SOURCES OF ERROR IN STATE REVENUE FORECASTS OR HOW CAN THE FORECAST POSSIBLY BE SO FAR OFF Thomas F. Stinson* ABSTRACT. Recently state and federal revenues have fallen well short of projections. While revenues normally turn down in a recession, current revenue shortfalls have been much greater than would have been expected given how mild the 2001 recession turned out to be. This paper examines some of the reasons for the large forecast variances observed in recent years using specific examples from forecasts made for the state of Minnesota. Key factors identified include inaccurate forecast for U.S. economic growth, inadequate, untimely and inaccurate data; imperfect models; and unrecognized changes in the structure of the economy. These factors came together and reinforced each other, ultimately producing a larger reduction in state revenues than could have been anticipated in advance. INTRODUCTION On December 4, 2002, Minnesota’s Department of Finance announced the state faced a $4.5 billion budget deficit. The $300 million deficit forecast for the 2002-03 biennium would have to be resolved quickly since Minnesota’s constitution prohibits borrowing across biennia for operating expenditures and less than 9 months remained before the biennium closed. But, the outlook for the 2004-05 biennium posed a more serious problem. There, a $4.2 billion gap between revenues and current law expenditures was now projected (Minnesota Department of Finance, 2002c, p. 1).1 A sizeable budget deficit had been widely anticipated, but few expected the projected shortfall would reach an amount equal to more than $900 per Minnesota resident. -------------------* Thomas F. Stinson, Ph.D., is an Associate Professor, Department of Applied Economics, University of Minnesota. Since 1987 he has also served as the Minnesota state economist where his duties include supervising preparation of the state’s revenue forecast. Minnesota was not alone in facing fiscal problems. Ray Schapack, executive director of the National Governor’s Association characterized the financial situation facing state government as the most severe since World War II. Somehow state revenues had not only failed to meet expectations, they had declined dramatically. The drop in revenue was particularly hard to accept and deal with because the shortfall had occurred following a period of what seemed like almost magical revenue growth. State policy makers had grown used to finding substantial surpluses each time the budget outlook was updated. Some efforts had been made to build up reserves and rainy day funds, and some money had been put away in “hidden reserves” to provide flexibility for dealing with an economic slowdown that was becoming evident. But, no one had prepared for budget problems of the magnitude that ultimately appeared. It is convenient to assume that state and federal revenue estimators should have used more sophisticated forecasting techniques, and that, had more complex modeling approaches been used, the eventual shortfalls would have been more modest. And, it is true that alternative approaches to revenue forecasting might have produced some marginal reduction in projected shortfalls. But it is unlikely that substantial shortfalls could have been avoided. Revenue forecasting models used in most states are quite sophisticated and much of the error in forecast comes from institutional factors outside the control of the forecaster and unrelated to the sophistication of the forecasting models used. Each state, of course, had different experiences and the sources of their budget problems were not all the same. Still, the forecasts of most states, and of the federal government, were adversely affected by many of the same forces, and understanding them will help policymakers better understand the need for risk management strategies including reserves and longer-term structural balance. Recent revenue shortfalls are a notso-gentle reminder that the economy is continuously changing, and that even the best economic forecasting models have great difficulty forecasting turning points or changes in regime. This paper describes sources of the budget deficit projected in Minnesota for the 2004-05 biennium. It begins with a short discussion of Minnesota’s forecast process. Then the evolution of the 2004-05 budget outlook is traced, beginning with legislative decisions made in June 2001. Key factors contributing to the widening gap between forecast and actual receipts are described in some detail. Those factors include the overly optimistic forecasts of economic growth and stock market growth provided by the state’s national forecasting service; inadequate, untimely and inaccurate data; imperfect models; and unrecognized changes in the structure of the economy. Each played a substantial role in Minnesota’s revenue shortfall. And like the “perfect storm” when these factors joined together they reinforced each other, ultimately producing a change in state revenues that was much larger than could possibly have been anticipated. FORECASTING AND BUDGETING IN MINNESOTA Minnesota is a biennial budget state with general fund expenditures totaling about $27 billion in the 2002-03 biennium (Minnesota Department of Finance, 2003, p. 66). Budgets are prepared on a biennial basis, and the Finance Department produces official planning estimates for the following biennium as well. Those planning estimates, while not legally binding, provide budget discipline, helping to forestall passage of programs with minimal costs in the current biennium, but with budget busting potential in following years. Although there is no explicit requirement that the budget be balanced in the out-years it is generally accepted that structural balance, defined as revenues equal to or exceeding on-going expenditures in the final year of the out-biennium, is consistent with good financial practice. Revenue and expenditure forecasts are prepared twice yearly, in November and February.2 The November forecast in even numbered years is the basis for the governor’s biennial budget. The forecast released the following February updates the revenue outlook, establishing the current law deficit or shortfall against which proposed future spending and tax changes are measured, and triggering budget action in the legislature. Forecasts the following November and February update the outlook, indicating whether there is likely to be a surplus or deficit at the end of the current biennium. The forecast prepared by the state’s Department of Finance is the basis for Minnesota’s general fund budget. The state does not have a consensus forecast process, there is no forecast council that passes on the final revenue estimates, and no competing forecasts are prepared by legislative staff. By law forecast results must be made public at the time they are revealed to the Governor. The state’s forecasts have the reputation of being professional and independent from political considerations. To help insulate the forecast from political pressures the state’s Finance Department contracts with the University of Minnesota for half time of a tenured faculty member to supervise the forecast’s preparation. The four-person revenue forecasting staff is very experienced, averaging more than 20 years of forecasting experience. In 2000, Minnesota held AAA bond ratings from all three rating services. Today, despite its recent financial problems, Minnesota retains AAA ratings from two rating agencies and is rated AA1 by the third. The income and sales taxes are the largest sources of revenue, generating 43 percent and 30 percent of general fund revenues respectively. The corporate franchise tax and a statewide property tax on commercial, industrial, and seasonal recreational property and the motor vehicle sales tax provide less than 5 percent each. Other revenue sources, including revenue from the state’s tobacco settlement made up about 16 percent of Minnesota’s general fund revenues (Minnesota Department of Finance, 2003, p.37). Minnesota’s income tax uses federal taxable income as its base. Capital gains are taxed at the same rate as ordinary income; they receive no special treatment. In 2000, wages were more than 71 percent of federal adjusted gross income (FAGI) for Minnesota taxpayers (see Table 1). Capital gains were 7.8 percent of FAGI in 2000. Minnesota applies a sales tax at a rate of 6.5 percent to purchases of household and business items. Major items excluded from the sales tax base include food, clothing, and prescription drugs. Purchases of business capital equipment are subject to tax, but eligible for a sales tax refund. The narrow coverage of the sales tax makes receipts more volatile than those from broader based sales taxes levied in a number of states. TABLE 1 Components of Minnesota Adjusted Gross Income, Tax Years 2000, 2001 2000 FAGI Wages Interest/Dividend Business Capital Gain Pensions/IRA Taxable Income 120.028 85.588 5.737 3.602 9.403 7.530 84.932 % of FAGI 71.3 4.8 3.0 7.8 6.3 70.8 2001 116.938 88.096 5.097 3.573 4.502 7.739 80.616 Source: Statistics of Income Bulletin (2002; 2003). % of FAGI 75.3 4.4 3.1 3.8 6.6 68.9 % Change -2.5 2.9 -11.2 -0.8 -52.1 2.8 -5.1 THE EVOLUTION OF THE 2004-05 DEFICIT The Prelude, 2000-2001 When the Finance Department released its preliminary forecast of revenues and expenditures for fiscal 2002-03 and its first planning estimates for FY 2004-05 the state had an embarrassment of riches. An ending balance of $924 million was projected for the end of the current (2000-01) biennium. State law required that balance be rebated to taxpayers. (Minnesota Department of Finance, 2000, p. 1) A balance of nearly $2.1 billion (in addition to the $924 million projected for fiscal 2001) was projected for the 2002-03 biennium, and the first planning estimates for the 2004-05 biennium showed total revenues of nearly $31.3 billion, with a structural balance of $1.5 billion in fiscal 2005. The forecast report contained a note of caution, however. It warned that “After several years of forecasts more cautious than most, (the state’s national forecasting service’s) new national model projects stronger than consensus growth through 2003”(Minnesota Department of Finance, 2000, p. 5). The baseline forecast for the current year called for national real GDP growth of 3.6 percent, only modestly stronger than the 3.4 percent growth anticipated by the November’s Blue Chip Consensus. In future years, though, the gap was larger. Minnesota’s national forecasting service expected the U.S. economy to grow at a real rate of 4.3 percent in 2002 and 4.8 percent in 2003. Those growth rates were well above the Blue Chip panel’s, 3.3 percent and 3.2 percent (p. 5). The Congressional Budget Office (2001) forecast, released the following January, was built on real growth rates of 3.4 percent and 3.1 percent for 2002 and 2003. Because forecasting services are generally believed to be hesitant to forecast the start of a recession and because even good forecasting models can occasionally produce projections that are decidedly too optimistic (or too pessimistic) Minnesota’s Finance Department uses a panel of local economists to help assess whether the national forecast provided by the state’s forecasting service is reasonable.3 That Council of Economic Advisors’ review, which is summarized in each forecast report, noted that “for the first time since February, 1995, the Council formally considered replacing the (national baseline) forecast with a more conservative forecast.” But, “after a lengthy discussion a majority of the Council recommended that the Department of Finance base its November revenue forecast on the (national baseline forecast provided by the contractor)” (Minnesota Department of Finance, 2000, p. 14) The summary of the Council meeting was filled with cautions noting that “the optimistic nature of this forecast greatly increases the risk that revenues will fall short of forecast at some point during the 2002-03 biennium” (November Forecast, 2000, p. 15). Three months later, when the February 2001 revenue forecast was released, economic conditions had deteriorated slightly, but only slightly. The forecast service’s U.S. baseline had been cut back modestly, Council members while in general agreement about the outlook for 2001 were again less optimistic than the forecast service for 2002 and 2003. The state’s national economic consultant saw additional risk, raising the probability of an early recession to 40 percent from November’s probability of 25 percent (Minnesota Department of Finance, 2001a, p. 1). Projected real GDP growth rates for 2001 and 2002 had been reduced, but the baseline forecast still anticipated 4.8 percent growth in 2003. The inflation outlook remained low, with the consumer price index projected to grow at just a 1.8 percent rate in 2002 and a 1.6 percent rate in 2003 (Minnesota Department of Finance, 2001a, p. 56). The nominal (current dollar) growth rate used in the forecast for 2002 was 6.4 percent, still noticeably above the 5.1 percent CBO amount used by CBO. The Council requested Finance Department economists construct and run several alternative national scenarios that assumed slower productivity growth rates in 2002 and 2003. Real growth rates slowed, but nominal (current dollar) growth rates changed little. Since Council members typically expected higher inflation than the forecasting service projected, and since revenue forecasts are driven by nominal growth rates, the lower inflation rate provided a slight conservative offset to the more optimistic real growth rate projection. Following another extended discussion among Council members, it was agreed to use the U.S. forecast provided by the forecasting service as the foundation for the state’s February revenue forecast (Minnesota Department of Finance, 2001a, p. 10). When completed, the February 2001 forecast showed the expected ending balance for the 2000-01 biennium had fallen by $67 million, to $856 million (Minnesota Department of Finance, 2001a, p. 1). The projected surplus for fiscal 2002-03 had shrunk by nearly $600 million, to $1.5 billion, but remained nearly 5 percent of the existing budget. The 2005 the structural balance had fallen only slightly to $1.4 billion. Legislative deadlock over how much of the projected 2002-03 surplus should go for tax cuts and how much for increased spending forced a special legislative session in June 2001. By the time the special session was convened, the weakening of the national economy was visible and state revenues were under-performing the forecast. The Finance Commissioner and the state economist held a press conference to warn that the economy appeared weaker than projected and that state revenues were unlikely to meet February’s forecast. The official forecast remained in place, but the legislature acceded to the Governor Ventura’s demand that additional funds be added to the reserve. As further insurance against lower revenues, an ending balance of $253 million was left at the end of the biennium. Other actions taken to improve future financial flexibility included both recognizing and paying for some future costs and buying back some shifts in payment timing that had been made in prior years. (Minnesota Department of Finance, 2001c) At the end of the 2001 special legislative session, Minnesota policy makers had reason to believe that they had positioned the state’s finances well for a weakening economy. Revenues would likely be below forecast when the next forecast was released in November, but the state had reserves of $1 billion, and a structural balance of $242 million in fiscal 2005. Events to follow, though, would soon make apparent that the coming problem would dwarf those attempts to provide for financial stability. The Shock --The Economic Aftermath of September 11 On September 11, 2001, everything changed. Admittedly, the economic outlook had deteriorated markedly before the terrorist attack. The September U.S. baseline forecast was substantially below that used to drive February’s revenue forecast. But, by October the outlook was for a recession in 2001 and no real GDP growth in fiscal 2002 (Minnesota Department of Finance, 2001b). The economy was expected to recover in fiscal 2003, but the 3.6 percent growth rate now projected while historically healthy was well below the 4.7 percent growth assumed when February’s revenue forecast was prepared. And, that growth would come from a lower base level of economic activity. Minnesota’s receipts during the first quarter of fiscal 2002 were equally troubling. Revenues were down $100 million (3.5 percent) from forecast. Large declines in the individual income tax – off $70 million (5 percent)-- and the corporate franchise tax – off $47 million (24 percent) were only partially offset by smaller than expected tax rebate payments and increased miscellaneous revenues. November’s forecast provided a more complete picture of the grim news. For the 2002-03 biennium the projected $253 million balance at the end of the legislative session had disappeared and the Finance Department now was now projecting a biennial budget deficit of nearly $2 billion dollars (Minnesota Department of Finance, 2001c). And, as would be expected, the outlook for the 2004-05 biennium was equally dreadful with a deficit of more than $2.5 billion projected. The $242 million structural balance for fiscal 2005 was no more. In its place was a structural deficit of $1.234 billion. Many hoped February’s revenue forecast would bring better news, but it did not. The projected deficit for 2002-03 grew by more than $300 million to nearly $2.3 billion, and the planning estimates for the 2004-05 biennium showed a projected shortfall of $3.2 billion (Minnesota Department of Finance, 2002a). The outlook for the U.S. economy had improved slightly from November to February, but revenues were even failing to keep pace with the lower projections made in November. Net non-dedicated revenues were now expected to be down $2.4 billion, almost 9 percent below levels forecast a year earlier, after adjusting for legislative action. Legislative Action – A Solution for FY 2002-03, But FY 2004-05 Left Unresolved The Governor submitted a set of revised budget recommendations that balanced the 2002-03 budget and made a start toward addressing the emerging, longer-term structural budget problem, but his recommendations were too austere to receive legislative approval in an election year. Instead, the legislature quickly approved a set of shortterm measures including drawing down the state’s budget reserve and shifting some payments to the next fiscal year to solve the projected deficit for the current biennium. Some of those actions reduced projected expenditures slightly for 2004 and 2005 as well, but most changes affected only the current biennium. The Governor vetoed that measure, the legislature overrode the veto, and the legislative session ended with a projected ending budget balance of zero for the 2002-03 biennium, and an official budget shortfall for the 2004-05 biennium of $1.65 billion (Minnesota Department of Finance, 2002c). The official projected shortfall for 2004-05 was substantially lower than had been projected in February’s planning estimates. But, the decline was not due to an improvement in the economic outlook or ongoing spending reductions. Instead, it reflected a new law prohibiting the Finance Department from adjusting expenditure estimates in the outyears for inflation, except where specifically required by law (Minnesota Department of Finance, 2002c). If expenditures had been fully adjusted for inflation the projected shortfall would have been in excess of $2.1 billion. Over the summer of 2002, the economic and revenue outlook continued to worsen. The Finance Department (2002b) noted that while total revenues were currently on forecast for 2002-03, there were some very troubling signs. After analyzing data on tax year 2001 individual income tax final payments and refunds (due April 15, 2002), tax year 2001 individual income tax liability appeared substantially less than had been forecast. The lower final payments and refunds were reflected in the fiscal year 2002 closing balance, but the longer-term impacts were not. Lowering the base liability for the income tax affects not only the year in which receipts were lower; it affects all future years as well, since the base from which all future projections are made is lower as well. By itself that lower base liability level for the individual income tax would reduce expected revenues for the 2004-05 biennium by more than $800 million, bringing the expected deficit for the 2004-05 biennium to more than $2.1 billion, with no inflation adjustment, even with no change in the economic outlook. And, as noted in the Update, the short-term economic outlook had also weakened. The Finance Department’s November 2002 revenue and expenditure forecast made clear to everyone the size of the state’s unresolved budget problem. In the absence of legislative action Minnesota’s would have a $4.56 billion deficit by the close of the 2005 fiscal year. In less than two years the outlook for the 2004-2005 biennium had gone from one showing a substantial surplus and a healthy structural balance in 2005, to a situation where major changes would be necessary if the state’s finances were to be put back in order. SOURCES OF MINNESOTA’S $4.56 BILLION DEFICIT Not all of Minnesota’s budget deficit came from lower than expected revenues. Spending increases triggered by the weakened economy also contributed. November’s expenditure forecast projected biennial spending at $30.975 billion; under current law with no inflation factored in, the increase of $975 million from end-of-session estimates was 21 percent of the total projected budget deficit. While revenue shortfalls were clearly the major source of the huge projected budget deficit, expenditure projections had also been unduly optimistic. Even though most state expenditures are limited to the amount appropriated, and thus capped at a known level until the next budget is approved, entitlement expenditures proved to be a major source of state budget instability. These expenditures, primarily on health care and income maintenance, tend to amplify state financial problems since caseloads increase substantially when the economy weakens and revenues turn down. Changes to the forecast for state health care costs for needy families alone accounted for $541 million of Minnesota’s $4.5 billion projected deficit for fiscal 2005-05 (Minnesota Department of Finance, 2002c). Lower state revenues were the major cause of the projected deficit, however. Projected individual income tax revenues for fiscal 2004-05 were down $1.333 billion from levels expected at the end of the 2002 legislative session, and total revenues were down $1.673 billion. More dramatically, expected revenues for fiscal 2004 and 2005 had fallen by more than $4.8 billion dollars, or more than15 percent from the level anticipated when the 2001 legislature was making its budget plans. Projected individual income tax revenues alone were down by $2.8 billion, and sales tax revenues by more than $1 billion. The United States economy had entered a recession, there had been a terrorist attack, and the United States was on the brink of war with Iraq. But, even given those largely unforecastable events, the decline in state revenues seemed surprisingly large and out of proportion to the shock to the national economy. Several institutional factors contributed to the large revenue shortfalls observed. Each is discussed in turn below. Overly Optimistic Forecasts of U.S. Economic Growth. State revenues vary with the outlook for the local economy, but each state’s economic outlook depends heavily on the national economy’s performance. Minnesota and many other states contract with national forecasting firms for the national projections needed to drive their state economic models.4 Relying on a national forecasting firm for the national outlook has obvious benefits. It is cost efficient, providing access to forecasts for a larger number of elements of the economy than would be possible if an internally generated forecast were used. More important though, use of a national forecasting firm lends credibility to the revenue forecast since it ensures that key assumptions driving the revenue forecast are based on outside professional judgement and not local politics or decisions by state employees. Minnesota’s first revenue projections for the 2004-05 biennium, the out year revenue planning estimates, were prepared when economic euphoria was at its peak. The economy’s performance had been surprising even optimistic forecasters, with real GDP growing at an average annual rate of 4.3 percent between 1996 and 1999. Late in 2000, it appeared the economy was on track for a year of more than 5 percent growth. That would make it the first time since the early 1960s that real GDP growth rates had exceeded 4 percent for four consecutive years. Unemployment was holding below 5 percent, and yet there were no noticeable inflationary pressures. And, after a two decade long slump, productivity was booming, hinting that further good economic news was likely in the future. There was even media talk of a “new” economy in which the business cycle was obsolete. Every forecaster was saying that this was the best economy our generation would see. And, while no one expected that the extremely strong economic growth would continue forever, there were no obvious signs that it was quickly coming to an end. Now, with the benefit of hindsight, it is clear that the November 2000 national baseline forecast provided by Minnesota’s economic consultant was far too optimistic. Projected real GDP growth rates for 2001, 2002, and particularly 2003, anticipated an economy that was going to continue to grow at an extremely rapid pace for at least an additional three years. The national economy was already in the longest expansion in postwar U.S. history, and the forecast called for three more years of extremely strong growth, without even a pause for consumers to catch their breath. Real GDP in 2003 was projected to grow at an annual rate of 4.8 percent, the strongest growth rate since 1984, and that was to come after four years of four plus percent annual growth followed by the already healthy 3.6 percent and 4.3 percent growth expected in 2001 and 2002. Those growth rates were well above the 2.5 percent growth rate that most economists had previously thought possible on a sustainable basis. Three years of stronger than expected productivity were beginning to convince forecasters that economic growth rates in excess of the 2.5 to 3 percent range could be sustainable without triggering inflation, but there was no reason to suspect that the economy’s sustainable real growth rate had jumped to more than 4 percent, nor was there any pentup demand to drive GDP growth higher. The national economic consulting firm serving Minnesota appears to have bought in to the productivity turn-around story, and made significant adjustments to their forecast in the second half of 2000. Their longer term forecast contained growth rates nearly 50 percent higher than had been projected just a few months earlier. And, for 2003, more than 24 months into the future, the November baseline called for growth at nearly twice the rate that had been thought to be the maximum sustainable rate just a few years earlier. In the forecasting service’s defense, productivity had skyrocketed and the economy had grown faster than anyone had thought possible during the past few years. Still, as the state’s economic advisors and Finance Department economists noted, even though it was not impossible for the economy to continue to grow at an extremely strong pace through 2003 and beyond, a slightly more conservative forecast for two and three years in the future would not have been considered pessimistic. Inaccurate national economic forecasts are an easy target when one is searching for reasons why revenue forecasts miss their mark. Forecasters admit their failings, and no responsible forecaster will claim that his model will successfully identify turning points in the economy. One has only to look at the Wall Street Journal’s semi-annual listing of their forecasting panel’s projections for economic growth during the coming year to see the substantial differences in projections among forecasters, and a quick review of the two-year history of the Blue Chip panel’s consensus forecast illustrates how wide the swings in the general outlook can be.5 Unfortunately, Minnesota’s first set of projections for the 2004-05 biennium was produced at the time when forecasters were feeling the most optimistic about the national economy’s future prospects. By February 2001, the U.S. baseline forecast was slightly more subdued, with projected real GDP growth rates of 2.1, 4.0, and 4.8 percent growth for 2001 through 2003 (Minnesota Department of Finance, 2001a, p. 56). But that forecast was still overly optimistic and by early September 2001 the economic outlook had weakened sufficiently, Minnesota would have faced budget difficulties, even had there been no terrorist attacks. The actions taken during the 2001 special session to cushion the state’s finances against potential revenue shortfalls might have been sufficient to deal with the losses brought on revenue by the weakening economy pre-September 11. But the revenue losses coming from layering the terrorist attacks and subsequent events on top of an already struggling economy overwhelmed Minnesota’s earlier efforts to provide for the possibility of weaker than projected revenue growth. Econometricians know that even well constructed economic forecasting models have random errors that will lead to differences between projected values and actual results. They also know that while the mean forecast error of a particular model may be zero, in any particular year the observed error can be substantial, and even the best model can over-forecast or under-forecast for several years in a row. But even with those caveats, the national economic forecasts provided clients during late 2000 and early 2001 appear to have been overly optimistic, particularly for 2002 and 2003. It appears national economic forecasters all began to believe existing structural relationships in the economy, particularly with respect to productivity, had changed dramatically, and that that change would affect real GDP growth rates far into the future. From the point of view of state revenue forecasts, a more circumspect set of expectations about future growth would have been desirable. The optimistic national economic forecasts were accompanied by equally optimistic stock market forecasts. Everyone knows it is impossible to consistently predict the stock market’s performance. But, historical data on the stock market is often useful in constructing models explaining economic activity and growth in many sectors of the economy. Because their clients use stock market data in their own internal forecasts, many national forecast services include a stock market forecast with their economic forecast. In Minnesota the state’s model of capital gains realizations makes use of a stock market forecast. And, as noted earlier, capital gains realizations are an important part of taxable income, so the stock market forecast affects the revenue forecast. Through the late 1990s the stock market forecast of Minnesota’s national economic consultant had been quite conservative. The valuation model used did not come close to matching the performance of the market during the late 1990s. We now know that the market peaked in March of 2000, but in November 2000, it appears the forecasting service had given up on their former stock market model. The S&P 500 stock index was projected to average 2051 in 2005, up from 1457 in 2000. The implied compound growth rate of just over 7 percent was not unreasonable, had it not been coming from what now in retrospect is surely a classic market bubble. Unfortunately the market bubble had already begun to collapse and stock market gains were to be turned into losses over the next three years. In late June, 2004 the S&P 500 index stood at just over 1140, or about 78 percent of the level observed in 2000. Forecasters know there will always be unexpected shocks that will accelerate or slow economic growth temporarily. Because the number and timing of those shocks are unknown, and because those shocks can bring bad news as well as good news, revenue forecasters are uncomfortable when the three to five year forecast calls for growth rates averaging near the top of the sustainable range. Implicitly such forecasts are building in all potential good news, and dismissing the possibility of bad news for an extended period. Still, when a respected national forecasting service is making those projections and there has been a series of surpluses in the past, choosing to use lower projections is not feasible without damaging the credibility of the revenue forecast. The optimistic national forecast may come true, but should the economy under perform projections early in the forecast horizon, relatively small changes in the economic growth rates in year one or year two can easily compound into large declines in projected revenues by year five. The arithmetic is similar to the well-known effects of changes in base year spending on future spending. More conservative forecasts for years four and five can limit some of the damage done by a rosy scenario early in the forecast horizon, but seldom can they fully offset the revenue losses caused by an overly optimistic forecast for early years of the forecast horizon. Optimistic economic and stock market forecasts were major factors in the change in the revenue outlook, but they are not the whole explanation by any means. As recessions go, the recession of 2001 was not particularly severe, and its impact on state revenues was far out of proportion to the decline in the national economy. Revenue losses were intensified by the substantial declines in the stock market (stock market gains and losses are not included in GDP), but even with that taken into account the revenue losses seem excessive. More conservative national economic and stock market outlooks certainly would have reduced the difference between forecast and actual receipts, but other factors contributed to the revenue loss as well. Inadequate, Untimely, Incomplete, and Inaccurate Data Much of the responsibility for recent state and federal revenue shortfalls should be assigned to the unduly optimistic national economic projections underpinning the revenue forecasts. But, while it is easy to criticize the results generated by forecasting models, the problems faced by revenue forecasters go much deeper. Without good contemporaneous data, not even the best model will yield timely, accurate forecasts. The national economic data produced by U.S. statistical agencies is the best and most complete in the world, but that data is still inadequate and untimely. Eventually, after data has been released, revised, re- revised, and benchmarked, a reasonably complete picture of what has gone on in the economy is available. But, that refining process takes years. It is important to have an accurate historical record of the economy’s performance, but by the time good data is finally available the turning point is long past and forecasters are searching for data releases for signs of the next decline. Individuals unfamiliar with the national income accounts may be surprised by how much the official estimates of the economy for any particular time period change. For example, in November 2000 when Minnesota first prepared estimates for 2004-05 revenues, real GDP for the recently completed the third quarter of 2000, was officially reported to be $9.382 trillion, an increase of 2.7 percent at an annual rate over the level observed in the second quarter of 2000. By February 2001 real GDP for that same quarter had been revised to $9.369 trillion, a growth of 2.2 percent at an annual rate over the second quarter. The June 2001 revisions showed even lower real GDP growth, just 1.3 percent. And today, after benchmark revisions, real GDP in the third quarter of 2000 is reported to have actually declined by 0.5 percent. Had forecasters known in late 2000 that the economy not only had not grown at an annual rate of 2.7 percent during the previous quarter, but actually declined, their late 2000 and early 2001 forecasts would almost certainly have been less optimistic. Forecasters know they work with flawed data, and that future revisions coming next month or next year may change their view of how well the economy is faring. Policy makers and the public, however, seldom appreciate how much the recent history upon which forecasts are based can change. Because they lack that background, they are not fully aware of the risk to the forecast. Some data series are reported relatively quickly, but most economic data are available only after lags of a month or more. Almost all key series are subject to substantial and sizeable revisions, and the most contemporaneous data is often, as would be suspected, the most heavily revised. The data series relied on most heavily by short-term forecasters is the monthly payroll employment report. This report offers an important clue about the economy’s recent performance, and it is available with less than a one-month lag. Unfortunately, that data is revised twice more, often to a significant degree, before a final monthly number is available. And, that final estimate is again only temporary, since it is also readjusted in an annual benchmark revision. Random revisions, while troubling, can be dealt with more easily by forecasters and modelers than can systematic biases. Unfortunately, revisions to the payroll employment estimates around turning points in the economy are pro-cyclical. Because the original data are obtained from a sample of existing firms, jobs created by new firms go uncounted since by definition those firms are not in the survey’s sample frame. That bias can easily be adjusted for when the economy is performing at its normal rate, but when growth slows, the number of new firms creating jobs also slows and when the economy booms, the number of new firms surges as well. Calibrating a bias adjustment to various phases of the economic cycle would be possible if it were known where in the cycle we were, but since this data is the first indicator of how the economy is performing, the bias adjustment often misses the mark and yield misleading results. The flawed bias adjustment means that when the economy is weakening, initial estimates of payroll employment tend to overstate the growth in the economy, and when the economy is showing signs of strength, later revisions will add to the employment estimates. While the biases in the data are known, they remain barriers to rapid recognition of changes in the economy’s progress. For example, initial Department of Labor reports showed payroll employment increasing slightly in each month between February and July of 2002, reinforcing the general impression that the economy was coming out of recession. Now the official data shows a loss of employment in four of the first seven months of 2002, with total payroll employment shrinking by 364,000 jobs. Forecasters using that set of information, rather than the information available at the time, would almost certainly have been less optimistic in their expectations for both 2002 and 2003. Corporate profit estimates are another example of delayed and inadequate data. Corporate profits affect revenues directly in most states through the corporate profits tax. They also have an indirect impact by shaping perceptions about the health of the national economy. Historical data on before tax corporate profits are available in the national income accounts with just over a one-quarter lag, but those estimates are subject to substantial revisions. Ultimately corporate profits can only be determined from tax filings, and most firms file well after the close of the year in question. Estimated tax bills are paid on a timely basis, but many corporations request and receive extensions to the deadline for submitting the actual tax forms to the IRS. This means that forms containing the data necessary to establish corporate income and profits are not available to the IRS for 10 months or more after the close of the firm’s fiscal year. It is only after two or more years that the national income accounts data on corporate profits reflect actual corporate earnings. Again, at turning points in a business cycle the differences between estimates and actual results can be dramatic. In February 2001, before-tax corporate profits for 2000 were reported to be $929 billion, up 13 percent from 1999. Current estimates are that corporate before-tax profits were $773 billion in 2000, down 0.3 percent from 1999. There are also important lags in the availability of state tax data, particularly individual income tax data. Individual income tax forms are due on April 15 of the year following the tax year under consideration. Many taxpayers take advantage of provisions allowing an automatic extension and delay filing until mid August. Some taxpayers, particularly those with large complicated returns, are allowed further extensions. It is not until the end of the calendar year following the tax year in question that the total tax liability for the prior year, as well as the actual levels of wage income and other non-wage income can be accurately estimated. Thus, in February of 2001, when the forecast actually used for setting budgets for the 2002-03 biennium was made, state income tax liability and the composition of taxable income for tax year 2000 was unknown, and could not be known for almost another year. For those sources of income that had been showing rapid growth, such as capital gains and pension incomes, more current data would certainly have been welcome and could conceivably have improved the forecast. But the nature of income tax filing makes that impossible. Imperfect Models Forecasts come from econometric models, systems of statistically estimated equations that describe economic relationships identified from historical data. Those models then have to be driven with a set of assumptions about future values for key variables. For national economic forecasts those assumptions are part of a scenario specified by the forecasting service. Each scenario typically includes assumptions about items influencing the economy such as Federal Reserve policy, global economic growth rates, federal budget activity, oil prices and the underlying rate of productivity growth. State economic models typically rely on the national economic projections provided by their forecasting service for future values of needed variables, and state revenue models use results from the state economic forecast. Typically the standard error of an equation is used to measure the variation of output, and results within two standard errors of the expected value are seen to be within the range of the model. Obviously, the larger the standard error, the larger the potential variation in results, so when evaluating forecasting equations one seeks to have as small a standard error as possible. Unfortunately, some components of the tax system simply cannot be forecast with the same precision as other items. And, some of the income types with large standard errors are important components of the individual income tax base. Sometimes a large error in one component of income can be cancelled by offsetting errors in another source of income, but when errors are reinforcing and not offsetting, large swings in projected revenues can occur. The better the model, the more limited the likely range of outcomes, but even with good, well specified models, actual results often differ substantially from projections, particularly with sources of individual income and corporate income tax receipts. Capital gains – profits from the sales of assets – illustrate some of the problems in modeling income tax receipts. Capital gains were an important and growing item of income in the late 1990s, with net realizations reaching $644 billion nationally in 2000, more than 250 percent above the 1995 level (Congressional Budget Office, 2004, p. 82). In Minnesota growth had been similar and for tax year 2000 net taxable capital gains realizations of Minnesota residents reached just over $9 billion (Minnesota Department of Finance, 2002c, p. 11).6 In February 2001 revenue forecasters did not know the level of capital gains realizations for tax year 2000. The most recent data available was for tax year 1999. From that base they would have to make projections for capital gains realizations for the just completed tax year 2000, as well as tax years 2001, 2002, and 2003. To produce that forecast Minnesota analysts rely on a single equation model based on an underlying assumption that there is a normal, desired relationship between the level of unrealized capital gain and the total stock of capital assets. That desired ratio of unrealized gains to household equity holdings varies with changes in economic and tax policy variables. Realizations in any single year are an attempt to re-establish that ratio, although like other models of this form only part of the necessary adjustment is done in any particular year. The model has reasonable statistical characteristics and its results are relatively robust, but there are some major problems.7 The model requires a forecast of the total value of household holdings of corporate equities in future years and, equally difficult, an estimate of the level of unrealized capital gains. The combination of a relatively high standard error for the estimating equation, and highly uncertain forecasts for the driving variables creates a situation where there can easily be substantial differences between actual and forecast revenues. Slightly different models are used in other states, and by the Congressional Budget Office, but all face similar problems. Because economists lack both knowledge of the factors influencing taxpayers’ decisions to realize gains and good data on the variables affecting the realization decision, capital gains forecasts will continue to be one of the most problematic portions of state and federal revenue forecasts. And, because capital gains had been increasing as a share of taxable income, revenue forecasts had become increasingly sensitive to changes in the outlook for capital gains. During the late 1990s the incredibly large growth in capital gains realizations (45 percent in tax year 1996, 40 percent in 1997, 25 percent in 1998 and 22 percent in 1999) was largely unanticipated. In Minnesota that growth led to large surpluses, and led policy makers to expect future surpluses. Even though conservative forecasting assumptions were used, when stock market conditions changed, no one was ready for the sudden drop in capital gains revenue that followed. Minnesota’s February 2001 forecast called for no growth in capital gains realizations in 2001. By February 2002, taking advantage of a an additional year’s information and the refitting of the model to just released revisions to the household balance sheet as reported in the Federal Reserve’s Flow of Funds Accounts, net realizations of capital gains were projected in Minnesota to decline by more than 33 percent (Figure 1) FIGURE 1 Minnesota Wage Growth: 1959-2003 [Please reduce the size of the table to fit the size of the journal (L/R Margins = 2”] Ann Pct Chg 4 Qtrs Apart 18 15 12 9 6 3 0 1959 1963 1967 1971 Source: U.S. Department of Commerce, (year? 1975 1979 1983 1987 1991 1995 1999 2003 Legislative leaders greeted the forecast of a decline in capital gains realizations with substantial skepticism. Almost all doubted that capital gains realizations could fall that precipitously. Unfortunately, the forecast turned out to be too optimistic, with capital gains realizations falling by more than 46 percent nationally. The loss in tax base in Minnesota was substantial. Net capital gains realizations by Minnesota resident tax payers fell from $9 billion in tax year 2000 to just over $4 billion in 2001, a decline of more than 55 percent. And, it is important to note that the revenue lost is not just limited to one year. All future changes in projected capital gains realizations will come from that lower base. Assuming no changes in future growth rates, that means that all future forecasts would have $5 billion less in capital gains realizations as well. Over the entire forecast cycle from fiscal 2001 through fiscal 2005, the reduction in capital gains realizations would amount to a reduction of more than $20 billion in taxable income. The stock market bubble and its subsequent bursting created two quite different challenges for modeling. Since economic models are estimated using historical relationships between the variable of concern (in this instance realizations and other variables driving the amount of unrealized gain) when those values go outside the range of the historical data the predictive power of the model is less certain. During the upward turn of the bubble, the growth in unrealized gains was exceeding the range of prior observations, so there was no suitable reference point to compare against. Then, once the bubble burst, there was an equally difficult problem to model: How much of the paper losses accumulating would be realized immediately, and how much would be left to ride out the slump. Since only $3000 of long term capital losses can be applied against ordinary income (but an unlimited amount can be used to offset long term capital gains) the existence of these sizeable capital losses, realized or not, will allow much of the population to offset all of their capital gains, plus a small amount of ordinary income for several years into the future, reducing future taxable income by a substantial, if unknown amount. Changes in the Structure of the Economy During the late 1990s forecasters noticed that withholding receipts were growing rapidly, indicating unusually large increases in nominal wages. With the low inflation rates of the time that meant real wages of workers were growing substantially as well. That was taken as just one more sign that a “new” economy was in place, one where the benefits from productivity increases were being shared with workers. Some of that was true: hourly wages of production workers did increase, but that was not sufficient to explain all the wage growth that occurred. In retrospect it is clear that the use of performance-based compensation, principally bonuses and options, was increasing, and growing more rapidly than base salary. That change had important implications for state revenues, but it went largely unnoticed. There had been an indication of the importance of options and bonuses in late 1992, when there were reports that high income taxpayers expecting a top bracket tax rate increase were seeking to minimize their tax liability by shifting bonus payments into 1992 from early 1993, and exercising in the money stock options. That activity produced an unusual spike in wages in the fourth quarter of 1992, followed by a steep decline in the first quarter of 1993, providing some of the first evidence of the amount of income received through end-of-year bonuses and the exercise of options. By the late 1990s corporate America’s desires to tie compensation more closely to productivity, coupled with the financial advantages options offered to both the firm and the employee, appears to have led to an even greater percentage of compensation being tied to performance based measures. Wage growth rates of more than 8 percent were seen in some quarters, a surprising event given that employment growth and inflation were much more modest. Revenue forecasters now know that much of that wage growth came from bonuses and the exercise of options and bonuses, not by growth in base salaries. The money generated was real, but as was proven in 2001, it was not permanent and it was not something that long term spending plans could be based on. When the economy weakened, nominal wage growth slowed dramatically. Nationally, the growth rate for nominal wages, four quarters apart, has been less than 2 percent just 20 times since 1960, 10 of those times have occurred since the third quarter of 2001. In Minnesota the results were even more dramatic. Since 1960 nominal wage growth four quarters apart fell below 3 percent only once between 1960 and the third quarter of 2001 (Figure 2) Since then, nominal wage growth four quarters apart has not risen above 3 percent. The increased importance of performance based compensation, something that had gone largely unrecognized, created an unexpected break in the structure of wage estimates. That break in structure, caused wage forecasts to be overly optimistic, and contributed greatly to the negative revenue variances observed during the period. Wages, of course, are the single largest component of the income tax base, accounting for more than 75 percent of adjusted gross income for Minnesota resident taxpayers in 2001. Minnesota forecasters recognized this problem in late 2001 and by February of 2001 they were incorporating performance based compensation estimates in the forecast. But the damage was already done since spending plans for 2002-03 had been based on the outlook before the precarious nature of the changes in wages had been recognized. There is every reason to believe that the performance based compensation proportion of total compensation will continue to grow. It is popular with firms because it provides wage flexibility and ties compensation to the firm’s performance. For workers, the increased volatility in pay is offset by the possibility of substantial boosts in compensation when times are good. For revenue forecasters and policy makers, however, this is not good news since it increases the volatility of the revenue stream – increasing revenues even more when times are good, and depressing them when the economy turns down. FIGURE 2 Minnesota Wage Growth: 1959-2003 [Please reduce the size of the table to fit the size of the journal (L/R Margins = 2”] [Table 1 and 2 are the same. Please check] Ann Pct Chg 4 Qtrs Apart 18 15 12 9 6 3 0 1959 1963 1967 1971 1975 1979 1983 1987 1991 Source: U.S. Department of Commerce (year) CONCLUSIONS Minnesota and other states have faced significant budget challenges after revenues failed to meet expectations following the 2001 recession. An overly optimistic national economic forecast was responsible for much of the revenue shortfall, but policy makers cannot realistically expect future economic forecasts to be sufficiently accurate, or so conservative that similar shortfalls will not occur in the future. Revenue forecasting is imprecise under the best of circumstances. National economic projections are often flawed, the data delayed and subject to 1995 1999 2003 substantial revisions, and economists understanding of factors underlying household behavior is often insufficient to produce models with small margins of error. When the possibility of undetected structural changes in the economy is added in, the confidence interval surrounding estimates in any particular revenue forecast expands further. Substantial reserves will not prevent the next budget deficit. But, if used responsibly, they will ease the adjustments necessary until revenues recover again. NOTES 1. All of the Finance Department’s forecast reports cited are available at www.finance.state.mn.us. 2. A more complete description of Minnesota’s forecast process is available in Stinson, 2002. Each of the Department of Finance’s forecast releases also contains a brief section, “Forecast Fundamentals” which describes forecast procedures. 3. A more complete description of the role Minnesota’s Council of Economic Advisors plays is provided in Stinson, 2002. 4. Typically state economic models are used to adjust the externally provided national economic forecast to reflect the specific structure of the state’s economy and the way employment and income in particular sectors have historically responded to national economic growth. For example, GDP growth driven by expanded oil and gas production would have little impact in Minnesota because Minnesota has no oil exploration and production activity. And, the number of individuals employed in the production of non-electrical machinery may be less volatile in Minnesota than nationally due to differences in the nature of contracts and productivity for Minnesota firms. 5. The Blue Chip Consensus Forecast for 2001 in January 2000 called for 3.0 percent real GDP growth. The panel grew more optimistic as 2000 went on and by October real GDP growth at an annual rate of 3.5 percent was projected. By January 2001, the consensus forecast had fallen to 2.6 percent, and by October 2001, 1.1 percent. (Blue Chip Economic Indicators, 2001, p. 4) The U.S. Department of Commerce now is reporting that the actual growth rate for real GDP in calendar 2001 was 0.5 percent. 6. The Minnesota capital gains data reported below differs slightly from the SOI data reported in Table 1. This data comes from a sample of Minnesota resident tax filers used to calibrate Minnesota’s income tax micro-simulation model. The SOI data includes income from part-year Minnesota residents whose filing address is in Minnesota. 7. Minnesota’s capital gains model is outlined in November Forecast 1997, pp 27-29. REFERENCES Blue Chip Economic Indicators (2001, October). Kansas City, MO: Aspen Publishing. Congressional Budget Office (2001, January). The Budget and Economic Outlook: Fiscal Years 2002-2011. Washington, DC: Author. Congressional Budget Office (2004, January). The Budget and Economic Outlook: Fiscal Years 2005-20141. Washington, DC: Author. Department of Commerce (year). Title of publication as cited in Figures 1 & 2. Minnesota Department of Finance (1977) November Forecast, Name of City, MN: Author. Minnesota Department of Finance (2000, November). November Forecast. Name of City, MN: Author. Minnesota Department of Finance (2001a, February). February Forecast. Name of City, MN: Author. Minnesota Department of Finance (2001b, October). Economic Update. Name of City, MN: Author. Minnesota Department of Finance (2001c, November). Fund Balance, General Fund. Name of City, MN: Author. Minnesota Department of Finance (2002a, February). February Forecast. Name of City, MN: Author. Minnesota Department of Finance (2002b, October). Economic Update. Name of City, MN: Author. Minnesota Department of Finance (2002c, November). November Forecast. Name of City, MN: Author. Minnesota Department of Finance, 2003 [Cited in text] Statistics of Income Bulletin (2002; 2003). [Cited in text. Need full references here] Stinson, T. F. (2002, June). “State Revenue Forecasting: An Institutional Framework.” Government Finance Review, Vol # (Issue #): 12-15.