The Cost of Inadequate Yardstick Models Per J. AGRELL Emili GRIFELL-TATJE

UCL/CORE UAB, Spain

NAPW 2016, Quebec City, June 14-18, 2016

Reference This presentation draws on Agrell, P J and E Grifell-Tatjé (2016), A Dynamic Model for Firm-Response to Non-Credible Incentive Regulation Regimes, Energy Policy 90, 287–299.

Outline Incentive regulation Game theoretic model Empirical method Empirical testing Results

Part 1

YARDSTICK REGULATION

Incentive regulation in a nutshell Regulator Services y

Contract M(y,t)

Operator

Effort e

Cost C(y,e) Infrastructure access, unbundled firm, inelastic demand for service Cost is observable and verifiable, effort is unobservable, multi-output service provision High-powered regulation is optimal: Laffont (1994), et al. Practical implementations: yardstick regimes: Schleifer (1985), Laffont and Tirole (1986)

Yardstick design and DEA Bogetoft (1994) › A cost norm built on DEA provides optimal incentives for deterministic models, and stochastic output under some conditions (truncation property and half-normal, exponential distributions)

Agrell, Bogetoft and Tind (2005) › A DEA-based cost norm is optimal also in a dynamic setting, the optimal implementation has a linear form for linear cost of effort.

DEA minimizes the rent given to the firm

EU Regulatory landscape (Energy) EU Regulatory landscape – Methods (Energy)

IS

Cost recovery Revenue Cap (CPI-X) Revenue Cap (CPI-DEA/SFA) Finland

Norway

Yardstick - DEA

Sweden Estonia

IRL

Latvia

DK

GB

Lithuania NL BE.

Poland

Germany

AT SL

HU HR

BH Portugal

Spain

Yardstick - MNA

Sweden under re-reform: Rate-of-return regulation

Switzerland under reform: incentive regulation (DEA pilot)

SK CH

Yardstick - Other

Belgium under re-reform: De facto cost-recovery

CZ France

Price-Cap

Iceland: reform not implemented

RO

Finland: revenue cap with StonED

SB

BG

Italy MN. FYROM. AL

Greece

Normative models are popular Country AUSTRALIA AUSTRIA DENMARK FINLAND GERMANY NETHERLANDS NEW ZEELAND NORWAY ICELAND PORTUGAL CHILE SPAIN ENGLAND BELGIUM SWITZERLAND SWEDEN 7

Approach Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante

Method CPI-DEA DEA/EngM COLS DEA->StonED DEA/SFA Yard Cost Yard CPI-DEA DEA Yard CPI-DEA SFA EngM EngM CPI-X CPI-DEA -> CR (RoR)->? (EngM)->RoR

Analysis x x x x x x x x x x x x x x x x

Operation x x x x x x x x ? x x x x

Irrelevance of cost norm Revenue cap = R0 CPI (1 – X – Xi) Incentive regulation, corollaries › A profitmaximizing firm do not care about the level of the cap › A utilitymaximizing firm cares about the incentive power › What matters are the commitment to and duration of the regime › No importance of the used cost norm

Does it hold in practice? The regulation is based on the cost norm Regulation must hold for all firms without bias It is not sufficient to be right on expectation Judicial recourse to protect from expropriation › Firms may appeal rulings › If a ruling shows a flaw in the model, the regime falls

Part II

ANECDOTAL EVIDENCE

Failing regulation in Europe Netherlands › Frontier model revoked 2004, debacle 140 M in welfare losses › Nillesen and Pollitt (2007) › Moratorium and average cost model

Belgium › Preparation for incentive regulation, overturned and decentralized in 2012 › Agrell and Teusch (2015) › Cost-plus regulation by region since 2012 …

Sweden › Network performance assessment model (NAPM) falls in 2006 › Moratorium and cost-plus regulation until 2014 …

Credibility Commitment is based on a rational expectation of durability The robustness of a regulation depends on › Participation of the regulated firms › Sustainability of rents left to stakeholders › Properties of the cost norm (soundness)

A regulation regime not satisfying these criteria is not credible

Idea Intuition: › A rational firm reveals only its full efficiency for a regime with a credible commitment and cost norm.

Method: › Decision model for a firm evaluating a proposed regime › Methodology to test the hypotheses for firm behavior › Validation with productivity data for a failed regime

Feasible and infeasible cost norms CAPEX/output

A

Benchmarking model Statistical model Normative model True frontier

OPEX/output

Part 1II

MODEL

information R(y) that assures full participation of all efficient firms in the tempting in long run and a regulatory commitment structure conducive n possibility to sustain truthful revelation of information by firms for the wever, given operation of the regulation. Assume now that a firm facing a in network non-credible high-powered regime R considers the probability e estimation of it falling to v per year and that failed regime is replaced Model models are by a low-powered regime. Let us assume ex-post verifiable with varying cost and single-dimensional service data as xt , yt for year t. regulated are used toOneThe firm isfirm maximizing horizon utility consisting of profit and able, cost forMulti-period game, discount factor slack using a discount factor . The initial situation is that of Furthermore, wx is the observed cost andR R = 0 Slack will=observe that high-powered regime where r 2 [0,optimal 1[. The slack is here an convenient expression for the acase of linear cost of effort, such that 1 euro of surplus cost (for example poor procurement policies, del. The risk is non-credible if it violates the participation constraint for obsolescence, luxurious equipment) has the value of r euro in disposable profits (that s maximizes in a For concession or licensing by imposing a could be paid any as e.g.firm dividends). a cost-plus policy with R = wx,system equation (1) simplifies to reimbursement R < x? . Note that the failure of a regime is not u(x, wx) = r(wx c(y, w)). (2) inimum loss instantaneous, its downfall is brought by sometimes lengthy The expected horizon utility is the sum of the discounted single-period utilities with a cted from a factor judicial theperiod inefficient incumbent discounting d 2 ]0, 1[,appeals meaning thatand 1 eurolobbying. at the end of of If the first is worth euro at the start of the horizon. The reservation utility for the firm is normalized to ely tod result zero. The firm does not participate unless the expected utility is higher or equal to the considerable reservation utility. Regulatory game system can 3.2. Dynamic game ? observes a regime R x for all firms, the optimal policy may be conThe regulator and the firm are playing an infinite horizon game for a stationary dePeriod 1:bew.toInitially, will optimize costselects as toa regime achieve = inx? , leaving no {R, v} c(y) y and fixed prices the regulator consisting isted. mand If the a revenue levelno derived some (e.g. average cost or ideal › R Launch of⇡high-powered regime R(y) profitusing = 0 methodology but infinite participation. If,network on the other hand, s candidates models) associated with a failure risk v. The objectives of the regulator are to assure firm ? Period t = 2,…,T the incumbent faces a regime R > x for itself participation and to incite cost minimization by the participating firms. Observing the but observes to form the regime {R, v} the firm chooses topotential participate in infeasibility the game or to exit.of Exiting reser› In each period, the regime is challenged at least one the gives costnorm, the choice vation utility. If participating, the firm maximizes its expected horizon utility EU(x) in › vnot = P(Regime revoked) ? , x] 5 . The the is trivial. that the regime bution(3)operover an input vector as x 2 [x ¯ validRecalling for the horizonthat lowerprobability input bound x? de› efficient If not Rtpolicy = R(y) a fully cost policy. Any such x > x?aisfailed defined asregime cost inefficient falls isrevoked: v each year andthatthat is replaced by a ile [9],fines [18], behaviour. The› payouts for a participating firm in the game If revoked: cost-plus regime Rt = xt-1 are illustrated in Figure 1. cost-recovery, Rt =v and xt the1 regulation for a reform In each period,pure the regime is repealed withi.e. probability resorts toin year t. The cost-plus, i.e., firm the firmmaximizes receives the reimbursement wx leading a utility and u(x, wx). With as an objective of toprofit slack R EVISITED 4 To simplify notation, all vector products are assumed to be done in the appropriate dimensions, i.e. ation suppressing is 5that the sign for the transposed vector. The upper bound x¯ for the support may be seen as existing observations or as the maximum value ? us demand y could award without u(x|R) =regulation. ⇡ + ⇢s = (R x) + ⇢(x x ) (1) that the regulator resetting the 6

Game timeline

U(x,R)

1-v

U(x,R)

1-v

U(x,R)

Incumbent regime R = R’

1-v

1-v v

v

v

obability 1 v, the regulator stands the appeal and enforces the payment R with the rm level utility u(x, R). The vfailure of the1 regime is irreversible for1 the horizon, i.e., 1 Low-powered regime e regulator must continue paying wx for the rest of the horizon. R = wx U(x,wx) U(x,wx)vector x for the firm In this game, the expected horizon utility forU(x,wx) a stationary input probability 1 v, the regulator stands the appeal and enforces the payment R with the given as t = 1 of the regime t=2 t=3 firm level utility u(x, R). The failure is irreversible for the horizon, i.e., • • the regulator must continue paying wx for the rest of the horizon. Figure 1: Dynamic regulation t 1model with failure probability t v. U(x) In =this u(x, wx)vd {(1utility v)t for +a 1} + Â u(x, R)dvector (1 v)xt for the firm game, the+expected horizon stationary input Â u(x, wx)vd t=2 t=1 is given as 2 vd vd (1 v) d (1 v) = u(x, wx) +• + u(x, R) • 1 d + 1 u(x, d (1wx)vd v) t {(1 v)t 11 + d1}(1+ v) u(x, R)d t (1 v)t EU(x) = u(x, wx)vd Â Â vd vd 2 (1 v) d (1 v) t=2 t=1 = r(wx c(y, w)) +2 + (R wx + r(wx c(y, w))) . 1 d (1 v) vd1 d vd1(1 d (1 v) v) d (1 v) = u(x, wx) + + u(x, R) (3) 1 d 1 1 1 d (1 v) 1 d (1 v) vd vd 2 (1 v) d (1 v) The first-order condition respect to x is obtained = r(wx c(y,of w))(3) with + + (R as: wx + r(wx c(y, w))) . 0.8 1 d 1 d (1 v) 1 d (1 v) vs ( d( 0.05) , rho) 2 (3) dEU(x) vd Firm’s v) d (1 v) policy 0.6vd (1 optimal multi-period = rw + w(r 1) . (4) vs ( d( 0.1 , rho) )+ Costinefficient is given dx 1 asd 1 d (1 v) 1 d (1 v) behaviour vs ( d( 0.5) , rho) The first-order condition of (3) 0.4 with respect to x is obtained as: vs ( d( 0.01) , rho) • • • t regime (v = 0), t texpected 1 In the case of a perfectly regulation the utility = Â u(x, EU(x) credible wx)vd + u(x, wx)vd (1 v) + u(x, R)d t (1 v)t 2 (1 v) Â Â dEU(x) vd vd d (1 v) 0.2 ollapses to the expression t=1 t=2 + w(r t=1. = rw + 1) (4) 2 dx 1 d Costefficient 1 d (1 v) 1 d (1 vd (1 v) dv) (1 v) behaviour vd = 0u(x, + d + u(x, R) 0 wx) 0 wx + r(wx 0.4 d 10.2 d 1c(y, (1 0.6v) . 0.8 1 1d (1 v) EU(x) = (R w)) (5) v=0 In the case of a perfectly credible regulation regime (v = 0), the expected utility 0 rho 1 2 d 1 vd vd (1 v) d (1 v) collapses to the expression = r(wx c(y, w)) +v = 0 + (R wx + r(wx c(y, w))) . Optimal response to credible regime: dfor d 1= {0.99, dthat (10.952, 1 d (1 v) all a regulation regime such that Rfailure c(y, w) feasible, av)cost efficient Figure 2: Critical probability v(d ˆ 1, r)meaning 0.909, 0.667}. firm d ideal regulation, (3) ould participate in the game under = a credible model 0).w)) For such EU(x) (R wx + r(wx(v =c(y, . (5) v=0 1 dconsider a flawed e firm response would be full cost efficiency. On the other hand, The first-order condition of (3) with respect to x is obtained as: gulation with certain demise (v = 1). In this case, the expected utility becomes response regime: v = 1 that a cost efficient firm Call a regulation Optimal regime such that Rto non-credible c(y, w) meaning feasible, 2 (1 v) dEU(x) vd vd d (1 v) could participate in the game under a credible = rw model + w(rideal1)regulation, . (4) 27 + d(v = 0). For such dx 1 w)) dOn 1the .dother (1 v) 1 a (6) d (1 v) EU(x) r(wx c(y, the firm response would be v=1 full = cost efficiency. hand, consider flawed 1 d regulation with certainIndemise (v = 1). In this case, the expected the case of a perfectly credible regulationutility regimebecomes (v = 0), the expected utility hus, the firm will follow an optimal policy leading to cost padding, selecting the upper collapses to the expression ound x, ¯ leading to the suboptimal cost wx¯ > c(y, w). Wedstate our findings in the EU(x)v=1 = r(wx c(y, w)) . (6) d roposition below. 1 d EU(x) = (R wx + r(wx c(y, w)) . (5) v=0

1

d

Thus, the firm will follow an optimal policy leading to cost padding, selecting the upper

Model predictions Proposition 1: › The optimal cost policy of a firm in a multi-period policy depends on the probability of regulatory failure (credibility), the time preferences of the firm (impatience) and the utility of inefficient cost (cost of effort).

above. above. Note that the level of the allowed revenue R does not affect the optimal cost policy, thatparticipation the level of the allowed revenue R does not affect the optimal cost policy, butNote only the constraint. above. but only the participation constraint. Corollary 1. Assume a given cost of effort r > 0 and discounting factor d . Then, there exists a finite failure v(d ˆ allowed , r) above which a dominated Note that levelrisk of the does not discounting affectisthe optimal policy,there Corollary 1. the Assume a given cost ofrevenue effort rRcost-efficiency > 0 and factorcost d policy. . Then, but only the participation constraint. exists a finite failure risk v(d ˆ , r) above which cost-efficiency is a dominated policy.the Proof. Follows from the first order condition (4) as a function h(v, d , r). Define

Corollary 1. Assume given effort r> and discounting ., Then, there {v critical Follows failure rate v(d ˆ athe , r)first =cost :ofh(v, d , r) =00}. failurefactor rate v(d ˆ ,Define r) cost-the Proof. from order condition (4) For as aa function h(v,vdd r). exists a finite failure risk v(d ˆ , r) above which cost-efficiency is a dominated policy. minimization is optimal and for v > v(d ˆ , r) the firm has a monotonously increasing critical failure rate v(d ˆ , r) = {v : h(v, d , r) = 0}. For a failure rate v v(d ˆ , r) costutility in x. minimization is optimal and for v > v(d ˆ , r) the firm has a monotonously increasing Proof. Follows from the first order condition (4) as a function h(v, d , r). Define the {v critical failure rate v(d ˆ , r) = : h(v, d , r) = 0}. For a failure rate v v(d ˆ , r) costutility in x. The function v(d ˆ , r) is illustrated in Figure 2. As a consequence of Corollary 1, minimization and for v > will v(d ˆ ,select r) thethe firm haslevel a monotonously increasing the The firm function facingis aoptimal credible regulation input x? , giving aof cost efficient 1, v(d ˆ , r) is illustrated in Figure 2. As a consequence Corollary utility in x. ? level wxfacing = c(y,aw). For a non-credible regime higher failure risk v >a v(d ˆcost, r), the the firm credible regulation will selectwith theainput level x? , giving efficient firm will adopt an input maximization behaviour, that is selecting the upper bound ¯the v(d ˆ For , r) aisnon-credible illustrated in regime Figure with 2. Asa ahigher consequence of Corollary 1, x. levelThe wx?function = c(y, w). failure risk v > v(d ˆ , r), ? Thefirm associated wx¯ >regulation c(y, w) implies cost the inefficiency firm.a As anefficient example, the facing credible will select input x the , giving firm will adoptacost an input maximization behaviour, thatlevel isby selecting thecost upper bound x. ¯ ?a imagine firm withFor a alow cost of effort (r with = 0.2) and a failure discount factor d ,= 0.99%. level wx = c(y, w). non-credible regime a higher risk v > v(d ˆ r), the The associated cost wx¯ > c(y, w) implies cost inefficiency by the firm. As an example, Thenwill in Figure 2 the firm would be cost minimizing any regime failure x. firm adopt an input maximization behaviour, that for is selecting the with uppera bound ¯rate imagine a firm with a low cost of effort (r = 0.2) and a discount factor d = 0.99%. above. less associated than v(d ˆ =cost 0.99, 0.2) 0.173.cost On inefficiency the other hand, a firm a high cost of The wx¯r>=c(y, w)=implies by the firm.with As an example, Then in Figure 2 thea firm would be cost(rminimizing for any regime with a 0.99%. failure rate imagine athat firm with low of effort = and a discount factor d were = policy, effort (e.g. rthe= 0.8) of would only abstain inefficiency if the policycost almost Note level thecost allowed revenuefrom R 0.2) does not affect the optimal less than v(d ˆ = 0.99, r = 0.2) = 0.173. On the other hand, a firm with a high cost of Then in Figure 2 the firmvconstraint. would be0.99, cost minimizing for any regime with a failure rate impossible overturn, < v(d ˆ = r = 0.8) = 0.004. but only the to participation effort (e.g. 0.8)r would abstain inefficiency if the almost less than v(d ˆ r==0.99, = 0.2) only = 0.173. On from the other hand, a firm withpolicy a highwere cost of Corollary 2. Assume a non-credible regime v > 0 and a given cost of effort r. Then, Corollary 1. Assume a given cost of effort r > 0 and discounting factor d . Then, there impossible v < v(d ˆ =abstain 0.99, rfrom = 0.8) = 0.004.if the policy were almost effort (e.g. to r overturn, = 0.8) would only inefficiency for any cost-efficient firmv(d there exists an upper bound dˆ foristhe discount factor. exists a finite failure risk ˆ , r) above which cost-efficiency a dominated policy. impossible to v(d ˆ = 0.99, r = 0.8)v = Corollary 2. overturn, Assume av < non-credible regime >0.004. 0 and a given cost of effort r. Then, Proof. The bound obtained from thean fixed point v( ˆa dˆfunction , dr) = v. Proof. fromis the first order condition asand h(v,discount dof, r). Define the for any Follows cost-efficient firm there exists upper for the factor. Corollary 2. Assume a non-credible regime v (4) > 0bound aˆgiven cost effort r. Then, critical rate v(d ˆ firm = {vexists : h(v,an dregime , upper r) = v0}. failure rate v factor. v(d ˆfactor , r) dcostfor any failure cost-efficient there bound dˆafor the discount Corollary Assume a, r) non-credible > 0For and given discount . Then, Proof. The3.bound is obtained from the fixed point v( ˆhas dˆa, r) = v. minimization is optimal and for v > v(d ˆ , r) the firm a monotonously increasing ˆ for any cost-efficient firm there exists an upper bound r for the cost of effort. Proof. bound is obtained from the fixed point v( ˆ dˆ , r) = v. utility inThe x. 3. Corollary Assume a non-credible regime v > 0 and a given discount factor d . Then, Proof. Follows directly from the inverse function of the critical failure rate v(d ˆd . Then, , r), the Corollary 3. Assume a non-credible regime v > 0bound and a given ˆconsequence for any firm there exists upper for discount the costoffactor ofCorollary effort. Theiscost-efficient function v(d ˆ as, r) is illustrated inan Figure 2.unique As a rand ˆ ˆ bound obtained v(d ˆ , r) = v. Note that r is bounded for all v > 0. 1, for any cost-efficient firm there exists an upper bound rˆ for the?cost of effort. the firm facing a credible regulation will select the input level x , giving a cost efficient Proof. Follows directly from the inverse function of critical failure rate v(d ˆ is, r), Corollary aafrom non-credible regime v > of 0. Thethe cost efficiency forv(d firm thenthe ? = 4. Proof. Follows directly the inverse function the critical failure ˆaall r), the level wxis c(y,Assume w). For non-credible regime with a higher failure risk vrate > ˆv(d ,,r), the ˆ ˆ bound obtained as v(d ˆ , r) = v. Note that r is unique and bounded for v > 0. inversely proportional to the discount factor d and the cost of effort r.

Corollaries

ˆ = v. Note that bound is obtained v(d ˆ ,maximization r) rˆ is unique boundedtheforupper all v > 0. x. firm will adopt an as input behaviour, that isand selecting bound ¯

Corollary 4. Assume non-credible regime v >the 0. Theby cost fororder a firm is then The associated cost wx¯aa> c(y, w)Proposition implies cost1v inefficiency theefficiency firm. As example, Proof. Follows directly from and for thefor first condiCorollary 4. Assume non-credible regime > 0. Thefunction cost efficiency aanfirm is then inversely proportional tonecessary the discount factor d0.2) and the of effort imagine firm low cost of effort (r = and a cost discount factor = 0.99%. tions (4),aproportional h(v, dwith , r). a A condition athe cost-efficient optimal by the inversely to the discount factor d for and cost of effort r. r.d policy Then in Figure 2 the firm would be cost minimizing for any regime with a failure rate firm is that h(v, d , r) 0, inducing cost-minimization with the optimal solution x =condix? . Proof. Follows directly directlyfrom fromProposition Proposition 1 and the function for the first order Proof. Follows 1 and the function for the first order condiless than ˆh(v, =d0.99, rAthere =necessary 0.2)is=an0.173. On thefor other hand, a firm with highpolicy cost2ofby Corollary 3 gives that upper bound rˆ for h given v and d . aCorollary protions (4),v(d ,r). r). condition a cost-efficient optimal tions (4), h(v, d , A necessary condition for a cost-efficient optimal policy by the the ˆ effort (e.g. r = 0.8) would only abstain from inefficiency ifConsider the policy were almost vides analogously an upper bound d for h given v and r. the curvature of h ? firm that h(v, h(v,dd,,r) r)0,0,inducing inducingcost-minimization cost-minimization with optimal solution firm is is that with the the optimal solution x = xx? .= x . impossible to overturn, v < v(d ˆ = 0.99, r = 0.8) = 0.004. with respect to r:that Corollary gives thatthere thereisisananupper upper bound rˆ for h given v and d . Corollary 2 proCorollary 33 gives bound rˆ for h given v and d . Corollary 2 proˆ ˆ Corollary 2. Assume a non-credible regime v > 0 and a given cost of effort r. Then, vides analogously an upper bound d for h given v and r. Consider the curvature vides analogously an upper bound d for h given v r. Consider the curvature of h of h 2 (1 v) + d (1 v d 2firm EU(x) vd vd bound ˆ for any cost-efficient there exists an upper d for the discount factor. with respect =w + >0 (7) with respect to tor: r: dxdr 1 d 1 dˆ (1 v) the fixed point v( Proof. The bound is obtained from ˆ d , r) = v.

0.8

dv( 0.2 , rho) 0.6 dv( 0.4 , rho) dv( 0.6 , rho) 0.4

0.2

There will always some laggards … 0

0

0.2

0.4

0.6

0.8

1

rho

ˆ for v = 0.2 and r = {0.1, 0.2, ..., 0.9} Figure 3: The indifference curve dˆ (r)

As the credibility of the regulation increases, v decreases and the set (area) in Figure 3 increases. Remark 1. Given n independent firms each having a cost of effort drawn from a distribution with density function f (r) and cumulative density function F(r) on the support [0, 1], then the probability that all firms are cost efficient under a non-credible regime is ˆ n. equal to 1 (F(r)) U(x,R)

U(x,R)

1-v

1-v

U(x,R)

1-v

The intuition behind Remark 1 is clear: the hope of incentivizing all firms to efficiency in a weak regulation regime is thin. In practice, there will always be inefficient 1-v firms in the set of regulated operators, see Figure 4. Thus, the empirical conjecture v would then v be a higher vincidence of non-profitmaximizing behavior from firms that have a plausible case of a failing regulatory regime. In particular, firms with stable semi-public ownership can represent the case of longv 1 1 of effort. 1This is frequently range time preferences and high cost the case for energy distribution in Europe. The opposite extreme would be a set of privately owned franU(x,wx) U(x,wx) U(x,wx) chisees in countries with high inflation or political risks, here the time preferences is short-run and the relative cost we test the validity of t = 1of effort low. t = 2 In the nextt section, =3 our model on an interesting case of regulatory failure in Scandinavia, a region otherFigure 1: Dynamic regulation model with failuresolutions, probability v. wise characterized by early adoption of market-oriented cf. Amundsen and Bergman (2007). 4. Empirical model The previously presented model leads to a number of empirically verifiable hypotheCritical failure probability ses for non-credible regimes; in particular that firms exhibit lower cost efficiency fol1

1

11

0.8 vs ( d( 0.05) , rho) vs ( d( 0.1 , rho) )

0.6 Costinefficient behaviour

vs ( d( 0.5) , rho) vs ( d( 0.01) , rho)

0.4

0.2 Costefficient behaviour

0

0

0

0.2

0.4

0

0.6 rho

0.8

1 1

Figure 2: Critical failure probability v(d ˆ , r) for d = {0.99, 0.952, 0.909, 0.667}.

27

6. Analysis Initially, we observe that the facts in the case correspond relatively well to those

Parttheoretical IV of the model. First, given the mixed ownership situation among the oper-

ators (50 per cent privately owned20 , 40 per cent publicly owned21 and 10 per cent consumer-cooperatives, (Agrell et al., 2005b)), we can safely assume that there is some heterogeneity with respect to the cost of effort r in the sample. At least some firms should have a strictly positive r, meaning that they are not exclusively profit maximizing. Second, given the lukewarm reception of the new regulatory model and the industrial and academic critique raised against it, we safely assume that all firms observed a high probability of regime failure v >> 0. Without adventuring into what value each firm (or group of firms) attributed to v, it suffices to recall our results from Remark 1 6. Analysis where adverse effects appear already from modest levels of v. Initially, we observe that the facts in the case correspond relatively well to those For clarity, we restate our hypotheses and proceed into the analysis step by step. of the theoretical model. First, the drop mixedinownership situation among oper-is virlaunch of the NPAM, there is agiven radical technological change, thethe value 20 , 40 per cent publicly owned21 and 10 per cent ators (50 per cent privately owned tually at its floor of unity (0.1 per cent). The difference in mean technical Hypothesis 1. Firms exhibit a lower cost efficiency CE during a non-crediblechange regimeis consumer-cooperatives, (Agrell et al., 2005b)), we can safely assume that there is some and of the expected sign. Looking closer at the data in Figure 6 shows a vsignificant > 0. heterogeneity with at respect to the In costthe ofanticipation effort r in the sample. least some model, firms alstriking difference firm level. of the fall ofAtthe NPAM t for the should have a strictly positive r, meaning that they are not exclusively profit maximizThe average cost efficiency CE operators by year is listed in Table 2 below. most no operator shows technological progress (the black curve). As predicted by the ing.note Second, given the lukewarm receptionefficiency of the newduring regulatory model and theonindusWe a clear tendency of decreasing the period, from average model, the absence of incentives is leading to a stand-still of the investments in new trial per and cent academic critique raisedto against per it, we safely assume that all firms observed a 74.5 before the NPAM cent during the NPAM. In fact, the overall technology and processes, resulting71.9 in the observed stagnation of technological change. highinprobability of regime v >> year 0. Without valueis each fall cost efficiency fromfailure the initial (2000)adventuring to the last into yearwhat (2006) 5.4 per Hypothesis 2 isofnot rejected by the data. firm (or firms) v, it suffices to per recall our is results from Remark 1 cent. Thegroup difference in attributed mean costtoefficiency (2.6 cent) statistically significant. where adverse effects appear already from modest levels Hypothesis 3. NPAM, The for productivity of theinin firms isoflow or nilchange, for red the curve duration of Additional support this finding is found Figure 5,v.where the shows launch of the there is achange radical drop technological the value is avirt For clarity, we restate our hypotheses and proceed into the analysis step by step. non-credible regime > 0.before the mean CE floor by operator the NPAM and the black curve depicts the analogous tually at its ofvunity (0.1 per cent). The difference in mean technical change is cost efficiency after the NPAM introduction. As seen in the Figure, the fall in cost a significant and of theexhibit expected sign. closer at the data in Figure 6(13), shows Hypothesis 1. Firms a calculated lower costLooking efficiency CE during a non-credible regime The productivity change, as the right-hand side of expression is 22 efficiency is generalized, except forInsome initially highly inefficient operators . Hence, striking difference at firm level. the anticipation of the fall of the NPAM model, v > 0. presented in Table 4. As a consequence of the previous results in terms of cost efficiency alwe conclude that Hypothesis 1 is not rejected by the data. most no operator shows technological progress (the black curve). As by and technological change, the productivity change prior toisthe reform waspredicted positive andthe t The average cost efficiency CE for the operators by year listed in Table 2 below. model, on the absence ofper incentives is leading to aintroduction, ofthe the investments strong, 3.5 per year. After the productivity change Hypothesis 2. The technical change ofefficiency the firms isstand-still stagnating for the duration ofinanew We note a average clear tendency ofcent decreasing during the period, from on average technology and processes, in the observed stagnation of is on per average negative perto cent). thetechnological means of thechange. two non-credible vthe >(-0.7 0. resulting 74.5 centregime before NPAM 71.9 perThe centdifference during thebetween NPAM. In fact, the overall Hypothesis 2 is not rejected by the data. periods is significant and of the right sign. Presented in more detail in Figure 7, fall in cost efficiency from the initial year (2000) to the last year (2006) is 5.4 per the The technical change DTC and the variation in cost efficiency DCE from the methodfindings aredifference without any ambiguity. Here note that of theof a cent. The in mean cost change efficiency (2.6 per cent) is confidence statistically significant. Hypothesis 3. The productivity ofwe the firms isthe low or nil for intervals the 23 .duration ological framework, expression (17), are calculated and reported in Table 3 The Additional support for this finding is found in Figure 5, where the red curve shows three years during thev > reform are below the horizon average productivity changepre(1.4 non-credible regime 0. reform technological change ratetheisNPAM strong, on the average 4.8 per cent per year. At the t by operator the cent). mean CE before and black curve depicts the analogous per Thus, we find that Hypothesis 3 cannot be rejected. productivity theseen right-hand side oftheexpression costThe efficiency after thechange, NPAM calculated introduction.as As in the Figure, fall in cost(13), is Hypothesis 4. The profitability of the firms is lower on average, and decreasing through22 . cost presented in Table 4. As a consequence of the previous results in terms efficiency efficiency is generalized, except forassome inefficient operatorsof Hence, 20 Including international firms such EDF, initially owner of highly Graninge, and the inter-Nordic operator Birka. out the duration of a non-credible regime v > 0. we21conclude that Hypothesis 1 is not rejected the data. and technological change, the productivity change prior to the reform was positive and Mostly municipal utilities with the exception of by the (then) state-owned operator Vattenfall. 22 An inefficient policy under a credible regime may be optimalin atthe very high of effort strong, average 3.5 per cent year. After theis introduction, productivity change The on average profitability Pt per by regulatory firm and year presented Table 2 cost below. The Hypothesis 2.theThe technical change firms isofstagnating for duration of to a this r, in which case policy is independent ofofthethe credibility the regime. Our the results are robust is on average negative (-0.7 per cent). The difference between the means of the findings document significant and monotonous fall in profitability over the period,two non-credible regime av > 0. assumption. periods is significant andcent of the right sign. Presented in more detail in Figure 7,Atthe 23 The from on results average per to the NPAM to 10.5the perperiod centwithout after the for 14.7 DSC from (17) prior are equal to unity throughout anyreform. significant The technical and the variation in cost methodfindings are without anyDTC ambiguity. Here we note that theDCE confidence intervals differences. operator-level, thechange results are illustrated in Figure 8.efficiency Hypothesis 4 from is notthe rejected by of thethe 23 . The preological framework, expression (17), are calculated and reported in Table 3 three years during the reform are below the horizon average productivity change (1.4 empirical data. reform technological change rate is strong, on average 4.8 per cent per year. At the per The cent). Thus, that Hypothesis 3 cannotpresented be rejected. 18 result canwe byfind explained by the previously methodology. Profitability variation PV 4.is The driven by productivity change and price recoveryand PR,decreasing as shown in exHypothesis profitability of the firms is lower on average, through-

VERIFIABLE HYPOTHESES

Research hypotheses

Part V

EMPIRICAL METHOD

Objective We are interested in a framework that links › › › ›

Profitability changes Cost changes Revenue changes Efficiency changes

Productivity development Simple approach: efficiency changes vs index ? Not conclusive, since price changes may be due to › Input price changes (price recovery) › Output price changes (profit margin) › Economies of scale (volume) › Allocative efficiency (mix) and, as a consequence, thatefficiency the productivity › Technical changes of the sector will suffer. In order to investi-

gate this phenomenon, we need both a test-case with a failed regime and a methodology that is capable of differentiating between profitability, cost efficiency and dynamic efNeed decomposed analysis fects for multi-input, multi-output production. For the empirical case, we collected data for the period before and up to the failing of a transient regulation regime in Sweden, described in more detail in Section 5. The methodology for the study of this empirical case is based on profitability6 , the firm financial indicator defined by the ratio revenue to cost P = py/wx. GeorgescuRoegen (1951) introduced profitability, called return to the dollar, as a financial performance indicator into the economic literature. It is independent of the scale of production, a virtue not shared by cost, revenue or profit measures. This property of independence of the scale of production is particularly relevant in sectors with a wide range in the size of operation. Moreover, it allows for the direct comparison between the remuneration from the regulator (R = py) and the observed cost of the firm (wx). We are Profitability change interested in the study of the evolution throughout time of the ratio revenue to cost, i.e. profitability change. It is defined as Pt+1 = Pt

pt+1 yt+1 /wt+1 xt+1 pt yt /wt xt pt+1 yt+1 /pt yt Revenue change = , wt+1 xt+1 /wt xt Cost change

(10)

which is equal to the ratio of revenue change to cost change. The next step is to identify the factors that cause changes in profitability. These factors are associated with changes in quantities and prices of individual outputs and inputs. Hence, we want to isolate the changes in prices of the changes in quantities, either of which influences profitability change. We can decompose cost change in (10) as wt+1 xt+1 wt+1 xt+1 wt xt+1 = wt xt wt xt+1 wt xt wt+1 xt wt+1 xt+1

(11)

s equal to the product of a Laspeyres input price index and a Paasche input quantit ndex. Both these pairings satisfy the product test. Laspeyres-Paasche index numbe airs are widely used, but because they use different weights, they generate differen esults. The third line solves this problem because it takes the geometric mean of th Laspeyres-Paasche index number pairs and creates a Fisher input price index and Fisher input quantity index7 . The last line of (11) defines a more compact notation as Revenue change Fisher input price index WF and a corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for rev nue change as pt+1 yt+1 pt yt

Fisher output price index

= PF (pt+1 , pt , yt+1 , yt )YF (yt+1 , yt , pt+1 , pt ),

(12

Fisher output quantity index

where PF is called a Fisher output price index and yF defines a Fisher output quantit ndex. Combining (11) and (12) yields an expression for the relative change in prof tability, pt+1 yt+1 /wt+1 xt+1 PF (pt+1 , pt , yt+1 , yt ) YF (yt+1 , yt , pt+1 , pt ) = . pt yt /wt xt WF (wt+1 , wt , xt+1 , xt ) XF (xt+1 , xt , wt+1 , wt )

(13

This profitability change is the product of a price recovery term PR for output an nput prices, respectively, and the Fisher productivity change index YF /XF for outpu nd input quantities, respectively. The price recovery PR compares the variation in th rices on the inputs with the corresponding input price changes between two periods The interpretation of PR is intuitive: a value lower than unity implies that the firm ha which is equal to the ratio of revenue change to cost change. The next step is to ncreasedidentify the output prices theininput price changes. Under high-powered regu the factors that less causethan changes profitability. These factors are associated with change ation, where theinoutput prices are fixed, this would expected changes quantities and prices of Cost individual outputsbe andthe inputs. Hence,outcome. we want to Inversely isolate the changesaninincrease prices of the changes in quantities, either of which influences PR > 1 would indicate of the output prices compared to the input prices, i.e profitability change.margin. We can decompose cost change in (10) as n increase of the gross We are interested in the decomposition of the Fisher productivity index by its eco t+1 t+1 t+1 t 1/2 t+1 t+1 t t+1 1/2 wt+1 xt+1 objectwof xattention w x by Diewert w x w x omic drivers. It has been (2014) (see also(11) Grifell-Tatj = t xt+1 t+1 xt wt xt wt xt (2003,w2015b); wtKuosmanen xt nd Lovell (2015a)), Grifell-Tatj´ e wand Lovell and Sipil¨aine t+1 t t+1 t t+1 t t+1 t = WF (wWe, wfollow , x , xGrifell-Tatj´ )XF (x , x , we and , w )Lovell (2015b) that ca 2009); Ray and Mukherjee (1996). e relatedExpression to the approach of geometric Ray Mukherjee (1996). conventional approach t Fisherand input priceofindex Fisher input An quantity index (11) takes the mean the Laspeyres-Paasche index number pairs eal withand thecreates generic production possibilities introduced in Section 3.1 is a Fisher input price index and a set Fisher input quantity index.6 The lastdefined b line of (11)programming defines a more compact as aetFisher input price indexon WFthe andData a he mathematical modelsnotation in F¨are al. (1985), based Envel corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for revenue change as

7 The

Fisher index has a t+1 set of t+1appealing axiomatic properties, see Balk (2012) p

y pt yt

= PF (pt+1 , pt , yt+1 , yt )YF (yt+1 , yt , pt+1 , pt ),

(12)

and creates a Fisher input price index and a Fisher input quantity index.6 The last line of (11) defines a more compact notation as a Fisher input price index WF and a corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for revenue change as

echnology, which corresponds to the expected behav decomposition to (16) can be done based on the Paas Fisher productivity index in (13). Here we obtain t p y = P (p , p , y , y )Y (y , y , p , p ), (12) py efficiency change, thechange technical change and the si Relative in profitability where P is called a Fisher output price index and Y defines a Fisher output quantity size change, has(11)the same as above, w index. Combining and (12) yields aninterpretation expression for the relative change in profitability, Profitability change(1978). Its empirical Fisher productivity by Charnes et by al. unded below unity. Thus, taking the geometric m P P (p , p , y , y ) Y (y , y , p , p ) = . (13) P W (w index , w , x , xand ) X (x the , x , w Paasche ,w ) aspeyres productivity product recovery This profitability change is thePrice product of a Fisher price recovery index P /W for ) on of the productivity index. We output Fisher and input prices, respectively, and the Fisher productivity indexobtain Y /X for output t+1 t+1

t+1

F

t t

t

t+1

t

F

F

t+1

t

t+1

t

F

t+1 t

F

F

t+1

t+1

t

t+1

t

t+1

t

F

t

F

t+1

t

t+1

t

t+1

t

t+1

t

F

F

F

and input quantities, respectively. The price recovery index compares the variation in s Ray s the s with the andinputs Mukherjee (1996), Kousmanen input and Sipiläinen (2009), between two periods. the prices on corresponding price changes i i i Diewert (2014), Grifell-Tatje Lovell (2003, 2015) The interpretation of price recoveryand index is intuitive: a value lower than unity implies st that the firm has increased the output prices less than the input price changes. Under t high-powered t+1 regulation, t t bet the expected where the output pricestaretfixed,tthis would outcome. Inversely, a price recovery index higher one than indicates an increase of the prices compared to the input prices. t outputt+1 t t+1 t+1 t+1 t+1 t+1 We are interested in the decomposition of the Fisher productivity index by its economic drivers. It has been object of attention by Diewert (2014) (see also Grifell-Tatj´e

ÂÂl i

F

y ,l

0 ,

(14)

+1 , y

,p ,p ) w x /c (y , w ) = s, respectively. ector for firm i in period +1 , x , w ,w ) w x /c (y , w ) ant returns to scale, which is a common t t t b and Pollitt (2003). Additionally, c (y , w ) thect (yt+1 , wt+1 ) The Fisher index a set of appealing axiomatic properties, inear surface, orhas frontier, over the t , wsee tdata t+1 (yt+1 , wt+1 ) ct+1 (y )Balkc(2012) from year 1 to year t inclusive. 2 13 Wet obt+ t+1 6

p py Fisherallow productivity index al way which does not for technit yt p p 4 t t t t t t t t re c (y , w ) = minx {w x : (xw,t xyt+1 )2 T }(yt ,wt ) wt xt / /ct+1 tt+1 tt+1 tt+1 t t+1 t+1 fined as CE t (yt , wt ) = cCostt (y , w )/w x w x /c (ychange ,w ) wt+1 xt /c Size efficiency vity index YF /XF = in (13) based· DSC on DCEis· DTC Technical Fisheroutput productivity quantities because the areefficiency conpeyresthe input quantity index component term, cost efficiency change DCE indicates whet Efficiency measured using non-parametric approach (DEA) 2 outputs (energy LV, HV,) 4 inputs (assetconnections, grid capital, cost OM, energy losses, energy transit)

between periods t and t + 1. The second term, the

wt )

wt xt+1 /ct+1 (yt , wt )

, (15)

Part VI

FULL DATASET

The Network Performance Assessment Model Electricity Act (2000) › Regulated revenue based on “objective performance”

NPAM › Green-field planning model, based on GIS-positioned load points, feedin points, standard costs › Critique from industry and academics, model suffers from several methodological flaws (Lantz, 2003; Wennerström and Bertling, 2008; Jamasb and Pollitt, 2008, Jamasb and Söderberg, 2008)

Green-field vs brown-field planning

NPAM rise and fall 2003 Start of implementation 2005 Rulings I for 2003 = 21 DSO for 76,3 MEUR › All DSO appeal

2006 Reduced claims for 2003: 8 DSO for 23 MEUR › DSO appeal to higher court

2007 New regulator › Out-of-court settlement: 8 DSO for 16.5 MEUR.

2009 NPAM suspended (cost-recovery) 2012 New regime: rate-of-return

Data DSO audited data from the regulator (EI) for Swedish electricity distributors (LV and MV only, no generation, retail or transmission) Balanced panel, 128 firms for 2000-2006, in all 896 DMU

Data: DSO 2000-2006 Table 1: Descriptive statistics and model variables. Category

Unit

Definition

mean

median

sd

Revenue R = py

kSEK kSEK kSEK

Total revenue Revenue LV Revenue HV

137,764 118,394 19,371

49,967 41,876 6,707

387,118 335,470 53,213

Costs wx

kSEK kSEK kSEK kSEK kSEK

Total cost (TOTEX) Cost transmission Cost energy losses Operating expenditure (OPEX) Capital expenditure (CAPEX)

119,515 33,791 7,878 46,766 31,082

46,483 13,285 2,864 18,615 8,602

346,036 100,420 21,395 130,483 102,922

MWh MWh

Energy delivered low voltage (LV) Energy delivered high voltage (HV)

488,052 221,633

204,662 71,037

1,235,396 623,509

SEK/kWh SEK/kWh

Price per energy delivered LV Price per energy delivered HV

0.228 0.109

0.226 0.104

0.043 0.057

MWh MWh km kSEK

Energy transported, total Energy losses, total Connection-weighted network LV+HV Network capital, total

742,112 32,427 41,415 458,831

281,796 11,952 14,198 100,737

1,913,920 86,027 121,128 1,521,204

SEK/kWh SEK/kWh SEK/m %

Transmission price Cost per energy losses OPEX per connection-line unit Cost of capital

0.049 0.260 1.379 0.086

0.048 0.252 1.332 0.083

0.019 0.120 0.543 0.033

Outputs y

Output prices p

Inputs x

Input prices w

Part VII

EMPIRICAL TEST

H1: Slumping cost efficiency H4: Profitability sacrifice

Table 2: Profitability Pt and cost efficiency CE t , mean per year, 2000-2006. year Pt

CE t

period

2000

2001

2002

2003

2004

2005

2006

2000-02

2003-06

Diff

1.150 0.762

1.149 0.732

1.141 0.741

1.128 0.732

1.128 0.723

1.086 0.713

1.079 0.708

1.147 0.745

1.105 0.719

-0.042*** -0.026***

Notes: ***p < 0.001; **p < 0.05; *p < 0.01.

Table 3: Cost efficiency DCE and technology change DTC, before and after NPAM. n period

All

Pre NPAM

Post NPAM

768 2000-2006 Mean SD

384 2000-2002 Mean SD

384 2003-2006 Mean SD

384 Diff

p-value

Table 2: Profitability Pt and cost efficiency CE t , mean per year, 2000-2006. year Pt

CE t

period

2000

2001

2002

2003

2004

2005

2006

2000-02

2003-06

Diff

1.150 0.762

1.149 0.732

1.141 0.741

1.128 0.732

1.128

1.086

1.079

1.147

1.105 0.719

-0.042*** -0.026***

0.723 0.713 efficiency 0.708 0.745 H1: Cost

Notes: ***p < 0.001; **p < 0.05; *p < 0.01. 1.0

C E 2000 2002 C E 2003 2006

●●●● ● ● ●●●

0.6

0.7

CE

0.8

0.9

●● ●● ● ●● ● ●● ●● ●● ● ● ● ● ● ●●● ●●● ●● ● ●● ●●● ●●●●● ● ●●●●● ●● ●●● ● ● ●● ●● ●●●● ●● ●●● ●●● ●●● ●●● ●●●● ●●●●● ●●● ● ● ●● ● ●●● ● 0.745 ● ●●●● ●●●●● ●●● ●●●●●● ● ●●● ●●●● ●●●●●● ●●● ●●● ●● ●●●● ● 0.719 ●●●●●● ●●●●● ●●● ●●●● ●●●●●●● ●●● ●● ●●●●● ●● ●●●● ●●●●●● ● ●● ●●●●● ● ●●● ●●●● ● ●● ●● ●●●● ●● ● ●●●●● ●●● ● ● ●

0.5

● ● ●●● ●● ● ●● ● ● ●●● ● ●● ●● ●

0

20

40

60

80

100

120

D S O (sorted) U C L/C O R E/Agrell uab.stemdata.rev2 2015 07 23

Figure 5: Cost efficiency CE t , average per DSO, development before and after NPAM.

ological framework, expression (17),cost are efficiency calculatedCE andt ,reported in year, Table2000-2006. 3.20 The preTable 2: Profitability Pt and mean per reform technological change rate is strong, on average 4.8 per cent per year. At the year period launch of the NPAM, there is a radical drop in technological change, the value is vir2000 2005 2006 in mean 2000-02 Diff tually at2001 its floor2002 of unity2003 (0.1 per 2004 cent). The difference technical2003-06 change is significant and of the expected sign. Looking closer at the data in Figure 6 shows a Pt 1.150 1.149 1.141 1.128 1.128 1.086 1.079 1.147 1.105 -0.042*** striking difference at firm level. In the anticipation of the fall of the NPAM model, alt CE 0.762 0.732 0.741 0.732 0.723 0.713 0.708 0.745 0.719 -0.026*** most no operator shows technological progress (the black curve). As predicted by the Notes: model, ***p

UCL/CORE UAB, Spain

NAPW 2016, Quebec City, June 14-18, 2016

Reference This presentation draws on Agrell, P J and E Grifell-Tatjé (2016), A Dynamic Model for Firm-Response to Non-Credible Incentive Regulation Regimes, Energy Policy 90, 287–299.

Outline Incentive regulation Game theoretic model Empirical method Empirical testing Results

Part 1

YARDSTICK REGULATION

Incentive regulation in a nutshell Regulator Services y

Contract M(y,t)

Operator

Effort e

Cost C(y,e) Infrastructure access, unbundled firm, inelastic demand for service Cost is observable and verifiable, effort is unobservable, multi-output service provision High-powered regulation is optimal: Laffont (1994), et al. Practical implementations: yardstick regimes: Schleifer (1985), Laffont and Tirole (1986)

Yardstick design and DEA Bogetoft (1994) › A cost norm built on DEA provides optimal incentives for deterministic models, and stochastic output under some conditions (truncation property and half-normal, exponential distributions)

Agrell, Bogetoft and Tind (2005) › A DEA-based cost norm is optimal also in a dynamic setting, the optimal implementation has a linear form for linear cost of effort.

DEA minimizes the rent given to the firm

EU Regulatory landscape (Energy) EU Regulatory landscape – Methods (Energy)

IS

Cost recovery Revenue Cap (CPI-X) Revenue Cap (CPI-DEA/SFA) Finland

Norway

Yardstick - DEA

Sweden Estonia

IRL

Latvia

DK

GB

Lithuania NL BE.

Poland

Germany

AT SL

HU HR

BH Portugal

Spain

Yardstick - MNA

Sweden under re-reform: Rate-of-return regulation

Switzerland under reform: incentive regulation (DEA pilot)

SK CH

Yardstick - Other

Belgium under re-reform: De facto cost-recovery

CZ France

Price-Cap

Iceland: reform not implemented

RO

Finland: revenue cap with StonED

SB

BG

Italy MN. FYROM. AL

Greece

Normative models are popular Country AUSTRALIA AUSTRIA DENMARK FINLAND GERMANY NETHERLANDS NEW ZEELAND NORWAY ICELAND PORTUGAL CHILE SPAIN ENGLAND BELGIUM SWITZERLAND SWEDEN 7

Approach Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante Ex ante

Method CPI-DEA DEA/EngM COLS DEA->StonED DEA/SFA Yard Cost Yard CPI-DEA DEA Yard CPI-DEA SFA EngM EngM CPI-X CPI-DEA -> CR (RoR)->? (EngM)->RoR

Analysis x x x x x x x x x x x x x x x x

Operation x x x x x x x x ? x x x x

Irrelevance of cost norm Revenue cap = R0 CPI (1 – X – Xi) Incentive regulation, corollaries › A profitmaximizing firm do not care about the level of the cap › A utilitymaximizing firm cares about the incentive power › What matters are the commitment to and duration of the regime › No importance of the used cost norm

Does it hold in practice? The regulation is based on the cost norm Regulation must hold for all firms without bias It is not sufficient to be right on expectation Judicial recourse to protect from expropriation › Firms may appeal rulings › If a ruling shows a flaw in the model, the regime falls

Part II

ANECDOTAL EVIDENCE

Failing regulation in Europe Netherlands › Frontier model revoked 2004, debacle 140 M in welfare losses › Nillesen and Pollitt (2007) › Moratorium and average cost model

Belgium › Preparation for incentive regulation, overturned and decentralized in 2012 › Agrell and Teusch (2015) › Cost-plus regulation by region since 2012 …

Sweden › Network performance assessment model (NAPM) falls in 2006 › Moratorium and cost-plus regulation until 2014 …

Credibility Commitment is based on a rational expectation of durability The robustness of a regulation depends on › Participation of the regulated firms › Sustainability of rents left to stakeholders › Properties of the cost norm (soundness)

A regulation regime not satisfying these criteria is not credible

Idea Intuition: › A rational firm reveals only its full efficiency for a regime with a credible commitment and cost norm.

Method: › Decision model for a firm evaluating a proposed regime › Methodology to test the hypotheses for firm behavior › Validation with productivity data for a failed regime

Feasible and infeasible cost norms CAPEX/output

A

Benchmarking model Statistical model Normative model True frontier

OPEX/output

Part 1II

MODEL

information R(y) that assures full participation of all efficient firms in the tempting in long run and a regulatory commitment structure conducive n possibility to sustain truthful revelation of information by firms for the wever, given operation of the regulation. Assume now that a firm facing a in network non-credible high-powered regime R considers the probability e estimation of it falling to v per year and that failed regime is replaced Model models are by a low-powered regime. Let us assume ex-post verifiable with varying cost and single-dimensional service data as xt , yt for year t. regulated are used toOneThe firm isfirm maximizing horizon utility consisting of profit and able, cost forMulti-period game, discount factor slack using a discount factor . The initial situation is that of Furthermore, wx is the observed cost andR R = 0 Slack will=observe that high-powered regime where r 2 [0,optimal 1[. The slack is here an convenient expression for the acase of linear cost of effort, such that 1 euro of surplus cost (for example poor procurement policies, del. The risk is non-credible if it violates the participation constraint for obsolescence, luxurious equipment) has the value of r euro in disposable profits (that s maximizes in a For concession or licensing by imposing a could be paid any as e.g.firm dividends). a cost-plus policy with R = wx,system equation (1) simplifies to reimbursement R < x? . Note that the failure of a regime is not u(x, wx) = r(wx c(y, w)). (2) inimum loss instantaneous, its downfall is brought by sometimes lengthy The expected horizon utility is the sum of the discounted single-period utilities with a cted from a factor judicial theperiod inefficient incumbent discounting d 2 ]0, 1[,appeals meaning thatand 1 eurolobbying. at the end of of If the first is worth euro at the start of the horizon. The reservation utility for the firm is normalized to ely tod result zero. The firm does not participate unless the expected utility is higher or equal to the considerable reservation utility. Regulatory game system can 3.2. Dynamic game ? observes a regime R x for all firms, the optimal policy may be conThe regulator and the firm are playing an infinite horizon game for a stationary dePeriod 1:bew.toInitially, will optimize costselects as toa regime achieve = inx? , leaving no {R, v} c(y) y and fixed prices the regulator consisting isted. mand If the a revenue levelno derived some (e.g. average cost or ideal › R Launch of⇡high-powered regime R(y) profitusing = 0 methodology but infinite participation. If,network on the other hand, s candidates models) associated with a failure risk v. The objectives of the regulator are to assure firm ? Period t = 2,…,T the incumbent faces a regime R > x for itself participation and to incite cost minimization by the participating firms. Observing the but observes to form the regime {R, v} the firm chooses topotential participate in infeasibility the game or to exit.of Exiting reser› In each period, the regime is challenged at least one the gives costnorm, the choice vation utility. If participating, the firm maximizes its expected horizon utility EU(x) in › vnot = P(Regime revoked) ? , x] 5 . The the is trivial. that the regime bution(3)operover an input vector as x 2 [x ¯ validRecalling for the horizonthat lowerprobability input bound x? de› efficient If not Rtpolicy = R(y) a fully cost policy. Any such x > x?aisfailed defined asregime cost inefficient falls isrevoked: v each year andthatthat is replaced by a ile [9],fines [18], behaviour. The› payouts for a participating firm in the game If revoked: cost-plus regime Rt = xt-1 are illustrated in Figure 1. cost-recovery, Rt =v and xt the1 regulation for a reform In each period,pure the regime is repealed withi.e. probability resorts toin year t. The cost-plus, i.e., firm the firmmaximizes receives the reimbursement wx leading a utility and u(x, wx). With as an objective of toprofit slack R EVISITED 4 To simplify notation, all vector products are assumed to be done in the appropriate dimensions, i.e. ation suppressing is 5that the sign for the transposed vector. The upper bound x¯ for the support may be seen as existing observations or as the maximum value ? us demand y could award without u(x|R) =regulation. ⇡ + ⇢s = (R x) + ⇢(x x ) (1) that the regulator resetting the 6

Game timeline

U(x,R)

1-v

U(x,R)

1-v

U(x,R)

Incumbent regime R = R’

1-v

1-v v

v

v

obability 1 v, the regulator stands the appeal and enforces the payment R with the rm level utility u(x, R). The vfailure of the1 regime is irreversible for1 the horizon, i.e., 1 Low-powered regime e regulator must continue paying wx for the rest of the horizon. R = wx U(x,wx) U(x,wx)vector x for the firm In this game, the expected horizon utility forU(x,wx) a stationary input probability 1 v, the regulator stands the appeal and enforces the payment R with the given as t = 1 of the regime t=2 t=3 firm level utility u(x, R). The failure is irreversible for the horizon, i.e., • • the regulator must continue paying wx for the rest of the horizon. Figure 1: Dynamic regulation t 1model with failure probability t v. U(x) In =this u(x, wx)vd {(1utility v)t for +a 1} + Â u(x, R)dvector (1 v)xt for the firm game, the+expected horizon stationary input Â u(x, wx)vd t=2 t=1 is given as 2 vd vd (1 v) d (1 v) = u(x, wx) +• + u(x, R) • 1 d + 1 u(x, d (1wx)vd v) t {(1 v)t 11 + d1}(1+ v) u(x, R)d t (1 v)t EU(x) = u(x, wx)vd Â Â vd vd 2 (1 v) d (1 v) t=2 t=1 = r(wx c(y, w)) +2 + (R wx + r(wx c(y, w))) . 1 d (1 v) vd1 d vd1(1 d (1 v) v) d (1 v) = u(x, wx) + + u(x, R) (3) 1 d 1 1 1 d (1 v) 1 d (1 v) vd vd 2 (1 v) d (1 v) The first-order condition respect to x is obtained = r(wx c(y,of w))(3) with + + (R as: wx + r(wx c(y, w))) . 0.8 1 d 1 d (1 v) 1 d (1 v) vs ( d( 0.05) , rho) 2 (3) dEU(x) vd Firm’s v) d (1 v) policy 0.6vd (1 optimal multi-period = rw + w(r 1) . (4) vs ( d( 0.1 , rho) )+ Costinefficient is given dx 1 asd 1 d (1 v) 1 d (1 v) behaviour vs ( d( 0.5) , rho) The first-order condition of (3) 0.4 with respect to x is obtained as: vs ( d( 0.01) , rho) • • • t regime (v = 0), t texpected 1 In the case of a perfectly regulation the utility = Â u(x, EU(x) credible wx)vd + u(x, wx)vd (1 v) + u(x, R)d t (1 v)t 2 (1 v) Â Â dEU(x) vd vd d (1 v) 0.2 ollapses to the expression t=1 t=2 + w(r t=1. = rw + 1) (4) 2 dx 1 d Costefficient 1 d (1 v) 1 d (1 vd (1 v) dv) (1 v) behaviour vd = 0u(x, + d + u(x, R) 0 wx) 0 wx + r(wx 0.4 d 10.2 d 1c(y, (1 0.6v) . 0.8 1 1d (1 v) EU(x) = (R w)) (5) v=0 In the case of a perfectly credible regulation regime (v = 0), the expected utility 0 rho 1 2 d 1 vd vd (1 v) d (1 v) collapses to the expression = r(wx c(y, w)) +v = 0 + (R wx + r(wx c(y, w))) . Optimal response to credible regime: dfor d 1= {0.99, dthat (10.952, 1 d (1 v) all a regulation regime such that Rfailure c(y, w) feasible, av)cost efficient Figure 2: Critical probability v(d ˆ 1, r)meaning 0.909, 0.667}. firm d ideal regulation, (3) ould participate in the game under = a credible model 0).w)) For such EU(x) (R wx + r(wx(v =c(y, . (5) v=0 1 dconsider a flawed e firm response would be full cost efficiency. On the other hand, The first-order condition of (3) with respect to x is obtained as: gulation with certain demise (v = 1). In this case, the expected utility becomes response regime: v = 1 that a cost efficient firm Call a regulation Optimal regime such that Rto non-credible c(y, w) meaning feasible, 2 (1 v) dEU(x) vd vd d (1 v) could participate in the game under a credible = rw model + w(rideal1)regulation, . (4) 27 + d(v = 0). For such dx 1 w)) dOn 1the .dother (1 v) 1 a (6) d (1 v) EU(x) r(wx c(y, the firm response would be v=1 full = cost efficiency. hand, consider flawed 1 d regulation with certainIndemise (v = 1). In this case, the expected the case of a perfectly credible regulationutility regimebecomes (v = 0), the expected utility hus, the firm will follow an optimal policy leading to cost padding, selecting the upper collapses to the expression ound x, ¯ leading to the suboptimal cost wx¯ > c(y, w). Wedstate our findings in the EU(x)v=1 = r(wx c(y, w)) . (6) d roposition below. 1 d EU(x) = (R wx + r(wx c(y, w)) . (5) v=0

1

d

Thus, the firm will follow an optimal policy leading to cost padding, selecting the upper

Model predictions Proposition 1: › The optimal cost policy of a firm in a multi-period policy depends on the probability of regulatory failure (credibility), the time preferences of the firm (impatience) and the utility of inefficient cost (cost of effort).

above. above. Note that the level of the allowed revenue R does not affect the optimal cost policy, thatparticipation the level of the allowed revenue R does not affect the optimal cost policy, butNote only the constraint. above. but only the participation constraint. Corollary 1. Assume a given cost of effort r > 0 and discounting factor d . Then, there exists a finite failure v(d ˆ allowed , r) above which a dominated Note that levelrisk of the does not discounting affectisthe optimal policy,there Corollary 1. the Assume a given cost ofrevenue effort rRcost-efficiency > 0 and factorcost d policy. . Then, but only the participation constraint. exists a finite failure risk v(d ˆ , r) above which cost-efficiency is a dominated policy.the Proof. Follows from the first order condition (4) as a function h(v, d , r). Define

Corollary 1. Assume given effort r> and discounting ., Then, there {v critical Follows failure rate v(d ˆ athe , r)first =cost :ofh(v, d , r) =00}. failurefactor rate v(d ˆ ,Define r) cost-the Proof. from order condition (4) For as aa function h(v,vdd r). exists a finite failure risk v(d ˆ , r) above which cost-efficiency is a dominated policy. minimization is optimal and for v > v(d ˆ , r) the firm has a monotonously increasing critical failure rate v(d ˆ , r) = {v : h(v, d , r) = 0}. For a failure rate v v(d ˆ , r) costutility in x. minimization is optimal and for v > v(d ˆ , r) the firm has a monotonously increasing Proof. Follows from the first order condition (4) as a function h(v, d , r). Define the {v critical failure rate v(d ˆ , r) = : h(v, d , r) = 0}. For a failure rate v v(d ˆ , r) costutility in x. The function v(d ˆ , r) is illustrated in Figure 2. As a consequence of Corollary 1, minimization and for v > will v(d ˆ ,select r) thethe firm haslevel a monotonously increasing the The firm function facingis aoptimal credible regulation input x? , giving aof cost efficient 1, v(d ˆ , r) is illustrated in Figure 2. As a consequence Corollary utility in x. ? level wxfacing = c(y,aw). For a non-credible regime higher failure risk v >a v(d ˆcost, r), the the firm credible regulation will selectwith theainput level x? , giving efficient firm will adopt an input maximization behaviour, that is selecting the upper bound ¯the v(d ˆ For , r) aisnon-credible illustrated in regime Figure with 2. Asa ahigher consequence of Corollary 1, x. levelThe wx?function = c(y, w). failure risk v > v(d ˆ , r), ? Thefirm associated wx¯ >regulation c(y, w) implies cost the inefficiency firm.a As anefficient example, the facing credible will select input x the , giving firm will adoptacost an input maximization behaviour, thatlevel isby selecting thecost upper bound x. ¯ ?a imagine firm withFor a alow cost of effort (r with = 0.2) and a failure discount factor d ,= 0.99%. level wx = c(y, w). non-credible regime a higher risk v > v(d ˆ r), the The associated cost wx¯ > c(y, w) implies cost inefficiency by the firm. As an example, Thenwill in Figure 2 the firm would be cost minimizing any regime failure x. firm adopt an input maximization behaviour, that for is selecting the with uppera bound ¯rate imagine a firm with a low cost of effort (r = 0.2) and a discount factor d = 0.99%. above. less associated than v(d ˆ =cost 0.99, 0.2) 0.173.cost On inefficiency the other hand, a firm a high cost of The wx¯r>=c(y, w)=implies by the firm.with As an example, Then in Figure 2 thea firm would be cost(rminimizing for any regime with a 0.99%. failure rate imagine athat firm with low of effort = and a discount factor d were = policy, effort (e.g. rthe= 0.8) of would only abstain inefficiency if the policycost almost Note level thecost allowed revenuefrom R 0.2) does not affect the optimal less than v(d ˆ = 0.99, r = 0.2) = 0.173. On the other hand, a firm with a high cost of Then in Figure 2 the firmvconstraint. would be0.99, cost minimizing for any regime with a failure rate impossible overturn, < v(d ˆ = r = 0.8) = 0.004. but only the to participation effort (e.g. 0.8)r would abstain inefficiency if the almost less than v(d ˆ r==0.99, = 0.2) only = 0.173. On from the other hand, a firm withpolicy a highwere cost of Corollary 2. Assume a non-credible regime v > 0 and a given cost of effort r. Then, Corollary 1. Assume a given cost of effort r > 0 and discounting factor d . Then, there impossible v < v(d ˆ =abstain 0.99, rfrom = 0.8) = 0.004.if the policy were almost effort (e.g. to r overturn, = 0.8) would only inefficiency for any cost-efficient firmv(d there exists an upper bound dˆ foristhe discount factor. exists a finite failure risk ˆ , r) above which cost-efficiency a dominated policy. impossible to v(d ˆ = 0.99, r = 0.8)v = Corollary 2. overturn, Assume av < non-credible regime >0.004. 0 and a given cost of effort r. Then, Proof. The bound obtained from thean fixed point v( ˆa dˆfunction , dr) = v. Proof. fromis the first order condition asand h(v,discount dof, r). Define the for any Follows cost-efficient firm there exists upper for the factor. Corollary 2. Assume a non-credible regime v (4) > 0bound aˆgiven cost effort r. Then, critical rate v(d ˆ firm = {vexists : h(v,an dregime , upper r) = v0}. failure rate v factor. v(d ˆfactor , r) dcostfor any failure cost-efficient there bound dˆafor the discount Corollary Assume a, r) non-credible > 0For and given discount . Then, Proof. The3.bound is obtained from the fixed point v( ˆhas dˆa, r) = v. minimization is optimal and for v > v(d ˆ , r) the firm a monotonously increasing ˆ for any cost-efficient firm there exists an upper bound r for the cost of effort. Proof. bound is obtained from the fixed point v( ˆ dˆ , r) = v. utility inThe x. 3. Corollary Assume a non-credible regime v > 0 and a given discount factor d . Then, Proof. Follows directly from the inverse function of the critical failure rate v(d ˆd . Then, , r), the Corollary 3. Assume a non-credible regime v > 0bound and a given ˆconsequence for any firm there exists upper for discount the costoffactor ofCorollary effort. Theiscost-efficient function v(d ˆ as, r) is illustrated inan Figure 2.unique As a rand ˆ ˆ bound obtained v(d ˆ , r) = v. Note that r is bounded for all v > 0. 1, for any cost-efficient firm there exists an upper bound rˆ for the?cost of effort. the firm facing a credible regulation will select the input level x , giving a cost efficient Proof. Follows directly from the inverse function of critical failure rate v(d ˆ is, r), Corollary aafrom non-credible regime v > of 0. Thethe cost efficiency forv(d firm thenthe ? = 4. Proof. Follows directly the inverse function the critical failure ˆaall r), the level wxis c(y,Assume w). For non-credible regime with a higher failure risk vrate > ˆv(d ,,r), the ˆ ˆ bound obtained as v(d ˆ , r) = v. Note that r is unique and bounded for v > 0. inversely proportional to the discount factor d and the cost of effort r.

Corollaries

ˆ = v. Note that bound is obtained v(d ˆ ,maximization r) rˆ is unique boundedtheforupper all v > 0. x. firm will adopt an as input behaviour, that isand selecting bound ¯

Corollary 4. Assume non-credible regime v >the 0. Theby cost fororder a firm is then The associated cost wx¯aa> c(y, w)Proposition implies cost1v inefficiency theefficiency firm. As example, Proof. Follows directly from and for thefor first condiCorollary 4. Assume non-credible regime > 0. Thefunction cost efficiency aanfirm is then inversely proportional tonecessary the discount factor d0.2) and the of effort imagine firm low cost of effort (r = and a cost discount factor = 0.99%. tions (4),aproportional h(v, dwith , r). a A condition athe cost-efficient optimal by the inversely to the discount factor d for and cost of effort r. r.d policy Then in Figure 2 the firm would be cost minimizing for any regime with a failure rate firm is that h(v, d , r) 0, inducing cost-minimization with the optimal solution x =condix? . Proof. Follows directly directlyfrom fromProposition Proposition 1 and the function for the first order Proof. Follows 1 and the function for the first order condiless than ˆh(v, =d0.99, rAthere =necessary 0.2)is=an0.173. On thefor other hand, a firm with highpolicy cost2ofby Corollary 3 gives that upper bound rˆ for h given v and d . aCorollary protions (4),v(d ,r). r). condition a cost-efficient optimal tions (4), h(v, d , A necessary condition for a cost-efficient optimal policy by the the ˆ effort (e.g. r = 0.8) would only abstain from inefficiency ifConsider the policy were almost vides analogously an upper bound d for h given v and r. the curvature of h ? firm that h(v, h(v,dd,,r) r)0,0,inducing inducingcost-minimization cost-minimization with optimal solution firm is is that with the the optimal solution x = xx? .= x . impossible to overturn, v < v(d ˆ = 0.99, r = 0.8) = 0.004. with respect to r:that Corollary gives thatthere thereisisananupper upper bound rˆ for h given v and d . Corollary 2 proCorollary 33 gives bound rˆ for h given v and d . Corollary 2 proˆ ˆ Corollary 2. Assume a non-credible regime v > 0 and a given cost of effort r. Then, vides analogously an upper bound d for h given v and r. Consider the curvature vides analogously an upper bound d for h given v r. Consider the curvature of h of h 2 (1 v) + d (1 v d 2firm EU(x) vd vd bound ˆ for any cost-efficient there exists an upper d for the discount factor. with respect =w + >0 (7) with respect to tor: r: dxdr 1 d 1 dˆ (1 v) the fixed point v( Proof. The bound is obtained from ˆ d , r) = v.

0.8

dv( 0.2 , rho) 0.6 dv( 0.4 , rho) dv( 0.6 , rho) 0.4

0.2

There will always some laggards … 0

0

0.2

0.4

0.6

0.8

1

rho

ˆ for v = 0.2 and r = {0.1, 0.2, ..., 0.9} Figure 3: The indifference curve dˆ (r)

As the credibility of the regulation increases, v decreases and the set (area) in Figure 3 increases. Remark 1. Given n independent firms each having a cost of effort drawn from a distribution with density function f (r) and cumulative density function F(r) on the support [0, 1], then the probability that all firms are cost efficient under a non-credible regime is ˆ n. equal to 1 (F(r)) U(x,R)

U(x,R)

1-v

1-v

U(x,R)

1-v

The intuition behind Remark 1 is clear: the hope of incentivizing all firms to efficiency in a weak regulation regime is thin. In practice, there will always be inefficient 1-v firms in the set of regulated operators, see Figure 4. Thus, the empirical conjecture v would then v be a higher vincidence of non-profitmaximizing behavior from firms that have a plausible case of a failing regulatory regime. In particular, firms with stable semi-public ownership can represent the case of longv 1 1 of effort. 1This is frequently range time preferences and high cost the case for energy distribution in Europe. The opposite extreme would be a set of privately owned franU(x,wx) U(x,wx) U(x,wx) chisees in countries with high inflation or political risks, here the time preferences is short-run and the relative cost we test the validity of t = 1of effort low. t = 2 In the nextt section, =3 our model on an interesting case of regulatory failure in Scandinavia, a region otherFigure 1: Dynamic regulation model with failuresolutions, probability v. wise characterized by early adoption of market-oriented cf. Amundsen and Bergman (2007). 4. Empirical model The previously presented model leads to a number of empirically verifiable hypotheCritical failure probability ses for non-credible regimes; in particular that firms exhibit lower cost efficiency fol1

1

11

0.8 vs ( d( 0.05) , rho) vs ( d( 0.1 , rho) )

0.6 Costinefficient behaviour

vs ( d( 0.5) , rho) vs ( d( 0.01) , rho)

0.4

0.2 Costefficient behaviour

0

0

0

0.2

0.4

0

0.6 rho

0.8

1 1

Figure 2: Critical failure probability v(d ˆ , r) for d = {0.99, 0.952, 0.909, 0.667}.

27

6. Analysis Initially, we observe that the facts in the case correspond relatively well to those

Parttheoretical IV of the model. First, given the mixed ownership situation among the oper-

ators (50 per cent privately owned20 , 40 per cent publicly owned21 and 10 per cent consumer-cooperatives, (Agrell et al., 2005b)), we can safely assume that there is some heterogeneity with respect to the cost of effort r in the sample. At least some firms should have a strictly positive r, meaning that they are not exclusively profit maximizing. Second, given the lukewarm reception of the new regulatory model and the industrial and academic critique raised against it, we safely assume that all firms observed a high probability of regime failure v >> 0. Without adventuring into what value each firm (or group of firms) attributed to v, it suffices to recall our results from Remark 1 6. Analysis where adverse effects appear already from modest levels of v. Initially, we observe that the facts in the case correspond relatively well to those For clarity, we restate our hypotheses and proceed into the analysis step by step. of the theoretical model. First, the drop mixedinownership situation among oper-is virlaunch of the NPAM, there is agiven radical technological change, thethe value 20 , 40 per cent publicly owned21 and 10 per cent ators (50 per cent privately owned tually at its floor of unity (0.1 per cent). The difference in mean technical Hypothesis 1. Firms exhibit a lower cost efficiency CE during a non-crediblechange regimeis consumer-cooperatives, (Agrell et al., 2005b)), we can safely assume that there is some and of the expected sign. Looking closer at the data in Figure 6 shows a vsignificant > 0. heterogeneity with at respect to the In costthe ofanticipation effort r in the sample. least some model, firms alstriking difference firm level. of the fall ofAtthe NPAM t for the should have a strictly positive r, meaning that they are not exclusively profit maximizThe average cost efficiency CE operators by year is listed in Table 2 below. most no operator shows technological progress (the black curve). As predicted by the ing.note Second, given the lukewarm receptionefficiency of the newduring regulatory model and theonindusWe a clear tendency of decreasing the period, from average model, the absence of incentives is leading to a stand-still of the investments in new trial per and cent academic critique raisedto against per it, we safely assume that all firms observed a 74.5 before the NPAM cent during the NPAM. In fact, the overall technology and processes, resulting71.9 in the observed stagnation of technological change. highinprobability of regime v >> year 0. Without valueis each fall cost efficiency fromfailure the initial (2000)adventuring to the last into yearwhat (2006) 5.4 per Hypothesis 2 isofnot rejected by the data. firm (or firms) v, it suffices to per recall our is results from Remark 1 cent. Thegroup difference in attributed mean costtoefficiency (2.6 cent) statistically significant. where adverse effects appear already from modest levels Hypothesis 3. NPAM, The for productivity of theinin firms isoflow or nilchange, for red the curve duration of Additional support this finding is found Figure 5,v.where the shows launch of the there is achange radical drop technological the value is avirt For clarity, we restate our hypotheses and proceed into the analysis step by step. non-credible regime > 0.before the mean CE floor by operator the NPAM and the black curve depicts the analogous tually at its ofvunity (0.1 per cent). The difference in mean technical change is cost efficiency after the NPAM introduction. As seen in the Figure, the fall in cost a significant and of theexhibit expected sign. closer at the data in Figure 6(13), shows Hypothesis 1. Firms a calculated lower costLooking efficiency CE during a non-credible regime The productivity change, as the right-hand side of expression is 22 efficiency is generalized, except forInsome initially highly inefficient operators . Hence, striking difference at firm level. the anticipation of the fall of the NPAM model, v > 0. presented in Table 4. As a consequence of the previous results in terms of cost efficiency alwe conclude that Hypothesis 1 is not rejected by the data. most no operator shows technological progress (the black curve). As by and technological change, the productivity change prior toisthe reform waspredicted positive andthe t The average cost efficiency CE for the operators by year listed in Table 2 below. model, on the absence ofper incentives is leading to aintroduction, ofthe the investments strong, 3.5 per year. After the productivity change Hypothesis 2. The technical change ofefficiency the firms isstand-still stagnating for the duration ofinanew We note a average clear tendency ofcent decreasing during the period, from on average technology and processes, in the observed stagnation of is on per average negative perto cent). thetechnological means of thechange. two non-credible vthe >(-0.7 0. resulting 74.5 centregime before NPAM 71.9 perThe centdifference during thebetween NPAM. In fact, the overall Hypothesis 2 is not rejected by the data. periods is significant and of the right sign. Presented in more detail in Figure 7, fall in cost efficiency from the initial year (2000) to the last year (2006) is 5.4 per the The technical change DTC and the variation in cost efficiency DCE from the methodfindings aredifference without any ambiguity. Here note that of theof a cent. The in mean cost change efficiency (2.6 per cent) is confidence statistically significant. Hypothesis 3. The productivity ofwe the firms isthe low or nil for intervals the 23 .duration ological framework, expression (17), are calculated and reported in Table 3 The Additional support for this finding is found in Figure 5, where the red curve shows three years during thev > reform are below the horizon average productivity changepre(1.4 non-credible regime 0. reform technological change ratetheisNPAM strong, on the average 4.8 per cent per year. At the t by operator the cent). mean CE before and black curve depicts the analogous per Thus, we find that Hypothesis 3 cannot be rejected. productivity theseen right-hand side oftheexpression costThe efficiency after thechange, NPAM calculated introduction.as As in the Figure, fall in cost(13), is Hypothesis 4. The profitability of the firms is lower on average, and decreasing through22 . cost presented in Table 4. As a consequence of the previous results in terms efficiency efficiency is generalized, except forassome inefficient operatorsof Hence, 20 Including international firms such EDF, initially owner of highly Graninge, and the inter-Nordic operator Birka. out the duration of a non-credible regime v > 0. we21conclude that Hypothesis 1 is not rejected the data. and technological change, the productivity change prior to the reform was positive and Mostly municipal utilities with the exception of by the (then) state-owned operator Vattenfall. 22 An inefficient policy under a credible regime may be optimalin atthe very high of effort strong, average 3.5 per cent year. After theis introduction, productivity change The on average profitability Pt per by regulatory firm and year presented Table 2 cost below. The Hypothesis 2.theThe technical change firms isofstagnating for duration of to a this r, in which case policy is independent ofofthethe credibility the regime. Our the results are robust is on average negative (-0.7 per cent). The difference between the means of the findings document significant and monotonous fall in profitability over the period,two non-credible regime av > 0. assumption. periods is significant andcent of the right sign. Presented in more detail in Figure 7,Atthe 23 The from on results average per to the NPAM to 10.5the perperiod centwithout after the for 14.7 DSC from (17) prior are equal to unity throughout anyreform. significant The technical and the variation in cost methodfindings are without anyDTC ambiguity. Here we note that theDCE confidence intervals differences. operator-level, thechange results are illustrated in Figure 8.efficiency Hypothesis 4 from is notthe rejected by of thethe 23 . The preological framework, expression (17), are calculated and reported in Table 3 three years during the reform are below the horizon average productivity change (1.4 empirical data. reform technological change rate is strong, on average 4.8 per cent per year. At the per The cent). Thus, that Hypothesis 3 cannotpresented be rejected. 18 result canwe byfind explained by the previously methodology. Profitability variation PV 4.is The driven by productivity change and price recoveryand PR,decreasing as shown in exHypothesis profitability of the firms is lower on average, through-

VERIFIABLE HYPOTHESES

Research hypotheses

Part V

EMPIRICAL METHOD

Objective We are interested in a framework that links › › › ›

Profitability changes Cost changes Revenue changes Efficiency changes

Productivity development Simple approach: efficiency changes vs index ? Not conclusive, since price changes may be due to › Input price changes (price recovery) › Output price changes (profit margin) › Economies of scale (volume) › Allocative efficiency (mix) and, as a consequence, thatefficiency the productivity › Technical changes of the sector will suffer. In order to investi-

gate this phenomenon, we need both a test-case with a failed regime and a methodology that is capable of differentiating between profitability, cost efficiency and dynamic efNeed decomposed analysis fects for multi-input, multi-output production. For the empirical case, we collected data for the period before and up to the failing of a transient regulation regime in Sweden, described in more detail in Section 5. The methodology for the study of this empirical case is based on profitability6 , the firm financial indicator defined by the ratio revenue to cost P = py/wx. GeorgescuRoegen (1951) introduced profitability, called return to the dollar, as a financial performance indicator into the economic literature. It is independent of the scale of production, a virtue not shared by cost, revenue or profit measures. This property of independence of the scale of production is particularly relevant in sectors with a wide range in the size of operation. Moreover, it allows for the direct comparison between the remuneration from the regulator (R = py) and the observed cost of the firm (wx). We are Profitability change interested in the study of the evolution throughout time of the ratio revenue to cost, i.e. profitability change. It is defined as Pt+1 = Pt

pt+1 yt+1 /wt+1 xt+1 pt yt /wt xt pt+1 yt+1 /pt yt Revenue change = , wt+1 xt+1 /wt xt Cost change

(10)

which is equal to the ratio of revenue change to cost change. The next step is to identify the factors that cause changes in profitability. These factors are associated with changes in quantities and prices of individual outputs and inputs. Hence, we want to isolate the changes in prices of the changes in quantities, either of which influences profitability change. We can decompose cost change in (10) as wt+1 xt+1 wt+1 xt+1 wt xt+1 = wt xt wt xt+1 wt xt wt+1 xt wt+1 xt+1

(11)

s equal to the product of a Laspeyres input price index and a Paasche input quantit ndex. Both these pairings satisfy the product test. Laspeyres-Paasche index numbe airs are widely used, but because they use different weights, they generate differen esults. The third line solves this problem because it takes the geometric mean of th Laspeyres-Paasche index number pairs and creates a Fisher input price index and Fisher input quantity index7 . The last line of (11) defines a more compact notation as Revenue change Fisher input price index WF and a corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for rev nue change as pt+1 yt+1 pt yt

Fisher output price index

= PF (pt+1 , pt , yt+1 , yt )YF (yt+1 , yt , pt+1 , pt ),

(12

Fisher output quantity index

where PF is called a Fisher output price index and yF defines a Fisher output quantit ndex. Combining (11) and (12) yields an expression for the relative change in prof tability, pt+1 yt+1 /wt+1 xt+1 PF (pt+1 , pt , yt+1 , yt ) YF (yt+1 , yt , pt+1 , pt ) = . pt yt /wt xt WF (wt+1 , wt , xt+1 , xt ) XF (xt+1 , xt , wt+1 , wt )

(13

This profitability change is the product of a price recovery term PR for output an nput prices, respectively, and the Fisher productivity change index YF /XF for outpu nd input quantities, respectively. The price recovery PR compares the variation in th rices on the inputs with the corresponding input price changes between two periods The interpretation of PR is intuitive: a value lower than unity implies that the firm ha which is equal to the ratio of revenue change to cost change. The next step is to ncreasedidentify the output prices theininput price changes. Under high-powered regu the factors that less causethan changes profitability. These factors are associated with change ation, where theinoutput prices are fixed, this would expected changes quantities and prices of Cost individual outputsbe andthe inputs. Hence,outcome. we want to Inversely isolate the changesaninincrease prices of the changes in quantities, either of which influences PR > 1 would indicate of the output prices compared to the input prices, i.e profitability change.margin. We can decompose cost change in (10) as n increase of the gross We are interested in the decomposition of the Fisher productivity index by its eco t+1 t+1 t+1 t 1/2 t+1 t+1 t t+1 1/2 wt+1 xt+1 objectwof xattention w x by Diewert w x w x omic drivers. It has been (2014) (see also(11) Grifell-Tatj = t xt+1 t+1 xt wt xt wt xt (2003,w2015b); wtKuosmanen xt nd Lovell (2015a)), Grifell-Tatj´ e wand Lovell and Sipil¨aine t+1 t t+1 t t+1 t t+1 t = WF (wWe, wfollow , x , xGrifell-Tatj´ )XF (x , x , we and , w )Lovell (2015b) that ca 2009); Ray and Mukherjee (1996). e relatedExpression to the approach of geometric Ray Mukherjee (1996). conventional approach t Fisherand input priceofindex Fisher input An quantity index (11) takes the mean the Laspeyres-Paasche index number pairs eal withand thecreates generic production possibilities introduced in Section 3.1 is a Fisher input price index and a set Fisher input quantity index.6 The lastdefined b line of (11)programming defines a more compact as aetFisher input price indexon WFthe andData a he mathematical modelsnotation in F¨are al. (1985), based Envel corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for revenue change as

7 The

Fisher index has a t+1 set of t+1appealing axiomatic properties, see Balk (2012) p

y pt yt

= PF (pt+1 , pt , yt+1 , yt )YF (yt+1 , yt , pt+1 , pt ),

(12)

and creates a Fisher input price index and a Fisher input quantity index.6 The last line of (11) defines a more compact notation as a Fisher input price index WF and a corresponding Fisher input quantity index XF . Symmetric derivations for the revenue side obtain an analogous expression for revenue change as

echnology, which corresponds to the expected behav decomposition to (16) can be done based on the Paas Fisher productivity index in (13). Here we obtain t p y = P (p , p , y , y )Y (y , y , p , p ), (12) py efficiency change, thechange technical change and the si Relative in profitability where P is called a Fisher output price index and Y defines a Fisher output quantity size change, has(11)the same as above, w index. Combining and (12) yields aninterpretation expression for the relative change in profitability, Profitability change(1978). Its empirical Fisher productivity by Charnes et by al. unded below unity. Thus, taking the geometric m P P (p , p , y , y ) Y (y , y , p , p ) = . (13) P W (w index , w , x , xand ) X (x the , x , w Paasche ,w ) aspeyres productivity product recovery This profitability change is thePrice product of a Fisher price recovery index P /W for ) on of the productivity index. We output Fisher and input prices, respectively, and the Fisher productivity indexobtain Y /X for output t+1 t+1

t+1

F

t t

t

t+1

t

F

F

t+1

t

t+1

t

F

t+1 t

F

F

t+1

t+1

t

t+1

t

t+1

t

F

t

F

t+1

t

t+1

t

t+1

t

t+1

t

F

F

F

and input quantities, respectively. The price recovery index compares the variation in s Ray s the s with the andinputs Mukherjee (1996), Kousmanen input and Sipiläinen (2009), between two periods. the prices on corresponding price changes i i i Diewert (2014), Grifell-Tatje Lovell (2003, 2015) The interpretation of price recoveryand index is intuitive: a value lower than unity implies st that the firm has increased the output prices less than the input price changes. Under t high-powered t+1 regulation, t t bet the expected where the output pricestaretfixed,tthis would outcome. Inversely, a price recovery index higher one than indicates an increase of the prices compared to the input prices. t outputt+1 t t+1 t+1 t+1 t+1 t+1 We are interested in the decomposition of the Fisher productivity index by its economic drivers. It has been object of attention by Diewert (2014) (see also Grifell-Tatj´e

ÂÂl i

F

y ,l

0 ,

(14)

+1 , y

,p ,p ) w x /c (y , w ) = s, respectively. ector for firm i in period +1 , x , w ,w ) w x /c (y , w ) ant returns to scale, which is a common t t t b and Pollitt (2003). Additionally, c (y , w ) thect (yt+1 , wt+1 ) The Fisher index a set of appealing axiomatic properties, inear surface, orhas frontier, over the t , wsee tdata t+1 (yt+1 , wt+1 ) ct+1 (y )Balkc(2012) from year 1 to year t inclusive. 2 13 Wet obt+ t+1 6

p py Fisherallow productivity index al way which does not for technit yt p p 4 t t t t t t t t re c (y , w ) = minx {w x : (xw,t xyt+1 )2 T }(yt ,wt ) wt xt / /ct+1 tt+1 tt+1 tt+1 t t+1 t+1 fined as CE t (yt , wt ) = cCostt (y , w )/w x w x /c (ychange ,w ) wt+1 xt /c Size efficiency vity index YF /XF = in (13) based· DSC on DCEis· DTC Technical Fisheroutput productivity quantities because the areefficiency conpeyresthe input quantity index component term, cost efficiency change DCE indicates whet Efficiency measured using non-parametric approach (DEA) 2 outputs (energy LV, HV,) 4 inputs (assetconnections, grid capital, cost OM, energy losses, energy transit)

between periods t and t + 1. The second term, the

wt )

wt xt+1 /ct+1 (yt , wt )

, (15)

Part VI

FULL DATASET

The Network Performance Assessment Model Electricity Act (2000) › Regulated revenue based on “objective performance”

NPAM › Green-field planning model, based on GIS-positioned load points, feedin points, standard costs › Critique from industry and academics, model suffers from several methodological flaws (Lantz, 2003; Wennerström and Bertling, 2008; Jamasb and Pollitt, 2008, Jamasb and Söderberg, 2008)

Green-field vs brown-field planning

NPAM rise and fall 2003 Start of implementation 2005 Rulings I for 2003 = 21 DSO for 76,3 MEUR › All DSO appeal

2006 Reduced claims for 2003: 8 DSO for 23 MEUR › DSO appeal to higher court

2007 New regulator › Out-of-court settlement: 8 DSO for 16.5 MEUR.

2009 NPAM suspended (cost-recovery) 2012 New regime: rate-of-return

Data DSO audited data from the regulator (EI) for Swedish electricity distributors (LV and MV only, no generation, retail or transmission) Balanced panel, 128 firms for 2000-2006, in all 896 DMU

Data: DSO 2000-2006 Table 1: Descriptive statistics and model variables. Category

Unit

Definition

mean

median

sd

Revenue R = py

kSEK kSEK kSEK

Total revenue Revenue LV Revenue HV

137,764 118,394 19,371

49,967 41,876 6,707

387,118 335,470 53,213

Costs wx

kSEK kSEK kSEK kSEK kSEK

Total cost (TOTEX) Cost transmission Cost energy losses Operating expenditure (OPEX) Capital expenditure (CAPEX)

119,515 33,791 7,878 46,766 31,082

46,483 13,285 2,864 18,615 8,602

346,036 100,420 21,395 130,483 102,922

MWh MWh

Energy delivered low voltage (LV) Energy delivered high voltage (HV)

488,052 221,633

204,662 71,037

1,235,396 623,509

SEK/kWh SEK/kWh

Price per energy delivered LV Price per energy delivered HV

0.228 0.109

0.226 0.104

0.043 0.057

MWh MWh km kSEK

Energy transported, total Energy losses, total Connection-weighted network LV+HV Network capital, total

742,112 32,427 41,415 458,831

281,796 11,952 14,198 100,737

1,913,920 86,027 121,128 1,521,204

SEK/kWh SEK/kWh SEK/m %

Transmission price Cost per energy losses OPEX per connection-line unit Cost of capital

0.049 0.260 1.379 0.086

0.048 0.252 1.332 0.083

0.019 0.120 0.543 0.033

Outputs y

Output prices p

Inputs x

Input prices w

Part VII

EMPIRICAL TEST

H1: Slumping cost efficiency H4: Profitability sacrifice

Table 2: Profitability Pt and cost efficiency CE t , mean per year, 2000-2006. year Pt

CE t

period

2000

2001

2002

2003

2004

2005

2006

2000-02

2003-06

Diff

1.150 0.762

1.149 0.732

1.141 0.741

1.128 0.732

1.128 0.723

1.086 0.713

1.079 0.708

1.147 0.745

1.105 0.719

-0.042*** -0.026***

Notes: ***p < 0.001; **p < 0.05; *p < 0.01.

Table 3: Cost efficiency DCE and technology change DTC, before and after NPAM. n period

All

Pre NPAM

Post NPAM

768 2000-2006 Mean SD

384 2000-2002 Mean SD

384 2003-2006 Mean SD

384 Diff

p-value

Table 2: Profitability Pt and cost efficiency CE t , mean per year, 2000-2006. year Pt

CE t

period

2000

2001

2002

2003

2004

2005

2006

2000-02

2003-06

Diff

1.150 0.762

1.149 0.732

1.141 0.741

1.128 0.732

1.128

1.086

1.079

1.147

1.105 0.719

-0.042*** -0.026***

0.723 0.713 efficiency 0.708 0.745 H1: Cost

Notes: ***p < 0.001; **p < 0.05; *p < 0.01. 1.0

C E 2000 2002 C E 2003 2006

●●●● ● ● ●●●

0.6

0.7

CE

0.8

0.9

●● ●● ● ●● ● ●● ●● ●● ● ● ● ● ● ●●● ●●● ●● ● ●● ●●● ●●●●● ● ●●●●● ●● ●●● ● ● ●● ●● ●●●● ●● ●●● ●●● ●●● ●●● ●●●● ●●●●● ●●● ● ● ●● ● ●●● ● 0.745 ● ●●●● ●●●●● ●●● ●●●●●● ● ●●● ●●●● ●●●●●● ●●● ●●● ●● ●●●● ● 0.719 ●●●●●● ●●●●● ●●● ●●●● ●●●●●●● ●●● ●● ●●●●● ●● ●●●● ●●●●●● ● ●● ●●●●● ● ●●● ●●●● ● ●● ●● ●●●● ●● ● ●●●●● ●●● ● ● ●

0.5

● ● ●●● ●● ● ●● ● ● ●●● ● ●● ●● ●

0

20

40

60

80

100

120

D S O (sorted) U C L/C O R E/Agrell uab.stemdata.rev2 2015 07 23

Figure 5: Cost efficiency CE t , average per DSO, development before and after NPAM.

ological framework, expression (17),cost are efficiency calculatedCE andt ,reported in year, Table2000-2006. 3.20 The preTable 2: Profitability Pt and mean per reform technological change rate is strong, on average 4.8 per cent per year. At the year period launch of the NPAM, there is a radical drop in technological change, the value is vir2000 2005 2006 in mean 2000-02 Diff tually at2001 its floor2002 of unity2003 (0.1 per 2004 cent). The difference technical2003-06 change is significant and of the expected sign. Looking closer at the data in Figure 6 shows a Pt 1.150 1.149 1.141 1.128 1.128 1.086 1.079 1.147 1.105 -0.042*** striking difference at firm level. In the anticipation of the fall of the NPAM model, alt CE 0.762 0.732 0.741 0.732 0.723 0.713 0.708 0.745 0.719 -0.026*** most no operator shows technological progress (the black curve). As predicted by the Notes: model, ***p