MERGERS IN NORWEGIAN ELECTRICITY DISTRIBUTION: A COST SAVING EXERCISE?

Transcription

1 MERGERS IN NORWEGIAN ELECTRICITY DISTRIBUTION: A COST SAVING EXERCISE? A UGUST 2006 P ETER B OGETOFT INSTITUTE OF FOOD AND RESOURCE ECONOMICS PRODUCTION AND TECHNOLOGY DIVISION THE ROYAL AGRICULTURAL UNIVERSITY, DENMARK PB@KVL.DK T HOR E RIK G RAMMELTVEDT DEPARTMENT OF ECONOMICS, BI - NORWEGIAN SCHOOL OF MANAGEMENT NORWAY THOR.E.GRAMMELTVEDT@BI.NO

2 2 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION ABSTRACT We investigate whether the mergers of Norwegian electricity distribution companies in the period can be rationalized as cost saving exercises or whether they should be interpreted in other ways. We use Data Envelopment Analysis (DEA) to estimate the potential cost savings from mergers ex ante, and we decompose the gains into gains from learning best practice and from improving economies of scale and scope. Moreover, we compare with post merger performances to identify extraordinary costs and benefits of the mergers. In addition to the cost saving aspect, we also discuss the possibility to gain from playing the regulation. We compare the regulated revenues and the regulated efficiency improvement requirements before and after the mergers. We find limited potential cost gains ex ante but some improvement in allocative efficiency ex post. Merged companies tend to rely more on outsourcing and less on internal labor. Moreover, merged companies show more rapid technological progress and catch up. We also find that the new companies face somewhat more favorable regulatory conditions, i.e. higher allowed revenues and fewer required efficiency improvements. The latter is the consequence of the - possibly flawed - specification of the regulatory cost function using a so-called vrs DEA model. DEA frontiers are biased and more so the fewer comparators. The regulator s estimate of efficient costs for the newly formed and relatively large units with few comparators may therefore be relatively large. Lastly, we find that mergers may generate gains in related activities, most notably electricity retail. KEYWORDS Data Envelopment Analysis (DEA), Mergers, Productivity, Electricity Distribution, Norway.

3 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 3 1. INTRODUCTION Mergers and takeovers seem to play an important role in the restructuring of many sectors, including the Norwegian electricity distribution sector. The general economic literature has identified two reasons to merge, namely to affect costs and competition, cf. e.g. Perry and Porter (1985) and Farrell and Shapiro (1990). The cost side includes the possibility to exploit economies of scale, economies of scope, risk sharing, scarce managerial skills, etc. The competition aspects include the possibility to gain market power via scale, scope, or collusive behavior. In the electricity distribution industry in Norway, the firms have local monopolies and their revenues are regulated by The Norwegian Water Resources and Energy Directorate (NVE). The competition aspects are therefore related to the possibility of playing the regulation or to strategic considerations in the related competitive wholesale and retail markets for electricity. There are of course many difficulties in mergers as well. They include the integration of different business cultures and the coping with competition policies directed against the exercise of market power. In this paper we focus on the cost side, i.e. the internal or organizational effects of mergers, but we will also consider some competition motives, i.e. the external or market effects, related to the regulation. We model the multiple inputs multiple outputs production process using an activity analysis or Data Envelopment Analysis (DEA) approach. This approach is easy to use and it has proved to be a flexible and powerful tool in a large number of empirical studies. We use the DEA framework to measure and decompose the potential and realized cost savings from mergers. In this respect, we build on work by Bogetoft, Jellesmark and Strange (2003) and Bogetoft and Wang (2005) on how to measure the potential merger effects empirically using existing production data from a cross section of firms. The idea is to estimate ex-ante what a merged unit can save and to decompose the saving into contributions from learning best practice and from improving economies of scale and scope. A related line of literature is concerned with estimating the potential gains from resource and task reallocations. An early reallocation study using DEA is Bännlund, Chung, Färe and Grosskopf (1998). They estimate the potential gains from allowing pollution permits to be traded among the firms in an industry. In Andersen and Bogetoft (2003, 2004) we extended these ideas by studying the effects of reallocations off the frontier, i.e. among inefficient firms. The difference between reallocation and merger studies lies primarily in the restrictions we impose on the reallocations. In a merger study, reallocations are allowed among partners but not across all market participants. In the present paper, we restrict attention to the mergers that have actually been realized, but along the lines of the reallocation studies - we extend the previous merger papers by distinguishing the resources according to their controllability and transferability. This allows us to make more refined models of the likely short and long run effects of mergers.

4 4 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION Methodologically, we deviate by also estimating both the potential gains a priori and the realized gains (or losses) ex post. Comparison of ex ante potential and ex post realized gains captures extraordinary gains and losses from mergers. Lastly, we deviate from the above studies by investigating some of the market aspects of mergers as well. We find that mergers may in part be rationalized as a cost saving exercise but that the potential savings are difficult to identify ex ante using aggregate data. One explanation may be that most organizations operate with slack and that a merger is an event that enables the organizations to reduce old slack. That is, the merger process may generate saving potentials rather that simply harvesting them. The changes in consumption of multiple inputs in actual mergers give insights into the integration process following a merger. We show, for example, that most savings are on labor and service costs and that the factor mix changes after a merger. Also, by including details of the incumbent regulation model, we can show that the regulation may benefit newly merged companies. This may be a consequence of larger bias for large companies in the cost norm derived form the regulator s model of efficient costs. Finally, by combining other available information, we investigate which of a series of other characteristics seem to distinguish the attractive from non-attractive mergers. We show, for example, that strategic positioning in the retail electricity market is a likely consideration in merger decisions. The outline of the paper is as follows. In Section 2, we introduce a class of production models and we review how they can be estimated using DEA. Also, we propose aggregate measures of the potential gains from a merger and we discuss how these measures can be decomposed into technical efficiency, size, and harmony effects. The specification of the DEA model, the data used, and the results of a traditional analysis of the efficiency within the distribution industry are discussed in Section 3. The mergers in the period are analyzed in Section 4. Conclusions and some suggestions for further research are given in Section DEA AND MERGER GAINS In this Section, we discuss the DEA approach to production modeling and we show how potential gains from mergers can be estimated and decomposed using such models. DEA MODELS Since it was first proposed by Charnes, Cooper and Rhodes (1978,79), Data Envelopment Analyses (DEA) has become a tremendously popular relative performance evaluation tool. A

5 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 5 recent bibliographic survey ( identified more than 1000 papers from all sectors of society. Many of these have been published in high quality international journals within economics as well as management science and operations research. Currently, DEA is also used as the basis for regulation regimes in different areas. In particular, it has been used in incentive regulation of private, semi-private and public utilities. In the regulation of electricity distribution, for example, countries like Austria ( Norway ( Holland ( and Finland ( ) have introduced DEA based revenue and price cap systems, and DEA has together with more traditional statistical methods been used to determine reasonable cost norms in countries like Australia, England, New Zealand, and Sweden. The great popularity of DEA among researchers and theorist alike also means that the basics of DEA modeling is described in several text books, cf. e.g. Charnes, Cooper, Lewin and Seiford (1994), Coelli, Rao and Battese (1998) or Cooper Seiford and Tone (2000), and are known to many readers of this journal. We shall therefore not give any in-depth introduction to DEA. It suffices to introduce a bit of notation and to recall a few basic definitions. Consider the case where each of n Decision Making Units (DMUs), i I={1, 2,, n}, transforms p inputs x = ( x1,..., xp ) R 0 i i i p i i i q to q outputs y = ( y1,..., yq ) R. Also, let T be the 0 production possibility set p T = {(x,y) R + q 0 x can produce y} The DEA approaches estimate T from observed productions (x j,y j ), j I, and they evaluate the efficiency of observed productions relative to the estimated technology. The estimate of T, the empirical reference technology T*, is constructed according to the minimal extrapolation principle: T* is the smallest subset of R + q 0 p that contains the actual production plans (x i,y i ), i I, and satisfies certain technological assumptions specific to the given approach. Given a technology T (or an estimate hereof T*) inefficiency (or relative inefficiency) of a given DMU, DMU i, reflects the possibility to reduce inputs and expand outputs. It is most commonly measured using the so-called Farrell (1957) measures E i = Min{E R 0 (Ex i, y i ) T} or F i = Max {F R 0 (x i, Fy i ) T} where E i is the maximal contraction of all inputs and F i is the maximal expansion of all outputs that are feasible in T. Different DEA approaches are distinguished by the technological regularities they impose on T and hereby on T*. Classical assumptions are A1 disposability, i.e. (x,y ) T and x x and y y (x,y ) T, A2 convexity, i.e.: T convex and A3 s-return to scale, i.e. (x', y') T k(x',

6 6 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION y') T for k K(s), where s = "crs", "drs", "vrs", or "irs" corresponding to constant, decreasing, varying, or increasing return to scale, and where K(crs) = R 0, K(drs) = [0,1], K(vrs) = {1}, and K(irs) = [1, + ), respectively. A less common, but very relevant assumption here, is the replicability assumptiona4 additivity, i.e. (x,y ) T and (x,y ) T (x +x,y +y ) T. Different DEA models invoke different assumptions about the technology. The original constant returns to scale (crs) DEA model proposed by Charnes, Cooper and Rhodes (1978,79) assumes A1, A2 and A3(crs) while the decreasing returns to scale (drs) and (local) variable returns to scale (vrs) models developed by Banker (1984) and Banker, Charnes and Cooper (1984) appeal to A1, A2 and A3(drs) and A1, A2 and A3(vrs), respectively. It is easy to see, cf. e.g. the references above, that A1, A2 and A3(s) lead to the empirical reference technology p T*(s) ={(x, y) R + q 0 λ R n 0 :x i λ i x i, y i λ i y i, λ Λ(s)} where Λ(crs) = R 0 n, Λ(drs) = {λ R 0 n i λ i 1} and Λ(vrs) = {λ R 0 n i λ i = 1}. The assumptions A1-A3 have been relaxed in the free disposability hull (fdh) model used by Deprins, Simar and Tulkens (1984), and the free replicability hull (frh) model briefly proposed in Tulkens (1993). The fdh model invokes only A1 and T*(fdh) and therefore has the structure above with Λ(fdh) = {λ R 0 n i λ i = 1, λ i {0, 1} i}. The frh model presumes A1 and the additivity assumption A4 such that T*(frh) has the structure above with Λ(frh)={ λ R 0 n λ i integer i}. DEA models which partially relax the convexity assumptions are suggested in Bogetoft (1996) and Petersen (1990). We note that DEA by the minimal extrapolation principle and assuming no noise in data, provides an inner approximation of the underlying production possibility set. The efficiency estimates are therefore optimistic and the potential input savings and output expansions are underestimated. The extent of this bias can be estimated using bootstrapping, cf. e.g. Simar and Wilson (2000). MEASURES OF POTENTIAL MERGER GAINS The idea of making a priori estimates of potential production economic (cost) gains from mergers is developed in Bogetoft, Jellesmark and Strange (2003) and in particular in Bogetoft and Wang (2005). We briefly recall the basic ideas here before turning to the development of refined measures below. Consider a possible merger of the units in a subset J of the available units I= {1, 2,..., n}. The merged unit is denoted DMU J. Direct pooling of the inputs and outputs gives a unit which has used j J x j to produce j J y j. This corresponds to having a completely decentralized (or divisionalized) organization where the decentralized units correspond to the J-units.

7 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 7 A radial input based measure of the potential overall gains from merging the J-DMUs is therefore E J = Min{E R 0 ( E[ j J x j ], j J y j ) T} E J is the maximal proportional reduction in the aggregated inputs j J x j that allows the production of the aggregated output profile j J y j. If E J <1, we can save by merging. If E J >1, the merger is costly. Similarly, one could of course measure the potential gains on the output side. We focus on the input measure here since distribution demand is not easy to control at least in the short run. The mathematical program above can be solved using several models of the technology T. In particular, we can use the above DEA estimate T*. In this case, the above program becomes a simple linear programming (LP) problem. The same goes for the other programs below as detailed in the appendix. The appendix is hereby useful as an operational guide for readers interested in using the methodology in other industries. Our measure of the potential overall merger gains, E J, encompasses several effects. We now decompose it into technical efficiency, scale, and mix effects, and we discuss the organizational relevance of this decomposition. Some or all of the units in J may be technically inefficient and this may be captured in E J. A merger may bring in new management which may facilitate the elimination of such inefficiencies. However, it is also possible to reduce technical inefficiencies through other means, e.g. by imitating the better performers, sometimes referred to as the peer units. To avoid compounding the effects, it is useful to adjust the overall merger gains for the technical efficiency effect. To do so, we can project the original units to the production possibility frontier and use the projected plans as the basis for evaluating the remaining gains from the merger. Thus, for example, we may project (x j,y j ) into (E j x j,y j ) for all j J, where E j =E {j} is the standard efficiency score for the single DMU j, and use the projected plans (E j x j,y j ), j J as the basis for calculating the adjusted overall gains from the merger Letting T J = E J /E *J we get E *J = Min{E R 0 ( E[ j J E j x j ], j J y j ) T} E J = T J * E *J where T J [0,1] indicates what can be saved by individual technical efficiency adjustments in the different units in J. Assuming that individual technical inefficiencies have been dealt with, we are left with the two most interesting production economic effects of a merger.

8 8 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION One is the scaling or size effect. A merger leads to a unit that operates at a larger scale. This may or may not be advantageous, depending on the scale properties of the underlying technology. The left panel in Figure 1 below illustrates this. We see how the original productions A=(x 1,y 1 ) and B=(x 2,y 2 ) are efficient and yet there are improvement potentials in the merged unit A+B using x 1 +x 2 to produce y 1 +y 2. It suffices to use E*(x 1 +x 2 ) to produce (y 1 +y 2 ) as in C. Output Input 2 y 1 +y 2 y 2 y 1 A B C A+B B x 2 A A+B x 1 +x 2 C E (x 1 +x 2 ) x 1 L(y) L(2y) O x 1 x 2 E (x 1 +x 2 ) x 1 +x 2 Input O Input 1 Figure 1 Size (S) and Harmony (H) Effects The other main effect of a merger is that it leads to other input and output mixes. This may be advantageous by taking us into more "productive" directions of the product space. We shall refer to this as the harmony, scope or mixture effects. The right panel in Figure 1 illustrates a case with a positive harmony effect. Here, two units, A and B, produce the same output, y, using two different input vectors, x 1 and x 2. The combined unit A+B is able to produce the original total output using only a fraction, E, of original inputs as in point C. Without further assumptions about the technology, we cannot say whether size and harmony effects favor a merger. This issue is discussed in details in Bogetoft and Wang (2005). Here it suffices to note that with a convex technology the harmony effect generally favors a merger, while the size effect may or may not favor a merger. To formalize the harmony gains we examine how much H J of the average input could have been saved in the production of the average output H J = Min{H R 0 ( H[ J -1 j J E j x j ], J -1 j J y j ) T} where J is the number of elements in J. We look at the average input and average output, since we do not want the expansion of size to come into play yet. Using the average is most relevant if the units in J are not too different in size to begin with. If the sizes differ considerably, we may be picking up scale effects, e.g. if some units are larger than and some are smaller

9 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 9 than the "most productive scale size" as defined by Banker (1984). Note that H J <1 indicates a savings potential due to improved harmony, while H J >1 indicates a cost of harmonizing the inputs and outputs. Next, we capture the size gains by asking how much could have been saved by operating at full scale rather than average scale, i.e. by the measure S J S J = Min{S R 0 ( S[H J j J E j x j ], j J y j ) T} The re-scaling is advantageous, S J <1, if we have economies of scale, and costly, S J >1, if the return to scale property does not favor larger units. Using these definitions, we have E *J = H J * S J and by E J = T J * E *J we get our basic decomposition E J = T J * H J * S J This corresponds to a decomposition of the basic merger index E J into a technical efficiency index T J, a harmony index H J, and a size index S J. The technical efficiency measure T J captures what can be gained by making the individual units efficient. The remaining potentials to save, E *J, are created by the harmony effect, H J, and the size effect, S J. The decomposition of the potential gains from merging DMUs into a technical efficiency measure, a harmony measure, and a size measure is important because full scale mergers are typically not the only available organizational option, and it may be that alternative organizational changes may be easier to implement. In particular, we suggest that the following may guide the organizational restructuring: Low technical efficiency measure T J : One could let the inefficient DMUs learn from the practices and procedures of the more efficient ones. If the problem is not lack of skills but rather motivation, one could improve the incentives, e.g. by using relative performance evaluation and yardstick competition based on the technical efficiency measures, cf. Bogetoft (1994,95,97,00). Of course, if the problem is scarcity of managerial talent, it may still be necessary to make a genuine merger to transfer control to the more efficient administrative teams and hereby improve the managerial efficiency (X-efficiency). Low harmony measure H J : One could consider reallocating the inputs and outputs among the DMUs to create more "powerful" input mixes and more easily produced output mixes. This can be done a) inside a hierarchy, b) by long term contracts or perhaps c) by creating a market for key inputs and outputs, cf. also Andersen and Bogetoft (2003,04) and Brännlund, Chung, Färe and Grosskopf (1998).

10 10 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION Low size measure S J : In this case, full scale mergers may be the only alternative. If we need large amounts of fixed capital, highly specialized staff, long run-lengths or simply a critical mass to obtain sufficient returns from scale, it may be relevant to merge. Also, and perhaps most importantly, this may be relevant if the reallocation through contracts or a market is associated with too many transaction costs to make it attractive, cf. the general discussion of the size of the firm in the industrial organization literature, e.g. Tirole (1988). The decomposition developed above gives one natural way, we believe, to define and distinguish between the technical efficiency, the size and the harmony effects. In Bogetoft and Wang (2005) additional motivations for these measures are provided. Most importantly, it is shown that with a convex technology, the harmony effect measures the most that can be gained by any kind of reallocation between the units in J. Assuming that we were to pick new inputs and outputs (x* j,y* j ) for each j J such that total inputs and outputs stay feasible, j J x* j j J x j and j J y* j j J y j, and such that all of the new production is possible (x* j, y* j ) T*, the largest proposal saving in j J x j is precisely the harmony effect. RESTRICTED CONTROLLABILITY AND TRANSFERABILITY In the estimates of potential merger gains above, we have assumed that all inputs and outputs can be redistributed in the merged entity J. In many cases, this approach is too restrictive. At least in a short run perspective, some dimensions are easier to change and reallocate than others. It may for example be easier to reduce the labor input than the capital input which is largely based on sunk investments. Also, some services may have to be provided on location and can therefore not be transferred to another unit, located elsewhere. It may for example be possible to transfer accounting and customer services to another location, but the physical net installations of the individual households can not be transferred. Lastly, some variable in actual models typically describe the context rather than choice variables, and they are therefore not transferable. Population density and climate characteristics are for example difficult to transfer (although minor adjustments to average values are possible by redefining the concession areas). We will now show how to evaluate efficiency and calculate potential gains when only some of the inputs and outputs can be adjusted and transferred among the members of the new and merged entity. First we consider the relatively straightforward case with restricted possibilities to control inputs and outputs, and then we extend by introducing restrictions in the transferability of some of the resources and services. Consider now the case of restricted controllability. Let the inputs and outputs be split in two types, x=(x V,x F ) and y=(y V,y F ) corresponding to the variable (controllable) and fixed (noncontrollable) dimensions. The relevant measures in this case are straightforward to derive. As it is customary in the DEA literature, we propose to use sub-vector efficiency measures, i.e. to account for the non-discretionary character of some dimensions by only looking for improvements in the other directions, cf. e.g. Charnes, Cooper, Lewin and Seiford (1994). It follows that the appropriate measure of the efficiency of DMU i is

11 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 11 E i V = Min{E R 0 (Ex i V,x F, y i ) T} We can now calculate the potential gross gains from a merger of the J units by E J V = Min{E R 0 ( E[ j J x j V], [ j J x j F], j J y j ) T} i.e. E J V is the maximal proportional reduction in the variable (discretionary) inputs that together with the fixed (non-discretionary) resources allows the production of the aggregated output profile j J y j. If E J V <1, we can save the fraction (1- E J V) of the variable inputs by merging. If E J V >1, the merger is costly and requires that the total usage of the variable resources is increased. Now, if we stick to a DEA estimate of the underlying technology, we find (as explained in the appendix for the usual cases above) that the potential merger gains with restricted controllability, E J V, are determined by solving the following operational LP program Min E E, λ s.t. E[ j J x j V] i I λ i x i V [ j J x j F] i I λ i x i F [ j J y j ] i I λ i y i λ Λ(k) As previously, we may also filter out the effects of individual inefficiencies by determining the adjusted overall gains in the direction of the variable inputs as the harmony effect as and the size effect as E *J V = Min{E R 0 ( E[ j J E j V x j V], j J x j F, j J y j ) T}, H J V = Min{H R 0 ( H[ J -1 j J E j x j V], J -1 j J x j F, J -1 j J y j ) T}, S J V = Min{S R 0 ( S[H J j J E j x j V], j J x j F, j J y j ) T} The interpretations and organizational implications of these scores are as previously explained except that all evaluations are now in savings of only the controllable inputs and they are calculated conditional on the given levels of the non-controllable inputs. Thus, for example, rescaling is advantageous, S J V <1, if we have economies of scale in (x V,y) for given x F, and costly, S J V >1, if the return to scale property does not favor larger units for the given values of the fixed inputs. We also note that the operational calculation of these measures with a DEA approximation of the underlying technology follows the same principles as demonstrated for the overall gains measure E J V above, and as explained in all the usual cases in the appendix. Using the above definitions, we once again get a decomposition

12 12 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION E J v = T J v * H J v * S J v This corresponds to a decomposition of the basic merger index E J V into a technical efficiency index T J V, a harmony index H J V, and a size index S J V. So far we have dealt with the possibility that only some of the variables are actually discretionary within a given time horizon. Another obstacle to the reallocation among DMUs may, however, be non-transferable (local, L) resources and services as opposed to transferable (global, G) ones. Allowing for this, we get a 2 x 2 taxonomy of all variables: They may be LF (local and fixed such as net installations), LV (local and variable such as perhaps maintenance personnel), GF (global and fixed such as perhaps specialized trucks or measurement hardware), and GV (global and variable such as for example accounting and billing personnel). To simplify notation, we can in the usual way indicate vectors of such variables by suppressing the sub-scripts that we include. We will for example refer to the L variables as the combination of the LF and LV variables, and to the F factors as the combination LF and GF factors In such a setting, it is more demanding to measure the possible gains from a reallocation of resources and services among the otherwise autonomous units in J. A harmony measure H could in this case be calculated in the following way. We consider what can be saved after individual learning - by mere reallocations of the global inputs and outputs among the units in J. Let the new input and output combinations after reallocation be (x* j, y* j ) for j J. If we measure the savings potential in the variable inputs and stick to proportional improvements, the savings from pure reallocations can be determined by solving the following pure reallocation problem: Min H H, (x* j, y* j ), j J s.t H[ j J E j Vx j V] j J x* j V, x* j L x j L j J x* j GF j J x j GF y* j L y j L j J y* j G j J y j G (x* j, y* j ) T all j J H R 0 The choice variables in this program are the contraction factor H and the new input and output combinations (x* j, y* j ) for j J. Since the original adjusted productions (E j Vx j V,x F,y j L,y j G) j J satisfy all the constraints, we always have H 1 corresponding to a potential saving. The constraints in this program ensure i) that we measure proportional reductions of all vari-

13 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 13 able inputs, ii) that the local consumption of local inputs is not increased, and iii) that the global consumption of the transferable fixed inputs does not increase. The output constraints similarly ensure iv) that the local services do not fall and v) that the total production of globally transferable services does not suffer. Finally, we require vi) that the new production plans are feasible on an individual level. It is worthwhile to note that when all resources and services are being global and variable, the above program leads to the original harmony measure. This is a simple application of Proposition 2 in Bogetoft and Wang (2005). 3. A DEA MODEL OF NORWEGIAN ELECTRICITY DISTRIBUTION BACKGROUND In Norway, electricity distribution is restricted to the task of transporting electricity from the regional transmission level to the end consumer. It is considered a monopoly service and as such subject to a revenue cap regulation by the Norwegian Water and Energy Directorate (NVE). The current regulation was introduced in 1997 replacing the Rate of Return Regulation (ROR) that had been in place since The revenue cap regulation was continued in a second regulation period starting in 2002 with only minor changes, see Grammeltvedt (2003). We have access to panel data of Norwegian electricity distribution companies covering the years 1995 to There were 198 companies performing distribution services in Norway in Of these, 86 distribution companies where involved in a series of mergers reducing the total number of distribution companies to 130 in In this Section we develop a DEA model of electricity distribution in Norway to evaluate individual firms. The same model defines the technology in the merger analysis in the next Section. MODEL SPECIFICATION To guide the model specification one can build on numerous studies from academics and practitioners alike. International models of electricity distribution use many different specifications. In a survey study, Jamsab and Pollitt (2001) conclude that there seems to be no obvious choice of variables. They do, however, find that the most frequently used inputs are operations costs, number of employees, transformer capacity and network length. The most widely used outputs are energy delivered, number of costumers and the size of the service area.

14 14 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION The variation in international model specifications may, in part, reflect different intended uses of the models, different access to data and different time horizons. Edvardsen and Førsund (2003) point out that although electricity distributors in the short run take the exiting lines, transformer capacity and customers as given, this is not the same as to say that these factors must be regarded as constants in the analysis. Both past and present decisions are reflected in the configuration of the grid and even the areas served by the companies can be rearranged through mergers or divisions. A conceptually appealing approach is to think of the outputs as capturing three dimensions: (i) customer service, (ii) transportation work and (iii) capacity provision, cf. Agrell and Bogetoft (2004). The first dimension is usually covered by the total number of clients, potentially divided into voltage levels or market segments. The second corresponds to total delivered energy, if need be differentiated by voltage level. The third dimension is covered by proxies for capacity such as installed transformer power or peak power. In terms of input, most studies focus on the operating costs, either in terms of physical (labor) or monetary units, while using proxies for capital in terms of line length and/or installed transformers. Note that the asset base can hereby enter as input (capital proxy), as environmental factor (proxy for operational complexity), and as output (proxy for capacity provision service) depending on the focus of the study. Our choice of inputs and outputs follows that used by NVE. It is except for a few refinements - based on Kittelsen (1993) and involves the inputs Cost of labor in knok (LABEX) Cost of capital (measured as annual depreciation plus the regulated rate of return times the book value) in knok (CAPEX) Cost of energy losses in knok (LOSEX) Cost of goods and services in knok (SERVEX) and the outputs Number of customers (CUSTOMERS) Energy delivered in MWh (ENEREGY) Km of low voltage lines (LINES LV) Km of high voltage lines (LINES HV) The use of book value as a proxy for capital input is problematic since the networks have different ages and the depreciations used in the accounts may not accurately reflect the services that a distribution company derives from its assets. Indeed, recent analyses of the Norwegian distribution industry show that a replacement value measure leads to much better model speci-

15 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 15 fications, cf. Agrell and Bogetoft (2004). Unfortunately, such measures were not available for the whole sample period. Both the price of energy and the regulated rate of return varied considerably during the study period. This does not affect the calculation of the ex ante potentials of the mergers, but it does affect the calculation of the Malmquist productivity index. To control for this we apply the average prices for energy and the average return over the period adjusted for inflation. We also used a model with OPEX (LABEX plus SERVEX), LOSEX and CAPEX as inputs. The advantage of this is to control the effect a change in input mix might have on the results of the Malmquist productivity analysis. We discuss this in more detail in Section 5. In NVE (1997), sea cable was introduced as a proxy for difficulty of the service area and a separate analysis was conducted using replacement value instead of book value, cf. also above. These variables are not included in this analysis due to lack of data. The results of this analysis may therefore over-estimate the efficiency potential in a given year to some extent. On the other hand, this bias is probably not too important. Our aim is not to evaluate the relative efficiency of the companies as such. Rather, we want to study the effect of mergers by analyzing the (potential) changes in inputs before and after the merger, as well as the development in efficiency after the merger. These changes may be less affected by a possible bias in the general level. In addition to the choice of inputs and outputs, a series of a priori assumptions must be specified. The NVE models have traditionally been based on the assumption of variable return to scale (vrs), and we shall therefore primarily use this assumption. We note, however, that this may put the large units in an overly favorable light. First of all, there may well be some positive economies of scale. Some previous studies have shown quite large differences in relative efficiency and significant increase in returns to scale in retail and distribution, cf. Kumbhakar and Hjalmarsson (1998). Secondly, the sample size bias may be more important for large units with only few comparators. Salvanes and Tjotta (1994) offer an interesting refinement of the returns to scale discussion. They study the Norwegian electricity distribution and distinguish between returns to scale and returns to network density. Returns to network density measures the economies of increasing the amount of electricity supplied when the network is held constant. Returns to scale measures the combined economies of increasing the amount of electricity supplied and also expanding the network. They find no evidence for economies of scale except for very small output levels, but they find significant evidence for economies of density. A merger between two distribution companies leads to an increase in both network and energy delivered. Initially, the merged unit will transport the same amount of energy to the same customers over the same grid, and there are as such no benefits of increasing returns to network density.

16 16 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION DATA The data in this analysis are taken from the annual economic and technical reports which the distribution companies are obliged to submit to NVE. These data are also used in the calculation of the revenue caps and individual efficiency requirements, and the data are therefore audited in several steps. We note, however, that companies may have incentives to manipulate these figurers in order to receive a higher revenue cap or a softer efficiency target. Summary statistics for the 4 inputs and 4 outputs are provided in Table 1, below. The large difference between the minimum and maximum illustrates a large difference in firm size. The relatively low average (compared to the maximum) and the even smaller median indicate that the sample contains a majority of small companies. Observe also that the standard deviation is somewhat larger for energy delivered and number of customers. This suggests that the economics of density might have an impact. Over time we see that labor cost in the sector is reduced while the cost of services increases. Likewise, we see that total capital cost is reduced. This is probably because of falling replacement investments. In terms of the structural development, we see that both the averages and the variations for all variables are increasing over time while the median is decreasing. This indicates that the restructuring process has led to a sample with a few relatively large companies while the majority of the companies in the sample remain relatively small. Table 2 highlights the difference between merged and non-merged companies. On the output side, the merged companies serve more than half of the customers and deliver more than half of the energy over less than half of the lines. This shows that the mergers have mainly taken place in the more densely populated areas. On the input side, the most interesting difference is that the merged companies seem to rely more on services and less on labor. We return to this in Section 4 below. Table 1 Summary statistics for the data panel of Norwegian electricity distribution companies No. LABEX SERVEX LOSEX CAPEX COSTUMERS ENERGY LINES LV LINES HV Sum 198 2,761,675 2,780, , ,760,656 2,409,909 65,729, ,518 88,774 Min 1,326 1, , Max 176, ,484 62, , ,339 8,301,033 7,644 4,900 Average 13,948 14,043 3,850 18,993 12, , Median 7,082 7,364 1,740 9,433 5, , St.dev 20,394 20,667 6,524 31,996 25, ,490 1, Sum 130 1,642,761 3,574, ,539 3,450,091 2,635,530 67,033, ,352 94,184 Min , , Max 168, ,864 78, , ,290 11,246,937 12,297 5,347 Average 14,363 24,548 4,815 26,571 18, ,579 1, Median 6,035 7,417 1,490 9,339 5, , St.dev 24,854 50,148 9,742 53,308 42,177 1,228,282 2,276 1,005 Average 1995 to 2004 Sum 162 2,173,238 3,297, ,720 3,735,844 2,544,689 68,064, ,151 91,771 Min , Max 130, ,521 71, , ,225 10,085,099 11,615 5,598 Average 13,371 21,043 4,051 23,405 16, ,607 1, Median 6,553 7,650 1,519 9,771 5, , St.dev 19,259 40,653 7,948 43,786 36, ,927 1,

17 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 17 Table 2 Summary statistics for 2004 for companies that have merged before 2004 and for companies that have not merged before 2004 No. LABEX SERVEX LOSEX CAPEX COSTUMERS ENERGY LINES LV LINES HV Not merged companies Sum ,050 1,587, ,974 1,453,141 1,042,667 25,794,920 90,393 51,842 Min 1,587 1, , , Maks 55, ,285 22, ,601 95,308 2,253,029 6,817 5,013 Average 8,747 8,555 14,434 2,227 13,210 9, , Median 4,945 5,898 7,526 1,137 7,228 4, , Stdv 8,840 23,920 3,629 19,254 15, ,440 1, Merged companies Sum ,240 1,972, ,889 1,981,036 1,583,295 41,031, ,992 41,906 Min 7,337 9,201 1,914 14,095 8, , Maks 114, , , , ,152 14,407,454 20,061 7,941 Average 38,291 38, ,608 19, ,058 87,961 2,279,544 5,611 Median 26,277 26,277 46,245 7,542 44,003 30, ,800 2,537 Stdv 30, ,274 30, , ,739 3,267,044 5,437 2,177 EFFICIENCIES Over the years, the average relative efficiency of Norwegian electricity distribution companies range between 85.4% and 91.9% in the vrs model, cf. Table 3. The trend seems to be increasing although the changes are moderate and certainly not monotonous. Also, the share of companies with low efficiency, say below 80%, seems to be declining somewhat over time. The regulation may therefore have contributed to less variation in performance across companies. In a crs model, we get similar results, cf. Table A1 in the Appendix. Table 3 Distribution of efficiencies in vrs model Efficiency interval All Average 88.60% 85.40% 87.10% 88.60% 90.40% 89.90% 91.90% 90.50% 90.30% 90.40%

18 18 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION 4. MERGER ANALYSIS MERGERS IN REGULATED INDUSTRIES As pointed out in Section 2, there are two main reasons why companies merge: Strategic reasons related to the possibility of gaining market power and production economic reasons related to synergies and economies of scale. For regulated monopolies, such as the Norwegian electricity distribution companies, the strategic aspects of mergers are connected to the possibility of playing the regulation, either by exploiting the way the regulated income or prices are set or by expanding into other, usually unregulated, activities. The production economic reasons for merging are connected to the possibility of realizing higher, short, or long run returns by exploiting the scale and scope properties in the industry. Horizontal expansions or mergers with other types of companies can in some cases also be motivated by the possibility of realizing other synergies, for instance between different types of network transportation systems like water, natural gas, etc. There may also be other and less economic reasons why companies merge. Managers might pursue non-economic motives and financial constraints might force some owners to realize their assets. The analyses in this chapter try to quantify both the potential and the realized production economic gains form mergers. In addition, we try to evaluate some possible strategic gains from playing the regulation, and we seek insight into the changes in factor usage, etc., after a merger. This is important information from a regulatory perspective If a merger is motivated by possible production economic gains it is attractive for society. If a merger primarily serves to play the regulation it is not attractive, and competition policy measures or perhaps even changes in the regulatory system should be considered. PRODUCTION ECONOMIC GAINS Using the theory developed in Section 2, we decompose the potential effects of a merger ex ante into the Technical (T) effect, i.e. what the companies might gain from adjusting to best practice on an individual level, the Harmony (H) effect, i.e. what can be gained from a more efficient output and input mix, and the Size (S) effect, i.e. what can be gained from improved scale economies. In the illustrations, we map the improvements 1-T, 1-H and 1-S. A value of say 1-H=5% means that the merger has the potential to lower all inputs by 5% due to better harmony. We evaluate the potential gains from the 42 mergers that took place in the period We use the DEA models from Section 3 to define the technology. Specifically, we evaluate a given merger in the technology as it looked the last year before the merger. The intuition is

19 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 19 that this reflects current information at the time when the merger was planned. We realize that alternative years could be used as the basis for the analyses. Indeed, a merger is probably driven more by the expected future technology than by the current technology. In reality, data from a given year is only made available after 1-2 years delay and our model can therefore be thought of as a projected technology (short sighted but rational). The results are summarized in Figures 1 and 2 using the vrs model. We get similar results for the crs case; full results for the vrs and crs models are provided in Tabel A2 in the Appendix. In Figure 2, we see that the technical effect (T) dominates with an average potential of 11% under vrs. The potential gains from mergers measured by the harmony effect are limited with an average potential of 4%. Compared to previous studies of gains from mergers (and reallocations) in other industries, these harmony effects are relatively small, cf. e.g. Andersen and Bogetoft (2003,04), Bogetoft, Jellesmark and Strange (2003) and Bogetoft and Wang (2005). One possible explanation may be that the electricity distribution technology is quite linear, i.e. that the isoquants do not have much curvature, as also suggested in Agrell and Bogetoft 2004). The fact is that the size gains (1-S) are 0 or negative for all 42 mergers, cf. Table A2. This suggests that the mergers, from the point of view of economies of scale, have actually become too large. Alternatively, it indicates that the DEA estimates are more biased for the relatively large units with few comparators. The latter is the more likely explanation. Conceptually it is hard to justify diseconomies of scale since large distribution companies could always form say geographical divisions that could operate like individual, but smaller distribution companies. Empirically, the crs approximation seems to also fit the data well, at least for companies with over 6000 customers above, cf. e.g. Agrell and Bogetoft (2004).

20 20 M ERGERS IN N ORWEGIAN E LECTRICITY D ISTRIBUTION 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% 6,773 8,151 9,578 10,294 10,665 11,315 13,097 14,259 17,618 18,440 20,668 21,290 21,820 23,018 23,892 24,101 25,277 25,669 26,477 26,999 31,049 CUSTOMERS Technical Gains (1-T) Harmony Gains (1-H) 48,497 51,684 53,903 54,030 54,524 60,223 65,327 67,648 67,965 79, , , , , , , , , , , ,393 Figure 2 Technical (1-T) and Harmony (1-H) Gains from Mergers in vrs Model The merged companies deliver more than half of the energy to more than half of the customers using less than half of the lines, especially high voltage lines. This is clear from Table 4, where we have aggregated the inputs and outputs in the given years for the merger companies, i.e. the companies that were to become members of one of the 42 subsequent mergers (1995) and those that resulted from these mergers (2004). Table 4 Merger companies share of inputs and outputs in 1995 and 2004 LABEX SERVEX LOSEX CAPEX CUSTOMERS ENERGY LINES LV LINES HV % 50.26% 55.96% 57.96% 60.35% 61.06% 53.08% 43.82% % 55.41% 59.09% 57.69% 60.29% 61.40% 52.77% 44.70% This suggests that the mergers are happening in the more densely populated regions of the country. Indeed, direct inspection of the individual mergers confirms that they are mainly concentrated around the large cities, Oslo, Kristiansand, Stavanger, Bergen and some other more densely populated large regions. This is compatible with the idea of economies to density. We note also that with one exception all mergers happen between companies in neighboring concession areas. This is also quite natural since the reallocation of most resources and services and the exploitation of economies of scale is clearly easier the closer the merging partners.

21 P ETER B OGETOFT AND T HOR E RIK G RAMMELTVEDT 21 Lastly, it is worthwhile to note that even though the production economic gains in general, and the harmony gains in particular are limited, we do find somewhat larger possibilities to save in the labor costs and service cost dimensions alone. The results of using the refined measures from Section 2 with labor costs, LABEX, and service costs, SERVEX, as the only variable (discretionary) input, V=(LABEX, SERVEX), are given in Table 5 below. Table 5 Full and Restricted Gains in crs Model All Directions LABOR and SERVICE Directions CRS Gains Adjusted Technical Harmony Gains Adjusted Technical Harmony Gains Gains Gains Gains Gains Gains Merger CUSTOMERS 1-E 1-E* 1-T 1-H 1-E 1-E* 1-T 1-H 1 6, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , Average 82,