Forecasting Transportation Revenue Sources: Survey of State Practices (2015)

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.

9 FEDERAL HIGHWAY REVENUE FORECASTING Because many states incorporate estimates of transportation revenues from the federal government into their own forecasts, and because the procedures used to forecast revenue by the fed- eral government are of general interest to those in the states making similar forecasts, the authors queried the FHWA on its revenue forecasting procedures. The FHWA Office of Trans- portation Policy Studies developed and maintains a Highway Revenue Forecasting Model (HRFM) for use in preparing Federal Highway Cost Allocation Studies. Staff of that office reported that the revenue estimation procedure is composed of several spreadsheets into which data are entered manually each time the model is updated. Estimates of federal transpor- tation revenue are currently available through 2020; the model is currently under review and subject to modification. Data used in developing these forecasts are generally not published and are subject to frequent revision. The HRFM currently can estimate revenues derived from federal motor fuel sales taxes and can incorporate the poten- tial indexing of federal motor fuel taxes to inflation. It also has the capability to estimate taxes based on axle-weights or gross vehicle weights of all vehicles; proceeds from poten- tial weight-distance truck fees; and revenues that could be collected based upon the pricing of vehicle air pollutant or greenhouse gas emissions. The spreadsheet models allow estimation of user fee pay- ments in 20 different classes of vehicles and 30 vehicle weight groups. For each year, vehicle class and weight group entries are made of estimates of the number of vehicles in each category and VMT are then estimated for each category. An overall average of miles per gallon is used to estimate fuel efficiency for all heavy and medium-weight trucks to esti- mate revenues from motor fuel taxes. Truck and trailer excise tax revenues are estimated based on approximated sales of power units in each category, which are estimated outside of the model. Similarly, revenues from excise taxes from the sale of tires are included; policy studies staff reported that the tire tax revenues estimates are based on data from 2005. Because the HRFM is out of date and of limited scope, states tended to rely more frequently upon the National Energy Modeling System (NEMS), created at the Energy Informa- tion Administration of the U.S. Department of Energy. NEMS projects the production, consumption, conversion, import, and pricing of energy. The model relies on assumptions for economic variables, including world energy market inter- actions, resource availability (which influences costs), tech- nological choice and characteristics, and demographics. The modules that address transportation energy consumption are updated annually, and are linked with the models represent- ing other sectors of the economy. Although each individual module is simple in concept, the system of models is complex because of its size and numer- ous linkages. The system amounts to a detailed accounting system, with linkages among regions of the United States, representations of social and economic activity, and between expenditures on transportation and the economic activities from which they are derived. Of most relevance to states, the NEMS model contains a light-duty vehicle submodule and a freight transportation submodule. These rely on cur- rent economic data, information on stocks of vehicles, and current measures of fuel efficiency. Because the EIA devotes substantial resources to updating every element of this model and makes the results and the inputs available for use through its websites, and because states lack resources with which to prepare their own detailed models, many states reported that they rely on the NEMS model when estimating and forecasting energy consumption in transportation, which is the principal determinant of fuel tax revenues. Readers may consult the latest publications of the National Energy Out- look (Energy Information Agency 2014a) for an overview of trends in transportation and other energy consumption and to review recent changes in projections. Those working on transportation revenue forecasts might be especially inter- ested in the EIA’s web publication entitled Transportation Demand Module of the National Energy Modeling System: Model Documentation (Energy Information Agency 2014b) for details as to how the transportation demand estimates are produced and data tables that could be useful when preparing state revenue forecasts. SHORT-TERM VS. LONG-TERM REVENUE FORECASTING All of the states reported preparing short-term forecasts, having time horizons of less than five years. Short-term forecasts vary in frequency of modeling and publication. One state reported chapter three FEDERAL AND STATE REVENUE FORECASTING PRACTICES

10 that short-range forecasts were created by applying subjective judgment to the interpretation of more formal long-range fore- cast results. Thirty-one (31) states reported that the state DOTs are either partially or exclusively responsible for the forecast- ing efforts. Partial involvement exists when the process is undertaken jointly with other agencies such as departments of finance or departments of motor vehicles. (Institutional arrangements for accomplishing the forecasts are described in more detail in chapter four.) Thirty-eight (38) states reported that they prepare long- term forecasts of future revenues, that is, with a time horizon of longer than five years, usually in support of State Transpor- tation Improvement Plans (STIPs). The Iowa DOT produces a five-year forecast of revenues each year. West Virginia uses a six-year forecast for state budgeting purposes. Seven states—Alabama, Arkansas, Colorado, Maine, New Jersey, Tennessee, and Vermont—do not produce long-term revenue forecasts. One respondent said that long-term fore- casts had too many uncertain factors, especially concern- ing the price of gasoline and the status of federal funding. In other states, the DOT produces both long- and short-term forecasts, but does so using different methodologies. For example, Michigan DOT’s short-term, one- to two-year fore- cast is done in “consensus” with the Michigan Department of Treasury, while forecasts for years three to five are carried out entirely by the DOT. TRADITIONAL SOURCES The majority of forecasting activity focuses on traditional revenue sources: disbursements of federal funds as reimburse- ment for expenditures, state fuel and excise taxes, state vehicle registration fees, tolls, and local taxes and fees. Figure 2 shows the widely reported types of revenue that are forecast by states, according to survey results. Distributions of Federal Funds The federal motor fuel tax is a per-gallon tax that is levied on gasoline, diesel, and other special fuels. Along with other fed- eral revenues such as fuel excise taxes, a portion of the motor fuel tax is allocated to states through the Highway Trust Fund through legislative programs (AASHTO Center of Excellence in Transportation Finance 2004). Thirty-nine (39) state DOTs estimate annual levels of federal funds that will be received as the dollar amount from the federal HTF in a given year. In recent years, projections of disbursements of federal funds to the states have also included general fund revenues as well as user fees, because the HTF has been augmented by infusions of general funds authorized by Congress. State Motor Fuel Tax State motor fuel taxes are, for many states, the largest single source of dedicated revenue for transportation programs. Revenues Routinely Forecast by States Type of Revenue Forecast Number of States FIGURE 2 State revenue forecasting by revenue type.

11 These include per-gallon gasoline and diesel excise taxes and ad valorem sales taxes levied on fuel. Forty-three (43) states responded that they typically forecast state motor fuel tax revenue, and 42 states forecast diesel fuel tax revenue. State Vehicle Registration Fees State vehicle registration fees are another significant source of dedicated state revenue. Thirty-nine (39) states responded that they forecast this type of revenue. In some states, vehicle reg- istration fees are not available for programs administered by the state DOTs; hence, they do not forecast revenues produced by those fees. For example, in California, vehicle registration fees are earmarked to support the Department of Motor Vehi- cles and the California Highway Patrol, which are not part of the California Department of Transportation (Caltrans). Revenue from Tolled Facilities Nine states regularly forecast toll revenues. In Texas, while some toll roads are operated by the state DOT, others are operated by a separate entity that is responsible for forecast- ing those toll revenues. In other cases, the state DOT collects and forecasts reve- nues from tolled facilities, or supervises consultants’ forecasts of revenues. An example of this arrangement is in Washing- ton State, where the DOT (WSDOT) supervises consultants’ forecasting of toll revenue on certain Puget Sound bridges. Several states forecast toll revenues using travel demand models and current per-trip tolling rates; however, a Tacoma Narrows Bridge forecasting work group reported that a travel demand model used by the WSDOT consultant in 2013 and 2014 overestimated both bridge traffic volume and toll rev- enue. Based on this work group’s review, WSDOT is evalu- ating the performance of an alternative econometric forecast model. Since 2010, other consultants have used a travel demand model to forecast toll revenue on another bridge in the Puget Sound region; those forecast projections have come in close to the actual figures. State General Fund Allocations Fifteen (15) states forecast general fund allocations to trans- portation programs. Some state transportation programs’ general fund revenue allocations are determined by formula; others are determined by appropriations processes in their legislatures. State Sales Taxes State sales tax revenues can be earmarked for transportation programs, or general sales taxes can include some alloca- tions to transportation programs. These taxes are forecast by 23 states. Local Taxes and Fees Six states forecast such taxes and fees as local or county excise taxes that are levied for transportation projects in par- ticular areas, as in Arizona’s Maricopa County. Similarly, several counties in Georgia have a special local option sales tax that is used to fund capital transit projects. Other Sources The remaining traditional revenue sources vary widely by state. Table 1 lists the additional revenue sources that states reported forecasting on a regular basis. This list is not exhaus- tive, as there may be other streams of revenue being forecast that were not identified by survey respondents. REVENUE FORECASTING METHODS This section discusses in more depth the most common meth- ods that state DOTs use to forecast revenues: simple histori- cal trend extrapolation, expert consensus, and econometric Sources of Revenue Forecast Oil company gross receipts tax Investment income and small revenue sources Aviation fuel tax Rental car surcharge Off-highway sales tax on dyed diesel Documentary stamp tax Motor carrier surtax Fuel tax transfers and refunds Oversize/overweight permits Damage to state property Toll road lease proceeds Miscellaneous permits and fees Weight distance tax Federal revenue reimbursements 2% Special fuel excise tax on dyed fuel usage Driver’s license revenue Truck regulation and enforcement fees Projected unencumbered cash balances Oil company—franchise tax Contribution from PA Turnpike Commission Vehicle code fines Vehicle sales tax Local participation Interest on cash balances Miscellaneous revenue Dedicated taxes and fees Ferry boat fees DMV fees Traffic violation fees Airspace leasing TABLE 1 SURVEY-REPORTED REVENUES THAT ARE FORECAST BY STATES

12 models, including econometric regression analysis. It is impor- tant to note that few states use one method exclusively. A state may use different forecasting methods for different revenue sources, or use a combined approach. Differing methods were also reported for short- and long-term forecasting. Trend Extrapolation The most straightforward revenue forecasting tools involve applying a mathematical formula to past revenue levels to determine future revenues. Historical trends are extrapolated into the future, either in a linear manner or with some adjust- ment to account for expected future changes in underlying conditions. As noted previously, some Georgia jurisdictions rely on local general fund appropriations and a Special Purpose Local Option Sales Tax (SPLOST) to provide the local share of funding for state highway projects. Georgia’s DOT devel- ops its forecasts for these local funds using historical alloca- tions. For transit, local funding comes from either general fund allocations or a local option sales tax. To forecast the amounts of these local sales taxes, historical data indicated a 3.5% compounded annual growth rate, and it was assumed that this growth rate would continue throughout the 30-year plan (Georgia Statewide Transportation Plan 2006). To fore- cast revenue collected from Georgia’s statewide gasoline tax, a regression model is used. Trend extrapolation can also be used to anticipate fed- eral funding disbursements. Montana bases its short-term (five-year) federal revenue forecasts on historic funding levels, known changes in federal program funding levels, anticipated changes in state share, and an estimated moder- ate inflation rate. The long-term revenue forecast method is similar but somewhat less detailed. According to the respon- dent from Montana’s DOT, “These long-term forecasts are better used for planning efforts—and to ensure that we iden- tify areas of future need.” It appears that trend extrapolations are most relied upon for revenue sources that are historically stable, such as gen- eral fund allocations, local contributions, and other sources that are determined by a fixed formula. For revenues that are dependent upon fluctuations in vehicle miles traveled or fuel prices (i.e., motor fuel taxes), states appear to be in transi- tion and some are adopting more sophisticated forecasting methods. In addition, the 2007–2010 recession has made projecting historical trends increasingly risky, and states are reconsidering their practices. Expert Consensus A commonly used method of forecasting is referred to as “expert consensus,” which relies on the professional judg- ment of a selected panel or conference of economists, ana- lysts, academics, and others who discuss and try to agree upon future revenue projections or on critical inputs that will affect those projections. Sometimes this method is used in conjunction with econometric models; for example, Arizona uses an econometric model, and the results are reviewed and adjusted as needed by a panel of experts. Florida’s DOT participates in a Revenue Estimating Con- ference at least twice per year where the 10-year forecast of the Department of Revenue is officially agreed upon and adopted. The other agencies represented at the conferences are the Florida legislature, the governor’s office, and Depart- ment of Highway Safety and Motor Vehicles. Similarly, Utah’s DOT relies on revenue projections from the Office of the Legislative Fiscal Analyst, the Governor’s Office of Management and Budget, and the state treasurer’s office. Expert judgment is also used to estimate independent eco- nomic variables. In 1992, Arizona introduced a risk analysis process for deriving the independent variables for its regres- sion-based forecast. A panel of experts was asked to judge the likely range, from a high of 90% to a low of 10%, for each independent variable. A spreadsheet of the assumptions of each expert and each variable is regularly published as part of Arizona DOT’s Risk Analysis Forecast (RAF) process, shown here in Table 2. One state respondent mentioned that the persons who provide the forecasts are not necessarily particularly expert at transportation funding. According to this respondent, the economists relied upon to provide forecasts are “more inter- ested in the state general fund forecast” and “sometimes do not fully analyze the factors affecting the forecast of gasoline taxes.” This statement may provide insight into why many states choose to empanel experts having different perspec- tives that may be important in building a meaningful consen- sus when forecasts are based on judgment. Econometric Models According to the survey, the majority of states forecast rev- enue using some version of econometric analysis. The fol- lowing section examines the independent variables that are used most often in states’ forecasting models, and presents case examples of two states, Oregon and Washington, that use regression analysis to forecast transportation revenues. Regression models are commonly used to quantify rela- tionships between variables. In their simplest form, linear regression models specify a relationship between indepen- dent variables (per-gallon price of gasoline, state population, state employment) and a single dependent variable (e.g., gasoline consumed or total revenue). Regression functions measure the extent to which the dependent variable changes based on changes to the independent variable(s). Regression modeling is often used to forecast a quantity, such as number of passenger vehicles sold or gallons of gasoline consumed, which is then multiplied by the per-unit tax rate to reach a

13 forecast for that particular revenue source. Each revenue source might have its own set of regression equations. Time series analysis, including ARIMA analysis, is a method used to analyze a series of successive data points that are col- lected at uniform intervals. This type of analysis is used to pre- dict future variables based on historical trends. State DOTs often use either univariate (meaning a single independent variable) or multivariate regression models to perform time series analy- ses of historical price and consumption data points. In some cases, models are adjusted to account for seasonal variability. CASE EXAMPLE: WASHINGTON WSDOT uses multiple models for short- and long-term forecasts of diesel and gasoline consumption in the state of Washington. In 2010, WSDOT revised its forecasting methodology to use a multivariate regression model with economic independent variables for both its quarterly and annual models. However, WSDOT found that monthly time series analysis provided more accurate forecasts in the short term than the quarterly econometric forecasts (Table 3). It adopted a new model in 2013, which uses historical monthly Nominal Personal Income Growth Pop- ulation Growth Construction Employment Growth 30 Year Mortgage Rate Phoenix CPI Growth Sky Harbor Passenger Traffic Growth Total Non- Farm Employment Growth FY2013 Upper 10% 6.74% 2.14% 14.14% 4.74% 2.74% 4.26% 3.55% Lower 10% 2.94% 0.99% 4.16% 3.66% 1.00% 1.25% 1.25% Median 5.17% 1.63% 8.21% 4.19% 1.87% 2.93% 2.53% FY2014 Upper 10% 7.83% 2.69% 16.45% 5.18% 3.21% 4.38% 3.83% Lower 10% 3.52% 1.30% 5.22% 3.84% 1.42% 1.31% 1.38% Median 5.85% 2.06% 10.53% 4.53% 2.28% 2.95% 2.89% FY2015 Upper 10% 8.10% 3.02% 16.67% 5.74% 3.61% 4.57% 4.43% Lower 10% 3.86% 1.44% 4.41% 4.08% 1.39% 0.93% 1.35% Median 6.25% 2.34% 10.56% 4.91% 2.40% 2.78% 3.18% FY2016 Upper 10% 8.21% 3.01% 13.68% 6.44% 3.79% 4.49% 4.62% Lower 10% 3.86% 1.42% 2.88% 4.58% 1.41% 0.73% 1.42% Median 6.15% 2.34% 8.15% 5.55% 2.37% 2.64% 3.33% FY2017 Upper 10% 8.24% 2.97% 12.34% 7.13% 4.01% 4.47% 4.63% Lower 10% 3.50% 1.33% 0.90% 4.94% 1.21% 0.50% 1.20% Median 5.92% 5.92% 6.20% 5.94% 2.28% 2.45% 3.31% FY2021 Upper 10% 8.59% 3.10% 3.10% 7.78% 4.26% 5.03% 5.30% Lower 10% 2.01% 0.79% 0.79% 5.10% 0.87% -0.18% 0.12% Median 5.24% 1.96% 1.96% 6.37% 2.32% 2.26% 2.77% FY2026 Upper 10% 8.45% 3.00% 10.43% 7.91% 4.70% 5.06% 5.19% Lower 10% 1.66% 0.67% -5.62% 5.15% 0.92% -0.31% -0.49% Median 4.86% 1.74% 3.54% 6.45% 2.38% 2.19% 2.58% TABLE 2 ARIZONA’S DOT RISK ANALYSIS FORECAST PANEL GROWTH FORECASTS Dependent Variable: 1st Difference LOG(GAS) Variable Coefficient Standard Error T-statistics Prob. Intercept 0.00175 0.00204 0.8564 0.401 Log_WA_Emp 0.63306 0.09357 6.7656 0.000 LOG_ US Fuel Efficiency * WA_Gas Price –0.09749 0.02326 –4.1922 0.0004 AR(1) –0.50424 0.18475 –2.7294 0.0122 MA(1) 0.19693 0.07612 2.58718 0.0168 MA(2) –0.15832 0.08271 –1.91407 0.0687 MA(3) –0.90648 0.06293 –14.4055 0.0000 Adjusted R Squared 0.5614 Durbin-Watson 2.0302 Root Mean Square Error 38.6597 Schwarz Bayesian Criterion –5.393011 TABLE 3 WASHINGTON DOT’S UPDATED ANNUAL GAS CONSUMPTION ECONOMETRIC MODEL OUTPUTS

14 fuel consumption levels rather than using historical economic data as independent variables. The new model operates using an ARIMA model that is better able to account for seasonal variables that affect fuel consumption. These monthly models also take into account the number of trading days in each month and some holidays; and unusual and unpredictable patterns in the data, such as adverse weather conditions, that could influ- ence fuel consumption. WSDOT has found the ARIMA models to be accurate in the near-term (two to three years). In the long term, it still uses the first-difference multivariate regression model with economic variables as independent variables. of new EPA regulations. The revised long-term diesel forecast model has two independent variables: an economic activity variable—state employment in the trade, transportation, and utilities sectors—and residents’ real personal income. Independent Variables WSDOT uses 10 independent variables for its quarterly forecasts of transportation revenues. Generally, the Washington Economic and Revenue Forecast Council provides WSDOT with the economic variable forecast through FY 2019. WSDOT extends the economic variables’ forecast for 14 years. Certain long-term forecasts for economic variables, such as popu- lation and employment, are provided by the state’s Office of Financial Management (Washington Transportation Rev- enue Forecast Council 2013). The sources for these variables are listed in Table 4. Gasoline Consumption Variables Forecasting transportation revenue levels can be a multi- step process. It is common for states to develop forecasts for variables such as statewide gasoline consumption or fuel price that are then inserted into other models. For example, Minnesota uses a national gasoline consumption forecast produced by the consulting firm IHS Global Insight (GI) and regional motor gasoline consumption forecasts produced by the EIA. Minnesota’s DOT combines these two sources into an average, and multiplies this average by the current motor fuel tax rate to estimate its revenues. Table 5 shows how Variables Sources WA personal income Based on the Washington Economic and Revenue Forecast Council in short-term (through 2017) based on forecasts from Blue Chip average US GDP growth rates and NYMEX fuel prices; and long-term Global Insight forecast Population Preliminary Office of Financial Management population projections Inflation (2 measures: CPI and IPDC) Washington Economic and Revenue Forecast Council for short-term and Global Insight forecast for long-term Employment WA non-ag. imp; WA TTU, WA retail trade and national unemployment rates Oil price index 2014 Global Insight forecast Fuel efficiency 2014 short and long-term Global Insight forecast U.S. sales of light vehicles 2014 Global Insight forecast and November 2013 long-term Global Insight forecast3 U.S. fuel prices (retail gas and diesel and index of petroleum products) EIA for short-term and Global Insight for long-term TABLE 4 SOURCES OF DATA FOR WASHINGTON DOT’S ECONOMETRIC MODELS Case Example: Washington DOT’s Regression Equations The equation for annual gasoline consumption in Washington is logarithmic in its formulation. The final model is a log-log functional model, with first difference of natural logs used on both sides of the equation. The equation is customarily referred to as using ordinary least squares. The equation for annual gasoline consumption in Wash- ington is: ln _ _ ln _ ( ) ( ) ( ) ( ) = α + ϕ + φ + δ + ε pLn Gas WA Emp WA GasP Eff WA pop Where: Gas = annual gross gasoline consumption WA_Emp = annual Washington non-farm employment WA_GasPEff = annual Washington gasoline prices  U.S. average fuel efficiency e = Stochastic disturbance on gasoline con- sumption. The individual regression coefficients were selected because they are statistically significant and have economic reasonable values. According to published reports, the model fits historical gasoline consumption data well. “Overall, the independent vari- ables are able to explain most of the variation in gasoline con- sumption.” Table 3 shows the manner in which model outputs are presented for users of the modeling results. According to Washington DOT staff, using the first difference multivariate regression model has led to a more pessimistic forecast and flat- tened the gas consumption forecast in the long-term than the prior forecast model predictions. This change in Washington state forecasting methodology was made in November 2010. Diesel Variables WSDOT has also updated its quarterly and annual diesel con- sumption models in similar fashion. Prior to 2010, WSDOT’s diesel consumption models used historical diesel consump- tion, with adjustments being made for trading days and certain holidays. This monthly model has proved to be quite accurate. One difficulty that WSDOT found with the long-term diesel fuel consumption model was that it was not able to account for the truck fuel efficiency variable. Truck fuel efficiency has increased slightly and is projected to grow in the future because

Source: Minnesota’s DOT 2014 Transportation Funds Forecast 2014. TABLE 5 MINNESOTA’S FUEL CONSUMPTION INPUTS, 2013–2017 Case Example: Oregon Over 200 stochastic equations comprise Oregon’s DOT (ODOT’) short-run revenue forecasting framework. The quantities in the model (“transactions”) span taxable gallons of motor fuels (principally gasoline and use fuels including diesel), weight-mile taxes paid by heavy vehicles in lieu of paying diesel fuel taxes (26,001 lb or more), to a very broad array of drivers and vehicles transactions. The latter range from driver’ licenses, vehicle registrations, and license plates to titling, as well as to business regulation of auto-related business activities. As an illustration of the specifications in the model, the fore- cast equation for motor fuels, which account for nearly 50% of the agency’s state-generated revenues, can be initially represented by the general function: , , , , , , , ,0 1 2 3 4 5 6( )= αQ f X X X X X X DBFBt where the independent variable vector (X) is made up of a0 = Intercept term, X1 = Real retail price index for motor fuels, with a distributed lag structure of 5 quarters, X2 = Fuel efficiency of the existing stock of light-duty vehicles, X3 = Oregon total non-farm employment, X4 = Oregon real aggregate personal income, X5 = Oregon labor force participation rate, X6 = Consumer sentiment index, and DBFB = Binary variable for implementation of an ethanol blend- ing mandate in 2008. Traditional functional forms of linear and multiplicative (linear in logs) specifications are used in parallel. The former forms have mathematically varying elasticities, while the latter have constant elasticities and, therefore, are somewhat more restrictive. The addi- tive specifications are usually the most used for final forecasts, as they permit the elasticities to vary with the point of evaluation. The equations are more representative of a reduced form structure, rather than as structural demand functions per se. As a result, the estimated elasticities are more properly viewed simply as “sensitivities,” not as demand elasticities. Notwithstanding, they are sometimes used interchangeably. The quarterly observations used in the econometric equa- tions are based on monthly data, tested for seasonality. The equa- tions are estimated using a Generalized Least Squares Estima- tor (GLSE). The fitted equations explain in excess of 98% of the variation in gallons consumed. The relative standard errors of the regression equations are generally in the range of 1.3–1.6& when compared to mean usage. The sensitivities routinely obtained from estimation (that is, the percentage change in usage relative to a percent change in an inde- pendent variable) are illustrated below. Elasticity Estimates Real fuel prices -0.073 Fuel efficiency of vehicle stock -0.13 Employment +0.3 Real personal income +0.25 Labor force participation rate +0.17 For example, a 10% increase in the real fuel price brings about a 0.7% decrease in fuel usage, distributed over a five-quarter response period. This relative “inelasticity” conforms closely with a preponderance of other empirical findings, given the largely derived-demand nature of gas consumption. ODOT modelers also maintain vector auto regression models for a number of usage transactions, although they do not fre- quently serve as a basis for final forecasts. ARIMA time series mod- els are also used in various driver and vehicle transaction forecast equations.

16 Minnesota’s DOT uses these inputs for its five-year (2013– 2017) forecasts. Several states, including Washington, rely on fuel con- sumption forecasts from GI and/or EIA as data inputs. Other states, such as Oregon, use GI and EIA for exogenous data only, with fuel consumption observations performed by the state DOT. Good, continuous data for state fleet fuel econ- omy is scarce, making it difficult to consider the regional differences in the composition of the national fleet. Demographic and Socioeconomic Variables for Econometric Analysis States rely on a multitude of demographic and socioeco- nomic data as inputs to their models. For example, Arizona estimates its transportation excise tax revenue with what it describes as a “comprehensive regression-based econo- metric model.” Approved in 1985 by voters in Maricopa County, which includes Phoenix, the excise tax is a half- cent sales tax dedicated to constructing a freeway system within the county. In 2004, the sales tax was extended for another 20 years, with proceeds dedicated to construction and improvements of new and existing freeways, road- ways, and transit services. The model uses seven indepen- dent variables that relate to state and county demographic information: • Maricopa County nominal personal income growth • Maricopa County population growth • Maricopa County construction employment growth • Phoenix consumer price index (CPI) • Sky Harbor Airport passenger traffic growth • Maricopa County total non-farm employment growth • The 30-year mortgage rate. The independent variables used by other states surveyed can be found in Appendix B. ACCURACY AND SHORTCOMINGS OF FORECASTING METHODS Respondents were asked to evaluate the accuracy of their cur- rent revenue forecasting processes. They were also asked to identify major shortcomings of these practices, and whether any action has been taken to improve their accuracy. While some states are very confident of their findings, seven respon- dents reported having updated or changed their forecasting methods in the past two years, either for reasons that were internally generated (staff desire to improve accuracy) and/ or in response to changes in external factors. In one case, North Dakota’s impetus for change was that the state has been experiencing a large increase in fuel usage as a result of the oil extraction activities in the western part of the state. The traditional means of forecasting did not address the new levels of fuel usage. In Oregon, a review of forecasting accuracy is published quarterly, and the forecast itself is updated every six months. However, most states do not report publishing regular accu- racy reviews. In Washington, accuracy has improved with a revised methodology. The biennial revenue forecast variance in the fuel consumption revenue has fluctuated less than 1.5% around the midpoint, which is a direct result of the forecast model changes. Table 6 presents some of the most informative state responses regarding accuracy. Many respondents mentioned that they are experienc- ing increasing uncertainty and difficulty in forecasting fed- eral revenues, particularly in relation to the future decisions of Congress. In some cases, political uncertainty leads to states’ choosing not to include federal revenues in their forecasting models. In other cases, the respondents reported that their DOTs simply use their own judgment to pick an estimate of federal distributions that appeared reasonable. WSDOT examines the CBO national forecast for future revenue and expenditures to predict when the HTF will become insolvent, and reduces the Washington share of national total of federal transportation revenue projections by an amount that keeps the HTF from going into deficit. By contrast to federal funding, revenues from state sources were generally considered more stable and predict- able. Illustrative responses about this issue are summarized in Table 7. Other significant hurdles that states reported were per- ceived shortcomings in the accuracy of their input data, a paucity of historical information, or a lack of sophistication of methodology. Table 8 presents a variety of additional rea- sons some states gave as challenges to forecasting accuracy. Finally, some states expressed concern about the accuracy of their forecasting techniques during economic downturns or recessions; these are presented in Table 9. For example, ODOT reports that staff members have generally been satis- fied with the precision of forecasts except during the Great Recession, when error rates were higher. RESPONSE TO POLICY PROPOSALS Twenty-four (24) of 45 states reported that they had been asked within the past two years to evaluate proposals to raise additional transportation revenue, either by raising rates or changing the structure of traditional transportation revenue sources. Table 10 presents sources of increases to current revenue sources that were enumerated, and Table 11 lists new sources of revenue. One state DOT respondent reported feeling constrained as a result of political sensitivities, reporting that the agency

17 Texas “Overall state revenue forecasts have proved to be accurate within ±5% over the last seven years. However, their forecasts of motor fuel tax and registration fees have been much more accurate.” Arizona “ADOT has a very robust forecasting process that has been in place since 1992, which has proven highly accurate in all but the most volatile of economic environments.” Iowa “Our forecast amounts vs. actual receipts have been within 98% each of the last several years due to the relative stability of the traditional transportation revenue sources. Estimation of non-traditional sources would be more difficult.” West Virginia “The forecasts work for our internal purposes, but they are basic and provide only general information.” Source: State DOT survey respondents. TABLE 6 ILLUSTRATIVE STATE RESPONSES REGARDING FORECAST ACCURACY Alaska “The current Congressional bill was only two years long and we are in the last year. No one knows how the next multi-year will be structured or if there will be a next bill. State allocations are just as ephemeral as Alaska's state government is funded by oil production taxes from a rapidly declining field and very uncertain oil prices.” Indiana “Inability to accurately predict Federal Highway Transportation Funds.” Mississippi “Our state revenues are fairly consistent each year so our forecasts for these are accurate. Since our federal revenues are dependent on contract expenditures and available obligation authority, they are much harder to forecast.” Missouri “Comfortable with state revenue sources. Weakness due to uncertainty of federal funding in the current environment.” Nevada “No problems with techniques - uncertainty of Federal highway Administration program makes forecasting difficult.” Ohio “The forecasts used by ODOT are fairly accurate. The biggest challenge is the uncertainty of funding, especially at the federal level.” Montana “Uncertainty in the federal program revenue generation. As mentioned above, the federal program funding size and timing has been generally unstable since the early to mid-2000s, leaving the states that heavily depend on the federal program with little ability to plan/project future program levels. Planning these kinds of infrastructure projects and making the right asset management decisions rely heavily on future program levels, it's difficult to know if we are making the right investment decisions with the current state of the federal program.” Louisiana: “Federal funds are extremely difficult to forecast because of its dependence on the wishes of Congress and not necessarily the motor fuel taxes collections.” Utah “Federal funds are basically projected as "flat" increase/decrease. The inability to know whether general fund infusions will actually continue as in the past is difficult.” Source: State DOT survey respondents. TABLE 7 ILLUSTRATIVE STATE RESPONSES REGARDING FEDERAL FUNDING IN FORECASTS North Dakota “Initially when the oil activity began in the state, the traditional techniques did not fit the traditional model for revenue forecasting. Now that we are gaining additional historical data, the forecasting of revenue projections has been much closer to the actual revenue received.” California “Lack of sophistication in forecasting revenues leaves us with ‘long-term trends’ basis rather than an economic basis.” Nebraska “Stability in motor fuel prices. Two components of our motor fuel tax are derived from a percentage applied to the motor fuel prices.” Utah “Projections of fuel efficiency of future vehicle fleets on the highways pose challenges.” Source: State DOT survey respondents. TABLE 8 ILLUSTRATIVE STATEMENTS ABOUT CHALLENGES TO REVENUE FORECAST ACCURACY Oregon “Major shortcoming of the revenue forecasting techniques currently used by the department is the ongoing inability to more accurately estimate economic downturns and associated detrimental impacts on transportation revenues and more accurately consider fuel efficiency and vehicle miles of travel growth uncertainty in future years.” South Dakota “They are becoming more unreliable due to economic volatility and slowing VMT rates of growth.” Delaware “Revenue forecasting has been very accurate, except during recessionary periods. FY09 and Fy10 were difficult.” Source: State DOT survey respondents. TABLE 9 STATEMENTS REGARDING THE IMPACT OF THE ECONOMIC DOWNTURN ON FORECAST ACCURACY

18 Type of Change State Providing Response Half-cent sales tax AR Increase in petroleum transfer fee ID Increases to the state excise taxes IA Gas tax increases MN, ID Increased motor vehicle sales taxes MN Increased vehicle registration taxes and fees MN, ID TABLE 10 CHANGES IN REVENUE SOURCES RESPONDENTS WERE ASKED TO EVALUATE New Revenue Source Evaluated State Providing Response A dedicated regional transportation tax passed in three regions of the state which fund transportation improvements within the counties and cities throughout the regions GA Various bonding proposals MN Indexing the fuel tax IN Turnpike Revenue Bonds OH Application of a sales tax on the price of fuel IA Transfer of general funds to transportation IA Appropriating a set dollar amount of the state's general fund to transportation progra ms IN Appropriating a % of state sales tax for transportation program funding IN Electric/hybrid car fee ID Rental car fees ID Increased dyed diesel enforcement ID Increase in sales tax by 1% with proceeds directed to highway account ID Transfer of 4.5% sales tax on motor vehicle goods to highways (unsuccessful) AR Passage of increase in tax on LPG and CNG (unsuccessful) AR A change (revenue-neutral, at least in the short term) in the way we levy motor fuel taxes, from a cents-per-gallon to a wholesale tax rate DC Sales tax on tires and other automotive accessories directed to the highway account ID Source: State DOT survey respondents. TABLE 11 NEW REVENUE SOURCES RESPONDENTS WERE ASKED TO EVALUATE cannot openly evaluate a full range of revenue options. The respondent also mentioned that the agency feels pressured by outside parties to generate overly optimistic revenue results. A more typical response came from Oregon. In forecast- ing revenue, ODOT bases projections on “current law” only; it does not speculate on legislative proposals until they have been enacted. For the purposes of estimating the amount of revenue that would be generated by various proposals, the respondent said that a simple approximation is sufficient. In addition, the respondent finds that attempting to model rev- enues from new sources is difficult because it is not often clear which variables are most appropriate to consider. To the extent that some options for future revenue are more novel than others, their very novelty makes them inherently more difficult to forecast.