******************************************************** NOTICE ******************************************************** This document was converted from WordPerfect to ASCII Text format. Content from the original version of the document such as headers, footers, footnotes, endnotes, graphics, and page numbers will not show up in this text version. All text attributes such as bold, itallic, underlining, etc. from the original document will not show up in this text version. Features of the orginal document layout such as columns, tables, line and letter spacing, pagination, and margins will not be preserved in the text version. If you need the complete document, download the WordPerfect version or Adobe Acrobat version, if available. ******************************************************* THE USE OF COMPUTER MODELS FOR ESTIMATING FORWARD-LOOKING ECONOMIC COSTS A Staff Analysis January 9, 1997 Jay Atkinson Chris Barnekov David Konuch William Sharkey Brad Wimmer Table of Contents I. INTRODUCTION II. BACKGROUND III. CRITERIA FOR EVALUATING THE UTILITY OF ECONOMIC COST MODELS IV. MODEL STRUCTURE AND INPUT REQUIREMENTS A. Underlying Structure of Models 1. Existing Wire Center Approach 2. Geographic Unit of Analysis 3. Specification of Demand 4. Specification of Network Elements B. Modeling of Network Investments 1. Loop Plant - Feeder and Distribution 2. Loop Plant - Fill Factors 3. Loop Plant - Cable and Structures 4. Switching Investment 5. Other Investments C. Modeling Expenses 1. Capital Expenses 2. Operating Expenses 3. Treatment of Joint and Common Costs D. Summary V. CONCLUSION The opinions and tentative conclusions expressed in this paper are those of the authors and do not necessarily reflect the views of the Federal Communications Commission or any of its Commissioners, or other staff. I. INTRODUCTION 1. In the Telecommunications Act of 1996 ("1996 Act"),Congress sought to establish "a pro-competitive, de-regulatory national policy framework" for the United States telecommunications industry. This past year, the Commission has undertaken proceedings on universal service, interstate access charge reform, and local exchange competition to overhaul its current regulations in light of the 1996 Act. In these proceedings, the Commission has examined, in varying degrees, the use of forward-looking economic cost methodologies as a basis for determining universal service support levels, cost-based access charges, and pricing for interconnection and unbundled network elements. 2. Forward-looking economic cost computer models (also referred to as "cost proxy models" or "models") could enable regulatory authorities to estimate the forward-looking cost of network facilities and services without having to rely on detailed cost studies, prepared by incumbent local exchange carriers, that otherwise would be necessary. In addition, a publicly available cost proxy model could be useful to regulators by providing an independent check on the accuracy of incumbent LEC cost studies. This paper is intended to stimulate discussion on criteria for the evaluation of forward-looking cost proxy models in determining universal service support payments, cost-based access charges, and interconnection and unbundled network element pricing. At various points in this paper, we present the staff's current views on specific issues that have arisen in the course of our examination of the cost proxy models submitted in recent rulemaking proceedings. At other points, we identify issues and questions about which we have not reached any preliminary conclusions and that we are continuing to analyze. We anticipate that the Common Carrier Bureau shortly will issue a public notice seeking comments on the views, questions, and issues set forth in this paper. II. BACKGROUND 3. The 1996 Act has fundamentally changed telecommunications regulation by replacing the framework of government-recognized monopolies with one in which federal and state governments work in tandem to promote efficient competition and to remove entry barriers and regulations that protect monopolies. The 1996 Act, when fully implemented, should greatly reduce the legal, regulatory, economic, and operational barriers to entry in the local exchange and exchange access market, while preserving and advancing enhanced universal service goals. The local competition provisions of the Act confer three fundamental rights on potential competitors to incumbent local exchange carriers ("LECs"): the right to interconnect with other carriers' networks at rates based on cost; the right to obtain unbundled network elements at cost-based rates; and the right to obtain an incumbent LEC's retail services at wholesale discounts in order to resell those services. The Act also requires a fundamental restructuring of current regulatory mechanisms for funding universal service goals. The Commission, after receiving the recommendations of a Federal-State Universal Service Joint Board ("Joint Board"), is to define the services to be supported by federal universal service mechanisms, to support such services in a manner that is "explicit and sufficient," and to ensure that "every telecommunications carrier that provides interstate telecommunications services shall contribute, on an equitable and non-discriminatory basis, to the specific, predictable and sufficient mechanisms . . . to preserve and advance universal service." In its recently initiated access reform proceeding, the Commission seeks to reform its system of interstate access charges to make it compatible with the competitive paradigm established by the 1996 Act and with state actions to open local networks to competition. 4. Recent decisions by the Commission and the Joint Board have found that forward-looking economic cost models hold promise as a regulatory tool. For example, the Joint Board recommended that a forward- looking economic cost computer model of the switched public telephone network be used to identify high cost service areas and help determine the level of universal service support payments. The Joint Board declined to adopt a particular model, but instead recommended that the Commission continue to work with the state commissions to develop an adequate model that can be adopted by the Commission by May 8, 1997. Prior to a Commission decision in the universal service proceeding, state members of the Joint Board will submit a report to the Commission on the use of proxy models and their application for funding universal service. Similarly, in the recently initiated Access Reform proceeding, the Commission sought comment on using economic cost models to move interstate access charges toward economic cost. 5. Staff consideration of economic cost models is proceeding on two tracks. In its Recommended Decision on universal service issues, the Joint Board urged the Commission to conduct a series of workshops to enable federal and state staff to work with industry participants to refine the models for universal service purposes. On December 12, 1996, the Commission issued a Public Notice scheduling a set of workshops for January 14-15, 1996, and sent a letter to model proponents requesting that they provide the Joint Board with further information about the models prior to the workshops. In this paper, we discuss the possible use of the models in the Commission's universal service, access reform, and local competition proceedings, and whether a single model, or combination of models, might be sufficient for all three purposes. 6. Parties have submitted for consideration by the Commission and the Joint Board several forward-looking, economic cost models. They include the Cost Proxy Model ("CPM"), filed jointly by Pacific Bell and INDETEC International in June; the Benchmark Cost Model 2 ("BCM2"), submitted by Sprint and US West in July; and the Hatfield Model, version 2.2, release 1 ("Hatfield 2.2.1"), submitted by AT&T and MCI in May. In late August, we received the Hatfield model, version 2.2, release 2 ("Hatfield 2.2.2"), which is an updated version of Hatfield 2.2.1. These models originally were designed to determine high cost service areas and calculate universal service support payments, although they may also be used in setting interconnection, unbundled network element, and transport and termination prices. 7. In this paper, in Section III, we discuss the criteria for evaluating the utility of economic cost models in determining universal service support payments, cost-based access charges, and in setting prices for interconnection and unbundled network elements. In Section IV, we discuss model structure and input requirements. III. CRITERIA FOR EVALUATING THE UTILITY OF ECONOMIC COST MODELS 8. In this section of the paper, we discuss the criteria for evaluating forward-looking economic cost models. 9. Use of Forward-looking Economic Cost as a Basis for Pricing. In dynamic, competitive markets, firms base their actions on the relationship between market-determined prices and forward-looking economic costs. We define forward-looking economic costs as the costs that would be incurred if a new element or service were provided, or that could be avoided if an existing element or service were not provided, assuming that all input choices of the firm can be freely varied. This is often referred to as long-run economic cost. This "long run" approach ensures that rates recover not only those operating costs that vary in the short run, but also fixed investment costs that, while not variable in the short term, are necessary inputs directly attributable to providing the element or service. If market prices exceed forward-looking costs, new competitors will efficiently enter the market and bring pressure to bear on prices. If forward-looking economic costs exceed market prices, new competitors will not enter, and incumbent firms may decide to exit. These voluntary actions by firms produce efficient resource allocation by adjusting price and output until the value to consumers of additional output is just equal to the cost of the resources required to produce it. In contrast, basing prices on embedded costs would fail to establish the critical link between economic production costs and market prices, and would be inconsistent with the goal of efficient competition. Pricing based on forward-looking costs enables efficient providers to cover their costs and make new investments, while facilitating efficient market entry by potential competitors. We therefore believe that models should not include sunk or historically incurred costs. We also believe that this view is consistent with the Joint Board's conclusion that basing universal service support levels on the forward-looking economic costs of an efficient carrier will preserve and advance universal service by providing carriers with the correct signals for entry, investment, and innovation. 10. Ability to Measure Costs Relating to a Narrowband Network. We believe that a model for pricing services and unbundled network elements should, at a minimum, be able to estimate the full stand-alone cost of the minimum set of network elements capable of delivering traditional voice telecommunications service and narrowband data services, at currently acceptable quality levels, to customers of the public switched network and to private line users. We base this belief on the view that the inability to purchase elements required to provide these narrowband services creates a "bottleneck" that could prevent competitors from entering the market. We realize that incumbent local exchange carriers may choose to construct network facilities capable of providing services that require higher transmission speeds ("broadband" services). We are currently analyzing how we should evaluate the utility of models when an incumbent LEC is offering both narrowband and broadband network components. 11. Use of Proxy Models for Multiple Objectives. Proxy models may be utilized for multiple regulatory objectives, such as in a prescriptive approach to access reform, determining levels of universal service support in high cost areas, and the pricing of unbundled network elements. It is not clear from our analysis to date whether a single proxy model, or combination of models, can or should be used to achieve all of these objectives. For example, does a network specifically dedicated to universal service objectives differ in a significant way from the summation of network elements envisioned in Section 251? How should common costs be treated in the different applications -- e.g., universal service or access reform -- of the models? If broadband networks become prevalent, will a single model be capable of measuring costs of providing supported services (which are narrowband) and access services or unbundled network elements that are provided over a broadband network? What modifications, if any, would be required if models were used for multiple objectives. 12. Consistency with Independent Evidence. It may be possible to obtain independent estimates of the costs of some network elements as a check on the validity of model estimates. For example, to what extent would it be feasible and valid to compare estimates of loop costs with competitive bids for installing loops? Would cable systems' costs of installing similar elements provide a comparable estimate? Can econometric studies provide any check on model results? Another option would be for parties to provide engineering studies for a representative sample of Census Block Groups ("CBGs") that would evaluate the networks derived by the models by comparing them to engineering plans used to build actual networks using today's technology. This approach would help us determine whether the models accurately estimate the level of facilities necessary to provide service, or whether the derived networks under or over-build facilities. 13. It may also be instructive to compare estimates calculated by the models with data from Automated Record Management Information Systems ("ARMIS"). As discussed in section IV of this report, all of the existing models report levels of forward-looking investment that are significantly lower than embedded levels of investment reported in ARMIS data. In addition, some of the models report significantly lower levels of expense than are reported in ARMIS data. We note that there are several possible explanations for these differences: (1) Technological change may have reduced the level of investment required to provide a satisfactory level of service. If these changes were unanticipated, they may have resulted in underdepreciation of incumbent carrier's assets. (2) Existing models may choose investment levels that are insufficient to provide satisfactory levels of basic local telephone service. More broadly, as explained in detail below, the model proponents' network design assumptions and choices of inputs, such as fill factors and percentages of sharing of structures, can significantly affect the costs estimated by the models. (3) Input prices may have fallen over time, so that forward-looking costs are less than embedded costs for equivalent levels of network plant. (4) LECs may have engaged in systematic overinvestment or other non-cost minimizing behavior. 14. To understand further the source of the differences in investment, detailed engineering analyses of existing networks and the networks derived by Hatfield, CPM, and other cost proxy models, as discussed above, would be useful. It would also be instructive to compare physical measures of network investment, such as loop length, as reported by the models, with independent sources of such data. Finally, an examination of telephone plant price indices should be useful in measuring the effect of changing input prices. We believe that these, or other methods, may provide a useful independent check on the accuracy of a model's estimates of the costs of providing the network services or facilities at issue. 15. Potential for Independent Evaluation. We believe that the algorithms and judgments made by a proxy model's designer or operator should be clearly identified and explained so they can be independently evaluated by state or federal regulators. We recognize, however, that this criterion could be satisfied in different ways. For example, a model could utilize only publicly-available information. This would be allow full independent evaluation, but may sacrifice some accuracy. Alternatively, models could utilize proprietary information (such as vendor pricing data), which would be made available to third parties in regulatory proceedings under protective order. This approach may produce more accurate results but could be administratively more cumbersome to evaluate. We are currently analyzing the relative advantages and disadvantages of these approaches. 16. Flexibility. Some states may possess detailed information about important model inputs, such as discount prices offered by switch vendors, that model designers could only estimate. In addition, states may possess detailed information on local conditions, such as zoning restrictions and labor rates, that they may wish to add as inputs to a model. We believe that cost proxy models should permit states to utilize such information where available. Also since the models may be used at different levels of aggregation (e.g., state density zones for pricing purposes, as compared to wire centers or CBGs for universal service), a model should be sufficiently flexible to permit a user to vary model inputs. In our view, the more model inputs that users can vary, the more useful a model will be. IV. MODEL STRUCTURE AND INPUT REQUIREMENTS 17. An economic cost proxy model for estimating the cost of network elements starts with an engineering model of the physical local exchange network, and then makes a detailed set of assumptions about input prices and other factors. Such models estimate the total monthly cost for each network element. In this section, we explore both model design and models' use of variable input factors for network investment, capital expenses, operating expenses, and common costs. We seek to identify advantages and disadvantages of existing models. We will also consider alternative modeling assumptions and algorithms. A. Underlying Structure of Models 1. Existing Wire Center Approach. 18. Each of the currently available models is based on an assumption that wire centers will be placed at the incumbent LEC's current wire center locations. Subject to this constraint, all remaining network facilities are assumed to be provided using the most efficient technology currently in use. We note that the Recommended Decision of the Federal- State Joint Board on Universal Service included "the understanding that the models will use the incumbent LECs' wire centers as the center of the loop network for the reasonably foreseeable future." While the constraint to existing wire center locations is not fully consistent with a forward-looking cost methodology, we believe that a cost proxy model should, in the near term, include the above assumptions in estimating the cost of unbundled network elements and supported services, and that these assumptions are consistent with the Recommended Decision of the Joint Board. 19. We recognize, however, that over time an existing wire center model may become less representative of actual conditions faced by new entrants and incumbents. For example, after existing wire center locations were chosen by incumbent LECs, larger capacity switches became available. Because of ongoing advances in technology, facilities-based new entrants, and incumbent LECs, may in the future choose a much different network topology that will result in different forward-looking costs than today's network. We therefore believe that the models' assumption regarding the locations of LEC wire centers could be relaxed at some future time. 20. We also recognize that when the existing wire center approach is used, the specific interpretation of this assumption may affect the final cost estimates produced by a model. Therefore, we are continuing to explore various interpretations of the fixed wire center assumption. For example, we are examining whether the placement of remote switching units should be restricted to existing wire center locations, and if the models should assume that every wire center includes a (host or remote) switch. At this time we believe that the models do not need to assume a switch must necessarily be placed in each of the incumbent LEC's current wire centers. Such an assumption restricts the models' ability to estimate accurately the forward- looking costs of an efficiently designed network. 21. Similarly, wireless technologies may in the future be capable of providing narrowband telecommunication services at a lower cost than wireline technologies. We therefore are examining how models should incorporate wireless technologies into their estimates of forward-looking costs. We are currently considering whether there should be a cost-cutover, or threshold cost per loop that would trigger the use of wireless technology instead of wireline. We are not, however, aware of any study that attempts to estimate what this threshold should be. 2. Geographic Unit of Analysis 22. The BCM2 and Hatfield 2.2.2 models both use, as the basic unit of analysis, the CBG, as defined by the Bureau of the Census. Each CBG contains approximately 400 households, and therefore the number of square miles contained within a CBG varies inversely with population density. The CPM, by contrast, uses a geographic grid structure. The CPM's geographical unit is 1/100th of a degree of latitude and longitude (approximately 1/4 square mile), which its sponsors characterize as a "grid." This allows the CPM the flexibility to model the cost of various types of serving areas, such as wire centers or political jurisdictions, as well as CBGs. A grid structure may be preferable because it allows households to be matched more accurately with existing wire centers. The large number of grids relative to CBGs, however, increases the computing costs of running the model. This may require a simplification in the computations undertaken for each grid in order to offset the increase in computing costs caused by the large number of grids. Such simplifications could lessen the accuracy of a model's estimates. 23. Having recognized the relative advantages of these two approaches, we are continuing to investigate whether models should use either the grid or the CBG as their basic unit of analysis. Specifically, we recognize that when using a model to determine the size of universal service subsidies, a more precise geographic unit of analysis may be preferable, especially in rural areas, because the cost of providing telecommunications services may vary widely within a given area. 24. These issues also lead us to question whether the models provide for the appropriate number and size of pricing zones when reporting aggregate cost estimates above the CBG or grid level. We believe that if a model is used for pricing unbundled network elements or in setting cost- based access charges, the cost differences within each zone should be insignificant, compared to the differences across zones. Furthermore, if models are used for multiple objectives, we are concerned that any differences in the definition of zones may lead to uneconomic arbitrage. We are continuing to examine whether other aspects of the models that pertain to geographic deaveraging of rates should be considered. 3. Specification of Demand 25. An accurate estimate of the cost of serving a CBG or any other serving area depends on a reliable forecast of customer demand patterns within the area, and the number of residential and business lines. Each model relies on census data to determine residential demand. However, because census data does not report the number of business lines, model designers must use indirect methods to estimate business demand. The potential for error in estimating business and residential demand creates certain difficulties. First, as noted below, fill factors or utilization rates may be expected to vary between business and residential lines. Second, loop lengths are typically shorter for business lines than for residential lines. Thus, unless the differences in costs associated with different fill factors for business and residential areas happen to offset exactly the differences in costs associated with differences in loop lengths, the cost of serving an area will depend on the ratio of business to residential lines. An understanding of the magnitude of these competing effects, however, requires an accurate estimate of the number of business and residential lines in a particular area. The authors of Hatfield 2.2.2 assert that incumbent LECs maintain data on the number of business lines in a CBG, but refuse to make this information available. 26. Hatfield 2.2.2 incorporates access line demand data from the Operating Data Reports, ARMIS 43-08, submitted to the Commission annually by all Tier 1 LECs. The Hatfield 2.2.2 model therefore incorporates data on the number of: (1) residential access lines, both analog and digital; (2) business access lines, which include analog single lines and multi-line analog and digital lines, PBX trunks, Centrex trunks, hotel and motel long-distance trunks, and multi-line semi-public lines; and (3) special access lines. Hatfield 2.2.2 computes residential lines in each CBG by multiplying the number of households in a CBG by the ratio of total residential lines, as reported by ARMIS, to the total number of households in a study area. Similarly, Hatfield 2.2.2 multiplies the ratio of total reported business lines to total employees in a study area by the total number of employees in a CBG, as reported by Dun and Bradstreet, to estimate the number of business lines in a CBG. The BCM2 and CPM use similar approaches to estimate the number of business lines in an area. Each of the models' approaches to estimating business and residential loop demand appears to have drawbacks that may lead to significant modeling inaccuracies. We expect that different industries have different demands for telephone use per employee. For example, service industry demand for telephone service is most likely greater than demand in the manufacturing sector. We are continuing to explore whether there are alternative publicly available databases that might be used to estimate business demand for loops. 27. The Hatfield 2.2.2 model also uses data on the total number of employees in a CBG to estimate the number of special access lines and public access lines, where the latter includes lines associated with pay phones, but excludes customer-owned pay phone lines, in that area. The BCM2 makes an allowance for special access lines through a user-specified "Special Access Ratio." The default value for this ratio is 0.13, which means that there are 0.13 special access lines per every business line. Using these estimates, the BCM2 estimates the number of lines in the CBG provisioned at the DS-1 level. ETI reports that the BCM2 sponsors assume that CBGs with greater than 2,016 lines terminate a variable percentage of lines at the DS-1 level to "reflect the costs of providing service to digital PBXs and providing wideband private line services." 28. We believe that models should include the total demand for telecommunication services, which, at a minimum, should include the demand for first and second residential lines, business lines, public access lines, and special access lines. We are in the process of evaluating how second residential lines and business lines, as well as broadband loops should be incorporated in a model used to estimate the forward-looking cost of network elements and supported services. We note, however, that these different types of lines may be provided using shared equipment, and the exclusion of any lines may lead to an overestimation of per-line costs when economies and scale and scope are present in the delivery of telecommunications services. We also note that all three models rely on current demand patterns to estimate the demand for loops, rather than employing forward-looking estimates of loop demand. Because it is costly to increase a network's capacity or to build plant that will be under-utilized, we believe that the use of current demand, such as that found in ARMIS, rather than a forecast of demand over the service life of the network may lead to significant modelling inaccuracies. 4. Specification of Network Elements 29. In general, cost proxy models seek to estimate the forward- looking economic cost of a network used to provide local telephone services. Different models, however, may estimate the cost of networks that are not comprised of exactly the same network components. We believe, therefore, that model sponsors should be required to state precisely the elements included in the network and the services those elements are capable of providing. We are continuing to evaluate the appropriate set of network elements that model sponsors should include in any model used to price interstate access, supported services, or unbundled network elements in the near future. In general, we believe that models should be updated or modified as the range of services, and network elements used to deliver these services, evolves over time. 30. Hatfield 2.2.2 estimates the cost of providing the following network elements: Loop Distribution; Loop Concentrator/Multiplexer; Loop Feeder; End Office Switching; Operator Systems; Dedicated Transport; Common Transport; Tandem Switching; Signaling Links; Signal Transfer Point ("STP"); and Service Control Point ("SCP"). The original Benchmark Cost Model, BCM1, was designed to produce "benchmark" costs for the provision of local telephone service. As one of its sponsors has noted, the model was designed to identify high cost areas for the purpose of universal service funding, and not to set prices of network elements. Consequently, BCM1 did not explicitly model many components of the local exchange network necessary to provide local service. Proponents of the BCM2 model argue that it has been modified to correct for the deficiencies of the BCM1, and that the current version of the model includes all elements necessary to provide local telephone service. The BCM2, however, does not currently provide forward-looking cost estimates for unbundled elements. The CPM was also designed to identify the costs of universal service support, although CPM proponents claim that current versions of the model are able to identify specifically the costs of the following unbundled network elements: Loops; Local Switching, which includes the non-traffic sensitive cost of line ports and traffic-sensitive switching costs; and Tandem Switching. B. Modeling of Network Investments 31. To model the forward-looking cost of an efficiently designed network, cost proxy models must accurately estimate the quantity of facilities required for an efficiently designed network to deliver the services at issue. Each of the three existing models on record claim to produce such estimates. When applied to regional Bell Operating Companies as a whole, Hatfield 2.2.2 estimates a total forward-looking investment in network facilities of $769 per line, BCM2 estimates a corresponding value of $960, and CPM a value of $1057. Each of these values is substantially lower than the per-line investment of $1609 reported in 1995 ARMIS data for the RBOCs. The differences between forward-looking model investments and ARMIS investment could be attributed to various factors, including insufficient model investment, the model proponents' network design assumptions and choices of inputs, technological change, changing input prices, or overinvestment by incumbents. 32. In this section we examine the methodology used by the different models to estimate the quantity and type of physical facilities that an efficiently designed network would deploy. Several commenters have asserted, in proceedings where cost proxy models have been discussed, that such models do not generate networks capable of delivering telecommunications services. These commenters, however, have generally provided neither a detailed analysis of the models' flaws nor put forth any alternative proposals that would improve the models. Similarly, although sponsors of models claim to design networks capable of delivering telecommunications services, they have provided little independent evidence to verify their claims. 1. Loop Plant - Feeder and Distribution 33. The largest portion of a network's investment consists of its investment in loop plant. It is therefore vitally important that models estimate accurately the cost of loop plant sufficient to satisfy demand. "Loop plant" consists of all network facilities, including wires, telephone poles or conduits, drops, etc., connecting the end office switch and customers' premises. It would be helpful to our analysis for all model sponsors to list and define the loop plant components derived by their models. For example, the Hatfield model estimates the cost of loop plant by estimating the cost of the following elements: network interface devices; wire drops; block terminals; distribution cables; and feeder facilities. 34. The BCM2 and Hatfield 2.2.2 both rely, at least in part, on the original BCM1 to estimate the feeder and distribution plant required by an efficiently designed network to meet demand. The BCM1 used data from the Census Bureau, the National Exchange Carriers Association, and United States Geological Service Satellite Survey to estimate population densities, and the length of feeder, sub-feeder and distribution portions of the loop. In addition, the BCM1 incorporated data that accounted for the effect of terrain on the cost of outside plant. In calculating loop distances, BCM1 assumed that CBGs are served from the nearest existing wire center, and that all CBGs are square and that households are uniformly distributed. In addition the BCM1 assumed that four main feeder routes leave each wire center with sub- feeder routes placed at 90 degree angles from the main feeder route. BCM1 assumed that feeder plant reached only the border of a CBG and that four equidistant legs of distribution plant served the interior of the CBG. Using these estimates, BCM1 estimated the per-foot cable investment for distribution and feeder and allocated feeder costs to multiple CBGs. 35. In a California proceeding, GTE asserted that the BCM1 underestimated loop plant because it assumed that CBGs are square, households are uniformly distributed, and that feeder plant runs only to the edge of the CBG. GTE contended that these assumptions were especially troublesome when applied to rural areas. GTE also criticized the BCM1 for using ratios, which were multiplied by investments in cable, to estimate investment for structures (e.g., poles, manholes, conduit) that support the loop plant. As a result of this approach, structure costs are directly related to the cost of cable. Thus, a decrease in the price of cable would result in a decrease in the BCM1's estimate of structure investments, even if the cost of building these structures remained unchanged. 36. The BCM2 and Hatfield 2.2.2 retain many of the BCM1's basic features, but both have incorporated modifications in response to criticisms of the BCM1. For example, both models estimate structure costs on a per-foot basis, rather than multiplying total investment by a factor. Both the BCM2 and Hatfield 2.2.2 also alter the BCM1's methodology for addressing the effect of terrain on loop costs. The BCM2 includes a variable that captures the effect of a terrain's slope and increases the length of a CBG's loop plant when the slope variable exceeds a specified minimum value. Hatfield 2.2.2 increases the length of feeder and distribution cables in rocky areas. Additionally, the Hatfield 2.2.2 and BCM2 both replace the BCM1's assumption of four equal distribution legs with an approach that allows the number of equidistant legs to vary. 37. The Hatfield 2.2.2 updates BCM's 1990 data with 1995 household counts and adds data on the number of employees per CBG. The Hatfield 2.2.2 model also assumes that digital loop carriers ("DLCs") are integrated into the switch, and its sponsors state that this is consistent with current LEC practices. We are currently attempting to verify that integration of DLCs and switches is the least-cost method of delivering telecommunications services. 38. BCM2's sponsors claim that the assumption that households are distributed uniformly across the CBG is not reasonable in low-density areas. The BCM2 therefore assumes that all households fall within 500 feet of either side of the road in CBGs with less than 20 households per square mile. The BCM2 excludes the areas outside of this 500 foot buffer zone and assumes that the population is uniformly distributed throughout the remaining areas. By excluding areas outside of this 500 foot buffer zone, the BCM2 estimates the cost of constructing a network for an area that is smaller than the actual CBG's surface area and has a uniformly distributed population. BCM2 also includes a user specified "Maximum Copper Distance" that may trigger an extension of feeder plant into the CBG if the CBG's width is greater than twice the Maximum Copper Distance. Additionally, the BCM2 assumes that customers with loop costs greater than $10,000 would be served more efficiently by a wireless system and has, therefore, capped per-loop investment at $10,000. 39. The CPM uses a methodology similar to the BCM2 and the Hatfield 2.2.2 model to estimate the loop plant required to meet demand for telecommunication services. As discussed above, the CPM uses a grid, rather than a CBG, as its geographical unit of analysis. The CPM assumes that each grid is served by the wire center that is currently serving the majority of customers located in that grid. The CPM uses the latitude and longitude of each grid's centroid and the actual location of switches to calculate loop distances. These distances are then used to determine the amount of outside plant facilities that are needed and what type of loop technology will be used. The CPM then incorporates population density, terrain, soil type, and other geological factors to estimate the cost of loop plant. The CPM relies on the relationship between these factors and the cost Pacific Bell has incurred when placing loop plant in areas with, for example, a particular population density or soil type, to determine the effect these factors will have on the cost of provisioning loop plant. The cost estimates derived by the CPM therefore reflect the particular characteristics of Pacific Bell's embedded network. The BCM2 and Hatfield 2.2.2 models, by contrast, attempt to estimate the cost of providing loop plant that would be incurred by an efficient provider given current wire-center locations. The BCM2 and Hatfield 2.2.2 models employ algorithms based on what their sponsors claim estimate the minimum forward-looking cost of deploying loop plant. To the extent that changing market and technological factors make past decisions for deploying loop plant non-optimal, the CPM's approach does not accurately estimate the forward-looking cost of deploying loop plant. 40. The BCM2, Hatfield 2.2.2, and CPM have all attempted to estimate the costs of low-density areas more accurately than the BCM1, and have adopted different algorithms to do so. In order to evaluate fully these different approaches, we believe that model sponsors should provide us with independent evidence that their approach leads to an accurate estimate of the forward-looking cost of providing telecommunications service in rural areas. 2. Loop Plant - Fill Factors 41. All the models include assumptions regarding feeder and distribution utilization rates (also called "fill factors"). In each model, lower utilization rates increase total loop investment because the increase in capacity associated with lower fill factors increases the amount of loop plant used to deliver telecommunication services. Thus, the choice of fill factor can have a significant effect on total cost. While all models allow user inputs for these quantities, it is not obvious what levels should be used as inputs. In a well-engineered network, it is necessary to include unused capacity when constructing loop plant to reduce the likelihood of outages in the case of breakages and to account for growth in demand. Furthermore, optimal fill factors should vary over the service life of the plant, increasing as demand grows until more plant is put into service. While the BCM2 and Hatfield 2.2.2 have chosen very similar default fill factors, neither has provided a detailed justification for these values. For example, neither of the sponsors has justified the differences in the default fill factors between feeder and distribution plant. In addition, the sponsors have not explicitly stated whether a fill factor should be determined by taking the average fill used over the projected service life of the plant, whether fill factors should differ for fiber plant and copper plant, and whether optimal fill factors should take account of anticipated competitive interactions among firms. 42. Fill factors also may differ between business and residential markets. In residential markets, LECs traditionally place between one and one-half and two wire pairs per home in order to be able to provide a second or third line to premises without incurring construction costs. Thus, fill factors that are less than 50 percent may be reasonable for residential markets. In business-dominated wire centers, the rate of utilization depends on the proportion of businesses using Centrex service rather than PBX terminal equipment, because PBXs serve to concentrate traffic between the customer and the central office. Customers using PBX equipment therefore require fewer lines than customers using Centrex service. Depending on the relative use of Centrex to PBX equipment, and LECs' plans for marketing Centrex services, business fill rates could be either lower or higher than residential fill rates. We are therefore not convinced that the models accurately incorporate the effect that the ratio of business to residential lines will have on optimal fill factors. 43. In general, we believe that model sponsors have not adequately justified the default fill factors that they have utilized. We believe that model sponsors should include an analysis that addresses the effect that a variety of factors will have on fill rates. For example, in addition to the factors mentioned above, sponsors should discuss how fill factors should vary according to population density, network reliability standards, and the effect of special service obligations associated with a carrier's eligibility for universal service support payments. 3. Loop Plant - Cable and Structures 44. Having estimated the length of loops, all three models estimate the forward-looking cost of this loop plant. With respect to cable investments, all three models use default input prices to estimate the cost of loop plant, but allow users to specify different input prices. We believe that a model should be supported by independent evidence that the default prices chosen for cable, fiber, and other loop-related facilities, such as drops, pedestals, and network interface devices, are equal to the actual market prices of these inputs. We also note that the model sponsors have not indicated whether their default prices include any quantity discounts that may be offered when purchasing these inputs. In general, we expect that the prices for inputs, such as cable and fiber, that would not be subject to non- disclosure statements by their vendors, should be based on publicly available prices. 45. All three models also estimate the forward-looking cost of installing loop plant, which includes the cost of building or obtaining access to structures that support the loop plant. Structures for cable plant consist of aerial, buried, and underground (i.e., cable in conduit) facilities. The BCM2 and Hatfield 2.2.2 models assign different default values for the proportions of each type of installation and for the cost of installing each type of cable. These differences can have a significant effect on estimated model costs. We believe that the treatment of forward-looking structure costs raises difficult modelling issues, and that none of the models is satisfactory in this regard. A crucial variable is the proportion of plant that is installed in new developments (where installation costs are relatively low) to plant installed for existing business and residential users. We recommend that model sponsors provide additional independent evidence on the appropriate mix of aerial, buried, and underground cable for use in a forward-looking cost study. 46. Different assumptions about sharing of structure costs can also have a significant effect on estimated model costs. Hatfield 2.2.2 allows users to specify the percentage of structure costs that are assumed to be shared with other users of these structures. Thus, Hatfield 2.2.2 includes only a fraction of total structure costs -- the default value is one third. BCM2 does not provide for such sharing, and includes the total cost of such structures. We believe that the default assumptions of BCM2 (no sharing) and Hatfield 2.2.2 (equal sharing by three utilities) are both simplistic, and that further investigation is needed by model sponsors on the sharing fraction that is most appropriate for the estimation of forward-looking costs. 47. Some parties have argued that, as a consequence of the fixed wire center assumption, the computation of forward-looking structure costs should take account of existing sunk investments in structure facilities. Consistent with the Universal Service Recommended Decision, we believe that proxy model estimates of structure investment, like all other model investments, should be based on forward-looking costs without regard to sunk investment. Because of this approach, as discussed in more detail below, we believe that choosing an accurate estimate of the economic life for these structures is crucial to estimating accurately the forward-looking cost of loop-plant structures. 4. Switching Investment 48. Having determined the number of lines assigned to a wire center, the BCM2 and Hatfield 2.2.2 determine the number and size of the switches to be placed in these wire centers. The BCM2 determines switching capacity based on the number of lines to be served by the switch and the nationwide average of dial equipment minutes ("DEMs"). The BCM2 assumes that LECs install one of four different switch sizes or a remote switching unit. BCM2 assigns fixed switching costs based on the number of lines served by the switch, and assigns a variable per-line investment of $100 for each line served by the switch. Hatfield 2.2.2 determines the investment in switches and interoffice transport based on the number of lines and DEMs, along with Bellcore assumptions on, among other things, busy hour call attempts. Hatfield 2.2.2 uses data from a McGraw-Hill study of the central office equipment market to derive average per-line prices for switching investment. Hatfield 2.2.2 also includes separate costs for the buildings, land, and other inputs to determine investment in switching. The CPM determines switch size based on the population density in a grid and uses Bellcore's switching cost information system ("SCIS") to estimate the investment in switching. While Hatfield 2.2.2 and BCM2 appear to take different approaches to estimating the forward-looking cost of switching, their estimates are very similar. We note, however, that the models do not currently identify any other factors, such as expected growth in demand, that may affect the switching capacity installed in an efficiently designed network. 49. The Hatfield 2.2.2 and BCM2 sponsors both assert that switch vendors typically grant carriers substantial discounts when selling switches, and require carriers to sign nondisclosure covenants that require carriers to keep actual prices for which switches are sold confidential. These sponsors also make similar claims for other electronic equipment, such as digital loop carriers. Hatfield 2.2.2 typically assumes higher discounts than BCM2. The proprietary nature of these discounts does not allow us to determine what level of discount should be applied to switching and electronic equipment prices in these models. We recommend that models should be supported by information on the actual level of discounts, provided the information remains proprietary. 50. The BCM2 assumes that the total cost of switching increases with the number of lines served by a switch. Hatfield 2.2.2 assumes, by including flat-rated port charges, that a portion of a switch's cost is sensitive to the number of lines served by a switch, but that these costs do not vary according to the number of minutes switched. The models all assume that the proportion of a switch's cost that is not traffic sensitive is constant across all switches in the network. Ameritech, however, claims to the contrary that this percentage varies according to the type of switch. If these non-traffic sensitive costs are not constant across all switches, we expect that cost- minimizing carriers would install switches whose costs are largely traffic or non-traffic sensitive depending on the type of traffic that will be switched in an area. For example, in an area that switches a large amount of traffic with long holding times, it may be cost minimizing to install a switch whose costs are largely non-traffic sensitive. We are, therefore, not convinced that the models' current treatment of non-traffic sensitive switching costs produces an accurate estimate of the relative proportion of traffic-sensitive and non- traffic-sensitive costs. 5. Other Investments 51. Hatfield 2.2.2 explicitly models the interoffice network, including the SS7 network, by calculating the distance between existing wire centers and using assumptions concerning traffic patterns. BCM2 gives much less detail and simply accounts for the interoffice network by increasing total investment by a small percentage. We believe that Hatfield 2.2.2's approach of modeling the cost of each element of the interoffice network is superior to BCM2's approach of adding a lump sum to its estimate of loop plant and end office switching investment. C. Modeling of Expenses 52. A cost proxy model ultimately produces an estimate of the annual or monthly cost of producing interstate access service, supported services, or a set of network elements. Annualized cost consists of the sum of the return on equity, taxes, interest, depreciation, network operations and support expense, customer operations, and corporate overhead. In a comparison of Hatfield 2.2.2, BCM2, and CPM, we find that differences in estimated monthly cost per line are substantially greater than the differences in the model's estimates of total investment. Hence the expense side of a model can have a significant effect on its final cost estimate. 1. Capital Expenses 53. In this section we analyze the relative advantages and disadvantages of each existing model's method for computing capital expenses, which include return on equity, taxes, interest, and depreciation expenses. We first discuss the methodologies used by existing models to compute capital-related expenses. Each of the three models calculates capital costs somewhat differently. BCM2 does not explicitly reveal its cost of capital and depreciation rates. These values are implicit in three annual costs factors that are used to derive annual expenses. In the CPM and Hatfield 2.2.2, capital expenses are computed as the sum of a return on investment, taxes, and depreciation. In Hatfield 2.2.2, the return on investment is equal to the net investment base (gross investment minus accumulated depreciation) multiplied by a rate of return equal to a weighted average of the cost of equity and the cost of debt, with weights equal to the corresponding percentages of equity and debt in total investment. Taxes in Hatfield 2.2.2 are equal to the product of the net investment base, the percentage return on equity, the percentage share of equity and a "tax gross up" factor determined by the following equation: 54. For each category of plant, the capital cost is computed for each year of the economic life of the plant and the resulting stream of returns is "levelized" through a net present value calculation to give a constant annual cost of capital for that category of investment. Aggregate capital costs are then computed as the sum of the capital costs for each category of plant. 55. The CPM computes capital costs in a manner that is conceptually similar to the Hatfield 2.2.2. The CPM approach, however, includes an adjustment for the difference between book depreciation and tax depreciation in computing its net investment base. These differences in investment base will produce differences in capital expenses between the CPM and Hatfield 2.2.2 even if both models produce the same network investments. The sponsors of the CPM have not, however, furnished any justification for the use of tax depreciation rates in a forward-looking cost study. 56. In addition to methodological issues, a number of additional issues must be resolved in order to obtain accurate capital cost estimates. In both the CPM and Hatfield 2.2.2, depreciation rates for categories of network investment may be specified by the user of the model. In addition, Hatfield 2.2.2 allows the user to specify the composite tax rate, shares of debt and equity in total investment, and the costs of debt and equity financing. Each of these factors will have a direct impact on the total capital expenses predicted by the model. 57. The forward-looking cost of capital is a weighted average of the forward-looking cost of debt and the forward-looking cost of equity. Hatfield 2.2.2 specifies default values of 7.7 percent for the cost of debt, 11.9 percent for the cost of equity, and a 55 percent proportion of equity financing. These assumptions imply a value of 10 percent for the cost of capital. We believe that, when estimating the forward-looking cost of capital, models should rely on market-determined costs for debt and equity as well as debt-equity ratios chosen by firms. 58. We are in the process of evaluating several alternative approaches to determining the market-based cost of capital that do not require a cost of capital proceeding. For example, USTA, in another proceeding, proposed using the cost of capital implicit in the U.S. National Income and Product Accounts to compute capital cost in a Total Factor Productivity Study. USTA argued that because capital markets are national and because risk levels for telephone assets are similar to those for other assets in the U.S. economy, year-to-year changes in the telephone industry's cost of capital should follow year-to-year changes in the U.S. economy's cost of capital. Alternatively, an implicit rental price for capital could be computed by dividing property income by the real capital stock, where property income is the difference between revenues and expenses on labor and materials. The real capital stock could be constructed by using the Perpetual Inventory Model. Although we recognize that these methods are inherently not forward-looking, we are continuing to investigate whether either approach can be used to obtain an accurate estimate of the forward-looking cost of capital. 59. The second component of a capital expense computation is a model's choice of depreciation rates. As described above, higher levels of depreciation lead to lower levels of investment base, and consequently lower annual expenses associated with return on investment and income taxes. Thus, changes in annual capital costs caused by changes in depreciation rates will automatically be mitigated to some extent by offsetting changes in return and taxes. 60. Hatfield 2.2.2 uses default asset lives that result in a depreciation rate of 6.56 percent for the Regional Bell Operating Companies, which corresponds to an average plant life of approximately fifteen years. The CPM uses a composite depreciation rate of 8.9 percent, corresponding to an average asset life of 11.9 years. The 1995 ARMIS data provide a composite depreciation rate of approximately 7 percent for Regional Bell Operating Companies, corresponding to an average plant life of 14 years. This rate is greater than the Hatfield 2.2.2 depreciation rate, but less than that used in the CPM. 61. We believe that depreciation schedules specified in a proxy model should be based on forward-looking costing principles and should reflect projected economic lives of investments rather than physical plant lives. As discussed above, we believe that the reported plant lives for loop-plant structures, such as conduit, manholes, and poles, are particularly important. Because of the relatively large investment necessary to construct such facilities, inaccurate estimation of the expected economic lives of such facilities may result in a significant under or overestimation of the forward-looking cost of these facilities. We also believe that the depreciation rates reported by incumbent LECs for financial purposes may provide information to determine the appropriate economic lives of facilities. We are continuing to evaluate the use of depreciation rates reported in ARMIS data. 62. We are also aware of alternative measures of depreciation that could be used to estimate forward-looking depreciation rates. For example, USTA has proposed that asset lives computed by the Bureau of Economic Analysis ("BEA") and the Bureau of Labor Statistics could be used to calculate economic depreciation rates for the LECs. Alternatively, depreciation rates derived from the Hulten-Wykoff formulas, which link depreciation rates to expected lifetimes (BEA lifetimes), may be appropriate. Finally, USTA proposed the use of economic depreciation rates from a study by Jorgenson for determining capital costs in a total factor productivity study. We are currently investigating whether the economic depreciation rates published by Jorgenson are appropriate for use in a model. It may be important to determine whether depreciation rates should differ depending on what services carriers expect to provide over an existing facility or the facility that will replace the existing facility. For example, the depreciation rate for copper cable may be affected by a carrier's plan to offer broadband services. Because broadband service may not be a supported service, should the depreciation rate used to determine the level of support for universal service differ from that used to price unbundled elements? 63. As noted above, all of the models estimate a forward-looking level of network investment that is significantly less than total investment levels recorded in ARMIS data. In addition, net investment (gross investment minus accumulated depreciation) reported by ARMIS is significantly greater than a comparable measure of net investment derived from Hatfield 2.2.2. We are unable to conclude at this time whether these differences are a result of insufficient levels of model investment for providing required facilities or services, the model proponents' network design assumptions and choices of inputs, the non-economic depreciation policies utilized in the past, or inefficient overinvestment decisions by incumbent carriers. For example, past depreciation policies may have resulted in under-depreciation of assets because of unanticipated technological change, or because they did not account for changes in input prices that may have reduced the forward-looking cost of provisioning a network. We believe that it may be important to determine the extent to which these and other factors may account for the above noted differences in network investment reported by ARMIS and estimated by the models. 2. Operating Expenses 64. In this section we discuss methods of computing non-capital- related expenses. These account for over one-half of the total annual cost of the network in some models, and include expenses related to both plant- related operating expenses and non-plant-related expenses. As noted above, the variation in the estimates of the total monthly cost of providing network elements, which includes operating and overhead expenses, produced by Hatfield 2.2.2, BCM2, and CPM is significantly greater than estimates of underlying network investments. Based on our analysis of these models to date, we believe that differences in the treatment of operating expenses may account for significant differences among the models and between the models and ARMIS data. 65. Both BCM2 and Hatfield 2.2.2 use annual cost factors to calculate non-capital-related expenses. An annual cost factor is the ratio of expense booked to a specific account and the gross investment booked to the same account. Typically, the expense associated with investment is the product of the model-generated investment and the associated annual cost factor. Annual cost factors are used by models, as well as by companies in individual cost studies, because methods for developing forward-looking expenses are complex and contentious. In the BCM1, a single expense factor, derived from nationally averaged accounting data, was used to convert network investments into monthly costs at the CBG level. In the BCM2, three separate factors are applied to three aggregate categories of plant investment: cable and wire; circuit equipment; and switching equipment. These factors are used to estimate the total level of capital costs, operating expenses, and corporate overheads. In Hatfield 2.2.2, network operations expense, and attributable support expenses are computed for each plant account. Operating expenses are based on historical expense factors calculated from balance sheet and expense account information in carriers' ARMIS reports on a state-by-state basis. Network-related expenses, which vary with capital investment or number of lines, are allocated accordingly. Non-network operating expenses are allocated based on data from comparable support expenses in competitive industries. 66. An alternative to the annual cost factor approach is used by the CPM, which employs an activity-based costing approach that uses accounting methods to trace expenses at a highly disaggregated level. The CPM uses Pacific Bell's 1994 per-line maintenance and repair expenses, adding a fixed amount per loop. Some adjustments are made to reflect a forward-looking methodology. For example, maintenance expenses for analog switches are excluded. While this approach is potentially able to provide an accurate accounting of expenses at any point in time, there may be two potential problems with it. First, the underlying data required may be proprietary and specific to each operating company, and thus, verification of model results may be difficult. Second, the methodology uses historical data rather than forward-looking data. 67. We are currently in the process of evaluating specific alternatives to the use of annual charge factors or accounting-based methods. For example, a different annual charge factor could be computed by taking the ratio of current expenses to a measure of current investment, which could be computed by revaluing embedded investment at current input prices using telephone plant price indices. We are evaluating the feasibility of this approach, and the consequences of using it in the determination of forward-looking operating expenses. We also recognize that quality of service practices and guarantees differ by type of customer. Typically, lines used by interexchange carriers and multi-line business customers are repaired faster than residential customer lines. If adequate data were available, these practices could potentially justify specific maintenance factors. 68. A different approach to estimating expenses might be to make use of yardstick comparisons in which, for each category of expenses, explicit comparisons would be made of current year expenses (or an average of expenses over the past three years) among all companies of a given size or type. Assuming that the methods of accounting for expenses across companies were consistent with each other, the forward-looking cost for each expense category would then correspond to the lowest observed cost. 69. Another approach, based on econometric methods, might be to specify non-capital-related expenses as a function of the amount of investment and the volumes of output. Historical data would then be used to estimate the parameters of the assumed functional form. This approach could be used to estimate expenses given levels of investment from an engineering study. We note, however, that any econometric approach is based on a relationship among historical variables, and we believe that such approaches must be cautiously interpreted in estimating forward-looking expenses. An econometric approach could also be used to estimate the total cost of network elements, as a function of loops, DEMs, and trunking facilities. Given appropriate treatment of the price of capital (based on the risk-adjusted cost of capital and economic depreciation rates) such an econometric cost function could represent the forward-looking cost of network elements. 3. Treatment of Joint and Common Costs 70. If proxy models are used to estimate forward-looking economic costs, the question of joint and common costs must be addressed. In the case of pricing of unbundled network elements, costs that are common to a set of network elements can be allocated among the individual elements in that set. For example, shared maintenance facilities could be allocated to the elements that benefit from those facilities. Common costs also include costs incurred by the firm's operations as a whole. Given these joint and common costs, setting prices for individual network elements based on forward-looking incremental costs alone would not recover the full forward- looking cost of the network. In the Local Competition Order, the Commission concluded that recovery of forward-looking joint and common costs is appropriate under a forward-looking economic cost paradigm, and that a reasonable measure of such costs should be included in the prices for interconnection and unbundled network elements. 71. If proxy models are used in determining universal service support payments or in setting cost-based access charges, additional issues are raised in the treatment of joint and common costs. Each of the proxy models addresses these issues differently. BCM2 assumes common costs are equal to 75 percent of the ARMIS per-line common costs. Hatfield 2.2.2 assumes that corporate overhead expenses vary with the size of the firm, and the model attributes a fixed proportion of aggregate total cost, set by default at 10 percent, to overhead expenses. The CPM assigns a fixed amount of joint and common costs to universal service based on Pacific Telesis accounting data. Current versions of the CPM also allow for a variable overhead allocation similar to the Hatfield 2.2.2 approach. 72. Based on our review, we believe that proxy models do not currently offer adequate justification for their calculation of forward-looking joint and common costs. Additional evidence is needed to justify their treatment of these costs. We are also examining alternative methods outside of the models, including econometric approaches, that might be used to establish an appropriate level of forward-looking joint and common costs. D. Summary 73. As explained in detail above, acceptable cost proxy models should estimate accurately the forward-looking cost of operating a telecommunications network providing unbundled network elements, supported services, or access services. While treatment of all modeling variables is important, our current understanding of the models leads us to highlight a number of areas in which we believe that additional modelling effort or supporting studies may be warranted. The core of a proxy model consists of the algorithms that it uses to determine total network investment. We are particularly interested in evaluating a model's ability to estimate total loop investment. For example, we believe that additional justification of a model's choice of fill factors and treatment of structure costs would be desirable. On the expense side, we believe further study is required to determine the appropriate forward looking cost of capital and rates of depreciation. We also believe that model proponents should further refine the methodologies that current models use to estimate forward-looking operating expenses. Since these expenses may comprise, in some models, over one-half of the total costs of network elements or supported services, we believe that additional supporting studies of non-capital expenses by model sponsors and outside parties would be desirable. V. CONCLUSION 74. By releasing this paper, we seek to stimulate discussion that will assist state and federal regulators in evaluating, and industry participants in designing, cost proxy models for possible use in pending Commission proceedings. We look forward to working with all interested parties in developing reasonable approaches to using economic cost models on these critical telecommunications policy issues.