ATTRIBUTES OF
A USEFUL
COST PROXY MODEL

2.1 Attributes of a reliable cost proxy model that would be useful for public policy purposes

Since the primary objective of this report is to evaluate the BCM as a tool for quantifying the need, if any, for high cost support, it is important to first identify the desirable attributes of a cost proxy model. Describing these attributes and assigning some priority to them will serve as a "yardstick" against which to measure the BCM. For the reasons discussed below, we conclude that a cost proxy model should possess the core attributes of relevancy, reliability (or accuracy), accountability (verifiability), and simplicity. It is both desirable and achievable for a cost proxy model to incorporate each and all of these attributes, although there will on occasion be tensions among them.

Of these four core attributes, clearly the most important is relevancy, i.e., the model must be addressing the correct policy question, "the cost of universal service" in this instance. The model must correctly address and provide cost estimates for only those services that are to be eligible for universal service support.[1] The model must, of course, provide reliable and accurate (realistic, but not necessarily precise) results. This means not only that all "hardwired" data must be based upon the requirements of "basic" service, but that user-specified inputs must also be carefully scrutinized to ensure that they are, in fact, modeling the correct definition.[2]

The goal of universal service is to provide universal connectivity of all US households to the public telecommunications network. That goal is satisfied by the availability, in each household, of one primary residential exchange access line; additional residential access lines in the same dwelling unit are not necessary for the achievement of universal service and should not be included in a USF analysis. This condition must be accurately reflected in the model; i.e., the model must be designed to examine the incremental cost of providing one and only one residential access line per legal dwelling unit. This restriction affects a number of the engineering assumptions in the model because, in practice, local telephone switching and distribution infrastructures are designed to satisfy more than the stand-alone demand for primary residential access lines.[3] Similarly, where the requirements of basic local exchange service can be met by a simpler and lower cost technology or architecture, that approach, rather than any "conventional" method of providing service, must be modelled. For example, in some isolated and insular areas, wireless technology may be a far less costly means of providing basic network connectivity than the traditional wireline architecture. In such cases, the appropriate cost for inclusion in the model is that for the wireless method of service. Similarly, where a LEC chooses, for strategic reasons unrelated to the provision of the primary residential access line, to serve a particular rural exchange with a dedicated central office switch rather than employing a remote service unit (RSU) connected to a more distant host, the relevant cost for purposes of universal service funding is the least cost means of satisfying the primary line residential demand, even if that is not the least cost means of serving the entire demand in the community.[4]

It is also important that each investment and expense input assumption reflect the actual forward-looking costs that LECs are likely to incur. While it might seem that such information would be readily available from companies that have operated, for many years, under extensive and detailed corporate and regulatory accounting regimes, this has not been the case. In a recent submission to the California PUC, Hatfield Associates, Inc. reported that "[t]he inputs necessary for the model to determine plant investment in a forward-looking network design are notoriously unavailable from equipment manufacturers and their customers, the Local Exchange Carriers." The report goes on to say:

It is difficult to obtain even the list prices of most types of telecommunications equipment. Even if they were readily available, the discounts large LEC purchaser obtain are necessary to determine the effective prices the LECs paid. Hatfield Associates Inc. ("HAI") understands that such discounts can be impressive: e.g., in the vicinity of 50% or more for switching equipment. The discount is therefore vital to the investment estimation.[5]

Thus, a key issue that regulators must address, if a cost proxy model is to be adopted, is how to obtain and then to verify the accuracy of investment and expense data.[6]

Another aspect of reliability is the ability to update critical inputs and assumptions. Both technological and economic forces work on the prices of many of the major inputs to the proxy model. As we enter a more competitive era, competition may result in deeper discounts for switching and other equipment, and in significant technological advances that lead to further decreases in costs overall. In some cases, the effective cost reduction is reflected in the form of capacity increments rather than in cost decreases per se; for example, by 1997, advances in electronic multiplexing equipment will make it possible for a single strand of fiber optic cable to carry 40 Gbps of digital bandwidth, as compared with a maximum capacity of only 2 Gbps in 1989.[7] Thus, while the installed cost of fiber cables might be relatively stable (or perhaps even increasing slightly), on a per-unit-of-capacity basis the cost of fiber optic transport is dropping like a rock.

In discussing the need for accurate and reliable data, a note of caution is necessary concerning the temptation to judge the reliability of a cost proxy model by whether its results can precisely match the cost results reported by the LECs through ARMIS or other embedded cost studies. It is neither necessary nor desirable that such data be completely comparable. First, it cannot be assumed that any failure to match is attributable to an error in the cost proxy model, rather than in reported costs. Second, since the cost proxy model is modeling long run incremental costs for an optimized network that is limited to providing basic universal residential connectivity, rather than the actual multipurpose networks that are constructed in practice, the costs are by definition not comparable, even if the embedded ARMIS-type costs were otherwise equivalent to the long run incremental costs of the proxy model.

Without a rigorous examination of the core assumptions and inputs, there is a substantial risk that the cost inputs of the model will be inflated and the engineering assumptions overstated to produce a higher than necessary support requirement. Thus, it is hard to separate the attribute of reliability from that of verifiability. The key to ensuring that the information being used to drive the model is accurate is to make it available for thorough public examination, with full disclosure of underlying data sources and an opportunity for in-depth analysis of that information by interested parties. Moreover, because the model is intended to be a tool for public policy deliberations, the model should be "open," that is, any user should be able to replicate the results, modify key inputs, and determine the source and accuracy of the underlying data.

The expectation in developing a cost proxy model, such as the BCM, is that it be widely deployed across multiple jurisdictions. This is far more likely to happen if the basic algorithms in the model are easy to follow. Along the same lines, it is important that any formulas incorporating inputs that will be user-designated or that will require regular updating be designed to facilitate such changes.[8]

Balancing of core attributes

Not surprisingly, optimization for any one of the goals described above may diminish the cost proxy model's ability to fully capture one or more of the other goals. For example, as the detail of specifications increases, accuracy may increase, but some simplicity and verifiability may be sacrificed. Similarly, the best publicly available data may not reflect market reality. This is best illustrated with respect to switch and other component prices, for which the "list" price is commonly understood to be far greater than the net discounted prices that are actually being paid by LECs under sales agreements that have, traditionally, been treated as competitively sensitive and thus not subject to public disclosure. In our example, given the large impact that switching costs and other component prices have upon the accuracy of the BCM, it is clearly worthwhile to require LECs to disclose the net prices actually paid to vendors, even though the need to afford some confidential treatment to this information may pose some limitations on public discussion of the numbers. Finally, there must be a balance between getting the model near perfect and getting it completed and implemented. In balancing these competing considerations, policy makers should be guided by the Act's overriding purpose of promoting competitive opportunity and defining universal service support in a manner that is explicit, specific and competitively neutral. We have attempted to balance these competing considerations in our analysis and recommendations.

2.2 Defining the costs to be estimated by a cost proxy model for Universal Service

A prerequisite to formulating or evaluating a cost proxy model for universal service is to carefully define the nature of the costs that are to be modelled. As is generally true for any cost study, the appropriate definition of costs depends, in turn, upon how the results of the model are to be used. In this case, the public policy objectives for a cost proxy model have been established in part by the universal service provisions of the Telecommunications Act of 1996,[9] which calls for the FCC to assemble a Federal-State Joint Board to define the scope of universal service, after which the FCC and state PUCs must develop "specific, predictable, and sufficient Federal and State mechanisms to preserve and advance universal service."[10] Importantly, in the competitive local exchange marketplace also envisioned in the Act, universal service funding is not limited to incumbent LECs, but to any and all "eligible carriers," including new entrants, that accept universal service responsibilities and are so designated by the state PUC.[11]

These requirements have two direct consequences for defining how costs should be viewed in a proxy model. First, the basic telecommunications service to be costed in the proxy model ultimately will have to reflect the definition of universal service recommended by the Joint Board and implemented by the Commission. Second, the proxy model should estimate the costs to supply the defined universal service as they would be confronted by any local exchange services provider, rather than to determine the costs incurred today by the incumbent LEC under its particular network configuration and architecture. The latter objective is central to the "proxy model" concept,[12] and has been overlooked by some incumbent LECs that have focused instead upon modeling costs specific to their own networks.[13]

Although the Joint Board investigation of universal service is just beginning, it is essential to consider how basic telephone service is likely to be defined. The Act provides specific guidelines to the Joint Board to assist with the task of defining the scope of universal service. When determining whether specific services should be recommended for universal service treatment, the Joint Board is required to consider whether the service:

* is essential to education, public health, or public safety;

* has been subscribed to by a substantial majority of residential customers as a result of the operation of market choices;

* is being deployed by carriers in public networks; and

* is consistent with the public interest, convenience, and necessity.[14]

As a starting point, we offer a definition drawing upon proposals advanced in industry discussions and pending state-level proceedings. The basic residential service for universal service purposes may encompass the following capabilities:

* single-line, single-party residential voice grade dial tone line providing access to local and toll calling;

* a defined, minimum level of local usage;

* access to emergency numbers, directory and operator assistance;

* touch tone dialing; and

* white pages listing and an annual directory.

The FCC, in the NPRM, identifies a similar, although somewhat less extensive, set of services that it suggests should be "among those core services receiving universal service support."[15] All of the elements enumerated above satisfy the Act's guidelines, particularly with respect to subscription by the majority of residential customers. In contrast, custom calling features (like Call Waiting, Selective Ringing, Call Return, and Caller ID) do not. In the same vein, additional residential access lines are ordered by only about 12.3% of residential subscribers,[16] with the greatest demand coming from customers in upscale communities.[17] Indeed, there are many more cellular telephones in use by consumers than there are additional residential access lines, yet no one would seriously propose that cellular be included within the scope of universal service at the present time.

Thus, the basic elements listed above are likely to be found necessary and sufficient to ensure that all citizens can participate in the public telecommunications network for social and business contact, regardless of their income or place of residence. Until a universal service definition has been formally adopted, however, a cost proxy model should permit the assumed elements of basic/universal service to be modified.

Once the basic residential service has been defined for universal service purposes, the goal of the proxy model is to estimate the economic costs of providing that service to all residential households in a given market area. Conceptually, a proxy model should develop the total service long run incremental cost (TSLRIC) for that defined basic service, taking the total output of the basic service as the increment of demand under study. One approach might be to consider only the costs of serving those residential customers that the LEC would otherwise not serve voluntarily at a defined "affordable" price level. Since the LEC would presumably serve all customers except those whose costs exceeded that price, the LEC would be capable of generating "contribution" from the remaining, lower cost customers (the significant majority of most LECs' customer base), at least some (if not all) of which could be used to offset the cost of serving that small minority of "highest cost" subscribers. Any claimed subsidy requirement would have to be developed relative to the total customer base for the defined basic service, net of the revenues generated by residential customers served at costs lower than the "affordable" threshold; hence, even if one were to limit subsidy funds to the highest cost customers, it would still be necessary to determine the costs and revenues associated with serving residential subscribers overall.

As with all TSLRIC applications, the definition of "long run" is crucial to the development of reasonable cost estimates for universal service. In a properly structured TSLRIC study, one seeks to identify and quantify all costs that would not be present in the long run if the LEC was no longer going to provide the service or functionality at issue. For this purpose, the "long run" should be defined to mean a planning horizon in which full replacement of all plant, equipment and other investment, as well as a major organizational restructuring, is physically possible if economically justified, but not so long that all existing plant will have become worn out and in need of replacement merely because of its age or physical condition. It is a time interval over which consideration of various architectural strategies, including but not limited to total plant replacement, incremental enhancement of existing resources, or some combination thereof, can be considered and acted upon in a prudent and economically efficient manner, or in which comprehensive organizational changes, such as the creation and/or elimination of entire functional units, can be considered and implemented. This TSLRIC approach will permit the identification of all costs that in the long run are caused by the defined basic service, that would not exist if it were not being provided.

One important parameter in creating a proxy model using TSLRIC cost methodology is the manner in which the geographic placement of network facilities is modelled. For modeling simplicity, it may be useful to begin with the existing wire center locations and outside plant routings to serve known customer locations (sometimes referred to as the "scorched node" approach). However, it should be recognized that the resulting "scorched node" approach will not necessarily produce the absolute, least cost network for providing universal service on a forward-looking basis. Advances in switching and distribution technology are affecting the optimal placement of network facilities.[18] Shifts in demographics and telecommunications usage patterns over time may also change the optimal locations of wire centers from those currently in use by incumbent LECs. Finally, the design of a network whose purpose is limited to serving primary residential access line demand may be far less complex -- and less costly overall -- than the multipurpose networks that are actually being constructed. All such adjustments that may have significant cost impacts, and that can be reasonably defined and modelled, should be incorporated into the proxy model in order to best approximate the forward-looking economic costs of providing universal service.

[2]In this regard, all assumptions/inputs that are directly affected by public policy determinations (e.g., the affordability threshold) should be specified as user inputs, rather than "hardwired" into the model.

[3]Significantly, the BCM itself excludes business access lines even though the actual LEC network is engineered and constructed to jointly serve both residential and business demand. It is entirely reasonable, for the purpose of developing a cost proxy model, to introduce engineering assumptions that are relevant to the specific question being addressed but that may profoundly impact the ultimate architecture of the network. That the LEC can realize significant scale and scope economies by designing its network to satisfy multiple sources of demand is not disputed: The question before us is the cost of providing universal residential connectivity, which consists of the stand-alone cost of an infrastructure design specifically for this limited purpose and offset by an appropriate share of the economies of scale and scope that are properly conferred upon the primarily residential exchange access line market.

[4]This is not to suggest that the LECs should not have the flexibility to construct their networks to include capabilities beyond those necessary for universal service, such as the enhanced capabilities provided by a digital switch rather than a more stripped-down RSU. The point is simply that the higher costs are not the appropriate ones to model for purposes of determining universal service support.

[5]AT&T Communications of California, Inc., MCI Telecommunications Corporation, and Hatfield Associates, Inc., A Discussion of Input Assumptions Used in the Hatfield Proxy Model, Response to Pacific Bell's Second Set of Data Request in California PUC Universal Service Fund Proceeding, March 11, 1996 ("California PUC, Universal Service Proceeding, A Discussion of Input Assumptions Used in the Hatfield Proxy Model"), at 2.

[6]This role is hardly a new one, and the process of obtaining and verifying such input data for the limited inputs required in a proxy model should be considerably narrower than what is required to investigate and verify the costs of each LEC to provide service in a wide assortment of high-cost exchanges.

[7]Brand, T.L. et al., "An Updated Study of AT&T's Competitors' Capacity to Absorb Rapid Demand Growth," AT&T Bell Laboratories, April 19, 1995 submitted as Attachment C to Pacific Bell's Second Set of Data Request in California Universal Service Proceeding, dated February 28, 1996, at 4-5.

[8]The BCM as originally published was implemented as a set of Microsoft Excel spreadsheets, in an extremely complex and physically imposing spreadsheet modelling effort. In fact, while probably permitting it to be completed more rapidly, the use of spreadsheet technology is entirely inappropriate for a model involving a data set of nearly a quarter of a million individual Census Block Group (CBG) records, each one of which contains some 14 elements of data. Even with the fastest Pentium processors available, a single replication of the entire model would require at least 110 hours and involve manual intervention of at least 20 to 30 hours. ETI estimates that a single full replication run would take a minimum of three weeks to complete. It is our understanding that a future revision (probably not the version that is expected to be issued in July) is being implemented in a more appropriate software technology, which should facilitate replication and analysis by others. We are concerned, however, that the implementation of the BCM in a high-level programming language (such as Visual Basic or C++), while vastly improving the program execution time, may make it even more difficult to examine and to understand the specific assumptions and algorithms that are hard-wired into the programming code. Accordingly, we would urge the Joint Sponsors to provide detailed documentation, as well as source code, for the revised model.

[9]State PUCs may have additional objectives for a cost proxy model.

[10]Telecommunications Act, Secs. 254(a)(1) and 254(b)(5).

[11]Id., Sec. 254(e), Sec. 102.

[12]NPRM, op. cit., footnote 1, at 19.

[13]See the discussion of Pacific Bell's Cost Proxy Model, Chapter 9, infra.

[14]Telecommunications Act, Sec. 254(c)(1).

[15]The FCC's initial definition does not include local usage or access to a white pages listing and annual directory. We have followed the FCC's practice of excluding access to telecommunications relay services (TRS) from the universal service definition since TRS is already funded by explicit support mechanisms adopted pursuant to the Americans with Disabilities Act. See, NPRM, op cit., footnote 1, at para. 16 and note 42.

[16]Statistics of Common Carriers ("SOCC"), FCC, 1994/1995 Edition, Table 2.5; Monitoring Report, CC Docket No. 80-286, Federal-State Joint Board ("FCC Monitoring Report"), May 1996, Table 3.6; Bureau of the Census, 1990 Census Population and Housing, available from http://www.census.gov.

[17]California PUC, Universal Service Proceeding, Deposition of Richard Scholl, Volume VII, March 26, 1996, at 964.

[18]For example, the penetration of fiber optic technology into the feeder network is reducing the distance sensitivity of feeder costs, just as it had previously done for interoffice and interexchange transport. Second, the increased modularity of newer-generation digital switches is allowing LECs to deploy switching resources more flexibly. As a result, many LECs have been consolidating wire centers, chiefly by deploying host/remote systems in which remote switching units operate as an extension to a centralized, remotely-located host switch.