The ICP is embarking on the next benchmark comparison work in 2003 with the ..... Programme (ICP) and on the use of Purchasing Power Parities (PPPs) in the Study ... Discussion with NSO, Ministry of Commerce (Dept. of Internal Trade), ..... Engel law which states that households that spend the same proportion of total.
Construction of Purchasing Power Parities (PPPs) for the Study of Global and Regional Poverty Report on A Pilot Project for the Statistics Division of the ESCAP
D.S. Prasada Rao Centre for Efficiency and Productivity Analysis School of Economics University of New England Armidale, Australia
This is a revised version of the Report prepared for discussion at the Expert Group Meeting on the International Comparison Program at the Economic and Social Commission for the Asia and Pacific (ESCAP) during 27 Feb-1 March, 2002.
April, 2002
1 CONTENTS
Section 1: Introduction 1.1 Background 1.2 Statement of the Problem 1.3 Outline of the Report Section 2: PPPs and Global and Regional Poverty Measurement 2.1 The Current Methodology for Global Poverty Measurement 2.2 A critique of the Methodology 2.3 Major issues Section 3: The ESCAP Pilot Project 3.1 Principal objectives 3.2 Countries covered in the Project 3.3 Terms of Reference and the Timeframe 3.4 Agenda items for discussion Section 4: Country Practices in Poverty Measurement 4.1 Determination of Poverty Lines 4.2 Adjusting Poverty Lines for Temporal and Spatial price differences 4.3 Use of Household Expenditure Surveys 4.4 A brief summary Section 5: PPPs for regional and global poverty measurement 5.1 Spatial comparisons of Prices and PPPs - A Digression 5.2 Aggregation methods for Spatial comparisons 5.3 Suggested methods for Poverty PPPs 5.3.1 Adjustment Factors for use with existing PPPs 5.3.2 Estimation of poverty PPPs using CPI prices and HES weights 5.3.3 Estimation of PPPs using only HES data 5.3.4 Estimation of PPPs using nutritional requirements 5.3.5 An assessment of relative merits of these methods 5.4 Conclusions Section 6: Summary and Recommendations 6.1 Summary of discussions 6.2 Concluding Remarks and Recommendations Report on the Deliberations of the ESCAP Expert Group on ICP, February, 2002. REFERENCES
2 SECTION 1: INTRODUCTION 1.1 Background The present project on the purchasing power parities for the global measurement of poverty is cast within the context of two major developments in the global scene. The first context is the need for reliable and internationally comparable estimates of global poverty in an increasingly globalized world. The second context relevant to the pilot project is the timing of the next Phase of the International Comparison Program (ICP). What are the principal benefits of globalisation to the poorer countries of the world? Has globalisation resulted in a significant reduction in the number of poor in the world? Answers to these questions require an appropriate methodology for the estimation of the number of global poor and for studying the nature and extent of poverty. Within the general parameters of such endeavours, there is also an increasing need to assess the global inequality and the growth performance of some of the most populous nations in the world. In addition to the research agenda on global poverty resulting from the growing debate on globalisation, several international organisations including the World Bank have taken a lead role in poverty alleviation programs in the world. These organisations have a mandate to eradicate poverty and create a world without poor. As a part of their ongoing activities in this area, these organisations routinely obtain estimates of the number of poor in the world. The Human Development Report provides estimates of global poor using the one and two dollar a day poverty lines. The ICP is embarking on the next benchmark comparison work in 2003 with the World Bank taking a leading role in providing funding and coordinating the regional and global comparisons of prices and real income. The World Bank has been a major user of the purchasing power parity (PPP) estimates derived from the ICP in its work on global poverty. There has been an increasing realisation within the academic circles and international organisations that PPPs from the ICP in its current form are not the most appropriate parities that can be used in making the one or two-dollar poverty lines operational. The increasing recognition of the need for an alternative to the PPPs from the ICP for purposes of poverty measurement has led the World Bank to explore the possibility of
3 constructing PPPs for poverty measurement. The current pilot project has arisen out of the need to bring poverty-PPPs into the mainstream work of the ICP. The pilot project is quite significant in its timeliness and the potential it has to influence the methodology underlying the estimation of the number of poor in the world. 1.2 Statement of the Problem In its simplest form, the main issue involved here is the relevance of the PPPs from the ICP in the context of global comparisons. Suppose the poverty line for global poverty is set, for instance, at one-dollar a day. How can this poverty line be used in practice? The one-dollar amount needs to be converted into respective national currency units prior to the actual estimation of poor in different countries. In order to obtain conversion factor, it is important to measure the relative purchasing power of the dollar with respect to various national currency units. The problem is to identify an appropriate basket of goods and services that needs to be priced in different countries in order to establish the parities. The current use of ICP PPPs for purposes of converting the one-dollar poverty line implies that prices of goods and services used for purposes of ICP are quite adequate and relevant for purposes of poverty measurement. There have been a number of papers (see for example, Deaton, 2001; Rao, 2001 and Heston, 2001) that point towards the inadequacy of the current approach of using ICP PPPs. If the purchasing power parities are considered inappropriate, then how does one derive a set of PPPs that are more suited to the needs of global poverty measurement? The current pilot project examines the issues involved and the current practices of selected countries with respect to poverty measurement and price level comparisons. The principal objective of the project is to articulate a methodology for the construction of purchasing power parities for global poverty measurement. 1.2 Outline of the Report Section 2 of this Report elaborates on the use of PPPs for the compilation of global and regional poverty statistics. A critical appraisal of the World Bank methodology for global poverty measurement is provided. Section 3 briefly sets the scene for the Pilot Project outlining the general terms of reference and the principal objectives. The main findings from the visits to the selected countries with respect to the determination of poverty lines, status of household expenditure surveys and the
4 estimation of poverty incidence are summarised in Section 4. Section 5 outlines a number of alternative approaches to the compilation of PPPs for poverty measurement. These approaches are all designed to provide PPPs that are better suited to the needs than the PPPs from the International Comparison Programme. A brief summary of the findings and recommendations are provided in the last Section.
5 SECTION 2: PPPS AND GLOBAL AND REGIONAL POVERTY MEASUREMENT This Section outlines the role of purchasing power parities in measuring regional and global poverty. A simple approach to regional and global poverty measurement is to first estimate incidence of poverty in different countries and then progressively aggregate them to yield estimates for various regions and country groupings. Though this approach is simple, in an operational sense poverty estimates from different countries are not easily comparable. Section 4 of this Report describes the concepts and methods employed in different countries and highlights the differences in the methodologies that render poverty estimates from different countries incomparable. Such differences suggest the need for a more unified approach to global poverty measurement. The World Bank employs such a methodology based on the concept of the one-dollar a day or two-dollars a day as poverty lines. This methodology is described in Section 2.1 followed by a critique of the methodology in Section 2.2. A list of issues are identified and listed in Section 2.3. 2.1 The Current Methodology for Global Poverty Measurement The current approach to global and regional poverty measurement is based on a simple and unifying concept of poverty line that is set at an arbitrary level of $1-a-day or $2-a-day. The estimated number of poor in the world for the year 1993 is now known to be 1.3 billion. These figures are then disaggregated into regional and subregional totals. While the poverty line used appears to be somewhat simplistic and that there is some likelihood of it being quite remote from what different countries may consider to be an adequate poverty threshold level, these estimates are still of considerable value for purposes of mobilising global attention to the problem of poverty alleviation. It is quite possible to create an impression about how poor are poor when $1-a-day is considered to be poverty line, in a developed world the purchasing power of a dollar is so limited that these estimates are bound to impress the general populous in the developed world. During the course of the country visits and discussions held with various national statisticians involved in poverty measurement, it was found that the respective national poverty lines which are themselves constructed after considerable research and effort are themselves not that far off from this somewhat simple notion of $1-a-day poverty line.
6 Given that the poverty lines of $1 or $2-a-day are acceptable as reasonable proxies for the poverty threshold, the current methodology used by the World Bank since the 1990 World Development Report (World Bank, 2000) may be described as below. The very first and crucial step involved in this exercise is one of converting dollars into national currency units. The World Bank uses the latest available Purchasing Power Parities (PPPs) for the "consumption" sub-aggregate of the gross domestic product as the starting point. For example, if poverty estimates are being compiled for the year 2001 the Bank starts with the PPPs for the benchmark year that is closest to the year under consideration. For example if 1993 is the latest benchmark for which PPPs for consumption are available, then these PPPs are used in converting the dollar poverty line into a national currency unit. Thus a poverty line for 1993 is obtained. This poverty line in local currency unit is then updated to the year 2001 using the best available consumer price index numbers. The poverty line then can be considered is a true conversion of the $1 dollar a day into local currency unit for purposes of estimating poverty incidence. The poverty line in national currency units is then used in conjunction with the distribution of income or consumption in finding estimates of the number of poor within each country. In practice detailed household expenditure survey data are utilised in this process of identifying poor. The current methodology used by the Bank in producing estimates for publication in the World Development Indicators has several advantages. The first and foremost is the consistency of poverty line implicit in the use of $1-a-day. Such consistency ensures international comparability of the poverty estimates and such comparisons can be used in assessing the effectiveness of poverty alleviation policies of national and international organisations with a mandate on poverty reductio n. The second advantage is attributable to the use of PPPs in the place of exchange rates. It is now well recognised that exchange rates of currencies of most of the developing countries do not accurately reflect the purchasing power of the respective currencies. In fact there is considerable volatility in the exchange rates resulting from large capital flows in and out of countries and also from various domestic political instabilities. For example, a number of currencies in the Asia region lost value during the Asian crisis and the use of exchange rates for purposes of converting $1-a-day poverty line into a
7 local currency, such as the Indonesian Rupiah, would have the effect of a serious upward movement of poverty incidence almost overnight. 2.2 A critique of the Methodology While the current World Bank methodology is based on sound principles and has several practical advantages discussed above, the Bank methodology suffers from a number of shortcomings. The most eloquent critique of this methodology is presented in Deaton (2001) and also discussed in Rao (2001) and Heston (2001). The first principal criticism concerns the use of PPPs, for the private consumption aggregate, for purposes of currency conversion. The PPPs are supposed to reflect the purchasing power of currency in different countries, and prices of consumption goods and services in particular. The question to be raised here is whether or not the PPPs from the ICP exercise accurately reflect the prices paid for goods and services that are typ ically consumed by poor in different countries. Answer to this question is an emphatic no. To appreciate this, it is necessary to examine the main steps and procedures employed in the ICP. According to the procedures employed, the ICP uses a list products (goods and services) which are developed in consultation with experts from different countries (usually within a region), and a list of the main characteristics of these products are prepared to undertaking price surveys. This process is a consequence of one of the main principles, known as the "identical product" rule, enunciated within the Handbook of ICP (United Nations, 1992). This rule is used as a way of dealing with possible quality variations across countries. Since the ICP lists are designed to achieve comparability across countries while maintaining constant quality, usually these lists have little overlap with the goods and services that belong to the consumption of the poor. Given the basic framework employed for purposes of ICP, the PPPs for consumption can at best reflect differences in prices of goods and services that belong to the consumption baskets of upper or upper- middle class households in these countries. It is not at all clear whether the price levels reflected in these PPPs are relevant for purposes of deriving poverty lines in different countries, and it is difficult to speculate on the direction of bias. There are very few studies that can help identify this direction. Decoster (1999) has examined, in depth, the question as to whether poor are
8 likely to pay more for goods and services (for identical products) than higher income households. Rao (2000) also examined this question using micro- level studies conducted in a South Indian Village and concluded that generally poorer households tend to pay higher prices. The main reasons offered in support of this observation is the fact that poorer households often purchase smaller quantities and also have less flexibility about the timing of purchase and where they buy their goods and services. Higher income households have more scope to plan their purchases to take advantage of lower seasonal prices and discounts offered for bulk purchases. Biru (1998) studied this aspect in Zambia and found that incidence of poverty increases when PPPs are made to reflect prices of poor. Another closely related criticism concerns the choice of weights used in aggregating price data within the ICP to derive PPPs for consumption. The weights used are normally those from the national accounts and usually represent average for the country as a whole. Are these weights of any relevance when it comes to aggregating price data for the poor? Again, answer to this is in the negative. Firstly, these weights are an average and, therefore, differ significantly from those for the poor. Secondly, the national accounts consumption aggregates have the tendency to differ significantly from the household consumption. These are due to the conceptual framework employed in the construction of the national accounts weights. Fore example, the national accounts consumption includes consumption of non-profit and own-account enterprises as well as imputed rental of owner occupied dwellings. It is also due to the fact that the consumption aggregates are sometimes derived as residuals, and, therefore, include errors and omissions. Based on these criticisms concerning the weights employed within the ICP framework, the PPPs for consumption do not accurately relate to the poorer segments of the population. In addition to these fundamental objections to the use of PPPs from the ICP, there are also issues relating to the problem of updating ICP PPPs for a given benchmark year, say 1993, to the current period, say 2001, under consideration. The use of consumer price index numbers for updating poverty lines within each country is a practice that requires further scrutiny. For example which CPI is the most appropriate? Should regional differences in prices be accounted for in applying these poverty lines?
9 Another related problem is one of consistency of PPPs in the benchmark comparisons. For example, suppose the 1993 ICP comparisons are used as the basis for converting $1-a-day poverty lines using the procedures described above. The 1993 PPPs forms the basis until a new benchmark is available. Suppose the next benchmark PPPs available are for the year 2003. When the PPPs from the new benchmark are employed, there could be a major shift in the relativities implicit in the PPPs. Such shifts may result in abnormal or counter-intuitive changes in global poverty, observed rather abruptly. This is a problem that has been long recognised, but no tangible solutions are available at the present time. 2.3 Major Issues The current research project is designed to study the issues raised above. In particular, the main question to be addressed as a part of the project is to examine the feasibility of constructing an alternative set of purchasing power parities that can accurately reflect differences in price levels across countries. Within this broad framework, the project is also to investigate the current practices in poverty measurement with particular focus on the nature of price data used in poverty line determination and also subsequent adjustments to update these poverty lines over time and across different regions within a given country. The question of weights which reflect the relative importance attached to different commodities and commodity groupings and identifying the principal source of such information is an issue to be investigated as a part of this project.
10 SECTION 3: THE ESCAP PILOT PROJECT The current pilot project initiated by the Statistics Division is commissioned as a forerunner to the next round of the International Comparison Programme which is expected to start in 2003 benchmark year. The World Bank and other users of PPP for the study of global inequality and poverty recognise the limitations of the ICP PPPs described in the previous Section. The present project is expected to provide a set of recommendations about the feasibility of modifying PPP measure in order to make it more suitable for the estimation of incidence of poverty in the world. 3.1 Principal Objectives The main objectives of the Pilot Project are: • • • • •
To provide an assessment of country practices with respect to poverty measurement with special focus on the methods and concepts used in the determination of poverty lines; To examine the methodologies used in adjusting poverty lines over time for temporal price changes and across regions to account for spatial price differences; To study the general structure of the household expenditure surveys and the underlying questionnaires to determine the feasibility of identifying the consumption patterns of the poor; To examine the feasibility of constructing PPPs for poverty measurement to be used in conjunction with the World Bank methodology of the $1 or $2-a-day poverty lines; and Finally to provide a set of alternative methods and procedures for the construction of PPPs using the existing price and expenditure data from the survey framework that is currently in existence in these countries.
3.2 Countries covered in the Project Within a limited timeframe it is impossible to examine the statistical systems and practices in all the countries in the Asia-Pacific region. The current project is limited to a pilot study covering only four countries. The countries included in the study are some of the major countries in the region both with respect to the size of the respective populations as well as the size of their economies. Another important consideration in the choice of the countries for the project is that these countries need to have well-established practices with respect to poverty measurement. Household expenditure surveys and the existence of institutional framework for data collection and subsequent dissemination of the survey data were also considered essential in the process of selection. The focus could then be exclusively on PPP construction rather than other issues relating to statistical infrastructure.
11 The countries included in the project are: India, Indonesia, the Philippines and Thailand. These countries satisfy all the basic requirements in terms of their expertise and the infrastructure availability for poverty measurement. These countries also complement another set of four countries selected for inclusion in a pilot project that is parallel to the current project. Countries included in that project are China, Fiji, Hong Kong and Malaysia. 3.3 Terms of Reference and the timeframe The following are the Terms of Reference for Consultants to Conduct Research on Integration of Work on Consumer Price Indices (CPI) and International Comparison Programme (ICP) and on the use of Purchasing Power Parities (PPPs) in the Study of Poverty A.
Background and General Directions: The pilot project grew out of recognition
that a) accounting for differences in consumption patterns and relative prices across income groups and regions is a first and critical step toward integrating purchasing power differences into strategic poverty alleviation policies; and b) compiling stylised facts pertinent to the impacts of poverty alleviation policies and adjustment programmes on different income groups requires comparable cross-country data. Objectives: The research proposal has two objectives. The primary objective is to develop an appropriate survey framework for measuring differences in purchasing power of income across income-groups and regions (subnational areas) that could be replicated in a large number of countries. In particular, the proposal aims, as its long term objective, to regularize poverty data collection by integrating income- group and region-specific price data collection into the larger context of System of National Accounts (SNA) implementation by national statistical systems in the ESCAP region. As such, the proposal is a step towards enhancing statistical capacity in the ESCAP region. A secondary objective is to begin the integration of the International Comparison Programme (ICP) techniques with work on cross-country poverty comparisons, with a view to develop an appropriate method for constructing "PPPs for the poor". To this end, the proposal's ultimate objective is to bring poverty-specific PPPs into the mainstream of ICP.
12 Under the general supervision of the Director of the Statistics Division and the Chief of the Statistics Development Section, the Consultant will undertake a study on the use of Purchasing Power Parities (PPPs) in the Study of Poverty in selected countries. The consultant will conduct a study to help determine the methodology necessary to obtain poverty focused PPPs. He will prepare a report to document his findings and recommendations. The report will examine the current practices in determining poverty lines and price indexes for the poor, and provide recommendations on how current practices and the ICP methodology might be integrated. He/she will:
(a) visit and hold discussions at national statistical offices and other relevant agencies in the following selected countries: India, Indonesia, Philippines and Thailand; b) in close collaboration with NSOs, develop a standard poverty basket and propose alternative methodologies to estimate the cost of such a basket in different countries; (c) develop a survey framework for collecting poverty focused prices by regions and income groups; (d) identify unresolved technical and methodological issues and flag problem areas on which further research should be focused; (e) establish objective criteria to establish a balance between exercising expert judgment and applying scientific survey framework; and (f) jointly with the other consultant (see B) assist the secretariat in servicing an expert group meeting to discuss the research findings;
The research is estimated to require three work- months. Period of the contract:
Three work- months between 4 November 2001 and 28 February 2002.
Work Timetable (tentative) 1. Discussions at NSO and NSCB in Manila, Philippines.5 & 6 November 2001 2. Hold meetings with ESCAP staff and the other consultant. Morning of 7 November 2001 3. Discussion with NSO, Ministry of Commerce (Dept. of Internal Trade), Thailand. 7, 8 & 9 November 2001 4. Consultations with ESCAP staff and other consultant at ESCAP. 9 November 2001 5. Discussions at CSO and other agencies in New Delhi, India. 12 – 16 November 2001 6. Discussions at BPS, Jakarta and other agencies. 19 – 23 November 2001 7. Mission report and discussions with ESCAP staff and the other consultant, prepare detailed outline of paper on survey framework and methodology issues, give a short talk to the Working Group of Statistical Experts about the research. 26 – 30 November 2001 8. Return home and prepare paper. 9. Prepare for the Expert Group Meeting, and finalize paper. 11 – 22 February 2002
13
During the implementation of the project, it became necessary to make further changes to the dates proposed in the original terms of reference. Dates for the Expert Group Meeting are now set for the last few days of February. 3.4 Itinerary for the project The following is a schedule of meetings held in the countries included in the pilot project. Thailand 7.11.2001 7.11.2001 8.11.2001 9.11.2001 9.11.2001
Meetings with NESDB Poverty Measurement Meetings with the NSO Household Exp. Surveys Meetings with Dept of Internal Trade Consumer Price Indices Meetings with the Director and Staff, Statistics Division, ESCAP Meetings with Dr. Sultan Ahmad, Consultant on ICP-CPI Project
India 12.11.2001 12.11.2001 12.11.2001
Meetings with the CSO Meetings with Planning Commission Meetings with NSSO
National Accounts Poverty Measurement National Samp. Surveys
13.11.2001 13.11.2001
Meetings with the CSO Meetings with NSSO
Consumer Price Indices National Samp. Surveys
15.11.2001
Meetings with Nat. Informatics Centre
16.11.2001 16.11.2001
Meetings with NSSO Meetings with the CSO
Computerisation of Surveys and use of computers in data collection Exp. Survey tabulations National Accounts ICP Participation
Indonesia 19.11.2001 19.11.2001 19.11.2001
Meeting with the Director General Meeting at BAPPENAS Meeting at BPS
Briefing Meeting Research on Poverty Poverty Measurement
20.11.2001 20.11.2001 20.11.2001 20.11.2001
Meetings at BPS Bank of Indonesia BPS BPS – meeting with the Director General
Consumer Price Index Research on Poverty Household Exp. Surveys Debriefing
Technical Committee on Prices (NSCB) -National Statistical Coordination Board
The Philippine Statistical System
Philippines 22.11.2001
14 22.11.2001
NSCB
22.11.2001 22.11.2001
Poverty measurement National Accounts: Exp. Side Lecture By Rao: ICP and PPPs for the Study of Global Poverty Visit to the National Statistical Information Centre (NSIC)
23.11.2001
Meeting with the Administrator – The Philippines Statistical Office
23.11.2001
Household Statistics Department
23.11.2001 23.11.2001
Industry & Trade Statistics Dept. ITSD and EIID
FIES -Family Income and Expenditure Survey Consumer Price Index ICP
The consultant gratefully acknowledges the help received from various statistical organisations, and to all the participants in various meetings for their time and discussions. The success of the pilot project greatly depended upon the enthusiastic cooperation of respective officers in charge of poverty measurement, household expenditure surveys and the measurement of the consumer price index. 3.5 Agenda items for discussion during visits to Participating Countries The following is the document that was mailed to the national statistical organisations outlining the main objectives of the project along with a list of agenda items for discussion during the consultant's visit. The project has emerged from the recent debate on the suitability of PPPs from the International Comparison Program (ICP) for the purpose of studying global poverty. Apart from the fact that the ICP PPPs cover all the goods and services included in the computation of gross domestic product, there is a recognition that accounting for differences in consumption patterns and relative prices across income groups and regions is a first and critical step towards integration of the PPP concepts into strategic poverty alleviation policies. In order to be able to develop a suitable methodology for the computation of PPPs for poverty measurement and analysis, it is necessary to identify the principal elements of such an exercise after a detailed examination of the practices of National Statistical Offices of selected countries. Under the general supervision of the Director of the Statistics Division and the Chief of the Statistics Development Section, a pilot project is being undertaken by a consultant of ESCAP. To facilitate a proper appraisal of the current practices, the consultant will be visiting the NSOs of these countries and collecting information on poverty measurement in these countries. The consultant will
15 have discussions with the heads of sections who are primarily responsible for the measurement of poverty in these countries. The following are the main agenda items for discussion during the visit of the consultant to each of the NSOs. These items are directly relevant to the problem of constructing PPPs for the study of global and regional poverty. 1. Poverty Measurement 1.1 Sections in the NSO Responsible for Poverty Measurement 1.2 Methodology used in constructing national Poverty Line 1.3 Type and Source of Price Data used in Poverty Line determination 1.4 National versus Regional Poverty Measurement 1.4.1 Adjusting national Poverty Line for regional price differences 1.4.2 Price Data used for price level adjustment 1.5 Estimation of poverty 1.5.1 Head count ratio 1.5.2 Poverty Gap 1.5.3 Any other measures 1.6 Decomposition of poverty by regions 1.6.1 Rural versus urban 1.6.2 By geographical regions 1.7 Production and Dissemination of poverty estimates 1.7.1 Frequency 1.7.2 Modes of dissemination 1.7.3 Part of the NSO responsible for these activities 1.8 Policy Formulation and Perspectives 2. Consumer Price Index 2.1 Organisational structure for CPI computation 2.2 Survey Framework for the CPI 2.2.1 Outlet Selection 2.2.2 Item specifications Treatment of quality change 2.2.3 Coverage over regions Geographical regions Rural versus urban regions 2.3 Computation of the CPI 2.3.1 Elementary indices 2.3.2 CPI headings 2.3.3 Index number formula 2.3.4 Weights for CPI calculations 2.4 CPI for different regions 2.4.1 Data related issues 2.4.2 Methods 2.4.3 Inter-regional comparisons of consumer prices 2.5 Production and Dissemination 2.6 Use of CPI in the adjustment of Poverty Lines over time 2.6.1 Sections responsible
16 2.6.2 Procedures actually employed 2.6.3 Actual price data or CPI results used in the adjustments 3. Household Expenditure Surveys (HES) 3.1 Division of the NSO responsible 3.2 Survey Design and details 3.2.1 Design description 3.2.2 Sample size and distribution 3.2.3 Weights 3.3 Questionnaire for the HES 3.4 Frequency of HES 3.5 Household consumption details 3.5.1 Expenditure 3.5.2 Quantities 3.5.3 Treatment of own consumption 3.5.4 Coverage of services 3.6 Household characteristics 3.6.1 Regional 3.6.2 Other (employment etc) 3.7 Dissemination of HES data 3.7.1 Summary reports 3.7.2 Availability of Unit Record Data 3.8 Use of HES data in the study of poverty 3.8.1 Sections of NSO responsible for the work 3.8.2 Methods employed 3.8.3 Publication of results 3.9 Income Distribution Surveys and the HES 3.9.1 Are there specific income distribution surveys? 3.9.2 NSO Sections responsible for income distribution work 3.9.3 Frequency of data collection 3.9.4 Compilation and publication of results 3.9.5 Income distributions and poverty measurement 4. NSO participation in the ICP 4.1 Sections dealing with NSO participation in the ICP 4.2 Data collection procedures 4.2.1 Specia l surveys 4.2.2 Use of CPI data 4.2.3 Quality issues 4.2.4 Representativeness of ICP commodity lists 4.3 Weights for ICP work 4.3.1 HES Sources 4.3.2 National Income Sources 4.4 Liaison with ESCAP on matters relating to the ICP participation 4.4.1 Administrative responsibility 4.4.2 Technical Group within the NSO The author of this report is greatly indebted to the enthusiastic participation of various officials in each of the countries included in the project. These officials were very
17 generous with their time and ideas during a free exchange of ideas and discussion of problems involved with the ICP and PPPs for poverty measurement in particular.
18 SECTION 4: COUNTRY PRACTICES IN POVERTY MEASUREMENT This section provides a brief summary of the methods and procedures followed in the course of poverty measurement in the four countries included in the current pilot project. It is very difficult to provide detailed accounts of the methods employed by various national statistical offices in a short project report. The principal objective of this section is to provide a background to some of the methods suggested in Section 5 for purposes of constructing PPPs for regional global poverty measurement. The analysis presented in this Section is based on the extensive discussions the author had with persons in national statistical offices responsible for various items listed in the previous sections. A number of publications in these countries, supplied to the author during his visit to these countries, are also used in the preparation of this summary. The current section is not meant to provide a detailed description of all methods and procedures employed in these countries. The principal objective is to find some themes underlying these procedures that may provide some hints and clues towards the feasibility of constructing purchasing power parities for global poverty measurement. An obvious route to follow is to examine closely the country practices to see if a bundle of goods and services may be identified as the mo st representative of the consumption of the poor in these countries. If price data were available for such a bundle in various countries involved, then it may be feasible to undertake international price comparisons that are purely focused on the poor. 4.1 Determination of Poverty Lines The first and the most important step in the study of poverty is the determination of the poverty line. Poverty lines are used in estimating the incidence of poverty as well as in examining the nature and severity of poverty in any given country. Poverty lines may be determined in different countries using the absolute or relative concepts. In most developing countries the absolute concept of poverty line or some variant of it is used. Even within a country, several poverty lines are usually in existence. These poverty lines refer to different geographical regions, for example rural and urban, or for different time periods or for households of different sizes and composition. There is considerable literature on the determination of poverty lines. Ravallion (1998) provides an excellent summary of some of the analytical issues and Kakwani
19 and Krongkaew (2000) provides a very good introduction to the problem of poverty line determination with details of its application to Thailand. There is considerable similarity in the approaches used in different countries in arriving at poverty lines, but there also exist subtle differences that distinguish each other. The most sophisticated, and the most recently established, procedure for determining poverty line was developed by Kakwani. A more general description with some possible applications to other Asian countries can be found in Kakwani (2001). All the current practices relate to the problem of finding the cost of obtaining a stipulated caloric intake. It is evident that caloric requirements vary from individual to individual depending upon the age-sex characteristics as well as the basic metabolic rates (BMR) which takes a value of one when the body is at rest. Values of BMR around 2.0 are indicative of high labour activities. The calorie requirement for an individual, with a given set of characteristics, can be computed using a simple system of weights based on norms that are country specific. The average nutritional requirement for a region or the country as a whole is computed through a weighted average of per-person calorie requirements, with weights reflecting the population weights given to each sampled household. Thus, the average or per capita requirements can vary across regions or over time depending upon the actual composition of the population. Once the average nutritional requirements are identified for the country as a whole or for a particular region within each country, the next step is to find the cost of attaining this nutritional requirement using data on food consumption and prices paid by households. This is a point of departure in terms of country practices. In Thailand, the procedure followed involves the estimation of average price per calorie intake. This is done by finding the total number of calories from an average basket of food consumption using a table of calorie conversion (in a region or for a sub- group of population) and simply finding the expenditure per calorie. In simple terms, the food poverty line of a household can be calculated by multiplying the per capita calorie requirement of a household by the average cost of obtaining one calorie of energy in that region. Obviously, the average calorie costs vary across regions and across different income groups. These differences in costs form the basis for stipulating different food poverty lines for different regions.
20 The food poverty lines are then used in finding the total poverty line using the famous Engel law which states that households that spend the same proportion of total expenditure on food enjoy the same standard of living. It is quite easy to construct total poverty line if the ratio of food expenditure to total expenditure is known for the poor. At this stage, there are possibilities for using simplified procedures or sophisticated econometrics involving regressions that relate energy intakes to food and total expenditures and the estimated regression equations can be used in finding the levels of expenditure that correspond to a particular calorie intake. The poverty line determination in Thailand adjusts for spatial price differences in food and non- food prices and the normal regional CPI's are used in updating poverty lines that were originally constructed for the year 1992. An important point to make here is the fact that the poverty line specification methodology described here is actually undertaken without ever really specifying a commodity bundle that can be considered typical of the consumption of the poor. The Indian approach to poverty determination is similar in terms of its use of caloric requirement, but the operational procedures used are different. Since the 1993 review of the procedures by the Expert Group on Estimation of Proportion and Number of Poor, the practice of identifying a poverty line at the national level has been discontinued. Instead, the poverty lines are first determined at the State level and are then used in estimating the number of poor within each State. These estimates provide an estimate of the total number of poor in the whole country. Using a description of the expenditure class-wise distribution of persons, the expenditure level that corresponds to the estimated total number of poor is taken as an approximate poverty line for the whole country. The Task Force had also recommended the determination of poverty lines for rural and urban areas separately. The current poverty lines used are based on poverty lines set in 1973-74, adjusted for inter-state price differentials in that year, and then adjusted for price differences over time using appropriate price indices. To work out the monetary equivalent of the caloric norms, the 28th Round NSS data for 1973-74 on household consumption both in quantitative and value terms were used. Using appropriate conversion factors, the calorie content of consumption baskets corresponding to various per capita expenditure classes were worked out.
21 Inversion liner interpolation method was applied to the data on average per capita monthly expenditure and the associated calorie content of food items in the class separately for rural and urban areas. This approach is designed to focus on the purchasing power needed to meet the specific calorie intake standard with some margin for non- food consumption needs. The methodology used in the determination of poverty lines in India does not explicitly identify the basket of goods and services that are considered typical of the consumption of the poor. In contrast the methodologies used in Indonesia and the Philippines have an explicit recognition of the commodities that enter the consumption basket. In Indonesia, the starting point is the use of 2100 calories as the minimum standard for the nutritional requirement. The construction of food poverty line is based on a list of 52 commonly used consumption items, these items were identified through the Food Basic Needs Survey (see page 40 and 124, 125 Crisis, Poverty and Human Development in Indonesia, 1998). The calorie content of the average consumption bundle is calculated for rural and urban areas separately, which are then used in finding the cost per calorie in these areas. These costs are used in arriving at the food poverty line. The approach used in Indonesia in finding the non-food component of the poverty line is more involved and sophisticated. The minimum expenditure for non- food commodities includes 27 items for urban areas and 26 for rural areas, covering expenditure on housing, clothing, education, health, transportation, durable goods and other essential goods and services. The item list for non-food items is identified using a special Survey on basic Commodity basket, 1995. The items are identified if these are considered essential by a majority of those surveyed. The average value of expenditure for each of the selected non-food items per month, for the reference population, are added up to get the minimum standard for non- food sufficiency. Once both of these components are identified, the total poverty line is taken as the sum of the two components. An examination of the process of determining the poverty line in Indonesia reveals that it is indeed possible to identify a list of food and non- food items that are considered important in the consumption basket of the poor.
22 In the Philippines, the process of determining the poverty line is further refined. While the basic idea of food and non- food component of poverty lines is retained, the methodology used in determining these is more refined. The food poverty line or threshold is determined for baskets that yield 2,000 calories and at the same time meet 100% adequacy of recommended daily allowances for protein and energy; and 80% adequacy for vitamins and other nutrients as determined by the Food and Nutrition Research Institute. At the next stage, menus consisting of daily consumption of various food items that can satisfy various nutrition requirements are constructed for the National Capital Region and also for other regions and provinces separately. An interesting feature of these menus is the identification of various consumption items as well as quantities that are deemed necessary to meet these requirements. These menus are then costed using prices from national statistical offices. The resulting costs form the "food threshold" or food poverty line. The total poverty line is the n defined as the ratio of the food poverty threshold to the ratio of food expenditure to the total basic expenditure at the average level. The basic expenditure items include items of food, clothing and footwear, housing including fuel and light, medical care, education, etc. However, expenditure on alcoholic beverages, tobacco, durable furniture and equipment, etc are excluded from the basic expenditure items. Determination of poverty lines based on regional and provincial menus (77 different menus) is a major novelty. This approach, while it is sophisticated, may still not address all the issues relating to the representativeness of the menus and availability of items included in the menus. In summary, all the countries in the project have well-established procedures for the determination of the poverty lines. David (2001) provides an exposition of the methods of compilation of poverty statistics. Continuous research is being undertaken to enhance the scope of poverty measurement in all these countries. The energy intake appears to be the very basic ingredient in the determination of the poverty lines in each of these countries. There are also differences in the approaches used in finding the food and non- food components of the poverty lines. A somewhat distinguishing feature of these practices is that none of them really identify a list of goods and services as being typical of the consumption of the poor. Indonesia seems to be an exception in that it is based on a list of food and non- food items.
23 4.2 Adjusting Poverty Lines for Temporal and Spatial Price differences As a part of this project, extensive discussions were held with those responsible for the construction of the poverty lines as well as the statisticians actively involved in the compilation and dissemination of consumer price index numbers. The main purpose was to determine if any particular price indices were considered more important in the process of updating the poverty lines. Discussions also encompassed the adjustments for spatial price differences. The first and foremost conclusion is that there are no systematic efforts made to adjust for spatial price level differences. In the case of Thailand, the poverty lines were constructed recognizing spatial price difference per calorie in 1992. Thus the 1992 may be considered as a base period where spatial price differences have been accounted for. Since then the food and non- food components are updated using movements in the consumer price index for these two components separately and for different regions and subregions of Thailand. A further study of the CPI compilation in Thailand revealed that price indices are compiled for three different groups: general consumer; low income and rural people. Approximately around 250 items (ranging from 240 to 270) are priced in rural and urban areas. Roughly 50% more non- food items than food items are priced. Detailed specification of the items priced are available from the Department of Internal Trade. Information is also available on the type of outlet from which the price quotations are obtained. The price data, for different item specifications and outlets, are then used in constructing price indices specific to regions and subregions over time. No spatial indices are constructed at the present time. There was considerable enthusiasm for instituting processes for compilation of spatial price index numbers. In the case of India, the rural and urban poverty lines in different States are undertaken using consumer price indices specific to population groups that may be considered closest. For example, the rural poverty lines are adjusted for price movements using State-wise consumer prices indices for agricultural labourers within a specific base year. The urban poverty lines are adjusted using the consumer index data for the industrial workers. In the case of consumer price index numbers for agricultural and rural labourers, the item specifications are determined reflecting the local preference and popularity of the
24 varieties, and all the relevant identifiable physical characteristics of each item is described by specifications such as variety, quality, make/brand, unit of price quotations, size dimensions, material content, etc. The sample of markets and outlets popular with the rural labour households is selected through a market survey. The type of markets and the outlets are also explicitly recognised when data are collected. These data are aggregated using standard index number formulae used in CPI computations. A similar process is used in compiling indices for industrial workers. The industrial workers are identified by sectors in which they are employed. The Consumer Price Index numbers in Indonesia are computed following all the standard procedures. Price data are collected on a regular basis from 43 cities, including all capital cities and 17 other main cities. Detailed specifications, indicating quality differences, are available from the agency. Information on the outlet specification is also available. The selection of outlets is based on standard CPI practices. However, the more interesting aspects of CPI are available through the 1996 cost of living survey which identifies the commodities with quality specifications. It appears feasible to obtain index numbers for rural areas from the Rural Terms of Trade indices. These indices use a smaller list of items (150 items compared to 249 normally). However, the rural indices are based on data collected from land owner farmers, but not including tenant farmers. In all the countries visited, a detailed list of all the items included in the CPI compilation is available along with a description of the product characteristics. These characteristics provide a measure of quality and these are, therefore, useful in adjusting price data for quality differences across countries or over time. Details of the outlets from which the prices are collected are also clearly identified. Information on outlets makes it possible to identify those prices that are collected from outlets that are used by poorer sections of the community. Outlet characteristics can also be used in adjusting prices for differences in the quality of service associated with certain outlets. The price index compilation in the Philippines is also very encouraging in terms of the applicability of the data compiled for the construction of price indices that are relevant for poor. The CPI in the Philippines uses 1994 as the base year. The coverage
25 and outlets are determined through the commodity and outlet surveys. It appears that such surveys are available for the poorest 30% of the population. Thus it is feasible to identify the commodities and outlets that are directly relevant to the poor. Prices of agricultural related products are compiled through the Bureau of Agricultural Statistics (BAS) and price data for the remaining items are compiled through the National Statistical Offices. Outlets from which prices are collected are clearly identified. Though such data are collected, they are not currently available in electronic form. Consumption patterns from the Family Income and Expenditure Survey are used in determining weights for index calculation. A close examination of the price data compiled by the Philippines National Statistical Office indicates the possibility of identifying a basket of goods and services that are relevant to the poorest 30% of the population. Discussions have indicated that there are no attempts made at this stage to construct spatial price index numbers even though Laspeyres-type indices are used in calculating spatial indices with the national capital region used as the base. In summary, while all the countries visited use standard CPI practices in terms of price compilation and index dissemination, there appears to be some valuable data on prices paid by persons and households that may be closely similar to poor households in these countries. An implication is that such price data, collected on an annual basis, can be used in conjunction with weights from household expenditure surveys in constructing spatial price indices for the adjustment of poverty lines within the countries and possibly using this data for purposes of constructing PPPs for poverty measurement on a global or regional scale. How do the national poverty lines compare with the $1-a-day poverty line? This question was raised and discussed at length during various meetings at different sections of the national statistical organizations. As described above, all these countries have well- established methods to derive poverty lines and these lines are expressed in national currency units. At the official exchange rates, it was generally found that the $1-a-day poverty lines were well above the national-poverty lines in national currencies. However, when rough estimates of purchasing power parities for the private consumption expenditure component of GDP are used the converted $1-aday lines are roughly around the national poverty lines and in some cases even lower.
26 By implication, use of $2-a-day poverty line converted into national currency units using PPPs would be significantly above the national poverty lines and, therefore, result in much higher rates of poverty incidence in these countries. These discussions and findings lend a lot of support for the use of a simple and intuitive $1-a-day poverty line which eliminates most of the variations in the methodologies employed in different countries, and as a result yield internationally comparable estimates of poverty incidence. 4.3 Use of Household Expenditure Surveys It was no surprise that the pilot project has confirmed the existence and intensive use of extensive surveys of household expenditure surveys in all the countries visited. All the four countries visited have considerable experience in conducting these surveys and all the details of the survey methodologies and summary statistical findings of these surveys are widely disseminated. Household expenditure surveys (HES), which are referred with different names in different countries, are a valuable source of data on expenditure patterns and weights that are used in the construction of the consumer price index numbers. The HES data are used in deriving the poverty lines, using food consumption data, and subsequently in obtaining estimates of incidence and extent of poverty based on the per capita expenditure distributions derived from these surveys. Within the framework for the pilot project, the HES data were examined in order to see if it is feasible to identify price and consumption data for those households which are identified as poor. If it is possible to tabulate, by household, price and quantity of each of the commodities listed in the household expenditure survey then it is feasible to analyse such data with the aim of constructing PPPs for poor. It is here, that there was mixed success. In Thailand, the HES questionnaires only ask for details of expenditure on various consumption items, and some times these items are within broad commodity groups. Even though it is possible to estimate the levels of expenditure on various commodities, no quantity data are available. Reasons for not including such data are obvious in that it is quite difficult to quantify consumption or purchases especially when households purchase small quantities usually worth a given value. For example, a household may purchase potatoes from the market for 2 Bahts without being aware
27 of the actually quantity involved. Specially trained staff may be required to collect such data that is reliable. The latest survey data are available from the 2000 round of the HES. During discussions with the National Statistical Office the possibility of recording quantity data along with expenditures was discussed. An approximate costing of modifying the existing questionnaire and the work involved in subsequent tabulation of the data collected was provided and discussed at one of the meetings. It is encouraging to see the active cooperation of the Thai NSO in setting up a system that may provide valuable information that can be used in the compilation of PPPs for poverty measurement in the future. Turning to data availability from India, the National Sample Survey Organisation (NSSO) is responsible for the conduct of the surveys. The 55th Round of the NSS for the period June 1999-2000 has been completed and results from this survey are now compiled and available. In contrast to the Thailand surveys, the NSSO surveys include the quantity of consumption as a part of the questionnaires. These surveys include information on consumption from own production as well as details on the frequency of purchase. Therefore it is possible to obtain detailed consumption data, including prices and quantities separately, either in a detailed format like the unit record data sets or in the form of a more aggregate data. The latest NSSO publication entitled, Consumption of Some Important Commodities in India, provides detailed tabulations by various States and for different income classes. It should be possible to identify those commodities that are consumed by the poor and obtain price and quantity data for these commodities. The household expenditure surveys in Indonesia are conducted under the auspices of the National Social and Economic Survey (SUSENAS) which has a format and sampling design that is consistent with international practice. Details of the methodology employed are provided in Surbakti (1997). In terms of the availability of price data that may be used in the measurement of PPPs for poor, the SUSENAS data fall in between the Thai and Indian scenarios. The SUSENAS questionnaire permits collection of detailed purchase information, including quantities, for items in the food component. However, for non- food items it is only possible to obtain expenditure data. The latest tabulated data currently available refers to 1999 Survey. Data are available in unit record format as well as more aggregated data in the form of income classes. In order to use data from SUSENAS, it may be necessary to augment price
28 data for non-food items from sources other than SUSENAS. It may be feasible to obtain data from the CPI sources to augment food price and quantity data. This strategy requires further examination, but it appears to be feasible at this time. The Philippines also has a well established survey framework under the auspices of the Family Income and Expenditure Survey (FIES) conducted by the National Statistical Office. The latest survey refers to year 2000. The Enumerator's Manual and a copy of the 2000 FIES Questionnaire have provided valuable information on the detail of the information compiled in the Philippines. As in the case of India, the FIES includes details on quantity consumed, with specific units of measurements provided in the questionnaires. Data on the value as well as quantity of purchases of various food and non- food items are collected through these questionnaires. A close examination of the FIES questionnaire suggests that the detail of data available is more extensive than the NSSO data from India. Thus Philippines has the potential to provide the most detailed data that can be used as an input into the construction of PPPs for poor. 4.4 A brief summary The material presented in this Section provides a framework within which a number of alternative strategies to construct PPPs for poverty measurement can be formulated. Discussions held with various NSOs and the published material examined thus far points towards a degree of consistency in the approach towards poverty measurement and uniformity in the availability of data. The Pilot Project has opened a number of possible avenues for compilation of price data without instituting sample surveys for the collection of price data that are more relevant to the poor. The discussion also points towards subtle variations in the approaches used and also the type of data collected through the household expenditure surveys and the consumer price index surveys for the lower income groups. A number of alternatives are explored in the next Section.
29 SECTION 5: PPPS FOR REGIONAL AND GLOBAL POVERTY MEASUREMENT - SUGGESTED METHODS This section describes a number of approaches that can provide PPPs for poverty measurement which can be used in the place of what the World Bank uses for converting the $1 or $2-a-day poverty lines into national currencies based on the PPPs derived from the International Comparison Program. All the suggestions made here are based on an assessment of the existing data from the National Statistical Offices. Since PPPs here are supposed to reflect spatial differences in prices, it is necessary to have a quick review of some of the issues involved in spatial price comp arisons of consumer prices. Sections 5.1 and 5.2 provide a conceptual framework and a brief summary of some of the index number methods used. Exposition in these sections is drawn from the author's paper, Rao (2001b), which will appear as an Annex to the forthcoming CPI Manual to be published by the ECE and ILO. Section 5.3 provides a description of a number of alternatives that can be considered and it is concluded with recommendations with respect to the most suitable methodology for the present problem. 5.1 Spatial Comparisons of Prices and PPPs - A Digression There are several major qualitative differences in the nature of price comparisons involved in the standard CPI comparisons over time and price comparisons over space involving regions or countries. These differences highlight the need for specialised methods for aggregating price data in deriving summary measures of price levels as well as specific types of data requirements associated with cross-country or inter-area comparisons. The first and foremost difference is the absence of a natural ordering of price and quantity observations in the context of cross-country or inter-area comparisons. The CPI framework and methods are devised to measure changes over time and, therefore, the price observations appear in a chronological order. The presence of a natural ordering of price observations makes it possible to examine the feasibility and relative merits of the fixed and chain index numbers. For example, in the context of constructing price comparisons across countries within the Asia-Pacific region, or across states within Indonesia or India, it is impossible to arrive at an ordering which facilitates chained comparisons.
30
The multilateral nature of spatial comparisons is a distinguishing feature of price comparisons across regions and countries. When price levels of goods and services across different countries are compared, it is essential that such comparisons are undertaken for every pair of regions being considered. If the World Bank is interested in comparisons of real income of different countries, it is necessary for the Bank to be able to make comparisons between all countries involved. This multilateral nature of comparisons creates several problems. First, the number of comparisons (one for each pair) can be quite large and presentation and utilisation of such results may be quite difficult. For example, if a particular comparison exercise involves 20 countries, then it requires 190 (20 x 19/2) separate binary comparisons involving distinct pairs of countries. Secondly, results from such a large tableau of binary comparisons require a degree of consistency. This requirement translates itself into the "transitivity" condition described below. International comparisons of prices, in the form of PPPs from the ICP, are used by international organisations and individual researchers in assessing growth and productivity performance of countries and also for making meaningful comparisons of various national income aggregates (including government expend iture) across different countries. Currently, consensus is emerging among researchers and practitioners that price comparisons and PPPs are necessary in assessing the nature and extent of global poverty and its distribution across countries and regions of the world. Data Requirements For Spatial Comparisons The basic data requirements for spatial comparisons are very similar to the data required for standard consumer price index calculation. The main components are the data on prices of a large range of products representative of the consumption baskets of households and information on weights associated with various product categories reflecting the importance attached to different products. Within the CPI, it is common practice to collect price quotations from different outlets scattered throughout the country. The selection of the outlets and areas from which prices are collected are based on complex multi-stage sampling designs.
31 Spatial comparisons pose several problems in terms of identifying products that are to be priced from different areas, regions or countries involved in a comparison exercise. This problem is less severe when fairly similar or homogeneous areas are being compared. In cases where comparisons involve areas that are fairly heterogeneous two problems arise. The first arises from major differences in the consumption baskets. For example, when comparisons are made between two provinces in India, say Uttar Pradesh and Tamil Nadu, there may be major differences in the consumption baskets at the detailed level even though the major expenditure categories may be identical. This problem is somewhat similar to the treatment of disappearing and new goods in the context of the CPI , but is more serious when cross country comparisons are being attempted. The second problem is due to major differences in the quality differences in items. The quality differences may be measured through several product characteristics with allowances and adjustments being made at appropriate stage of index number calculation. Kokoski et al. (1999) demonstrate the feasibility of making inter-area price comparisons for heterogeneous goods. Changes in quality are likely to be more gradual in the case of temporal comparisons, but can be a serious problem when comparisons across countries are attempted. The ICP follows the principle of identity in dealing with the problem of quality differences across countries. A comprehensive list of products with detailed product specifications is developed at the planning stages of any cross-country comparison exercise. These items are priced in different countries from various outlets distributed across the country, a procedure very similar to that used in the CPI. However, development of the product listing is a difficult step, with the degree of difficulty depending upon the size and heterogeneity of the group of countries involved. Use of a product listing, based on the identity principle, can have serious implications for the representativeness of the product list of the consumption baskets in different countries. There are several operational procedures used by international organisations in handling these problems relating to the compilation of price data. A more detailed account of the problems and recommended solutions can be found in ICP Handbook ( UN, 1992). Spatial comparisons of consumer prices, pose specific problems relating to the nonoverlapping nature of the consumption baskets, major differences in the quality of items prices in different regions and countries, and, finally, the non-availability of
32 crucial data on region specific expenditure patterns. These problems require the development of new analytical techniques that can handle major differences in quality. Further, this may require additional financial resources by the national statistical offices in order to provide reliable and meaningful price comparisons between different cities, areas and regions within the countries and thus reliable data for the more difficult inter-country comparisons of prices and real consumption. 5.2 Aggregation Methods For Spatial Comparisons This section briefly describes the types of aggregation methods that are commonly used in cross-country comparisons of prices. Since most of these methods have been developed in the context of ICP, and are equally valid for inter-area or regional comparisons, the discussion below uses the countries as spatial entities. This section is further divided into three parts. The first deals with the notation and conceptual framework necessary to deal with multilateral spatial comparisons. The second describes the construction of elementary indices for aggregation of prices when no quantity or expenditure information is available. The last part presents a small sample of index number methods used in spatial price comparisons. Notation and Conceptual framework Consider the case involving comparisons across M countries and price and quantity data on N commodities. These commodities refer to goods and services that are priced in all the countries. If the commodities refer to items below the elementary level at which no quantity or expenditure share data are available, we make use of only the price data. At this stage all the problems relating to non-overlapping commodity lists and existence of quality difference are set aside so that focus is just on the aggregation issues. Let pj = [ p1 j, . . . pNj ] and qj = [q1 j, . . . qNj ] represent the price and quantity vectors from country j (j=1,2,…,M). In the case of international comparisons, all the prices are expressed in respective national currency units. As in the case of the CPI computation, the problem is one of decomposing the differences in the value aggregates (5.1)
Vj =
∑
N i =1
p i jq i j
into measures of price and real expenditure components.
33 Let Ijk denote the (consumer) price index number for country k with country j as the base. If j and k are, respectively, the USA and India, and if Ijk = 22.50, then the index is interpreted to mean that 22.50 Indian rupees have the same purchasing power as one US dollar for the goods and services covered in computing the index. Thus the index can also be interpreted as the purchasing power parity (PPP) between currencies of j and k. This interpretation is consistent with the meaning accorded to the CPI. Since currency denominations are involved here, a proper measure of relative prices can be obtained if the PPP is compared to the exchange rate prevailing at the time when comparisons are made. Due to the multilateral nature of spatial comparisons, when M countries are involved, it is necessary to provide comparisons between all pairs of countries. Thus, it becomes necessary to compute each and every entry in the following matrix of binary comparisons.
(5.2)
I 11 I I = 21 I j1 IM 1
I12 I 22 I j2 IM 2
I 1k I 2k I jk I Mk
I 1M I 2 M I jM I MM
Several points concerning the matrix, I, are worth noting. First, the matrix can be quite large if the number of countries (or regions) involved is large. Second, the results recorded in the matrix need to be internally consistent. It is possible to apply Fisher, Tornqvist, Walsh or other index number formulae used in standard index number applications. In order to ensure meaningful interpretation of the results from multilateral crosscountry comparisons, index number methods applied need to satisfy a number of basic requirements. Only the most important of these are discussed below. Kravis et al. (1982), OECD (1999) and UN (1992) provide a complete list of these requirements. Transitivity: An index number formula Ijk is said to satisfy the transitivity property if and only if for all choices of j,k and l (j,k, l = 1,2,..,M), the index satisfies (5.3)
Ijk = Ijl x Ilk
34 Equation (5.3) requires that the application of the formula to make a direct comparison Ijk , should result in the same numerical measure as an indirect comparison between j and k through a link state l . Note that the transitivity property ensures internal consistency of the index numbers in the matrix given in equation (A.2). It guarantees that the PPP for two currencies, say A and B, is the same whether it is derived through a direct comparison of A and B or through an indirect comparison which compares A with C and C with B, which are then combined to provide an indirect PPP for A and B. This requirement is mostly due to the spatial nature of the comparisons where no natural ordering of the countries involved could be imposed without a value judgement. Most of the commonly used index number procedures do not satisfy this requirement. The following result is useful in constructing transitive index numbers. An Implication of Transitivity Property: An index number formula Ijk satisfies the transitivity property in (5.3) if and only if there exist M positive real numbers λ1 , λ2 ,..., λM, such that (5.4)
I jk =
λk λj
for all j and k. The result is quite important since it shows that when the transitivity property is satisfied, all that is necessary is to measure M real numbers λ1 , λ2 ,..., λM, and then all the necessary indices in (5.2) can be calculated using these M numbers, thus reducing the dimension of the problem involved. Two important points may be noted. First, λj's in equation (5.4) are not unique, since any scalar multiplication of a vector of λj's can also lead to the same matrix of index numbers as that from the original λj's. Therefore, these λj's need to be determined (in any empirical exercise) up to a factor of proportionality. Second, these λj's can be interpreted as the purchasing power parities of currencies involved. This particular result formed the basis of the work of statisticians like Geary (1958), Khamis (1970) and others who proposed aggregation methods which are designed to compute PPPs directly from the price and quantity data without invoking the index number literature.
35 Base invariance: An index number formula is said to be base invariant if comparison between any pair of countries (j,k) is invariant to the order in which the countries are listed. This implies that multilateral comparisons should be invariant to all possible permutations of the data set. For example, under this condition, transitive comparisons derived using a particular country (say the US) as a star-country, with all other comparisons being derived using the US as the link, would be inadmissible, since the US is accorded a special status in the comparisons. Thus the base invariance condition requires symmetric treatment of all the countries involved in the comparison. Characteristicity: This is a requirement outlined in Drechsler (1975). This property requires that any set of multilateral comparisons satisfying the transitivity property should retain the essential features of the binary comparisons constructed without the transitivity requirement. Since condition (5.3) implies that a transitive comparison between a pair of countries j and k is necessarily influenced by the price and quantity data for all the other states, the characteristicity property requires that distortions resulting from adherence to the transitivity property should be kept at a minimum. Balk (2001) shows that a complete adherence to the characteristicity principle in its extreme (complete preservation of all binary comparisons) would imply that price indices, and hence purchasing power parities, cannot depend upon any quantity or expenditure share weights. This is an extreme result which is to be avoided in all index number comparisons. The Elteto-Koves-Szulc (EKS) method for multilateral comparisons, discussed below, has its origins in the characteristicity property. Index Number Methods for Spatial Comparisons Spatial price comparisons in general, and international comparisons in particular, use index number methods for aggregating price and quantity data at two different levels. The first is the basic heading level. This is the lowest level of aggregation at which expenditure data and weights are available. These basic headings usually comprise of fairly homogeneous group of items that are priced in different outlets in the countries. The next level of aggregation leads to indices for broad expenditure categories and finally to the whole consumption basket. Since Poverty PPP computation is contemplated at a level where quantity data are available, only aggregation methods above the basic heading level are discussed
36 below. Rao (2001) provides a brief exposition of some of the commonly used indices for aggregation below the basic heading level. Methods for Aggregation above the Basic Heading Level This level of aggregation is similar to the stage where elementary indices are aggregated to derive the overall CPI. However, the multilateral nature of spatial comparisons necessitates slightly different approaches to their construction. A number of index number methods for aggregation above the basic heading level have been developed over the last three decades. In the interest of brevity, only the principal methods are discussed below. These are the Geary-Khamis and EKS Methods for international comparisons, the principal aggregation methods used in various international comparison exercises by the ICP, the OECD, EUROSTAT and the FAO of the United Nations. Several approaches to the construction of multilateral index numbers which satisfy the transitivity and base invariance properties are discussed below. Four distinct approaches have emerged during the last three decades of ICP work. The first and the most straightforward approach is the EKS approach, which uses binary results as the building blocks for multilateral comparisons. The second approach is the GearyKhamis approach which provides a methodology for computing "purchasing power parities" of currencies and "international average prices" of commodities using the price quantity data at the basic heading level. The third approach is the stochastic approach based on the CPD method and its generalisations that can be used in econometrically estimating the purchasing power parities in a regression framework. The fourth and last approach discussed here is the linking approach to construc ting chained comparisons based on the concept of the minimum spanning tree. The last approach is generating considerable interest and is explored further below. These four approaches are by no means exhaustive, but they represent major strands of research and development in this area. EKS Method:
The Elteto-Koves-Szulc (EKS) system is a simple method of
generating transitive multilateral index numbers from a system of binary index numbers, with the property that the resulting multilateral indices deviate the least (according to a specific criterion) from the binary indices. Since the seminal paper by Drechsler (1975), it has been well recognised that (transitive) multilateral systems
37 necessarily deviate from their binary counter parts and therefore result in a loss of “characteristicity”. The EKS system is designed to minimise such loss of characterisiticity. The original EKS system uses the Fisher binary indices, but work of Caves, Christensen and Diewert (1982) and Rao and Banerjee (1984) recognise that other binary indices could be used in conjunction with the EKS technique. For any pair of countries j and k, if Fjk represents the Fisher binary index, then M
(5.5)
[
EKS jk = ∏ Fj l .Flk l =1
]
1/ M
provides the EKS index. The are several notable features of the EKS technique. First, it is based on the premise that that direct binary comparisons, derived using any chosen formula, provides the best comparison between pairs of countries. Secondly, even though the EKS index in (5.5) is defined using the Fisher index, this approach can be applied with any other choice of the index number formula. For example, the Fisher index in (5.5) may be replaced by another superlative index like the Tornqvist index. Caves, Christensen and Diewert (1982) suggest the use of Tornqvist-based EKS formula for spatial comparisons. Third, EKS index in (5.5) is the multilateral index that deviates the least from the matrix of non-transitive binary indices, when the deviations are measured using a logarithmic distance function. Finally, the EKS index can be interpreted as a simple geometric mean of all indirect comparisons between j and k through all possible link countries. The Geary-Khamis Method: The Geary-Khamis (GK) method was originally proposed by Geary (1958) and subsequently developed by Khamis (1970,1972 and 1984). The Geary-Khamis method has been the principal aggregation method in most of the ICP phases to date. Since 1996, the OECD has produced and published international comparisons based on both the EKS and GK methods. The Geary-Khamis method provides a way of calculating PPPs of currencies of different countries from the observed price and quantity data (applied at the basic heading level). The concept of PPP is applicable even when the currency unit is the same in several areas of a country. The GK method simultaneously determines international average prices of different countries. Let Pi denote the international
38 average price of i- th commodity. The GK method is defined through the following system of inter-related equations, defined for j and i, M
(5.6)
Pi =
∑p j =1
j i
N
j
q i / PPPj and
M
∑q
j i
PPPj =
∑p
qi j
i =1 N
∑P q i =1
j=1
j i
i
. j i
These simultaneous equations are then solved to yield numerical values of PPPs and Ps after selecting one of the currencies as a numeraire. Once the PPPs are solved, the spatial price index numbers are simply defined as I jk =
(5.7)
PPPk PPPj
One of the main reasons for the continued use of the GK method is "additivity". Additivity requires that aggregates, such as real domestic product, derived by converting national aggregates using PPPs, should be equal to aggregates derived through valuation of quantities at international prices. Thus additivity requires (5.8)
n
n
i =1
i =1
∑ p i jq i j / PPPj = ∑ Pi q i j .
This requirement is satisfied automatically by the PPPs and P's derived from the GK system defined in (5.6). The GK system is also useful in analysing the structure of real GDP and shares of different components across different countries. This system provides a framework within which internationally comparable national accounts could be constructed. Weighted CPD Method : It is possible to generalise the CPD method discussed in the context of aggregation below the basic heading level. Rao (1995) has generalized the CPD method by making use of quantity and value data directly into the CPD method. The basic idea behind this generalisation comes from the fact that the standard CPD regression model attempts to track the logarithm of the observed prices using an unweighted residual sum of squares. However, in the spirit of the standard index number approach where price index numbers are required to track price changes of more important commodities more closely, a more appropriate procedure would be to find estimates of the parameters that are likely to track important commodities more closely. This is achieved by minimising a weighted residual sum of squares, with each
39 observation weighted according to the expenditure share of the commodity in a given country. Thus the generalised CPD method suggests that estimation of the equation ln p ij = π1 D1 + π 2 D 2 + ... + π M D M + η1 D1* + η 2 D *2 + ... + η n D *n + u ij
(5.9)
is conducted after weighting each observation according to its value share. This is equivalent to the application of ordinary least squares after transforming the equation premultiplied by (5.10) wij ln p ij = π1 wij D1 + π 2 wij D 2 + ... + π M wij D M + η1 wij D1* + ... + η n wij D *n + vij where wij =
p ij q ij N
is the value share of i-th basic heading in j-th country..
∑ pij qij i =1
Rao (1995) has shown that the international prices and purchasing power parities resulting from the estimates of parameters in (5.9) are identical to those derived using the Rao- method for international comparisons described in Rao (1990). Thus the weighted CPD method may be considered as a bridge between the GK approach to international comparisons and the standard stochastic approach to index numbers. The weighted EKS method has the potential to provide PPPs while making adjustments for quality variations in products across countries. Information regarding outlets can also be incorporated into the regression equation which results in PPPs that are devoid of any influence (to the extent possible) of various quality characteristics thus enabling the PPPs to reflect pure price level differences. This approach will be referred to among the alternative approaches described in the next section. Spatial Linking and Chaining Approach: Spatial linking in a conceptual sense is of limited use within the context of only a few countries. But it can be immensely useful when price comparisons are being made across a large number of geographic regions that differ in climatic and other conditions that influence consumption patterns. In recent years this new approach to inter-area and inter-country comparisons of prices has been given serious consideration. It advocates spatial chaining of binary comparisons where links are identified using a procedure based on a measure of distance or reliability of binary comparisons involved. This approach is in sharp contrast to the general approach to multilateral comparisons where either all the
40 binary comparisons are used, as in the case of the EKS method, or all the price and quantity data are simultaneously used, as in the case of the GK and CPD approaches. Using the graph theoretic concept of Minimum Spanning Trees, Hill (1999a, 1999b) proposed a method of deriving a system of transitive multilateral comparisons from a matrix of binary comparisons. The Hill approach is based on the fact that direct binary comparisons may not always be the best. For any pair of countries j and k, Hill suggests a measure of distance (indicating the reliability of the binary comparison) using the Laspeyres - Paasche spread defined as (5.10)
L ( j, k P ( j, k )
D ( j, k ) = ln
where L(j,k) and P(j,k) are, respectively, binary Laspeyres and Paasche price index numbers. Note that the same distance function emerges if price index numbers are replaced by quantity index numbers. D(j,k) is equal to zero if the price structures or quantity structures are identical in countries, j and k. Thus this distance function serves as an indicator of similarity of price and quantity structures in these countries. Using a matrix of distances calculated for all pairs of countries, Hill (1999a, 1999b) suggests that a Minimum Spanning Tree (MST) be extracted which provides chained links between all pairs of countries, with the property that total distances in the tree is minimum. For purposes of illustrating the concepts involved, the following MST is provided using ICP data for Asia for the 1985 benchmark year.
41
Minimum Spanning Tree: Asia Bangladesh
Hong Kong; China
Pakistan
Thailand
Republic of Korea
Philippines
Sri Lanka
Turkey Japan
Iran
India
It is evident from the MST presented here that a comparison between Japan and Pakistan is through a chain involving the Republic of Korea and Hong Kong, China. This chained comparison is deemed to be better than a direct comparison between these two economies. There are a number of yet to be resolved issues regarding the use of MST. Nonetheless, a MST provides a formalization of a somewhat intuitive notion of linking dissimilar countries using a chain of similar countries. Once the minimum spanning tree is identified, transitive comparison between any two countries in a particular exercise is constructed using binary indices calculated using chosen formula, like the Fisher or Tornqvist index, and the links indicated in the above spanning tree. Thus, if a comparison between Sri Lanka and Bangladesh is needed, then the MST approach suggests the following index for this comparison. MST
I Sri Lanka,Bangladesh
= FSri Lanka, Thailand FThailand, Bangladesh
42 where F denotes the Fisher index. Since the minimum spanning tree provides a unique chain of links between any two countries, comparisons are uniquely defined. However, the spanning trees are sensitive to the countries included and the types of measures used in assessing the degree of reliability or comparability of any two countries. Heston et al. (2001) examine the sensitivity of the spanning trees and the resulting comparisons based on a range of measures including some similarity indices. Rao et al. (2002) applied the spanning tree approach to the construction of multilateral, agricultural input, output and productivity indices using US State level agricultural production data. The section has presented a small selection of the range of aggregation methods used in the context of spatial comparisons. A more comprehensive analysis of the spatial aggregation methods developed over the last three decades is presented in a recent paper by Balk (2001). In terms of the relevance of these methods to the present problem of constructing PPPs for poverty measurement the most relevant are the Geary-Khamis, the EKS and the weighted CPD methods. Reference is made to these methods in the next section. 5.3 Suggested Methods for Poverty PPPs In order to apply the methods described in the previous Section, it is necessary to start with a tableau of prices and quantities, or expenditures that can be used in conjunction with the formulae attached to different methods. In discussing the suggested alternative methods below, a quick review of the existing scenario may be useful. The current state of the art is to use the PPPs for the private consumption heading from the ICP for converting the $-poverty lines into national currency units. The main criticism of this approach is that these PPPs do not accurately reflect the prices paid by the poor. It may be stated that the PPPs, in view of the methodology of their estimation, necessarily reflect the prices paid by households/persons belonging to the upper or middle income groups of the countries involved in the comparison. The methods suggested below are designed to provide better estimates of purchasing power parities that are conceptually more suited to the conversion of a $1 or $2-a-day poverty lines into national currency units. Methods described here differ in their data requirements and therefore selection of a suitable method has to balance the
43 advantages and conceptual superiority underlying a particular method with associated data requirements. 5.3.1 Derive Adjustment Factors for use with Existing PPPs In a conceptual sense, the simplest alternative would be to explore the possibility of deriving "adjustment factors" that can be applied to the existing PPPs before their use in converting the dollar-poverty lines. Suppose PPPj is the purchasing power parity that is currently in use for converting the dollar-poverty line into the currency unit of country j. The suggested alternative here is to derive an adjustment factor, δ j, such that the revised parity, denoted as, PPPj* is given by PPP *j =
πj π US
x PPP j
where π j and π US are the indices of price level differences in the prices paid by the lower income groups relative to the upper income groups in countries j and the US respectively. Now the problem is one of estimating the differences in the relative price differences between the rich and the poor in different countries. These adjustment factors are essentially price index numbers computed for different income groups. These can be computed using: •
Standard index number formulae such as the Laspeyres, Paasche or Fisher formulae;
•
Price and quantity data can be drawn from one of the two following sources. •
The first source could be the standard CPI price data which, in all the countries in the pilot project, have prices paid by poorer sections and the average consumers. For example in India, price indices are computed for the agricultural labourers in rural areas, for industrial workers in urban areas and non- manual employees in urban areas. So price data for the non-manual urban employees may be considered as representative of the upper classes and the industrial workers and agricultural labourers could be considered as poorer classes. The expenditure share data for weights can be drawn from either CPI sources or more directly from the household expenditure surveys.
44 •
The second source of potential data for this purpose is the household expenditure survey itself. From the description of the data on expenditure and quantities purchased, it is possible to obtain price as the ratio of expenditure to quantity. This is a type of unit value ratio. Once these price data are obtained it is possible to use the generalized CPD method described in equation (5.10) using different income classes as population sub-groups, denoted by j in the equation, in the regression equation. If the top income (or top-half of the population) is considered as the reference group represented by the standard ICP PPPs, then the bottom income groups will form the remaining groups. In this case there is no need to use any other index number formula.
•
These adjustment factors are derived for each of the countries separately using data from the countries concerned alone.
This procedure is quite simple and a plausible extension to the present methods employed. However, it is conditional on the availability of price and quantity data available from either the CPI or the HES sources. When a large number of countries are involved, as is the case in global poverty comparisons, it may be possible that not all the countries may have well established and extensive HES or CPI data available. In such cases it may be feasible to use adjustment factors derived for countries that may be considered similar or those from countries that are geographically contiguous. It may be more desirable to make use of reliable adjustment factors for a limited number of countries instead of attempting to derive factors for all the countries involved in the comparisons. 5.3.2 Estimation of PPPs using CPI prices and HES weights Under this approach, PPPs are derived utilizing data on prices from the CPI price surveys and using them in conjunction with expenditure weights for the poor derived from the household expenditure data. For this purpose, we need to define prices and expenditure shares relating to the products. It may be possible that procedures similar to those employed in the standard CPI calculations using aggregation below the basic heading level without the use of any weights and then follow it with weighted indices. Data required for this approach are the following:
45 •
Price data in the form of price of each product, averaged over the outlets, at a given point of time in a given country. Let pij represent the price of i-th commodity in country j.
•
Corresponding to these items and prices suppose average expenditures, denoted by eij, are available. Then the quantities implicit in these values are obtained by qij = eij/pij.
•
The expenditures, and implicit quantities are in the form of averages over all those households that are considered to be poor. This poses a problem of circularity since these PPPs are in turn needed for the determination of the poor. In such cases, a simple alternative is to consider the bottom 30% as representative of the poor for purposes of deriving the weights. One would not expect major shifts in expenditures within the neighbourhood of a given cut-off mark.
Once the price data are compiled for each of the countries involved, then a suitable method of aggregation could be used. If the standard ICP approach is used then either the Geary-Khamis method or the EKS method may be used in constructing PPPs that are based purely on the price data representative of the consumption of the poor and the expenditure weights that are also representative of only the poor. 5.3.3 Estimation of PPPs using only Household Expenditure Data One of the main problems associated with the alternative described in 5.4.2 is the problem of establishing a one-to-one correspondence between the CPI item prices for the poor and the commodity classification used in the HES data. Given the existence of possible inconsistencies within these two survey frameworks, it may be useful to focus on a single source for both price and quantity data. Such a source is the household expenditure survey where quantity and expenditure data are recorded for all the households. From the discussion of the findings in the pilot project countries, it is obvious that at this stage, survey data from Thailand cannot be used since no record of purchased quantities is made. Similarly, the Indonesian data is restricted to only the food items and no quantities are recorded for non- food items. In such cases it may be possible to restrict the PPP computation only to the food prices since about 70% (approximately, this may vary from country to country) of the poverty line is made up of food items. It may be possible to mimic the country practice of arriving at the nonfood component using some applications of Engel's law.
46 In the discussion below, suppose the expenditure and quantity data are available for each household that is included in the HES. For the moment, let these data correspond to N commodities (this could refer to food alone or both food and non- food items inclusive). There are M countries and Mj households in country j (j = 1,2,…,M). Let pikj = price of i-th commodity paid by the k-th household in j-th country eikj = expenditure on -i th commodity by the k-th household in j- th country, this expenditure is expressed in national currency units. Within this strategy two further options are available: Option 1: This involves the following steps. Simply compute the average price and expenditure, averaged over all the households. Then for each commodity, there is a single price and expenditure in each country. Then apply the Geary-Khamis, EKS or the weighted CPD methods to the data available. A disadvantage with this option is that some of the information that is available in the form of household data is lost due to the averaging process followed in the initial stage. Option 2: Make use of all the household level data along with the country-productdummy method, with each price observation referring to a household within a particular country. The weighted version of the CPD method allows the use of expenditure data, and the resulting indices have some very useful properties. The weighted CPD model is shown in (5.11) (5.11)
wij ln p ij = π1 wij D1 + π 2 wij D 2 + ... + π M wij D M + η1 wij D1* + ... + η n wij D *n + vij
where
wij =
p ij q ij N
is the value share of -i th basic heading in j- th count ry The
∑ pij qij i =1
model in (5.11) can be applied on data that is specific to each household. It is then necessary to have the regression model estimated using data on commodities (i=1,2,…,N); countries (j=1,2,…M) and also households using an additional subscript h (h=1,2,…H) where H is the number of households in the data set. It is possible within Option 2 to incorporate dummy variables for different regions within each country into the regression equation. Such a model will automatically generate PPPs for different regions in different countries, all derived simultaneously.
47 This model will eliminate the need for making adjustments for regional price differences within a given country separately, and therefore, can be considered superior to all the methods discussed so far. An additional advantage with both options 1 and 2, which make use of the weighted CPD model discussed in equation (5.10), is that it is possible to handle the problem of quality differences within the regression model. The CPD model can be extended to incorporate product characteristics, when prices are well- specified items, and the type of outlets from which prices are recorded. Such a model is very similar to the standard hedonic regression models used in making quality adjustments. 5.4 Conclusions On the basis of the currently available data from the household expenditure surveys, CPI surveys and other work undertaken in association with poverty measurement, it appears that it is feasible to derive estimates of PPPs that can replace the existing PPPs from the standard ICP. However, the new methods discussed in 5.3 require careful data preparation before the actual empirical implementation. Since there are several alternative approaches that can be employed, it is necessary to examine the sensitivity of the empirical estimates of PPPs to the use of different methods and data sources. This could form the next stage of this project which can be implemented in conjunction with the work on the next round of the International Comparison Programme in 2003.
48 SECTION 6: SUMMARY AND RECOMMENDATIONS This section provides a summary of the major findings with respect to data related issues that have resulted from the pilot project. A brief summary of the discussions held with statisticians in India, Indonesia, the Philippines and Thailand is provided in this section. Instead of describing deliberations in different countries separately, an attempt is made here to synthesize the discussions under different topic headings. The Section is concluded with some general remarks and recommendations. 6.1 Summary of discussions Poverty Measurement Poverty measurement is considered a significant activity in all the countries visited. In all these countries estimates of the number of households and number of people in poverty are estimated on a regular basis. In addition, other poverty measures such as the poverty gap ratio and the Foster-Greer-Thorbecke indices are also calculated. These measures are designed to provide an indication of the depth of poverty. Separate government entities, such as the Planning Commission in India and BAPPENAS in Indonesia, are entrusted with the work of analysing and formulating suitable poverty alleviation policies. Poverty incidence is also estimated at the regional or provincial level. In all the countries visited, poverty estimation is based on the absolute poverty approach and, therefore, depends on the determination of poverty lines. There is a common structure present in the country methodologies used in deriving poverty lines. All the pove rty lines are primarily based on per capita energy intake requirements around 2100 kCal per day. These caloric requirements are then converted into nominal values in the form of expenditure on food necessary to meet the energy requirements. Methodologies used in translating these requirements are quite different in these countries. Procedures followed in Thailand and the Philippines seem to be more comprehensive. In fact this is a major point of departure from the standard practices in other countries. The next step in determining the total poverty line, which is the sum of food and nonfood poverty lines, is again addressed in different countries differently. The most sophisticated among these is the procedure used in Thailand. However, all the
49 countries use the concept of Engel ratios in one form or another in arriving at the nonfood component of the poverty line. Poverty estimates are provided on a regional basis in the study countries. The countries find it necessary to ensure that poverty lines adequately reflect regional differences in nutritional requirements, expenditure patterns and prices of goods and services. Country practices, however, are quite diverse in this area. Household expenditure surveys are the primary source for the estimation of poverty incidence. Income concept is used in the Philippines. Based on a recent review of the poverty measurement methodology in India, poverty estimates are based on household expenditure survey data instead of the national accounts data on the private consumption expenditure. It was found that updating the poverty lines is largely based on the price movements from the consumer price index (CPI) numbers. However, in some countries, these indices are available for low income or some equivalent groups. Such CPIs are more appropriate for the purpose of poverty line adjustments than the national CPI. Household Expenditure Surveys (HES) All the countries visited have well established methods in terms of sampling design, data collection, tabulation and dissemination of results from the HE Surveys. In the Philippines, these surveys refer to families and hence the surveys are known as the Family Income and Expenditure Surveys. For the most recent years, all the data collected from these surveys are available in electronic medium in the form of unit record data on CD-ROMs. There are major differences in the frequency, data collection periods and the actual questionnaires used in the surveys. The HES data are used in estimating the proportion of poor in the population. These surveys are also used in regional poverty estimation. Expenditure and income data are available from these surveys and, therefore, both income and expenditure approaches to poverty estimation can be used with this data. These surveys also form the principal source for the derivation of the food poverty lines. While all the countries make use of these surveys, the actual methods of use differ significantly. An important feature of the HES is that these surveys collect household-specific expenditures on specific food and non- food items along with quantities. With the exception of Thailand, all the countries tabulate both expenditure and quantity data
50 which can then be used in deriving commodity-specific unit value ratios. While these ratios mask essential quality differences, they could be of considerable relevance in the case of poorer households. All the countries recognize the fact that quantity measures are of a less reliable quality than the expenditures. All the surveys attempt to adequately account fo r consumption out of own production using standard imputation procedures. Given the extensive nature of these surveys and their availability in all the pilot study countries, data from these surveys will form a major input into the computation of PPPs for poverty measurement. Discussions during these visits also focused on the divergence between the HES and National Accounts estimates of per capita consumption expenditure. All the countries are aware of the sources of divergence and that the national accounts are likely to underestimate poverty while the use of HES may slightly overstate poverty incidence. The general consensus is to rely on the household expenditure survey data. Consumer Price Indices Consistent with prior expectations, all the statistical offices have fairly wellestablished structures in place for the purpose of CPI computation and dissemination. The survey instruments are well developed, with commodities and outlets selected on sound statistical principles. The dissemination of CPI is based on the standards dictated by the IMF, with monthly CPIs made available. The main focus of the discussions during these visits was to assess the extent to which CPI figures and the underlying data are used in establishing the poverty lines and their subsequent updating for temporal changes in prices. Suitability of using CPI price data for poverty studies was also an item for discussion. In all the countries visited, there are CPIs for different populations groups of which one or two groups could be considered close to the poor. For example CPI for the bottom 30 per cent of the population could provide a reasonable approximation to the price movements for the poorer sections of the population. In India, CPI for the rural agricultural labourers and urban manual labourers are computed separately. Both of these indices could provide a better approximation of the price movements relevant for the poor.
51 Although unpublished, it appears that very useful information is available from the price data collected for CPI purposes. For all the items of the CPI baskets, a description of the item-characteristics are available along with a general indication of the type of outlet (supermarket vs. traditional markets etc) are available. However, these details are not computerized at this time in any of the four countries. Such data could be made available in an electronic format if sufficient resources were made available. Availability of price data that is collected as a part of the CPI, especially for population groups that closely reflect the poorer sections of the population, provides an opportunity for the construction of the PPPs for poor. This is an approach that was pursued in Section 5. Another important finding that emerged from these meetings is that none of the statistical offices are currently constructing spatial consumer price index numbers on a regular basis. In India, spatial price indices computed in the 70s still form the basis for the poverty lines for different states of the country. In Thailand, these were computed for the year 1992 as a part of the overhaul of the procedures used for poverty measurement. The regional poverty lines are then updated using movements in prices of food and non- food items. In Indonesia and the Philippines, spatial price indices were computed using a Laspeyres-type index formula. The meetings have a revealed a distinct lack of knowledge of the methods used in spatial price comparisons. These offices have expressed a strong desire to learn and use these methods regularly for purposes of spatial price comparisons. 6.2 Concluding Remarks and Recommendations The consultant is satisfied with the visits to the countries and is quite impressed with the level of preparation of the statistical offices and the time they had devoted in preparing the background material. The officers present in various meetings were very patient with the consultant and his searching questions aimed at gaining an understanding of the procedures employed in the course of poverty measurement. The consultant gratefully acknowledges the help provided by these offices which has facilitated efficient use of the limited time available during these visits.
52 A few comments and recommendations are listed below: 1. All the countries visited, including India, expressed their interest in participating in the next round of the International Comparison Programme. This is an encouraging sign. 2. Officials who were associated with the ICP work in the past have complained about the lack of transparency of the ICP programme. They felt that they were not sufficiently informed of the ICP programme once the price data were supplied. They expressed their wish to be regularly informed of the progress of ICP work throughout the Phase. 3. There was considerable demand for statistical training on the construction of spatial price comparisons. Regional comparisons are deemed important in all the four countries the consultant had visited, but not much effort has been devoted to spatial price comparisons. They felt a distinct lack of knowledge in this area. It is recommended that a training program on spatial price comparisons be set up prior to the next round of the ICP. 4. During the discussions, a consensus has emerged that the $1-a-day poverty line may not be as obscure as it may sound. In most countries, PPP conversion of the one-dollar poverty line appears to be reasonably close to the poverty lines used in these countries. It is recommended that further work be undertaken to empirically implement the methods described in Section 5. 5. There is a general willingness to provide data necessary for purposes of testing any methods that may be proposed for the construction of PPPs for global and regional poverty measurement. But provision of some of the data may require additional work in the form of computerizing data that is not currently available in electronic format. This is an initiative that has considerable benefit in the long run. 6. Detailed HES data are available in electronic format but the cost is quite high. It may become necessary to acquire these data in order to empirically implement some of the methods that may be proposed in the final report. 7. Given the positive outcomes reported here, it appears that the next logical step is to implement some of the procedures and obtain alternative sets of PPPs for poverty measurement. 8. In order to establish a link with the US$ and to derive a PPP with respect to US dollar, it is necessary to examine if household expenditure and CPI data are available within the United States in a format similar to that found in the Asian countries covered in the present study.
53 REFERENCES Aten Balk, B.M. (2001) "Aggregation Methods for International Comparisons: What have we learnt in the last decade?" Paper presented at the Joint World Bank-OECD Seminar on Purchasing Power Parities: Recent Advances in Methods and Applications, 30 January-2 February, 2001, Washington DC. Biru, Y. (1998) "The Purchasing Power of the Poor in Zambia", World Bank Seminar on Prices and Purchasing Power Parities, Washington, D.C. Caves, R.E., L.R. Christensen and W.E. Diewert (1982), "Multilateral Comparisons of Output, Input and Productivity using Superlative Index Numbers", Economic Journal, 92, 73-86. David, J.P. (2001), "Issues and Recommendations for Improving Poverty Statistics", Paper presented in the Poverty Statistics Session of the ESCAP Working Group of Statistical Experts, 27-30, November, Bangkok. Deaton, A. (2001), "Counting the world's poor: problems and possible solutions", World Bank Economic Review, 2001. Decoster, R. (1999), "Proposal for Comparative Poverty Assessment Using Purchasing Power Parities for Low Income Households", Paper prepared for the DECDG, World Bank. Drechsler, L. (1975), "Weighting of index numbers in multilateral international comparisons", Review of Income and Wealth. Geary, R.C. (1958) "A note on comparison of exchange rates and purchasing power parities between currencies", Journal of Royal Statistical Society, 121, Part 1. Heston, A.H., R. Summers and B. Aten (2001), " Some Issues in Using Chaining Methods for International Real Product and Purchasing Power Comparisons", Paper presented at the Joint OECD-World Bank Seminar on Purchasing Power Parities: Recent Advances in Methods and Applications, Jan 30-2 Feb, 2001, Washington DC. Heston, A.H. (2001), "Which PPPs should we use in comparing Poverty Levels?", Paper prepared for DECDG of the World Bank, Washington DC. Hill, R.J. (1999a), "Chained PPPs and Minimum Spanning Trees" in Lipsey and Heston (eds.) International and Interarea Comparisons of Prices, Income and Output, NBER, Chicago University Press, pp. 327-364. Hill, R.J. (1999b), "Comparing Price Levels Across Countries Using Minimum Spanning Trees", The Review of Economics and Statistics, 81, 135-142. Kakwani, N. (2001), "On Specifying Poverty Lines", Paper presented at the Asia and Pacific Forum on Poverty: Reforming Policies and Institutions for Poverty Reduction, 5-9, February, Manila. Kakwani, N. and M. Krongkaew (2000), Poverty in Thailand: Defining, Measuring and Analysing, Report prepared for the National Economic and Social Development Board (NESDB), Thailand.
54 Khamis, S.H. (1970) "Properties and Conditions for the Existence of a New Type if Index Numbers", Sankhya, Series B., 32. Khamis, S.H. (1972) "A New System of index Numbers for National and International Purposes", Journal of the Royal Statistical Society, Series A, 135. Khamis, S.H. (1984) "On Aggregation methods for International Comparisons", Review of Income and Wealth, 30. Kokoski, M.F., B.R. Moulton and K. D. Zieschang (1999) "Interarea Price Comparisons for Heterogeneous Goods and Several Levels of Commodity Aggregation", " in Lipsey and Heston (eds.) International and Interarea Comparisons of Prices, Income and Output, NBER, Chicago University Press, pp. 327-364 Kravis, I.B., A.W. Heston and R. Summers (1982) World Product and Income: International Comparisons of Real Gross Domestic product, Johns Hopkins University Press, Baltimore. OECD (1999) Purchasing Power Parities and Real Expenditures, Department of Economics and Statistics, OEC D, Paris. Rao, D.S. Prasada (1990), "A System of Log-Change Index Numbers for Multilateral Comparisons", in Comparisons of Prices and Real Products in Latin America Eds. J. Salazar-Carrillo and D.S. Prasada Rao, North-Holland, Amsterdam. Rao, D.S. Prasada (1995), "On the Equivalence of the Generalized Country-ProductDummy (CPD) Method and the Rao-System for Multilateral Comparisons", Working Paper No. 5, Centre for International Comparisons, University of Pennsylvania, Philadelphia. Rao, D.S. Prasada (2001), Integration of CPI and ICP: Issues, Possibilities and Recommendations, A Report Prepared for DECDG of the World Bank, Washington DC. Rao, D.S. Prasada (2001b), "Spatial Comparisons Of Consumer Prices, Purchasing Power Parities And The International Comp arison Program" To appear as an ANNEX to the ECE-ILO CPI Manual to be published by the end of 2002. Rao, D.S. Prasada and K.S. Banerjee (1984), “ A Multilateral Index Number System Based on the Factorial Approach”, Statistische Hefte, 27, 297-313. Rao, D.S. Prasada, C.J. O'Donnell and V.E. Ball (2002), "Transitive Multilateral Comparisons of Agricultural Output, Input and productivity: A Nonparametricf approach", in Ball and Norton (eds.) Agricultural Productivity: Measurement and Sources of Growth, Kluwer Academic Publishers, Boston. Rao, V. (2000), "Price Heterogeneity and "Real" Inequality: A Case-Study of Prices and Poverty in Rural South India", The Review of Income and Wealth, 46:2, 201-212. Ravillion, M. (1998) Poverty Lines in Theory and Practice, LSMS Working paper No. 133, Living Standards Measurement Study, The World Bank, Washington, DC. Ravallion, M. (1998), "On the determination of Poverty Lines", Working Paper, Living Standards Measurement Study (LSMS), World Bank, Washington DC.
55 Surbakti, P.(1997), Indonesia's National Socio-Economic survey - A Continual Data Source For analysis on Welfare Development, Central Bureau of Statistics, Jakarta. United Nations (1992), The Handbook of the International Comparison Program (ICP), Series F. No.61, New York. World Bank (2000) World Development Report, Washington DC.
56
Material used from country sources INDIA CSO
(2000), Consumer Price Index for Urban Non-Manual Employees [CPI(UNME)], Base: 1984-85=100, Central Statistical Organisation, Ministry of Statistics and Programme Implementation, New Delhi.
Government of India (2001), Consumer Price Index Numbers for Agricultural and Rural Labourers (Base: 1986-87 = 100), Note provided by the Labour Bureau, Ministry of Labour, Simla, India. Government of India (2001), Consumer Price Index Numbers Industrial Workers (Base: 1982 = 100), Note provided by the Labour Bureau, Ministry of Labour, Simla, India. NSSO (1999), Instructions to Field Staff, Volume -1Schedules of Enquiry, SocioEconomic Survey, NSS 55th Round (July 1999-June 2000), Department of Statistics & Programme Implementation, Government of India, New Delhi. NSSO (1999), Instructions to Field Staff, Volume -2: Design, Concepts, Definitions and Procedures, Socio-Economic Survey, NSS 55th Round (July 1999-June 2000), Department of Statistics & Programme Implementation, Government of India, New Delhi. NSSO (2001), Level and Pattern of consumer Expenditure in India, 1999-2000, NSS 55th Round, ), Department of Statistics & Programme Implementation, Government of India, New Delhi. NSSO (2001), Consumption of Some Important Commodities in India, 1999-2000, NSS 55th Round, ), Department of Statistics & Programme Implementation, Government of India, New Delhi. PCL (2001), Compilation of Consumer Price Index for Urban Non-Manual Employees with a copy of the Field Questionnaire, Notes supplied by the Prices and Cost of Living Unit, Central Statistical Organisation, New Delhi. Planning Commission (1979), Report of the Task force on Projections of Minimum Needs and Effective Consumption Demand, Government of India, New Delhi. Planning Commission (1993), Report of the Expert Group on Estimation of Proportion and Number of Poor, Government of India, New Delhi. Sharma, G.D. (2001), "Compilation of consumer Price Indices (CPI) and Wholesale Price Index (WPI) in India", Mimeographed, Central Statistical Organisation, Department of Statistics & Programme Implementation, Government of India, New Delhi. INDONESIA BPS - Statistics Indonesia (1998), Crisis, Poverty and Human Development in Indonesia, BPS Katalog: 1115, Jakarta. BPS - Statistics Indonesia (1999), Pengukuran Tingkat Kemiskinan Di Indonesi, 1976-1999, Seri Publiksi Susenas Mini 1999, Katalog BPS: 2320.
57 BPS - Statistics Indonesia (1999), Penyempurnaan Metodologi Penghitungan Penduduk Miskin Dan Profil Kemiskinan 1999, Katalog BPS: 2321, Jakarta. BPS - Statistics Indonesia (2000), Indonesia in Figures, Katalog BPS: 1403, Jakarta. BPS - Statistics Indonesia (2000), Statistick Harga Konsumen, Harsha, HK 1-6 BPS - Statistics Indonesia (2000), Methodologi penentuan RumaH Tangga Miskin 2000, Studi Penentuan Kriteria Penduduk Miskin, Katalog BPS: 2324, Jakarta. BPS - Statistics Indonesia (2000), Indikator Kesejahteraan Rakyat (Welfare Indicators), Katalog BPS: 4103.Jakarta. BPS - Statistics Indonesia (2001), Berita Resmi Statistik: Perkembangan Indeks Harga Honsumen/In flash, Jakarta. BPS - Statistics Indonesia (2001), Monthly Statistical Bulletin - Economic Indicators, Katalog BPS: 1201, Jakarta. Rosidi, Ali (2000), "Ind ustrial Production Index, wholesale/producer price Index and Consumer price Index of Indonesia, Country paper for the Joint OECD/ESCAP Workshop on Key Economic Indicators, 22-25, May, Bangkok. Surbakti, S., P. Surbakti, L.O. Syafiuddin, Maesuroh and W. Imawan (2001), Poverty Data Policy, Management, and Implementation, Country Paper: Indonesia, Workshop on Strengthening Poverty Data Collection and Analysis, April 30May 3, 2001, Manila. Sutanto, A. (1999) "The December 1998 Poverty estimate: Methodological Issues", Paper presented at the Workshop on Poverty Number Computational Method at BAPPENAS (National Development Board), 25 June, Jakarta. THAILAND NESDB (1998), New Poverty Thresholds for Thailand with Policy Applications, A Newsletter of National Economic and Social Development Board, Vol 3, No. 1, Bangkok. NESDB (1998), Poverty profiles for Thailand, A Newsletter of National Economic and Social Development Board, Vol 2, No. 3, Bangkok. NESDB (1999), Poverty and Inequality During the Economic Crisis in Thailand, A Newsletter of National Economic and Social Development Board, Vol 3, No. 1, Bangkok. NESDB (2000), Poverty and Income Distribution in 1999, A Newsletter of National Economic and Social Development Board, Vol 4, No. 1, Bangkok. NSO (1998), Socio-Economic Survey 2000, Household Expenditure Survey Questionnaire. National Statistical Office, Bangkok. NSO (2000), Socio-Economic Survey 2000, Household Expenditure Survey Questionnaire Parts 1 and 2). National Statistical Office, Bangkok. NSO (2000), Report of the 2000 Household Socio-Economic Survey, Whole Kingdom, National Statistical Office, Bangkok.
58
THE PHILIPPINES NSCB (2000), Profile of Censuses and Surveys Conducted by the Philippine Statistical system, Manila. NSCB (2000), Philippine Poverty Statistics, National Statistical Coordinating Board, Makati City. NSCB (2001), Press Release on Poverty in the Philippines, July 25, 2001, National Statistical Coordinating Board, Makati city.. NSO (2000), 2000 Family Income & Expenditure Survey, Manila. NSO (2000), Family Income and Expenditure Survey- Detailed Schedules, Manila. NSO (2001), 2000 Family Income and Expenditure Survey (FIES), Preliminary Release on Poverty, Manila. NSO (2001), 2000 Family Income and Expenditures Survey, Preliminary Results, Manila. OTHER Fiji (2000), "Fiji's Consumer Price Index", Country paper for the Joint OECD/ESCAP Workshop on Key Economic Indicators, 22-25, May, Bangkok. Fujiang, L. (2000), "Brief Introduction on Labour Force, Retail Sales and price Statistics of China", Country paper for the Joint OECD/ESCAP Workshop on Key Economic Indicators, 22-25, May, Bangkok.
