Chapter 19 Cost Analysis and Cost Modelling for Regulatory Purposes: UK Experience

Martin Cave

1.0

Introduction

The development of competition in telecom in recent years, and more particularly the decision taken within the European Union to mandate competition in all telecom markets from the beginning of 1998, have focused new attention on estimating and modelling the costs of providing telecom services and their component elements. Historically, European monopoly PTOs attached little importance to costs, and their cost accounting data were often rudimentary. But the emergence of competition has forced some operators themselves to carry out more sophisticated cost modelling in order to achieve or maintain profitability, and a further important motive for cost studies has emerged, associated with regulation. Regulators need cost information for at least four major purposes. Firstly, competing networks have to be interconnected in order to ensure any-to-any connectivity. Operators thus trade with one another in wholesale markets, to permit a call made by a subscriber to Operator A to be terminated on Network B. Both first principles and experience suggest that, especially when small entrants are competing against a dominant incumbent, ordinary commercial negotiation between the parties will not lead to satisfactory agreements on price and other conditions. Regulatory involvement will be required, and the regulator will typically try to set interconnection prices which bear some relation to costs. Some form of cost modelling or cost estimation is thus required. A second source of demand for cost data arises because governments and regulators want to maintain universal service obligations (USOs) under competition, and often do so by imposing such obligations on the dominant operator, with the net costs being shared on a proportional basis on all operators. This requires an estimate of USO costs. Thirdly, regulators in their role of a competition authority often have to determine whether prices are predatory or involve cross-subsidy, and these judgements require cost information. 1 The above three motives are associated with competition, but the regulator will often require data on the costs of the dominant operator for a fourth purpose, to control its prices before competition has become fully effective. If an attempt is made to control individual prices, then detailed cost of service information would be required. But even if a price cap mechanism is employed which is intended to allow the dominant operator to

265

recover its (efficiently incurred) costs but give it a degree of freedom in choosing individual prices, cost information will still be required to allow the regulator to project recoverable revenues over the duration of the price cap. 2.0 Costing Interconnection

Of all of these separate needs, probably the most ubiquitous and important is that related to interconnection pricing, because it is the level of such prices which determines the viability of competition. As networks interconnect more flexibly at different levels, the need for disaggregated interconnection prices increases. For these reasons, the focus of this chapter is upon approaches to estimating or modelling network costs for the purpose of establishing interconnection prices, although the final section discusses how the results can also be employed for other purposes. The task of setting interconnection charges is currently being undertaken in one form or another in most EU countries, where an army of officials and consultants is devising costing principles and beginning to implement them on operators data. The process is most advanced in the UK, where the regulator has been involved in setting interconnection charges since a path-breaking determination of the charges which BT could levy on Mercury for access to its network was made by OFTEL in 1985 (see WIK 1994). This covered a small number of services (six in all, on three charge bands) and interconnection charges were set in accordance with BTs licence condition on the basis of fully allocated costs including an appropriate return on capital. The determination was indexed in succeeding years, but in 1993 OFTEL issued a further determination, covering four conveyance services but using the same costing principles. The cost data employed in these determinations were derived from BTs management accounts, in which capital assets were valued on an historic cost basis. Effectively, OFTEL had to determine what direct costs were associated with the provision of particular interconnection services, and then augment those with a contribution to overhead costs, covering a proportion of company expenditure which could not be attributed to BTs other (i.e. non-regulated) lines of business. The decision to adopt nationally averaged interconnection charges, corresponding to nationally averaged retail tariffs, simplified the process. In 1994, however, OFTEL embarked upon a major review of interconnection pricing, in three stages (OFTEL 1994a). The first was a generalisation of the 1993 determination. In the second stage, which came into effect in 1995-96, the same costing principles (fully allocated costs) were to be employed, but in application to a much more highly disaggregated list of services. This reflected a growing requirement on the part of the operators to purchase unbundled interconnection services as they developed their own networks, which now took a variety of forms, from long distance networks to local access networks operated by access providers (CAPs) and cable companies. The last and most radical change in network costing for interconnection pricing is now underway. Crucially, it involves switching the starting point of the costing process from fully allocated costs based upon historic cost accounts to a system based upon the computation of forward-looking incremental costs, reflecting the replacement cost of capital assets. At the same time, OFTEL proposes to move away from individual pricing of interconnection services in favour of a more flexible arrangement, based on a network price cap.

266

The switch to forward looking incremental costs aims to price interconnection on the basis of the costs which an entrant into the industry might incur. Because of general inflation and changes in relative prices in the industry, the prices at which assets were originally purchased (the basis of historic cost accounts) no longer reflect their replacement costs. Broadly speaking, while the relative cost of civil works such as the construction of trenches and ducts has gone up, the price of electronic components such as switches, has declined because of technological change. If interconnection prices are based on historic costs, they run the risk of sending the wrong entry signals to the firms in the industry. The need to regulate interconnection in this new way has provoked OFTEL to undertake, in conjunction with the industry, a substantial program of cost analysis, which has adopted two alternative methodologies. The first, known as bottom-up cost modelling, involves the construction of an engineering model of parts of a telecom network. The resulting network design is then costed, and the average incremental cost of providing a particular (in this case, fairly broadly defined) service is then computed. The alternative approach, known as top-down involves utilisation of BTs management accounts. Starting from a summation of all network costs, including overheads, an attempt is made to identify from the accounts costs associated with progressively more narrowly defined services. In principle, the two methods should ultimately lead, after appropriate adjustments, to the same result. Because the two methods should produce the same outcome, it is natural to ask why both are employed. The reason is that each acts as a check on the other and that they have complementary strengths. If the regulator relies entirely on a top-down method, the resulting cost estimates may embody, and pass on to competitors in interconnection charges, inefficiencies in the dominant operators cost structure. Moreover, that operator may be able to manipulate the cost data to its advantage; the complexity of a top-down model facilitates this practice. The bottom-up model, by contrast, has the advantage of greater transparency, but because it relies upon a hypothetical rather than a real network, it may be unrealistic in its assumptions about what is attainable. There are thus advantages in applying both modelling approaches. The following two sections describe these two approaches and their implementation in more detail, and the third section discusses the work which has been undertaken to reconcile them. A fourth section reviews the case of cost data in other contexts. For completeness, it is worth noting that a third approach to cost modelling is also available, using econometric or statistical techniques. This involves estimating a statistical cost function on the basis of cost observations taken from a number of equivalent operators (normally local exchange carriers in the United States where comparable data are most readily available). Using the estimated equation, it is then possible to calculate the incremental costs of individual telecom services. A number of studies of this kind are available (see for example, Shin and Yin 1992). However, its implementation requires cost observations which are not generally available in sufficient numbers in Europe and for this and other reasons, econometric methods are not employed by regulators for this purpose.

267

3.0

The Bottom-Up Model

As noted above, cost modelling using the bottom-up engineering approach involves designing a hypothetical telecom network of a pre-specified capacity. The inputs required to deliver the services are then costed, and unit costs, are then computed. We use as an example OFTELs bottom-up network model, intended to compute the average incremental cost of interconnection services. The network model includes local exchanges, tandem switches (which are similar to local exchange switches except generally larger), and transmission. (This account is based on OFTEL 1996a and 1996c, studies prepared for OFTEL by NERA, an economic consultancy.) In modelling local exchanges, costs are separately computed for the local exchange switch, the host concentrator site and the remote concentrator site. For the modelling to proceed, a given technology has to be chosen. In OFTELs case it was System X. The bulk of switch costs comprise, (1) port costs, driven by busy hour erlangs (BHEs) a measure of the throughput of calls in the network in the busiest hour of the day; (2) processing costs, driven by the number of busy hour call attempts (BHCAs) successful and unsuccessful calls initiated in the same hour; and (3) unattributable costs driven by neither BHEs or BHCAs. Given assumed levels of busy hour erlangs and busy hour call attempts, the necessary port and processor capacity is identified. Investment per BHE is then translated into investment per busy hour minute, by dividing investment per busy hour erlang by the total number of busy hours minutes in a year (60 minutes in each of 250 working days). The investment data are then annualised. This involves two stages. The first converts the capital cost of switches into an annual value, using an annuity formula. Operating costs are then added. A similar calculation is done for the case where the cost driver is busy hour call attempts. This involves in addition an estimate of the holding time per call and the ratio of total call attempts to successful attempts. The cost of transmission is estimated on the basis of separate costings of transmission equipment, duct and fibre. Transmission network links can be provided at a number of capacities, and, in order to derive an average cost, a distribution by capacity has to be assumed. In addition, repeaters required to boost transmission may also have to be included. If charges are levied for rights of way, these would be included here. The cost of duct, inclusive of installation cost, also has to be estimated. In the OFTEL model this is taken to be 40,000 per kilometre. Assumptions on route length also have to be made. The duct contains fibre, demand for which is determined by an assumption about fibre utilisation. Adding these three components together, total transmission cost, normalised per 64 Kbit link, is calculated as the sum of duct cost and optical fibre cost per kilometre (multiplied by route length) and transmission costs. This total transmission cost is then converted into an annualised cost per busy hour minute using the method shown above. This brief account of some representative elements in the bottom-up network model indicates the range of assumptions which have to be made. For example, equipment prices depend significantly upon the volume of purchases made. Moreover, operators are reluctant to discuss such information. Duct costs too are highly sensitive to terrain, requiring the assumption of a representative terrain mix. A further difficulty arises in the allocation of costs. Thus in discussing exchange costs above, we focused on port costs, driven by BHEs, and processing costs driven by

268

BHCAs. However, local exchanges contain, in addition, substantial unattributable costs incremental to conveyance, which have to be allocated in one form or another. There are also difficulties in collecting information on operating costs, which cannot be derived from basic engineering principles. A further set of key assumptions is necessary to translate investment costs per unit (for example, per BHE) into annualised costs per busy hour call minute. As well as requiring the cost of capital, assumptions have to be made about the rate of real price change of the relevant assets. Table 1 gives an illustration of how the identification of component costs in the bottom-up model can be used to establish the average incremental cost of interconnection services.
InterTandem Transmission Interconnection Service*
(pence per minute)

Concentration

Local Switch

Tandem Switch

Remote Local Transmission

Local Tandem Transmission

Cost per minute (p)

0.17

0.20

.0.06

0.28

0.04

0.22

Number required for: Local 1 exchange segment 1 Single Tandem segment

0.65

0.75

1.03

* Found by multiplying each component required by its cost per minutes

Table 1 Component cost analysis

4.0

Top-Down Costing Methods

Top-down costing models start from the actual cost incurred on an existing network, rather than costs of a hypothetical network derived from a spreadsheet. In the UK context, the task of preparing top-down cost estimates naturally fell to BT, whose data form the basis of the calculation. Their work has, however, been scrutinised by OFTEL and its consultants. BT started the exercise from its own management cost accounts or integrated product reporting (IPR) systems. It produces fully allocated costs of services at a highly disaggregated level and is based on actual costs broken down into numerous cost categories. In order to derive an estimate of incremental costs from these, it is necessary to investigate cost volume relationships, and strip out common costs. This process is illustrated in Figure 1, which shows the relationship between current output and cost. The figure assumes that half the total output is not related to a firms regulated activity (the PSTN), while half is, and that of the latter half, four-fifths is related to conveyance

269

activities and one-fifth to providing access to customers. Of the total cost of 1,000, 200 is a common cost and 800 is volume dependent. The Figure is used to calculate the incremental cost of conveyance as follows. First, the incremental cost of non-PSTN activities (400) the distance AB is stripped out, to yield the stand-alone PSTN cost of 600. The decrement in costs associated with eliminating conveyance is then calculated. If output were further reduced by 40 percent of the total (conveyance demand) costs would come down by a further 320. The process is repeated for access demand, starting from a stand-alone PSTN. The elimination of access demand would reduce costs from 600 to 520, implying an incremental cost of 80. These successive calculations leave a common cost of 1000-400 (320+80), or of 200.

COST

1 000

600
B

200

PSTN 100 200 Volume

Figure 1 A cost function with costs common to access and conveyance (80% conveyance, 20% access) Clearly, the relationship between incremental and fully allocated costs depends upon the precise form of the cost volume relationship. Thus in the case illustrated in Figure 2 the incremental costs of successive segments of output differ. Average incremental cost thus depends on the size of the increment. In the event that the total of PSTN activity is accounted for by conveyance (i.e. there is no access-related demand) then the incremental cost of conveyance is the full stand-alone cost, which may (see Figure 3) exceed the fully allocated cost calculated from the cost function, including non PSTN activities. This is a direct consequence of the procedure adopted for defining the starting point of the calculation a stand-alone PSTN.

270

COST

1 000
B

600

200

PSTN 100

NON-PSTN

200

Volume

Figure 2 A cost function with varying incremental cost By establishing cost-volume relationships for each category of costs associated with a particular service, it is possible to eliminate common cost elements in fully allocated costs and arrive at an estimate of incremental cost. A further complication encountered in early attempts by BT to implement the topdown model arose because the asset value data employed were in historic cost accounting (HCA) terms, with no adjustment for past inflation or changes in relative prices, but in some cases fully depreciated, because the accounting lives of assets such as trenches and ducts have expired, even though they are still in service. Such data cannot produce forward looking cost estimates, for which asset valuations in current cost accounting (CCA) terms require. This difficulty has now been eliminated through use of disaggregated CCA valuations. The top-down model is thus a development of BTs traditional costing practices, in two directions. It uses fully allocated costs to generate incremental costs, and it has been developed to incorporate forward looking asset valuations, through the replacement of HCA by CCA asset values. However, it retains many features of the original approach, in particular, the very high level of detail.

271

COST

1 000

FAC

600

200

PSTN 100

NON-PSTN

200

Volume

Figure 3 A cost function with all PSTN costs attributable to conveyance (100% conveyance)

5.0

Comparing the Two Approaches

The preceding sections have outlined in broad terms the procedures underlying the topdown and bottom-up approaches. In principle, provided they are based upon identical assumptions they should produce identical results. In particular, the bottom-up model attempts to cost a network of the same basic topology as BTs existing network. Secondly, the definition of relevant increments, which our discussion of the top-down modelling has shown to be of key importance, is the same in both cases. Thirdly, the same input prices are employed. Not surprisingly, however, the results of the incremental cost of the same interconnection service derived using the two models were not identical. As a result, OFTEL commissioned a further report from NERA designed to reconcile the results. The consultants focused upon a number of key possible causes of differences. Capital equipment, arising from differences concerning utilisation, pricing or asset mix. Operating costs arising from the use of the bottom-up model of an operating cost to capital cost ratio reflecting the experience of a range of operators, rather than BTs experience alone.

272

Cost causation arising from differences in the implicit or explicit cost volume relationships in the two models. Economic and accounting depreciation associated with alternative depreciation profiles.

When an attempt was made to adjust for these differences, the resulting estimates in some cases proved to be similar. Thus, Table 2 shows the estimates from the two models for local exchanges, where the top-down estimate has been adjusted for two alternative values of the cost of capital (10.5 percent and 11.5 percent), whereas BT used 15 percent; and the generic operating cost data in the bottom-down model has been replaced by BTs actual figures. Thus the difference between the adjusted figures is only 4 percent 0.426 pence per minute from the bottom-up model per minute as compared with about 0.412 pence per minute for the top-down. This difference could easily be explained by a variety of factors and its small size enhances the credibility of a hybrid estimate obtained by averaging. Top-Down Model
Component Cost as estimated by BT Estimated Component Cost at 10.5% Estimated Component Cost at 11.5%
Bottom-Up Model
Component Cost as Estimated by Generic Model Component Cost Following Adjustment 0.369 0.426

(Pence per Minute)

Table 2 Reconciliation of top-down and bottom-up models for local exchanges

In the case of tandem exchanges the reconciliation proved more difficult and the bottom-up estimate was 30 percent lower than the top-down estimates. In transmission too, major differences in the estimated costs of individual segments were found, although the estimated annual costs of all transmission services were similar in both cases. NERA have identified possible reasons for this difference, but it has not proved possible to estimate their significance. The results of the attempted reconciliation are thus mixed, with differences observed in the level of costs (for tandem exchanges) and the structure of costs (for transmission). In the face of these discrepancies, OFTEL is continuing to work to reduce them. 6.0 Other Uses of Incremental Cost Data

I noted at the outset that cost data are also needed for other regulatory purposes, notably the estimation of the cost of universal service obligations (USOs) and to establish whether any particular set of prices offered by a dominant operator, whether in wholesale
273

(interconnection) or retail markets, is anti-competitive. It is therefore useful to consider whether the modelling described above can also be utilised for these purposes. The issue of universal service has been much debated in Europe, in the context of liberalisation. Some operators have argued that, if competition is introduced, universal service will be sacrificed because operators will not be able to afford to provide service to remote areas or disadvantaged customers. In response, it has been argued that the costs of providing universal service can be computed, and then shared among all operators (Cave et al. 1994). Clearly, computing the costs of universal service obligations involves identifying those customers or groups of customers who impose a net loss on their current operator customers whom that operator would not normally serve on the basis of ordinary commercial considerations. In order to identify such customers, it is necessary to establish the revenues associated with their membership of the network and the costs which they impose upon it. The revenues of any group of subscribers can be computed from their current bills, which should be augmented by any profits to the operator associated with incoming calls which they receive, on the argument that if they left the network, revenues from incoming calls would be lost as well. It is generally agreed that the costs associated with serving any group of customers are best calculated on the basis of long run avoidable costs. In other words, the question to be asked is: If a particular group of customers left the network, what costs would the operator save in the long run, having made the necessary adjustments to network capacity? This concept of long run avoidable cost is effectively the same as long-run incremental cost, which can be estimated using methods described above. However, there is an important difference in the nature of the cost data required for the purposes of setting interconnection charges and establishing the cost of universal service obligations. In the former case, geographically averaged data are normally acceptable, because both retail and wholesale prices are typically uniform within a particular service area. Thus costs are computed for average routes within a charge band. In the case of estimating the cost of USOs, geographical differentiation is required, in order to focus upon especially high costs associated with particular geographical areas or customers types. Hence, although similar cost concepts are employed for both purposes, entirely different estimation procedures are required. An illustration of the calculation of the cost of USOs can be found in OFTEL (1995a). Acting as a competition authority, a regulatory body may also be asked to determine whether a particular wholesale or retail price is anti-competitive. This will normally involve comparison of the tariff for a particular service with some estimate of its costs. Conventional analysis from the standpoint of competition law and policy often regards as suspect prices below long-run incremental cost (which may in some conditions be predatory) or above the stand alone cost of providing the particular service in question as a sole product (as perhaps involving excessive returns). In order to implement these tests a regulator would thus have to compute the long-run incremental cost and stand alone cost of a particular service. In principle, this could be done using either the bottom-up or the top-down model, or a hybrid. Because tariffs are typically geographically averaged, information relevant to geographically averaged interconnection prices can be applied directly. Where allegations of predatory or excessive pricing are made with respect of retail services, sold to end users, the regulator would also have to investigate retail costs, such as marketing and

274

billing. For example, a competitor may claim that a dominant operator is operating a price squeeze in which the difference between that operators retail price and the cost of the network component required to deliver it, reflected in interconnection charges, is insufficient to cover retailing costs. Where data on the incremental costs, both of network services and of retailing are available, such an allegation can be investigated quite readily. In practice, because of the extreme levels of dominance which incumbent operators enjoy in many telecom markets, regulators sometimes impose rather tougher conditions on pricing than those adopted by competition authorities. Thus in the United Kingdom, OFTEL has recently proposed guidelines for the investigation of alleged anticompetitive pricing by BT, which state that pricing below long-run average incremental cost is likely to distort competition and would require justification by the dominant operator. In addition, prices above long-run average incremental cost but below fully allocated costs may have an anti-competitive effect: e.g. the fact that prices are below fully allocated costs could indicate the activity may be in receipt of a cross-subsidy or there could be significant barriers to entry which means that the pricing policy has the effect of deterring entry into the market. Where prices fall between long-run average incremental costs and fully allocated costs it would be up to a complainant to justify how the pricing policy of the dominant operator had an anti-competitive effect on the market. (OFTEL, 1996c, Annex F, Paras 41-42.) The final use of incremental cost data noted in the introduction related to the setting of price caps. Conventionally, price caps in telecom have controlled retail prices, but the same methodology can also be applied to network or wholesale prices. This approach is now being adopted in the UK by OFTEL, which now sets a network price cap, covering most network services, and a retail price cap, now applying only to certain residential customers (excluding those with atypically high bills). In setting the network price cap, OFTEL intends to allow BT to recover in its wholesale prices the incremental costs of providing the relevant services (including an appropriate return on capital) and a proportion of common costs. The process of setting the price cap thus begins with an estimate of incremental costs. OFTEL has indicated that for these purposes it proposes to use a hybrid of the top-down and bottom-up methods of computing incremental costs described above. The manner in which the cost data are to be used are described in more detail in OFTEL 1996c. 7.0 Conclusions

The UK modelling work described above is being replicated in a number of European countries, including France. The need for satisfactory incremental cost numbers will certainly not go away, and the jury is still out on which method of estimation is satisfactory. In all probability, regulators will choose to employ a mixed method, requiring the construction of both bottom-up and top-down models.
Endnote 1 Financial support from the ESRC under its Contracts and Competition Programme is acknowledged.

275