POLITICIZATION, REPUTATION COSTS, AND MACROECONOMIC FORECASTS

High-quality macroeconomic and fiscal forecasts are often requisite for both efficiently allocating scarce public resources and facilitating effective public and private sector planning. If government forecasts are too optimistic, then the federal government will have unanticipated declines in revenue and unanticipated demands for services and transfers. If private sector agents rely on government forecasts to allocate resources or anticipate government policy choices, then poor government forecasts will result in suboptimal decision making by private sector actors.

Although political scientists have recently begun to study the quality of U.S. federal macroeconomic and fiscal projections (Engstrom and Kernell 1999; Corder 2005; Krause and Douglas 2005, 2006), scholars have yet to focus on the intertemporal aspects of these agency decisions. Public agencies are compelled to weigh the current benefits of providing good news to incumbents versus long-run damage to their reputation for policy competence if forecasts are too optimistic. Incumbent politicians generally prefer that agencies provide optimistic estimates of future policies and conditions as a way to signal voters that the incumbent is a competent manager (Weatherford 1987; Keech 1995; Alesina and Roubini [with Cohen] 1997; Carlsen 1999). Incumbent politicians also wish to run on “good news” since an election is a referendum on their past performance in office (Fiorina 1981). Executive branch agencies thus possess tangible incentives to bias their forecasts towards serving a president's interests at the expense of objective quality.³

Presidents can sanction executive agencies that fail to produce sufficiently optimistic forecasts by proposing a cut in the agency's budget, marginalizing or removing political executives, or advocating agency reorganization or outright termination. For instance, Mike Parker, Director of the Army Corps of Engineers in the Bush II administration, testified before Congress that proposed administration budget cuts would have adverse consequences for agency performance. Following his testimony, Parker was presented with a choice between tendering his immediate resignation or facing outright termination (Christian Science Monitor December 17, 2002).

Reputation costs can, however, constrain agency forecast optimism in several ways. First, professional norms influence bureaucratic behavior (Wilson 1989). If, for example, government economists value the esteem of their peers and act in ways that maintain their professional reputation, they will resist efforts by the White House to produce optimistic forecasts. Second, possessing a reputation for bureaucratic competence enhances an executive agency's credibility as an honest broker of information (Heclo 1975; Rourke 1992; Carpenter 2001).⁴

For instance, if the electorate is to view the incumbent president as a competent economic manager, then executive branch macroeconomic forecasts must be viewed as credible sources of information for voters.⁵ Finally, the erosion of reputation can result in budgetary or auditing sanctions that place agencies in jeopardy (Bendor, Taylor, and Van Gaalen 1985; Banks and Weingast 1992). Overall, agencies suffer if they consistently err on the side of forecast optimism.⁶

In the absence of reputation costs pertaining to bureaucratic competence, agencies possess a strong incentive to produce the most highly optimistic forecasts feasible. In the presence of such reputation costs, agencies must balance the risks and rewards associated with forecast optimism. We argue that public agencies routinely face political pressure to produce optimistic forecasts, and that these reputation costs impose a constraint on their intertemporal decisions. Next, we present a theory that explains how variable reputation costs influence forecasting decisions made by public agencies when strong political incentives exist for forecast optimism.

BUREAUCRATIC FORECAST OPTIMISM AND INTERTEMPORAL DECISION MAKING

The level of forecast optimism tolerated by an agency is a function of the present value of discounted future reputation costs.⁷

These costs pertain to the loss of an agency's favorable reputation when providing information that is influenced by the desire of elected officials for “good news.” When agencies place a premium on minimizing reputation costs, they will be more concerned with maintaining credibility than with accommodating politicians' demands. Our theory predicts that, for a given agency, reputation costs will be discounted more heavily for forecasts involving longer forecast horizons. Because of decline in organizational memory attributable to personnel turnover, it is more difficult to sanction an agency for its more distant past decisions compared to its more recent past decisions. This leads to our first hypothesis: In other words, the political benefits corresponding to forecast optimism will outweigh future reputation costs as the forecast horizon increases.

We also maintain that the tradeoff between political pressures and reputation costs will vary according to an agency's level of organizational stability. An agency's level of organizational stability is largely determined by its institutional design. Agencies exhibiting high levels of organizational stability are characterized by low personnel turnover and considerable autonomy from political influence. Because it is easier to trace the source of decisions in highly stable agencies back to specific individuals or units, these reputation costs are discounted at a lower rate. Stable agencies also possess a greater incentive to build a durable reputation for bureaucratic competence than less stable agencies. This, in turn, can enable an agency to preserve, or even expand, its future level of political independence (Wilson 1989, Chapter 10). An agency that places greater weight on maintaining a favorable reputation for bureaucratic competence than accommodating short-run political pressures enhances its autonomy in the long run (Carpenter 2001). In contrast, less stable agencies often rely on close ties to elected officials and retain only low levels of discretionary authority. In addition, unstable agencies experience greater personnel turnover across both political appointee and civil servant positions (Lewis 2003), reducing both organizational memory and bureaucratic accountability. Because it is more difficult to blame specific individuals or units for poor decisions, less stable agencies can produce more optimistic forecasts than more stable agencies. Thus, our second hypothesis is:

H2 states that an agency's level of forecast optimism is inversely related to its degree of organizational stability.⁸ Because a less stable agency's reputation costs are discounted at a higher rate compared to stable agency, a more stable agency will be less sensitive to short-term political pressures to produce optimistic forecasts than a less stable agency.

We can also expect that organizational stability will exert a dynamic impact on forecast optimism as the forecast horizon increases. Specifically, we posit that an “optimism gap” will emerge between less stable and more stable agencies as the forecast horizon increases. The theoretical mechanism underlying this phenomenon is simple. For a fixed discount rate, the present value of discounted reputation costs associated with low bureaucratic competence in distant future periods should decline at a faster rate for an unstable agency compared to a stable agency. As the forecast horizon extends into the future, the unstable agency should exhibit a successively greater level of forecast optimism relative to that exhibited by the stable agency. Our third theoretical hypothesis is:

Put simply, H3 states that as the forecast horizon increases, a less stable agency will make increasingly more optimistic forecasts relative to a more stable counterpart.

A summary of our theory and its empirical implications are in order. Our basic claim is that public agencies possess an incentive to accept greater forecast optimism as reputation costs are more heavily discounted, ceteris paribus.⁹

Reputation costs will be discounted more heavily for forecasting decisions involving longer forecast horizons. Therefore, the level of forecast optimism associated with a given agency forecast will rise as the forecast horizon increases (H1). Reputation costs will also be discounted more steeply by less stable agencies. Accordingly, we predict that the level of forecast optimism will be greater for less stable agencies vis-a-vìs more stable agencies at a given forecast horizon (H2). This difference in discounted reputation costs, and the resulting gap in forecast optimism will grow between less stable and more stable agencies as the forecast horizon increases (H3). Next, we test our theory with macroeconomic forecast data produced by the OMB and SSA.

AN APPLICATION TO U.S. FEDERAL EXECUTIVE AGENCY MACROECONOMIC FORECASTS

Macroeconomic forecasting within federal executive agencies serves as a suitable empirical laboratory for analyzing how intertemporal decisions might be affected by the discounting of reputation costs. OMB and SSA publish forecasts of identical macroeconomic indicators. Further, these forecasts are directly comparable because the agencies follow similar reporting conventions. OMB and SSA each annually publish macroeconomic projections for current year outcomes and extend the forecast to 5 years into the future.¹⁰

OMB and SSA macroeconomic forecasts provide a natural experiment to test our theory because both agencies are housed within the executive branch, but have different levels of organizational stability. These differences permit a test of specific expectations about the link between intertemporal decision making and organizational stability. OMB is located directly in the Executive Office of the President and functions as a political support agency for the White House. The SSA, specifically the Office of the Chief Actuary, is more insulated from the president. SSA was part of a cabinet agency from 1953 to 1994 (initially, Health Education and Welfare and, later, Heath and Human Services), but the Office of Chief Actuary maintained a close direct working relationship with the congressional committees overseeing Social Security and Medicare (Rosenblatt and DeWitt 2004). In addition to differences in agency insulation, the rate of leadership turnover differs among they key agency executives largely responsible for these macroeconomic forecasts. The average OMB director has tenure of a little over 2 years: 11 directors were appointed between 1981 and 2005 (OMB 2005). The Office of the Chief Actuary experiences remarkably low turnover: since 1981 two Chief Actuaries have directed the office. Robert J. Myers held the post of Chief Actuary from 1945 to 1970 and established many of the norms of professionalism and independence that characterize this particular office (Rosenblatt and DeWitt 2004).

A recognized tension exists between the optimistic biases of elected officials and the credibility of the Office of Chief Actuary. “The political leaders, whether Democrat or Republican, believe that their economic and fiscal policies will produce positive results in the short term. They would like the trustees reports to reflect their optimism” (Former Chief Actuary Harry Ballantyne quoted in Rosenblatt and DeWitt 2004, 4) We maintain that the relatively high level of organizational stability afforded SSA vis-a-vìs OMB by its institutional design translates into a smaller discount rate for reputation costs within SSA. OMB should thus be more accommodating of politicians' desire for optimistic macroeconomic forecasts relative to SSA.

Our research design allows us to analyze a rich data set consisting of multiple comparable indicators produced by a pair of executive branch agencies for varying forecast horizons. Our statistical analysis examines both OMB and SSA forecasts for three U.S. national macroeconomic variables: the annual percentage growth rate of real Gross Domestic Product (GDP), the inflation rate based on the annual percentage growth of the Consumer Price Index (CPI), and the unemployment rate (reported as a calendar year average).¹¹

SSA macroeconomic forecasts are embedded in a larger forecasting enterprise, which includes demographic forecasts, and are ultimately designed to assist in constructing estimates of the long-term financial health of the trust funds that support Social Security and Medicare. The SSA produces several alternative economic assumptions and publishes optimistic, pessimistic and intermediate projections every year. The intermediate projections are used for the statistical analysis because they reflect the “Trustees' best estimate of future experience” (SSA 2005, 14).

One might presume that SSA macroeconomic forecasts are less optimistic than OMB's because SSA experiences more serious policy consequences than OMB for optimistic forecast errors. SSA forecasts are closely linked to a core mission—keeping the social security trust fund solvent for the long haul. An unanticipated depletion of social security trust funds can arise if SSA produces overly optimistic projections of real output growth or unemployment.¹²

This is a particular concern for inflation forecasts since Social Security benefits are indexed to the price level. We contend, however, that both executive branch agencies confront equally serious consequences regarding macroeconomic forecast optimism. For example, optimistic OMB forecasts of real output growth and unemployment might result in negative policy repercussions ranging from overly restrictive monetary and fiscal policies that induce an economic downturn to understating future demand for social welfare programs or the true magnitude of fiscal deficits. The basic difference between these agencies is simple. Current OMB personnel are not likely to bear the consequences of long-term decisions given the lower level of organizational stability in the agency. The SSA is likely to bear much greater reputation costs for long-term decisions because it is more difficult for agency personnel to escape future blame for poor performance. Observed differences in forecast optimism between OMB and SSA reveal how these executive agencies balance political pressure with the desire to maintain a reputation for bureaucratic competence. Next, we discuss model specification issues and the econometric testing of our theory.

MODEL SPECIFICATION AND ECONOMETRIC TESTING

The three empirically testable hypotheses derived from our theory are investigated using OMB and SSA annual forecast data on U.S. real GDP growth, inflation, and the unemployment rate for 1979 to 2003. We propose that an agency's level of forecast optimism can be measured as its forecast error. Each forecast error is specific to a given macroeconomic variable, forecast horizon, agency, and calendar year. This can be represented by the general two-equation system,

where the forecast error dependent variable in each equation is a function of a series of dummies representing forecast horizons for successive years (h = 1, h = 2, …, h = 5), a vector of control variables (j = 1, j = 2, ……, j = 6), and a stochastic disturbance term. To preserve positive forecast errors as an indicator of forecast optimism, real GDP growth forecast errors appearing in (1)–(2) are appropriately amended by switching the order of the forecast and actual values for the dependent variable such that the terms on the left hand side of the equal sign are

, and

. We pool all of the forecasts (h = 0, ……, h = 5) made in a given year for the 1979 to 2003 period. Each agency equation contains 137 usable observations for the real GDP growth model, and 139 usable observations for both the inflation and unemployment rate models.

Our statistical control variables account for the independent effects of staffing politicization within each agency, the president's party, the presence of divided government, presidential elections, and past macroeconomic conditions. Staffing politicization is measured as the annual percentage of political appointees within each agency at the staff levels for each executive agency.¹³

Because agency forecast optimism is motivated by political considerations, we predict that staffing politicization will be positively related to forecast optimism. Therefore, any differences that we observe between OMB and SSA will be independent of the percentage of political appointees located within each agency at any given time. We account for partisan differences between administrations with a dummy variable that equals 1 for a Democratic president, 0 for a Republican president. We expect Democratic administrations to provide more optimistic real output growth and unemployment rate forecasts and for Republican administrations to provide more optimistic inflation rate forecasts, consistent with the preferences held by the parties' core constituent groups (e.g., Keech 1995; Alesina and Roubini [with Cohen] 1997). We also include a dummy variable for divided party government. Proponents of the creation of the CBO noted that it could contest and challenge White House numbers and assumptions (Engstrom and Kernell 1999); thus we expect that executive branch forecasts will be less optimistic under conditions of divided party control of the White House and Congress. We also expect to observe greater forecast optimism during presidential election years because the incumbent administration can benefit from favorable economic news. A dummy variable for presidential election year is incorporated into the empirical models to test for this possible effect. Finally, we incorporate a pair of macroeconomic conditions as independent variables in each model specification—the 3-year lagged moving average and standard deviation of the actual macroeconomic outcome of interest. These statistical controls account for the impact of past observed macroeconomic conditions and volatility on forecast optimism.

We estimate our two-equation system via the Seemingly Unrelated Regression (SUR) estimator (Zellner 1962). This statistical technique has two important advantages for our particular research design. First, we wish to directly compare OMB and SSA, yet treat them as conceptually distinct agencies with varying behavior and responses to exogenous political and economic conditions. The SUR estimator also permits us to test cross-equation coefficient equality restrictions across agencies, which are useful for statistically analyzing both static and dynamic orgainzational stability hypotheses (H2 and H3). Also, the SUR estimator increases statistical efficiency by exploiting the cross-equation correlation of contemporaneous residuals. Useful information can be gleaned from analyzing each model's contemporaneous residuals—that is, the forecast error shocks experienced by OMB and SSA. We have strong reason to believe that these forecast error shocks are highly correlated. Similar responses to exogenous shocks might be attributable to either shared unobserved forecasting technologies or commonly held assumptions about future conditions. This type of shared agency response is likely since bureaus possess a strong incentive to avoid being singled out as the inferior agent when output quality can be easily observed (e.g., Kunioka and Rothenberg 1993), and also wish to reduce information costs in the presence of bureau competition (Miller and Moe 1983). It is thus rather plausible that similar responses to exogenous shocks are indicative of private information sharing between these agencies (Krause and Douglas 2006).¹⁴

Because we are concerned with several possible violations of least squares' assumptions involving the disturbance term, we calculate bootstrapped standard errors in all of our statistical analyses. These bootstrapped estimates are robust to heteroskedasticity and nonnormality. This approach also handles the moving average error processes induced by an overlapping data problem—specifically, the relationship between regression disturbances and the forecast horizon (Hansen and Hodrick 1980). The bootstrapped standard errors will not only yield more accurate inferential tests of our statistical hypotheses compared to those generated from SUR-GLS estimates, but are also superior to those which correct for any single statistical problem.¹⁵

EMPIRICAL FINDINGS

The SUR regression results appear in Table 1. The first five rows in the table report coefficients on the forecast horizon dummy variables. The estimates indicate that OMB and SSA forecasts of unemployment rates include positive forecast errors at each forecast horizon. The OMB forecast of the inflation rate also contains positive errors at each time horizon. Both variables reveal positive forecast errors, net of other exogenous influences, consistent with the desire of elected officials to report good news. But, unlike OMB, SSA inflation forecast errors are significantly different from zero only for the current year (h = 0: α₀ = −1.18). SSA forecasts at distant time horizons are less optimistic. Further, and surprisingly, both executive branch agencies' forecast errors for real GDP growth are negative, and thus pessimistic in absolute terms. OMB's real GDP growth forecast errors are not statistically distinguishable from zero (except for the current-year forecast), whereas SSA forecasts contain marginally significant pessimistic bias for only longer time horizons (h = 4: α₄ = −1.03 and h = 5: α₅ = −1.04). One possible explanation for these results is that real GDP growth might be less tangible to voters than unemployment or inflation; thus incentives for OMB and SSA to generate optimistic forecasts for this variable are low.

The statistical control variables also uncover several interesting empirical findings. For instance, staffing politicization generally has no discernible bearing on each executive agency's forecast biases. In only a single instance, SSA unemployment forecasts, do we observe a significant effect of staffing politicization. In this particular case, however, the sign of the coefficient is inconsistent with our expectations. Higher staffing politicization actually led to lower SSA forecast optimism. Moreover, the standard difference in staffing politicization's impact between OMB and SSA is negligible.¹⁶

Contrary to conventional expectations, both executive branch agencies generate less optimistic forecasts of real output growth and unemployment rates under Democratic presidents. For OMB, the forecast optimism for real GDP growth and unemployment rates are lower by about 1.13% and 0.86% per annum in absolute terms, respectively. The effects of party are greater for OMB forecasts than SSA forecasts. Wald tests indicate the difference between the agencies is statistically significant (real GDP growth rate model: χ² ≈(1) = 37.37, p = 0.00; unemployment rate model: χ² ∼(1) = 40.72, p = 0.00).¹⁷

These findings do not suggest that OMB generates forecasts favorable to the core constituencies of the party controlling the White House. When a Democrat is in the White House, the OMB does not forecast lower unemployment or higher economic growth. It could be the case that Democrats prefer less optimistic forecasts of real output growth and unemployment to retain a compelling justification for entitlement and social insurance programs that comprise the social safety net. Alternatively, Democratic administrations' might just simply do a poor job signaling competent macroeconomic management.¹⁸

Consistent with our expectations, we obtain some empirical evidence that divided party government is associated with a lower level of agency forecast optimism. Under divided party government, forecast optimism is lower for SSA real GDP growth projections and both OMB and SSA unemployment rate projections. The effects of divided party government are greater for OMB unemployment rate forecasts than for SSA. OMB (SSA) forecasts of the unemployment rate are typically 1.05% (0.67%) lower under divided government compared to unified government. The Wald test indicates the difference between the agencies is statistically significant (χ² ∼(1) = 21.06, p = 0.00). The existence of divided party government fails to affect inflation forecast biases for either agency. We find no evidence indicating that any of these executive branch agency forecasts are influenced by the timing of presidential elections.¹⁹

Finally, the levels of recent past observed real GDP growth and inflation have a significant influence on the level of forecast error in both agencies. Higher levels of past observed real GDP growth leads to higher output growth forecast optimism for both SSA and OMB. Similarly, lower levels of past observed inflation tends to have a positive effect on the level of forecast optimism for both SSA and OMB. Such adaptive behavior suggests that recent recent macroeconomic conditions do affect the quality of future forecasts. Past volatility of the macroeconomic indicators has no impact on forecast optimism. Thus, OMB and SSA forecasting behavior are risk neutral: neither public agency responds to recent past volatility in macroeconomic conditions when formulating forecasts. The contemporaneous forecast error shocks are highly correlated in a positive direction [

]. This result indicates that OMB and SSA respond very similarly to exogenous shocks.²⁰

The statistical results for the forecast horizon hypothesis (H1) appear in Figure 1. These results assess the difference in each agency's forecast optimism in an exhaustive pairwise manner. Positive changes in agency forecast errors represent an increase in forecast optimism from the baseline forecast horizon in relation to a more distant forecast horizon located on the X-axis. Statistical significance (or lack thereof) is assessed through a Wald test where the null hypothesis is that the change in agency forecast error is equal to zero. The statistical evidence only supports the Forecast Horizon hypothesis (H1) in a few cases. These particular cases involve OMB current year and near-term real GDP growth rate forecasts (Figure 1A) and OMB current year and long-term unemployment rate forecasts (Figure 1C). Specifically, the change in OMB real GDP growth rate forecast errors moves closer to zero from the current-year forecast (h = 0: α₀ = −1.18) for each of the subsequent three forecast horizons (h = 1: α₁ = −0.47, h = 2: α₂ = − 0.41, and h = 3: α₃ = −0.40). The level of OMB unemployment rate forecast optimism increases significantly from the current year forecast (h = 0: α₀ = 2.30) to the long-term forecasts (h = 4: α₄ = 3.00 and h = 5: α₅ = 2.99). As expected, OMB introduces the most optimistic bias in the long-term forecasts of the unemployment rate. One potential explanation for this result is that the unemployment rate may be the most politically tangible indicator of macroeconomic performance to the electorate—so that OMB has a strong incentive to signal low and stable unemployment rates for the long term. Yet, many of the observed differences regarding the change in agency forecast errors are statistically indistinguishable from zero.

Paradoxically, we detect significant reductions in the level of forecast optimism from SSA current year inflation forecasts (h = 0: α₀ = 1.18) to those made two years (h = 2: α₂ = 0.32) through 5 years (h = 5: α₅ = 0.34) ahead into the future (Figure 1D). Why might the SSA wish to avoid generating highly optimistic forecasts of future inflation beyond the current year? Because the SSA has responsibility for administering social security programs, the SSA inflation forecasts might be influenced by the agency explicitly factoring in the automatic Cost-of-Living Adjustments (COLAs) for these programs.²¹

The SSA may be minimizing the probability of unanticipated costs in the form of higher than expected COLAs for Social Security benefits. Although many other U.S. federal government programs of interest to presidents are indexed to inflation, OMB may not possess an equally strong incentive to generate forecasts for high inflation. Our statistical evidence does not reveal any significant drop in OMB inflation forecast optimism as the forecast horizon increases. It is possible that observed differences in OMB and SSA inflation rate forecasts for a given forecast horizon arise from the use of COLAs for Social Security benefits. This potential problem should not affect our empirical tests of H2 or H3 for real GDP growth or unemployment rate forecasts. Nor should our statistical tests of H3 involving inflation rate forecasts necessarily be affected by such considerations.

On a substantive level, the empirical evidence indicates that these executive branch agencies possess an internal conflict between their short-run desire to curry favor with presidents by providing optimistic forecasts and a long-run preference for making less rosy forecasts which enhance their bureaucratic reputation. Although we typically observe an upward swing in forecast optimism for longer–term forecasts (h = 4, h = 5), they are only statistically discernible for OMB forecasts of U.S. unemployment rate (Figure 1C). It could be the case that public agencies' desire to maintain a positive organizational reputation (Carpenter 2001) simply outweighs immediate political pressures to generate optimistic forecasts. This should be especially true in the executive branch insofar that presidential administrations are relatively short-lived. Nonetheless, while both OMB and SSA may care more about possessing a reputation for bureaucratic competence on a general level, they may systematically differ in how much weight they place on this goal vis-a-vìs political responsiveness.

We now turn our attention to the statistical testing of the static organizational stability hypothesis (H2). This hypothesis involves examining differences in OMB-SSA forecast optimism at each fixed forecast horizon. These results appear in Figure 2. The points in the figure are computed by subtracting the SSA forecast horizon dummy coefficient from the OMB forecast horizon dummy coefficient at each forecast horizon. The OMB-SSA forecast error difference is hypothesized to be positive because OMB is the less stable agency. Wald tests allow us to ascertain whether the coefficient differences are statistically different from zero. At a given forecast horizon, our statistical evidence shows that in most cases OMB forecasts are more optimistic than those generated by SSA at p < 0.05. This holds true across all forecast horizons for both unemployment rate forecasts (Figure 2C), and in all but the current-year inflation forecasts (Figure 2B). Support in favor of H2 with respect to real GDP growth rate forecasts is only observed for longer forecast horizons: h = 3, h = 4, and h = 5 (Figure 2A).

Strong statistical support for H2 indicates that OMB discounts reputation costs at a higher rate than SSA. It is interesting to note that these institutional design effects are statistically negligible for current year forecasts of real GDP growth and inflation rates, as found elsewhere (Corder 2005; Krause and Douglas 2005, 2006). This empirical finding is compatible with our theory any difference involving intertemporal discounting of reputation costs by agencies operating under different institutional designs will be minimal in the current period. The ability to sanction agencies for poor performance should only marginally differ across agencies if errors are immediately revealed (Krause and Douglas 2005, 303). Politicians can more easily distinguish between an inferior and superior agency in the immediate period, thus agencies will have a strong incentive to behave similarly to avoid either political sanctions for being singled out as being too pessimistic—or loss of bureaucratic reputation for being singled out as being too optimistic (Krause and Douglas 2006).

The statistical results for the dynamic organizational stability hypothesis (H3) appear in Figure 3. Each entry represents the change in OMB-SSA forecast error differences at varying forecast horizons in relation to a specified baseline forecast horizon. Statistical significance is determined by a series of Wald tests for each of these pairwise combinations. Our statistical evidence shows abundant, if not unanimous, support for H3 across all three macroeconomic forecast variables. In every instance, the graphs clearly show that change in OMB-SSA forecast optimism differences is both positive in magnitude and is positively related to forecast horizon length. Substantively, this means that less stable agencies discount reputation costs at a higher rate than more stable agencies. As the forecast horizon is extended, the difference in the present value discounted reputation costs between these executive branch agencies appears to be amplified. The statistical support for H3 is strongest when comparing cross–agency differences in forecast optimism for current year and longer term forecasts. That is, evidence in favor of H3 is strongest when the forecast horizon grows in relative terms (i.e., the time between the baseline and more distant forecast horizons become greater). In addition, the sharper upward slopes of these estimated forecast error differences for both the inflation and unemployment rate forecasts (Figure 3B and 3C) indicate that differences in agency forecast optimism grow at a faster rate for these indicators than compared to real GDP growth rate forecasts (Figure 3A). Overall, the empirical findings from Figure 3 suggest that OMB-SSA differences with respect to the discounting of their reputation costs for bureaucratic competence has more important implications for long-term forecasts than for short-term projections.