V. Analysis of Models
We first examine which model better complies with TELRIC, the Commission's CCPs, and our modeling criteria. Parties have filed extensive critiques of the two models at issue and there is a long litany of alleged flaws. It is often a daunting task to sift through and delineate critical flaws from superficial ones. The following section examines the more significant flaws that parties have alleged.
A. Flaws in the Verizon Model
Beginning with the Verizon model, JC, TURN, ORA and XO contend VzCost is seriously flawed and does not adhere to FCC, TELRIC or Commission modeling criteria. While Verizon claims its model replicates its existing network while updating equipment and operating costs, JC and others criticize Verizon's model as "part fish, part fowl." They contend VzCost artificially distorts current network design and creates a network with structural flaws resembling nothing that was or ever will be built. (JC, 11/9/04, p. 3.) Ultimately, the parties allege the Verizon model does not adhere to TELRIC because it is too linked to Verizon's embedded network, and fails the Commission's cost modeling criteria. JC, ORA, TURN, and XO describe the following significant flaws in the Verizon model:
1) VzLoop is not forward looking because it replicates Verizon's existing network of distribution areas and cable routes and does not size facilities to meet demand,
2) VzLoop contains preprocessing structural flaws that result in inefficient and overlapping distribution areas (DAs) and equipment,
3) The components of Verizon's model are not integrated, which leads to duplicative feeder and distribution networks, lack of structure sharing, and unnecessary model complexity,
4) The model contains flaws in expense modeling, and
5) Verizon's switching model is not open and reviewable and relies on extensive preprocessing, and can't modify assumptions of new vs. growth lines.
These five flaws are discussed in greater detail below.
The FCC's approach to TELRIC modeling requires the least cost configuration using the existing location of incumbent's wire centers. This is often referred to as a "scorched node" approach because existing network facilities are assumed to be non-existent except for the wire centers, or "nodes." According to JC, Verizon violates this approach by replicating its embedded network configuration and performing a loop investment analysis that makes no effort to efficiently size and deploy current technology. (JC/Donovan-Pitkin-Turner Decl., 8/6/04, paras. 100-102, 310-317, 380-386.)
First, according to JC, Verizon's model does not start with the forward-looking approach of sizing facilities to meet current and reasonably foreseeable future demand. Instead, Verizon's loop model attempts to replicate Verizon's existing network by using existing distribution areas and cable routes. These routes and distribution areas were designed decades ago, when demand, technology, and plant design were quite different from what engineers would consider if reconstructing a forward looking network today. (Id., paras. 377-379.) JC contend that a network design that mimics the current network bears no resemblance to a forward-looking network that an efficient engineer would design today, taking into account real-world constraints. (JC, 8/6/04, p. 52.)
Specifically, Verizon relies on its embedded network configuration for the sizing and placement of serving area interfaces (SAIs), digital loop carrier (DLC) equipment, and distribution areas. Then, Verizon overlays modern equipment onto this embedded framework. JC contend this produces meaningless and massively inefficient results because it is forcing modern equipment into fixed network locations and routes selected decades ago for older equipment or for other purposes. JC contend this has the effect of creating a "super-sized" network that is always adding to, but never subtracting from, the plant and equipment Verizon has in place today. Verizon maintains all existing equipment and then adds more to meet new engineering requirements, without considering the possibility of configuring existing equipment or distribution areas more efficiently. (Id.) In JC's view, Verizon's embedded approach leads it to model very small distribution areas, over half of which are less than 200 lines. As a result, JC allege that Verizon's model bulks up the network by increasing SAI investment 25% and DLC investment 600% above embedded base, dramatically overstating the number of DLC systems required in a forward-looking environment. (JC/Donovan-Pitkin-Turner, 8/6/04, para. 380.)
TURN echoes this complaint that distribution areas and clusters in the VzLoop are not forward-looking. As TURN explains, TELRIC requires cost models to be based on the most efficient technology and lowest cost network configuration, which requires a cost model to allow users the flexibility to choose from a menu of efficient equipment and to vary the line size of the cluster. TURN argues it is critical that the size of the distribution area should be an output of the model process, not an input to a TELRIC model. (TURN, 11/9/04, p. 7.) TURN's witness Loube found that "VzCost is incapable of producing TELRIC-compliant costs, because the distribution areas in the model are pre-determined by Verizon's embedded network." (Id., p. 3.) TURN asserts that because VzCost retains embedded network design instead of constructing the most efficient network possible, it does not comply with the FCC's requirement that models reflect the "long run" where all inputs are variable. (TURN, 8/6/04, p. 7.)
As an example, TURN notes VzCost associates each distribution terminal with the SAI with which it is currently connected, and therefore, VzCost designs a network that retains the basic design of the embedded local network instead of constructing the most efficient network possible. (Id., pp. 6-7.) TURN maintains the placement of SAIs drives the placement of other equipment, such as DLC systems, and thereby drives total cost in the model. The impact of Verizon's modeling approach is that VzCost places more and smaller DLCs than HM 5.3, thus generating higher costs. (Id., p. 8.)
XO provides its own analysis questioning whether VzLoop produces forward-looking results. According to XO, VzLoop produces basic loop, DS1 and DS3 loop rates that range from 100% to 145% higher than average rates in 38 to 40 areas served by major ILECs. XO contends the Verizon model is suspect because Verizon's California operations make it the tenth larges ILEC in the U.S. which should exhibit significant economies of scale rather than rates more than double national averages. (XO, 8/6/04, p. 11.) XO hypothesizes that Verizon loop rates are high because the model relies on embedded network characteristics. (Id., p. 13.) Further, XO claims that the failure of Verizon's model to aggregate demand for DLC equipment significantly overstates DLC investment. (Id., p. 21.)
Second, despite Verizon's claims it modeled masses of embedded data to ensure that its model replicates "actual" routes from its existing network, JC contend VzLoop fails to reflect actual routing in Verizon's current network. JC maintain that while Verizon has attempted to reflect its "real network," its underlying data is neither sufficiently detailed nor accurate enough to reflect even its embedded plant and it has not validated the data to ensure it reflects forward-looking design considerations. (JD/Donovan-Pitkin-Turner, 8/6/04, paras. 71-76, 89-90.) ORA questions whether Verizon's model actually captures the "real-world constraints" Verizon claims to model. While Verizon states constraints such as land usage plans, zoning maps, and landscape features should be incorporated into a cost model, it does not explicitly provide data demonstrating how particular constraints are reflected in its inputs. (ORA/Watts-Zagha, 11/9/04, p. 5.) Verizon itself admits that the systems it relied on for data do not always reflect the actual location of its equipment. (Workshop Tr., 1/13/04, at 3308-3311.) According to JC, Verizon prepared maps intending to show that HM 5.3 does not mirror Verizon's current network. These maps indicate the Verizon model "fails to reach many customers and forces modern telecommunications equipment onto embedded network layouts that it is ill-suited to serve." (JC, 11/9/04, p. 22.)
Finally, JC contend Verizon's model does not accurately calculate the economically efficient cross-over point from copper to fiber. To determine the efficient cross-over point, one must evaluate the loop distance at which the combination of high cost fiber electronics plus low cost fiber cable is less costly than the per mile cost for copper cable. Verizon admits that its attempt to include a cross-over analysis in VzLoop includes errors which are not yet corrected. (Verizon Rebuttal Panel on VzCost and VzLoop, 11/9/04, pp. 60-65.) JC contend that the cross-over analysis logic in VzLoop is structurally flawed and that turning the feature off actually lowers loop costs. (JD/Donovan-Pitkin-Turner, 8/6/04, paras. 339-342.)
In response to all of these criticisms, Verizon claims there is value in using real network locations as the starting point for its cost studies because use of existing location information reflects the context-specific judgments of Verizon's engineers who make decisions based on geographical and other constraints. (Verizon, 11/9/04, p. 62.) Verizon contends its loop routes are far superior to those generated by HM 5.3 (Verizon Rebuttal/Tardiff-Murphy-Dippon, 11/9/04, p. 45.) Verizon defends its method of locating loop facilities such as distribution terminals, DLCs and SAIs, contending that even if all JCs' comments were accepted at face value, they would have only minimal impact on costs. (Verizon, 11/9/04, p. 68.) Verizon describes how the network data it used as a starting point for its model could be reinterpreted, relying only on customer locations, existing roads, and existing wire centers. (Verizon Rebuttal on VzCost and VzLoop, 11/9/04, p. 211.) According to Verizon, when the data is reclustered into new distribution areas under this approach it results in total investment only 14% less than that proposed by Verizon. (Id., pp. 212-214.) Moreover, Verizon contends JC have erroneously interpreted Verizon's data to conclude that 29% of DAs have fewer than 50 lines. Rather, Verizon contends that a correct interpretation of its data shows only 13% of DA's have fewer than 50 lines. (Id., pp. 206-207.)
With regard to the criticism that Verizon's modeled network does not follow current network routes, Verizon admits it had to rely on "surrogate data" in limited instances where there is not sufficient reliable data to model a particular aspect of the network. For example, Verizon describes how data for six of its 275 wire centers was either missing or "not producing a logically constructed modeled network," so it modeled costs for these wire centers by reference to modeled unit costs from wire centers with similar characteristics. (Id., 11/9/04, pp. 192-193.) Verizon claims any discrepancies between SAI locations in VzLoop and the real world lead to insignificant changes in loop costs. (Verizon, 11/9/04, p. 67.)
On the third criticism that Verizon's model does not calculate an economically efficient cross-over point, Verizon admits that corrections to the economic cross-over analysis have been identified and could be reflected in any compliance filing ordered by the Commission. (Verizon Rebuttal Panel on VZCost and VzLoop, 11/9/04, p. 60.) According to Verizon, the net effect of these corrections is a 2% decrease in modeled investment because the reduction in DLC investment is largely offset by increases in copper feeder investment. (Id., p. 64.)
We find merit in the criticism of JC, TURN, ORA and XO that Verizon has not modeled a forward-looking network. Our main concern is that Verizon has attempted to replicate its existing network rather than reconfigure and re-size facilities to meet current and reasonably foreseeable future demand. Verizon's model starts with a layout to serve today's customer base, but it has layered on new equipment without considering more efficient network configurations. As a result, VzLoop assumes it would now require vastly more equipment to serve current demand, namely a 25% increase in SAI investment and 600% increase in DLC investment found by JC. While some of this increased SAI and DLC investment may be explained as forward-looking replacement of Verizon's currently deployed network, the increases are troubling when coupled with Verizon's modeling of such small DAs.
We consider it a key flaw that Verizon is using today's network layout with new equipment rather than considering more efficient configurations. We agree with JC this is backward looking and inefficient, and fails to consider the efficient alternative of reducing the amount of equipment by aggregating smaller distribution areas into larger ones. (JC/Donovan-Pitkin-Turner, 8/6/04, para. 381.) Essentially, VzCost models the SAIs that currently exist in its network rather than considering network reconfiguration with fewer and larger SAIs. Verizon's SAI placement drives the placement of DLC plant. Because VzCost places more DLCs for smaller distribution areas than HM 5.3, the smaller DLC systems generate higher costs than HM 5.3. (TURN, 8/6/04, pp. 7-8.)
Second, we are concerned that even though Verizon has attempted to model its current network routes, it has not done this entirely successfully. Although Verizon touts its use of current network routes as a superior feature of its model, we cannot actually rely on these claims because Verizon itself admits it has had to use some surrogate data and make certain adjustments. Verizon admits it could not model its actual network in all respects, but its main response is that these differences will have an insignificant impact on the modeled costs. So we are left with a model that attempts to replicate the network today, and claims this is a superior feature, but does not actually do so in all respects. We are unwilling to rely on Verizon's assertions from its rebuttal comments that these discrepancies are insignificant.
Finally, Verizon admits its model contains errors in its cross-over analysis, which is part of the forward-looking analysis that a carrier would undertake in constructing a local exchange network. It would be unreasonable to rely on the Verizon model when such a critical economic variable, namely the point at which the network should be built with either copper or fiber, is not operating correctly.
JC contend Verizon uses embedded data that is preprocessed and used as a modeling input, and that errors in this preprocessed data lead to structural flaws in the model.
First, Verizon's model indicates a very high percentage of collocated distribution terminals, which raises suspicions over whether the resulting modeled network is truly efficient or forward-looking. JC contend that in one instance, Verizon models over 100 distribution terminals at the exact same location and this indicates serious problems either with Verizon's data or the way in which Verizon processed the data. Multiple small terminals at the same location suggest Verizon has inefficiently undersized distribution terminals or that it failed to properly model the location of terminals. An appropriately designed network would model a single, larger terminal for economies of scale rather than multiple smaller ones. (JC/Donovan-Pitkin-Turner, 8/6/04, para. 78.) In response, Verizon contends many of these are separate terminals that would be required in any network to serve a multi-tenant environment such as an apartment complex or trailer park. (Verizon, 11/9/04, p. 69.)
Second, Verizon's use of embedded data leads to overlapping and inefficient distribution areas. Specifically, Verizon has modeled distribution routes where cable from one distribution area extends far into another distribution area to serve a location that cable in the second distribution area could have more efficiently served. JC maintain this results in significant amounts of overlapping cable between different distribution areas that is not efficient and not forward-looking. JC surmise this result occurs because Verizon has either made errors in the preprocessing of its location data, or it is using embedded, overlapping routes that may actually exist in its network today but that an efficient carrier would not replicate. (JC/Donovan-Pitkin-Turner, 8/6/04, paras. 81-82.)
Verizon acknowledges this overlap, but contends the impact of this overlap has not been quantified, is likely insignificant, and it cannot easily remove such small, inconsequential overlaps. (Verizon Rebuttal on VzCost and VzLoop, 11/9/04, pp. 176-177.) Verizon contends that while removal of some of these overlaps is possible, "this level of perfection is unnecessary to accurately model engineering constraints and loop costs and ... is also unachievable in any cost modeling process that relies on geocoding." (Footnote omitted.) (Id., p. 180.)
Third, JC allege Verizon's SAI locations are inefficient because they are either far outside the distribution areas they serve, or multiple SAIs appear stacked at the same location. (JC/Donovan-Pitkin-Turner, 8/6/04, paras. 89-90.) Thus, it appears that Verizon's preprocessed data is not capable of identifying actual outside plant locations as Verizon intended. JC maintain the result of inefficient and inaccurate SAI placement is an overstatement of distribution cable and a corresponding impact on feeder investment as well. Ultimately, if SAIs are not in real world locations, Verizon's criticism of HM 5.3 applies equally to its own model, which appears to not accurately model facilities where they actually exist today.
VZ responds SAI investment is a small part of total investment, and any minor discrepancies in SAI location can be addressed in a compliance filing. Verizon contends its modeling approach is sensible, particularly when compared to the HM 5.3 approach, which always places the SAI in the center of a distribution area. (VZ Rebuttal Panel on VZCost and VZLoop, 11/9/04, pp. 181-187.)
Fourth, JC contend the network Verizon models will not actually operate or work as Verizon intended. JC maintain Verizon admits various errors in its VzLoop model, such as the surrogate data and cross-over analysis errors described above, as well as errors in how VzLoop limits the maximum copper loop length. Indeed, Verizon admits VzLoop models loops with copper segments more than 18,000 feet due to an error in one of its programs. Verizon maintains it did not have time to correct this error, despite its claims to have limited copper segments to a maximum of 12,000 feet. (Id., p. 142.) According to JC, this error is significant and results in 22% of wire centers with loops exceeding 18,000 feet, which will not provide an acceptable level of basic phone service. (JC/Donovan-Pitkin-Turner, 8/6/04, paras. 318-321.)
In response, Verizon claims HM 5.3 is equally, if not more, at fault with regard to loops violating the 18,000-foot copper limit. Verizon contends that 66% of HM 5.3's wire centers have copper lengths in excess of 18,000 feet. Verizon estimates if this were fixed in HM 5.3, it would raise loop rates 6%. (Verizon/Tardiff-Murphy-Dippon, 11/9/04, p. 47.)
Again, we are concerned with the issues JC and other parties have raised with the preprocessed data and how it is used by the Verizon model. Parties raise numerous issues with multiple, overlapping facilities and distribution areas and the likelihood that this has increased local loop plant investment above and beyond a level that a forward-looking network design should incorporate. The criticisms call into question the extent to which Verizon's model accurately depicts the current local exchange network, which Verizon considers an advantage of its model. Verizon does not deny that the examples provided by its critics exist as described. Instead, its main defense centers around its assertion that these errors, once corrected if so ordered by the Commission, will lead to insignificant changes in the rates. Verizon's response is not reassuring because it primarily falls back on claims that corrections would be insignificant, errors can be fixed later, or JC has not quantified the error.
Given our experience with Verizon cost models over the past several years, including our prior requests for Verizon to fix modeling problems in the prior OANAD proceeding, we are not amenable to ordering corrections and waiting for the results. We prefer to choose between the models as filed rather than ask either model proponent to make corrections, forcing yet another round of critique of the supposed fixes. Verizon provides no support for its claim that any preprocessing errors lead to insignificant rate changes. We recognize neither model is perfect so we must asses which model best embodies forward-looking principles combined with ease of use.
Several parties maintain the various components of the Verizon cost model lack integration, which can lead to several modeling problems.
First, lack of integration in VzCost means it may not model the most efficient, forward looking network design. JC claim "Verizon's disjointed array of disconnected models for each component of its network contributes to its failure to properly account for costs of the entire network." (JC, 8/6/04, p. 35.) Specifically, JC allege Verizon's failure to integrate feeder and distribution in its model causes the model to install duplicative cable and structure. In other words, JC contend Verizon models costs as if a carrier would build feeder structure down a street and then build additional distribution structure right on top of the existing feeder structure. (JC/Donovan-Pitkin-Turner, 8/6/04, para. 270.) While Verizon's model accounts for some sharing of poles by feeder and distribution facilities, it does not in any way reflect sharing of buried or underground structure or routes. (Id., paras. 273-5.) JC contend this structural flaw cannot be fixed because it is buried in model code. Verizon admits that it uses different approaches to model feeder and distribution routes, but contends the two approaches produce duplicative feeder and distribution only in a rare number of instances. (Verizon, 11/9/04, p. 79.) Additionally, Verizon claims that in the real world, feeder often runs underground while distribution runs in aerial structure. (Id.)
Second, parties charge that lack of integration makes it difficult for a user to work with and change inputs in the Verizon model. As an example, JC contend that multiple steps in the loop costing process require extensive manual effort. JC describe 50 steps and an alleged three day process to change DLC material costs. (JC/Donovan-Pitkin-Turner, 8/6/04, paras. 16-18.) Also, JC allege Verizon does not provide even a basic flow chart or otherwise provide documentation to help an experienced professional understand the model's logic in linking input data to output. (Id., para. 19.)
TURN agrees that it is significantly more difficult to work with the Verizon model and change model inputs, than with HM 5.3. TURN notes that while Verizon and JC each criticize the other for offering a model that is difficult to work with, Verizon's complaints focus on the clustering process which is an input to HM 5.3, while the problems JC focus on in the Verizon model occur at each stage of the modeling process. (TURN, 11/9/04, p. 4.) For example, TURN explains that a relatively simple change to the cost of equity input requires 12 steps and an hour to complete, while the change can be made in one step and much less time in HM 5.3. TURN also notes that VzCost is not integrated across all modules. Depending on the module, VzCost generates either per-line results or total investment levels. TURN comments that these varying results make "apples to apples" comparisons difficult, and hamper the ability to audit the results and ensure the various network pieces add-up logically. (Id., p. 6.)
Similarly, XO maintains the Verizon model contains unneeded complexity and fails to integrate critical network assumptions. For example, XO points out that the Verizon model relies on eight proprietary software and database systems from six different vendors. (XO, 8/6/04, p. 4.) Moreover, the Verizon model contains numerous disaggregated inputs for business and residential network characteristics that create unneeded complexity and lead XO to question whether Verizon improperly modeled business and residential demand when it could have aggregated it. (Id., p. 26).
In response, Verizon defends its model as complex and highly sophisticated "in order to assimilate and process the massive amounts of information regarding Verizon CA's real-world network that must be reflected in a forward-looking cost study." (Verizon, 11/9/04, pp. 70-1.) Verizon claims that if VzLoop were simplified, it would not successfully account for real world attributes of routing and topography. Further, Verizon maintains its modeling inputs are entirely transparent and can easily be modified. In general, Verizon contends its model is no more complicated or less integrated than HM 5.3.
We are troubled by the parties' claims that lack of integration in Verizon's models cause them to install duplicative facilities and fail to capture the economies and efficiencies that a forward-looking analysis should entail. Again, Verizon says this is a minor problem which occurs only rarely. However, it is unclear how to test this assertion, particularly when lack of integration makes it difficult or impossible to test various outcomes or modify inputs. During the course of this proceeding, Commission staff met numerous times with Verizon to understand how to run VzCost. We find the amount of time invested in learning how to run the Verizon model impractical for quick UNE updates. Ultimately, when we compare the time required to master and manipulate VzCost and HM 5.3, we conclude that the Verizon model requires too many resources and is not "user-friendly." The time required to run learn the model and master sensitivity analyses to test input changes is simply not reasonable. On the other hand, Commission staff has familiarity with HM 5.3 from its use in the SBC UNE reexamination proceeding and staff is able to run multiple scenarios quickly and without excessive guidance from outside modeling consultants provided by the parties.
JC allege numerous flaws with how Verizon models the expenses that must be incorporated into forward-looking network modeling. First, JC criticize the Verizon model because of difficulty adjusting expense factors that rely on poorly documented data and untraceable calculations. (JC/Brand-Menko, 8/6/04, paras. 84-105.) As JC explain, adjusting expenses in Verizon's element specific cost studies requires manipulating multiple files and a complicated sequence of steps. As a result, it is difficult to audit the results and verify the change was made correctly. (Id., paras 89-97.) Verizon responds that these assertions should be rejected because it has fully explained its studies and supporting data. (Verizon, 11/9/04, p. 56.)
Second, JC maintain Verizon's expense modeling methodology is flawed because it starts with 2002 embedded costs and adjusts them with a "Forward-Looking Calibration" (FLC) factor. The factor has the effect of restating expenses at 2002 booked levels despite any projected decrease in forward-looking investment. JC claim that Verizon's FLC factor maintains Verizon's embedded expense levels and fails to capture network efficiencies that it has experienced or will experience. (JC/Murray, 8/6/05, p. 38.) The FLC works to ensure that embedded inefficiencies are carried forward into UNE cost results.
(JC/Brand-Menko, 8/6/04, paras. 50-63.) Therefore, the VzCost is not based on a least-cost, most efficient technology.
TURN also criticizes the FLC factor for failing to produce expenses that are forward-looking. As Loube explains, Verizon's methodology is circular because when the formulas used in Verizon's expense to investment factors are compared, it becomes clear that embedded expenses equal forward-looking expenses. (TURN/Loube, 8/6/04, pp. 28-29.) In an arbitration involving Verizon Virginia (often referred to as the Virginia Arbitration Order18), the FCC's Wireline Competition Bureau found Verizon's FLC factor circular and flawed, and it instead recommended the use of ratios similar to those used in HM 5.3. (Virginia Arbitration, paras. 139-140.)
Verizon responds its use of the FLC is appropriate because Verizon starts with book investments and forward-looking expenses in developing its cost factors. The FLC simply converts booked investment into forward-looking investment so that the cost factor expresses the correct relationship: forward-looking expense to forward-looking investment. (Verizon Recurring Cost Testimony, 11/3/03, pp. 156-161.) Verizon says it must make this conversion using the FLC because it would be inappropriate to compare forward looking expenses to booked investment. (Verizon, 11/9/04, p. 48.)
Verizon explains that without the FLC, expenses would be distorted because TELRIC investment is typically lower than book investment. For example, Verizon illustrated that if a switch with a book cost of $40,000 has forward looking maintenance costs of $1,000, the unadjusted cost factor would be .025 ($1000/$40,000). If the TELRIC price of the switch is $10,000, use of the .025 cost factor leads to the assumption that forward looking maintenance expenses for the switch are only $250. However, forward-looking maintenance costs should equal $1000. (Id., p. 49.)
Finally, JC give further examples of why Verizon's expense modeling approach merely perpetuates its embedded expense levels and fails to calculate forward-looking expense estimates. Specifically, Verizon ignores cost savings from employee reductions, projected productivity gains, and forward-looking building space requirements, and instead, uses 2002 actual expense levels. (JC/Brand-Menko, 8/6/04, pp. 99-111.) Furthermore, JC maintain that Verizon's expense methodology fails to remove numerous non-UNE related expenses such as those for non-regulated services, pension costs, retail marketing costs, costs related to affiliate transactions, and DSL and broadband-specific investments. (Id., paras. 19-21 and 212-228.) While Verizon claims it has made forward-looking expense adjustments, JC take issue with this claim. They note that most of the reductions Verizon claims to have made merely remove costs that should never have been included in a recurring UNE study in the first place such as non-recurring costs, and expenses associated with obsolete plant. (Id., para. 21.)
Verizon contends its cost studies do not reflect embedded expenses because although it started with 2002 booked expenses as a starting point, Verizon made significant adjustments to those expenses to make them appropriately forward-looking, such as reductions in copper maintenance costs, merger costs, and retail-related costs. (Verizon, 11/9/04, p. 36.) These adjustments reduced 2002 booked expenses by over 22%. Verizon then made further adjustments to reduce expenses another 6% to account for lower expenses associated with forward-looking technology. (Id., p. 37.) The FCC itself stated in its rulemaking on TELRIC that the best method of projecting expenses in a UNE cost proceeding is to make forward-looking adjustments to actual expenses. (Id,. p. 37.)
We find criticisms in this area valid. First, we agree that the expense portion of Verizon's model is difficult to audit and verify because it relies on numerous factors that are difficult to trace and changes to the factors require multiple steps. Second, the key problem with the entire FLC debate is that Verizon states it has determined the level of forward-looking expenses, and starts with those before adjusting investments to match expenses. The problem we find is that Verizon does not explain how it deduced the level of forward-looking expenses. In fact, the usual purpose of E:I ratios is to use the relationship of current expenses and investments to deduce forward-looking expenses. Verizon starts by apparently knowing the answer and creates a ratio that will provide a plug number to always produce the level of forward looking expenses that it has determined are proper. This is indeed circular, as the FCC found in the FCC's Virginia Arbitration.
The HM 5.3 model, in contrast, looks at the relationship of current expenses and investments, and applies that ratio to forward looking investments under the assumption that there is a correlation between expenses and investments, and as investments change, expenses will change as well. Verizon's FLC factor has the opposite result of assuming that expenses will remain at current levels even as investments change. In support of its method, Verizon argues that "it costs the same amount, for example, to dry clean a cheap suit as it does to dry clean an expensive one." (Verizon/Jones Decl., 11/9/04, p. 53.) This example, however, assumes that suits and highly sophisticated switching and loop equipment are analogous. There may be plenty of reasons why it can be more expensive to dry clean silk than cotton, and it can be more expensive to clean a 50-year old wool or silk suit that needs special handling than a brand new, machine-washable microfiber suit that never needs dry cleaning at all. Verizon's analogy is too simplistic and fails to capture the reality that as telecommunications technology advances, operational and maintenance expense savings may be realized.
Third, we agree there is uncertainty whether Verizon has properly made forward-looking adjustments to its expenses. Our own review suggests that some of these categories still require adjustment, particularly productivity, building space requirements, and non-UNE related expenses for DSL and broadband. The complexity of the ACFs used in the Verizon model do not lend themselves to modifying these expense levels easily.
Verizon uses two complex and data intensive models to determine switching costs. The SCIS model determines investments for Lucent 5ESS and Nortel DMS switches, while Verizon's own Costmod analyzes investment for GTD5 switches. According to JC, both SCIS and Costmod are fatally flawed in cost model design and methodology, and the model inputs cause severe switch cost overstatements. The complexity and rigidity of these proprietary cost models make it impossible to correct all aspects of the cost study. The primary flaws in the Verizon switching models are that they are difficult to modify and change assumptions regarding pricing inputs and equipment, and that they rely on a hypothetical switch configuration from Verizon's national data rather than California specific inputs. These are discussed in greater detail below.
First, the model methodology and structure is not open, reviewable, and able to support sensitivity analyses. According to JC and ORA, Verizon uses an array of several massive investment cost models to develop switching costs and changing a fundamental input requires a monumental effort. JC contend that as a result, it is difficult to modify assumptions regarding the percentage of new and growth line purchases and the technology mix and price of switches. Specifically, critical assumptions in SCIS are developed in a pre-processing program called the "partitioning process," which contains the basic intelligence of the model. JC complain this partitioning process is proprietary, and its results are hard-coded data items in the model's investment tables. Thus, the true intelligence of the model was not available for review. (JC/Pitts, 8/6/04, para 30, ORA/Lehman, 11/9/04, pp. 3-4.)
As an example, Verizon's proposed UNE switching prices are dominated by GTD-5 switch costs. Specifically, Verizon's switching cost studies assumes a mix of 63% GTD-5 switches, 22% Lucent supplied switches, and 15% Nortel switches. (Verizon Switching Rebuttal, 11/9/04, p. 23.) In contrast, Pitts uses only a mix of Lucent and Nortel switches to match the switch purchase data provided by Verizon. Verizon defends its GTD-5 assumptions because it expects to use the GTD-5 technology for the foreseeable future and it deploys different switch types for functional and strategic considerations. (Verizon Recurring Cost Testimony Verizon 11/3/03, p. 79; Verizon 11/9, p. 84.) Pitts claims the critical question is whether Verizon would purchase GTD-5 switches today if it were purchasing new switches. According to Pitts, the GTD-5 switch is not forward looking because evidence shows Verizon purchased only one since 1990, while it has purchased 42 Lucent and Nortel switches since 1991. (JC/Pitts, 8/6/04, p. 25.) TURN and ORA echo support for removal of GTD-5 switches from the cost studies. (TURN/Kennedy, 8/6/04; ORA/Lehman, 11/9/04.)
Moreover, JC contend the Verizon switching models use no information concerning Verizon's actual cost for new switches or growth line switching costs. (JC/Pitts, 8/6/04, para 20.) Instead, Verizon uses a complex calculation to back into the "discount" level it uses to convert list prices into what it actually pays its vendors. (Id.) Moreover, little of the data Verizon uses relates to new switch purchases. Instead, 98% of the investment relied on to calculate switch costs relates to Verizon's purchases to add growth lines to its embedded switching base. (Id., paras. 57-58.)
JC's witness Pitts attempted to modify the critical input of switch price but found prices are entered only as a discount off list price. Verizon admits that in order to change switch pricing inputs, one must use a "work around" to accommodate a change in switch price. Pitts contends that without the work around, changing investment discounts would require 13,000 manual data entries. (Id., pp. 9-10.) The work around adjusts output results rather than modifying the switch price discount inputs. Pitts considers this an inferior approach and less reliable than changing pricing inputs directly.
Furthermore, JC and ORA contend it is difficult to modify Verizon's switching models to reflect forward-looking technology. Specifically, the Verizon models are based on an outdated version of SCIS that is not forward looking because it relies on switch components that are no longer offered and excludes newer, more efficient switch components. (Id., pp. 2 and 18; ORA/Lehman, 11/9/04, p. 4.) Pitts maintains that amending SCIS to incorporate forward-looking switch components would be extremely burdensome.
Verizon responds that JC's criticisms are hard to understand since their primary witness, Ms. Pitts, was a primary designer of Verizon's switching cost model, SCIS, and she was able to manipulate SCIS to restate its outputs. (Verizon, 11/9/04, p. 80.) Verizon claims it provided the source code for SCIS to the parties and they have failed to use it to perform any analysis. Verizon defends its switching model as detailed and technical for good reason because its network is very complex and requires significant engineering design. (Id., pp. 81-82.) Verizon also discounts the criticism that an older version of SCIS was used, stating that each new version makes very little changes to the basic model structure and any outcome differences are inconsequential.
(Id., pp. 82-83.) As further support for this claim, Verizon performed a comprehensive analysis of switch components and their prices from 2000 to 2003 and claims that changes in components and prices since 2000 do not have a major effect on Verizon's switching cost analysis. (Verizon Switching Rebuttal, 11/9/04, pp. 10-11.) Verizon contends its calculations of switch discounts are based solely on recent purchase data. (Id., p. 86.)
Second, JC and ORA maintain the Verizon switching models are flawed because they are based on a hypothetical switch configuration from Verizon's nationwide network. Verizon did not use the models as they were designed to be used, but instead entered input data into switch models for a small sample set of fictional "representative" switch configurations, then adjusted generic cost outputs using outboard calculations in an attempt to estimate California specific investments (JC/Pitts, 8/6/04, p. 6; ORA/Lehman, 11/9/04, p. 3.) According to JC's witness Pitts, the SCIS and Costmod models are designed to accept user input data for each switch in the network and produce detailed results about each switch. Verizon, however, did not enter data about actual individual California switches, but developed theoretical switches to represent all switches in the entire Verizon footprint and ran these theoretical switches through its investment models. Verizon then uses a true-up process to attempt to modify the SCIS output to reflect California switches. Pitts maintains this approach produces unreliable results because the unit costs calculated by SCIS and Costmod do not accurately reflect each California switch and are, therefore, incorrect. The true up cannot produce accurate results because it is using incorrect unit costs. (JC/Pitts, 8/6/04, pp. 18-21.)
Verizon responds that its representative office approach closely matches Verizon's California switches, is more manageable and less time consuming than modeling every office, and does not significantly impact results. (Verizon Switching Rebuttal, 11/9/04, p. 15.) Further, Verizon counters that only a finite set of switching model inputs have any significant impact on investments. Verizon ran hundreds of sensitivity runs of its switching models to verify which inputs are significant. These are the inputs that Verizon varied in its model office runs to capture the investment differences between different switches. (Id., p. 14.) Finally, changes to the Verizon switching models are not unduly burdensome because the process has been simplified by using the representative office approach and changes can easily be input to the key cost drivers to see immediate results. (Id., p. 16.)
We find that Verizon's switching model is indeed highly complex and difficult to manipulate. Our main concern is that due to the model complexity and hard coding of input data, it is difficult to run sensitivity analyses with varying assumptions for the percentage of new and growth lines and switch discounts. We agree with JC that using a work around approach is less reliable than actually changing model inputs. If Verizon has introduced a truly superior, albeit complex, model, then the Commission should be able to run it and test changes in inputs rather than performing outboard calculations to mimic input changes. Furthermore, we agree with JC that the GTD-5 is not a forward-looking switch technology because Verizon has purchased only one since 1990. We find it unduly burdensome to change the technology assumed in the switching model and remove the GTD-5 switch.
Verizon claims that using an updated version of SCIS or using actual switch data rather than its hypothetical approach would lead to an inconsequential difference in results. This has been a popular response from Verizon and one that we are asked to take at face value without the ability to actually test the assertion. We are reluctant to accept Verizon's claims without testing them and we are reluctant to rely on a model that requires outboard calculations to run scenarios rather than actual model runs.
In summary, we find five major flaws with Verizon's model. First, Verizon models a network that is not forward-looking because it attempts to replicate its embedded network configuration, albeit not always successfully. In so doing, Verizon fails to efficiently size and deploy current technology. By overlaying modern equipment on an embedded network, Verizon models numerous small distribution areas that produce inefficient results, with SAI and DLC investment far exceeding current levels. In addition, Verizon's loop model does not effectively calculate the economically efficient cross-over point from fiber to copper facilities.
Second, Verizon's model appears to contain errors in preprocessed data due to a high percentage of collocated distribution terminals, overlapping distribution areas, and inefficient SAI locations.
Third, lack of integration in the various modules in Verizon's cost model increases the likelihood duplicative facilities are modeled, and makes the model cumbersome if not impossible to test input sensitivity.
Fourth, Verizon's expense modeling is difficult to adjust and based on a flawed FLC factor.
Finally, the structure of Verizon's switching model makes it difficult to modify and test varying inputs, particularly for purchases of new and growth lines and equipment technology.
B. Flaws in HM 5.3
Verizon's criticisms of HM 5.3 are very similar to SBC's criticisms of HM in the SBC UNE Reexamination, completed in 2004. Verizon's essential criticisms of HM 5.3 are:
1) It contains errors in the customer location process leading to hypothetical clusters that do not resemble real world distribution areas;
2) It ignores engineering standards and network design principles by sizing for current demand and not allowing enough spare capacity;
3) It assumes unrealistic efficiencies from ubiquitous network replacement, unreasonable labor and productivity rates, and understated expenses;
4) It relies on unreasonable expert opinions; and
5) It results in unrealistic investment levels compared to Verizon's current network.
In the SBC UNE case, the Commission examined virtually the same arguments by SBC along with SBC's claim that as a result of these flaws, HM 5.3 produced a network that is unrealistic because it has far less outside plant than SBC's actual network today. (See D.94-09-063, Section V.B.) In particular, SBC alleged HM 5.3 modeled a network with fewer distribution areas, less distribution pairs, less fiber equipment, less trunks, and less interoffice network equipment than SBC's current network. The Commission reviewed SBC's criticisms in depth in D.04-09-063 and found merit to some of SBC's criticisms, but not all of them. The Commission concluded that many of SBC's criticisms could be addressed with input modifications to HM 5.3, particularly in areas relating to engineering and design standards, spare capacity, and expense levels. The Commission did not agree with all of the assumptions built into the HM 5.3 customer location process and expressed concern that it was not possible to modify this area and test various scenarios. Nevertheless, the Commission found that HM 5.3 complied with TELRIC and could be relied on to set SBC's UNE rates. In addition, the Commission agreed that many of the "expert judgments" underlying HM 5.3 were questionable and appeared biased to produce low results. However, the Commission found it could replace many of these expert judgments with assumptions and inputs from SBC's own model, although it was not possible to fully replace assumptions regarding labor costs. Finally, the Commission found that criticisms of the HM 5.3 transport model were valid. (Id., p. 108.)
In this case, the version of HM 5.3 that we are asked to examine is virtually identical to the version in the SBC case, although with a different clustering database based on Verizon specific data. Given the similarity in arguments, our findings are similar to those in the SBC case and we will review them here only briefly in the sections below.
We will not rely on the rebuttal version of HM 5.3 filed by JC on November 9, 2004 to set UNE rates for Verizon. In its rebuttal, JC provided numerous corrections and changes to HM 5.3. Verizon objected that the scope and magnitude of the HM 5.3 model changes were sweeping and should be stricken. (Verizon Motion to File Surrebuttal Testimony, 12/3/04, p. 2.) JC defended their rebuttal filing of HM 5.3 as appropriate, claiming all changes were made in response to criticisms by Verizon. The ALJ directed JCs to file a summary table describing all modeling changes in the rebuttal version of HM5.3 including citations to the original criticism and to a description of the modeling change in JC's rebuttal filings.19
Verizon responded to this summary table on March 15, 2005, stating the vast majority of modeling changes listed in the summary table were not described in JC's rebuttal filing, not responsive to a specific criticism levied by Verizon, or not implemented in such a way as to be responsive. In addition, Verizon claims JC failed to disclose a number of significant changes to the rebuttal version of HM 5.3 with regard to model platform, inputs, and preprocessing. Verizon requests the Commission strike all the modeling changes in the rebuttal version of HM 5.3.
After review of the summary table and the response, the ALJ determined that the rebuttal version contained some changes that were not responsive to Verizon's criticisms or were not reasonably explained in the JC rebuttal filings. For example, JC modified code in HM 5.3 relating to ADSL lines, admitting it was not responsive to any criticism by Verizon. (Summary Table, 1/21/05, p. 3) Further, the rebuttal version contains new hard-coded information on switching, updated cost factors, and an increase in the number of customer locations without an adequate description. (Verizon, 3/15/05, pp. 19, 21, and 24.) While some changes in the rebuttal version were responsive and reasonably explained, other changes would require significant resources to examine adequately. Therefore, it is not appropriate to rely on the entire package of changes which comprise the rebuttal version. Instead, we will use the original filing of HM 5.3, as amended on February 6, 2004, for our model runs. To a limited extent, we may use minor corrections to HM 5.3 suggested by JC in their rebuttal filing, without relying on the entirety of the HM 5.3 rebuttal version. If any rebuttal corrections are used, this is noted in the text of this order.
Similar to SBC's criticisms of HM 5.3 in the SBC UNE proceeding, Verizon criticizes HM 5.3 for relying on a cluster input database developed by a third-party vendor, TNS. Verizon contends the hypothetical customer clusters developed by TNS bear no resemblance to any reasonable depiction of real world distribution areas. Further, because these clusters are a proprietary input, they cannot be modified.
Verizon provides several examples of specific problems it sees with the HM 5.3 clusters. First, Verizon alleges the clusters are not realistic because they assume customers are uniformly spread in rectangular shaped clusters, with parcels of equal size and shape. (Verizon/Tardiff, 8/6/04, pp. 26-27.) Second, HM 5.3 uses a simplistic design of distribution cables in a grill pattern throughout these rectangular serving areas, which undersizes cables. (Id., p. 31.) Further, Verizon maintains the HM 5.3 clustering data ignores existing rights of way and physical obstructions (e.g., freeways and bodies of water). (Verizon/Dippon, 8/6/04, pp. 4, 30-36.) In sum, Verizon contends that despite the use of data precisely locating existing customers with a longitude and latitude, HM 5.3 underestimates costs and fails to model plant to any existing customer locations. (Id., pp. 27-28.)
JC respond that HM 5.3 does not ignore real world obstacles, but instead uses "right angle routing" to ensure that more than enough loop plant is designed to accommodate existing rights of way, easements, and obstacles. As a result, HM 5.3 produces an average loop length that exceeds loop lengths modeled by Verizon. (JC, 11/9/04, p. 28.) In addition, JC maintain FCC regulations require that the TELRIC of an element "should be measured based on the use of the most efficient telecommunications technology currently available and the lowest cost network configuration, given the existing location of the incumbent LEC's wire centers." (47 C.F.R. Sec. 51.505(b)(1).) The FCC has consistently held that "existing incumbent LEC plant is not likely to reflect forward-looking technology or design choices." (JC, 11/9/04, pp. 24-5, citing FCC's 5th Report and Order, para 66.) JC admit HM 5.3 does not model Verizon's embedded network, but an exact replica of the current network is not required by TELRIC. As final support, JC cite the Supreme Court's admonishment in its review of TELRIC that "regulation does not and should not guarantee full recovery of embedded costs." (Id., p. 29, citing Verizon v. FCC at 1681.)
Our findings with regard to a comparison of HM 5.3 and VzLoop mirror those from the SBC UNE case where we considered these same criticisms. While we do not agree with all aspects of HM 5.3's customer location and loop modeling, it is no more a "black box" than Verizon's own preprocessed network information and input modeling assumptions related to the VzLoop. Both HM 5.3 and VzLoop lack transparency, limit the Commission's ability to test various scenarios, and can be faulted for the accuracy of their customer location process. HM 5.3 is based on a detailed examination of current customer locations, and makes simplifying assumptions not unlike the assumptions underlying VzLoop. The HM 5.3 model ultimately ignores customer locations when modeling loop plant. As a result, although HM 5.3 starts with current customer location data, it does not model all loops in the exact locations where they exist today. Nevertheless, HM 5.3 has one advantage over VzLoop because it starts with actual customer locations to cluster customers into efficient groupings, whereas VzLoop makes no attempt to determine efficient customer groupings based on current population density characteristics.
We find that the method used by HM 5.3 to model customer locations, create forward-looking customer clusters, and estimate the costs of reconstructing Verizon's loop network falls reasonably within TELRIC guidelines, even if the reconstructed network does not follow Verizon's actual outside plant routes. We find the approach used in HM 5.3 more forward-looking and in compliance with the Commission's CCPs and TELRIC than Verizon's approach. Verizon assumes all existing distribution areas are unchanged and attempts to replicate its existing network, although not entirely successfully. As noted in Section V.A.1 above, we concluded Verizon's approach overstates DLC and SAI equipment investment and does not mirror existing customer locations in all instances.
This does not mean there are not other valid criticisms of the clustering process underlying HM 5.3. Significantly, we were unable to run our own analyses to test HM 5.3's sensitivity with different clustering inputs. We would have preferred to test the results of different cluster sizes. At the same time, our inability to run sensitivity analyses of cluster sizes is not unlike our inability to run sensitivity of VzLoop's preprocessed cluster assumptions. In other words, both models involved extensive preprocessing of data that, for various reasons, was difficult to modify. Thus, we find that both models contain aspects of their loop modeling that we were unable to modify to our satisfaction.
Verizon maintains that HM 5.3 ignores widely accepted engineering standards and deviates from well established network design principles. Although Verizon provides numerous examples, its criticisms boil down to three major claims.
First, Verizon maintains that HM 5.3 focuses solely on current demand and ignores requirements for additional spare capacity. According to Verizon's witness Tardiff, HM 5.3 excludes costs by modeling insufficient capacity for fluctuations in demand over time. (Verizon/Tardiff, 8/6/04, pp. 13-15.) JC respond that HM 5.3 does not ignore growth, customer churn or fluctuations in demand. Rather, JC contend that HM 5.3 uses real-world engineering guidelines to install sufficient capacity for growth and changing conditions and uses cable sizing factors based on a reasonable projection of actual total usage, consistent with FCC requirements.
Verizon's criticism mirrors SBC's criticism of HM 5.3 in the SBC UNE proceeding. In that case, we agreed with the proponents of HM 5.3 that the model should not build to "ultimate demand," as suggested by SBC. We discussed how the FCC's TELRIC guidelines clarify that model inputs should reflect current demand, defined as a "reasonable amount of excess capacity to accommodate short term growth." (D.04-09-063, mimeo. at 74.) Here, as in the SBC case, we find that we can ensure HM 5.3 provides reasonable spare capacity for growth by reviewing the fill factors used in the model, which are discussed further in Section VI.D below.
Second, Verizon claims the distribution clusters in HM 5.3 are too large, and would be inefficient and unmanageable in the real world. The use of these overly large clusters produces fictitious economies of scale through the use of unrealistically large equipment and the modeling of cables that are larger than those typically deployed. (Verizon/Tardiff , 8/6/04, pp. 32-33.) JC respond Verizon is relying on outdated guidelines, and Verizon's newer guidelines call for larger distribution areas. (JC, 11/9/04, p. 31.) JC's witness Donovan defends HM 5.3 for accumulating demand in clusters using a bottom-up approach, and criticizes Verizon for confusing detailed engineering designs and their myriad requirements, with TELRIC cost modeling. Donovan maintains that a TELRIC model attempts to provide a reasonably conservative estimate of costs but does not attempt to precisely locate each piece of outside plant equipment or right of way. (JC/Donovan, 11/9/04, pp. 3-5.) Donovan further claims that Verizon's own data indicate a large discrepancy between Verizon's alleged engineering practices (i.e., distribution areas between 200 and 600 households) and its cost study, (i.e., thousands of clusters below 200 lines and larger clusters in urban areas). (Id., p. 17-22.)
This disagreement is again reminiscent of the SBC case. There, we expressed concern with the larger distribution areas modeled by HM 5.3, and we suggested a preference to run our own scenario with smaller clusters, although not as small as those modeled by SBC. We found SBC's cluster sizes based on a limit of 200 to 600 households too restrictive because a forward-looking network configuration would most likely recognize today's dense customer groupings and the availability of larger equipment sizes. Thus, a forward-looking network would likely contain distribution areas larger than SBC's historic configuration. We concluded we would have preferred a middle ground between the two extremes, although we were not able to modify the clustering process used as an input to HM 5.3 to accomplish this. (D.04-09-063, mimeo. at 90-92.)
Our conclusion here is exactly the same. While we are unable to modify the somewhat large distribution cluster assumptions used as an input to HM 5.3, we note Verizon's model has numerous unrealistically small clusters we cannot modify. While we would have preferred to run HM 5.3 with different, slightly smaller cluster assumptions, neither can we rely on Verizon's approach rooted in distribution area limits that, according to witness Donovan, Verizon does not follow itself. We prefer a model with larger clusters based on a forward-looking reconfiguration to take advantage of technology advances, rather than an approach that appears to rebuild today's existing network configuration with fixed distribution areas incapable of maximizing the efficiencies of forward-looking equipment technologies.
Verizon's third engineering criticism involves numerous alleged flaws in HM 5.3, including the assumption that a stand-alone loop can be unbundled using IDLC technology and insufficient pole investment. (Verizon/Tardiff, 8/6/04, p. 56.) Regarding the use of IDLC technology, we are somewhat perplexed by Verizon's criticism since it runs its own model with a 90% IDLC technology, discussed in more detail in Section VI.C below. Regarding pole investment, a similar criticism was levied in the SBC UNE case, and it is not difficult to change input assumptions in HM 5.3 to increase pole investment accordingly. Overall, we find the problems identified in HM 5.3 can be remedied more readily than the problems we find in the Verizon model. Specifically, it is easier for the Commission to modify engineering assumptions such as pole spacing, fill factors, and structure sharing in HM 5.3 than in the Verizon model.
Verizon criticizes HM 5.3 for its aggressive interpretation of TELRIC that assumes ubiquitous replacement of the network, with the assumption of efficiencies unattainable in the real world by a real-world carrier.
According to Verizon, HM 5.3 contains numerous flawed assumptions, particularly for inputs related to structure sharing, labor costs, and switching inputs, which lead to unrealistically low UNE rates.
Specifically, Verizon claims structure sharing inputs in HM 5.3 ignore Verizon's actual experience and assume all networks, including those of other utility and cable providers, are rebuilt simultaneously so that each entity shares structure costs. (Verizon, 8/6/04, p. 63.) In addition, HM 5.3 contains understated labor costs and labor productivity assumptions that do not remotely resemble values actual carriers can attain. (Id., pp. 69-72.) With regard to switching, HM 5.3 estimates switching costs based on the unrealistic assumption that over 90% of equipment is purchased at the new switch discount price. (Verizon/Tardiff, 8/6/04, pp. 16-17.)
JC defend the level of sharing assumed in HM 5.3 as based on rational economic behavior. (JC, 11/9 pp. 57-8.) Regarding labor costs, JC counter that vast portions of Verizon's own placement and productivity assumptions are not supported by anything other than subject matter expert input, and Verizon does not supply data from its own operations to support its complaints that HM 5.3 values are too low. (Id., p. 60.) JC asserts Verizon relies on unrealistic timelines for installation that include "wait time" and other bureaucratic inefficiencies rather than a pure estimate of efficient installation time. (JC/Donovan, 11/9/04, p. 74.)
Regarding switching inputs, JC contend low prices for new switch purchases are not isolated, but have been available from 1995 through at least 2002. (JC, 11/9/04, p. 61.) Further, since Verizon switch discounts are based on volume, one would expect even lower prices if larger volumes of new switches were purchased. Moreover, JC allege Verizon's plan to replace digital switches with all new packet switches contradicts Verizon's assertions it anticipates paying only "growth prices" for switches in the future. (Id., p. 63.)20
Similar to our findings in the SBC UNE case, Verizon's criticisms of HM 5.3 principally highlight questionable modeling inputs, but do not indicate HM 5.3 violates TELRIC requirements overall. Verizon suggests HM 5.3 inappropriately assumes a network is built overnight. While we agree this may be an unrealistic assumption, every cost model must make simplifying assumptions. We find HM 5.3 falls reasonably within TELRIC guidelines by modeling the location of existing wire centers coupled with forward-looking technologies and network configuration. We find that Verizon's criticisms regarding specific inputs can be remedied with input changes and do not prevent us from using HM5.3 to set UNE prices. Specifically, we can modify labor costs, structure sharing, switching inputs, and other assumptions to address Verizon's concerns.
Verizon contends the Commission should reject HM 5.3 because of its extensive reliance on unsubstantiated opinions. Verizon contends that HM 5.3 relies too heavily on dubiously supported "expert judgments" for a large share of modeling inputs, which renders the model's UNE cost estimates suspect. According to Verizon, the opinions are offered with little analysis or back up documentation, and often ignore empirical data that is readily available. (Verizon/Murphy, 8/6/04, pp. 118-20.) For example, Verizon contends JC's inputs for labor costs have declined without explanation, despite Verizon data that labor costs have steadily increased. (Id., pp. 123-126.) According to Verizon, it supplied detailed cost data to JC for numerous inputs such as engineering, cable placement and support structures which JC ignored in favor of other sources. (Id., pp. 132-33.) Verizon runs HM 5.3 with new input assumptions, which results in a loop cost of $32.10. (Verizon/Tardiff-Murphy-Dippon, 11/9/04, p. 26, and Attachment TMD-9.)
JC respond that Verizon inappropriately equates its actual cost data with JC's estimates of forward-looking costs based, in many instances, on data supplied by Verizon. (JC, 11/9/04, p. 42.) Moreover, Verizon itself uses numerous expert opinions and unnamed sources for its own inputs and assumptions. (Id.)
Once again, SBC levied the same criticism against HM 5.3 in the SBC UNE proceeding. There, we found that the use of expert judgments in HM 5.3 was usually adjustable. (D.04-09-063, mimeo. at 94-95.) We make the same finding here. We do not consider the reliance on expert opinions a fatal flaw in HM 5.3 because we can modify the inputs, as Verizon itself did when it re-ran HM 5.3 for its rebuttal filing. In fact, as we discuss in Section VI below, we accept many of the suggested input changes that Verizon offers.
Verizon contends the investment and expense levels in HM 5.3 defy common sense and sound economic reasoning. Even with technological advancements and efficiencies, Verizon claims there is no basis to conclude it Verizon could rebuild its entire network for only a small fraction of its current cost. Specifically, Verizon finds it implausible that HM 5.3 results suggest deployment of a brand new network at investment and expense levels far below Verizon's current experience. Specifically, Verizon faults HM 5.3 for proposing:
· Investment levels only 25% of reproduction cost
· Expense levels only 30% of Verizon's current levels
· Total investment of $2.9 billion
· Total labor force only one quarter of the current labor force.21
Verizon maintains these results defy common sense, particularly given that Verizon has been subject to price-cap regulation for over 13 years.
JC respond that HM 5.3 cost estimates have been validated against real world data and the experience of other industries that have transitioned to a competitive environment. (JC/Klick, 11/9/04, paras. 20-33.) For example, JC's witnesses contend HM 5.3 incorporates real-world information by ensuring sufficient distribution cable is modeled to connect all customer locations in a cluster to each other and to the SAI serving that cluster. (JC/Mercer-Pitkin-Turner, 11/9/04, pp. 4-5.) Further, simply because Verizon has been operating under price caps does not mean its embedded costs reflect the forward-looking efficiencies required by TELRIC. (JC, 11/9/04, p. 40.) JC contend price cap regulation is not as effective as competition in forcing incumbents to become efficient. (JC/Klick, 11/9/04, p. 12.) In sum, the "real world" test incumbents advocate is neither meaningful nor consistent with TELRIC.
Similar to our findings in the SBC UNE case, we find Verizon's comparisons of HM 5.3 modeling results to Verizon's current expense and investment levels are not meaningful. In our SBC UNE decision, we found SBC's arguments that HM 5.3 results were unrealistic echoed the ILECs unsuccessful arguments to the U.S. Supreme Court, where ILECs attempted to invalidate TELRIC. (D.04-09-063, mimeo. at 76-77.) Where Verizon has provided useable information regarding its actual network experience, we have attempted to incorporate that information into our HM 5.3 model runs. For example, in Section VI below, we modify labor inputs, switching, and asset lives based on Verizon specific information.
In summary, we find the method used by HM 5.3 to model customer locations, create customer clusters, and estimate the costs of reconstructing Verizon's loop network falls reasonably within TELRIC guidelines. While we were unable to modify the distribution clusters used as an input to HM 5.3, we found that compared to the smaller cluster assumptions in the Verizon loop model, the cluster sizes in HM5.3 were based on a forward-looking network configuration and maximized the efficiencies of new equipment technologies. Moreover, we find that HM 5.3 can be modified to overcome many of its alleged flaws. Specifically, the model can be modified to use different input and engineering design assumptions, spare capacity can be increased, labor rates can be increased, and expense assumptions can be modified to increase expenses such as cost of capital and depreciation. We do not agree that HM 5.3 contains unrealistic investment levels, particularly after key inputs are modified.
C. Adherence to Commission Modeling Criteria
Both parties made admirable efforts to comply with the Commission's cost modeling criteria, namely that the models allow the user to reasonably understand how costs are derived, replicate model results, and modify inputs and assumptions. Nevertheless, both models contain areas of concern.
With regard to the Verizon model, several parties commented that the model should be rejected because it is internet-based and not actually filed with the Commission. JC claim the Verizon model is not officially on the record because it resides on the internet, which means it is owned, controlled and maintained in Verizon's custody with access strictly limited. JC contend this could present a problem in the event of subsequent court review. (JC, 8/6/04, pp. 16-17.) They further claim that parties and staff cannot review or modify source code or run sensitivity tests of coding changes, and the model is hard to use and adjust because it lacks integration and extensive manual effort is required for input changes. (Id., pp. 19-21.)
First, we are somewhat troubled by the fact that Verizon's model is internet based. Our main concern mirrors the one raised by JC, that as Verizon issues updates to its model, old versions that reside only on the internet, and are not documented in the Commission's record, may not always be available for subsequent UNE pricing proceedings or court challenges to this decision. Second, we are not able to modify preprocessed network information used as an input to the VzLoop module. Third, it is difficult and time-intensive to run sensitivity analyses of the Verizon model.
As for HM 5.3, we have a similar concern that it does not allow modification of preprocessed customer location information leading to the clustering of customers into distribution areas. In addition, we are not able to reasonably understand the components of interoffice transport modeling in order to test criticisms of its demand assumptions and ring architecture.
Since both models exhibited areas of concern with regard to the modeling criteria, we did not eliminate either model based on these criteria.
D. The Commission Should Rely on HM 5.3 Because It is Less Flawed than VZCost
The analysis above describes why we have concluded that both HM 5.3 and the Verizon model contain flaws that we cannot correct completely. Verizon models a network that is not forward-looking because it replicates the embedded network configuration and fails to efficiently size and deploy current technology. Verizon's loop configuration assumes fixed distribution areas, based on its current network, which are not capable of maximizing the efficiencies of forward-looking equipment. Other troubling aspects of the Verizon model include numerous small distribution areas, excessive SAI and DLC investment compared to current levels, overlapping distribution areas, and a lack of integration between the various modules of the Verizon model that leads to potentially duplicative cable and structure and makes it difficult to test input sensitivity. Further, Verizon's model contains many inputs and assumptions that we conclude are not forward-looking - such as the FLC factor, switching equipment, structure sharing, cost of capital, and overhead markup.
Overall, we found the lack of integration in the Verizon model made it difficult to change numerous input assumptions quickly and efficiently. Relatively simple changes to one input, such as cost of capital, require numerous manual inputs to successfully implement in the various Verizon modules. When numerous inputs are modified for a sensitivity analysis, the amount of time needed to manually change the selected inputs in the various modules and ensure the changes are correctly integrated across all modules increases significantly. The lack of flow through from one module of the Verizon model to another makes it extremely challenging to successfully manipulate and excessively prone to errors when modifying inputs. For this reason, we find the Verizon model is less user-friendly than HM 5.3 and not given to easy updates in the event future UNE price adjustments are needed. It is unduly burdensome and therefore not reasonable to use the Verizon model, which requires extensive and time-consuming manual manipulation and is prone to human error in the input adjustment process.
We did not experience the same degree of difficulty in modifying and correcting our runs of HM 5.3. In general, we were able to understand how to make the necessary modifications, implement them quickly, and after making them, we could easily and consistently replicate our results in a reasonable time frame and with a high degree of certainty. Even though we disagree with many of the input assumptions used in HM5.3 - such as the cost of capital, asset lives, structure sharing, DLC costs, labor and productivity, and switching
assumptions - we can change these inputs and assumptions. In many areas, we have incorporated inputs from the Verizon model into HM 5.3, particularly in areas such as labor rates, asset lives, and certain switching assumptions. Despite these efforts, we could not cure all of the flaws we found in HM 5.3. We cannot perform sensitivity analyses on the clustering process that builds the initial estimates of outside plant, and we do not have complete confidence in HM 5.3's interoffice transport and high capacity loop modeling.
Therefore, we will adopt HM 5.3 model results for Verizon's permanent UNE rates. We conclude this approach is reasonable given the enormous complexity involved in TELRIC modeling exercises. It is reasonable to use a model with some flaws when the alternative is another model with more significant flaws that is also difficult to operate and modify.
18 In the Matter of the Petition of WorldCom, Inc. Pursuant to Section 252(e)(5) of the Communications Act for Preemption of the Jurisdiction of the Virginia State Corporation Commission Regarding Interconnection Disputes with Verizon Virginia, Inc., and for Expedited Arbitration, CC Docket No. 00-218, Memorandum Opinion and Order, DA 03-2738, (rel. Aug. 29, 2003.) ("Virginia Arbitration").
19 Joint Commentors' Filing of Summary Table Pursuant to ALJ's Ruling, 1/21/05.
20 In D.05-09-045, the Commission addressed issues relating to Verizon's intent to replace circuit switches with packet switches.
21 See Verizon/Tardiff, 8/6/04, pp. 6 and 50-59.
