Estimating Toll Road Demand and Revenue (2007)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

This chapter presents several conclusions and observations derived from the study. The chapter closes with suggestions for further methodological and procedural research on toll road demand and revenue forecasting taken from the litera- ture and the survey of practitioners. The survey of practice and the literature corroborated and detailed several issues of concern, most of which had been identified in the scope for this synthesis. Several conclusions were reached. • Many of the problems that had been identified with the performance of traffic and revenue forecasts were related to the applications of the models, less so to methods and algorithms. In particular, assumptions regarding land use, network inputs, values of time, and other inputs; the process of reviewing models and their results; and the treatment of uncertainties and risks were most often cited in the literature in explaining why the performance problems occurred. It is noteworthy that much of the literature that describes these problems in the context of toll road demand and revenue forecasts comes from the financial community, rather than the transportation modeling community. This suggests a disconnect between the two communities; the latter being the “traditional” users (and developers) of the models and the former representing the new users. The financial community’s concerns also parallel those of other new users, such as those involved in air quality conformity analysis, which suggest that the state of the practice in travel demand modeling has not kept pace with the issues that the models now must address. The disconnect is exemplified in different ways; for example, in the use of risk analysis (incorporating a range of outcomes) and stress tests (which assess extreme and multiple “shocks”); neither of which has been widely applied to transportation planning. Another example is the treatment of the impact of short-term eco- nomic recessions on long-range demand forecasts, which is starting to be considered. A third example is the more explicit understanding of the role of different economic influences, such as gender, age, and occupation, on toll road choice. It is also incumbent upon the new users to under- stand how the models work and how to interpret their results, as well as their inherent limitations; and it is equally incumbent on the developers of the models and 40 their data to provide this understanding. One traffic and revenue study reviewed for this synthesis noted that “professional practices and procedures were used in the development of the traffic and revenue forecasts included in this report” as a preface to its overview of the modeling process, which was brief and did not pro- vide many details. Doubtless this is true; however, the ensuing documentation provided very little informa- tion that could help analysts understand the input or modeling assumptions, let alone address the questions posed in the preceding chapter’s checklists. On the other hand, some traffic and revenue studies provided considerably more explanation of the modeling proce- dures and, especially, of the input assumptions and how they were derived. • Nonetheless, improvements in both aspects (application and method) are required to address the performance of the models in traffic and revenue forecasts. • Although the application of state-of-the-art method- ological improvements into common practice—such as activity-based models and network micro-simulation— should be expected to improve the state of the practice, it is likely that these alone will not improve the perfor- mance of traffic and revenue forecasts. Notwithstanding, several methodological improve- ments might be made. Two important improvements to the travel demand modeling process are time-of-day choice modeling and the modeling of commercial and truck traffic. The understanding of traveler behavior when faced with tolls continues to evolve and must be better understood; that is, a more comprehensive under- standing of the actual determinants of choice are re- quired, including values of time and willingness to pay when confronted with different factors (such as the toll collection method). The explicit incorporation of risk and uncertainty in all aspects of the modeling process also is needed, as is consideration of inputs and outputs in terms of ranges rather than as absolutes. It could be argued that each of these areas requires further research and development before it can be implemented properly (e.g., the literature indicates that time-of-day choice modeling is an emerging topic). Conversely, it also could be argued that simplified (or better) methods already exist for accounting for each topic and would represent an immediate improvement to the forecast if considered (assuming, of course, that assumptions and methods are treated transparently). CHAPTER FIVE CONCLUSIONS

41 Finally, more generally, a basic prerequisite to any toll demand and revenue forecast (and to the ability to account for any of the previously cited improvements) must be the ability to model the four components of travel (generation, distribution, mode choice, and route choice) as a starting point. • According to the literature, questions regarding the per- formance of the models and forecasts have been posed mainly by the financial community, rather than by the transportation modeling community (with some notable exceptions). The latter community has focused on improving the methodological basis of modeling and its underlying data, as demonstrated by the extensive litera- ture and research on the topic. However, as suggested in the previous point, this emphasis has not generally recti- fied the performance problems. In other words, this suggests that there is a disconnect between the develop- ers of the models and their users, with the latter having evolved from the traditional decision makers to those with different or more rigorous decision-making criteria. • In turn, this suggests that those in the transportation community who are making investment decisions regarding tolled facilities do not always know which questions to ask of their modeling and forecasting efforts—in other words, the analytical and modeling capabilities available to them have not always kept pace with the needs. This was demonstrated by the large number of explanations that were cited in the literature and in the survey as causes of performance failures. In some cases, the explanations were contra- dictory; in particular, the financial community cited application as a key problem, whereas survey respon- dents cited model method as the problem. This can be explained in part by the relative newness of toll roads in some parts of the country and the corresponding lack of a long-term performance history. It also can be explained by the changing nature of the tolled facili- ties, in which parts of a facility (i.e., individual lanes) are now being tolled: this changes the analytical requirements significantly. The problem is exacerbated by the “confidential” or “proprietary” nature of the forecasts and methods that are developed for toll roads, and also by “optimism bias” on the part of the sponsor, local elected officials, or other advocates of the pro- posed toll road. • Observers have remarked, informally, that there is no standardization in the toll road demand and revenue modeling and forecasting processes. Also, they have questioned whether such standardization could be pro- moted in the community. Given that this mirrors a sim- ilar lack of standardization in travel demand modeling in general, as noted, that the state of the art in modeling continues to evolve, and that there are several different and valid techniques (e.g., as in toll diversion modeling), it may be more appropriate instead to do two things. First, standardize the terminology or at least list and cat- egorize the different definitions for key terms; and sec- ond, develop a commonly used set of questions and attributes that should be considered. As an example of the first, there is no single, commonly understood defini- tion of what is meant by an “investment grade” traffic and revenue forecast: it may be more appropriate instead to develop a list of how different organizations under- stand, interpret and use this term. An example of the second is the Texas Turnpike Authority’s guidance (see chapter four), which provides a practical treatment that perhaps could be reviewed for general application across the United States, taking into account the differ- ent perspectives of the owners, sponsors, and financial community. The Traffic Risk Index similarly provides a useful starting point for elaborating on the specific questions that should be asked (more precisely, the vari- ables that should be taken into account and their impacts) in the development of toll road demand and revenue forecasts. In their own right, standardizing and understanding the terminology will not improve the process and results of traffic and revenue forecasts. However, these improvements will encourage practitioners and owners to understand more clearly the objectives and require- ments of decision makers and financial sponsors, and ensure in turn that the components of the forecasting process are more responsive to these needs. • Several observers have noted that “you get what you pay for.” In other words, in the United States, sufficient resources have not been devoted to procuring the required data or to updating older data, or to calibrating models to the level of detail that is required. The gen- eral practice in Europe, for example, is to prepare three sets of forecasts. This clearly requires an investment on someone’s part. Intuitively, better data, more detailed models, and multiple forecasts should improve model performance. Similarly, it is intuitive that a stronger role for peer review should improve the performance; however, the literature review did not uncover any eval- uations or specific assessments of the effectiveness of the peer review process for modeling in general. • It should also be noted that the comparisons in the liter- ature focused primarily on the revenues as opposed to the demand (i.e., the traffic and its composition). The litera- ture noted that revenue performance could be affected by changes in toll rates or by drawing on reserves, or other means; in other words, by actions that are not related directly to demand. Therefore, at least some of the avail- able information does not accurately reflect the outputs of the travel demand model and, accordingly, the linkage between the demand forecasts and the revenue perfor- mance is not always completely direct or explicit. Accordingly, there is a need to measure the performance of the travel demand models in their own right, specifi- cally examining how well the toll road demand models simulate each class of vehicle and traveler. The research literature cited in this synthesis largely focused on methodological issues; notably, the understanding

of the variables that affect the traveler’s decision to use a toll road, consideration of probability distributions to describe these variables as a means of analyzing and managing risk, development of time-of-day choice models, simulation of value of time and the role of stated preference surveys in esti- mating value of time or on the development of value-of-time models based on historical data that are now becoming avail- able. The need for continued research in these areas was inherent in the literature, with specific recommendations for research comprising the following. The financial community and practitioners made several explicit recommendations for further research. • Improve understanding of the impact of electronic toll collection (more generally, the type of tolling collec- tion) on value of time. • Incorporate risk analysis into the demand forecasting process, accounting for multiple possible inputs and outputs. • Account for the impacts of changed economic condi- tions (notably, recessions and lower than expected initial demographic and economic growth). • Improve the understanding of and the methods for esti- mating value of time and ramp-up, including the impact of existing congestion and expected development on travel demand and on toll road demand specifically. • Improve the treatment of trucks and commercial vehi- cles in toll road modeling. • Improve the validation of the travel demand models. • Develop a more improved understanding of the factors and assumptions that are used to develop the demand models and of the criteria that are used to assess their performance and calibration. A 2005 review of the practice of modeling road pricing by Spears made the following five recommendations for research. 1. Document case studies of transportation planning agencies that have incorporated road pricing in their 42 travel models to provide details “concerning changes in model structure, data requirements, value-of-time parameters, calibration and validation considerations, and specific application results from other modeling efforts.” 2. Compile and synthesize current and past empirical research on value of time and value of reliability, then compile the existing research on value of time into “an application-oriented document that provides travel modelers with reasonable ranges for [value of time], classified by income level, trip purpose, or other rele- vant parameters,” and includes practical guidance on value-of-time adjustments (such as the impact of elec- tronic toll collection). A similar compilation of the more limited research on value of reliability was also recommended, as well as the identification of priority areas for additional research and data. 3. Encourage data collection on travel behavior on road pricing projects. Funding for such projects should account for additional empirical data collection (notably, on value of time and value of reliability), as well as for an independent evaluation of the project. 4. Conduct basic and applied research to incorporate time-of-day and peak spreading models in current travel demand models to address the “principal limita- tion” in current travel demand models; that is, “the way in which they distribute daily trips by time-of- day.” New methods or models are needed to allow for more precise resolution of daily trip distributions (by the hour or half-hour), perform efficient multiclass assignments over multiple time periods, and allow for a systematic shift of trips between adjacent time peri- ods to reflect peak spreading. 5. Conduct basic research to better understand and measure the influence of traffic congestion (and the underlying factors that influence congestion on a day-to-day basis) on travel time variability. Archived operational traffic data exist to explore this relationship, although they have been largely unused for this subject.