|
Article Excerpt
Performance-based regulation (PBR) is influenced by the Bayesian and non-Bayesian incentive mechanisms. While Bayesian incentives are impractical for direct implementation, the insights from their properties can be combined with practical non-Bayesian mechanisms for application to transmission pricing. This combination suggests an approach based on the distinction between ultra-short, short and long periods. Ultra-short periods are marked by real-time pricing of point-to-point transmission services. Pricing in short periods involves fixed fees and adjustments via price-cap formulas or profit sharing. Productivity-enhancing incentives have to be tempered by long-term commitment considerations, so that profit sharing may dominate pure price caps. Investment incentives require long-term adjustments based on rate-of-return regulation with a "used and useful" criterion.
1. INTRODUCTION
Economists like to optimize. Performance-based regulation (PBR) should be optimal. What does that mean? Until about 1970, many of us believed in truly optimal or first best regulation, which meant marginal cost prices as proposed by Hotelling (1938). However, over time, the adjective "optimal" has received more and more qualifications. The first was that losses incurred under optimal prices in the presence of economies of scale led to second best Ramsey pricing associated with Boiteux (1956), a movement that peaked around 1980. The main insight here was that prices should deviate from marginal cost prices by markups that are inversely proportional to demand elasticities (or, more precisely, to super-elasticities). The deficiency of Ramsey prices was the regulator's lack of information about cost and demand functions. Thus, the next wave was third best regulation under incomplete information. The main insights from this wave were (a) that regulated firms might need to be able to make economic profits in order to reveal private information and (b) that such profits can be limited by giving firms a choice from a menu of regulatory options. This wave probably peaked with the publication of the Laffont and Tirole (1993) book on incentive regulation. What is the next step away from optimal price regulation? Is it fourth best regulation that makes theoretical models of regulation applicable under political and practicality constraints? In any case, regulation economists have moved further and further away from what was once perceived as optimal price regulation. Consequently, in order to be relevant, the price regulation mechanisms we consider here are not strictly optimal in that they maximize a well-defined social welfare function. Rather, the PBR schemes are meant for practical application and thus should have some desirable properties.
Performance-based regulation (PBR or incentive regulation) is characterized by two main properties. First, it gives the regulated firm some behavioral discretion, for example, in the choice of prices. Second, it rewards (punishes) the firm for improving (deteriorating) performance relative to the regulator's objectives. In addition, it is often more compatible with the opening of regulated markets to competition than traditional cost-of-service (or rate-of-return) regulation.
The current literature on PBR developed from two very different strands. The first strand reflects the theoretical literature on optimal pricing for public enterprises and regulated industries that we alluded to above. During the 1970s the principal-agent framework had been developed to deal with similar problems of asymmetric information in the context of managerial incentives under separation of ownership and control. The merger of the principal-agent approach and the optimal pricing literature then lead to the Bayesian approach to incentive regulation by Baron and Myerson (1982), Sappington (1983) and Laffont and Tirole (1986). This literature has grown substantially over time. It has arguably influenced the views of what regulators can achieve and may have led to more regulatory independence, but has had much less concrete and visible impact on the way regulation has been done. One reason for this is the difficulty to translate the approach into rules that regulators can apply directly. Another reason may be that some of the potential learning from this approach is not viable in the actual regulatory environment (Crew and Kleindorfer, 2002).
The second strand developed from the need to improve the perceived theoretical and practical inadequacies of U.S. rate-of-return regulation. It had its roots in the discovery of the Averch-Johnson effect, according to which rate-of-return regulation deviates from cost minimization (Averch and Johnson, 1962). But it began in earnest only with Baumol's "plausible policies for an imperfect world" (Baumol, 1967), leading to new developments such as price caps and yardstick regulation. This literature had a substantial impact on the regulation of network industries worldwide and particularly on telecommunications. PBR is also common in U.S. telecommunications but not as much in electricity. This lack of PBR for the electricity sector may be due to lumpy and long-term investments, which are hard to handle with incentive schemes of shorter duration, or due to the prevalence of cost increases (rather than declining costs) over time, which--in a perverse way--makes incentive regulation look unattractive, because consumers and politicians would tend to associate price increases with incentive regulation. While this short characterization of the two approaches would suggest that electricity transmission pricing be best served by the practical approach of plausible rules, we will argue below that both approaches are needed to deal with transmission pricing problems.
Electricity transmission pricing is a complicated task. The lack of storability of electricity, in combination with transmission capacity constraints, suggests that pricing in the very short run is important for rationing demands efficiently in the presence of congestion. At the same time prices have to guide operating and investment decisions by transmission companies, generators and load-serving entities (distribution companies). This suggests pricing approaches geared at highly differentiated time horizons. Furthermore, transmission cost functions are affected by loop flow problems (Kirchhoff's law), power losses and ancillary services.
2. BAYESIAN VS. NON-BAYESIAN INCENTIVE SCHEMES
2.1 Bayesian Incentive Schemes
2.1.1 Characterization
Under the Bayesian incentive approach the regulator is viewed as a principal who uses the regulated firm as an agent in order to fulfill the principal's objective, which is well-defined and expressed in monetary terms. A typical regulatory objective would be to maximize W(p) = V(p) + [alpha][pi](p), where W(p) represents welfare as a function of the firm's price, V(p) is consumer surplus, [alpha] is a weight between and 1, and [pi](p) is the firm's profit. The tool is usually a monetary transfer T(p) from the regulator with which the firm is induced to act in the principal's interest. Since the transfer occurs between the regulator and the firm, it affects the regulator's objective function not only through the behavioral effect on the firm but also directly. Here the assumption is that the effect on the firm's profit is weighted by [alpha] [less than or equal to] 1, while the effect on the regulator's budget is weighted by 1.
Information is asymmetric in that the agent has important knowledge about herself or about the situation that the principal does not have. Items that matter for performance, such as the firm's effort to reduce costs, are framed in terms of asymmetric information. If the regulator could costlessly observe such effort he would hold the firm to the optimal effort level directly. This asymmetry of information is captured by the assumption that the principal has probabilistic information about the agent (knows the distribution of types that the agent belongs to), while the agent knows her actual type. The principal and agent are further assumed to be fully informed about all the other relevant properties (functional forms, expenses, outputs, etc.) necessary to fulfill the principal's objective.
Since the agent acts in her own interest by maximizing [PI](T(p), p) = [pi](p) + T(p), principal and agent play a game against each other. One specific feature of the principal-agent game is that the principal usually moves first by setting the incentive scheme for the agent. In a single-period setting the principal therefore solves the game backwards by taking into consideration the agent's response to the principal's incentive scheme. This response is captured by two constraints on the principal's optimization problem. The first is that the agent will only be willing to act on the principal's behalf if she is rewarded at least with her reservation utility (participation constraint = PC), which is usually normalized to zero. (1)
The second constraint is that the agent is going to maximize her own utility under the incentives provided by the principal. Since these incentives depend on the actual type of the agent they have to be designed in such a way that the agent is induced not to mimic someone else's type (incentive compatibility constraint = ICC).
In the case where the regulated firm's type ([theta]) and effort (e) cannot be observed and where the unobservable cost of effort to the firm is [psi](e) the maximization problem for the regulator is to find an incentive scheme T(p) with
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)
The ICC results from the observation that a more efficient type [[theta].sup.+] can mimic a less efficient type [[theta].sup.-] simply by exerting less effort and still making the same profit and getting the same transfer. As a result, mimicking type [[theta].sup.-] would give type [[theta].sup.+] an extra profit from the reduced cost of effort.
The incentive schemes designed under this approach are called Bayesian because the probability distribution of types used by the regulator is subjective with density f([theta]) and cumulative distribution function F([theta]) and is updated (in a multi-period problem) according to Bayesian rules of probability updating. The prior probabilities are simply assumed as given.
Crew and Kleindorfer (2002) severely criticize (a) that the approach spends no intellectual effort on the derivation of the probability distribution F([theta]) and (b) that F([theta]) is assumed to be common knowledge. Both these criticisms are severe. However, some strands of the Bayesian literature deal with regulatory monitoring that can be interpreted as the gathering of information about agents' types and thereby dealing with the first criticism (Baron and Besanko, 1984). (2) Nevertheless, Crew and Kleindorfer are right that no part of the Bayesian literature reflects the actual information-gathering process of regulatory procedures. The second criticism may be countered by the argument that the regulated firm does not have to have the same information about the distribution as the regulator as long as she knows her own type. (3) However, even if no common knowledge assumption is required the assumption of subjective priors means that the scheme cannot be monitored by the general public that employs the regulator. (4) The public wants the regulator to be well informed and apply the information in an unbiased manner. Crew and Kleindorfer are therefore right that the use of the principal-agent framework is suspect because the regulator lacks the sovereignty of a real principal. (5) The answer already anticipated by the Bayesian incentive literature has been to complicate the model by treating regulators simultaneously as principals and agents in a two-stage principal-agent framework, where regulators are employed by the general public and themselves employ the regulated firm (Laffont and Tirole, 1993, chapters 11 and 15).
A further criticism of the Bayesian approach is its widespread use of transfers to the firm as the main tool for influencing the regulated firm's behavior. (6) Normally, regulators do not have the authority to pay transfers to firms. Nor can they tax firms if the transfers turn out to be negative (which they often are in the optimum). (7) The transfer issue, however, can often be solved by making consumers pay fixed fees equal to these transfers. (8) As long as the fixed fees do not exclude consumers, their allocative effects are negligible (depending on income effects or the like). But if they are charged in conjunction with and in addition to other fixed fees that is not always a safe assumption. In this case and in the absence of fixed fees or transfers the necessary allocative distortions are larger but the main insights of the approach remain intact (Laffont and Tirole, 1993, chapters 2 and 3).
The more realism is included in the regulatory problem under the Bayesian approach the less feasible it becomes to derive directly applicable quantitative regulatory rules. Rather, the results of the Bayesian approach only lend themselves to general qualitative prescriptions that have to be filled out by the legislature and by concrete regulators. In my view, this is best done in combination with the simple rules derived under the non-Bayesian approach. The results from the Bayesian literature could determine the design of non-Bayesian mechanisms and the choice among several non-Bayesian mechanisms.
2.1.2 Main Insights
What are the qualitative insights of the Bayesian approach?
(1) The full information optimum is generally not achievable. The claimed exception is the absence of distributional concerns ([alpha] = 1) and of any costs of raising public funds for transfers. The example for that is the Loeb-Magat mechanism (Loeb and Magat, 1979), which, however, additionally requires full information of the regulator about the demand function. Regulation is always imperfect. So is incentive regulation. The question is therefore primarily if it is an improvement compared to traditional cost-of-service regulation.
(2) The quality of Bayesian regulation depends crucially on the ability and intentions of the actual regulator. This requires that the application of regulation schemes be based on mutually observable data, meaning that incentives have to be compatible with regulatory accounting. The better the hard information of the regulator the better the goal achievement and the lower will generally the information rents be that have to be granted to the firm.
(3) The most important...
|
