The GASB No. 34 Impact of Budget-to-Actual Variances on Bond Ratings: Evidence from U.S. Cities

Within the accounting literature, public sector research to date generally focuses on the associations between accounting information and bond ratings, while controlling for economic and demographic factors (e.g., Reck, Wilson, Gotlob, and Lawrence 2004). The research gained momentum with the financial reporting model required by the Governmental Accounting Standards Board (GASB) in its Statement No. 34 (GASB 34, GASB 1999). Much of the recent research emphasizes metrics derived from the two sets of financial statements (i.e., fund-basis and government-wide statements on bond ratings [Plummer, Hutchinson, and Patton 2007; Johnson, Kioko, and Hildreth 2012; Pridgen and Wilder 2013]). Another important aspect of the GASB 34 reporting requirements has not been previously examined, namely comparisons of actual with budgeted financial information. To our knowledge, the resulting budget-to-actual variances have not been empirically examined for their potential linkage to bond ratings.

We focus on bond ratings because, with more than $3.7 trillion in municipal debt outstanding (Government Accountability Office [GAO] 2012), the bond ratings process is important to a wide range of municipal stakeholders affected by debt costs. Individual investors hold more than 75 percent of the outstanding balance, yet the markets are less liquid and less transparent than equities markets (GAO 2012). The bond ratings process, if accurate, can decrease information asymmetry between municipal bondholders and municipal managers.^1,2 While the Securities and Exchange Commission (SEC) cites concerns with the reliability of bond ratings in its recommendations for additional required municipal disclosures (SEC 2012), the complexity of the existing financial reporting model makes it unlikely that the need for the bond ratings agencies can be significantly reduced, particularly for individual investors. Therefore, it is important that we extend our understanding of the factors that potentially influence the bond ratings process. This study makes such an effort by examining the association between budget-to-actual variances and bond ratings.

GASB Concepts Statement No. 1 (GASB 1987) emphasizes the accountability created by the disclosure of budget-to-actual comparisons. Although state and local governments had previously been required to present this information, GASB 34 (GASB 1999) added the requirement that original budgets be reported. This change allows financial statement users to evaluate management's effectiveness in predicting revenues and expenditures at the outset of the fiscal year, rather than after amendments are made based on events during the year. Because changes to the budget can result from changes in external factors or policy (GASB 2012), the addition of original budgets to the budgetary comparison schedule may provide users with more comparable information, as these amounts reflect expectations at the outset of the fiscal year before any such changes occur (GASB 2012).

The ratings agencies themselves (Moody's, Standard & Poor's [S&P], and Fitch Ratings [Fitch]) cite the importance of budgeting in their ratings methodology documents (Moody's 2009; Fitch 2011; S&P 2011). However, ratings documents do not explicitly identify how budget ratcheting and budgetary slack are evaluated in the ratings process. Within governments, empirical evidence supports potential incentives with the budgetary process, including budget ratcheting and the creation of budgetary slack (Zimmerman 1977; Giroux and Shields 1993; Lee and Plummer 2007; Callahan, Waymire, and West 2011). If emphasis is placed on favorable variances (GASB 2012; Government Finance Officers Association [GFOA] 1998), then municipal managers may have incentives to inflate expenditure budgets and to be conservative in their revenue budgets to minimize the number of potential red flags that would otherwise be suggested by unfavorable variances. Moreover, the emphasis that ratings agencies place on the budgetary process could provide incentives to produce favorable variances.

We use a sample of 190 city-year observations of large U.S. cities during 2003 to 2006. We consider budget-to-actual variances using the original budget amounts required by GASB 34 (GASB 1999). We observe descriptive and statistical evidence of a disproportionate share of small favorable revenue and expenditure variances (between 0 and 5 percent). Furthermore, the variances associated with the net of revenues and expenditures, i.e., surplus/deficit, have a distribution with an even greater number of small favorable variances (between 0 and 5 percent). These results are consistent with the proposition that budgets are set to create the necessary slack to produce favorable variances.

We then examine associations between variances and Moody's general obligation bond ratings. Ceteris paribus, we expect that favorable variances are associated with higher bond ratings. We find that, despite the apparent incentives to report favorable variances, these variances are influential in the ratings process. Specifically, we find that the magnitudes of revenue variances, in either direction, are negatively associated with bond ratings. In contrast, more favorable expenditure variances are associated with higher bond ratings. These findings are consistent with Frank and Zhao (2009), who emphasize the importance of precision in revenue forecasting and conservatism in estimating expenditures. Associations are robust when we examine S&P and Fitch ratings.

These results contribute both to the accounting literature and to the understanding of municipal finance. That the variances, calculated using original budget amounts, are associated with bond ratings affirms the GASB's decision to include these disclosures as well as the final budget amounts. Compared to financial reporting for other entity types (e.g., for-profit, nonprofit), this GASB requirement is unique and appears to be useful. Furthermore, from the perspective of municipalities, these reported variances may have negative consequences for bond ratings and ultimately debt costs. These results therefore have practical implications for municipal stakeholders including bondholders, municipal managers, bond rating analysts, and others interested in the ratings process.

In the sections that follow, we develop our primary hypothesis in the context of a literature review in the second section, present our methodology and models in the third section, present our results in the fourth section, and then conclude in the fifth section.

The GASB's Statement No. 34, Basic Financial Statements and Management's Discussion and Analysis for State and Local Governments (GASB 1999), required extensive changes to the state and local government reporting model. Incorporating a government-wide set of financial statements that adopted an accrual basis of accounting, users of these statements now can view financial results through two accounting bases—accrual and modified accrual. This standard was followed by research to examine the bond rating implications of the required information in various settings. Within a sample of school districts, Plummer et al. (2007) find that the new government-wide statement of net assets influences the bond ratings process, while the statement of activities (similar to a consolidated corporate income statement) does not. Johnson et al. (2012) and Pridgen and Wilder (2013) extend these results to a sample of state governments and municipal governments, respectively, both finding evidence of the benefits of elements of the accrual-based financial statements in the bond ratings process. Using net interest costs as a more direct measure of debt costs, Reck and Wilson (2014) find that the accrual-based statements required by GASB 34 provide incremental explanatory power beyond the effect of pre-GASB 34 fund financial information.

In addition to this fundamental change in the reporting model, GASB 34 also added a requirement that original budget amounts be reported in the schedule that compares budgeted with actual results (GASB 1999). Prior to the implementation of GASB 34, budgetary comparison schedules were presented, but only the final budget was required. The requirement to show both original and final budgets reflects the GASB's acknowledgement that final budgets may not be sufficient for the readers of the financial statements. Final budgets incorporate changes that are based on external environmental factors (e.g., changes in property values) that may become known during the year and changes in policy (GASB 2012). In contrast, original budgets reflect a state or local government's expectations at the outset of the year with no such adjustments. Because the reasons for changes in budget are not easily disentangled (i.e., how much is related to changes in external factors versus policy), the original budget amounts may be useful to the readers of the statements.

The role of budgeting in the bond ratings process has received little attention in the academic literature. However, its importance is evident in the ratings criteria, which cite budgeting as an important consideration for evaluating both management and financial condition (Moody's 2009; S&P 2011; Fitch 2011). More specifically, budget-to-actual variances calculated using original budget amounts can reflect features of both management quality (i.e., how proficient are managers in predicting revenues and expenditures) and environmental uncertainty (i.e., what is the impact of external forces that cannot be predicted with accuracy). Despite difficulty disentangling the underlying reasons for variances, the language in the ratings criteria documents of the three dominant ratings agencies suggests that budgetary information is relevant information in the bond ratings process.

Bond ratings reflect the default risk associated with a debt issue, which may include managerial aspects of municipalities (Bond Market Association [BMA] 2001). The public administration literature has generally supported a direct relationship between financial management quality, using the money score produced by the Government Performance Project (GPP), and bond ratings (Rubin and Willoughby 2009; Jimenez 2011; Krueger and Walker 2010). Budget-to-actual variances capture similar information as the GPP money category score.³ Furthermore, budget-to-actual variances may be explicitly used by ratings analysts and other financial information users to assess management quality (Poister and Streib 1999; International City/County Management Association [ICMA] 2003). The suitability of this proxy, however, should be considered in light of Stinson's (2006) observation that inaccuracies in forecasting include items within management's control (e.g., imprecise models) and outside management's control (e.g., structural economic changes).

The impact of budget-to-actual variances must be placed in organizational context when making predictions about and evaluating bond-rating effects. The governmental setting has requirements and incentives that may have implications for the distribution of variances. Many state and local governments have balanced-budget requirements (GFOA 1998), although the specified requirements can differ. In addition, because governments cannot legally spend more than is appropriated, they may build slack into their budgets, resulting in expenditure variances that are mostly favorable. In contrast, revenues may not be required to exceed a minimum level, but Frank and Zhao (2009) cite the importance of precision in revenue forecasting. As a result, we expect state and local governments to aim for revenue variances of low magnitude, in either direction.

In addition to legal constraints in the budgetary process, governments face other incentives in setting budgets. With regard to the initial budget-setting process, Callahan et al. (2011) find empirical evidence of the incrementalism phenomenon—request “last year's plus a little more”—as well as potential negative operating performance effects of this practice in a sample of U.S. Army hospitals. Similarly, building upon the work of Leone and Rock (2002), who find evidence of budget ratcheting in a corporate setting, Lee and Plummer (2007) report evidence of this budget ratcheting process in a sample of Texas school districts. They demonstrate that budget increases associated with prior year overspending are greater in magnitude than budget decreases associated with prior year underspending (Lee and Plummer 2007). Indeed, citizens understand incrementalism, i.e., the “inching up of the budget” by governmental agencies, and may view it in a negative light (Wildavsky 1988). Although this practice can be used in positive ways (e.g., stockpiling of supplies to be used in the next fiscal period as described by Balakrishnan, Soderstrom, and West [2007]), it is commonly viewed as a negative feature of governments that encourages spending beyond what is necessary (Wildavsky 1988).

Absent legal requirements (appropriations as a legal limit on expenditures) and incentives (managers likely have a preference for favorable revenue and expenditure variances), we expect that the average budget-to-actual variance is 0. Whether for revenues, expenditures, or surplus/deficit, on average, municipal managers are expected to overestimate or underestimate by the same margin over time (i.e., the distribution of variances would be normal and have a mean of 0). However, given the legal requirements associated with the appropriations/spending process, we expect a disproportionate share of favorable expenditure variances. Moreover, despite the emphasis on precision in revenue forecasting (Frank and Zhao 2009), the flagging of unfavorable variances, including revenue variances, as potential causes for concern (GASB 2012) likely leads to small favorable revenue variances. Therefore, we expect that revenue, expenditure, and surplus/deficit variances to be disproportionately favorable than would be expected absent legal constraints and managerial incentives. We explore this possibility with both descriptive and statistical analyses.

We then consider the impact that these budget-to-actual variances have on bond ratings. The ability to predict revenues and expenditures and the demonstration of fiscal restraint are important elements of successful management. Frank and Zhao (2009) cite the importance of precision in revenue forecasting. Describing revenue forecasting as the “cornerstone of budget preparation,” Frank and Zhao (2009) find that local government officials attempt to under forecast revenue and constrain expenditures. City governments differ from their federal and state counterparts in that municipal revenue streams are less elastic with the emphasis on property taxes (Frank 1990; Mikesell 2003). They also tend to be subject to less political scrutiny than federal and state governments (Wildavsky 1988; Bretschneider, Gorr, Grizzle, and Klay 1989; Klay and Grizzle 1992). Particularly in large cities, these factors potentially create low tolerance for large variances by stakeholders.

We therefore expect that revenue variances in either direction (magnitude) result in lower bond ratings. Additionally, we expect that unfavorable expenditure variances (direction) result in lower bond ratings. However, it is unclear whether bond ratings analysts anticipate and then discount budgetary slack in the ratings process. If indirect evidence of managerial incentives is found with a nonrandom distribution of budget-to-actual variances, then we are uncertain whether and how bond ratings analysts adjust their evaluations. It could be that the variances lose their relevance, in which case we do not expect a statistical association between these variances and bond ratings. On the other hand, it could be that bond ratings analysts expect municipal managers to create slack as a means of maintaining existing budget levels and perhaps to create opportunities to redeploy resources to meet organizational objectives. If this is the case, then we expect revenue variances (in either direction) and unfavorable expenditure variances to be associated with lower bond ratings. Thus, we present H1 in the null form:

• H1: 

  Budget-to-actual variances and bond ratings are unrelated.

In testing H1, we consider three types of variances: revenue, expenditure, and the net (or surplus/deficit − revenues less expenditures). We consider the net variances, as well as the revenue and expenditure variances because net variances allow financial statement users to evaluate whether a municipality is able to adjust if unexpected events affect either revenues or expenditures.

Large cities tend to issue debt frequently and, therefore, large cities require the regular issuance of bond ratings. As a result, our sample of 190 city-year observations includes 54 of the largest U.S. cities over the 2003–2006 time period.⁴ As depicted in Table 1, this produced a maximum sample size of 216 observations. The sample is reduced for instances where the CAFR was unavailable (4 observations) or bond ratings were not reported (22 observations). The final sample is a panel of 190 city-year observations.⁵

The cities in our sample issue debt regularly. Table 2, Panel A provides descriptive statistics related to debt activity and budget changes. As depicted, the cities issue general obligation debt an average of 5.132 times during their 3.585 years in the sample period, for an average of 1.439 general obligation debt issues per city per year. The amount of debt issued annually ranges from $56.4 million to $11.0 billion, with a mean of $287.0 million. Bond ratings are issued an average of 2.755 times over the sample period.⁶ In addition, 94.2 percent of the city-year observations include at least one change in the original budget over the course of the fiscal year.

Panel B of Table 2 also presents descriptive statistics for control variables described later. Mean population for the cities in our sample is 800,315, ranging from 207,782 to 8,250,567. Unemployment over the sample period averages 5.4 percent, and per capita income averages $33,679. Mean leverage, measured as general obligation debt scaled by general fund revenues, is 1.224.

We report specifications for Moody's uninsured general obligation bond ratings obtained from the CAFRs.⁷ We select Moody's for our primary tests, as all 190 observations had Moody's ratings available both in the CAFRs and reported by Moody's. The self-reported ratings were generally found in the management discussion and analysis (MD&A) section of the CAFRs.⁸

The primary analyses utilize the Moody's general obligation ratings, but we also consider S&P and Fitch to allow for differences in ratings processes (Dreibelbis and Breazeale 2012; Allen and Dudney 2008). The ratings are converted to an ordinal scale, where higher values reflect better ratings. Moody's ratings potentially range from 1 to 21, and S&P and Fitch from 1 to 19, although not all possible ratings are populated by cities in our sample. As depicted in Panel C of Table 2, Moody's ratings are fairly evenly distributed among the highest six rating categories.

For our independent variable of interest, budget-to-actual variances, we hand collect information from the CAFRs. We calculate variances for revenues, expenditures, and surplus/deficit (i.e., revenues less expenditures) for the general fund using the original and final budget information, as the general fund is common to all cities in the sample.

We use the original budget to measure variances, as the timing of budget revisions likely varies significantly across the sample. Furthermore, budget revisions could be made for two primary reasons—to correct for forecasting errors and to incorporate unexpected events—which cannot be precisely disentangled. We use total revenues and total expenditures, rather than more detailed line items, for calculating variances because cities can differ in their line item detail, but all of the cities should be concerned about overall variances. Furthermore, when calculating the surplus/deficit variances, we use revenues less expenditures, excluding other financing sources or uses. Excluded amounts include proceeds of debt and transfers that may be less frequent in occurrence and not representative of ongoing operations. Control variables that capture financial condition are also obtained from publicly available CAFRs.

Two ordinary least squares (OLS) regression models are estimated: one that considers separate effects of revenue and expenditure variances, and another that considers the net, or surplus/deficit, impact in total. Although the dependent variable, general obligation bond rating, is an ordinal measure, we use OLS, as it has been determined to be a reasonable approach when there are a sufficient number of categories (Xia 2014; Kaplan and Urwitz 1979). Furthermore, OLS is preferable to ordered logit when panel data are used, as the latter may cause inconsistent estimators (Baetschmann, Staub, and Winkelmann 2011; Baghai, Servaes, and Tamayo 2014).⁹ We incorporate year fixed effects to address any potential time series dependence, as well as robust, clustered errors on city to address cross-sectional dependence (Petersen 2009).¹⁰

In the first model, for each set of analyses (revenue/expenditure and surplus/deficit), we begin with a simple model that includes the raw variances scaled as appropriate to remove the effect of size. We then extend the analyses to accommodate the potential impact of magnitude, rather than direction, of the revenue variances. This model is estimated for Moody's bond ratings in our primary tests and for the two other agencies in follow-up analyses as follows:

Appendix A provides full variable definitions. Notice that larger general obligation bond ratings reflect a better credit rating. Revenue variances (RevVar) and expenditure variances (ExpVar) are positive if favorable (revenues are more than budgeted, expenditures are less than budgeted), and are scaled by budgeted revenues and expenditures, respectively. Consistent with Reck et al. (2004) control variables include financial position (FinPos) measured as general fund balance scaled by general fund revenues, and Leverage measured as general obligation debt, scaled by general fund revenues. We also include demographic variables: population, per capita income, and unemployment rate. We also incorporate year fixed effect dummy variables.

In the second model, we consider associations for surplus/deficit, i.e., the net of revenues and expenditures (defined in Appendix A). Similar in our approach to analyzing revenues and expenditures in the first model, we first consider the direction of the net variance. We then incorporate additional analyses to consider the magnitude of the net variance. Control variables are consistent with those included in Model (1):

Table 2 presents descriptive statistics for the variables of interest. We note that the mean bond rating 18.5 for Moody's corresponds with a rating between an Aa-2 and an Aa-3. Mean bond ratings for S&P and Fitch are 16.9 and 16.5, respectively, which correspond with a rating between an AA and an AA−.¹¹ For the time period of the study, the averages correspond with strong creditworthiness under all three assessments. Budget-to-actual variances are computed as comparisons of actual with both original budgeted amounts and final budgeted amounts. Variances are scaled to remove the effect of size, and variances are favorable when positive and unfavorable when negative. Revenue variances (RevVar) are scaled by budgeted revenues. Revenue variances using original budget amounts range from 13.7 percent unfavorable to 22.0 percent favorable, with a mean of 2.0 percent favorable. Using final budget amounts, revenue variances range from 33.1 percent unfavorable to 14.5 percent favorable, with a mean of 0.3 percent favorable. Expenditure variances (ExpVar) are scaled by budgeted expenditures. Expenditure variances using original budget amounts range from 7.7 percent unfavorable to 22.1 percent favorable, with a mean of 2.4 percent favorable, similar to revenue variances. Using final budget amounts, expenditure variances range from 13.4 percent unfavorable to 24.8 percent favorable, with a mean of 3.7 percent favorable.¹² The net, or surplus/deficit, variances (NetVar) are calculated as actual surplus/deficit (revenues less expenditures) less surplus/deficit profit (budgeted revenues less budgeted expenditures), scaled by budgeted revenues. Net variances using original budget amounts range from 13.3 percent unfavorable to 44.7 percent favorable, with a mean of 4.9 percent of budgeted revenues. Using final budget amounts, net variances range from 43.7 percent unfavorable to 15.0 percent favorable, with a mean of 3.9 percent unfavorable.

Table 3 presents Pearson pairwise correlations for the variables employed in the OLS regressions. Revenue variances and expenditure variances, scaled by budgeted amounts, are negatively correlated (r = −0.359 with p-value < 0.01); i.e., a favorable revenue variance is more likely to be associated with an unfavorable expenditure variance. The three ratings are highly correlated (r = 0.944 between MoodyRating and S&PRating; r = 0.947 between MoodyRating and FitchRating; and r = 0.961 between S&PRating and FitchRating, all statistically significant at the 0.01 level). Bond ratings are not highly correlated with the revenue variances, but they are positively associated with expenditure and net (surplus/deficit) variances.

We include control variables, FinPos, Leverage, Pop, PerCapIncome, and UnempRate to capture financial, economic, and demographic information that may be associated with bond ratings. Financial position, or FinPos, is calculated as the fund balance in the general fund divided by general fund revenues. This variable provides a scaled measure of the reserves held by a municipality (Reck et al. 2004) and should be positively associated with bond rating. Leverage, measured as general obligation debt divided by general fund revenue, would intuitively vary with bond rating. We expect per capita income and unemployment rate to be positively and negatively associated with bond ratings, respectively.

A few relationships between the control variables and bond ratings are worth noting. As expected, the unemployment rate (UnempRate) is negatively associated with all three bond ratings, i.e., unemployment indicates greater uncertainty associated with a municipality's ability to repay its obligations. Additionally, financial position (FinPos), measured as general fund balance scaled by general fund revenues and a measure of the extent of municipal reserves (Reck et al. 2004), is positively associated with all three bond ratings. While these correlations are consistent with expectations, some correlations are less straightforward in their interpretation. Population and per capita income have a negative association with bond ratings. Furthermore, leverage appears to be positively associated with all three bond ratings, which may seem counterintuitive, i.e., the more highly levered a city, the more likely a default and, hence, a lower bond rating. However, it is also true that financially sound municipalities, with access to lower-cost debt, may be more inclined to issue debt than those that are financially stressed (Zhao and Guo 2011; Sengupta 1998). The inclusion of only very large cities in our sample may also have implications for the relationships above.

We first explore the distribution of the budget-to-actual variances, and the potential for managerial incentives to produce favorable variances. Absent incentives to the contrary, we expect equal numbers of favorable and unfavorable variances (i.e., a variance would have a 50-50 chance of being favorable). However, ratings agencies indicate that analysts consider budgetary results in their assessments of management. Furthermore, balanced-budget requirements and the legal requirements associated with the appropriations process impose additional pressure on municipal managers to produce favorable variances. Municipal managers have the advantage of private information when setting budgets. Outsiders may anticipate, and even accept, the creation of budget slack to increase the likelihood of meeting budgeted targets.

Figure 1 presents three histograms that depict the number of unfavorable and favorable variances. Consider first Panels A and B for revenue and expenditure variances, respectively. Similar to evidence reported in Burgstahler and Dichev (1997), the panel figures indicate a disproportionate share of “good news” just to the right of 0 (small favorable variances).

Panel C of Figure 1, which shows the histogram for the net, or surplus/deficit, variances, presents an even more distinct pattern in the reporting of favorable variances. 84.2 percent of the reported variances are favorable (actual surplus > budgeted surplus), and 87 of 190, 45.8 percent are small, favorable variances (> 0 and < 5 percent). Recall that the correlation matrix presented in Table 3 indicates an inverse relationship between revenue and expenditure variances. Given this relationship, it appears that balanced-budget requirements and other incentives may be more significant at the net level than at the line item level.

We follow the graphical presentation in Figure 1 with an empirical analysis of the means and distributions of the three variances depicted. In addition to the variances that include original budgets, we also include variances calculated using final, revised budgets. In the binomial test presented in Panel A of Table 4, we evaluate whether variances have an equal probability of being favorable versus unfavorable. For the variances computed using the original budgets, the evidence rejects this proposition of equal distributions of favorable and unfavorable for all three variances (revenue, expenditure, and net). Panel B, which reports the t-tests for means and medians of 0, suggests further that all three variances computed using original budgets differ from 0. These results, along with the evidence in Figure 1, are consistent with managerial incentives to create budgetary slack to increase their chances of meeting budgetary targets.¹³

Table 4 provides insight about revisions to the original budgets. While the original budget produces revenue variances that are positive and statistically different from 0 (132 of 190 variances are favorable), revenue variances calculated using final budgeted amounts produce a mean of 0.3 percent (104 of 190 observations are favorable), not statistically different from 0. Expenditures, on the other hand, become more favorable—130 of 190 expenditure variances using original budget amounts are favorable, while 151 expenditure variances using the final budget amounts are favorable. The mean expenditure variance moves from 2.4 percent to 3.7 percent. It should also be noted that the net, or surplus/deficit, variances move from being skewed to the right (favorable) to being skewed to the left (unfavorable). While expenditure variances become more favorable, revenue expenditures become less favorable in a greater magnitude, producing this result. The shift in the net variance may reflect emphasis on balanced budgets at the outset of the fiscal year as a result of policy and public attention, which may wane as the year goes on.

Table 5 presents the results for Model (1). The first column shows the control variables only; the second column adds the variance variables as scaled percentages; and the third column presents results when revenue variances are considered using dummy variables. In particular, in the third column, the indicator variable RevVarLP distinguishes observations with favorable revenue variances greater than 5 percent of budgeted revenues, and the indicator variable RevVarN distinguishes observations with unfavorable revenue variances. The benchmark classification is observations with favorable revenue variances less than 5 percent of budgeted revenues.

Considering the variances as percentages (Column 2) increases the explanatory power over the baseline (controls only) model from an R² of 0.302 to 0.325. The raw revenue and expenditure percentage variances are statistically significant. This suggests that, despite the potentially dysfunctional incentives associated with incorporating slack into expenditure budgets, bond ratings agencies reward favorable expenditure variances. Turning to Column 3, distinguishing revenue variances as RevVarLP (favorable revenue variances > 5 percent) and RevVarN (unfavorable revenue variances) further increases explanatory power to 0.354. RevVarLP, or large favorable revenue variances, and RevVarN, or unfavorable revenue variances, are indicator variables that capture the intercept shift relative to small favorable revenue variances, which are the most commonly reported variances. Large favorable revenue variances, RevVarLP, are not rewarded by the bond ratings agencies. Furthermore, in analyses using the ratings of the other two agencies, large favorable variances are associated with lower bond ratings. Unfavorable revenue variances are associated with lower bond ratings, and this holds true for the ratings of S&P and Fitch as well. Unemployment rate is inversely related to bond rating, while the statistical significance of other control variables differs across models.

Table 6 presents the results of our Model (2), which considers net, or surplus/deficit, variances. Relative to the baseline model, adding net variances increases explanatory power slightly, but the net variance variable is not statistically significant. However, when the model is replicated using Fitch ratings, favorable net variances are associated with higher bond ratings.

Given the disproportionate share of small favorable net variances (see Figure 1), we include not only indicator variables to capture large favorable net variances, NetVarLP, and negative net variances, NetVarN, but also the magnitude of the variance, absNetVar, as the absolute value of the net variance expressed as a percentage of budgeted revenues. These variables are incorporated in the third column. When we interact NetVarN with absNetVar, we find expected associations with bond ratings. In particular, ratings are lower when net variances are unfavorable.

We replicate Tables 5 and 6 using ordered logit or logit specifications that simply distinguish investment grade debt. Results (untabulated) are consistent with those presented. In fact, the logit regressions yield increased statistical significance for some independent variables of interest. Although our attempts to address any time series dependence (with year fixed effects) and cross-sectional dependence (with robust, clustered errors) are consistent with Petersen (2009), we also estimate OLS specifications for each year of our four sample years to provide additional certainty that cross-sectional dependence does not affect our results. Although the sample size for each of these regressions is small (between 46 and 50 observations), the effects of the variances hold. Specifically, in three of the four years, expenditure variances are associated with higher Moody's bond ratings and unfavorable revenue variances are associated with lower Moody's bond ratings. In addition, the results for the net variances are similar to those presented in Table 6.

We use self-reported underlying Moody's bond ratings in the OLS regression results presented in Tables 5 and 6, as well as S&P and Fitch ratings in untabulated results, which yield consistent results. To address concerns about self-reporting, we obtain the general obligation bond ratings directly from Moody's for each observation in our sample. The ratings we collected are those that were issued, as either a ratings change or an affirmation of the existing rating, on or after six months following the end of the fiscal year. We replicate Tables 5 and 6 using these ratings, which are not subject to the self-reporting or timing concerns. Results (untabulated) support our findings that unfavorable variances (for either revenues or expenditures) are associated with lower bond ratings and that favorable revenue variances are not associated with higher bond ratings. We consider the implications of budget revisions by replicating Tables 5 and 6 using variances calculated using final budgeted amounts. Although explanatory power decreases in some models, our basic inferences hold.

Finally, we consider whether revenue growth has implications within our models. Replicating Tables 5 and 6 including using current year revenues over prior year revenues, scaled by prior year revenues, resulted in no shift in our inferences.

In this study, we find evidence of two phenomena that should interest municipal stakeholders. First, the data from the post-GASB 34 financial statements of large U.S. cities suggest incentives for municipal managers to produce small favorable variances for revenues, expenditures, and the resulting surplus/deficit (i.e., revenues less expenditures). The disproportionate share of modest favorable variances (between 0 and 5 percent) over the 2003–2006 time period may suggest that managers may respond to incentives by creating budgetary slack. Second, we find evidence that, despite the potential for budgetary slack, variances are correlated with bond ratings. Consistent also with the academic literature that emphasizes conservatism in spending and precision in forecasting revenues (e.g., Frank and Zhao 2009), we find that bond ratings are positively associated with favorable expenditure variances, but that revenue variances are not associated with bond ratings. This is particularly important given that ratings agencies are not explicit in their documentation about how the incentives for budget-to-actual variances are incorporated into the ratings process.

While we provide indirect evidence of the managerial incentives in the budgetary process and direct evidence of the impact of the resulting budget-to-actual variances on bond ratings, we acknowledge certain limitations, including the potential for endogeneity that may be present despite our efforts to mitigate this concern. Furthermore, the evidence regarding the distribution of variances is merely indirect evidence of managerial response to incentives. In general, most stakeholders likely view variances consistent with GASB (2012) guidance, i.e., unfavorable variances are “red flags,” and worthy of further investigation. In spite of the potential for the creation of budgetary slack to create favorable variances, however, the evidence in consistent with a characterization that bond ratings analysts value the information gleaned from the budgetary comparison schedules.