator for

the treatment effect on the treated of starting a programme m at time t compared to not starting a programme until and by the end of month t, as described above. The aggregate effect sums up over all persons having already started their programme before t. Pˆ ( Ds = m D = m ) is the share of participants in programme m starting their treatment at

time t. The variance of single time effects is computed equally as above:

( )

ˆ ˆtm , s = Var

1 N m,s

ˆ Yt + Var

2 i I m ,s

1 N m,s

2

( w ) Varˆ Y m,s 2 j

j I0,s

j I 0,s

t

.

The variance of the composite effect has to take into account the covariances between effects of different starting dates, arising through the same persons being part of the "nontreated" group in different time periods, or first "nontreated" and then "treated".

91

( )

ˆ %tm = Var +2

S

s 1

s =1 r =1

S s =1

( ) ( Pˆ ( D

ˆ ˆtm , s Var i Im

s

= m D = m)

)

2

Pˆ ( Ds = m D = m ) Pˆ ( Dr = m D = m )

1 N m, s N m,r

j I 0,s

wmj , s 1( j

1 N m, s N m ,r

j I 0,s i I 0,r

I m , s ) Cov (Yt , s , Yt ,r ) j I 0,s

wmj ,r wmj , s Cov (Yt , s , Yt ,r ) j I 0,s

4. Validity of CIA CIA is a very strong assumption that states that all information that influences the participation decision and the outcome simultaneously is observed and controlled for. Nullity would mean that there are unobservable differences between the treated and untreated group that are not controlled for, but that jointly influence participation and outcome. This is in the case that the effects do not have any causal interpretation, but are a composite between the selection bias and the true causal effect. In the static setup, the CIA might be violated, despite the randomised assignment of programme starts. For the dynamic setup, scepticism about the validity of the CIA might emerge. Nevertheless, the comprehensive data (that among other variables also contains earnings profiles of the past 10 years as well as caseworkers’ subjective rating of the unemployed person’s employability) should support this case. With these variables a lot of generally unobserved but important heterogeneity should be accounted for.

3.5 Comparison of estimators The two estimators that are applied and compared in this section differ in several respects, first of all in the definition of a "nonparticipant". The "static method" assumes that the CIA holds true over all time periods. In reality it implies that the decision about the programme (including nonparticipation) as well as the optimal timing for it is decided by the caseworker at the beginning of the unemployment spell, based on entire knowledge at hand about the concerned person. An unemployed without a programme in his unemployment spell is observed in two cases: one, the initial decision not to participate at all (type 1), and two, the unemployment spell ends before the programme could start (type 2). The trimming procedure used for programme starting dates for nonparticipants takes into account particularly the second case, by assigning

92

a hypothetical starting date to every non-participant and deleting those observations with a starting date later than their observed deregistration. The problem faced here is that the distinction between the two types is impossible. The trimming method automatically deletes type 1 observations, particularly those with long unemployment spells, leaving only “better” cases. "Dynamic assignment of first treatment" allows the decision about participation in a programme be sequential. The CIA is weaker, as its validity is required only month by month. Programme participation is decided on a month-by-month basis, given that the jobseeker has not left unemployment. Nonparticipation is not defined over the total unemployment spell, but month by month only, leaving room for changes in opinion over the course of time, accounting for possible change in information. The interpretation of this is also the weakness of this method, as people who are earlier classified as non-participants might be later participants in a programme, and thus the effect can only be interpreted as “waiting effect”. An estimated positive effect therefore can occur due to negative effects of later treatments, and vice versa, whereas the overall nonparticipation does not contain any later programme. Depending on assumptions about treatment assignment and outcomes, it is not clear which effect is more relevant. An employment officer who assigns treatments to the unemployed will be interested in the effect of a programme compared to no programme at the moment. A policy maker interested in pure cost-benefit analysis might be interested in the overall participation effect. Another aspect is the properties of the estimation procedure. One problem is the assignment of randomised programme starting dates for nonparticipants. Usually, the overall distribution of programme starting dates has a lot of variation. Nonparticipants are dropped from the analysis if they are already employed at their hypothetical starting date. This leads to earlier starting dates for nonparticipants on average. Chances to find a job earlier in the unemployment spell are usually higher (e.g., Djurdjevic, 2003, studying Switzerland). Together with the varying ability of the estimator to match programme starting dates of treated and untreated exactly, this could be one cause for biased effects. The programme starting date is only one variable among others to be matched on.

93

4 Data and sampling 4.1 Data The sample is an inflow of all people who registered at an unemployment office in Switzerland between the 1st of January 2000 and the 30th of June 2001. The data is a combination of two information systems, namely the unemployment insurance system (placement/statistics and payments) and pension system records. The different data sets can be combined through use of a unique personal identification number. The following information is available in the data. Placement and labour market statistics database: Monthly information about each individual’s placement office, personal characteristics, characteristics of previous job and desired job, status of job seeking, work experience, skills and caseworker’s rating of chances to find a job, duration of unemployment spell, participation in active labour market programmes/job introduction allowances.17 Information both at the beginning of the spell as well as monthly information for the definition of unemployment status is used in the present study. Unemployment insurance payment system: Total benefits and other supplementary allowances paid by the unemployment insurance system, including sanctions, interim jobs, and benefit exhaustion (on a monthly basis). Information at the beginning of the spell is utilised, as well as monthly information for the definition of interim jobs and unemployment status. Pension system: All payments paid to the pension system by an employer, by the person for themselves, or by the unemployment insurance system (the source can be identified). Note that social assistance payments are not stated in the pension system. The data incorporates yearly information indicating the months of the beginning and ending of each income period (maximum January and December). This data is available from 1990 until 2002 for this study. Data obtained prior to the beginning of the period is used for the construction of unemployment and employment history on a yearly basis, thereafter for monthly definition of outcomes.

17

Though the data is very rich and contains a large amount of information, it needs to be mentioned that information about education is available only from 2002 onwards and is therefore not used in this study. Former income and general qualification variables serve as proxies.

94

In order to maintain protection of privacy, access to all of these databases is restricted, such that they are not publicly available. For a more in-depth description the reader is referred to Gast, Lechner & Steiger (2004).

4.2 Sample selection Data is available describing all persons registering at a placement office in Switzerland between the beginning of 2000 and mid-2001. The basic sample consists of all persons with at least one registration during that period, and for whom information about entitlement, personal characteristics and the pension system is available; it consists of 191,263 persons. However the present study applied further selection criteria. Persons below the age of 25 or above 55 were not included in these analyses, because other opportunities such as basic education and early retirement schemes might be available for people in these age groups. Furthermore, individuals were chosen who were not affiliated with disability insurance, since disability insurance provides its own programmes not visible in this data set. Foreigners without at least a yearly permit were also excluded, as well as home workers, trainees and students (see Appendix B for detailed selection criteria). Unemployment periods that did not overlap at least two consecutive calendar months were deleted. Only persons who had been employed at least once within ten years prior to registration were considered. Those few persons who participate in another programme not considered here (settling-in allowance, motivation semester, commuter allowance, basic education allowance) were also removed from the sample. For programme participants, another criterion was applied (that was be applied for corresponding nonparticipants as well), namely that they are not deregistered during the next calendar month following the start of the programme. In most cases, this must have implied that a person found the job before the programme already and thus employment is not likely to be an effect of the programme itself (see paragraph 2.3 for more about this point). Finally, using only those cases without missing data in important characteristics, 115,943 persons remain in the sample (for exact numbers of observation loss at each stage, see Table A11).

4.3 Outcomes This paper uses eight different outcomes. They are computed using only one of the above mentioned data sources (a-d pension system, h UI data) or a combination of the two (e,f,g).

95

a employment: A person has a positive entry if they have a salary coming from employment in the pension system for the corresponding month. During the period considered here, income during employment programmes and interim jobs can also lead to a positive entry for employment. b unemployment: A person has a positive entry if they receive an unemployment benefits payment in the pension system for the corresponding month. The unemployment benefits payment can be zero during an employment programme. c Inactivity: A person has no entry in the pension system for the corresponding month. This is interpreted as inactivity. Inactivity can also be related to dependency on social assistance, if that individual has no other means for subsistence. Note that once a person has exhausted benefits without finding a job, they will also be placed in this category. d Total income: Total income in corresponding month according to the pension system. e Stable employment: A person is employed according to the pension system data in an employment spell that lasts at least three consecutive months. At that the same time the individual does not receive any payments from the unemployment insurance system. f Stable employment without loss of income: A person is employed according to the same criteria as in outcome e (at least three consecutive months, no benefits during the same period). Additionally, the earnings are at least 85% of their last income, which is the basis for the computation of benefits. g Sustainable employment: A person is employed according to the same criteria as in outcome e (at least the consecutive months, no benefits during the same period). Additionally, the earnings are “sustainable” in the sense that they exceed a minimal income criteria that is computed according to the guidelines of the social welfare (basic needs, additional needs, health insurance, flat rent) on the basis of the number of persons in the same household and the area in which the person lives.18

18

An "equivalence income" is computed as the sum of recommended basic living allowances, health insurance costs, and costs for apartment rental, conditional on the number of persons living in the same household. Additional persons in the same household are assumed to be children below 16, except for the person themselves and a partner in case the person has civil status "married". Data sources: Schweizerische Konferenz für Sozialhilfe (www.skos.ch/deutsch/skos_richtlinien/index.html)

96

h Job-seeking: A person is registered at the regional employment office. All outcomes are measured in months after programme start (max. 30 months) as well as after registration (max. 33 months).

4.4 Programmes In line with earlier evaluation work for Switzerland, the first significant programme after first registration since 2000 and before the end of 2001 is considered (i.e., the first programme that has a duration of at least one week). Shorter programmes are not taken into account in these analyses. All later programmes are assumed to be part of the outcome of the first programme. The programme code in the unemployment insurance data differentiates between more than 30 programmes. Here they are aggregated to eight programme groups as follows (this aggregation is similar to the one used by Frölich, Lechner and Steiger, 2003): personality and other basic courses (1), language courses (2), basic computer courses (3), higher vocational training (4), lower vocational training (5), other training (6), employment programmes in single workplaces (7), employment programmes in collective workplaces (8). Another large group of programmes, basic programmes, are not taken into account in this conceptualisation because they are often followed by another programme and thus their effects should be assigned to the predominant programme. Interim jobs are also not classified as a programme in this study. 1 Personality and other basic courses: Courses which help persons to position themselves and their needs in the labour market. Additionally, there are courses for basic qualifications (3,405 observations). 2 Language courses: Either courses in a foreign language for persons with good knowledge in the corresponding Swiss language, or German/French courses for persons with another mother tongue (5,301 observations). 3 Basic computer course: This computer course provides basic knowledge, for instance Word courses, beginners courses, internet courses (7,455 observations).

for basic living allowances, Federal Office for Social Security (http://www.bsv.admin.ch/statistik/details/d/svs/kv_6_5.pdf) for average health insurance costs per canton, Swiss Statistical Yearbook for average apartment rent costs per canton. From gross income stated in social security data, 20% is discounted for taxes and contributions.

97

4 Higher vocational training: These courses are on a higher level than actual vocational education in Switzerland. They include computer courses on an advanced level for specialists, commercial, and technical courses (1,600 observations). 5 Lower vocational training: These courses are not beyond the level of a vocational education. They include commercial, technical, hotel/restaurant industry, nursing, and cleaning courses (1,617 observations). 6 Other training: This is a group that subsumes all remaining courses, such as laboratory firms, internship, self-employment courses (2,253 observations). 7 Employment programmes single workplace: Participants work in the public administration, in another public service, or in a non-profit organisation. They work together with other “normally” employed persons (2,055 observations). 8 Employment programmes collective workplace: Participants work in facilities specially designed for unemployed persons, for instance recycling (3,449 observations).

4.5 Definition of nonparticipants and programme starting dates Nonparticipants in the static method are defined according to Gerfin and Lechner (2002) as persons who do not participate in any programme during the observable time frame. For the outcome after programme start, a hypothetical programme starting date is assigned to persons in the following way: the programme start variable for all other persons is regressed on all available X characteristics, and the standard error of the residuals is computed. Based on the estimated coefficients and the corresponding X characteristics of the nonparticipants, plus a random variable with zero mean and the estimated standard error of the residuals, hypothetical starting dates for nonparticipants are computed. Nonparticipants who are de-registered two months later are excluded from the sample. For the "waiting" method, nonparticipation is defined on a monthly basis, such that a person in month t is a nonparticipant if they have not yet participated in a programme and will not be de-registered two months later. Furthermore, if a person is in an interim job during the same month, he or she is not eligible for a programme and is therefore excluded from the nonparticipants sample. Unfortunately, the reasons why individuals do not enter a programme is not observable with these methods. Three key reasons why people do not enter a programme need to be excluded, namely when a person has already found a job, when they are still in full

98

employment, or when they are not entitled to a programme. Therefore, the comparison (and the participant sample) excluded individuals who were de-registered within the next calendar month after starting a programme, and people who were employed without earnings loss according to the pension system in the corresponding month. Nonparticipants are required to have a positive payment from the unemployment insurance system at the month of a hypothetical programme starting date, in order to make sure that they are eligible for a programme. Furthermore, all persons are excluded from the sample after they have participated in a programme, because in this case that person (in a strict sense) is not part of the target group of a programme any more, and his or her present state is already an effect of a former programme. Out of the sample of 115,943 observations, 511 participants who are enrolled in programmes were deleted because they are employed without earnings loss at the beginning of their programme. In the nonparticipant group, 63,019 nonparticipants (of 88,808) are excluded during the static setup by trimming their pre-treatment durations and applying other criteria. In the dynamic case, 39,964 nonparticipants were never part of any control group at any time within their period of unemployment.

4.6 Descriptive statistics Selected descriptive statistics of the sample are presented in Table 11. An overview of all descriptive statistics of all variables available is given in the Appendix. Graphs with descriptive statistics of outcomes are displayed in Figure 27.19 At the beginning of the unemployment period, all persons are registered at the employment office and are searching for a job according to the unemployment insurance data. The pension system records show that a considerable number of these individuals still receive income from employment. The number of people without information in the pension system increases over time. Typically, the share of persons without information rises after the benefit entitlement period is exhausted (after 24 months). Similarly, the total available income drops after benefit exhaustion. At a first glance a small employment effect close to benefit exhaustion can be found.

19

The graphs display outcomes by programme participation. For nonparticipation, the static definition is assumed.

99

Table 11.

Selected descriptive statistics. Participants

Number of observations Current unemployment spell Beginning of spell in calendar months (1=January 2000) Unemployment benefit, daily, CHF Programme participation (in %) No course Personality course Language course Basic computer course Vocational training high level Vocational training low level Other training Employment programme single workplace Employment programme collective workplace Beginning of programme in calendar months (1=january 2000) Duration of current unempl. spell at beginning of first programme Duration of first programme (only people in courses / programmes) Other personal characteristics Age in years Female Marital status Single Married Divorced Nationality Swiss Foreigner with permanent permit Foreigner with yearly permit Qualification Skilled Semi-skilled Unskilled Caseworker's subjective rating of employability Very easy or easy Medium Difficult Monthly earnings in last job (CHF) Last job position High (management, etc.) Medium Low

– table to be continued –

27,135

Nonparticipants all 88,808

Nonparticipants nondyn. 25,789

8.68 142.38

9.02 139.99

9.45 142.69

1.00

1.00

0.13 0.20 0.27 0.06 0.06 0.08 0.08 0.13 12.69 5.01 62.99 38.14 0.53 0.29 0.57 0.13 0.58 0.26 0.16 0.55 0.17 0.28 0.13 0.71 0.15 4065.82 0.08 0.57 0.34

12.54 4.09

36.38 0.45 0.36 0.51 0.12 0.59 0.27 0.14 0.56 0.17 0.27 0.17 0.68 0.13 4021.63 0.07 0.58 0.34

37.17 0.48 0.33 0.54 0.13 0.58 0.27 0.14 0.56 0.17 0.27 0.16 0.69 0.14 4069.43 0.08 0.58 0.34

100

– Table 11 continued – Employment history from pension system data Income type: High income level (>60,000), increasing High income level (>60,000), decreasing Medium income level (40-60,000 CHF), increasing Medium income level (40-60,000 CHF), decreasing Low income level (20-40,000 CHF), increasing Low income level (20-40,000 CHF), decreasing No income (=200,000 0

(y

y tc,fi ) =

1 t ,i

1 N1

i , Ni > 0

1 N1

y t1,i

w

y t0, j

j

j

where w j is a weight attributed to every person in the comparison sample, which is a combination of the number of times it is used as comparison, and the weight it had in each comparison group: w j =

Ntre a te d i =1

1 1( j Ni

Ii ) .

The variance of this term is calculated as:

1 Ntre a te d

1

Vaˆ r Yt1 +

Nu n tre a te d

i Im

Finally, the treatment effect

[

E Y t 0 D = 0, X = x , Z = z 1

] E[Y

( w j ) 2 Vaˆ r Yt 0 .

2

t

0 t

j

j

is estimated by subtracting the estimation for

]

D = 0, X = x , Z = z 0 in form of the last pre-announce-

ment term (at time Ta)13:

ˆ = 1 t N1

i , Ni > 0

(y

1 t ,i

y T1a ,i

)

1 N1

w j

j

(y

0 t, j

)

y T0a , j .

Consequently, the estimated variance is:

Vaˆ r ( ˆt ) =

13

1 Ntre a te d

(Vaˆ r Y + Vaˆ r Y ) + N 1

i Im

t

i Im

1 Ta

1

u n tre a te d

2

j

(

)

( w j ) 2 Vaˆ r Yt 0 + Vaˆ r YT0a . j j

This share does not completely reflect the differences over time, since the chance to become unemployed is lower throughout, so differences would grow over time. In order to take this into account, the analysis would have to impose much more structure, which the current research seeks to avoid. The results will be interpreted accordingly, as this provides a lower bound.

138

3.2 Regression-discontinuity approach The second identification strategy uses a regression-discontinuity approach. It makes use of the fact that the policy change was age related. Only people below 55 were subject to the reduction. Hence, under further assumptions persons above 55 can be used as control group. They do not receive the treatment at all, but they face the same labour market conditions as the treated group. 3.2.1 Identification The exact identification strategy is a regression-discontinuity approach (cf. Hahn, Todd, and Van der Klaauw, 2001 for identification and estimation in regressiondiscontinuity approaches), with the age criterion of 55 as discontinuity threshold. A person who is beyond the required age is not affected by the reduction, while those under the age limit are. The effect is identified by comparing persons in a narrow environment either side of the boundary. The assumption of the regression-discontinuity approach is that there is a clear cut-off variable which determines whether a person is affected by a policy or not. Moreover, this cut-off itself is not related to the outcome of the policy, at least in a close environment either side of the boundary. In this context, this would mean that within a certain age bandwidth, age does affect the maximum duration of benefits directly, but not the chance of finding a job otherwise. As before, this research focuses on the average treatment effect on the treated. Under the pure regression-discontinuity design, this parameter could be identified in the close environment of the cut-off by the modified and much stronger assumption: Y0 D,

leading to

[

E Yt1

] [

] E[Y

Yt 0 = E Yt1 D = 1

1 t

]

D=0 .

However, the assumption of the pure regression-discontinuity approach might fail. If it was possible to compare persons with just one day of difference in their birth dates, it would be a rather convincing claim. However, the number of observations is relatively small, and in order to have reasonable sample sizes, a broader age group will have to be chosen (age 50 to 59 years). Age will thus become an important factor, since older people tend to have more difficulty finding a job. Hence, the assumptions for the re-

139

gression discontinuity approach are likely not to be satisfied. Instead, the corresponding effect again includes a bias term. However, it is assumed that the bias term is only due to age, and not due to different compositions of unemployed in the treated and untreated groups. In other words, it is assumed that other characteristics are equally distributed in the treated (younger) and untreated (older) group. The next assumption is that the pure age effect on the outcomes will be constant, at least over the period of one year (2002 to 2003). This strategy provides the opportunity to control for the effect of age by using data from one year earlier. We have to further assume that no pre-announcement effect of the treatment occured. The same parameter as above is calculated using the population of unemployed of one year earlier, satisfying the same criteria one year earlier, as if the reduction had taken place in 2002. The corresponding "treatment" effect is thus a pure age effect, because no benefit reduction actually took place. The age-controlled treatment effect is then identified by: E Yt12003 D = 1

0 E Yt 2003 D=0

E Y D =1 E Y D =0 ) (1444444 24444443 1 t 2002

0 t 2002

= E Yt 02003 D =1

= E Yt 1 D = 1

E Yt 0 D = 1 =

E Yt 02003 D = 0

t

3.2.2 Estimation Estimation is undertaken simply by computing differences in outcome means of the corresponding treated and untreated groups in 2002 and 2003 respectively. Denote the sample of treated in 2003 as I2003 and the sample of untreated in 2003 as J2003. Equivalently, definitions are used for the sample of virtually treated and untreated in 2003, namely those satisfying the same conditions one year earlier (same age, same contribution time) as I2002 and J2002. The estimated effect on the treated is then:

ˆ = t

1 NI 2003

i I 2003

y t12003 ,i

1 N J 2003

y t0, j j J 2003

1 "" $ NI 2002

k I 2002

y t12002 ,k

1 N J 2002

l J 2002

! y t0,l # # %

140

4 Data 4.1 Data source and sampling Data is available from the unemployment insurance system in Switzerland, incorporating all persons entering unemployment in Switzerland, covering the time period until December 2003. All persons who are potentially concerned by a reduction in benefits are selected, and for whom the outcome can be observed for a reasonable time. The general selection criteria for the base population are listed in Table 15. Table 15.

General selection criteria.

Selection criterion

Description

1 Beginning of benefit entitlement period

Only persons who are still in their benefit entitlement period at the end of June 2003 or later can be affected by the cut. Hence the start of the entitlement period has to be less than 2 years before that date (2nd of July 2001 or later). For a reasonable period of observable outcome, we select those with a benefit entitlement that starts until end of June 2002. Only persons with an entitlement period with duration of 2 years are selected.

2 Eligibility for benefits

Person is eligible for benefits and has a contribution time of at least 6 months. This selection criterion excludes those with a claim for other reasons than employment, since their total number of benefits within the entitlement period is generally lower.

3 No disability

Disability automatically makes automatically eligible for a higher number of benefits. Therefore, those with a claim or a request for benefits from the disability insurance are excluded.

4 Cantonal selection

Persons from the canton of Geneva are excluded from the study, since the canton of Geneva has been exclusively granted an extended maximum amount of benefits due to its high unemployment rate.

5 Valid work permit

This criterion applies for foreigners with a yearly permit (B permit). It should not have expired before the end of the entitlement period. The reason for this criterion is that in certain cases, the work permit might not be extended during unemployment.

The final total base sample consists of 92,802 persons. Note that it is the total population of all unemployed satisfying the restrictions defined by the general selection criteria one to five described above. For the data analysis different subsamples are defined. The first three samples use persons who are affected by the cut only. The first subsample contains all persons affected by the cut; the second subsample uses only those for whom the end of the enti-

141

tlement period is observable; a third subsample includes those who were still unemployed at the announcement date of the cut and actually at risk of losing benefits. Subsamples four to six are constructed for the regression-discontinuity analysis, and therefore use persons who did and did not experience the cut, but only of the age 50 to 59 in 2003. The fourth subsample contains all of them, the fifth again only those whose benefit entitlement is observed until the end, and the sixth subsample contains only those at risk of losing. Table 16 summarises the criteria for the different subsamples.

Figure 37 depicts the time sequence for the selection of those who are at risk of losing benefits. Until the end of January 2003, a minimum amount of 295 and a maximum amount of 413 daily allowances has to be received in order to suffer from an actual cut of at least one day in June 2003. All sample sizes are displayed in Table 17. Table 16.

Selection criteria for subsamples.

Selection criterion

Description

Samples 1

2

3

6 Affected by cut

The person has to be affected by the cut. This is the case X if a person is below 55 by the end of the entitlement period, or if the person has her 55th birthday between 1st of July 2003 and the end of the entitlement period, with a contribution time of less than 18 months.

X

X

7 End of entitlement period observable

Those with an entitlement period starting after 1st of July 2001 and before 15th of December 2001 are selected.

X

X

8 Unemployed in January 2003

Only those persons who are still registered as unemployed at the end of January 2003 and who have consumed a minimum and a maximum amount of benefits by 14 the end of January 2003.

9 Age 50-59

Only persons between 50 and 59 years of age in the corresponding year of the cut (2003 or 2002)

14

4

5

6

X

X

X

X

X

X

X

The time span February until June 2003 corresponds to 106 daily allowances (Feb: 20, Mar: 21, Apr: 22, May: 22, Jun: 21). In order to be at risk for a reduction, less than 520–106 = 414 benefits must be received by the end of June. Furthermore, until the end of the benefit entitlement period, the amount of benefits left and those already received must add up to more than 400.

142

Figure 37.

Time sequence for benefit reduction.

beginning of entitlement period

min. 295 benefits used max. 413 benefits used

Table 17.

January 2003: information

end of June 2003: cut

end of entitlement period

5 months (max. 106 benefits)

Sample sizes. All

All persons with cut Treated (2001-2003) Controls (2000-2002) Persons between 50 and 59 Treated (2001-2003) Controls (2000-2002)

Sample1 83,081 58,117 Sample 4 14,025 11,001

Starting between July and mid December Sample 2 35,969 27,488 Sample 5 5,598 4,959

At risk at announcement date Sample 3 3,508 1,684 Sample 6 1,010 581

The treatment variable is computed in the following way: 0 1

Person is over 55 by end of June 2003, or will be 55 after end of June, but before the entitlement period is over, and shows contribution time of at least 18 months Person will be under 55 at the end of the entitlement period, or person will be 55 between July and end of entitlement period, with a contribution time of less than 18 months

Note that the additional criterion of having 18 months of contribution time might make the groups with and without treatment in samples four to six more unequal, in the sense that in the category with the treatment there might be more persons who had been unemployed earlier, or who do not have ordinary employment histories. As a whole, 700 persons will become 55 after end of June 2003, but before their entitlement is over. 211 of these have an entitlement period of 18 or months or less and will be in the "treated" group. In sample five these figures are 138 (40 of them less then 18 months), and in sample six 20 persons (7 with less than 18 months contribution). This possible selection problem is therefore negligible. Descriptive statistics of the corresponding samples of all persons treated (Sample 1) and persons between 50 and 59 (Sample 4) are provided in Table 18.

143

Table 18.

Descriptive statistics of characteristics – full samples (treated versus old).

Number of observations Female Age in 2002 Foreigner Civil status: single Civil status: married or separated Insured earnings (CHF) Insured earnings lower than 3000 CHF Insured earnings higher than 4000 CHF Employability: difficult Employability: easy Qualification: skilled Qualification: unskilled Had earlier registration Has subsequent entitlement period Contribution time in months Number of daily allowances consumed in ent. period

Sample 11 83,081 45 35 26 49 41 4,291 25 52 18 17 62 22 34 4 18 158

Sample 42 14,025 43 53 20 13 62 4,804 21 61 33 10 61 22 38 4 20 108

Notes: 1Sample for all treated persons; 2Sample for elderly persons. Where not otherwise stated, figures are percent.

4.2 Outcome variables The primary outcome variables of interest in this study are three exclusive states: "unemployed", "found a job" and "left unemployment without finding a job". The information is defined daily according to the following procedure: Unemployed: a person is registered at the regional unemployment office. This is also the case if the person is currently in an interim job or any type of labour market programme. Found a job: when a person is de-registered and their de-registration code indicates that a job was found, the outcome "found a job" is 1 for every day of the subsequent period, until the person re-registers. While a person is registered, this outcome is always 0, and also if the person gets subsidies for an interim job. This outcome indicates that a person is in regular employment. Left unemployment: when a person is de-registered and their de-registration code indicates that no job was found, the outcome "left unemployment" is 1 for every day of the subsequent period, until the person re-registers. A possible job accep-

144

tance at a later time after de-registration cannot be observed. While a person is registered, this outcome is always 0. This outcome indicates that a person gave up searching for a job and possibly left the labour force. Both outcomes are measured daily from July 1, 2001, until December 31, 2003. All reregistrations during that period are taken into account. A third auxiliary outcome, "in interim job" measures to what extent persons are in subsidised employment while they are still registered at the unemployment office. Subsidised interim jobs are usually regular employment, but with a lower degree of employment, or paying a lower salary than before. If the total duration of the interim job/s exceeds 12 months during the benefit entitlement period, it allows for new benefit eligibility after exhaustion of the former benefit entitlement period. This outcome is measured daily, but based on monthly information. It is a binary variable as well, set at 1 if it can be concluded from the data that the person spent more than 50% of the corresponding month in an interim job, otherwise set at 0. Finally, two more outcomes are computed, one a binary variable that indicates if the person has a subsequent entitlement period following the old one, the other one the total amount of benefits received until the end of 2003. Descriptive statistics for these outcomes are displayed in Figure 38. The first diagram shows the development of the outcome states for all "treated" persons who entered unemployment between July 2001 and December 2002 (sample one). Note that until the end of December 2002 (day 549), new unemployed enter this group. The graph shows that persons either remain registered (23%) or find a job (59%), but relatively few of them leave unemployment for non-employment (17%). However, this fraction increases towards the end of the period. At the same time, unemployment remains constant over the last three months, and the employment rate even drops towards the end of the year. Persons with an entitlement period starting between July and December 2001 (sample two) show a similar pattern. However, a rise in the share of those leaving unemployment without finding a job is observable, as well as a further increase over time, while the employment share hardly increases or decreases towards the end. Around 17% of this sample are still registered at the end of 2003, 22% left without a job, and 61% held a job.

145

Figure 38.

Descriptive Statistics of outcomes for treated. Sample 2: Entitlement period starting July-December 2001

0

100

200

300

400 500 600 days, 1=1 July 2001

found a job left unemployment

700

800

900

0

0

10

20

persons in 1000 20 30

persons in 1000 40 60

40

80

Sample 1: All treated

0

unemployed

100

200

300

400 500 600 days, 1=1 July 2001

found a job left unemployment

700

800

900

unemployed

0

1

persons in 1000 2 3

4

Sample 3: Those unemployed in January 2003

0

100

200

300

400 500 600 days, 1=1 July 2001

found a job left unemployment

700

800

900

unemployed

All figures are based on persons who satisfy the conditions for benefit reduction, whose entitlement period starts after 1st of July 2001, and had their first observed registration relevant for the actual entitlement period before the end of 2002. Sample 1: All. Sample 2: Entitlement period starts before mid of December 2001. Sample 3: Registered at the end of January 2003 and at risk to lose benefits.

Examining the figures of those who are unemployed at the end of January 2003 (sample three) and at risk of losing benefits, it is clear that that many persons de-registered without finding a job right at the end of June. On the other hand, an increase in job finding rates also can be observed. At the end of 2003, 11 months after the announcement date, 17% of this sample is still registered as unemployed at the end of 2003, together with 51% who de-registered without finding a job. Only around one-third of the individuals in this sample actually found a job. Concerning the age group between 50 and 59, the statistics are displayed for treated and untreated separately (Figure 39). The graphs show a slight jump in the rate of those persons who leave unemployment without finding a job right at the end of June 2003. However, for those closer to the end of their entitlement period (samples five and six) the untreated "catch up" towards the end of the observation period. Through-

146

out the samples and periods, the "treated" and younger persons have a higher job finding rate. Figure 39.

Descriptive Statistics of outcomes for people aged between 50 and 59. Sample 5: Entitlement period starting July-December 2001

400

500

600 700 days, 1=1 July 2001

found job, untreated found job, treated

800

900

0

0

10

20

%

20

%

40

30

40

60

50

Sample 4: All

400

left, untreated left, treated

500

600 700 days, 1=1 July 2001

found job, untreated found job, treated

800

900

left, untreated left, treated

0

20

%

40

60

Sample 6: Those unemployed in January 2003

600

650

700 750 800 days, 1=1 July 2001 found job, untreated found job, treated

850

900

left, untreated left, treated

All figures are based on persons who are between 50 and 59 in 2003, whose entitlement period starts after 1st of July 2001, and had their first observed registration relevant for the actual entitlement period before the end of 2002. Sample 1: All. Sample 2: Entitlement period starts before mid of December 2001. Sample 3: Registered at the end of January 2003 and at risk to lose benefits.

4.3 Accuracy of outcome The outcome variables used so far in these analyses rely on the accuracy of the data. Unfortunately, the information is not as correct as it might appear. Lack of accuracy arises from two sources. First, the caseworker does not know exactly what happens to each individual they work with. This problem is worse at the end of a benefit entitlement period, when the unemployed no longer have to be registered and therefore are not obliged to announce job search efforts. The second problem begins at this point, as

147

once a person is de-registered, there is no follow-up unless the person comes back to the unemployment office. A person who de-registers without finding a job might find a job later, but they remain "inactive" in the sense of these outcomes. Equivalently, a person might find a job, but lose it later without coming back to the unemployment office. There is another data source from which information about subsequent labour market outcomes can be drawn, namely the social security system15. All sorts of employment subject to contribution to the pensions system are included in these records (including unemployment benefits). This data therefore allows for verification of the employment information system. Unfortunately, this data is available only with a time lag of around two years and cannot be used as outcome variable directly. Data for 2002 is available. From a combined social security / unemployment insurance database conclusions can be drawn about the reliability of the outcome measures. This analysis involves taking people at an earlier calendar time (2001) close to their benefit exhaustion date and comparing their outcomes for different months before and after this date. These comparisons provide reliability measures that can be applied to the results obtained for 2003. The first step involves taking all unemployed persons in Switzerland whose benefit exhaustion date is between September 2001 and August 2002, and who are still registered five months prior to the end of their benefit entitlement period. For each of these people the calendar month of benefit exhaustion is defined as a reference month. Four months before and four months after this reference month, parameters )( & ,' are computed. ( refers to the month (-4,…0,…4), &

{u , f , n } , refers to that person’s state in

the unemployment insurance system data (unemployed, left with job, left without job), '

{u , e , n } refers to their state in the pension system data (unemployed, employed,

out of labour force). The two graphs in Figure 40 show the results of this cross-validation for the two parameters ) f ,e and ) n , e . Between 85 and 91% of those who had announced they had found a job were indeed categorised as employed according to the social security data. However, those who left unemployment close to the exhaustion date have a lower employment rate, which shows in a significant drop in employment rates between -1 and 1. In contrast, between 36 and 48% of those who left without announcing a job were 15

In German: Alters- und Hinterbliebenenversicherung (AHV).

148

employed according to the pension system. The same pattern is revealed, such that those who left unemployment close to exhaustion time (compared to before) were less likely to have found a job according to the pension system data. Cross-validation UI / social security data ( ) f ,e and ) n , e ).

Figure 40.

left without finding a job

found a job 92

10'000

9'000 8'000

91

49

7'000

47

6'000

45

89

5'000

43

88

4'000

90

9'000 8'000 7'000 6'000 5'000 4'000

41

3'000

87

2'000 86 85 -4

-3

-2

-1

0

1

2

3

1'000 0

35

month b efore/after end of entitlement period AHV: employed

2'000 37

4

N (right scale)

3'000

39

1'000 0 -4

-3

-2

-1

0

1

2

3

4

month b efore/after end of entitlement period AHV: employed

N (right scale)

Based on all persons whose benefit entitlement period ended between September 2001 and August 2002 and who were still registered 5 months before the end. Each figure is based on persons who left in the meantime towards the corresponding state only (N is number of observations).

From this exercise two conclusions can be drawn. First, employment rates calculated from unemployment insurance data cannot be taken as real figures in absolute terms. There are individuals in the sample who de-registered towards employment, but did not find a job. On the other hand, there are other individuals who de-registered without employment, but actually were employed. The second conclusion is that de-registering at the exhaustion time itself is more likely to be towards non-employment than at other times. This conclusion will strengthen the findings presented in section 5.

149

5 Results The results are presented separately for the two identification strategies and for the different outcomes. Furthermore, results are presented in two different time scales, process time and calendar time. Process time is the time since the start of the benefit entitlement period for each individual. This outcome has the drawback that the announcement date is different for most people. Nevertheless, this outcome shows effects within the entitlement period, especially around the end of the entitlement period. The second time scale, calendar time measured in days, allows viewing of effects which change with the announcement date, but does not allow interpretation of where the underlying persons are within their individual entitlement period. Not only was the maximum entitlement period affected by the policy change, but also the minimum contribution time. This additional change might have made it more difficult for some unemployed people to "earn" a subsequent new entitlement period, which might have affected their behaviour. Therefore this section begins with results concerning some auxiliary outcomes, subsequent entitlement periods and number of benefits received in total (5.1), before discussing the results of the matching estimation (5.2) and regression-discontinuity design (0).

5.1 Number of benefits and consequent entitlement periods The results concerning the number of daily allowances and consecutive entitlement periods are displayed in Table 19. Effects for the older group only are shown, because a good correction for pre-treatment differences was not available. Effects are computed as differences, and differences in differences between the treated and nontreated. For the number of daily allowances there is a significant difference of around 14 allowances (almost three weeks) in the total population (sample four) after controlling for age. There are no significant differences for the sample of those registering early (sample five), possibly because most of this group are no longer unemployed at the date of the cut. Those still unemployed in January (sample six) clearly received a smaller number of instalments, by around seven weeks. Subsequent entitlement periods are only available for those starting their benefit entitlement before mid December 2001, and therefore results for the total sample are omitted. There is no significant effect on the probability of having a subsequent entitlement period based on these samples.

150

Table 19.

Results concerning entitlement period and number of daily allowances.

Number of daily allowances Treated mean Untreated mean Difference not age-controlled

Sample 4 195.63 222.81 -27.18*** (2.67) -13.98*** (3.92)

Sample 5 196.90 219.38 -22.48*** (4.45) -6.71 (6.26)

Sample 6 432.07 476.01 -43.94*** (3.21) -35.07** (5.80)

14,025 (7,881/ 6,144)

0.087 0.082 0.006 (0.008) 0.011 (0.012) 5,598 (3,311/2,287)

0 0.002 -0.002 (0.002) -0.008 (0.014) 1,010 (574/436)

Age-controlled Next entitlement period Treated Untreated Not controlled Age/controlled N (treated/untreated)

Numbers reflect differences in levels. Significance levels are indicated on * 10 percent level, ** 5 percent level, ***1 percent level.

5.2 Year-to-year comparison This section provides results for those people below the age of 55 who could potentially have been subject to the cut in benefits. Table 20.

Summary of matching procedure.

Treated Total number Match found Maximum number of matches Av. number of matches, if positive Untreated Total number Used as matches at least once Max. weight Average weight, if positive

Sample 1

Sample 2

Sample 3

83,081 72,286 235 24.31

35,969 31,000 187 22.32

3,508 1,479 12 2.26

58,117 53,259 17.27 1.36

27,488 25,118 17.27 1.23

1,684 956 7.33 1.55

5.2.1 Matching results The first step in the estimation is the matching of treated to untreated individuals. Table 20 provides a summary of the matching procedure. In the largest sample, for

151

87% of the population at least one match was found. The maximum number of matches was 235, and on average a person could be matched to 24 controls. These numbers are roughly the same for sample two. For sample three, only 42% of the population could be used for matching. For all others, no corresponding person was found. On average, 2.2 controls were found. Figure 41 displays the outcome "found a job" in the two time schemes (process and calendar time). First, it is clear that job finding is higher in the control group before and after the cut was announced, which reflects the generally lower labour market conditions for the treated group. The matching even increases the difference to some extent. There is a cyclical movement in entitlement period after one and two years, but also stagnation in calendar time during the cold season (days 450-600). The observed stepwise evolution in calendar time is due to the fact that de-registrations often take place at the end of a month. For those at risk of losing benefits (sample three) a rise in employment at the time of benefit exhaustion was observed for the control group. For the treated group this jump takes place at the calendar time of the cut. However, this should not be interpreted as direct evidence for individuals accepting employment after benefits are exhausted. Some people continue a job that they had before the exhaustion, but do not receive compensation payments via the interim job scheme any more. As indicated above, for identification of the causal effects of the cut it is necessary to control for the generally lower labour market conditions by differencing out the preannouncement "effects". An "eyeball test" on the data for job finding confirms that the difference between the two lines is more or less constant after a certain point. Note that this method is not possible for sample three, as their labour market status is conditioned at the announcement itself. Differences therefore cannot be attributed to the cut alone, but reflect the generally lower labour market as well; thus there can be no talk about causal effects in this case, but this study can still examine the results from a qualitative perspective, keeping this fact in mind.

152

Figure 41.

Descriptive outcomes for outcome "found a job". In process time

In calendar time

0

.3

.4

.2

.5

.4

.6

.6

.7

.8

Sample 1: all treated

100

200

300

400 500 600 days in entitlement period

700

800

100

900

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

.3

0

.4

.2

.5

.4

.6

.6

.7

.8

Sample 2: entering July-December

100

200

300

400 500 600 days in entitlement period

700

800

100

900

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

0

0

.1

.1

.2

.2

.3

.3

.4

.5

.4

Sample 3: unemployed at announcement date

400

500

600 700 days in entitlement period

Outcome treated Outcome matched treated

800

900

Outcome controls Outcome matched controls

400

500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003 Outcome treated Outcome matched treated

900

Outcome controls Outcome matched controls

Sample 1: Unmatched: 83,081 treated, 58,117 controls. Matched: 72,286 treated, 53,259 controls. Sample 2: Unmatched: 35,969 treated, 27,488 untreated. Matched: 31,000 treated, 25,118 controls. Sample 3: Unmatched: 3,508 treated, 1,684 untreated. Matched: 1,479 treated, 956 controls.

Figure 42 displays outcomes "left without finding a job". First, the control group was more likely to leave towards this non-employment state, but that the difference is quantitatively small compared to the first outcome. This pattern is in line with a priori beliefs about the effect of the cut. There is a jump at the exhaustion time itself towards this state. This jump is more substantial for sample two, which can be observed at the end of the entitlement period. Prior to exhaustion, there is a smooth pattern for the

153

control group, but a kink for the treated group after the cut came into effect. There is a significant jump at calendar date 730, when the cut took place. Figure 42.

Descriptive outcomes for outcome "left without finding a job". In process time

In calendar time

0

0

.05

.05

.1

.1

.15

.15

.2

.2

.25

Sample 1: all treated

100

200

300

400 500 600 days in entitlement period

700

800

100

900

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

0

0

.05

.05

.1

.1

.15

.15

.2

.2

.25

Sample 2: entering July-December

100

200

300

400 500 600 days in entitlement period

700

800

900

100

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

.6 0

0

.2

.2

.4

.4

.6

.8

Sample 3: unemployed at announcement date

400

500

600 700 days in entitlement period

Outcome treated Outcome matched treated

800

900

Outcome controls Outcome matched controls

400

500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003 Outcome treated Outcome matched treated

900

Outcome controls Outcome matched controls

Notes see Figure 41.

Here again, for the first two samples the assumptions about constant differences in employment rates between treated and non-treated, as compared to the last preannouncement date, seems to be reasonable. For sample three, there can be observed a slightly lower probability to be in a non-employment state after the cut, which might

154

again be due to the overall conditions, as this also occurs earlier in the larger samples. The results will indicate if this difference is significant, and the effects will be interpreted accordingly. Figure 43.

Descriptive outcomes for outcome "registered". In process time

In calendar time

.2

0

.4

.2

.6

.4

.8

.6

1

.8

Sample 1: all treated

100

200

300

400 500 600 days in entitlement period

700

900

800

100

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

0

.2

.2

.4

.4

.6

.6

.8

1

.8

Sample 2: entering July-December

100

200

300

400 500 600 days in entitlement period

700

800

100

900

200 300 400 500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003

900

0

.2

.2

.4

.4

.6

.6

.8

.8

1

1

Sample 3: unemployed at announcement date

400

500

600 700 days in entitlement period

Outcome treated Outcome matched treated

800

900

Outcome controls Outcome matched controls

400

500 600 700 800 calendar days, 1=1 July 2001, 580=end of January 2003 Outcome treated Outcome matched treated

900

Outcome controls Outcome matched controls

Notes see Figure 41.

To complete the picture, Figure 43 displays unemployment rates. Even though there are large differences in outcomes at earlier stages in the entitlement period, these differences more or less cancel out towards the end of the observation periods, as persons

155

de-register from unemployment insurance. Only a stock of 10 to 20% remain, which appear not to be affected by the labour market conditions directly. A subset of these will also have renewed entitlement periods. This observation means that the "effects" of the cut should be interpreted between the announcement time and the date of the actual cut, and until shortly after the expiry of the entitlement period. Any longer term persistent effect on unemployment must be zero, and might be due to the pre-treatment correction. 5.2.2 Estimated Effects This section presents the estimated effects on job finding and non-employment. Note that the figures are already corrected for pre-announcement differences.16 The effects are displayed together with a 95% confidence interval, for every point in time. If the corresponding point is above zero, then the persons subject to the cut are more likely to be in the corresponding state (employed or non-employed). The first block of results ("found a job") is displayed in Figure 44. In the total sample of those treated, the minimum date of the announcement is after 215 days, and the maximum after 580 days. The new total entitlement duration of 400 daily allowances can affect persons earliest after 365 days. After 730 days, the first entitlement period is expired for all. Those individuals subject to a cut were not more likely to find a job towards the end of the entitlement period (730 days). For sample two no significant effects can be found. Sample three has lower job finding rates after the date of the announcement, but a jump back to higher levels at the time of the cut itself. This cannot be readily interpreted as an effect of the cut, because this does not control for labour market conditions. The only conclusion which can be drawn is that at no point in time were people more likely to have found a job compared to the earlier year, given that they were unemployed in January prior to the end of the entitlement period. Unmatched effects in terms of the outcome "left without finding a job" are displayed in Figure 45. There is a clear effect on this outcome in all samples. There is a jump at the calendar time of the cut, which is around 20% for those in sample three, and 2.4% for sample two. The figures indicate a persistently higher non-employment rate compared to the earlier cohort. However, there are limitations of the pre-announcement

16

The following pre-treatment matching dates are used for correction for process time: 200 for sample one, 365 for sample two, and sample three does not need to be corrected. In calendar time, 585 is chosen for sample one and two.

156

correction, and as such the current study focuses on the time between announcement and implementation of the policy change, rather than longer term effects. The fact that unemployment effects cancel out towards the end of the observation period (see section 0) would suggest that a persistent effect on non-employment rates would have to be compensated by a persistent negative employment effect, which is not observed. This point will be discussed further in the context of the results obtained by the regression-discontinuity approach. Figure 44.

Results for "found a job" (matched).

-.02

-.02

-.01

-.01

0

0

.01

.01

.02

.02

.03

Sample 1: all treated

200

300

400

500 600 days in entitlement period

700

800

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

-.02

-.02

-.01

-.01

0

0

.01

.01

.02

.02

.03

.03

Sample 2: entering July-December

350

450

550 650 days in entitlement period

750

850

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

-.1

-.2

-.15

-.05

-.1

-.05

0

0

.05

.05

Sample 3: unemployed at announcement date

450

500

550

600 650 700 days in entitlement period

750

800

850

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

The displayed effects are the differences between outcomes of the treated and adjusted outcomes of the untreated, together with a 95 percent confidence interval. Sample 1: 83,081 treated, 58,117 controls. Correction based on day 205 (set to 0). Sample 2: 35,969 treated, 27,488 controls. Correction based on day 369 (set to 0). Sample 3: 3,508 treated, 1,684 controls. No correction.

850

157

Figure 45.

Results for "left without finding a job" (matched). Process time

Calendar time

-.02

-.02

0

0

.02

.02

.04

.04

.06

.06

Sample 1: all treated

200

300

400

500 600 days in entitlement period

700

200

800

300

400

500 600 days in entitlement period

700

800

-.02

-.02

0

0

.02

.02

.04

.04

.06

.06

Sample 2: entering July-December

350

450

550 650 days in entitlement period

750

450

850

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

-.1

0

0

.1

.2

.1

.3

.2

.4

.3

Sample 3: unemployed at announcement date

450

500

550 600 650 700 days in entitlement period

750

800

850

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

Notes see Figure 44.

The unmatched results are listed in the Appendix and do not significantly differ from these findings. The selection problem seems not to be a decisive component for the results.

5.3 Regression-discontinuity approach This section describes the results of the regression-discontinuity approach. Identification comes from a comparison between those just above the age criterion to those just below.

158

Figure 46.

Results for "found a job". Not age-controlled

Age controlled

-.05

0

0

%

% .05

.05

.1

.1

Sample 1: all

0

100

200

300

400 500 600 days, 1=1 July 2001

700

800

0

900

100

200

300

400 500 600 days, 1=1 July 2001

700

800

900

700

800

900

700

800

900

-.05

-.05

0

% 0

%

.05

.05

.1

Sample 2: registered between July and December 2001

0

100

200

300

400 500 600 days, 1=1 July 2001

700

800

0

900

100

200

300

400 500 600 days, 1=1 July 2001

-.05

-.1

0

-.05

% 0

% .05

.05

.1

.1

.15

Sample 3: those at risk at end of February 2003

0

100

200

300

400 500 600 days, 1=1 July 2001

700

800

900

0

100

200

300

400 500 600 days, 1=1 July 2001

Own calculations. Sample sizes: Sample 4: 14,025 (7,881 treated / 6,144 untreated); Sample 5: 5,598 (3,311/2,287); Sample 6: 1,010 (574/436). Sample sizes for age controlling: Sample 4: 11,001 (6,278 treated; 4,723 untreated); Sample 5: 4,959 (2,917/2,042); Sample 6: 581 (316/265).

5.3.1 Effect on job finding Figure 46 depicts the effects before and after controlling for age. Examining the pure differences between those who were affected by the cut and those who were not (left column), there is a significantly higher job-finding rate for the treated than for the nontreated throughout the observation period, in both sample one and sample two. Note that these figures are based on a population of the age 50 to 59, and thus might reflect a pure age effect. For those who are actually at risk at the time when the policy change

159

was announced (sample six), no significant effect between the announcement date (580) and the cut-off date (730) is found. After controlling for age as described above, a small significant effect on job-finding rates can be found only towards the end of the observation period for sample one. This might indicate an effect of the policy change after its implementation on the newly unemployed, but the size of the effect is still small (2 to 4%). It can be concluded that job-finding rates were hardly affected by the policy change in a short term. Figure 47.

Results for "leaving unemployment without job" (not age controlled). Not age-controlled

Age controlled

-.02

-.02

0

0

.02

%

%

.04

.02

.06

.04

.08

Sample 1: all

0

100

200

300

400 500 600 days, 1=1 July 2001

700

800

0

900

100

200

300

400 500 600 days, 1=1 July 2001

700

800

900

700

800

900

700

800

900

-.05

-.05

0

%

% 0

.05

.05

.1

Sample 2: registered between July and December 2001

0

100

200

300

400 500 600 days, 1=1 July 2001

700

800

0

900

100

200

300

400 500 600 days, 1=1 July 2001

-.1

-.1

0

0

.1

%

% .1

.2

.2

.3

.3

.4

Sample 3: those at risk at end of February 2003

0

100

200

300

400 500 600 days, 1=1 July 2001

Own calculations. See note in Figure 46.

700

800

900

0

100

200

300

400 500 600 days, 1=1 July 2001

160

5.3.2 Leaving unemployment without finding a job The results concerning leaving unemployment without finding a job are more conclusive. There is a jump at the cut-off date 730 for all samples. For sample two, this jump is around 5%. In sample three, around 30% more of those who actually were at risk of losing benefits (sample three) left unemployment without finding a job. In sample one the effect is (significantly) positive and persistent. These results do not change after controlling for age. However, the probability of leaving unemployment after the announcement date is even lower for those who are at risk of losing benefits. These results suggest that significantly more people left the unemployment registry without finding a job. For those who were concerned by the reduction in benefits, the difference is around 28%. Typically, the effect disappears towards the end of the year, when benefits run out for the comparison group as well, but not for sample one. 5.3.3 Interim jobs A special scheme in Switzerland allows, and also forces individuals to accept and employment during their unemployment period. There are various types of interim jobs, and they can either be permanent or temporary. They have one feature in common, that they pay less than the current unemployment benefits, be it because of a lower degree of employment, or because the person is overqualified for the job. If such a lower-paid job is accepted temporarily, a person is entitled to compensation by the unemployment insurance system, as long as benefits are not exhausted. While this subsidy is still being received, the person remains in the unemployment registry, but collects contributions to qualify for new benefit entitlements. If person accepts a lower-paid job, this could result in not being coded as "taken up a job", but rather that they still remaining in the registry. To check if this is the case, an additional outcome "in interim job" was constructed. The results are displayed in Figure 47. They are shown only for the regressiondiscontinuity design. No positive effects of taking up a lower-paid job can be discerned during the period of interest (580 to 730). The small negative effects found towards the end of the observation period in sample three could be because the benefits of the treated were already exhausted, and there was no claim on the subsidy any more.

161

Figure 48.

Results for "in subsidised interim job". Not age-controlled

Age controlled

-.01

-.015

-.01

-.005

%

% -.005

0

0

.005

.005

Sample 4

400

500

600 700 days, 1=1 July 2001

800

400

900

500

600 700 days, 1=1 July 2001

800

900

500

600 700 days, 1=1 July 2001

800

900

500

600 700 days, 1=1 July 2001

800

900

-.02

-.01

-.01

-.005

%

%

0

0

.005

.01

Sample 5

400

500

600 700 days, 1=1 July 2001

800

400

900

-.04

-.04

-.02

-.02

% 0

% 0

.02

.02

.04

.04

Sample 6

400

500

600 700 days, 1=1 July 2001

800

900

400

Own calculations. See note in Figure 46.

6 Conclusions This study compared the effect of a cut in maximum benefit duration in Switzerland using both matching estimation and a regression-discontinuity design. This study is based on comprehensive micro data of the unemployment insurance system utilised by all unemployed persons in Switzerland. The effect was estimated by imposing very little structure on the effects as well as on the processes. By matching estimation, a sample of people affected by the cut (below 55 years of age) was compared to a sam-

162

ple of people one year earlier, matched according to individual characteristics. Pretreatment differences in employment rates were used to control for a part of the changing labour market conditions. The second method used the age threshold of 55 years and compared persons above the threshold who were not affected by the cut to persons below, following a regression-discontinuity approach. Calendar time age effects were controlled for by subtracting estimated effects of the previous year. For the time span available (up to six months after the actual cut and 11 months after the announcement of the cut), it was found that cutting benefits did not lead to significantly or substantially higher job acceptance rates. No positive effect on the probability of taking up a non-suitable job (interim job) was found. There was also no effect on eligibility for subsequent benefit entitlement. Instead, significantly more people left unemployment for non-employment after the exhaustion of their benefit entitlement. Hence, the theoretical argument that an unemployed individual lowers their reservation wage and/or increases their job search behaviour in this situation (both of which would lead to a higher job adoption rate), cannot be supported with this data. Switzerland maintains a generous unemployment benefit system, with a high replacement rate and comparatively long benefit durations. At the same time, the system attempts to monitor unemployed persons' job search efforts tightly, by means of regular (at least monthly) meetings with an employment officer, and by the possibility of sanctions. Counselling services are available for every person. The findings of the current research suggest that this system seems to overcome possible disincentive effects fairly well. It also does not discourage persons from taking up employment in an advanced stage of the unemployment spell (after at least 1.5 years). With regard to the other key finding, the likelihood of leaving unemployment without finding a job, the question of optimality and efficiency arises. This researcher argues that a state in which persons have to leave the registry without finding a job cannot be socially desirable. As long as people remain under the umbrella of regional employment offices they receive counselling services and are motivated and encouraged to search for a job. While these people are entitled to benefits, there is the possibility of sanctions if they do not show sufficient job search efforts. During this time employment officers might succeed in motivating and encouraging such people to accept employment, be it by active placements or personal talks. However, if a person does not remain in the registry, these tools are very limited. This could even explain a decrease in the employment rate for persons who were subject to the cut (as was observed for persons who were actually threatened by losing benefits).

163

The figures suggest that the unemployment insurance system saved around 35 daily allowances for each unemployed person aged between 50 and 55, who were notified that their benefits would soon be cut in January 2003. This finding, together with the fact that no increased job finding was found amongst this group, leads to the conclusion that certain groups of persons were significantly affected by this cut, and its equity effects might have been considerably high compared to the possible improvement made by incentives. If there is a need for fewer expenses, policy makers could have considered an alternative strategy. A system with decreasing replacement rates over time, rather than a cut in benefits, is one possible alternative.

164

Appendix Table A12.

Probit selection estimation.

Dependent variable: starting entitlement period in 2001/2002 Variable Number of dependants Age in years Age below 30 Age above 50 Female Mother tongue French Mother tongue Italian Mother tongue others French speaking canton Italian speaking canton Mother tongue is main cantonal language Mother tongue is Swiss language Foreign language German good Foreign language English good Foreign language French good Foreign language Italian good Foreign language is cantonal language Civil status married Civil status widower Civil status divorced Foreigner permanent residence Foreigner annual residence permit Tertiary education Qualification semi-skilled Qualification unskilled Employability very good Employability good Employability medium Employability poor Employability very poor Mobility daily commuter Mobility parts of Switzerland Mobility whole Switzerland Mobility abroad

– table to be continued –

Coefficient

t-value

-0.0024 -0.0042 0.0859 -0.0074 0.0284 0.0346 0.0555 -0.0154 0.0297 -0.0736 0.0379 -0.0529 -0.0407 0.0492 -0.0073 0.0049 0.0385 -0.0360 -0.0687 -0.0387 -0.0417 -0.1452 0.2822 0.1825 0.0980 -0.0991 0.0253 0.0822 0.1699 0.1386 0.1930 0.0714 0.1018 0.0329

-0.6 -5.28 6.41 -0.48 3.35 2.19 2.57 -0.25 0.91 -2.81 2.88 -2.37 -2.77 5.1 -0.65 0.36 2.63 -2.97 -1.72 -2.97 -3.79 -7.35 19.69 18.05 8.83 -1.05 0.27 0.89 1.84 1.45 3.76 1.28 1.69 0.47

165

– Table A12 continued – Variable Insured salary Insured salary 2000-3500 CHF Insured salary 3500-5000 CHF Insured salary 5000-7000 CHF Insured salary >7000 CHF Position self-employed Position management Position specialised worker Region east Region central Region northwest Region romandie Region Zurich Unemployment rate in industry Unemployment rate in industry missing Constant

Coefficient

t-value

0.0001 -0.0519 -0.0062 0.0282 -0.0023 -0.2037 -0.1464 -0.0026 0.1081 0.1612 0.0815 -0.0706 0.1209 0.5977 1.6545 -1.5806

6.84 -3.25 -0.26 0.83 -0.04 -3.14 -8.28 -0.27 3.68 4.79 2.53 -4.82 3.79 122.3 98.59 -14.97

The results also include 24 dummies for profession (all significant) and 28 industry dummies (4 not significant) which are omitted. N = 152,091, log likelihood = -92,261.091.

166

Figure A2.

Outcome differences for "found a job" (unmatched). Process time

Calendar time

-.02

-.02

-.01

-.01

0

0

.01

.01

.02

.02

Sample 1: all treated

200

300

400

500 600 days in entitlement period

700

450

800

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

-.02

-.03

-.02

-.01

-.01

0

0

.01

.01

.02

.02

Sample 2: entering July-December

350

450

550 650 days in entitlement period

750

850

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

450

500

550

600 650 700 days in entitlement period

750

800

850

-.1

-.2

-.08

-.15

-.06

-.1

-.04

-.05

-.02

0

0

Sample 3: unemployed at announcement date

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

The displayed effects are the differences between outcomes of the treated and not adjusted outcomes of the untreated, together with a 95 percent confidence interval. Sample 1: 83,081 treated, 58,117 untreated. Correction based on day 205 (set to 0). Sample 2: 35,969 treated, 27,488 untreated. Correction based on day 369 (set to 0). Sample 3: 3,508 treated, 1,684 untreated. No correction.

850

167

Figure A3.

Outcome differences for "left without finding a job" (not matched). Process time

Calendar time

-.02

-.01

0

0

.01

.02

.02

.04

.03

.06

Sample 1: all treated

200

300

400 500 600 days in entitlement period

700

450

800

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

350

450

550 650 days in entitlement period

750

850

-.02

-.02

0

0

.02

.02

.04

.04

.06

.06

Sample 2: entering July-December

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

450

500

550

600 650 700 days in entitlement period

Notes see Figure A2.

750

800

850

-.1

0

0

.1

.1

.2

.2

.3

.3

.4

Sample 3: unemployed at announcement date

450

500 550 600 650 700 750 800 calendar days, 1=1 July 2001, 580=end of January 2003

850

168

169

Summary and Outlook

170

Labour market policies are measures attempting to facilitate and improve the functioning of the labour market. Passive measures provide income support and financial security during periods of involuntary joblessness, whereas active policies influence job matching in the labour market directly. Active labour market policies in Switzerland are wage subsidies, employment programmes, training, counselling, and different allowances for commuting, education and bad weather. The first part of this thesis discussed the development and key features of the unemployment insurance system and labour market policy instruments in Switzerland. In parts two and three, active and passive policy measures were evaluated in terms of their effects on labour market states and income. Comprehensive micro data from the unemployment insurance system and the pension system were used for this purpose. Active labour market programmes were evaluated by matching estimation with regard to different outcomes such as employment, unemployment and income. Vocational training (both high and low level), basic computer courses and language programmes were found to be successful. The results demonstrate that individuals who participate in these programmes are better off in almost every sense, such that this group were more likely to be employed, where "employment" was analysed not only quantitatively, but also qualitatively in terms of stability, purchasing power, and sustainability. Moreover, participants tended to keep in touch with the labour market longer than nonparticipants, in the sense that they did not leave the labour force and hence were less likely to become dependent on social assistance. Personality courses did not show negative effects in the longer run, but also not noticeable success. Results regarding employment programmes in single workplaces were slightly positive in the longer run. Employment programmes in collective workplaces were found to have the least success, since they did not succeed in keeping people attached to the labour market. However, there is anecdotal evidence that this effect might even be intended. These programmes are said to be used by caseworkers to test either willingness to work or availability (e.g., especially mothers and persons suspected to have illegal employment while receiving benefits). Therefore, exit towards inactivity as a "causal effect" of this programme should perhaps not be considered a "negative" impact, but rather a desired consequence. The effects were measured by two propensity score matching estimands, one based on Sianesi (2004) and one based on Lechner (1999). The results using Lechner’s (1999) method produced more “negative” results on almost every occasion. In terms of passive labour market policy, this research evaluated the effect of a cut in maximum benefit duration in Switzerland. Estimations were based on matching estimation and using a regression-discontinuity design using difference-in-differenced

171

estimation. Identification came primarily from comparisons to an earlier cohort of unemployed. These analyses revealed that cutting benefits did not lead to higher job acceptance rates. Rather, significantly more people left unemployment towards nonemployment after exhaustion of their benefit entitlement. Hence, the theoretical argument that an unemployed individual lowers her reservation wage and/or increases job search in this situation (which would both lead to higher job acceptance rates), could not be supported with this data. However, it is not clear whether these negative effects were persistent or not. This thesis argues that as long as people remain in the unemployment insurance system, they can be monitored and encouraged to return to employment. Leaving the registry can lead to the unemployed person being excluded from the labour market, such that any type of attachment to the market is lost. The Swiss unemployment insurance system provides a comparatively high level of insurance, in terms of duration as well as benefit level. Its institutions have undergone certain adjustments in the past 21 years, since the enactment of the unemployment insurance act. Switzerland had to deal with a large rise in unemployment in the 1990s. The results demonstrate that, despite its generosity, the Swiss system does not seem to provide undesired negative incentives for the unemployed, but rather achieves the aim of providing income support and facilitating job matching by counselling activities and monitoring. Moreover, some of its active policies are also effective in bringing people back to work earlier and increasing their future earnings. However, there are two caveats to these findings. First, the analysis does not incorporate effects on the labour market itself, such as different behaviour of job-seekers and employers due to the presence of certain policies. Particularly note that the interim job scheme can lead to distortions in the labour market. Furthermore, short or lower benefit entitlements might have effects on wages. Second, a cost-benefit analysis was not provided for active labour market programmes. It is argued that often the benefits do not justify the costs of programmes, and the success of programmes for participants is bought at the disadvantage of others. However, compared to the findings of the evaluation studies conducted in the 1990s, some lessons seem to have been learned, and the effectiveness of the system seems to have improved. In terms of passive maintenance, the reduction in 2003 might not have been the optimal strategy. If this reduction aimed to save costs, a replacement rate that decreases over time might have been an alternative. This thesis concentrated on average effects of policies only. Future work could address subgroup heterogeneity of effects. Furthermore, the time horizon of this research is limited to 1.5 to 2.5 years after programme participation in the case of active labour market programmes, and to six months in the case of the benefit reduction analyses.

172

Data for a longer time horizon was simply not available at the time of this study, but can systematically be updated for future research. Finally, a cost-benefit analysis was not provided here, which is often noted as a topic of specific interest for policy makers. This analysis was not undertaken due to a lack of reliable cost data. Costs of individual programmes are judged as sensitive information and are thus not publicly available, but could be requested and collected from cantonal and federal authorities.

173

References Acemoglu, Daron and Robert Shimer (1999), Efficient Unemployment Insurance, Journal of Political Economy, Vol. 107, No. 5 (Oct. 1999), 893-928. Addison, John T. and Pedro Portugal (2004), How Does the Unemployment Insurance System Shape the Time Profile of Jobless Duration?, IZA Discussion Paper No. 978. Aeppli, Daniel C., Brigitte Hoffmann, Roland Theiss (1998), Ausgesteuerte in der Schweiz: ein Situationsbericht, Hrsg. BWA, Bern: Haupt. Aeppli, Daniel C., Cora Hotz, Valerie Hugentobler, Roland Theiss (1996), Die Situation der Ausgesteuerten in der Schweiz, Hrsg. BIGA, Bern: Haupt. Aeppli, Daniel C., Roli Kälin, Walter Ott, Matthias U. Peters (2004), Wirkungen von Beschäftigungsprogrammen für ausgesteuerte Arbeitslose, Zürich: Rüegger. Angrist, J. D. and A.B. Krueger (1999), Empirical Strategies in Labour Economics, in Ashenfelter, O. and D. Card (eds.), Handbook of Labor Economics, Elsevier. Arulampalam, Wiji (2001), Is unemployment really scarring? Effects of unemployment experiences on wages’, Economic Journal, 111 (November), 585-606. ATAG Ernst & Young (1999), RAV Evaluationsstudie. Schlussbericht, Beiträge zur Arbeitsmarktpolitik Nr. 4, BWA Schriftenreihe.7 Atkinson, A. B. and Micklewright, J. (1991) Unemployment compensation and labour market transitions: a critical review, Journal of Economic Literature, XXIX, 1679–727. Battaglini, Monica; Olivier Giraud (2003), Policy Styles and the Swiss Executive Federalism: Comparing Diverging Styles of Cantonal Implementation of the Federal Law on Unemployment, Swiss Political Science Review, spring 2003, v. 9, iss. 1, pp. 285-308. Bauer Tobias, Baumann Beat, Künzli Kilian (1999), Evaluation der Regelung des Zwischenverdienstes in der Schweiz, seco Publikation, Beiträge zur Arbeitsmarktpolitik Nr. 1, Bern, seco. Betcherman, Gordon, Karina Olivas, and Amit Dar (2004), Impacts of Active Labor Market Programs: New Evidence from Evaluations with Particular Attention to Developing and Transition Countries, Social Protection Discussion Paper Series, No. 0402, World Bank, January 2004.

174

Blanchard, Olivier and Jean Tirole (2004), The Optimal Design of Unemployment Insurance and Employment Protection. A First Pass, NBER Working Paper, No. 10443. Boone, J., J. C. van Ours (2000), Modeling Financial Incentives to Get Unemployed Back to Work, IZA Discussion Paper No. 108. Bover, Olympia, Manuel Arrelano and Samuel Bentolila (2002), Unemployment Duration, benefit duration and the business cycle, The Economic Journal 112 (April), 223-265. Burdett, Kenneth, Unemployment insurance payments as a search subsidy: A theoretical analysis, Economic Inquiry, 17. 333-343. Card, David and Phillip Levine (2000), Extended Benefits and the Duration of UI Spells: Evidence from the New Jersey Extended Benefits Program, Journal of Public Economics, 78, 107-138. Clark, Andrew E. and Andrew J. Oswald (1994), Unhappiness and unemployment, Economic Journal, 104, 648-659. Curti, Monica (2002), L'évaluation de la politique du marché suisse du travail: une analyse d'efficacité par l'approche non-expérimentale, Dissertation, University of Fribourg. Curti, Monica, Erika Meins (1999), Politiques suisses du marché du travail – Différences resultant de sa mise en application au niveau du service de l'emploi, Die Volkswirtschaft – La Vie Economique, 2/99, pp.64-70. Dar, Amit, and Zafiris Tzannatos (1999), Active Labor Market Programs: A Review of the Evidence from Evaluations, Social Protection Discussion Paper Series, No. 9901, World Bank, January 1999. Davidson, C and S Woodbury (1997), Optimal Unemployment Insurance, Journal of Public Economics, 64, 359-387. Decreuse, B. (2002), On the Time Sequence of Unemployment Benefits When Search is Costly, Labour, 16(4), 609-633. Dehejia, Rajeev H. and Sadek Wahba (2002), Propensity Score-Matching Methods for Nonexperimental Causal Studies, Review of Economics and Statistics, February 2002, 84(1), 151-161. Djurdjevic, Dragana (2003), Unemployment and under-employment: the case of Switzerland, Diskussionspapier 2003-18, Universität St. Gallen.s

175

Djurdjevic, Dragana (2005); Unemployment and Under-Employment: The Case of Switzerland, Swiss Journal of Economics and Statistics, 141. Faesi, Thomas (1999), Arbeitslosenentschädigung und Zwischenverdienst: Ursachen und Wirkungen der zweiten Teilrevision des AVIG, Dissertation, Zürich: Schulthess, 1999. Falk, Armin, Rafael Lalive and Josef Zweimüller (2005), The Success of Job Applications: A New Approach to Program Evaluation, Labour Economics, 12(6) (2005), 739-748. Fay, Robert G. (1996), Enhancing the Effectiveness of Active Labour Market Policies: Evidence from Programme Evaluations in OECD Countries, OECD Labour Market and Social Policy Occasional Papers, No. 18, OECD Publishing. Fehr, Ernst and Lorenz Götte (2005), Robustness and Real Consequences of Nominal Wage Rigidity, Journal of Monetary Economics, 52, 779-804. Ferro-Luzzi, Giovanni, Yves Flückiger, José V. Ramirez, and Anatoli Vassiliev (2001), Performance Measurement and Determinants of Inefficiency of Regional Employment Offices: A Non-parametric Frontier Analysis for Switzerland, mimeo. Filippini, Paolo (1998), Regional Unemployment Disparities: The Case of the Swiss Cantons, Zurich: vdf. Flückiger, Yves (1998), The Labour Market in Switzerland: The End of a Special Case? International Journal of Manpower, Vol. 19, No. 6, 369-395. Fredriksson, Peter and Per Johansson (2004), Dynamic Treatment Assignment – The Consequences for Evaluations Using Observational Data, IZA Discussion Paper No. 1062. Fredriksson, Peter, and Bertil Holmlund (2001), Optimal Unemployment Insurance in Search Equilibrium, Journal of Labor Economics 19, 370-399. Fredriksson, Peter, and Bertil Holmlund (2005a), Optimal Unemployment Insurance Design: Time Limits, Monitoring, or Workfare? Revised version of Working Paper 2003:17, Department of Economics, Uppsala University. Fredriksson, Peter, and Bertil Holmlund (2005b), Improving Incentives in Unemployment Insurance: A Review of Recent Research, forthcoming in: Journal of Economic Surveys. Freiburghaus, Dieter (1987), Präventivmassnahmen gegen die Arbeitslosigkeit in der Schweiz. Methoden der Wirkungsanalyse und erste Ergebnisse. Bern/Stuttgart: Haupt.

176

Frölich, Markus (2004), Programme evaluation with multiple treatments, Journal of Economic Surveys, 18 (2), 113-224. Frölich, Markus, and Michael Lechner (2004), Regional treatment intensity as an instrument for the evaluation of labour market policy, CEPR Discussion Paper No. 4304. Frölich, Markus, Michael Lechner, and Heidi Steiger (2003), Statistically Assisted Programme Selection – International Experiences and Potential Benefits for Switzerland, Swiss Journal of Economics and Statistics, 139, 311-331. Gast, Jonathan, Michael Lechner, Heidi Steiger (2004), Swiss Unemployment Insurance Micro Data, Schmollers Jahrbuch, 1/2004, European Datawatch. Gerfin, Michael, and Michael Lechner (2002); Microeconometric Evaluation of the Active Labour Market Policy in Switzerland, Economic Journal, 112 (482), 854-893 Gerfin, Michael, Michael Lechner, and Heidi Steiger (2005), Does subsidised temporary employment get the unemployed back to work? An econometric analysis of two different schemes, Labour Economics, 12 (6), 727-838. Hahn, Jinyong, Petra Todd, and Wilbert Van der Klaauw (2001), Identification and estimation of treatment effects with a regression-discontinuity design, Econometrica 69 (1), 201-209. Heckman, James J., Robert LaLonde and Jeffrey Smith (1999), The Economics and Econometrics of Active Labor Market Programs, in O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics, Volume 3A. Amsterdam: North-Holland: 1865-2097. Hilbert, Christoph (2004), Performanzmessung und Anreize in der regionalen Arbeitsvermittlung: Der Schweizer Ansatz und eine Modellrechnung für Deutschland, Discussion Paper, Wissenschaftszentrum Berlin für Sozialforschung (WZB): Holland, Paul W. (1986), Statistics and causal inference, Journal of the American Statistical Association, 81 (396), 945-970. Holmlund, Bertil (1998), Unemployment Insurance in Theory and Practice, Scandinavian Journal of Economics, 100 (1), 113-141. Hug, Peter (1986), Die Entstehung der Massnahmen zur Förderung der Arbeitsaufnahme von Arbeitslosen. Ein Beitrag zur Geschichte der Arbeitsmarktpolitik in der Schweiz, Seminararbeit in Wirtschafts- und Sozialgeschichte an der Universität Bern, mimeo.

177

Hunold, Claude (1998), Qualitätsbeurteilung der Regionalen Arbeitsvermittlungszentren (RAV) und der arbeitsmarktlichen Massnahmen. Telefonische Repräsentativbefragung von Stellensuchenden und Unternehmungen. Publicom und M.I.S. Trend. Hunt, Jennifer (1995), The Effect of Unemployment Compensation on Unemployment Duration in Germany, Journal of Labor Economics, 13 (1), 88-120. ILO (2003), Active labour market policies, Committee on Employment and Social Policy, 288th Session, Geneva, November 2003. Imboden, Carlo, Kurt Baumann, Marcel Egger, Carlos Lenz (1999), Leistungsauftrag und Anreizmechanismen für die RAV. In: BWA Schriftenreihe Beiträge zur Arbeitsmarktpolitik (Nr. 17). Katz, Lawrence F. and Meyer, Bruce D. (1990) The impact of potential duration of unemployment benefits on the duration of unemployment, Journal of Public Economics, 41, 45– 72. Lalive, Rafael (2001), The duration of unemployment: institutions, labor market programs, and social interactions, Dissertation, University of Zurich. Lalive, Rafael, and Josef Zweimüller (2002), Benefit entitlement and unemployment duration: the role of policy endogeneity, CEPR Discussion Papers: 3363. Lalive, Rafael, Jan van Ours, and Josef Zweimüller (2001), The Impact of Active Labor Market Programs on the Duration of Unemployment, IEW Working Paper No. 41, University of Zurich. Lalive, Rafael, Jan van Ours, and Josef Zweimüller (2002), The Effect of Benefit Sanctions on the Duration of Unemployment, IZA Discussion Papers 469, Institute for the Study of Labor (IZA). Lalive, Rafael, Jan van Ours, and Josef Zweimüller (2005), How Changes in Financial Incentives Affect the Duration of Unemployment, CEPR Discussion Paper Nol 4986. Lechner, Michael (2001), A note on the common support problem in applied evaluation studies, Discussion Paper 2001-01, Department of Economics, University of St. Gallen Lechner, Michael (2002a), Program Heterogeneity and Propensity Score Matching: An Application to the Evaluation of Active Labor Market Policies, The Review of Economics and Statistics, 84 (2), 205-220.

178

Lechner, Michael (2002b), Some practical issues in the evaluation of heterogeneous labour market programmes by matching methods, Journal of the Royal Statistical Society - Series A, 165/1, 59-82. Lechner, Michael (2004), Sequential Matching Estimation of Dynamic Causal Models, Discussion paper, No. 2004-6, Department of Economics, University of St. Gallen. Lechner, Michael, Jeffrey Smith (2006), What is the value added by caseworkers? forthcoming in Labour Economics, 2006. Lenz, Carlos, Marcel Egger, Boris Zürcher, (2001), Die Vereinbarung zwischen Bund und Kantonen über den Vollzug des Arbeitslosenversicherungsgesetzes in der Schweiz. Tage der amtlichen Statistik, Neuchâtel, Bundesamt für Statistik. Leuven, Edwin, and Barbara Sianesi (2003), PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing, http://ideas.repec.org/c/boc/bocode/s432001.html. Version 1.2.3. Li Yunfei Paul, Propert Kathleen J., Rosenbaum Paul R. (2001), Optimal balanced risk set matching, Journal of the American Statistical Association 96:870-882. Ljungqvist, Lars and Thomas J. Sargent (1998), The European Unemployment Dilemma, Journal of Political Economy, 106 (3) pp. 514-550. Magnin, Chantal (2005), Beratung und Kontrolle: Widersprüche in der staatlichen Bearbeitung von Arbeitslosigkeit, Zürich: Seismo. Martin, John P. and David Grubb (2001), What works and for whom: A Review of OECD countries’ experiences with active labour market policies, Swedish Economic Policy Review, 8 (2), 9-56. Martinovits-Wiesendanger, Alex and Ganzaroli Dennis (2000), Evaluationsprogramm Arbeitsmarktpolitik: Wirkungsanalyse der Gastgewerbe-/Winwordkursen und Einsatzprogrammen, seco Publikation, Beiträge zur Arbeitsmarktpolitik Nr. 5, Bern. Meyer, Bruce D. (1990) Unemployment insurance and unemployment spells, Econometrica, 58(4), 757–82. Miquel, Ruth and Michael Lechner (2005), Identification of the Effects of Dynamic Treatments by Sequential Conditional Independence Assumptions, Discussion Paper, August 2005.

179

Mortensen, Dale (1977), Unemployment Insurance and Job Search Decisions, Industrial and Labor Relations Review 30, 505-517. Oberholzer-Gee, Felix (2005), Nonemployment Stigma as Rational Herding: A Field Experiment, forthcoming in Journal of Economic Behavior and Organization. OECD (1993), Employment Outlook, Paris. OECD (2002), Benefits and wages: OECD indicators, Paris. Prey, Hedwig (2000), Evaluationsprogramm Arbeitsmarktpolitik : Wirkungsanalyse zu Weiterbildungs- und Umschulungsmassnahmen in St. Gallen, seco Publikation, Beiträge zur Arbeitsmarktpolitik Nr. 3, Bern. Puhani, Patrick (2000), Poland on the Dole – The Effect of Reducing the Unemployment Benefit Entitlement Period during Transition, Journal of Population Economics, 13: 35-44. Puhani, Patrick (2003), Relative Demand Shocks and Relative Wage Rigidities During the Rise and Fall of Swiss Unemployment, Kyklos 56: 541-562. Røed, Knut and Tao Zhang (2003), Does unemployment compensation affect unemployment duration? The Economic Journal, 113 (January), 190-206. Rubin, Donald B. (1974), Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies, Journal of Educational Psychology, 66, 688-701. Rubin, Donald B. (1980), Discussion of "Randomization Analysis of Experimental Data: The Fisher Randomization Test" by D. Basu, Journal of the American Statistical Association, 75, 591-593. Rubin, Donald B. (1980), Bias Reduction Using Mahalanobis-Metric Matching, Biometrics, 36, 293-298. Saint-Paul, Gilles (1995), Some political aspects of unemployment, European Economic Review, 39(3-4), 575-582. Schmid, Hans und Eckehard F. Rosenbaum (1995), Arbeitslosigkeit und Arbeitslosenversicherung aus ökonomischer Sicht, Bern: Haupt. Schmid, Hans (1998): Arbeitslosigkeit in der Schweiz des 20. Jahrhunderts, in: T. Geiser, H. Schmid, and E. Walter-Busch (eds.): Arbeit in der Schweiz des 20. Jahrhunderts, Bern: Verlag Paul Haupt.

180

Sheldon, George (2000), Evaluationsprogramm Arbeitsmarktpolitik : Die Effizienz der öffentlichen Arbeitsvermittlung, seco Publikation, Beiträge zur Arbeitsmarktpolitik Nr. 4, Bern, seco. Sheldon, George (1989), Die Dynamik der Arbeitslosigkeit in der Schweiz, Bern: Haupt. Sianesi, Barbara (2004), An evaluation of the Swedish system of active labour market programmes in the 1990s, Review of Economics and Statistics, 86 (1), 133-155. Stancanelli, Elena G. (1999), Unemployment duration and the duration of entitlement to unemployment benefits: an empirical study for Britain, Applied Economics, 31, 1043-1051. Vassiliev, Anatoli (2003), Studies in Efficiency Evaluation of Labour Market Policies, doctoral thesis, University of Geneva. Winter-Ebmer, Rudolf (1998), Potential Unemployment Benefit Duration and Spell Length: Lessons from a Quasi-Experiment in Austria, Oxford Bulletin of Economics and Statistics, 60, 33-45. Wolter, Stefan C., Weber, Bernhard A. (1999). Skilling the Unskilled – A Question of Incentives? International Journal of Manpower, Vol. 20, No. 3, 254-269.

181

Curriculum Vitae

Heidi Maria Steiger

Born in Altstätten SG, May 3, 1974 Nationality:

Swiss

Two children: Manu Thomas Parikshit (June 4, 2003) Rhea Martha (January 18, 2005)

1994

Matura Typus B, Kantonsschule Heerbrugg

2000

Degree in Economics (lic. oec.) with focus on economics and econometrics, University of St. Gallen

2002

Program for beginning doctoral students, Study Center Gerzensee (Macroeconomics, Microeconomics, Econometrics)

2000-2005

Assistant to Prof. Dr. Michael Lechner, University of St. Gallen

2007

PhD in Economics, University of St. Gallen

Suggest Documents