state revenue forecasts and political acceptance: th e ...faculty.cbpp.uaa.alaska.edu/afgjp/padm628...

John L. Mikesell is Chancellor’s

Professor of public and environmental

affairs at Indiana University. His published

research focuses on sales and property

taxation, budget systems, fi scal federalism,

and tax administration. His textbook Fiscal

Administration is widely used in graduate

public administration programs. He has

served on the Indiana Revenue Forecast

Technical Committee since the mid-1970s.

E-mail: [email protected]

Justin M. Ross is assistant professor of

public fi nance and economics in the School

of Public and Environmental Affairs at

Indiana University. Previously, he forecasted

state and local economic indicators while

working for the Bureau of Business and

Economic Research of West Virginia

University. His research focuses on local and

state public economics, with an emphasis

on empirical investigations of government

behaviors under alternative systems of tax

administration.

E-mail: [email protected]

State Revenue Forecasts and Political Acceptance: The Value of Consensus Forecasting in the Budget Process 1

Public Administration Review,

Vol. xx, Iss. xx, pp. xx–xx. © 2014 by

The American Society for Public Administration.

DOI: 10.1111/puar.12166.

John L. MikesellJustin M. Ross

Indiana University

Concerns about political biases in state revenue forecasts, as well as insuffi cient evidence that complex forecasts outperform naive algorithms, have resulted in a nearly universal call for depoliticization of forecasting. Th is article discusses revenue forecasting in the broader context of the political budget process and highlights the impor-tance of a forecast that is politically accepted—forecast accuracy is irrelevant if the budget process does not respect the forecast as a resource constraint. Th e authors pro-vide a case illustration in Indiana by showing how the politicized process contributed to forecast acceptance in the state budget over several decades. Th ey also present a counterfactual history of forecast errors that would have been produced by naive algorithms. In addition to show-ing that the Indiana process would have outperformed the naive approaches, the authors demonstrate that the path of naive forecast errors during recessions would be easily ignored by political actors.

Revenue forecasts play a critical role in the development of state budgets because they establish the resource baseline within which

expenditure programs must fall if operations are to be executable and sustainable. As a result, many research-ers have devoted attention to the revenue forecast-ing process and its outcomes along two dimensions. Th e fi rst literature approaches the revenue forecast as a technical problem for which one process might outperform another, either through the use of quali-fi ed public employees or increasingly sophisticated methodological models. Th is literature has typically found little to no relationship between the sophisti-cation of the revenue forecast methodology and its subsequent accuracy.1 Th e second dimension of the previous research is a political critique of the actors in the process and tests for corresponding systematic bias. Th is literature has generally found forecast errors to contain evidence of political bias and risk aversion stemming from political offi cials and the bureaucrats they infl uence. Th ese dual fi ndings generally motivate the recommendations for institutional reforms of the forecasting process toward simple “naive” forecast methods that would be no worse than more complex

approaches in terms of forecast error but presumably would be free of ideological bias of the actors neces-sary in causal methods.2 Th is article argues that when the forecasting process is viewed in its full institutional context, as one step in a budgetary process devoted to the allocation of revenue, this emphasis on accuracy alone is insuffi ciently narrow.

Fiscal sustainability requires that state lawmakers have a baseline forecast of the amount of revenue expected during the budget period. Th e selection of this base-line fi gure is of direct importance to meeting the goals of the political actors involved, and subsequently, there exists an incentive for these actors to infl u-ence the forecast toward producing fi gures that favor their objectives. Some may wish to restrict the size of state government by choosing the most conservative forecast assumptions; others may select optimistic assumptions that make additional spending programs or tax cuts appear more aff ordable. In a rational model of the entire budget process, however, these biases should be anticipated and accounted for by the actors whose spending choices are informed by the forecast. For example, members of a legislature could reason-ably infer that a governor who wants to cut spending might propose a budget built from more conservative forecast assumptions. Recent examples of these dis-putes during 2013 include New Jersey and California (Hamilton 2013; Megerian 2013).

An accurate and binding forecast serves fi scal sus-tainability by providing the hard budget constraint of resources available for allocation across public services without shifting the cost of programs to the future. An inaccurate forecast or a distrusted forecast both represent ways in which the forecast process can contribute to a violation of sustainability. Th ese dueling concerns over accuracy and perceived biases potentially require a forecast process that fi nds the method of minimizing error that is accepted by the actors, perhaps even managing trade-off s between accuracy and acceptance. For instance, a highly sophisticated forecast model may provide an excellent

State Revenue Forecasts and Political Acceptance: Th e Value of Consensus Forecasting in the Budget Process

2 Public Administration Review • xxxx | xxxx 2013

of these errors to the forecasting process. To some extent, state revenue forecasting is just an application of economic forecasting, and as a result, there has been some tendency for scholars to view the evidence from a broader set of literature than only state revenue forecasts (e.g., Th ompson and Gates 2007).3 A variety of statisti-cal techniques have been applied to numerous economic variables (e.g., stock prices, gross domestic product, fi rm sales, employment,

consumer sentiment, etc.) in part to deter-mine which approaches, a priori, might be the most reliable in minimizing forecast error.4 Th e most famous evaluations are probably from the M-Competitions that have repeat-edly pitted forecasts against one another in a tournament-style setting on common data (e.g., Makridakis and Hibon 2000). Th e

conclusion of this competition series, which replicates the general fi ndings across much of the forecasting literature, is that increasingly sophisticated forecasting techniques do not necessarily improve performance.

State revenue forecasts usually build from a causal model that links forecasted economic conditions and state tax revenue. Th e complexity of the causal model varies across states, but it involves determining a relationship between the economy, the tax base, and tax revenue.5 On the basis of previous research on methods and accuracy in forecasting, some have challenged this conceptualization as not being worth the eff ort. Kliesen and Th ornton (2012) demon-strate that the defi cit and debt projections prepared by the federal Congressional Budget Offi ce are simply no better than a random walk forecast, that is, using last year’s actual as a forecast for the next year.6 At the state level, Th ompson and Gates suggest, drawing on the literature of business and fi nance forecasts, the implementation of a simplifi ed approach to identifying state revenue growth: “the simple average of past growth rates is the best estimator of expected growth rates” (2007, 826). In sum, important parts of the forecast-ing literature question the value added by attempting complex causal modeling of the variable to be forecast and propose that naive models do just as well, if not better, and are less prone to interfer-ence with the forecast result.

Much of the existing literature also reports evidence of systematic bias in revenue forecast errors. One common fi nding is that state revenue forecasters systematically bias their forecasts downward, and a considerable stock of the existing academic research assumes that forecasters deliberately guard against criticism from governors and legislatures by conservatively underforecasting.7 Forecasters gener-ally work with a forecasting range (even if they do not provide that range in their presentations), and, as Rodgers and Joyce state, “It is seen as much too foolish to use the high-end forecast, risky to accept the best estimate forecast, and fi scally responsible to endorse the low-end forecast” (1996, 49). Th is underforecast is their fi nd-ing for all states from 1975 through 1992. Feenberg et al. (1989) fi nd similar underestimation in New Jersey, Massachusetts, and Maryland over the years from roughly the end of World War II to 1987. Heins (1975) fi nds a negative bias in forecasts over 58 years in nine states (Alabama, California, Delaware, Illinois, Indiana, New Jersey, New York, Rhode Island, and Wisconsin). Rose and Smith (2012) fi nd underestimation in a 47-state panel over 1986–2007. Williams (2012), looking at the work of fi ve diff erent entities that

fi t to the historical data, but it may also aff ord insiders the opportu-nity to employ model assumptions that suit a particular ideological preference in the forthcoming budget cycle. By contrast, a simple naive forecast that uses only the previous year’s actual collections is transparent and therefore hard to manipulate, but it also can pro-duce predictable enough errors to generate additional debates over whether certain programs are truly “aff ordable” or not.

Th is article is the fi rst to bring the concern for the political acceptance of revenue forecasting to the forefront. It proposes that forecasting must be observed and understood within the broader context of the budgeting deliberation, in which there are multiple points for various actors to revise away from the original base-line values. It is conceivable that an independent and depoliticized forecast committee could produce widely accepted revenue projec-tions, but we suggest that is a hasty recommendation: a depoliti-cized forecast does not ensure that there are no political gains from criticizing and rejecting the forecast. We similarly argue that naive algorithms for producing forecasts, which may or may not improve upon causal forecasting approaches, are relatively easy for budget actors to reject on the occasions when doing so produces political gains. Th e absence of a human stakeholder without a reputational or political interest in defending the process causes the forecast to be a politically inexpensive and unresponsive target.

For purposes of concreteness, we demonstrate these points by employing the state of Indiana as a case study. Indiana’s budget process is reasonably representative of other states, involving delib-eration within the legislature, between branches of government, and across competing political parties. Indiana is also interesting for the study of forecast acceptance because it has been recognized as having one of the nation’s most accurate forecasts and has a deliber-ately political consensus process. Th e next two sections expand our outline of the previous literature of state revenue forecasting and provide the appropriate background on Indiana. To advance the argument for a new research emphasis on trust and accuracy, the next section examines the performance of the actual Indiana consen-sus forecasts against the competing naive models. Th is comparison to naive models renders a clear image of why such models would produce revenue baselines that would go unaccepted by state actors, especially during business cycle peaks or troughs, which inevitably produce signifi cant forecast errors. Th e following section further demonstrates the importance of forecast acceptance by following the political history of the Indiana Revenue Forecast Technical Committee. Th is political history begins with the controversies that led to the inception of the committee, which was designed to garner political acceptance, and further allowed it to weather several politi-cal challenges levied through the Great Recession. In the conclusion, we suggest several important research questions for a new literature on forecast adoption into the budget as a trusted constraint, which is a topic that the fi eld of public administration is particularly well suited to address.

Previous Literature: Revenue Forecast Methods and ErrorsTh e academic literature on state revenue forecasting has largely focused on an ex post analysis of forecast errors and the relationship

Th is article is the fi rst to bring the concern for the political

acceptance of revenue forecast-ing to the forefront.


political interests. Jonas, Rest, and Atkinson (1992) fi nd that fore-cast errors in Virginia were more attributable to economic factors than poor forecasting, although much of the forecast was based on undocumented judgmental inputs. Krause and Douglas (2012) fi nd that forecast error carries a quadratic eff ect with respect to political diversity on the consensus forecasting panel, suggesting that there can be both too much or too little comprehensive political inclusion involved with forecast accuracy.

In summary, the research record demonstrates evidence of system-atically biased forecasting along political dimensions, as well as an absence of convincing proof that sophisticated methods that lack

transparency are actually delivering reduced errors. Furthermore, conditional on political actors being involved with the forecast, their biases are attenuated when there are compet-ing political ideologies involved. Th is article proposes the view that the forecast represents the beginning of a political bargain over what assumptions are to be used in determining the policies that will take place in the near-term budget cycle, and this suggests that the study

of revenue forecasting must expand to include an understanding of what elements of the forecast process make its resulting output the accepted baseline from which these determinations are made.

State Revenue Forecasting ProcessesTh is section summarizes, in general terms, the diff erent organi-zational approaches employed by states to forecast revenue. State revenue forecast processes diff er considerably in type and the extent to which they involve political actors in the budget process.8 Also, some forecast processes are implemented by state statute, while oth-ers are structured on the basis of informal agreement. As a matter of necessity, the choice of biennial or annual budgeting cycle directly aff ects the forecasting process by determining the frequency with which forecasts occur. Furthermore, state revenue forecasts tend to follow one of three general paths, and each is directly related to the overall structure of the corresponding process that the state uses to produce a budget.9

1. Both legislative and executive branch agencies independently prepare revenue forecasts. Legislative forecasts may come from a legislative research agency or the staff of a legislative committee. Executive forecasts may come from the state budget agency or the state revenue department. Identifying a hard budget constraint for development of a sustainable budget can be complicated when there are multiple forecasts in the process.10 In Wisconsin, the Department of Revenue prepares a forecast for the governor’s use in developing the budget, and the Legislative Fiscal Bureau prepares a forecast for use by the Joint Committee on Finance as it reviews the budget.11 These entities cooperate, and there typically is little difference in their numbers. In another example, the New Jersey Department of Treasury and the Offi ce of Legislative Services each does a forecast, and the differences between them have sometimes created considerable political tension.12

2. A state executive agency may prepare the revenue forecast. For instance, the Idaho Division of Financial Management prepares the forecast of revenue in that state, the Offi ce of Economic Analysis in

prepared forecasts for New York City in fi scal years 2003 through 2007, fi nds underforecasting that increases as the horizon extends further into the future for most revenue categories but overforecast-ing bias with the property tax. Krause, Lewis, and Douglas (2013) look at distinctions between legislative, executive, and independent commission eff ects in states from 1987 to 2008 and fi nd underfore-casting in most states that is more pronounced when legislatures are divided on a partisan basis as opposed to unifi ed party government. However, not all state-level studies fi nd general underforecasting. For instance, Mocan and Azad (1995) fi nd no systematic over- or underforecasting bias in legislative forecasts for 20 states over 1986–92. Similarly, Cooke and McIntosh (2011) attribute Idaho’s recent fi scal problems to overly optimistic general fund revenue forecasts. Boylan (2008) fi nds an important political infl uence: budgets in year ending right before an election or starting right before an election tend to be based on optimistic forecasts.

National government forecasts do not, how-ever, follow the conservative forecast pattern generally claimed for the states. A recent study of offi cial forecasts of real growth rates and budgetary balances among 33 countries fi nds systematic optimism bias (Frankel 2011). Similarly, Auerbach (1999) fi nds that forecasting performance by the Congressional Budget Offi ce, the Offi ce of Management and Budget, and a major private forecaster (Data Resources, Inc.) was generally of equal quality and showed no underlying bias in either direction. Blackley and DeBoer (1993) fi nd no evidence of bias in federal Offi ce of Management and Budget economic and revenue forecasts in the 1963–89 period (although underforecasting of outlays), leading to underprediction of defi cits. In an investigation of 12 Organisation for Economic Co-operation and Development countries over 10 years in the late 1990s to the late 2000s, Buettner and Kauder (2009) fi nd small errors, most often with slight underes-timation. However, the accuracy depended signifi cantly on the sheer number of taxes being levied and on the importance of income and corporate taxes in the total, both infl uences working to increase the error. Independence of the forecasting entity reduced the error rate.

Th ere has also been great interest in identifying the role of politi-cal ideology, or perhaps direct political interference, as a source of forecast error. As stated by Cassidy, Kamlet, and Nagin, “as revenue forecasts are such an integral component of state and local govern-ment budgeting processes, any positive theory of resource allocation in this area of government would be seriously incomplete without an accompanying theory of revenue forecast behavior” (1989, 321). Studies fi nd overforecasting in election years, underforecast-ing by governments from the Right and when there is a provincial antidefi cit law, and underforecasting when economic conditions are improving (Couture and Imbeau 2009); more accurate forecasts when those doing the forecasts are qualifi ed to do the job (Krause, Lewis, and Douglas 2006); independent legislative forecasts improve accuracy (Bretschneider and Gorr 1987); and a more negative forecast error when the state has a budget stabilization fund (Rose and Smith 2012). Shkurti and Winefordner (1989) describe an instance in Ohio in the late 1980s in which the competing politi-cal parties produced their own forecast estimates and accused one another of producing manipulated numbers that served their own

Th e forecast represents the beginning of a political bargain over what assumptions are to be used in determining the policies that will take place in the near-

term budget cycle.


Table 1 Features of Consensus State Revenue Forecasting Entities

State Forecast Group Forecast Committee Membership

Connecticut Offi ce of Policy and Management and Offi ce of Fiscal Analysis

Delaware Delaware Economic and Financial Advisory Council 35 members from academic and business community, executive branch, and legislatureFlorida Revenue Estimating Conference Three members from legislature (House, Senate, and joint Legislative Management Commit-

tee), one from governor’s offi ceHawaii Council on State Revenues Seven members: three appointed by governor, two appointed by president of Senate, two

appointed by speaker of the HouseIndiana Revenue Forecast Technical Committee Director of Tax Review from State Budget Agency, majority staff member from House Ways

and Means Committee, majority staff member from Senate Finance Committee, minority staff member from House Ways and Means Committee, minority staff member from Senate Finance Committee, tax economist from state university

Iowa Revenue Estimating Conference Governor (or designee), director of Legislative Services Agency, third person agreed to by other two members

Kansas Consensus Revenue Estimating Group Representatives of Division of Budget, Department of Revenue, Legislative Research Depart-ment, and one consulting economist each from University of Kansas, Kansas State Univer-sity, and Wichita State University

Kentucky Consensus Revenue Group Economists (fi ve to seven) jointly appointed by state budget director and Legislative Research Commission

Louisiana Revenue Estimating Conference Governor, president of the Senate, speaker of the House, or their respective designees, and a faculty member with revenue forecasting expertise from private or public university in state selected by the other three principal members from a list of as many as fi ve, but not less than three names, submitted to them by the Louisiana Higher Education Executive Advisory Council

Maine Consensus Economic Forecasting Commission (CEFC) and Revenue Forecasting Committee (RFC)

CEFC: members appointed by governor, Senate, and House; RFC: state budget offi cer, state tax assessor, state economist, university economist, director of Fiscal and Program Review Offi ce

Maryland Consensus Revenue Monitoring and Forecasting Group

Director of Bureau of Revenue Estimates, Deputy Comptroller, State Treasurer, Department of Budget and Management, Department of Transportation, and Offi ce of Policy Analysis of Department of Legislative Services

Massachusetts Secretary of Administration and Finance and House and Senate Committees on Ways and Means

Forecasts considered come from Department of Revenue and Massachusetts Taxpayers Foun-dation (private entity); hears testimony from other interested parties

Michigan Consensus Revenue Estimating Committee Director of House Fiscal Agency, director of Senate Fiscal Agency, and director of Department of Management and Budget or state Treasurer

Missouri Director of Revenue in consultation with Commis-sion of Administration, State Treasurer, House and Senate appropriation staff, Division of Budget and Planning, University of Missouri staff

Nebraska Nebraska Economic Forecasting Advisory Group Five appointed by Legislative Council, four appointed by governor (start with independent forecasts from Department of Revenue and Legislative Fiscal Offi ce). Members to have expertise in tax policy, economics, or economic forecasting

Nevada Economic Forum (EF) with Technical Advisory Com-mittee (TAC) on Future State Revenues

EF includes fi ve members who cannot be state government employees including publicly supported institutions of higher education. Governor appoints all fi ve, with Senate majority leader and speaker of Assembly each nominating one member for appointment. TAC includes Senate Fiscal Analyst, Assembly Fiscal Analyst, Chief of Budget Division of Department of Administration, Chief of Bureau of Research and Analysis in Department of Employment, Training, and Rehabilitation, vice chancellor for fi nance of university and com-munity college system, state demographer, and chairman of Committee on Local Govern-ment Finance

New Mexico Economists from Taxation and Revenue Depart-ment, Department of Finance and Administration, legislative Finance Committee, Highway and Transportation Department

New York Division of Budget Offi ce of Fiscal Affairs, Assembly Ways and Means, Senate Finance

Consensus forecast from Senate Finance Committee (majority and minority party representa-tion), Assembly Ways and Means Committee (majority and minority party representation), and Budget Director.

North Carolina Offi ce of State Budget and Management and Fiscal Research Division of Legislative Fiscal Services Offi ce

North Dakota Executive Budget Offi ce with tax and fi nance legisla-tors, legislative fi scal offi ce.

Rhode Island Consensus Revenue Estimating Conference Budget Offi cer, Senate Fiscal Advisor, and House Fiscal AdvisorSouth Carolina Board of Economic Advisors One appointment by governor to chair, one appointment by chair of Senate Finance Commit-

tee, one appointment by House Ways and Means, and designated representative of Depart-ment of Revenue and Taxation

Utah Revenue Assumptions Committee provide inputs to economists in Legislative Fiscal Offi ce, Offi ce of Planning and Budget, and State Tax Commission

RAC includes economists from Offi ce of Planning and Budget, Legislative Fiscal Analyst’s Of-fi ce, State Tax Commission, state agencies, and experts from academia and private sector

Vermont Consensus Revenue Forecast adopted by Emergency Board

Forecast prepared by state economist representing administration and state economist representing legislature. Emergency Board includes Governor and chairs of Senate Appropriations, Senate Finance, House Appropriations, and House Ways and Means.


leadership (e.g., Maryland and Maine). Only Indiana, Washington, and New York explicitly recognize potential political party differences by requiring representation from both political parties on the legislative side of the consensus group. In most of these arrangements, the members of the consensus group bring forward their independent forecasts, and, after discussion, fi nal forecasts for each revenue source are adopted.

As described in the previous section, the structure of the forecast process has been an important source of exploitable variation for empirical research interested in testing hypothesized sources of forecast error. However, the heterogeneity of the forecast process is likely an emergent feature of the state’s institutional context rather than exogenous random variables and very likely serves as a source of simultaneous causation that could bias regression results.

The Indiana Forecast and Budget ProcessAccording to a recent review by Boyd, Dadayan, and Ward (2011), Indiana managed to achieve a better than average level of forecasting accuracy over the 1987–2009 period, although, like all other states, it suff ered considerable inaccuracy during the Great Recession. Indiana levies all three major taxes common to state tax systems (individual and corporate income taxes and retail sales tax), and the Indiana economy generally tracks the same economic cycles as the nation as a whole.

Figure 1 outlines the basic structure of the forecast process,14 and fi gure 2 demonstrates the position that the revenue forecast occupies in the larger budgeting system. For additional context, fi gure 3 merges certain elements of these fi rst two fi gures to dem-onstrate the actors’ relationships in the overall budgeting process. Th e budget process in Indiana (fi gure 2), as it is in all states except Texas, consists of numerous exchanges between political parties and branches of government. Th e method of forecasting also follows a standard pattern for creating the revenue forecast: a national eco-nomic forecast drives a state economic forecast (fi gure 1). Th e state economic forecast then drives the forecast for individual state taxes, based on an econometrically derived relationship between elements of the state economy and each tax. Th ese relationships are delivered

the Department of Administrative Services prepares the Oregon forecast, and the elected state comptroller prepares the Texas forecast. Rubin, Peters, and Mantell (1999) identify 23 states with single-agency revenue forecasting in their survey of state methods in the late 1990s.

3. A consensus approach of some variety. The National Association of State Budget Offi cers defi nes this approach as a “revenue projection developed in agreement through an offi cial forecasting group representing both the executive and the legislative branches” (2008, 105). The underlying objective of the method is to produce a single forecast that emerges from cooperative work between legislative and executive branches, thereby producing a reliable and accepted budget constraint for development and adoption of state expenditure programs. Table 1 shows that some manner of executive–legislative consensus, including use of a formal consensus group in many states, produces the revenue forecast in slightly more than half the states. Qiao observes that states use this approach “in the hope of improved revenue forecast accuracy and providing the executive and legislative branches a common ground about the amount of revenue available for the budget” (2008, 394). However, the table shows great variation in the specifi cs of the process. The group size ranges from two or three principals in Connecticut and Iowa to 35 in Delaware, while some are indefi nitely large in other states. For instance, Louisiana law establishes a Revenue Estimating Conference with voting members that include “the Governor, the president of the Senate, the speaker of the House of Representatives, or their respective representatives, and a faculty member with the revenue forecasting expertise from a public or private university in the state.”13 Michigan generates its forecast through a conference that includes the state budget director or the state treasurer (recently, the treasurer has been the executive representative), the director of the house fi scal agency, and the director of the senate fi scal agency (Ross 2001). But Kentucky forms its forecasting group entirely with professional economists appointed by the budget director and the Legislative Research Commission, and part of the Nevada structure requires that all appointees be from outside the public sector. The legislative representation comes sometimes from a nonpartisan legislative research organization or its appointees (e.g., Nebraska and Utah) and sometimes from the elected legislative

Table 1 Continued

State Forecast Group Forecast Committee Membership

Virginia Advisory Council on Revenue Estimates with Joint Advisory Board of Economists

JABE: Secretary of Finance, Staff Director of House Committee on Appropriations, Staff Direc-tor of Senate Committee on Finance, and 15 nonlegislative citizen members, 12 appointed by the governor and three appointed by Joint Rules Committee (all economists from private or public sector). ACRE: Governor chairs with Speaker and Majority Leader of House, Presi-dent pro tempore and Majority Leader of Senate, Chairs of House Committee on Appro-priations, House Committee on Finance, Senate Committee on Finance, two members of House appointed by Speaker, two members of Senate appointed by Chair of Senate Finance Committee, and 15 to 20 nonlegislative citizen members appointed by Governor.

Washington Economic and Revenue Forecast Council Two members appointed by Governor, two members from each caucus of Senate, two mem-bers from each caucus of House.

Wyoming Consensus Revenue Estimating Group Legislative Service Offi ce, Economics Analysis Administration, representatives from state auditor and state treasurer, superintendent of public education, director of Department of Revenue, state geologist, oil and gas commissioner, economics professor from University of Wyoming

Source: Authors’ compilation from state statutes and regulations and from correspondence with individual state offi cials.

Only Indiana, Washington, and New York explicitly recognize potential political party diff er-ences by requiring representa-tion from both political parties

on the legislative side of the consensus group.


House of Representatives. Th e House version of the bill is passed to the Senate for further debate and deliberation. After passing the Senate, the bill goes to a conference committee to reconcile diff er-ences between the House and Senate versions of the bill. Behind all legislative deliberations lurks the revenue forecast as a constraint on expenditures. Th e fi nal bill is returned to the governor, who must either sign or veto the bill in its entirety. Th e initial revenue fore-cast that was originally produced by the forecast committee may be disregarded by any actor at any point in the budget deliberation process.

to the Indiana State Budget Committee, which produces the fi nal numbers for all fi nancial resources and produces a set of recom-mendations on spending that is informed by the State Budget Agency’s review and analysis of programs that were requested by the various state agencies. Th e recommendations are delivered to the governor, but they are only recommendations, and the governor is under no obligation to accept any aspect of this initial budget report. Th e governor then delivers both a fi nal budget report along with a proposed budget bill to the General Assembly, the items and appropriations of which are taken as the basis of debate in the

Figure 1 Indiana’s Two-Step Forecast Process

Source: Authors’ representation of the process described at http://www.in.gov/sba/2372.htm.

Figure 2 Indiana Budget Process


Th e RFTC is not a product of state statute, which is not especially unusual, considering that there are 15 states without a statutory assignment of forecasting responsibilities.16 Indiana state code (IC 2-5-1.1–8) requires only that the budget committee of the State Budget Agency be able to provide the legislative council an estimate of revenues and “any other data which will enhance an understand-ing of the fi scal aff airs of the state.” Because the RFTC is an infor-mal institution, there are no formal rules to the membership of the RFTC, but the practiced norm has been to include a representative from the State Budget Agency (the head of the Tax Review Offi ce), House Republicans, House Democrats, Senate Republicans, and Senate Democrats, plus a tax economist from a state university as a completely neutral expert. Th is process for forecasting has been in place for about 35 years without signifi cant change.

Notably, the affi liations of the members are representative of the groups that will be involved in writing the budget as it progresses through the legislative process. Th e tax economist, having been on the RFTC for more than three decades, serves to maintain consist-ency of process and anchor it to the most relevant academic evidence on tax systems and their forecasting. Th e members of the commit-tee do not admit to any desire to manipulate the forecast, and they subsequently do not view the tax economist as a check on the process for that purpose. When originally formed, the appeal of having a university tax economist was that it granted them access to a com-puter that could be used to conduct the econometric analysis. More recently, the tax economist’s role shifted to suggesting alternatives for the budget agency to explore and then assisting the committee in selecting the fi nal methodology to be adopted.17 Representatives from the legislature are staff of the Senate Finance and House Ways

Th e fi rst element in the revenue forecast is an economic fore-cast.15 Originally, selected elements of the state and national macroeconomy were forecast by a group called the Economic Forecast Committee (EFC). Th e EFC was an ad hoc group of volunteers from private sector businesses and academic econo-mists from around the state that prepared forecasts of key macr-oeconomic variables (gross domestic product, infl ation, personal income, state personal income) for the upcoming two budget years. After many years of service, however, the regular con-tributors to the group retired, and no replacements were found. Since 2007, the economic forecast has been purchased from an outside consulting fi rm (IHS Global Insight). With that shift, many more macroeconomic variables have become available and are employed for driving the revenue equation. No formal rules, however, dictate that the macroeconomic forecast is separately produced from the Indiana revenue forecast, but that is the prac-ticed norm. Th e budget for this purchase comes from the State Budget Agency.

Th e second and more critical component of the system is the Revenue Forecast Technical Committee (RFTC). Th e RFTC creates a methodology that translates the macroeconomic forecast into the revenue baseline in a transparent manner, and the RFTC operates under no constraints in regard to how it manages that process. Th e RFTC forecasts several tax components, including the general sales tax, individual income tax, corporate income tax, and revenues from gaming, to produce the total revenue forecast. Th e sales and individual income taxes are the most signifi cant revenue contribu-tors, representing more than 70 percent to 80 percent of the total revenues being forecasted.

Note: Solid arrows depict the same process relationship as in fi gure 2. Dashed arrows depict the source of appointed members (e.g., State Budget Committee members are appointed by the governor and each party within each chamber of the General Assembly).

Figure 3 Indiana Budget Process with Actor Linkages


macroeconomic variables are presented on the State Budget Agency Web site, and a forecast accuracy report is published monthly. Th e RFTC faces no restrictions in regard to how it prepares its fore-cast, using any approach and/or variables to drive the forecast. Th e RFTC has a huge array of options at its disposal: economic variables used to drive the forecast, functional forms used for forecasting equations, lags in the link between economic and revenue variables in equations, number of years to use in developing the estimating equations, even whether to use extrapolation approaches rather than economic drive variables in the forecast. Th e sole expectation is that the forecast methodology will be presented to the State Budget Committee in a transparent fashion so that any interested party could replicate the methodology employed to produce the forecast.

Indiana is one of 19 states that operate on a biennial budget cycle (Snell 2011). In the legislative session of each odd-numbered year, a budget is adopted for the next two fi scal years. Th erefore, in December of each even-numbered year, the RFTC prepares a revenue forecast for each of the next two fi scal years starting the next July—that is, the fi scal year that begins roughly six months ahead and the fi scal year beginning roughly 18 months ahead. Th is is the critical revenue forecast because it is the one on which the budget for both years is going to be developed, and hence it provides the hard budget constraint that is necessary for fi scal sustainability. Th ere is an April update prepared in odd years, to provide the legislature one last view of revenue prospects for the budget biennium before appropria-tions are fi nal, but the December forecast is the critical one because, by April, the General Assembly has completed most of the political bargaining, and major adjustments are not likely.

Th e RFTC also prepares a biennial forecast review in December of odd-numbered years. In this review, the committee considers execution of the biennial revenue forecast fi ve months into the biennium. It adjusts its forecast if that seems appropriate, including reestimation of its forecasting equations, and presents the results of that execution review to the State Budget Committee. Usually, no actions are taken as a result of this report because appropriations for the full biennium are already in place.20

To track forecast accuracy, the annual forecast is broken into monthly segments based on historical monthly fl ow profi les, and there is a monthly comparison of year-to-date actual revenues against revenue levels that would be consistent with meeting the annual forecast. Th e report is posted on the budget agency Web site, a news release is provided, and the news media report on the results when there are signifi cant errors.

Forecast Committee versus the Naive Approaches: Forecast Error as a Barrier to AcceptanceOne of the common recommendations from the literature on forecast accuracy emerges from the fi nding that, on average, “naive” forecast methods produce similarly accurate estimates as human-based committees. Th is section begins by comparing realized errors from the Indiana forecast committee with naive alternates and then maps out the evolution of the errors over business cycles for both approaches. It is argued that the predictability of the errors made by naive approaches represents a reasonable basis for political actors to ignore and disregard forecasts during the peaks and troughs of the business cycle.

and Means committees. Th is allows turnover of committee member-ship to be low because a change in political control of one house of the General Assembly does not mean that representation on the RFTC will change (four of the six members in 2013 had fi ve or more years experience on the committee). Th e committee is also assisted by tax experts from the Legislative Services Agency and the Department of Revenue on an as-needed basis. Th e committee is expected to pro-duce a methodology report that will provide step-by-step instructions for converting the macroeconomic forecasts into a revenue baseline for each major tax and for total general fund revenue.

Th e forecast is based on current law, but it also includes the esti-mated impact of any changes to law that will go into eff ect during the forecast period, even though the eff ect is not yet in the historical revenue stream.18 In the preparation of the methodology, the RFTC reviews the forecasting equations employed for the year and analyzes errors to determine to what extent they emerged because of prob-lems with the revenue forecast model as opposed to random fl uctua-tions from the macroeconomic forecast. Th e methodology is always reviewed, models are always reestimated, and macroeconomic drive variables are often changed. Th e methodology is expected to be agreed upon before the RFTC knows the macroeconomic forecast, further establishing independence between the two elements. Th e RFTC is not known for voting on anything, except sometimes the scheduling of future meetings.

Table 1 provides a brief summary of the revenue forecasting struc-ture used by every state that employs a consensus forecast process. Th e Indiana process diff ers from that in most other states in two important respects. First, the RFTC, in addition to bringing execu-tive and legislative branches together to create the forecast, directly guarantees that members of both the Republican and Democratic parties will be involved. Most states include representatives from both the executive and legislative branches in a consensus com-mittee, but party only diff ers when there is division of control by the parties. After the 2012 elections, in 38 states, the same politi-cal party controlled both houses of the legislature as well as the governorship (Nelson 2012). In that environment, it is not hard to imagine that a legislative–executive consensus could be tinged with the political agenda of one party, which may be furthered by an artifi cially high or low forecast in the manner supported by the research of Krause, Lewis, and Douglas (2013). Second, the fore-casting group does not merely bring together competing forecasts developed by the representatives on the committee but also involves the forecasting group in developing the methodology employed to create the forecast. In contrast, for instance, the Michigan process produces a consensus from alternative economic and revenue fore-casts from the House Fiscal Agency, the Senate Fiscal Agency, and the state treasurer, with each independently developed. In Kentucky, the Governor’s Offi ce of Policy Management applies its forecasting model to national and state economic forecasts and presents the results to a consensus group of economic advisors for review and revision. Th ese approaches are considerably diff erent from the devel-opment process of the Indiana system. Formally, it is the method employed to create the forecast that is the consensus reached by the Indiana RFTC rather than the value.19

Th e methodology (the fi nal equations developed to translate the macroeconomy into revenue fl ows) and the IHS Global Insight


Using F to refer to the forecasted value, A the actual value, and t a time indicator, these four alternative models include the following:

1. Simple lag: the forecast is the most recently observed value.22

Ft = At–1

2. Trend: Th e forecast is based on a simple trend of the histori-cally observed values.

Ft = β0 + β1t

3. Autoregressive model: Th e forecast is based on a time-series regression with an intercept and lagged dependent variable as the determining indicators.

Ft = δ + ρAt–1

4. Exponential smoothing with trend: An exponential smooth-ing model that adjusts for a trend in the data. Here, α and β represent smoothing param-eters for the forecast error and trend estimate in the previous period.

Ft = FEt + Tt

FEt = αAt–1 + (1 – α)(Ft–1 – At–1)

Tt = β(FEt – FEt–1) + (1 – β)(Tt–1)

In the autoregressive and trend forecast mod-els, the coeffi cients (β0, β1, ρ, δ) are estimated using linear regression techniques based on the historical data that were available at the time the state’s revenue forecast was performed.

Likewise, the α and β parameters in the exponential smoothing approach result from a calibration exercise in which all possible com-binations from zero to one are used to predict over the contempora-neously appropriate historical data, and the parameters that provide the best historical fi t are used to generate the forecast values.

In a review of all state revenue forecast errors by Boyd, Dadayan, and Ward (2011), the Indiana state forecast was found to be among the most accurate in the nation as measured by single-year forecast errors. As the authors of the study note, it is diffi cult to truly compare the accuracy of forecasts across states because states levy diff erent taxes and operate in diff erent economies. For the purpose of this article, we propose to infer the “value added” of the revenue forecast process in Indiana by comparing it against four alternative naive forecast models that can be created on simple decision rules. Th ese models are often suggested by critics of contemporary forecast approaches using causal models and forecast committees.21

Th e analysis compares the actual Indiana forecasts against the forecasts that would have been produced from these alternative models, using exactly the data that would have been available when the revenue baseline would have been developed to guide the adoption of budgets. Because Indiana uses a biennial budget system, the primary test involves forecasts done for two budget years, one starting roughly six months after the date when the forecast is presented and one starting roughly 18 months after. Th is forecast date—December of even-numbered years—is just before the legislative session in which the biennial budget that would be based on that revenue would begin. Th is is the critical forecast time because this is the point at which the revenue baseline establishes the hard budget constraint for developing an executable budget and consideration of both the one-year-ahead and two-year-ahead forecasts prepared at that date provides the appropriate evaluation of the forecast for utility in the budget process. Reviewing a series of one-year-ahead forecasts, the standard approach used in the revenue forecast literature does not provide the appropriate test for biennial budget states.

Table 2 Mean Absolute Percentage Error and Value Added by Revenue Source and Forecast Method

Mean Absolute Percent Error State Value-Added Over Naive

Number of Forecasts Forecast Outlook Length State Forecast Simple Trend AR(1) Exp Smooth w/Trend Simple Trend AR(1) Exp Smooth w/Trend

Sales Sales12 One-Year 3.2% 9.6% 4.3% 4.9% 5.3% 6.4% 1.1% 1.7% 2.0%11 Two-Year 3.9% 14.9% 6.1% 4.9% 6.3% 11.1% 2.3% 1.0% 2.5%

CIT CIT

12 One-Year 12.7% 17.9% 19.3% 18.9% 26.1% 5.2% 6.6% 6.2% 13.4%11 Two-Year 18.5% 17.8% 19.3% 18.3% 21.3% –0.6% 0.8% –0.2% 2.9%

PIT PIT

12 One-Year 6.2% 11.9% 7.2% 8.7% 11.0% 5.7% 1.0% 2.6% 4.8%11 Two-Year 7.8% 13.5% 6.7% 10.0% 10.0% 5.7% –1.1% 2.2% 2.2%

Gaming Gaming

4 One-Year 8.6% 9.7% 17.9% 21.7% 17.6% 1.1% 9.3% 13.1% 9.0%3 Two-Year 14.0% 12.4% 22.6% 39.8% 20.2% –1.6% 8.6% 25.8% 6.2%

Other Other

12 One-Year 12.8% 11.5% 9.5% 15.9% 24.4% –1.4% –3.3% 3.1% 11.6%11 Two-Year 16.3% 16.6% 13.9% 20.7% 30.4% 0.4% –2.4% 4.4% 14.1%

Total Total

12 One-Year 3.5% 9.5% 5.7% 5.8% 6.6% 6.0% 2.3% 2.3% 3.1%11 Two-Year 4.5% 13.3% 5.4% 6.5% 6.3% 8.8% 0.9% 2.0% 1.8%

Note: “State Value Added over Naive” is calculated as (MAPENaive – MAPEState).

Th e analysis compares the actual Indiana forecasts against the

forecasts that would have been produced from these alternative models, using exactly the data that would have been avail-

able when the revenue baseline would have been developed to guide the adoption of budgets.


revenue forecast beats all other methods universally in predicting sales taxes. In the personal income tax, the revenue forecast only has a negative value added when compared to the trend’s forecast for two years ahead. For the corporate income tax, the revenue forecast is only outperformed by the simple and autoregressive forecasts in the two-year-ahead outlook. Th e simple forecast at two years ahead is the only instance in which the state forecast provided negative value added for gaming revenue, although this is a relatively new revenue instrument. For the remaining miscellaneous revenues agglomerated into the “Other” revenue category, both the one- and two-year state revenue forecasts hold the median MAPEs in the middle of the road relative to the other approaches.

In summary, an ex post evaluation of the MAPE statistics does not indicate that any other naive forecast approach would systematically outperform, or even match the performance of, that of the state rev-enue forecast committee. By this metric, the committee adds value in technical accuracy to the budget process beyond what a naive approach could provide state legislators.

Forecast BiasTh e previous literature has suggested that particular biases may exist that incentivize forecast committees to over- or underforecast rev-enues and that this may be attributable to risk aversion arising from a principal–agent problem or from ideological bias. It is worthwhile, once again, to consider the offi cial revenue forecast in this dimen-sion and compare it to the naive approaches. A state that presum-ably had unexpected good or bad fortune relative to their historical performance may arbitrarily imply a positive or negative forecast bias, so, once again, comparison against the naive approaches pro-vides a reasonable basis for value added. To evaluate this considera-tion, the mean percentage error (MPE) is computed for each of the two forecast periods to depict the level of bias:

MPE = ∑t=1

T

Ft – At ______ At

If revenues are routinely underforecasted, then the actual realized revenues will be greater than the forecasted values, and the MPE will become negative.25 Routinely overforecasting revenues would imply positive MPE statistics, with a bias-free forecast being one that approaches zero. For the state forecast’s value added, we calculate the diff erence of the MPE’s absolute value from that of the naive method.26

A positive value added outcome under this calculation occurs when the state’s forecast carries a MPE that is closer to zero than the naive counterpart, and, as a result, it embeds a normative assumption that overforecasting is no better, or worse, than underforecasting.

Table 3 indicates that the one-year forecast of the state was nearly zero in the MPE for the

one-year forecast and slightly underforecasted revenues at two years with an MPE of –0.6 percent.27 In addition to the MPE values, the individual forecast values tend to be as likely to be positive as they are negative. Th e individual total revenue forecast errors by the state were nearly evenly split between over- and underestimates, with 12 of the 23 periods forecasting less revenue than actually realized (not shown in the table). Th e simple look-back forecast was the only one-year MPE that underforecasted, whereas underforecasting was

Th e tests that follow work with the fi rst offi cial biennium Indiana state forecast, which predicted fi scal years 1989 and 1990 using data from fi scal years 1978 to 1988.23 Th e forecast generates estimates for the future values of tax revenues from personal income, corporate income, sales, other revenues, and by aggregation, it produces the total tax revenue. Th e forecast was extended to include the newly introduced revenues from gaming starting in fi scal year 2004. In each period, all methods have their parameter estimates updated with any new information that would have been available to the Indiana forecast committee at that time so that the information set is held constant in evaluating the error.

Comparison of Forecast AccuracyTh e common measure of forecast error is the mean absolute percent-age error (MAPE), computed for both one and two periods ahead:

MAPE = ∑t=1

T

|Ft – At| ______ At

Using actual observed revenues allows for all of the MAPE esti-mates to have the same denominator in all forecast methods.24 Th e one- and two-year MAPE statistics are calculated for the six revenue estimates for the fi ve forecasts, for a total of 60 results reported in table 2. Also provided in table 2 is the diff erence in error of the naive method against the state forecast, (MAPENaive – MAPEState), which is referred to as the “value added” comparison. Th is calcula-tion takes a positive value when the state’s MAPE is smaller than the naive alternative, giving rise to our normative interpretation of “positive value added.”

First examining the absolute performances, one would expect forecasts for the second year of the biennium tend to have larger MAPEs than their shorter, one-year counterpart. Among total revenues, all of the forecast methods have a MAPE under 10 percent for the one-year forecast, and only the simple forecast breaks above that level in the two-year forecast. Th e poorest-performing method is the simple look-back, whose MAPE is 9.5 percent for the one-year forecast and 13.3 percent for the two-year forecast. Within the diff erent types of revenue sources, it is the sales tax that is broadly the easiest to predict, at least in terms of achieving low errors. Among the major broad-based taxes, the corporate income tax has the largest errors across all forecast approaches.

Our proposal is to consider the value added of the state revenue forecast committee in its abil-ity to outperform the other naive methods. In terms of total revenue, the state forecast’s value added score for total revenue is positive against all other methods for both time horizons. Th e diff erence between the state revenue forecast and these naive approaches would imply that the value added of offi cial forecast committee is in the neighborhood of 2.3 percent to 6 percent for the one-year forecast and 0.9 percent to 8.8 percent for the two-year forecast.

Th e total revenue forecast is arguably the most critical component because it represents the bottom line in terms of total available resources, but because revenues can be earmarked according to particular uses, it is worthwhile to examine the value added of these instruments as well. Among these diff erent instruments, the state

Our proposal is to consider the value added of the state revenue forecast committee in its ability to outperform the other naive

methods.


behavior of the forecasters or an artifact of the state’s recent history of business cycles. Looking at total revenues, the ability to provide a relatively unbiased forecast is indicated by a positive value added against the naive alternatives, with the lone exception being the AR(1) model’s two year-forecast, which was relatively less biased by –0.2 percentage points. Th e pattern generally holds out across the subcategories, as there does not seem to be any approach on any particular revenue source that can provide relatively less bias than that of the state’s forecast.

Accepting Naive ModelsTh e analysis of table 3 shows no evidence of a systematic bias in the state revenue forecast, nor is a bias reliably produced in any of the naive methods. Figure 4 provides a graphical comparison of the best-performing naive forecast (the autoregressive model) against the RFTC forecast and the actual collections, providing a more compre-hensive look at the nature of the forecast errors than what might be ascertained from the MAPE. During the expansionary period of the mid- to late 1990s, the RFTC tended to slightly underestimate the actual revenue collected, while it tended to overestimate collections in contractionary periods surrounding recessions.

While both forecasts miss the recessions and subsequently produce large errors, the naive approaches tend to follow these dips and spread them into future periods, whereas the technical committee accommodates them into their statistical model for the recovery

period. Th is is likely a signifi cant aspect of forecast adoption into the budget constraint, as a naive lag model would be predictably low in the aftermath of a recession, and it would be reasonable for elected representatives to ignore it for their own individual estimates of what revenues will actually exist. In other words, it is reasonable in volatile times for the participants in the budget process to distrust the forecasts produced by naive methods. Furthermore, criticizing a naive forecast

universal in the two-year forecast. It would seem that there is no particular proclivity revealed in the data to suggest that the forecast committee is systematically under- or overforecasting, nor is it clear that any particular method would likewise produce that result, with the only exception being the simple-look back. Naturally, any revenue stream with positive growth will have a tendency toward negative MPEs under the simple-look back forecast. Among the sub-components of the revenue forecasts, there are no clear patterns to suggest that any particular revenue source produces a systematic over or under bias.

One of the advantages of these comparisons in table 3 is that they should be indicative of whether systematic biases are the result of the

Table 3 MPE by Revenue Source and Forecast Method

Mean Percent Error State Value-Added Over Naïve

Number of Forecast Forecast Outlook State Forecast Simple Trend AR(1) Smooth w/Trend Simple Trend AR(1) Smooth w/Trend

Sales Sales

12 One-Year –0.4% –9.6% –2.4% –0.1% –2.1% 9.2% 2.0% –0.3% 1.7%11 Two-Year –0.5% –14.9% –4.5% –1.2% –4.2% 14.4% 4.0% 0.7% 3.7%

CIT CIT

12 One-Year 4.6% 3.8% 8.8% 5.5% 15.4% –0.8% 4.2% 0.8% 10.8%11 Two-Year 3.8% –3.0% 4.4% –0.1% 9.3% –0.9% 0.5% –3.7% 5.5%

PIT PIT

12 One-Year 2.4% –6.3% 2.6% 5.0% 1.3% 3.9% 0.1% 2.6% –1.1%11 Two-Year –0.5% –14.9% –4.5% –1.2% –4.2% 14.4% 4.0% 0.7% 3.7%

Gaming Gaming

4 One-Year 2.1% –4.0% 13.7% 17.5% 13.4% 1.9% 11.6% 15.4% 11.3%3 Two-Year 14.0% –7.0% 22.6% 39.8% 20.2% –7.0% 8.6% 25.8% 6.2%

Other Other

12 One-Year –11.1% –8.8% –5.2% –12.7% 4.8% –2.3% –5.9% 1.6% –6.2%11 Two-Year –15.3% –14.0% –9.0% –18.5% 4.2% –1.3% –6.3% 3.2% –11.1%

Total Total

12 One-Year 0.0% –7.5% 0.0% 1.1% 1.0% 7.4% 0.0% 1.0% 0.9%11 Two-Year –0.6% –13.3% –2.5% –0.4% –1.5% 12.7% 1.9% –0.2% 0.9%

Note: State value added is calculated as (|MAPENaive| – |MAPEState|).

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 20105000

6000

7000

9000

8000

10000

11000

12000

13000

14000

Fiscal Year

Mill

ions

($)

Actual

State ForecastAR(1) Forecast

Figure 4 Total Indiana Tax Revenue: Comparison of Forecasts vs. Actuals

Criticizing a naive forecast amounts to disputing an unre-

sponsive target, easing the ability of actors to implicitly or explicitly substitute their own

estimation of what level of avail-able resources exist.


programs cannot easily proceed if the legis-lature cannot initially agree on the baseline fi scal constraint. Accuracy of the forecast matters for development of a fi scally sustain-able budget and is a test of the forecast itself. However, acceptance of the fi scal baseline is the test of the forecast process, just as accuracy is critical for the fi scal baseline. In the years since the development of the Indiana consen-sus revenue forecasting approach, there have never been dueling forecasts produced within the General Assembly. Th e Indiana legislature

and governor have always accepted the forecast coming from the RFTC as the one used for determining appropriations. As a result, it served as a hard budget constraint. Feigenbaum (2009b) sums up the results: “For decades, Indiana’s biennial budget process, characterized by its nonpartisan economic outlook and bipartisan revenue forecast, has served up numbers faithfully followed as gospel in assembling two-year budgets. Using projections from the forecasts, lawmakers argue between chambers and parties (with some input from the executive branch) over how much money to plug in for what programs, using the forecasts to determine the bottom line.”

But there have been two episodes when forecast acceptance was threatened. In the fi rst years of the twenty-fi rst century, an audi-tor of state became exasperated with the offi cial forecast (she was responsible for presenting a cash budget for the state and used a forecast in that work but was not part of the process of preparing or adopting the state budget) and threatened to do her own forecast. While she did issue an independent statement of monthly revenues, she did not carry through with generating her own forecasts and instead continued to use the RFTC forecast.

Th e more signifi cant challenge to the constraint of the consensus forecast came during the depths of the recent Great Recession, and it reveals an important advantage of political appointments involved with the process. Year-to-date actual collections lagged far behind forecasted collections through 2008 and 2009, requir-ing that the governor make signifi cant impoundments of state appropriations. As the appropriations for a new biennium were being considered during the spring of 2009, there was continu-ing concern about the accuracy of the revenue forecast because of the recent history. As the session continued, the governor began preparations to produce his own forecast (much lower than the forecast prepared by the RFTC), using advisors outside the con-sensus process and without any input from the legislature (or from the State Budget Agency representative on the RFTC). Ultimately, the governor and legislative leaders “agreed on commissioning the new ‘realistic’ forecast [from the RFTC] as part of a broader special budget process preceding the special session [held because the upcoming biennial budget had not been adopted before the legally required end to the legislative session]” (Feigenbaum 2009a). In practice, the RFTC continued with its methodology but changed the macroeconomic indicators from Global Insight being employed as the basis of the economic forecast. Because many of these vari-ables are highly correlated with one another, this change was largely superfi cial but apparently politically satisfying, as the new fore-casted values were once again incorporated into the state budget as developed in the special session.

amounts to disputing an unresponsive target, easing the ability of actors to implicitly or explicitly substitute their own estimation of what level of available resources exist. By contrast, a similar criticism produced by the politically diversifi ed forecast committee is not substantially diff erent from criticizing one’s own political party.

The Role of Politics in Producing Acceptable ForecastsTh e “acceptance” of the revenue forecast is somewhat more diffi cult to gauge than the value added measures already proposed, but we consider it here to mean that the budget debates do not include a debate over the available resources through the interjection of competing forecasts. Exploring the political his-tory of the RFTC, both in the immediate lead-up to its founding and during the challenges to its legitimacy, provides some evidence of the importance of forecast acceptance and the role that politics may play in obtaining this acceptance.28

Th e Indiana process emerged in the mid-1970s after a legislative session in which the governor and the General Assembly could not agree on a revenue baseline and, as a result, spent much of the legislative session arguing about how much revenue the state might collect during the budget biennium rather than making decisions about policies and resource allocation. Th e governor used one revenue baseline, the Republicans in the General Assembly used a diff erent one, and the Democrats used yet another. Indeed, there were even disagreements between the houses of the same party. After the session ended, all parties to the process came to the realization that the forecast should not be a policy question and that policy discussions would be improved if all budget participants worked from a single forecast.

Th e process installed then, by agreement and not by statute, seeks to produce an accepted forecast that neutralizes political infl u-ences (both real and accused) on the forecast result by involving all branches and both political parties. Th is creates checks and bal-ances in the system and incentivizes the forecast to be as accurate as possible. It is a scheme designed for maximum internal control over the forecast, and, as it has turned out, rather than engendering a climate of suspicion and eternal vigilance for transgressions by other committee members, it has arguably produced a common quest for accuracy. Furthermore, any lawmaker inclined to use a revenue baseline other than the one coming from the RFTC would be confronted by a member of his or her own party because representa-tion from the party was involved in the forecast process. As a process originating from political mistrust, the RFTC’s structure manages a balanced process that strives toward both forecast accuracy and acceptance in the budget process. No branch of state government nor any political party (should one party control all branches of gov-ernment) can easily manipulate the forecast for political gain. Th e process as initially created hinges on balancing all countervailing interests to produce an untainted forecast.

Acceptance of the forecast as a hard budget constraint, even when it turns out to be inaccurate, is an important juncture in the budget-ing process. Th e business of making policy choices among public

Accuracy of the forecast matters for development of a fi scally

sustainable budget and is a test of the forecast itself. However, acceptance of the fi scal baseline is the test of the forecast process, just as accuracy is critical for the

fi scal baseline.


Th e composition of the RFTC was critical in this preservation of the forecast process during the governor’s challenge. Not only did legislative and executive branch appointees argue for the process with those who had appointed them, they also worked through the channels of the two political parties to maintain support. Th us, it was not just through the organizational structure of government that the forecast process was maintained, it was also through the political parties.29 Much of the case invoked the need to avoid the deliberative anarchy that had prevailed prior to the existing process. Even if a naive forecast approach had produced identical numbers and history of accuracy to that of the RFTC, it would lack any institutional stakeholder to support it through the political parties, and it would not seem likely to have been eff ective if it were to have come from a nonpartisan research offi ce or consulting group.

During the Great Recession, the governor continued to use these RFTC’s monthly reports as the basis for expenditure impoundments across state government. Since the errors of the Great Recession, the RFTC has also conducted a quarterly review of its forecasts, focus-ing on economic developments that might require a formal revision of its methodology and its forecast. However, there have been no signifi cant revisions from these reviews.

In sum, the process develops a consensus forecast from a scheme involving all fi scal parties. Th e process institutionalizes counter-vailing interests—between legislature and executive branches and between the two political parties in Indiana state government—to prevent organized manipulation of the forecast. By forestalling manipulation, the strongest incentive among the actors in the proc-ess is the pursuit of accuracy, the fi rst test of revenue forecasting. But the process, by ensuring inclusion of all these interests, creates the basis for acceptance of the forecast as the binding baseline for the budget process.

Evidence from business forecasting shows that what is called “team-based forecasting,” a rough equivalent of Indiana consensus forecast-ing, in addition to improving forecast accuracy, also leaves managers more satisfi ed with the forecasts (Kahn and Mentzer 1994). As more actors are satisfi ed with a forecast, the more likely it is that the actors will operate within that forecast. In public sector terms, the actors are more likely to regard the forecast as binding. Evidence shows that the naive forecasts proved no more accurate than those from the RFTC. Indeed, most such approaches were less accurate. However, the big question for the alternative approaches is whether naive forecasts would be accepted as a hard budget constraint, given that they would be destined for similar periodic inaccuracy without the benefi t of political stakeholders to ensure acceptance during the budget process.

Conclusion and Directions for Further ResearchTh e literature on public revenue forecasting has a diversity of results on the role of political incentives that might explain forecast error and bias. Th is article has argued the forecast should be viewed in the context of the entire public budget deliberation process, with the recognition of potential for political manipulation and political gains for accusing the forecast of bias. In this broader view, it becomes clear

that the organization of forecast committees and the methods they employ must be understood for their ability to gain acceptance. In doing so, this article redirects the literature to understanding the forecast acceptance as an important institutional objective that can be as signifi cant as the forecast outcome. An accurate forecast is of no value in the budget process if process participants ignore it as they develop the expenditure program. Policy recommendations that have emerged from the research on political bias in the forecast-ing process treat the structure of forecast committees as exogenous variation and universally recommend approaches to depoliticization. For instance, Boyd, Dadayan, and Ward summarize the literature in stating that consensus forecasting processes seek “to remove politics from the estimating process as much as possible to limit lawmak-ers’ attempts, especially in election years, to present a rosier view of revenue” (2011, 974). Th ese recommendations may produce gains in accuracy but also may forgo acceptance by the budget actors by neglecting the broader political budgeting process.

To demonstrate the importance of forecast acceptability, this article examined the forecast errors and political history of revenue forecasting in Indiana. Indiana is particularly interesting because its forecasts have been recognized as generally accurate, and the process is explicitly politicized. In stark contrast to the recommendations to depoliticize, the Indiana process explicitly brings political diff erences into the forecasting methodology. Interestingly, this is diff erent from the “consensus” forecasts generated by more than one branch of state government that has been the focus of previous research. Th e legislature is not represented in the process by a nonpartisan research body, as is the case in some consensus states, but rather by majority party and minority party staff from the fi scal committees of each house. In an era of single-party statehouses, such explicit politicization and guaranteed representation from both political parties may be the best approach to inducing forecast accuracy, considerably more relevant than the standard legislative–executive consensus, and will be more likely to produce a controlling revenue forecast. Such an observation would be consistent with other recent empirical work by Krause and Melusky (2012), which demonstrates that unilateral executive authority over revenue forecasts or budget formulation results in fi scal spending growth patterns that are con-sistent with the respective party’s political ideology.

Th e Indiana consensus forecast record emerges from its utilization of countervailing interests that balance to make accuracy a primary objective and contribute to making the forecast, right or wrong, accepted in the budget process. Th e process puts executive and legislative branches and political parties in countervailing positions. Neither the legislative nor the executive branch can implement a fi scal strategy of constraint or expansion by intentionally over- or underforecasting revenue in the forecast deliberations because both branches are in the process. Th e legislature is represented in the con-sensus process by fi scal staff from the two political parties. In sum,

the countervailing process balances executive and legislative bias, as well as political party bias, to create a common objective of forecast accuracy. All parties—political, legislative, and executive—have been included in producing the forecast and, accordingly, have a stake in its acceptance. A naive forecast structure (and there are multiple possible naive methods,

All parties—political, legisla-tive, and executive—have been

included in producing the forecast and, accordingly, have a

stake in its acceptance.


with no a priori way of selected from among them) has no way of marshaling countervailing forces into the forecasting process and creates no environment in which budget participants might feel allegiance to the forecast that is expected to be binding.

We do not suggest that the outcomes of the Indiana forecast-ing process necessarily translate to other states, but we do sug-gest that observing the Indiana history is illustrative of a broader need to shift the emphasis of the academic literature on revenue forecasting. In particular, an emphasis that views the forecast as a stage of the budget facing the public administration problem of institutional trust is necessary. It may be the case that add-ing political diversity not only improves forecast quality, as has been found in previous research, but also systematically facilitates forecast acceptance. It is also possible that political diversity is not the only mechanism that can result in acceptability. Th e point is that acceptance is a vital component of the forecasting process, and future research should relate how diff erences in the structure of the budgeting process inform which revenue forecast processes become politically acceptable, as well as how they induce forecast acceptance. Th is article drew on an analytical narrative from media reports around the forecasting process and interviews with forecast members to determine the extent to which the forecast numbers were accepted. Alternative approaches, particularly ones that deter-mine forecast acceptability using a quantitative approach, would be of considerable interest.

AcknowledgmentsTh e authors appreciate discussion from the members of the Indiana Revenue Forecast Technical Committee. Th e authors also wish to express their appreciation for helpful comments and feedback from two anonymous referees, Don Boyd, Gary Wagner, NaLette Brodnax, Burney Fischer, and participants at the 2012 Annual Conference on Taxation, 2012 Public Budgeting and Finance Section, Western Social Science Association Conference, and the Elinor and Vincent Ostrom Workshop in Political Th eory and Policy Analysis Colloquium Series. Any mistakes are solely those of the authors.

Notes 1. A recent review of this literature on state revenue forecasting techniques and

accuracy is found in Boyd, Dadayan, and Ward (2011). 2. Our judgment is that most of this literature is found in economics and fi nance

journals, but political science and public administration have long articulated a larger and more nuanced view of the interactions between political actors and public bureaucrats (e.g., Falaschetti and Miller 2001; Lewis 2003; Moe 1989). Th is latter, albeit smaller, literature that has been applied to state revenue fore-casts generally fi nds that, conditional on having politics involved with the rev-enue forecast process, it is better to be politically diverse (e.g., Deschamps 2004; Klay 1985; Krause, Lewis, and Douglas 2006, 2013; Smith 2007; Voorhees 2004).

3. Grizzle and Klay (1994) raised the issue of the relative utility of diff erent quan-titative methods, in regard to the cost of econometric causal modeling and its relative accuracy, in state revenue forecasts in the mid-1990s. Lower costs of data management and analytic technology make some of this of lesser concern now.

4. For examples, see Makridakis and Hibon (1979). 5. General macroeconomic forecasting is not without its challenges as well. Faust

and Wright (2007) test the accuracy of gross domestic product (GDP) forecasts and fi nd that a simple univariate autoregressive forecast (forecasting GDP

growth by regressing growth against four lagged periods of growth) is as accurate as substantially more complicated models.

6. Th is observation is somewhat confounded by the fact that the Congressional Budget Offi ce is charged with forecasting revenues under the existing baseline, and that baseline may deviate on the basis of expenditure decisions made by Congress and the president.

7. Danninger, Cangiano, and Kyobe (2005) suggest that interference to reduce revenue forecasts provides governments of low-income countries greater capacity to extract resources (i.e., steal them) without being noticed. Th at is not likely an issue for American states.

8. Revenue forecasting (or preparing the revenue baseline) is distinct from revenue estimating or scoring the impact of proposed changes in the revenue system (Mikesell 2012). Th e revenue estimators may be in either the executive (budget agency, department of revenue, etc.) or the legislative (legislative service agency) branch of government.

9. Pennsylvania follows a diff erent path from the other states: its revenue forecasts are prepared in its Independent Fiscal Offi ce, an entity not connected with either the legislative or executive branch of government.

10. For instance, the California Legislative Analyst’s Offi ce forecast $5 billion less for fi scal 2013 than did the governor—out of a $92.5 billion budget (Buchanan 2012). In such a case, it is not simple to know what amount is available for appropriation.

11. Th e forecasts are developed independently, but they usually have proven to produce similar totals (Niederjohn 2004).

12. Th e governor described the legislature’s analyst as “Dr. Kevorkian” for producing forecasts not to his liking (Hamilton 2013).

13. Louisiana Revised Statutes (Title 39:26).14. Th e Indiana approach is described in greater detail in Mikesell (2008).15. Manufacturing is the largest single contributor to gross domestic product in

Indiana, and the state economy has historically been extremely sensitive to fl uctuations in national economic activity.

16. Th e states without a binding revenue forecast process covered by state statute include Alaska, Arizona, Colorado, Delaware, Georgia, Indiana, Montana, New Hampshire, New Jersey, New Mexico, North Carolina, Ohio, Texas, Utah, Vermont, West Virginia, and Wyoming.

17. Th is summary is according to comments from the tax economist.18. For example, when the RFTC prepared its forecast for the 2014–16 budget bien-

nium in December 2012, it included the impact of a use tax collection agree-ment reached with Amazon.com that would take eff ect in calendar year 2014.

19. Of course, it is almost certain that the members of the committee are aware of what the approved method implies for the forecasted amount of revenue.

20. Research that ignores the distinction between biennial and annual budget proc-esses by using only the revenue forecast prepared just before the start of each budget year in judging the accuracy of the revenue forecast misses an important detail of the budget process. In biennial budget states, it is the two-year revenue forecast prepared before the budget adoption session that matters for develop-ment of a useful hard budget constraint.

21. According to Medetsky (2012), the Russian Federation collects half its revenue from oil exports and recently sought a “precise mathematical formula” to esti-mate oil prices. Th e Ministry of Finance proposes to use the median price of oil for the past 10 years to “avoid personal judgment … and make it more solid.” Th e 2013 budget would be based on fi ve years, and the horizon would increase by one year each year until it reaches 10 years.

22. Although a frequently suggested naive model, the simple lag can be a limited operational option because budgets get developed, considered, and enacted before actual revenue for the prior year is known.

23. Th e Indiana offi cial forecast predates this time period, but the data for prior years seem to be lost. Data used in this analysis were generously provided by the state Department of Revenue (for earlier years) and from postings to the


State Budget Agency Web site (for more recent years) (see http://www.in.gov/sba/2601.htm).

24. Using actual instead of forecast values in the denominator also avoids the problem of the MAPE becoming a nonlinear function of the forecast error, which can be seen by taking the derivative of MAPE with respect to the denominator.

25. In the computation of MPE and MAPE, it is important that At, rather than Ft, be used in the denominator. Although some previous research has reversed this scaling of the error, this causes these two measures to become nonlinear func-tions of the forecast, which can be seen by taking the partial derivative of the formulas with respect to Ft.

26. Th is is not equivalent in nature to the previous value added statistic that was the diff erence in the mean absolute percentage errors. For any two functions, f and g, there are only special circumstances in which f(g(x)) is equal to g(f(x)).

27. To be clear, it is not the case that the time trend or the AR(1) regression–based forecasting would necessarily produce a mean error of zero because each forecast is an out-of-sample prediction based only on the historical data available at the time the forecast would have occurred.

28. Th is section is primarily based on discussions the authors had with current and past members of the RFTC.

29. According to the university tax economist serving on the RFTC, he may have played some role in assisting with the accuracy of the forecast, but he played no role in preserving the acceptance of the forecast process in delivering a hard budget constraint.

ReferencesAuerbach, Alan J. 1999. On the Performance and Use of Government Revenue

Forecasts. National Tax Journal 52(4): 765–52.Blackley, Paul R., and Larry DeBoer. 1993. Bias in OMB’s Economic Forecasts and

Budget Proposals. Public Choice 76(3): 215–32.Boyd, Donald J., Lucy Dadayan, and Robert B. Ward. 2011. States’ Revenue

Estimating: Cracks in the Crystal Ball. State Tax Notes, March 28, 945–72.Boylan, Richard T. 2008. Political Distortions in State Forecasts. Public Choice

136(3): 411–27.Bretschneider, Stuart I., and Wilpen L. Gorr. 1987. State and Local Government

Revenue Forecasting. In Th e Handbook of Forecasting: A Manager’s Guide, 2nd ed., edited by Spyros Makridakis and Steven C. Wheelwright, 118–34. New York: Wiley, 1987.

Buchanan, Wyatt. 2012. California Revenue Estimates Diff er by $5 Billion. San Francisco Chronicle, January 12.

Buettner, Th iess, and Bjoern Kauder. 2009. Revenue Forecasting Practices: Diff erences across Countries and Consequences for Forecasting Performance. Working Paper no. 2628, CESifo. http://www.cesifo-group.de/ifoHome/publica-tions/working-papers/CESifoWP/CESifoWPdetails?wp_id=14556414 [accessed November 25, 2013]

Cassidy, Glen, Mark S. Kamlet, and Dan S. Nagin. 1989. An Empirical Examination of Bias in Revenue Forecasts by State Governments. International Journal of Forecasting 5(3): 321–31.

Cooke, Stephen C., and Christopher S. McIntosh. 2011. Forecasting General Fund Revenue: An Analysis of Idaho FY 1998 to FY 2011. Public Budgeting and Finance 31(4): 51–73.

Couture, Jérôme, and Louis M. Imbeau. 2009. Do Governments Manipulate Th eir Revenue Forecasts? Budget Speech and Budget Outcomes in the Canadian Provinces. In Studies in Public Choice: Do Th ey Walk Like Th ey Talk?, edited by Louis M. Imbeau, 155–67. New York: Springer.

Danninger, Stephan, Marco Cangiano, and Annette Kyobe. 2005. Th e Political Economy of Revenue-Forecasting Experience from Low-Income Countries. Working Paper no. WP/05/2, International Monetary Fund. http://www.imf.org/external/pubs/ft/wp/2005/wp0502.pdf [accessed November 25, 2013]

Deschamps, Elaine. 2004. Th e Impact of Institutional Change on Forecast Accuracy: A Case Study of Budget Forecasting in Washington State. International Journal of Forecasting 20(4): 647–57.

Falaschetti, Dino, and Gary Miller. 2001. Constraining Leviathan: Moral Hazard and Credible Commitment in Institutional Design. Journal of Th eoretical Politics 13(4): 389–411.

Faust, Jon, and Jonathan Wright. 2007. Comparing Greenbook and Reduced Form Forecasts Using a Large Realtime Dataset. Working Paper no. 13397, National Bureau of Economic Research. http://www.nber.org/papers/w13397 [accessed November 25, 2013].

Feenberg, Daniel R., William Gentry, David Gilroy, and Harvey S. Rosen. 1989. Testing the Rationality of State Revenue Forecasts. Review of Economics and Statistics 71(2): 300–308.

Feigenbaum. Edward. 2009a. Lawmakers Th ink “Gloomy” in Spite of Rosy Forecast. Indianapolis Business Journal, April 27.

———. 2009b. Let the State Budget Bickering Begin Again. Indianapolis Business Journal, June 8.

Frankel, Jeff ery A. 2011. Over-Optimism in Forecasts by Offi cial Budget Agencies and Its Implications. Oxford Review of Economic Policy 27(4): 536–62.

Grizzle, Gloria A., and William Earle Klay. 1994. Forecasting State Sales Tax Revenues: Comparing the Accuracy of Diff erent Methods. State and Local Government Review 26(3): 142–52.

Hamilton, Billy. 2013. New Jersey’s “Dr. Kevorkian” Is Back, Possibly Killing a State Tax Cut. State Tax Notes, June 10, 837–41.

Heins, A. James. 1975. Th e Politics of Revenue Forecasting. Proceedings of the Annual Conference of the National Tax Association, 29–33.

Jonas, Kirk, Gregory J. Rest, and Terry Atkinson. 1992. Virginia’s Revenue Forecasting Process and Models. Public Budgeting and Finance 12(2): 70–81.

Kahn, Kenneth B., and John T. Mentzer. 1994. Th e Impact of Team-Based Forecasting. Journal of Business Forecasting Methods and Systems 13(2): 18–21.

Klay, William Earle. 1985. Th e Organizational Dimensions of Budgetary Forecasting: Suggestions from Revenue Forecasting in the States. International Journal of Public Administration 7(3): 241–65.

Kliesen, Kevin L., and Daniel L. Th ornton. 2012. How Good Are the Government’s Defi cit and Debt Projections and Should We Care? Federal Reserve Bank of St. Louis Review 94(1): 21–39.

Krause, George A., and James W. Douglas. 2012. Organizational Structure and the Optimal Design of Policymaking Panels: Evidence from Consensus Group Commissions’ Revenue Forecasts in the American States. American Journal of Political Science 57(1): 135–49.

Krause, George A., David E. Lewis, and James W. Douglas. 2006. Political Appointments, Civil Service Systems, and Bureaucratic Competence: Organizational Balancing and Executive Branch Revenue Forecasts in the American States. American Journal of Political Science 50(3): 770–87.

———. 2013. Politics Can Limit Policy Opportunism in Fiscal Institutions: Evidence from Offi cial General Fund Revenue Forecasts in the American States. Journal of Policy Analysis and Management 32(2): 271–95.

Krause, George A., and Benjamin F. Melusky. 2012. Concentrated Powers: Unilateral Executive Authority and Fiscal Policymaking in the American States. Journal of Politics 74(1): 98–112.

Lewis, David. 2003. Presidents and the Politics of Agency Design: Political Insulation in the United States Government Bureaucracy, 1946–1997. Stanford, CA: Stanford University Press.

Makridakis, Spyros, and Michele Hibon. 1979. Accuracy of Forecasting: An Empirical Investigation (With Discussion). Journal of the Royal Statistical Society, series A, 142(2): 97–145.

———. 2000. Th e M3-Competition: Results, Conclusions, and Implications. International Journal of Forecasting 16(4): 451–76.


Medetsky, Anatoly. 2012. Cabinet Backs More Precise Oil Calculation. Moscow Times, June 29.

Megerian, Chris. 2013. Analyst Disputes Brown’s Budget Outlook as Too Glum. Los Angeles Times, May 17.

Mikesell, John L. 2008. State Revenue Forecasting in the State of Indiana: A Consensus System in a Politically Divided State. In Government Budget Forecasting, Th eory and Practice, edited by Jinping Sun and Th omas D. Lynch, 415–30. Boca Raton, FL: CRC Press.

———. 2012. Revenue Estimation/Scoring by States: An Overview of Experience and Current Practices with Particular Attention to Dynamic Methods. Public Budgeting and Finance 32(2): 1–24.

Mocan, H. Naci, and Sam Azad. 1995. Accuracy and Rationality of State General Fund Revenue Forecasts: Evidence from Panel Data. International Journal of Forecasting 11(3): 417–27.

Moe, Terry M. 1989. Th e Politics of Bureaucratic Structure. In Can the Government Govern?, edited by John E. Chubb and Paul E. Peterson, 267–329. Washington, DC: Brookings Institution.

National Association of State Budget Offi cers (NASBO). 2008. Budget Processes in the States. Washington, DC: NASBO.

Nelson, Colleen McCain. 2012. Single Party Statehouses. Wall Street Journal, November 9.

Niederjohn, Scott. 2004. State Revenue Forecasting in Wisconsin: A Critical Examination. Wisconsin Interest 13(3): 33–40.

Qiao, Yuhua. 2008. Use of Consensus Revenue Forecasting in U.S. State Governments. In Government Budget Forecasting, Th eory and Practice, edited by Jinping Sun and Th omas D. Lynch, 393–414. Boca Raton, FL: CRC Press.

Rodgers, Robert, and Philip Joyce. 1996. Th e Eff ect of Underforecasting on the Accuracy of Revenue Forecasts by State Governments. Public Administration Review 56(1): 48–56.

Rose, Shanna, and Daniel L. Smith. 2012. Budget Slack, Institutions, and Transparency. Public Administration Review 72(2): 187–95.

Ross, Rebecca. 2001. Consensus Revenue Estimating: Th e Process. Fiscal Forum: A Legislative Briefi ng, April. http://www.house.mi.gov/hfa/Archives/PDF/consens.pdf [accessed November 25, 2013]

Rubin, Marilyn, J. L. Peters, and Nancy Mantell. 1999. Revenue Forecasting and Estimation. In Handbook on Taxation, edited by W. Bartley Hildreth and James A. Richardson, 769–99. New York: Marcel Dekker.

Shkurti, William J., and Darrell Winefordner. 1989. Th e Politics of State Revenue Forecasting in Ohio, 1984–1987: A Case Study and Research Implications. International Journal of Forecasting 5(3): 361–71.

Smith, Daniel L. 2007. Rules, Participants, and Executive Politics in State Tax Revenue Forecasting. Journal of Public Budgeting, Accounting and Financial Management 19(4): 472–87.

Snell, Ronald K. 2011. State Experiences with Annual and Biennial Budgeting. http://www.ncsl.org/documents/fi scal/BiennialBudgeting_May2011.pdf [accessed November 25, 2013].

Thompson, Fred, and Bruce L. Gates. 2007. Betting on the Future with a Cloudy Crystal Ball? How Financial Theory Can Improve Revenue Forecasting and Budgets in the States. Public Administration Review 67(5): 825–36.

Voorhees, William R. 2004. More Is Better: Consensual Forecasting and State Revenue Forecast Error. International Journal of Public Administration 27(8–9): 651–71.

Williams, Daniel W. 2012. Th e Politics of Forecast Bias: Forecaster Eff ect and Other Eff ects in New York City Revenue Forecasting. Public Budgeting and Finance 32(4): 1–18.

state revenue forecasts and political acceptance: th e ...faculty.cbpp.uaa.alaska.edu/afgjp/padm628...

Documents