Show Summary Details

Page of

PRINTED FROM OXFORD HANDBOOKS ONLINE ( © Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Handbooks Online for personal use (for details see Privacy Policy and Legal Notice).

Subscriber: null; date: 13 August 2020

Revenue Estimation

Abstract and Keywords

This article poses three questions: Where did the forecasting models go wrong? Who's to blame for the forecast errors? and How can [state and local] revenue estimation improve? To answer all three, the article starts out with a look at just how severe the 2007–2009 recession was, and then empirically examines how five common features of revenue-estimating models were affected by the second longest business-cycle contraction of the past eighty years. The Great Recession (eighteen months, peak to trough) was the most severe decline of any recession since the Great Depression of 1929–1933 (which lasted forty-three months). The article carefully works through each of the key indicators that a state revenue estimator watches, and it demonstrates that not only was the Great Recession dramatically deep (the last time all six major indicators contracted was 1948–1949, and even that was of shorter duration) but also that it has been very different than prior recessions.

Keywords: revenue estimation, forecasting models, recession, business-cycle contraction, state revenue, economic indicators

The Great Recession that began in December 2007 has exposed limitations in state and local revenue-estimating models that state and local policymakers rely on to make budget decisions. Even though the unemployment rates have not risen to the level or duration of the Great Depression, and today's social and financial safety net is much stronger than in the 1930s, an examination of several indicators reveals that in several ways the depth of the recession has been unprecedented. As the dust settles and the crises abate, each individual state—that is, state and local system—will go through a period of soul-searching because, as was true in many states, as each time the revenue estimators thought they had been conservative enough to have made what in their judgment was a “worst-case” forecast, the next forecast led to yet another report of a deteriorating revenue outlook.

Where did the models go wrong? Who's to blame for the forecast errors? How can the estimates improve in the future? These questions have been asked in the past but there are many features of the latest business cycle that make them harder to answer this time. This chapter looks at the recent history in the revenue-estimating context and attempts to draw some conclusions about the state of the science of revenue estimations.

 Revenue Estimation

Figure 19.1 State and local tax revenue

Source: US Census Bureau.

A look at key indicators reveals that the Great Recession has been very different than prior recessions.1 Of six key indicators—employment, wages, personal income, gross domestic product (GDP), consumer price index (CPI), and the Standard & Poor's 500 index (S&P500)—all six were contracting in 2009. In the previous two recessions, a maximum of two indicators contracted: employment and S&P500. (p. 498)

The last time all six contracted was in 1949. This had important implications for revenue estimating, particularly at the beginning of the recession:

  • The largest annual contraction in real GDP since 1946 had been less than 2 percent and there had not been more than one year of contraction in a row since the 1946 recession.

  • During the 2009 recession, real GDP contracted 0.3 percent in 2008 and 3.5 percent in 2009.2

  • Personal consumption expenditures (PCE) had never contracted two years in a row since the Great Depression and had not contracted for a single year since 1980.

  • During the 2009 recession, PCE declined two years in a row and the contraction in 2009 was the largest since 1942.3

  • Personal income (PI) had not contracted since 1949 and the wage component not since 1954.During the 2009 recession, PI contracted $197 billion from the fourth quarter 2007 to the second quarter 2009, or 1.6 percent, and wages contracted $245 billion, or 3.8 percent.4

  • Total nonfarm employment only contracted two years in a row twice since data collection started in 1939 and never more than the 3.6 percent, the contraction in 1945 after World War II ended.

  • During the 2009 recession, employment contracted 4.3 percent in 2009 and from 2007 to 2010 lost almost eight million private sector jobs and taking the employment level back to 2003. Employment fell three years in a row by 2010.5

  • Even the S&P 500, a notoriously volatile time series, had two consecutive years of decline worse than the bust that started in 2001.

  • From 2007 to 2009, the S&P500 dropped 36 percent compared to the 2000 to 2003 period when it dropped 32 percent.6

(p. 499)

As revenue estimators built models and analyzed data to compile revenue estimates for FY 2009 and FY 2010, the time-series data they relied on did not contain the information that would show the kind of downturn that was to occur.

In addition to the exogenous variables that drive the revenue estimates, history of collections is also used. Here again, collections provided no help in predicting the contraction in state and local revenues.

  • Since 1951, total state-level revenues have only contracted twice. In 2002, total state revenues contracted $24.4 billion, or 4.4 percent. The contraction in 2009 was $66 billion, or 8.5 percent.7 Forty-five states contracted in 2009.

  • Most revenue forecasts from the late summer and fall of 2007 for FY 2009 and FY 2010 were “pessimistic” at the time when compared to the high levels of growth that had just occurred. A slowdown was projected but not a big downturn.

  • In New Mexico, there were words of caution regarding the state of the economy and the estimate for personal income tax was for growth of 3 to 4 percent, “low by historic standards.”8

  • In Vermont, the economic outlook was revised down significantly but did not include a recession though the report did warn policymakers that “most recent estimates put Vermont's chances of two consecutive declines in real GSP at better than 50%.”9

The summer of 2008 was the end of the good times for revenue estimators across the states. Most states were just beginning fiscal year 2009 and were closing their books on one of the highest revenue years ever.10 All states had fully recovered from the 2001 recession, which had had a bigger impact on state budget experts than the taxpayers and workers. States had also built up what was thought to be adequate reserves.11 In fact, there was discussion in many states about the right level of reserves and whether the states were keeping too much revenue. By the end of FY06, states had built up reserves to 11.5 percent, up from 3.2 percent in FY 2003.12

When the revenues started turning and an official recession was declared, experience was still acting against accurate forecasts. Before December 2007, a handful of states had never been through a long recession. Since 1991, there were six states that had fewer than ten months (nonconsecutive) of contraction in employment prior to December 2007. Many of those states have now been in recession for one to two years. For example, New Mexico had only one month (June 1991) when employment declined from the prior year from January 1991 to November 2008, a period of 215 months; from December 2008 to August 2011, a period of 33 months, year-over-year employment declined for 31 of those months. When the first few months of contraction were reported at the end of 2008, it was simply implausible to consider a continued two-year string of declines that occurred.13

Personal income, likely found in more models than employment because of its stronger correlation with tax revenue, is even more complicated. From 1970 to the start of the recession at the end of 2007, a span of 152 quarters, only eighteen (p. 500) states had ever experienced a downturn.14 From the fourth quarter of 2007 to the third quarter of 2010, a span of eleven quarters, all fifty states and the District of Columbia experienced negative growth rates in personal income and the average number of quarters was 4.2 (table 19.2).

Table 19.1 States with fewer than 15 months of annual employment contraction before December 2007

January 1991 to November 2007 (203 Months)

% of Months

December 2007 to August 2011 (45 Months)

% of Months

New Mexico

























North Dakota















South Dakota















Note: The data are the number of months, consecutive or not, of year-over-year declines in seasonally adjusted nonagricultural payroll employment.

Source: BLS, author's calculations.

Unprecedented declines in all of the major variables made it a challenging environment in which to look to the future. In a report published by the Pew Center (p. 501) on the States in 2011, the median forecast error grew to over 10 percent in 2009, up from a median of 3.5 percent in prior years. The conclusion of the report is that the volatility of revenue estimates has increased in recent years and that “errors in revenue estimates have worsened progressively during the fiscal crises that have followed the past economic downturns. During the 1990–1992 revenue crisis, 25 percent of all state forecasts fell short by 5 percent or more. During the 2001–2003 revenue downturn, 45 percent of all state forecasts were off by 5 percent or more. And in 2009, fully 70 percent off all forecasts overestimated revenues by 5 percent or more.”15

Table 19.2 States and District of Columbia with at least one quarter of personal income contraction

Recession End

# of States













Source: US Bureau of Economic Analysis, author's calculations.

The Pew Center report concentrated on the revenue estimate at the time in the fall of 2007 for fiscal year 2009 and compared that with earlier data in order to frame the issue as an ominous trend in the accuracy of revenue estimates. However, a few missing elements added to the complexity. The variables noted above and many others—housing indicators, oil and natural gas prices, and interest rates—deviated from a fairly long history that formed the basis for revenue estimates.

Revenue Estimating Methods

There are as many methods of forecasting revenues as there are states but there are common features and a common sequence:

  1. 1. Adopt a forecast of the national economy.

  2. 2. Forecast the state economy.

  3. 3. Model and forecast revenues by using historical collections and exogenous forecasts of appropriate variables.

  4. 4. Compile recent changes in legislation.

  5. 5. Test and evaluate the completed forecast.

1. Adopt a forecast of the national economy

The starting point of a revenue estimate process is to evaluate the US economy. The interconnectedness of the states requires that attention be paid to the national economy and, in the case of states that have a higher concentration of exporting industries, the global economy. Typically, state governments rely on external contractors for this service.16 The largest and most widely used are IHS Global Insight (GI) and Moody's Analytics (also known as These vendors provide forecast data that are used in econometric models and narratives outlining the outlooks and the risks. In addition to a baseline forecast, they may provide scenarios and assign probabilities of their occurrence.17 These vendors also made presentations at conferences where state economists assemble, providing context to the forecasts being compiled.18

 Revenue Estimation

Figure 19.2 Moody's Analytics forecast of real GDP and actual

Source: Moody's Analytics, August 2007; US Bureau of Economic Analysis (BEA).

There were some warning signals in these forecasts but the baseline, or most plausible, forecasts from both Moody's and GI in late 2007 did not include a recession (figure 19.2).19 In the fall of 2008, the recession was predicted to be narrow because it was restricted to housing markets. The sentiment at the time was that (p. 502) housing market crises were localized and had a more limited impact on the national economy. At this point, only a few economists were waving the red flag that would be the financial crisis and reveal the web that lay beneath the mortgages.20

Another source of information, more as a benchmark, are the federal agencies charged with economic forecasts for federal budgeting purposes. The Office of Management and Budget (OMB) and the Congressional Budget Office both produce economic forecasts. Their track record was similar to the private forecasters (figure 19.3). When preparing the budget for FY 2009, the OMB and CBP were forecasting 4.7 to 5.1 percent growth in nominal GDP in FY 2009. One year later, the forecast was reduced to essentially zero growth, still missing the actual decline of 1.7 percent.

2. Forecast the state economy

 Revenue Estimation

Figure 19.3 OMB and CBO forecasts of 2009 GDP

Source: US Office of Management and Budget (OMB); US Congressional Budget Office (CBO).

The national forecast is subsequently used to generate a forecast of the local economy. States have a variety of approaches for determining the best forecast for the (p. 503) local economy. States with a consensus process for revenues will generally start by agreeing on a forecast for important economic variables. There are also states that have advisory groups that discuss and may agree on a forecast.21 In some states, a higher education institution has been charged with developing a forecast for the state's economy. These institutions have an appreciation developed by experience and research of the nuances of a particular geographic locale that help customize the forecast to the locale. Additionally, many of these institutions are integrated with government institutions by providing new graduates to work in the revenue and budget offices, by having experienced government staff as adjunct professors or by taking on special academic research that pertains to public policy. The universities also have connections in the private sector—for many of the same reasons—and may have a more direct connection to what is going on in the local economy that models built on national data would not be able to project

Some states also use national macroeconomic forecast vendors to compile forecasts of the local economy. This can either be an alternate forecast to compare to the locally developed one or used as the official forecast. Using a respected private vendor gives the imprimatur of independence that is sometimes helpful. Regionally, there are several private economic consulting firms that specialize in the geographic area. Forecasts can also be purchased for specific industries if that industry is sufficiently important that a general economic forecast will not capture the specialization. This is the case for states whose industrial makeup is far different than the national economy, such as Michigan (motor vehicles) or Kansas (agriculture) or Wyoming (energy). In these cases, specialty forecasts are included in the official forecast.22

Variable Selection. Regardless of the source, most revenue estimators rely on a handful of economic variables that, when combined, capture most taxable activity. Each revenue stream has a particularly meaningful or useful driver. The first criterion of variable selection is the availability of good data. There are some variables that are ideal for inclusion in a model but are difficult to collect, have too long a lag time in publication, do not occur frequently, or are difficult to forecast. It would be very useful to include GDP by state in revenue models, but the data are provided annually and are two years old by the time they are released.23 (p. 504)

Table 19.3 Examples of variables used to forecast revenues




Wages and Salaries


S&P500, Personal Income


S&P500, Corporate Profits

Sales Tax

Wages and Salaries, National Retail Sales

Motor Vehicle Excise

Car Registrations, Employment

Motor Fuel

Oil Price, Wages, and Salaries


Price at wellhead, Production volume

The withholding of personal income tax, for example, is correlated to wages and salaries or employment—these are direct measures of paychecks that are the source of withholding. Withholding makes up the lion's share of personal income-tax revenue. Nonwithholding is made up of final tax payments that are included with tax returns, estimated quarterly payments, and refunds of taxes. Because final payments and quarterly payments are often related to investment activity, the S&P500 is often used for the model. A reliable forecast of stock-market prices is a difficult proposition and is the cause of significant error in forecasting income-tax revenue. Also, the reliability of the S&P500 as an explanatory variable was brought into question in 2010 when it rebounded more than expected but nonwithholding continued to decline.24

The sales tax has historically been a stable revenue stream because it was thought to have a fairly straightforward relationship to employment.25 As long as the base was broad enough, regardless of economic conditions, people still had to buy things and that would keep up sales-tax revenues. In the two recessions prior to the 2007–2009 recession, sales growth rates slowed considerably but never contracted and then rebounded in the recovery.26 Beginning in 2009, the narrative changed and now there is a full year of contraction in the data. Further complicating conventional forecast tools, the environment of taxable sales is changing. Three trends have emerged that require attention: housing market exposure, a move away from goods toward services in the states' output, and the rise in Internet purchases.27

  • Housing market exposure. One of the most significant effects of the 2007–2009 recession was the magnitude of damage caused by the housing collapse. The complete shutdown of new housing construction caused an immediate loss of relatively high-paying construction jobs and an immediate loss of spending on construction materials. The former reduced both sales and income taxes and increased demand for public services, particularly unemployment compensation. The latter was felt mostly in lower sales-tax revenue. At the peak of the housing bubble, sales of building materials and furniture as a share of total retail sales (excluding food service) also peaked at 11.7 percent.28 It had been steadily growing since 2000 and was contributing more and more to the growth of retail sales. In 2007, building and furniture declined for the first time since the data began in 1992 and declined for three consecutive years. In 2010, these sales represented less than 10 percent of retail trade. This is exacerbated by mortgage equity withdrawal that was financing other consumption outside of the housing markets like automobiles and travel. In a discussion about risks to the forecast, the Minnesota Department of Finance in its November 2007 forecast directly confronts the challenges facing estimators: “Historically, increases in wealth, other things equal, have been found to produce small additions to consumer spending. But forecasters attempting to project the impact of the current decline in housing values on consumption find themselves in uncharted territory because there has not been a yearlong decline in housing values since the Great Depression.”29 (p. 505)

  • Services replacing goods. In the 1930s, goods made up 55 percent of personal consumption expenditures.30 By the 1970s, there was a shift to services as the major share of consumption and by 2009 the share of goods had dropped to 22 percent. This change in emphasis has decreased the efficiency of the sales tax for many states that exempt most or all services.31 Increasingly, states are attempting to add services to the base, starting with services that are less likely to be part of business-to-business transactions (thus to avoid pyramiding) but it has proved as difficult as any other type of tax increase.32 The FTA maintains a listing of the services that states include in their tax bases and over time there have not been significant moves toward broader bases.33 There has, however, been an increase in the rates, which drives up the costs of those items remaining in the base. This narrowing of the base makes the sales tax less efficient as a revenue generator.

  •  Revenue Estimation

    Figure 19.4 Goods versus services

    Source: US Bureau of Economic Analysis.

     Revenue Estimation

    Figure 19.5 E-commerce sales

    Source: US Census Bureau.

    Internet sales. The other trend is increased purchases over the Internet. This is not a new issue. For a state to collect tax revenue from a company, that company must have some nexus in the state.34 Sales over the Internet are comparable to sales made through catalogs—and like catalog sales they are generally exempt from sales tax unless the company has a physical presence. The increasing share of commerce that occurs over the Internet, however, is creating significant problems for sales-tax revenues. In 2003, the Census Bureau began keeping statistics about online commerce and its survey showed that about 1.8 percent of retail sales were through e-commerce. By 2008, the share had doubled to 3.6 percent. A significant amount of these sales are taxable due to the company's “brick and mortar” presence but there is a considerable amount that is not.35 A study from the University of Tennessee Center for Business and Economic Research projects that the total revenue loss to state and local governments is $12 billion in 2012.36 (p. 506)

Income taxes and sales taxes are the revenues that get the most attention from revenue estimators due to their relative importance to total revenue. The taxes and nontax revenue—fees, permits, fines, charges for service—that make up the remaining third of revenues also must be forecast.37 In many cases, such as fees and many permits, the models do not require external variables because the models are simple models. There are cases where it is helpful to aggregate different revenues and model as a group with an economic variable. Building permits and other construction-related revenues can be compiled into an aggregate, modeled using a construction-related variable that has a good forecast and then disaggregated for reporting purposes.

3. Model the data

To model revenue data appropriately, the historic collections have to be prepared for modeling. Collections data are often very “lumpy,” in that it is not as smooth as it should be because of problems associated with the collection process and the accounting processes. For example, in 2010, there was a tremendous snowstorm in the mid-Atlantic region that closed District of Columbia operations for several days. This closure resulted in delays in processing tax collections and so made January activity (which is reported in February) seem artificially low and February activity (reported in March) seem high due to the catch-up. These two months are called “paired opposite outliers” and should be averaged if the model is based on monthly data. Most forecast data are quarterly so this particular example would be corrected by transforming the monthly data into quarterly data.38

The estimator must also know the timing of collections and the accrual process used so that the collections data can be as closely matched to the selected explanatory variable as possible and also to select the appropriate frequency for modeling. Lags are used for this purpose to shift the data backward. In most jurisdictions, sales tax collected in the current month are required to be sent in by a certain date next month so the collections data are always one month off the activity month. Income-tax (p. 507) withholding follows the same pattern but nonwithholding does not. Estimated payments are quarterly but are generally equal distributions of prior year tax liability though they are adjusted as changes in a particular taxpayer's circumstances warrant. Because the actual data are more closely related to an annual series—last year's total tax liability—an annual model may be more desirable over a quarterly model.

Once the data are prepared for analysis, the next step is model selection. Selecting the model depends on the revenue stream. There are several types of models that are useful and they range from simple moving averages and trend models to more sophisticated linear regression and time-series models. It can be informative to model the same revenue using a simple and complex model as a way to validate the results. The model selection is also matched to the importance of the revenue. As with any activity, resources must be prioritized and the most attention paid to the largest revenues. In other words, the model sophistication increases with the importance of the revenue. Alternatively, as mentioned above, revenues may be aggregated and modeled using more sophisticated techniques than would be the case if they were modeled individually.

There are some revenues—fees and licenses, for example—that are required to be renewed or applied for. Simple models that calculate a multiyear average are often sufficient and can be adjusted for any increase in rates. There are other revenues that are so volatile and unpredictable that a simple average is often the best or even the only appropriate method. Examples are unclaimed property or estate taxes.39 A history of collections will indicate whether a simple model is appropriate.

A trend model is used when revenue follows a consistent trend. This is more sophisticated than a simple average because it allows for growth in the forecast. Revenues that increase with population or inflation can be modeled with a simple trend model, and trend models are useful as benchmarks to validate other more complicated models. Trend models are not useful in identifying turning points, unfortunately.

Some revenues, theoretically, can be modeled using an explicit ratio to another variable. For example, the sales-tax base is usually highly correlated to wages and salaries. A ratio model would apply the historic ratio to the forecast of wages and then apply the appropriate tax rate. This model uses an exogenous forecast as the primary driver.

Most estimators rely on econometric models for the most important revenues.40 The standard form is the original least squares (OLS) model modified for time-series analysis. Modification may be the inclusion of an autoregressive term to address serial correlation or using lags of the dependent variable (the tax revenue being modeled). There is an abundance of academic research on this level of model selection and there are infinite variations on the way to treat both the dependent variable and the independent or exogenous variables.41 Most estimators have software that will perform the most common tests and provide forecasts that are based on “best practices” in the field.

While model form is important, variable selection is critical. The model has to be easily understood and based on a theory about the movement of the revenue. Data mining for good “fits” may make a model's coefficients look robust but, when (p. 508) the model is used to forecast, the interactions have to be well understood. The estimator has to be able to explain why a particular variable is used and often to a public audience that is not composed of economists. An example is the use of dummy variables, which is a technique used to account for seasonality or regime shifting (such as when a new tax rate is applied). Dummy variables serve an important role in identifying shifts in the data (e.g., before and after a tax-rate increase, or summer sales of ice cream) but they should not be used as a substitute for variability. In cases where there is an observable outlier or spike in the data and there is a rationale for it, a dummy can be used to mitigate its effect on the model. An example of a case where there is justification for using a single period dummy is Microsoft: in the fourth quarter of 2004, Microsoft paid out a dividend that was large enough that it was noticeable in the quarterly data at the national level.42 In this case, because it was a one-time event, the model should reflect an adjustment for this quarter. There are other large dividend payouts by other companies but they occur with regularity so are already included in the base data.

As noted above, however, the forecast of explanatory variables upon which the data are being modeled is crucial. A simple model of total state-level revenue using wages and salaries as the primary driver has excellent model attributes and, as figure 19.6 shows, the modeled values closely track the actual values.43 This is the way the world looked in the fall of 2007. Figure 19.7 compares the forecast would have been using a fall 2007 forecast of wages and salaries with the actual outcome: a significant deviation. In this case, the forecast of the explanatory variable did not capture the downturn.

4. Change legislation, policy, and behavior

 Revenue Estimation

Figure 19.6 Fitted versus actual of total revenue model using wages

Source: US Bureau of Economic Analysis; author's calculations.

 Revenue Estimation

Figure 19.7 Forecast of total revenue model using wages

Source: US Bureau of Economic Analysis; author's calculations.

Once the variables are selected and forecast, the revenue estimators evaluate whether the base needs to be adjusted to reflect changes in tax law, tax policy, or taxpayer/tax-administrator behavior. There are not only recent changes in tax law (p. 509) but also changes that are phased-in or triggered by events that estimators have to be conscious of. At times, policymakers will not want to absorb the impact, positive or negative, of a particular policy in a single year and will instead phase it in over several years. A fiscal estimate that is generally calculated and reported at the time of passage must be reviewed as each phase is entered. Major reforms are often phased in to give taxpayers ample time to adjust; alternatively, they are phased out through sunsets.44 The danger is that these changes are reflected very slowly in the tax base that is used to forecast revenues and also are not usually reflected in the economic variables. The revenue estimator has to review all policies to determine if the impact is in the base or has to be added or removed. For example, a change in the sales-tax rate begins to show up in the base immediately but is not in the history that is used to calculate the estimation coefficients so it has to be added after the fact until the revenue estimator is comfortable that the model is capturing the rate change. One way to accommodate this is to model tax bases rather than tax collections and then any rate can be applied at any time.

According to an annual survey conducted by the National Association of State Budget Officers (NASBO), states other than Michigan enacted $1.6 billion of budget cuts in FY 2008.45 Forty-two states (including Michigan) enacted tax changes: more than half decreased either sales or personal income taxes, taxes that make up two-thirds of state revenue.

There are also other policy changes that are not legislated but that affect revenue collections. A concerted effort by the compliance division of the tax department increases revenues but may require an appropriation. Innovations in compliance such as bank-account attachments or reciprocal agreements with the Internal Revenue Service can also lead to more revenue. To the extent that a collection rate is either implicit or explicit in the revenue model, an increase in the rate has a direct impact on the revenue forecast. Similarly, an environment that discourages extensive and comprehensive auditing of taxpayer returns could lower the collection rate. These types of variables, however, are extremely difficult to (p. 510) quantify and so are not typically incorporated explicitly in the revenue forecast. It is important to be aware of collections activity because recent history of collections is often the strongest driver of the forecast.

Table 19.4 Tax legislation in 2008

NASBO Reported Changes

FY 2008 State Tax Revenue

Number of States Enacting Tax Changes








































Source: US National Association of State Budget Officers (NASBO); US Census Bureau.

A corollary of compliance is taxpayer behavior. The success of the US economy and the ability to fund federal, state, and local government services is the high compliance rate for the payment of major taxes, particularly income taxes. Most of the income tax is collected through the withholding from paychecks and employees have little control over that process. Without this established method of collecting taxes, the compliance rate could be more volatile as individuals weigh alternatives to paying a tax bill like businesses do. Particularly in times of economic contraction, taxpayers who control the amount of estimated taxes may decrease or delay the payment of taxes to improve cash flow despite penalties and interest. If the choice is not paying employees or shutting down and making an estimated tax payment, the taxpayer may choose the latter and risk the penalty with the goal of paying the tax and penalties when the environment improves.

In addition to taxpayer behavior around the payment of taxes, there is also avoidance behavior either through tax planning, illegal evasion, or substitution. Changes in tax law that affect high-income taxpayers, individual or business, may result in additional emphasis on tax planning to lower the taxpayers' taxable income. For example, including a service such as legal services in the sales-tax base may induce companies to hire lawyers as staff rather than hiring law firms, thus reducing the amount they spend on a taxable activity. Changes that increase the audit requirements of the tax administrators may increase opportunities to illegally evade the tax. Credits and deductions may be claimed with no supporting evidence if there is no mechanism for tracking the credits. Finally, taxpayers may simply stop the taxable activity to avoid the tax. An example is increasing Internet purchases in response to an increase of the sales tax. (p. 511)

5. Test the results

Once the models have been identified and run, the results have to go through another series of tests, outside of the econometric tests in the model output, that the revenue estimators must conduct: reasonableness, validation, and range of result. The irony of the past few years is that an accurate forecast of revenues in the fall of 2007 would likely have been rejected or modified because it would not have made sense given the history for the reasons outlined above.

Reasonableness. A test for reasonableness is simply asking the question: given the economic outlook and the policy environment, does this result make sense? In the last few years, the reasonableness test may have caused estimators to throw out results because the models wanted to return to a norm when the outlook was for continued contraction. A test for reasonableness may also be related to the particular characteristics of the revenue that would not be captured by a model. One place this occurred probably more than others was capital gains income. In the history of capital gains, a period of decline has been followed by a fairly sharp increase. This was the case with the stock market in the Great Recession. Following a decline of 30 percent in 2008, the S&P500 ended 20 percent higher in 2010. A model using the S&P500 would have shown a similar bounce-back. This is where the estimator might use expert judgment and reject the model output, knowing that there will be extraordinary losses or that at a certain point, the relationship between the stock market and capital gains no longer holds.46

Validation. Another test is a validation test. As mentioned, often a trend model or ratio model can be used to validate the results of a more complicated model. The validation is not to confirm the exact results but is more another test of reasonableness. A simple test may not point to the magnitude of a particular trend but it should point to the direction. If a simple test shows a negative direction and the model shows a positive direction, there must be something in the model data that is changing the course of the revenue and has to be explained or revised. Billy Hamilton, a frequent contributor to State Tax Notes, uses a simple benchmark to validate his analysis of sales-tax revenue: sales-tax growth should be about two to three times employment growth and so any model estimate can be compared to this rule of thumb.47

Range of result. Finally, econometric models will provide a confidence interval or band of probability results along with a point estimate. If there is concern that the forecast of the economic variable may be too optimistic, the estimator can pick the lower band and intentionally build in more pessimism to the model without disregarding the model output. The inability to provide a range estimate for revenues rather than a point estimate means that the band of uncertainty, the confidence interval, is not usually explicit in presentations and reports. Unfortunately, policymakers cannot appropriate in ranges but rather to a particular level so a point estimate is required.

The final step in the revenue-estimate process is to write the narrative and answer a set of questions about each revenue stream:

  1. 1. What are the primary drivers of the forecast and are they included in the economic outlook discussion? (p. 512)

  2. 2. How has the forecast changed since the last forecast?

  3. 3. Was the model reviewed? Did the model capture the latest data reasonably well?

  4. 4. What are the risks to this forecast?

Because the nature of estimating is uncertain by definition, it is important to include clues about the direction of uncertainty. The answers to these questions will help write the narrative and prepare for any questions regarding the forecast from policymakers or the public. As forecast accuracy is examined, looking back on these questions at the time that a revenue estimate is made provides additional depth. In the fall of 2007, there were many risks outlined that had to do with housing markets and energy prices but the financial crisis had not emerged yet and employment did not start falling until December 2007.

Revenue estimators are applied economists rather than theoretical economists and must balance the requirement of theoretical robustness with the ability to generate consistent and meaningful forecasts. In the final report, the revenue estimator must be able to explain how a forecast was arrived at and the more sophisticated the techniques, the more difficult the explanation. The unfortunate nature of revenue estimating is that, while it can be sheltered from political influence, the estimate is placed squarely in a political environment and the revenue estimator is often called upon to describe and explain the estimate in a public forum of noneconomists.

The recent years have provided a new set of information that in many ways conflict with historic data and add new extremes to the time series used for revenue estimating. The future of revenue estimating will have to take the new information and incorporate it into the existing models or develop new models. Not only are there new lows and new contractions to understand but also new data that were not fully appreciated before. Over the next few years, housing measures will be looked at much more closely as they relate to revenues and whether the collapse set up a new trend or reverted to a pre-bubble trend. For revenue estimators, the task of variable selection and forecast selection is harder because the traditional methods failed to capture the magnitude of the decline. In this environment, estimators will adapt by adding new models and new methods of evaluating models.


Bruce, Donald, William F. Fox, and LeAnn Luna (2009, August). “State and Local Government Sales Tax Revenue Losses from E-Commerce.” 511–518. (p. 515) Find this resource:

    Congressional Budget Office (2008, January). “The Budget and Economic Outlook: Fiscal Years 2008 to 2018.”Find this resource:

      Congressional Budget Office (2009, January). “The Budget and Economic Outlook: Fiscal Years 2009 to 2018.”Find this resource:

        Eckl, Cornia, and Jed Klee (2005). “Rainy day funds (budget stabilization, budget reserve funds)”., “In The Encyclopedia of Taxation and Tax Policy, edited by Joseph J. Cordes, Robert D. Ebel, and Jane G. Gravelle. Washington, DC: Urban Institute Press. 327–328.Find this resource:

          Geweke, John (2007, October 3). “Revenue Predictions,” memo from University of Iowa Institute for Economic Research to Revenue Estimating Conference, this resource:

            Hamilton, Billy (2011, January 24). “Worried Man: Trying to Predict the Turnaround in Tax Revenue.” State Tax Notes 59: 295–299.Find this resource:

              Kavet, Rockler and Associates (2008, January 16). “January 2008 Economic Review and Revenue Forecast Update.” Prepared for the State of Vermont Emergency Board and Legislative Joint Fiscal Committee.Find this resource:

                Lutz, Byron, Raven Molloy, and Hui Shan (2010). “The Housing Crisis and State and Local Tax Revenue: Five Channels.” Federal Reserve Board Finance and Economic Discussion Series, 2010–2049.Find this resource:

                  Maki, Wilbur R., Carlo del Ninno, and Peter L Stenburg (1980). “Forecasting State Economic Growth in Recession and Recovery.” Journal of Regional Analysis and Policy 13(2): 39–50.Find this resource:

                    Massachusetts Department of Revenue (2007, December 13). “Briefing Book: FY2009 Consensus Revenue Estimate Hearing.” Boston: Massachusetts Department of Revenue.Find this resource:

                      McCullen, Mark (2007, September 17–19). “Regional Economic Fallout.” Presentation to annual Federation of Tax Administrators’ revenue-estimating conference, Raleigh, North Carolina.Find this resource:

                        McNichol, Elizabeth, and Kwame Boadi (2011, February 3). “Why and How States Should Strengthen Their Rainy Day Funds.” Center on Budget and Policy Priorities.Find this resource:

                          Mier, Chris (2009, May 12–15). “The Current Environment in Financial Markets.” Presentation to the National Association of State Treasurers’ 2009 Treasury Management Conference, Atlanta, Georgia.Find this resource:

                            Minutes of the Delaware Economic and Financial Advisory Council (2008, September 15). Buena Vista.Find this resource:

                              National Association of State Budget Officers (NASBO) (2008). “Budget Process in the States, Summer 2008.” this resource:

                                National Association of State Budget Officers (NASBO) (2010). Annual Report on Fiscal Survey of the States. this resource:

                                  Nelson A. Rockefeller Institute of Government (current quarterly). State Revenue Report. Albany: State University of New York at Albany. this resource:

                                    New Mexico Department of Finance and Administration (2007, October 23). “General Fund Consensus Revenue.” Presentation to the New Mexico Legislature.Find this resource:

                                      Office of Management and Budget (OMB) (2008, February 4). “Budget of the United States Government: Fiscal Year 2009.”Find this resource:

                                        Urahn, Susan, and Thomas Gais (2011, March). “States’ Revenue Estimating: Cracks in the Crystal Ball.” Joint Project of Pew Charitable Trusts and Rockefeller Institute.Find this resource:

                                          US Bureau of Economic Analysis (BEA) (monthly and annual) National Income and Products Table. (p. 516) Find this resource:

                                            US Bureau of Labor Statistics (BLS) (monthly and annual) Employment Situation.Find this resource:

                                              US Census Bureau (2011). Annual Survey of State Tax Collections. this resource:

                                                Watson, Harry (2005), “Excess Burden.” In The Encyclopedia of Taxation and Tax Policy, edited by Joseph J. Cordes, Robert D. Ebel, and Jane G. Gravelle. Washington, DC: Urban Institute Press. 121–122.Find this resource:

                                                  Williams, Daniel W (2008). “Preparing Data For Forecasting.” In Government Budget Forecasting, edited by Jinping Sun and Thomas D. Lynch. Boca Raton, FL: CRC Press. Chapter 14.Find this resource:

                                                    Willoughby, Katherine, and Hai Guo (2008, March). “The State of the Art.” Government Budget Forecasting: 27–42.Find this resource:

                                             (2007). “WSJ Forecasting Survey—December 2007.”, accessed January 30, 2011.Find this resource:


                                                        (1) Mier (2009), slide 2. In a presentation to the National Association of State Treasurers, a strategist from Loop Capital Markets compared the conditions in the financial markets to the ten plagues of Egypt in the Bible.

                                                        (2) US Bureau of Economic Analysis (BEA).

                                                        (3) Ibid.

                                                        (4) Ibid.

                                                        (5) US Bureau of Labor Statistics.

                                                        (6) Yahoo! Finance (

                                                        (7) US Census Bureau, Annual Summary of State Taxes (last revised January 18, 2011). Note that this does not include the District of Columbia, which shows up in the local tax data: DC also contracted in 2009.

                                                        (8) New Mexico Department of Finance and Administration (2007) 10.

                                                        (9) Kavet (2008), 7.

                                                        (10) BEA and author calculations. In 2008 thirty-eight states and the District of Columbia had the highest revenues.

                                                        (11) Eckl and Klee (2005, 327–328) note that while there is no “ideal” amount that should be accumulated in a budget reserve (“rainy-day fund”), a combination of general fund surpluses and budget reserves should be at least 5 percent of total state expenditures.

                                                        (12) NASBO (2010).

                                                        (13) The author was the chief economist of the NM Legislative Finance Committee and participated in the consensus revenue forecast.

                                                        (14) BEA; the data are quarterly changes in four quarters of the moving average of personal income.

                                                        (15) Urahn and Gais (2011), 4.

                                                        (16) The larger states—California, New York, and Florida—may have resources to compile in-house forecasts of the national economy and use outside forecasts.

                                                        (17) In the fall of 2007, IHS Global Insight began increasing the probability of their pessimistic forecast.

                                                        (18) Every fall, the Federation of Tax Administrators convenes a revenue-estimating conference that includes tax-policy experts from many states as well as private and federal experts. The 2007 conference included presentations by both IHS Global Insight and that played down the probability of a broad recession.

                                                        (19) At the FTA conference, predicted that although a recession would be avoided, the probability increased from the low 20s to almost 40 percent, the highest probability since the 2001 recession.

                                                        (20) WSJ (2007). A Wall Street Journal forecast survey reported only three economists of fifty-four at the time projected a contraction in US GDP and only four thought that the probability of a recession was greater than 50 percent.

                                                        (21) NASBO (2008). Twenty-nine states have economic advisory groups, and twenty-eight use a consensus revenue process.

                                                        (22) Alaska, Colorado, New Mexico, Utah, and Wyoming all include forecasts of the price of oil and/or natural gas in the official forecasting reports; West Virginia includes the coal outlook; Kansas includes agricultural statistics in its forecast documents.

                                                        (23) The BEA released advanced estimates of 2009 GDP by state in November 2010 and then the final 2009 numbers plus revisions in July 2011.

                                                        (24) It's likely that losses from prior years are now offsetting gains and depressing nonwithholding.

                                                        (25) Hamilton (this volume).

                                                        (26) In the 2001 recession, sales-tax growth reported in the Census Annual Survey of State Tax Collections was 0.2 percent and grew 2.7 percent the following year. In 1991, the growth rate was 3.5 percent. Up-to-date quarterly state-by-state estimates are provided by the Nelson A. Rockefeller Institute of Government (quarterly at

                                                        (27) Fox (this volume).

                                                        (28) US Census Bureau Monthly Retail Trade (downloaded January 23, 2011).

                                                        (29) Minnesota Department of Finance (November 2007), forecast,

                                                        (30) US Bureau of Economic Analysis (BEA).

                                                        (31) Efficiency of a tax is defined as the ability to generate adequate revenue with minimal administration. A very efficient tax has the broadest base possible. See Watson (2005), 121–122.

                                                        (32) Pyramiding is when there are taxes applied throughout the production process rather than on the final sale.

                                                        (34) Nexus is usually defined as a physical presence through facilities or agents but the Internet has created new questions about what nexus is and is not. States have recently been trying to address this, because online sales become a much larger share of commerce. The Streamlined Sales Tax Initiative that has been joined at some level by every state that has a sales tax is an example. States are also challenging online retailers like eBay and Amazon or rewriting statutes to capture these sales.

                                                        (35) For example, a purchase made online at BestBuy is likely subject to sales tax because BestBuy has locations in every state but the identical purchase from would be exempt in most states.

                                                        (36) Bruce, Fox, and Luna (2009). To put it into context, the total decline in state sales-tax collections in 2009 was $12.9 billion, according to the Census Survey of Governments.

                                                        (37) Sjoquist and Stoycheva (this volume).

                                                        (38) Williams (2008), 351.

                                                        (39) It may be possible, using mortality models, to model the number of deaths but it is almost impossible to combine that with the wealth of the deceased!

                                                        (40) Some states also have microsimulation models for income taxes but these models are more useful for estimating the fiscal impacts of changes in parameters and not useful for forecasting. Similarly, input-output models (such as REMI) are generally not used for forecasting variables but for fiscal impact estimating.

                                                        (41) Willoughby and Guo (2008), 31.

                                                        (42) BEA Frequently Asked Questions. Answers can be found at

                                                        (43) The data are from a simple Cochrane-Orcutt model using GRETL econometric software where total state revenue is regressed against US wages and salaries and is for illustration only.

                                                        (44) The Joint Committee on Taxation issues a report every year of tax provisions that are expiring because of sunset provisions in the statutes.

                                                        (45) NASBO (2007). Michigan enacted $1.6 billion in tax increases because of the prolonged budget and economic problems there. Including Michigan would offset almost all of the other states' tax decreases.

                                                        (46) Losses are extremely difficult to model because (a) details about the magnitude are not well known, (b) they can be claimed on amended returns or have been carried forward from prior years and taken out of current year collections, and (c) modifying the way losses are treated is a popular form of tax relief, particularly for the federal government.

                                                        (47) Hamilton (this volume).