The Impact of Budget Stabilization Funds on State General Obligation Bond Ratings

Transcription

1 Page 1 of 24 The Impact of Budget Stabilization Funds on State General Obligation Bond Ratings Cleopatra Charles Postdoctoral Fellow Robert F. Wagner Graduate School of Public Service New York University cc2664@nyu.edu ABSTRACT This study examines the impact of budget stabilization funds on state general obligation bond credit ratings. While a number of past empirical papers have examined the effect of various fiscal institutions on state general obligation bond ratings; to date, BSFs have been largely ignored in the literature. The theoretical determinants of credit ratings are discussed and evaluated empirically here using maximum likelihood estimation (MLE). Model estimates show that while the choice to have a statutory BSF versus a constitutional BSF has no apparent impact on credit ratings, more stringent BSFs are associated with higher credit ratings. Further, weak deposit rules are associated with lower credit ratings while weak withdrawal rules are associated with higher credit ratings. INTRODUCTION Budget stabilization funds (henceforth BSFs); also called contingency funds or rainy day funds are now conventional in most states. By early 1995, 45 states and Puerto Rico had created a total of 51 funds. 1 Today, the only states without such funds are Alabama 2, Arkansas, 1 Eckl, C. (1995). States Broaden the Scope of Rainy Day Funds. The Fiscal Letter, XVII (2). 2 Note, while the National Association of State Budget Officers (NASBO) considers Alabama s Education Proration Prevention Fund to be a BSF due to its restrictive scope (funds can only be used for education purposes) most of the literature does not classify it as a BSF. This paper follows the trend in the literature. In addition, Alaska is treated as an outlier in most of the literature primarily because Alaska has a unique revenue structure that is heavily dependent on oil and gas revenue resulting in a larger than average rainy day fund balance. This study does the analysis with and without Alaska but the overall results remain the same (signs remain the same and coefficients change very little). Results reported include Alaska in the sample.

2 Page 2 of 24 Montana, and Oregon. The fundamental concept of a BSF is simple: money is saved when state finances are healthy for use during economic downturns. The ability of a BSF to effectively smooth out government expenditures without tax and spending changes is desirable for a number of reasons: volatility in government expenditure, overall size and growth of government, and the structure of taxation create fiscal instability and a bad investment climate. Individuals and businesses are left wondering whether a state will change its tax system (or the quality of its public services) during periods of fiscal downturn. This in turn, can cause the state to become a less attractive place to live, work and invest. An adequate BSF makes it easier for states to borrow money for financing capital investments in the long-term. Further, bond rating agencies like Moody s and Standard and Poor s consistently allude to healthy BSFs as a reason for giving states high bond ratings. A number of earlier studies 3, 4, 5 have shown that good bond ratings can directly reduce the cost of borrowing for a state. This in turn reduces the cost of providing public services and infrastructure in the long run. Largely because BSFs have the ability to aid states in alleviating periods of fiscal stress, they have the potential to reduce the default risk associated with state-issued municipal bonds. While several studies over the last decade have explored how various fiscal institutions affect state general obligation bond yields, the role of budget stabilization funds has been largely ignored in the literature 6. Whether BSFs have an impact on credit ratings is a matter that has significant consequences for the primary role of credit ratings as information providers. 3 Hsueh, L. P., & Kidwell, D. S. (1986). The Signaling Benefits of a Bond Rating: A Study of Rated and Nonrated Municipal Bonds. Tulane University. 4 Hsueh, L. P., & Kidwell, D. S. (1988). Bond Ratings: Are Two Better Than One? Financial Management, 17(Spring), Johnson, C. L., & Kriz, K. A. (2002). Impact of Three Credit Ratings on Interest Cost of State GO Bonds. Municipal Finance Journal, 23(1), Wagner, G. A. (2004). The bond market and fiscal institutions: Have budget stabilization funds reduced state borrowing costs? National Tax Journal, 57(4),

3 Page 3 of 24 Further, in the past, many researchers who have examined the impact of BSFs have taken a very narrow approach in their analysis. In many cases, researchers view BSFs as either existing within a particular state or not, frequently employing the use of indicator or dummy variables that take the value of one if a BSF exists and zero if it does not. This narrow approach does not take into account the fact that a BSF can take many forms and can vary considerably in design, scope, and restrictiveness. This paper aims to bridge this gap in the literature and examines the effect of the type of BSF (based on the strictness of the rules that determine how funds enter and leave the fund and whether the fund is statutory or constitutional in nature) as well as the size of the rainy day fund on state GO bond ratings. This study draws on the work of a number of earlier authors including Johnson & Kriz 7 and Wagner 8 and utilizes a set of information on state GO bond ratings from Moody s Investors Service for the years to examine whether BSFs have an effect on state underlying GO bond credit ratings. Model estimates show that more stringent BSFs are associated with higher credit ratings. Further, weak deposit rules are associated with lower credit ratings while weak withdrawal rules are associated with higher credit ratings. BACKGROUND AND LITERATURE REVIEW BSFs are an important part of a responsible state budget process first and foremost because taxes and public spending operate on different cycles. According to Navin and Navin 9 BSFs serve many functions among the states; BSFs can be used as buffers against countercyclical budget fluctuations and as an alternative revenue source to cope with cash flow 7 Johnson, C. L., & Kriz, K. A. (2005). Fiscal institutions, credit ratings, and borrowing costs. Public Budgeting & Finance, 25(1), Wagner, G. A., Navin, J. C., & Navin, L. J. (1994). An evaluation of state budget stabilization funds among Midwestern states. Growth and Change, 25(4),

4 Page 4 of 24 problems within and between fiscal years. They can also be used as a convenient account to be tapped as political needs arise. The existing literature on budget stabilization funds focuses on two fundamental and interrelated questions: how do budget stabilization funds affect state savings behavior and do budget stabilization funds affect fiscal stability? In addressing the first question, Knight and Levinson 10 examine the effect of rainy day funds (RDFs) on state savings behavior. They observe that states with a cap or limit on their BSF that is above nine percent save more than states with a low limit (cap below five percent). Further, states with the strictest withdrawal provisions, allowing access to funds only in recessions, save significantly more than states that can access funds through appropriation. While they cannot directly correct for the fact that states planning future savings may adopt RDFs, their findings are robust to the inclusion of measures of savings preferences. Wagner 11 examined the degree of substitutability between general fund and BSF balances by regressing a state s total savings (defined as the sum of the general fund surplus/deficit and BSF balance) on the existence, size, and structure of BSFs. Wagner found evidence regarding substitutability and found that states save significantly more following BSF adoption and that states with BSFs governed by strict deposit and strict withdrawal rules, meaning those funds that force deposits and limit withdraws, experience the largest savings gains. Hou and Duncombe 12 explore how the saving behavior of state governments is affected by adopting the budget stabilization fund (BSF) and by the state's balanced budget requirements 10 Knight, B., & Levinson, A. (1999). Rainy day funds and state government savings. National Tax Journal, 52(3), Wagner, G. A. (2003). Are state budget stabilization funds only the illusion of savings? Evidence from stationary panel data. Quarterly Review of Economics and Finance 43(2), Hou, Y., & Duncombe, W. (2006). State saving behavior: Effects of two fiscal and budgetary institutions. Available at SSRN:

5 Page 5 of 24 (BBR). They assembled a 25-year panel dataset ( ) that includes several budgetary institutions, and controls for state economy, social services, state politics, and business cycles. Their paper finds that state adoption of a BSF is associated with a 2.5 percent increase in savings; the effect, however, is mainly from encoding medium- or high-range allowable balance caps and merging cash flow into the missions of the BSF, but not by requiring a certain balance level of the general fund as source of BSF. The second primary question addressed by the literature involves the extent to which budget stabilization can help reduce fiscal instability. Wagner and Elder 13 use data that span the entire history of stabilization fund usage for nearly all states. Their article is the first to investigate how the existence, size, and structure of BSFs impact the cyclical variability of state government expenditures. They control for the effects of demographic, political, and economic factors, and find that a state s ability to smooth expenditures over the business cycle using a BSF is highly dependent on the structure of the deposit and withdrawal rules governing the fund. Specifically, their results show that states where policy makers have discretion over both deposits and withdrawals experience no significant reduction in expenditure volatility compared to states without stabilization funds, suggesting that such funds are tantamount to having no fund at all, at least from the perspective of expenditure smoothing. In contrast, however, stabilization funds governed by one or more strict rules, meaning those that actually require deposits or limit withdrawals, result in significant reductions in the cyclical variability of expenditures. Sobel and Holcombe 14 examine the role of BSFs in easing fiscal stress during recessions. They examined how states with BSFs fared during the 1990 to 1991 recession by 13 Wagner, G. A., & Elder, E. M. (2005). The role of budget stabilization funds in smoothing government expenditures over the business cycle. Public Finance Review, 33(4), Sobel, R. S., & Holcombe, R. G. (1996). The impact of state rainy day funds in easing state fiscal crises during the recession. Public Budgeting & Finance, 16(3),

6 Page 6 of 24 regressing a measure of fiscal stress, which is defined as the sum of discretionary expenditure reductions and tax increases over the period from 1989 to 1992, on indicator variables for if the state has a BSF, if the BSF has a strict deposit rule, if the BSF has a strict withdrawal rule, and the maximum allowable BSF size. The only BSF characteristic that they found to be significant is the existence of a strict deposit rule, which is correlated with lower fiscal stress. The results obtained by Sobel and Holcombe 15 provide some evidence that BSFs contributed to a smoother fiscal cycle during the 1990 to 1991 downturn. In their fiscal stress measure they assigned a zero value to states that did not have discretionary reductions in spending or tax increases over the period from 1989 to 1992 making the connection between economic volatility and fiscal stress somewhat unclear. In addition, their regression model does not control for fiscal institutions, such as balanced budget rules or tax and expenditure limitation laws, or the size of a state s general fund surplus. Thus, if two states were hit equally hard by the 1990 to 1991 downturn but one state used its general fund surplus balance to offset tax increases or expenditure reductions, these states are treated as having experienced different levels of fiscal stress in the empirical analysis. Navin and Navin 16 examined the movement of the fund balances over time ( ) in seven Midwestern states to see how the fund balances move in relation to a number of indicators of state fiscal health like gross state product (GSP) and state unemployment rate. The results of their analysis indicated that states use these funds for a variety of purposes. In addition, the level of funding and the ability of these funds to serve as effective tools for countercyclical state fiscal policy also vary significantly. 15 Sobel & Holcombe, Navin & Navin, 1994

7 Page 7 of 24 There is also a small literature on the optimal size of BSFs; for example Joyce 17 as well as some recent studies that have attempted to predict the determinants of BSF adoption as well as fund characteristics; see for example Rodriguez-Tejedo 18 and Wagner and Sobel 19. Finally, a more recent strand of the literature has recognized the potential link between credit ratings, bond yields, borrowing costs and BSFs. Wagner 20 explores how budget stabilization funds affect state bond yields using data that spans the entire history of stabilization fund usage for nearly all states. His empirical results reveal that the typical state experiences a modest reduction in bond yields following the adoption of a budget stabilization fund and further that stabilization funds with different types of deposit and withdrawal rules are found to affect borrowing costs differently. He observes that the largest reduction in yields occurs in states that have budget stabilization funds governed by strict deposit and withdrawal rules. EMPIRICAL MODEL A number of earlier analyses inform this study of the impact of budget stabilization funds on state general obligation bond ratings for example; Ederington, Yawitz, and Roberts 21 and Johnson and Kriz 22. In studies of credit ratings and other variables that are multinomial-choice and inherently ordered the ordered probit and logit models have been used extensively. Although the outcome is discrete, the multinomial logit or probit would fail to account for the ordinal nature of the 17 Joyce, P. G. (2001). What's So Magical about Five Percent? A Nationwide Look at Factors That Influence the Optimal Size of State Rainy Day Funds. Public Budgeting and Finance, 21(Summer), Rodriguez-Tejedo, M. I. (2007). State fiscal institutions: An evolution. University of Maryland, College Park. 19 Wagner, G. A., & Sobel, R. S. (2006). State Budget Stabilization Fund Adoption: Preparing for the Next Recession or Circumventing Fiscal Constraints? Public Choice, 126(1-2), Wagner, Ederington, H. L., Yawitz, J. B., & Roberts, B. E. (1987). The Information Content of Bond Ratings. Journal of Financial Research, 10, Johnson & Kriz, 2005

8 Page 8 of 24 dependent variable 23. Further, ordinary regression analysis would not take into consideration the ranking of the dependent variable. Both the ordered probit model and the binomial probit model are built upon latent regression models. 24 Assume that the unobserved continuous measure, creditworthiness y *, is a linear function of a set of explanatory variables x, with parameter vector β, and an error term ε: y * = x β + ε As usual, y * is unobserved. What we do observe are the credit ratings assigned to the states by Moody s Rating Agency which can range from Aaa to C (0-4). So, in this case: y = 0 if y * 0, y = 1 if 0 < y * µ 1, y = 2 if µ 1 < y * µ 2,... y= J if µ J - 1 y *, The µs are unknown parameters to be estimated by β. The ordered probit model assumes a standard normal distribution for the error term yields i.e. an ordered probit model giving the following probabilities: Prob(y = 0 x) = Φ(-x β) - 0, Prob(y = 1 x) = Φ(µ 1 - x β) - Φ(-x β), Prob(y = 2 x) = Φ(µ 2 - x β) - Φ(µ 1 - x β),... Prob(y = J x) = 1 - Φ(µ J x β). This research utilizes a panel approach which creates econometric problems for the ordered probit model, or any limited dependent variable model. Fixed effects would be standard 23 Greene, W. H. (2003). Econometric Analysis (Fifth Edition). New York: Macmillan. 24 See Greene, 2003 for a discussion.

9 Page 9 of 24 in a linear regression, but the fixed effects are not estimated consistently because there is a small time series (here, 10 years) for each fixed effect. Consistency requires unbiasedness and a large sample; the second condition is missing. By differencing out the fixed effects, linear regression avoids bias in other coefficients. This differencing is impossible in probit or most limited dependent variable models. Random effects can be used under the assumption that the random effects are uncorrelated with other explanatory variables. This is usually a problematic assumption for individuals, who tend to be quite idiosyncratic, but is more likely to be true of states, which are more likely to be well described by the principle explanatory variables. Note that the logit model can be estimated with fixed effects, but only if the dependent variable changes over time. For many states, bond rating changes very little if at all. This makes the logit model with fixed effects lose most of the observations. For a discussion of the difficult tradeoffs involved in fixed and random effects in probit and logit models, see Baltagi 25. Due to these inherent difficulties, 26 random effects ordered probit estimation is used here to deal with the unobserved heterogeneity and to account both for the ordinal structure of the dependent variable credit rating (CRATE) and for the panel structure of the data. Random effects ordered probit is a superior method in this case primarily because it considers the existence of an additional normally distributed cross-section error. The unconditional log-likelihood is obtained by integrating the conditional log-likelihood. The integration is approximated for each unit by Gauss-Hermite quadrature. See Butler and Moffitt 27 for details about using Gauss-Hermite quadrature to approximate such integrals. 25 Baltagi, B. H. (1995). Econometric Analysis of Panel Data (pp ). New York: John Wiley & Sons, Inc. 26 Fixed effects have been rarely used due to the lack of suitable econometric methods. An alternative method that has been used to incorporate fixed effects is to transform the ordinal variable into a binary variable that takes the value of one above or under a specific threshold. 27 Butler, J. S., & Moffitt, R. (1982). A computationally efficient quadrature procedure for the one-factor multinomial probit model. Econometrica, 50(3),

10 Page 10 of 24 The included variables were chosen because of their statistical significance in other earlier studies 28, 29 as well as their availability. The model estimated is defined as: CRATE= F (BSF CHARACTERISTICS, ECONOMIC FACTORS, DEMOGRAPHIC FACTORS, FINANCIAL FACTORS) Table 1 provides a brief summary of the variables, their sources, and the expected directions of their effects on average state general obligation bond ratings. State tax capacity or (total taxable resources per capita) is a measure of the state's underlying ability to raise revenues that can be allocated to higher education and other public purposes. States with a strong and diverse economy typically have a high tax capacity. The impact of this variable is uncertain; while a high tax capacity may signal an increased ability to service debt, which could potentially increase the credit rating, rating agencies also view states with high tax burdens negatively, in this case lowering the rating. State unemployment rate, percent of population living at or below the federal poverty line, and per capita GDP are proxies for the strength of a state s economy. States with a vibrant and robust economy will have low levels of unemployment, low poverty rates and high GDP and consequently higher credit ratings. Personal income per capita and homeownership rates of a state measure the wealth of the state. Wealthier states have a greater ability to weather periods of economic downturn resulting in higher credit ratings. The debt position of a state is measured by total debt outstanding at the end of the fiscal year. Higher levels of debt will be viewed negatively by rating agencies due to the potential detrimental effect on the state s ability to repay outstanding bonds. Weak balanced budget rules 28 Ederington, H. L. (1986). Why split ratings occur? Financial Management, 15(1), Johnson & Kriz, 2005

11 Page 11 of 24 are expected to decrease credit quality; strong GO debt limitations are expected to increase credit quality. This study also strives to take into consideration the fact that BSFs can take many forms and can vary considerably in design, scope, and restrictiveness by creating a BSF stringency index. The BSF stringency index is created in the following way to account for its stringency in terms of manner of codification (statutory rainy day funds are not as strong as constitutional rainy day funds and are easier to revise or abolish), deposit rules, withdrawal rules, and fund size, ranging from 0 to 5: coded as 0, if a BSF is not present in state i in year t and coded as 5 when (1) state has a BSF; (2) the BSF is constitutional; (3) money is either deposited each year according to a strict mathematical formula or money is deposited each year into the fund but the amount of the deposit is not strictly dependent on statistical measures of state economic conditions; (4) money can only be withdrawn by a mathematical formula dependent upon state economic conditions or with a supermajority vote of the state legislature, (5) and state has a cap on its fund that is greater than 5 percent 30. Figure 1 provides a map showing the relative stringency of state BSFs based on the design of the fund and the rules that govern how funds enter and leave the BSF. This study also examines individually, the impact of actual size of BSF at the end of the fiscal year, statutory BSFs, and stringency of deposit and withdrawal rules on state credit rating. RESULTS Table 5 provides a summary of statistics for the variables used in this study. 75 percent of states have a statutory BSF; 74 percent of states have weak deposit rules while 76 percent have weak withdrawal rules. On average, state unemployment rate is 4.66 percent but this ranges from 30 For a discussion of deposit and withdrawal rules and a summary of state BSFs see Wagner (2003) and Wagner (2004).

12 Page 12 of 24 a low of 2.3 percent to a high of 8.1 percent. State poverty rates range from 4.5 percent to 21.2 percent. The stringency scores of state BSFs range from 0 to 5. The state of Virginia has the highest BSF stringency score with a score of 5. Table 6 presents the results from the random effects ordered probit model. The overall model fit is statistically significant as identified by the chi-square test. Note that the credit rating variable (CRATE) is set up with better credit ratings being represented by higher numbers (e.g. the highest credit rating Aaa = 4). Larger BSF balances and weak withdrawal rules are both associated with higher credit ratings. Both results are statistically significant at the 0.01 level. The size of the BSF as a percentage of the general fund and weak deposit rules are both associated with lower credit ratings. Increased stringency of deposit rules (going from 1 4) is associated with higher credit ratings while increased stringency of withdrawal rules is associated with lower ratings. The decision to have a statutory BSF versus a constitutional BSF appears to have no impact on credit ratings. More stringent BSFs are associated with higher credit ratings. The other results are generally consistent with prior research on the determinants of municipal bond credit ratings. High unemployment, high levels of outstanding debt and high poverty rates all decrease credit rating while higher per capita GDP and homeownership rates increase credit rating. The coefficient for federal non-matching grants per capita is significant at the 0.05 level, indicating that a relative increase in state non-matching grant funds is associated with a lower credit rating. The coefficient of per capita income is negative and statistically significant at the 0.01 level indicating that a relative increase in state per capita income is associated with a lower credit rating. Weak balanced budget rules and less stringent GO debt limitations are associated with the likelihood of getting a lower credit rating.

13 Page 13 of 24 The interpretation of the effects of explanatory variables in ordered probit requires some explanation. (Random effects are irrelevant to this.) One disadvantage in the random effects ordered probit estimation is the difficulty encountered in interpreting its statistical results. This is primarily due to the non-linearity of the model. Unlike the case in linear regression models, the magnitude of effects in random effects ordered probit models cannot be inferred directly from the coefficients. Instead, we should interpret it in terms of marginal effects. This gives a more accurate depiction of the direction and extent of probability changes across the different categories of the dependent variable i.e. credit ratings. In binomial probit, increasing the latent index unambiguously increases the probability that the outcome is 1 as opposed to 0 (here, a higher bond rating). Given more than two categories, however, the interpretation becomes more complex. Increasing the latent index must reduce the probability of the lowest category (bond rating) and increase the probability of the highest category. In between, the effect is ambiguous depending on the amount of probability moving in from below and moving out from above. Technically, this depends on the standard normal pdf at the category limits. The total change in probability across all categories is zero, because the total probability must sum to 1. The marginal effects of an ordered probit model shown in Table 7 are calculated at the mean value of each independent variable unless otherwise noted. A state with the average amount of money in its BSF (in this case million dollars) has a 0.5 percent probability of having a rating of Aa3, 58 percent probability of having a rating of Aa1, 40 percent probability of a rating of Aa2, and 0.2 percent probability of having a credit rating of Aaa (holding all other explanatory variables at their mean).

14 Page 14 of 24 If we calculate the marginal effects for Model 2 31 with all variables at their means then a state in this case would have a 54 percent probability of having Aa1 rating. However, holding all other explanatory variables at their mean a state with a constitutional BSF, strong deposit and weak withdrawal rules would have a (higher) 77 percent probability of having Aa1 rating. A state with weak deposit and withdrawal rules, statutory BSF but high unemployment, poverty, and debt levels would have a 1 percent probability of having a rating of Aa1 (holding all other explanatory variables at their mean). Unemployment and poverty are difficult for states to change, but deposit and withdrawal rules are matters of policy, so states can affect their bond ratings by changing the stringency of these rules. Finally, if we calculate the marginal effects for Model 5 32 which contains the BSF stringency index then holding all variables at their means a state would have a 57 percent probability of having Aa2 rating and a 42 percent probability of having Aa1 rating. Note however, for a state with a stringency value of 0 (no BSF) and all other variables at their means the probability of Aa1 rating is 18 percent and there is a less than 0.1 percent chance of having Aaa rating. Compare this to state with a BSF stringency value of 5 (most stringent) and all other variables at their means the probability of Aa1 rating is 76 percent and there is now a 5 percent probability of having Aaa rating. An important issue in this study is the potential endogeneity of BSFs and total debt outstanding. It is possible that states with more constraining BSFs and more money in their BSFs also have fiscally conservative electorates and that the primary factor explaining the impact on credit ratings is the electorate, not the BSF. In other words, it is highly possible that BSFs are a reflection of voter preferences, and as such, that the correlation between BSFs and credit ratings 31 Model 2 adds the following variables to the general model for credit ratings (fund size, statutory BSF, weak withdrawal rules, weak deposit rules). 32 Model 5 adds the stringency variable to the general model for credit ratings.

15 Page 15 of 24 may just reflect an underlying correlation between voter tastes and fiscal policies. Further, states may decide to adopt more stringent BSFs and keep more money in the BSFs because they have good (Aaa for example) credit ratings and want to maintain their good credit rating. Typically, researchers in public finance have approached the endogeneity problem in two ways. The first approach involves using a fixed effects model and assuming that voters tastes and preferences are a time-invariant constant that is captured in the unobserved state effects. For the methodological reasons detailed earlier in this paper, this approach is not feasible in this case and is beyond the scope of this research. The second approach involves employing the use of a 2SLS/IV estimator to control for the simultaneity problem. However, there is no IV estimator for the random effects ordered probit model. In an attempt to explore the endogeneity issue further I first ran a reduced form model using all exogenous variables 33. I then saved the residuals from the analysis and included the residuals in my original model 34. Finally, I test for the significance of the coefficient of the added residual. Using this approach it became necessary to find valid instruments that were correlated with the BSF variables but did not have any direct influence on credit ratings nor were correlated with the unexplained variation in state credit ratings. Drawing on the work of Wagner 35 the following variables were chosen as instruments for the BSF variables: (1) an indicator variable that equals 1 if a state follows a biennial budget cycle; (2) an indicator variable that equals 1 if a state s governor faces a term limit; (3) an indicator variable that equals 1 if a 33 In this case my BSF stringency variable becomes the dependent variable and I include all exogenous variables and two instrumental variables (term limit for governor and term limit for legislature. 34 Now my dependent variable is credit rating and once again I include all exogenous variables, the possible endogenous variable BSF stringency and the residuals from the reduced form model. 35 Wagner (2004)

16 Page 16 of 24 state s legislature faces a term limit; and (4) an indicator variable that equals 1 if both the governor and legislature of a state face term limits 36. test residuals Chi-squared (1) = 0.47 Prob > chi-squared = The large p-value indicates that random effects ordered probit is consistent in this case however further research should address the question of simultaneous random effects ordered probit estimations. CONCLUSION Budget stabilization funds have the potential to aid states in mitigating periods of fiscal stress and have the potential to reduce the default risk associated with state-issued debt. This paper provides evidence that the form, design, and scope of BSFs can have an impact on credit ratings. Having a large BSF is positively associated with credit ratings, states especially need to pay attention to the rules that govern how funds enter and leave the BSF. A state with a value of 5 on the stringency scale (most stringent) has a 5 percent chance of getting Aaa rating and a 76 percent chance of Aa1 rating while a state without a BSF has almost a 0 percent chance of having Aaa rating and 18 percent chance of getting a Aa1 rating. Clearly, a more stringent BSF can improve credit ratings. It also appears that weak deposit rules are associated with lower credit ratings so states need to implement more stringent measures to ensure that adequate deposits are made periodically. A BSF with no money in it or a negative balance as was the case with some of the states in this sample is no help at all during an economic downturn. Similarly, having an adequately funded BSF but without the flexibility of drawing from it during an economic 36 For a discussion of the relevance and choice of instruments please see Wagner (2004)

17 Page 17 of 24 downturn is not good fiscal policy. Higher credit ratings are associated with weaker and hence more flexible withdrawal rules. Strict withdrawal rules can increase the likelihood that states will have to rely on tax hikes and budget cuts to get through an economic downturn. In this case, having a BSF and not using it, is the same as not having a BSF at all. Over the last six decades, the rating of municipal bonds by independent rating agencies has grown from an insignificant factor into a fundamental and, in many cases, controlling factor as to whether an investor or institution will buy a municipal bond. Rating agencies have consistently alluded to the fact that reserve policies are important criteria when assigning a credit rating. This paper provides evidence that indeed, the policies that govern a BSF can impact credit ratings. State and local governments rely on general obligation and revenue bonds to finance many infrastructure projects. Municipal bond ratings are important to citizens because they can alter the cost of public infrastructure; the higher the rating the lower the cost of borrowing for a government entity. Whether BSFs have an impact on credit ratings and subsequently borrowing cost is a matter that has significant implications for the role of credit ratings as information providers. Any attempt to understand the impact of BSFs on borrowing has important implications for understanding the internal workings of the municipal bond market. The results obtained here have significant implications for helping us understand the effects of BSFs and for informing future debates about the relevance and design of BSFs. There is still much work left to be done in this area. The results obtained herein indicate that the effects of BSFs on credit ratings and borrowing costs needs to be a continuing area of research in the field of public finance.

18 Page 18 of 24 Table 1: Independent variables to be used to predict bond ratings (dependent variable) Variable Name Description Expected Sign Source TAXCAP State Tax Capacity? The National Center for Higher Education Management Systems UNEM State unemployment rate - Bureau of Labor Statistics INCOME Per capita personal income + Census Bureau DEBTOUTS Per capita debt outstanding - Census Bureau WEAK_DEPOSIT Deposit requirements of BSF - Wagner (2004) according to the strictness of the rule from (1 2=weak; 3-4=strong) WEAK_WITHDRAW Withdrawal requirements of BSF - Wagner (2004) according to the strictness of the rule from (1 2=weak; 3-4=strong) STATUTORY Dummy variable indicating legal nature of BSF - National Conference of State Legislatures BBR_LAX Index capturing strength of a state s balanced budget rules + Poterba (1994) ACIR GO_DEBTLIM Index capturing strength of state + Bahl & Duncombe (1993) GO debt limitations FUND_SIZE Size of rainy day fund where 1= + Wagner (2004) 5% of budget or less; 2= between 5% and 25%; 3= no limit GDP Per capita state GDP in constant + Bureau of Labor Statistics 2000 dollars MGRANTS Federal matching grants per capita - Census Bureau (FAADS) NMGRANTS Federal non-matching grants per - Census Bureau (FAADS) capita POVERTY % of population living below - Census Bureau federal poverty line HOMEOWNERSHIP % of owner occupied houses + Census Bureau BSF_AMOUNT BSF balance in $1,000 at the end + NASBO of fiscal year BSF_PERCENT BSF balance as a percentage of + NASBO general fund balance at the end of fiscal year BSF_STRINGENCY BSF stringency index (0-5) + Author s calculations

19 Page 19 of 24 Table 2: Moody s Municipal Bond Ratings Long-term Debt Ratings (maturities of one year or greater) Investment Grade Aaa Highest rating, representing minimum credit risk Aa1, Aa2, Aa3 High grade A1, A2, A3 Upper medium grade Baa1, Baa2, Baa3 Medium grade Speculative Grade Ba1, Ba2, Ba3 Speculative elements B1, B2, B3 Subject to high credit risk Caa1, Caa2, Caa3 Bonds of poor standing Ca Highly speculative, or near default C Short-term debt ratings (maturities of less than one year) Prime 1 Lowest rating, bonds typically in default, little prospect for recovery of principal or interest Highest quality Prime 2 Prime 3 Not Prime Source: Moody s Rating System. Table 3: State General Obligation Credit Rating Distribution Moody s Underlying Ordinal Credit Frequency Percent Credit Rating Rating Aaa Aa Aa Aa A1 and below Total Source: Moody s Investors Service ( ). Note: The following states were omitted from the analysis because they had no general obligation debt so no rating was available: Colorado, Nebraska, South Dakota, and Wyoming. Note: The following states while they had no general obligation debt rating were given an issuer rating. However, they were also omitted from the analysis: Arizona, Idaho, Indiana, Iowa, Kansas, Kentucky, and North Dakota. Table 4: BSF Stringency Index BSF Stringency Index Frequency Total 500 Source: Author s calculations for years

20 Page 20 of 24 Table 5. Descriptive Statistics (N=390) Mean SD Min Max Per capita income Per capita GDP ($ 2000) Homeownership Rate Per capita debt outstanding Unemployment rate Total taxable resources Log (matching grants per capita) BSF amount ($1,000,000) BSF amount % of General Fund Poverty rate Log (non-matching grants per capita) Credit rating Fund size BSF_Stringency BSF dummy BSF statutory Weak balanced budget rules Weak deposit rules Weak withdrawal rules GO debt limit index

21 Page 21 of 24 Table 6: Random Effects Ordered Probit Model Predicting Credit Rating Model 1 Model 2 Model 3 Model 4 Model 5 Credit Rating Beta SE Beta SE Beta SE Beta SE Beta SE BSF balance ($1,000,000) Fund Size Weak withdrawal rules Weak deposit rules Statutory BSF Deposit rule Withdrawal rule BSF% of general fund BSF stringency index * *** ***1.484 *** *** ***0.690 *** *** *** Per capita income Per capita GDP ($ 2000) Homeownership rate Per capita debt outstanding GO debt limit index Weak balanced budget rules Unemployment rate Total Taxable Resources Poverty rate Non-matching grants Matching grants *** ***0.117 *** *** *** *** *** *** ** *** *** ***0.155 ** ** *** *** ** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ** *** *** ** *** ** **0.076 ***5.954 *** *** *** **0.051 *** *** *** N Log Likelihood LR chi-squared Prob > chi-squared NOTE: *** indicates significant at.01;** indicates significant at 0.05; * indicates significant at

22 Page of 24 Table 7A: Marginal Effects Model 1 CRATE Probability in CRATE Maximum BSF balance Minimum BSF balance A1 & below <0.001 <0.001 <0.001 Aa < Aa Aa Aaa ***NOTE: The marginal effects are calculated at the mean value of each explanatory variable in column 2. Marginal effects in other columns are calculated at different values of BSF variables. For example in column 3 above, marginal effects are calculated at the mean value of each explanatory variable but at the maximum BSF balance for the sample. CRATE Probability in CRATE Table 7B: Marginal Effects Model 2 Weak deposit Strong deposit Weak withdraw weak withdraw StatutoryBSF constitutional Strong deposit strong withdraw constitutional BSF Strong deposit weak withdraw constitutional fund size=3 BSF A1 & below <0.001 <0.001 <0.001 <0.001 <0.001 Aa < Aa Aa Aaa ***NOTE: The marginal effects are calculated at the mean value of each explanatory variable except where noted. CRATE Probability in CRATE Table 7C: Marginal Effects Model 3 Deposit rule=4 Withdraw rule=4 Deposit rule=1 Withdraw rule=4 Deposit rule=4 Withdraw rule=1 Deposit rule=1 Withdraw rule=1 A1 & below <0.001 < <0.001 <0.001 Aa < Aa Aa Aaa < ***NOTE: The marginal effects are calculated at the mean value of each explanatory variable except where noted. (4= strictest rule; 1= weakest rule) Table 7D: Marginal Effects Model 4 CRATE Probability in CRATE BSF= % of GF BSF= 1.470% of GF A1 & below <0.001 < Aa Aa Aa <0.001 Aaa <0.001 <0.001 <0.001 ***NOTE: The marginal effects are calculated at the mean value of each explanatory variable except where noted. Table 7E: Marginal Effects Model 5 CRATE Probability in BSF STRINGENCY=0 BSF STRINGENCY=5 CRATE A1 & below <0.001 <0.001 <0.001 Aa <0.001 Aa Aa Aaa < ***NOTE: The marginal effects are calculated at the mean value of each explanatory variable except where noted.

23 Page of 24 Figure 1: Stringency of State Budget Stabilization Fund Rules (Author s Calculations) 25% by Note: Stringency scale begin with an analysis of the various types of state BSFs and was developed using the classification of deposit and withdrawal rules derived by Wagner (2003) and Wagner (2004). Scale goes from 0 (weakest) to most stringent (5). The BSF stringency score of a state equals the sum of scores from each of the following features (Existence of BSF, Constitutional BSF, Strict deposit rules, Strict withdrawal rules, Cap on fund greater than 5 percent).

24 Page of 24 References Baltagi, B. H. (1995). Econometric Analysis of Panel Data. New York: John Wiley & Sons, Inc. Butler, J. S., & Moffitt, R. (1982). A computationally efficient quadrature procedure for the one-factor multinomial probit model. Econometrica, 50(3), Eckl, C. (1995). States Broaden the Scope of Rainy Day Funds. The Fiscal Letter, XVII(2). Ederington, H. L. (1986). Why split ratings occur? Financial Management, 15(1), Ederington, H. L., Yawitz, J. B., & Roberts, B. E. (1987). The Information Content of Bond Ratings. Journal of Financial Research, 10, Greene, W. H. (2003). Econometric Analysis (Fifth ed.). New York: Macmillan. Hou, Y., & Duncombe, W. (2006). State saving behavior: Effects of two fiscal and budgetary institutions (Publication.: Hsueh, L. P., & Kidwell, D. S. (1986). The Signaling Benefits of a Bond Rating: A Study of Rated and Nonrated Municipal Bonds. Tulane University. Hsueh, L. P., & Kidwell, D. S. (1988). Bond Ratings: Are Two Better Than One? Financial Management, 17(Spring), Johnson, C. L., & Kriz, K. A. (2002). Impact of Three Credit Ratings on Interest Cost of State GO Bonds. Municipal Finance Journal, 23(1), Johnson, C. L., & Kriz, K. A. (2005). Fiscal institutions, credit ratings, and borrowing costs. Public Budgeting & Finance, 25(1), Joyce, P. G. (2001). What's So Magical about Five Percent? A Nationwide Look at Factors That Influence the Optimal Size of State Rainy Day Funds. Public Budgeting and Finance, 21(Summer), Knight, B., & Levinson, A. (1999). Rainy day funds and state government savings. National Tax Journal, 52(3), Navin, J. C., & Navin, L. J. (1994). An evaluation of state budget stabilization funds among Midwestern states. Growth and Change, 25(4), Pollock, R., & Suyderhoud, J. P. (1986). The role of rainy day funds in achieving fiscal stability. National Tax Journal, 39(4), Rodriguez-Tejedo, M. I. (2007). State fiscal institutions: An evolution. University of Maryland, College Park. Sobel, R. S., & Holcombe, R. G. (1996). The impact of state rainy day funds in easing state fiscal crises during the recession. Public Budgeting & Finance, 16(3), Wagner, G. A. (2003). Are state budget stabilization funds only the illusion of savings? Evidence from stationary panel data. Quarterly Review of Economics and Finance 43(2), Wagner, G. A. (2004). The bond market and fiscal institutions: Have budget stabilization funds reduced state borrowing costs? National Tax Journal, 57(4), Wagner, G. A., & Elder, E. M. (2005). The role of budget stabilization funds in smoothing government expenditures over the business cycle. Public Finance Review, 33(4), Wagner, G. A., & Sobel, R. S. (2006). State Budget Stabilization Fund Adoption: Preparing for the Next Recession or Circumventing Fiscal Constraints? Public Choice, 126(1-2),