The Effect of Unfunded Pension Liabilities on State Borrowing Costs Part 3

Chapter 6. Discussion My study attempts to estimate the supply curve for state credit. The main question I am posing is whether state liabilities are factored into the supply of state credit through GO bond yields. My hypothesis is that state liabilities, both state unfunded pension liabilities and debt, will positively affect GO bond yields through the default risk premium. My study analyzes the economic significance of the effect of these liabilities on bond yields and whether these two liabilities, unfunded pension liabilities and state debt, differ in their effect on yields. Furthermore, I address the simultaneity issue of yields and debt through state fixed effects and later in Section 6.2, with instrumental variables. In addition, I test whether the financial crisis increased the effect of the parameters on bond spreads and whether the credit supply curve is nonlinear. Lastly, in Section 6.3, my study qualitatively assesses the financial crisis’ effect on various state indicators, such as debt, unfunded pension liabilities, revenues, expenditures, and deficits. 6.1 A Simply Panel Data Estimation of the Credit Supply Curve Table 6.1 presents the basic regression results. The regressions are divided into pairs, in which the first regression uses pooled data and the second uses state fixed effects. The pooled regressions use both within-state and between-state variation while the fixed effect regressions use only within-state variation. In other words, fixed effects control for the state-specific characteristics that affect the supply of credit and demand, and thereby partially address the simultaneity problem discussed in Section 4.2. The Ftests for state fixed effects in Table 6.1 show that the fixed effects are significant at the 1% level in all fixed effects regressions, which suggests that omitted state characteristics are important to include in the regression. 25 Table 6.1: GO Bond Spread and State Liabilities (Pooled Data and State Fixed Effects) Dependent Variable: GO Bond Spread Unfunded Pension Liabilities/GSP (1) (2) (Pooled) (FE) 0.062 0.152 (0.011)*** (0.015)*** (3) (Pooled) 0.062 (0.011)*** 0.000 (0.017) 0.117 0.287 0.431 4.29 (P=0.000)*** 55.02 (P=0.000)*** 7.53 1.14 0.20 (P=0.007)*** (P=0.287) (P=0.652) Net Tax-Supported Debt/GSP Unemployment (4) (FE) 0.144 (0.016)*** 0.069 (0.061) (5) (Pooled) 0.025 (0.010)*** 0.016 (0.013) 0.185 (0.016)*** NOCA (6) (FE) 0.062 (0.017)*** 0.104 (0.052)** 0.174 (0.019)*** (7) (Pooled) 0.023 (0.010)** 0.012 (0.014) 0.187 (0.016)*** -0.063 (0.059) [NOCA]*[Net Tax-Supported Debt/GSP] adjusted R 2 F-Test for state fixed effects F-Test for non-fixed effects coefficients 0.286 0.113 4.21 (P=0.000)*** 0.479 0.431 2.22 (P=0.004)*** 78.10 (P=0.000)*** 0.22 0.39 (P=0.642) (P=0.534) (8) (Pooled) 0.024 (0.010)** 0.005 (0.015) 0.184 (0.016)*** -0.298 (0.138)** 0.104 (0.055)* 0.438 F-Test for equality of debt and unfunded pension 0.97 liabilities (P=0.326) *significant at 10%; ** significant at 5%; ***signifcant at 1% Notes: Fixed effects in this table include only state fixed effects. The F-test for significance of coefficents exclude any fixed effects. The F-test for state fixed effects test for the significance of fixed effects. The F-test for equality of debt and unfunded pension liabilities test for the equality of the debt and unfunded liablities coefficients. 26 The regressions in Table 6.1 imply that unfunded pension liabilities have an economically important effect on yields. For example, in all the regressions, the coefficient for unfunded pension liabilities is statistically significant at the 1% or 5% levels. The estimate varies depending on the other variables in the equation, but the fixed effects regression with the most controls in column (6) has the lowest estimated coefficient among the fixed effects regressions. This column shows that a percentage point increase in the ratio of unfunded pension liabilities to GSP leads to a 6.2 basis point (.062%) increase in the default risk premium. Since I use fixed effects and are therefore analyzing the within-state variation, I also use the within standard deviation (1.81%) as a measurement unit, which shows that an increase of one within standard deviation leads to a yield increase of 11.2 basis points. The debt coefficient is insignificant in most regressions in Table 6.1. The insignificance may be due to the simultaneity issue between the debt level of a state and the state’s bond yield. As described in Chapter 4, although this study’s focus is on estimating the supply of credit, in a simple regression analysis it is difficult to separate the supply of credit from the demand for credit. The debt level and promised yield are simultaneously determined by the shifting supply and demand for credit, causing the debt level to affect and be affected by the promised bond yield. Since a higher debt level leads to investors requiring higher yields and higher yields leads states to demand less funds, the estimated coefficients for debt in the Table 6.1 regressions may be biased downward. One method of addressing this endogeneity issue is to use fixed effects. As mentioned in Chapter 4, fixed effects account for state-specific omitted variables that may affect the supply. If not controlled for, state-specific characteristics that shift supply 27 upward (downward) will lead to a decrease (increase) in the level of debt, biasing the estimated coefficient downward. If we use only within-state variation, the variation used in the regression will less likely be due to shifts in the supply curve and more likely be due to movement along the supply curve, since a large amount of supply shift variation is in the between-state variation. However, there are many other possible causes for supply shifts besides state-characteristics, and so I also use instrumental variables later to try to address this simultaneity issue. Using fixed effects and including unemployment, column (6) in Table 6.1 has a significant debt coefficient, and yields increase by 10.4 basis points with a one-percentage point debt level increase. This is equivalent to an increase of 4.5 basis points with an increase of one within standard deviation (0.43%). The second row of F-statistics test for the significance of all the parameters and show that in every column with fixed effects the hypothesis that all coefficients are zero can be rejected. Columns (7) and (8) test the effect of fiscal constraints on bond yields, specifically the effect of NOCA laws. These laws should act as a signal of fiscal responsibility to investors and should therefore lower bond yields. In column (7), the NOCA coefficient does have a negative sign, but it is insignificant. Column (8), which includes an interaction term between the debt and NOCA variables, has a significantly negative coefficient on the NOCA law variable and a weakly significant and positive interaction term. Using the overall standard deviation as a measurement unit for this pooled regression, the practical implication of the NOCA coefficient is -9.5 basis points (-29.8 + 1.95*10.4). These results imply that NOCA laws become less effective as a signal the greater the debt level. However, these results need to be interpreted with caution since NOCA laws may in fact be endogenously determined, and therefore the 28 coefficients in columns (7) and (8) would be biased. Because both NOCA laws and debt levels are determined by political preferences, there may be a correlation between NOCA laws and debt levels.8 The last row of F-statistics in Table 6.1 tests for the equality of the debt and unfunded liabilities coefficients. For all the fixed effects regressions, we cannot reject the null hypothesis that these coefficients are equal. Even in the fixed effect regression in column (6), which has a significant debt variable, we still cannot reject the null hypothesis. Because these are both obligations against future tax revenues, both variables should increase the probability of default by the same amount, which is in line with the results. However, a possible argument is that the unfunded pension liabilities should increase the probability of default more than the debt variable, since the unfunded pension liabilities are “enforceable contractual relationships” and therefore, have stronger legal protections. (Brown and Wilcox, 2009, p. 539) Table 6.2 shows the same regressions as Table 6.1 except using both state and time fixed effects. The variation left to estimate the unfunded pension liabilities and debt variables is the variation that differs from the state mean (state fixed effects) and differs from the year mean (time fixed effects). The unfunded pension liabilities variable is significant in columns (2) and (3) at the 10% level, although the coefficient is much lower, implying a 3.2 basis point increase with a one standard deviation increase above the unfunded pension liabilities mean. The pension liabilities coefficient becomes insignificant with the inclusion of unemployment in column (4), and debt is consistently 8 Other studies, such as Poterba and Reuben (1999), have tried to account for this endogeneity with limited success by controlling for political preference and eliminating the debt variable from the equation. The studies that most successfully analyze the effects of fiscal restrictions on bond yields, Poterba and Reuben (2001) and Lowry and Alt (2001), focus on deficits and how fiscal restrictions mediate the market’s reaction to deficits. 29 Table 6.2: GO Bond Yields and State Liabilities (State and Time Fixed Effects) Dependent Variable: GO Bond Spread (1) (FE) Unfunded Pension Liabilities/GSP Net Tax-Supported Debt/GSP Unemployment adjusted R 2 *significant at 10%; ** significant at 5%; ***signifcant at 1% Notes: Fixed effects in this table includes both time and state fixed effects 0.953 (2) (3) (4) (FE) (FE) (FE) 0.011 0.011 0.010 (0.006)* (0.006)* (0.006) 0.005 0.002 (0.017) (0.016) 0.033 (0.011)*** 0.954 0.954 0.955 insignificant. These regressions suggest that Table 6.1 may not be robust. However, column (1) of Table 6.2 also shows that the adjusted R2 for the equation with no variables and only time and state fixed effects is 0.953, indicating that a large part of the variation in yield spreads is explained by these fixed effects. The large explanatory power of fixed effects makes it difficult to estimate the more relevant liabilities variables. Therefore, using both state and time fixed effects may not be the best model for an analysis of the liabilities variables. Table 6.3 tests the effect of the financial crisis on state bond spreads and whether the financial crisis has changed the effect of the liabilities variables on spreads. The first two regressions divide the data into two sub-groups. Column (1) uses data from 1997 to 2007 year-ends and column (2) uses data from 2008 to 2009 year-ends. Since the yearend for most states is June 30th, the end of the data set in column (1) is June 30th, 2007 and the beginning of column (2)’s data set is July 1st, 2007, corresponding well with the onset of the financial crisis.9 Column (1) in Table 6.3 shows results very similar to the results in column (3) with the full data set (column (6) in Table 6.1), except here the debt coefficient is, once again, insignificant. Column (2) has no significant coefficients and 9 This cutoff date is around the time of the bankruptcy of the two hedge funds managed by Bear Stearns (July 31st, 2007). 30 Table 6.3: The Effect of the Financial Crisis Dependent Variable: GO Bond Spread Unfunded Pension Liabilities/GSP Net Tax-Supported Debt/GSP Unemployment (1) 1997-2007 0.041 (0.018)*** 0.062 (0.052) 0.181 (0.029)*** (2) (3) 2008-2009 1997-2009 0.035 0.062 (0.058) (0.017)*** 0.127 0.104 (0.256) (0.052)** 0.031 0.174 (0.030) (0.019)*** Post-2007 dummy (4) 0.040 (0.015)*** 0.069 (0.047) 0.104 (0.020)*** 0.582 (0.083)*** [Post-2007 dummy]*[Unfunded Pension Liablities/GSP] (5) 0.050 (0.0161)*** 0.067 (0.047) 0.107 (0.020)*** 0.707 (0.102)*** -0.040 (0.019)** [Post-2007 dummy]*[Net Tax-Supported Debt/GSP] Observations adjusted R 2 Chow test without post-2007 dummy 208 0.213 38 0.410 246 0.479 3.55 (P=0.000)*** Chow test with post-2007 dummy F-Test for coefficients 23.22 (P=0.000)*** F-test for equality of debt and unfunded liabilities variables 1.61 (P=0.226) 78.10 (P=0.000)*** 246 0.571 246 0.578 (6) 0.040 (0.015)*** 0.070 (0.048) 0.103 (0.020)*** 0.605 (0.141)*** (7) -0.006 (0.032) 0.050 (0.0162)*** 0.065 (0.047) 0.108 (0.020)*** 0.671 (0.144)*** -0.042 (0.020)** 0.012 (0.032) 246 0.570 246 0.576 1.30 (P=0.171) 83.64 68.76 66.63 57.09 (P=0.000)*** (P=0.000)*** (P=0.000)*** (P=0.000)*** 0.28 (P=0.598) Notes: Regressions in this table use state fixed effects. The first Chow test tests if the estimates are equal in columns (1), (2) and (3). The second Chow tests if the estimates are equal in columns (1), (2), and (4). The first F-statistic tests for the significance of the coefficients in the regression and the second F-statistics tests the equivalence of the debt and unfunded pension liabilities variables in column (3). 31 the F-test for the coefficients is insignificant, although this is not too surprising considering that there are only 38 observations and the within-state variation is severely limited. The first Chow test in Table 6.3 tests whether the estimates in columns (1) and (2) are equal to the estimates in column (3) in order to test for a structural break in the data between the pre- and post- financial crisis. The null hypothesis can be rejected at the 1% level. However, it is difficult to conclude much about the post-financial crisis period since the estimates in column (2) are all insignificant. Columns (4) through (7) use a post-2007 dummy with the same cutoff date as columns (1) and (2) to test the effect of the financial crisis. The post-2007 dummy is significant in all three columns, which corresponds with the jump in state yield spreads in 2007 from negative to positive. Columns (5), (6), and (7) test whether the increased risk aversion during the financial crisis led to investors reassessing the effect of liabilities on the states’ default risk. The expectation is that these interaction terms will be either positive or zero, which would means that, in the former case, investors believe that in this crisis period these liabilities have a greater effect on the states’ default risk or, as in the latter case, that these liabilities continue to have the same effect on risk as before the crisis. Column (5) shows the interaction term between the post-2007 dummy and the unfunded liabilities variable to be significant, but it is significant in the wrong direction. On the other hand, in column (6), the interaction term between the post-2007 dummy and the debt variable is insignificant. As expected, the column (6) results imply that investors believe debt to have the same effect on states’ creditworthiness before and after the crisis. The results in column (5) may indicate that in a more uncertain future, investors worry 32 more about debt than unfunded pension liabilities. This may be because states have mechanisms of reducing their unfunded pension liabilities, such as increasing contributions rates, decreasing cost-of-living adjustments, raising the retirement age, limiting eligibility of deferred retirement option, etc. Many states in 2010 and 2011 have already undergone many of these policy options. (Snell, 2012) In column (7), the debt interaction variable continues to be insignificant and the unfunded pension liabilities variable remains significant and negative. The debt interaction variable in column (7), although still insignificant, has switched signs and now is positive, suggesting that if there is any effect of the financial crisis on the debt variable, it may be in the upward direction. Interestingly, the null hypothesis for the Chow test cannot be rejected if we use column (4) as the restricted regression instead of column (3). This result is consistent with the financial crisis raising risk premiums, but not affecting the other parameters on yield spreads. The F-test of the hypothesis that the debt and unfunded liabilities variables in column (4) are equal cannot be rejected. This is consistent with the results in Table 6.1 and suggests, once again, that the hypothesis that the respective effects of debt and unfunded pension liabilities on yields spreads are the same cannot be rejected. However, the pension liability interaction term in column (7) suggests that this might not be true during a crisis. Table 6.4 reports tests for nonlinearity in the credit supply curve by using squared terms for both the unfunded pension liability and debt variables in the regression. The squared term for the pension liability variable is never significant. On the other hand, the squared term for the debt variable is weakly significant, but is negative, implying that the 33 Table 6.4: Testing for a Nonlinear Supply Curve Dependent Variable: GO Bond Spread Unfunded Pension Liabilities/GSP Net Tax-Supported Debt/GSP Unemployment [Unfunded Pension Liabilities/GSP]2 [Net Tax-Supported Debt/GSP]2 (1) 0.062 (0.017)*** 0.104 (0.052)** 0.174 (0.019)*** (2) 0.077 (0.020)*** 0.110 (0.052)** 0.179 (0.019)*** -0.003 (0.002) (3) 0.065 (0.017)*** 0.309 (0.118)*** 0.166 (0.019)*** -0.024 (0.013)* (4) 0.080 (0.020)*** 0.315 (0.118)*** 0.171 (0.020)*** -0.003 (0.003) -0.024 (0.013)* Adjusted R 2 F-Test for state fixed effects 0.479 0.480 0.485 0.487 2.22 2.18 2.36 2.27 (P=0.004)*** (P=0.005)*** (P=0.002)*** (P=0.003)*** F-Test for non-fixed effects coefficients 78.10 50.07 72.86 50.98 (P=0.000)*** (P=0.000)*** (P= 0.000)*** (P=0.000)*** *significant at 10%; ** significant at 5%; ***signifcant at 1% Notes: The F-test for significance of coefficents exclude any fixed effects. The F-test for state fixed effects test for the significance of fixed effects. The F-test for equality of debt and unfunded pension liabilities test for the equality of the debt and unfunded liablities coefficients. effect that debt has on yield spreads may decrease at higher levels of debt. This suggests that investors become less concerned with debt at higher levels, possibly due to an understanding that states will lower their debt levels when debt becomes too high. While it is only weakly significant, coefficient implies that as the debt to GSP ratio increases by one standard deviation (0.43% percentage points) borrowing costs decrease by 2.1 (2*0.43*2.4) basis points, which holds little economic significance. It is not too surprising that there is no evidence of a backward bending supply curve in these results because it is extremely rare for the equilibrium to be in the nonlinear portion of the supply curve. In the unlikely event that the equilibrium point is in the nonlinear portion, it is usually due to a supply shift, which is not the variation needed to estimate the supply curve. Nevertheless, the results in Table 6.4 suggest that, contrary to the market discipline hypothesis, the credit supply curve may not provide the incentive to limit excessive borrowing. 34 6.2 Addressing the Simultaneity Issue in Time-Series Variation Tables 6.5 and 6.6 use instrumental variables to address the endogenous relationship between yield spreads and debt in time series variation. Although fixed effects control for shifts in the supply curve between states, it does not address the problem of supply curve shifts over time. The four instruments I use are the residuals of the linear trends of each state’s general fund expenditures, total expenditures, general fund revenues, and general fund deficits each divided by the GSP. The rationale of using these instruments is that unexpected deficits are liable to lead to shifts in the demand curve, which is the necessary variation to estimate the supply curve. Table 6.5 shows the first stages of the two-stage least squares model and Table 6.6 shows the corresponding second stages. Columns (1a) in Table 6.5 and (1b) in Table 6.6 instrument debt with the de-trended general fund expenditures as a percent of GSP, columns (2a) and (2b) use the de-trended total expenditures as a percent of GSP, columns (3a) and (3b) use the detrended general fund revenues trend each as a percent of GSP, and columns (4a) and (4b) use the de-trended deficit as a percent of GSP. Columns (5a), (5b), (6a), and (6b) use combinations of these instruments, and column (7), as a comparison, is the same as Table 6.1’s column (6). Unfortunately, as seen in Tables 6.5 and 6.6, there may not be a strong enough correlation between these expenditures and revenues residuals and debt levels. Only the expenditures variables are significant at the 10% level in the first-stage regression, but these estimates have the wrong sign. The instrumented debt variable is significant in none of these regressions. These fluctuations in revenues and expenditures must not have a strong influence on states’ demand for debt. 35 Table 6.5: Expenditure and Reveunue Time Trend Residuals, first stage (IV) IV Dependent Variable: GO Bond Spread Unfunded Pension Liabilities/GSP Unemployment Residuals of general fund expenditures time trend (1a) 0.138 (0.019)*** -0.024 (0.024) -0.203 (0.119)* Residuals of total expenditures time trend (2a) 0.137 (0.019)*** -0.017 (0.025) (3a) 0.137 (0.019)*** -0.026 (0.024) (5a) 0.137 (0.019)*** -0.023 (0.024) -0.215 (0.121)* -0.091 (0.053)* Residuals of general fund revenues time trend 0.008 (0.054) Residuals of general fund deficits time trend Adjusted R 2 partial F-test for significance of multiple instruments (4a) 0.137 (0.019)*** -0.024 (0.024) (6a) 0.135 (0.019)*** -0.016 (0.024) 0.024 (0.055) -0.092 (0.054)* 0.014 (0.055) -0.047 (0.053) 0.229 0.227 0.219 0.221 0.229 1.53 (P=0.218) 0.229 1.48 (P=0.229) (3b) 0.478 (0.718) -2.933 (5.228) 0.094 (0.137) (4b) 0.087 (0.123) -0.077 (0.888) 0.168 (0.030)*** (5b) 0.021 (0.063) 0.401 (0.451) 0.181 (0.022)*** (6b) 0.061 (0.065) 0.108 (0.460) 0.173 (0.023)*** *significant at 10%; **significant at 5%; ***signifcant at 1% Table 6.6: Expenditure and Reveunue Time Trend Residuals, second stage (IV) IV Dependent Variable: GO Bond Spread Unfunded Pension Liabilities/GSP Net Tax-Supported Debt/GSP (instrumented) Unemployment (1b) 0.011 (0.066) 0.475 (0.466) 0.183 (0.023)*** (2b) 0.055 (0.065) 0.147 (0.465) 0.174 (0.022)*** (7) 0.062 (0.017)*** 0.104 (0.052)** 0.174 (0.019)*** 0.472 0.470 0.470 0.470 0.471 0.469 0.479 Adjusted R 2 *significant at 10%; **significant at 5%; ***signifcant at 1% Notes: These two table show in pairs the first stage and second stage estimates for six two-stage least squares regressions. The four instrumental variable in this table are the time trend residuals of general fund expenditures, total expenditures, general fund revenues, and general fund deficits, each divided by GSP. Columns (1a) and (1b) use the first of these instruments, columns (2a) and (2b) use the second instrument, columns (3a) and (3b) use the third, columns (4a) and (4b) use the fourth, columns (5a) and (5b) use the first and the third, and columns (6a) and (6b) use the second and the third. Fixed effects in this table include only state fixed effects. 36 6.3 The Effect of the Financial Crisis on State Indicators This last portion of my analysis examines whether the financial crisis had different effects on states that had higher or lower pre-crisis default-risk premiums. This section looks at a qualitative analysis of the change in various state indicators pre- and mid- crisis, which leads to some interesting differences. As in Section 6.1, the pre-crisis date is taken as June 30th, 2007 and the mid-crisis date is the final year in the data set, June 30th, 2009. As seen in Table 6.7, the states are divided into four quartiles with four or five states in each. The quartiles rank from lowest pre-crisis default-risk premium (or spread) in the first quartile to highest pre-crisis default-risk premium in the fourth quartile. As in Section 6.1 as well, I use within standard deviations to give a sense of the magnitude of these changes, since a regular sample deviation would be misleading due to the different means across states. With only a quick look at the spreads column, it becomes clear that states in these quartiles were affected differently by the financial crisis. In all the quartiles, spreads increased by a little less than two standard deviations, except for the fourth quartile, the worst pre-crisis states. This quartile increased by more than 2.5 standard deviations and almost a full standard deviation above the first quartile. In other words, while the differences between the first three quartiles do not exceed 10 basis points, the difference between the third and fourth quartile is almost 30 basis points. Although the change in spreads increases from the first to the fourth quartile, the change in debt decreases. The first quartile is the only quartile that increases its debt levels, the second quartile decreases its debt level slightly, and the third and fourth quartiles decrease their debt levels by about a half standard deviation. Because the 37 Table 6.7: Effect of the Financial Crisis on State Indicators Quartile ∆ Spread ∆ Unemployment ∆ Debt* ∆ Unfunded ∆ Per capita ∆ GF ∆ Total ∆ GF ∆ GF pension liabilities* real GSP expenditures* expenditures* revenues* surplus* 2.90 -1,743 -0.07 0.80 -0.35 -0.28 1.77 -2,422 0.14 0.46 -0.35 -0.49 1.58 -2,001 0.01 0.87 -0.36 -0.37 1.59 -2,117 0.39 1.32 -0.27 -0.66 0.02 -3,381 -0.58 -1.81 -0.72 -1.63 8.44 -548 1.40 1.79 -0.11 -0.05 1.81 2,395 0.34 0.78 0.50 0.51 1st Quartile** 0.95 4.96 0.20 2nd Quartile 1.05 4.96 -0.08 3rd Quartile 1.09 4.48 -0.26 4th Quartile 1.37 4.90 -0.19 Minimum 0.82 3.40 -0.83 Maximum 2.34 6.70 0.58 Times series standard deviation 0.53 1.55 0.43 * These variables are all as a percent of GSP **The unfunded pension liability for the 1st Quartile is relatively large due to Ohio, which saw the largest increase in unfunded pension liabilities out of all the states analyzed (8.44%). Excluding Ohio, the 1st Quartile would have an unfunded pension liabilities increase of 1.51%. Table 6.8: Annual Required Contributions (ARC) in 2007 and 2009 State California Connecticut Florida Georgia Illinois Maryland Massachusetts Michigan Minnesota New Jersey New York North Carolina Ohio Pennsylvania South Carolina Texas Virginia Washington Wisconsin Average Percent Paid 2007 2009 84.33% 82% 98.80% 96% 108.29% 108% 100% 100% 40.30% 71% 81.38% 84% 98.10% 66% 79.72% 100% 80.47% 78% 57.08% 36% 100% 100% 99.95% 100% 89.99% 94% 38.77% 31% 100.12% 100% 86.08% 99% 85.96% 82% 51.14% 73% 100% 100% 83.18% 84.21% Change -2.33% -2.80% -0.29% 0.00% 30.70% 2.62% -32.10% 20.28% -2.47% -21.08% 0.00% 0.05% 4.01% -7.77% -0.12% 12.92% -3.96% 21.86% 0.00% 1.03% 38 Figu ure 6.1: Bacckward Bending Supplyy Curve preads are so o high for wo orse pre-crissis states, theese states maay be unablee to issue debbt, sp an nd must decrrease their debt d level durring crisis ti mes when riisk premium ms dramatically in ncrease. Fig gure 6.1 show ws graphically a possiblee explanationn for the finddings for sprreads an nd debt levels in Table 6.7. 6 The finaancial crisis caused a shiift in the (baackwardbending) supp ply curve to the left, and d in order to remain withhin the upwaard sloping seection of thee supply curv ve, these stattes’ demand for funds m may have neeeded to decreease. The T question remains, however, as to o how these sstates, in a tiime of crisis, were able tto su ustain a conttractionary policy, p an isssue I will retturn to in a m moment. For th he change in unfunded peension liabillities, it is difficult to disscern differences accross states because b the averages within each quuartile are veery similar accross quartilles. Although A nott disclosed in n Table 6.7, within w each quartile, theere is a rangee of changess in un nfunded pen nsion liabilitiies, the mostt drastic beinng Ohio in thhe first quarttile, which siignificantly increases i thee first quartile average. However, itt is importannt to note thaat on av verage acrosss all states there t is an in ncrease in unnfunded penssion liabilitiees of almost one 39 standard deviation. There can be two possible reasons for this: states have decreased the percentage of their ARC contribution that they paid or the asset value of the pension funds has declined. Looking at Table 6.8, it appears that several states had large shifts upward or downward, but on average, the ARC payment remained the same. A stable, on average, ARC payment suggests that the dramatic increase in unfunded pension liabilities across all states was mainly due to a drop in asset value. Returning to the puzzle of high risk premium states decreasing their debt during crisis times, the expenditures, revenues, and surplus changes give some hints as to how these states compensate for their lack of access to credit markets. The first quartile is the only quartile that has a decrease in general fund expenditures, and the second and third quartiles have an increase in general fund expenditures by less than a half a standard deviation. The fourth quartile, on the other hand, increases its general fund expenditures by over a full standard deviation, which is almost three times as large as the change for the quartile with the second largest increase. The general fund deficit for the fourth quartile is the largest as well. Total expenditures increase by over 1.5 standard deviations for the fourth quartile while for the other quartiles the increase is either around one standard deviation or less. To understand the situation in which riskier states decrease their debt but increase their expenditure and deficits, I believe that it may be helpful to examine more closely two “bad” states, California and Illinois. Looking closely at their CAFRs, it is clear that California and Illinois used several alternative forms of borrowing. In 2009, California’s general fund has a large increase in liabilities to other California state funds. (State of California Comprehensive Annual Financial Report For the Fiscal Year Ended June 30, 40 2008, p. 160; State of California Comprehensive Annual Financial Report For the Fiscal Year Ended June 30, 2009, p. 160) Illinois borrowed from its own funds as well and had a dramatic increase in accounts payable, or delay of payment to vendors. Illinois also included in its general fund part of the American Recovery and Reconstruction Act (ARRA) funds. Comprehensive Annual Financial Report, Fiscal Year Ended June 30, 2008, p. 146-147; Comprehensive Annual Financial Report, Fiscal Year Ended June 30, 2009, p. 152-153) The NASBO calculation of general fund revenues, as described in the Data section in Chapter 5, clearly separates out all federal funds from state funds, meaning the 2009 Illinois general fund deficit calculated in this study is larger than Illinois’s calculation. If California and Illinois can tell us anything about riskier states as a whole, we can possibly conclude that risky states, lacking access to credit markets, were forced to undergo one-time solutions that pushed their problems into the future. Both increases in delayed vendor payment and in liabilities to the government’s own funds did not provide any permanent solution for these states, but only delayed the payment to the near future. As mentioned in the Introduction, both states pre-crisis had already had structural deficits that left little flexibility for additional deficits during the crisis. ARRA funds very possibly helped mitigate the size of these states’ deficits and helped decrease states’ demand for credit by allowing these states to not borrow in the market at a high price. Yet, despite this help, states facing the largest problems pre-crisis were compelled to resort to “pushing-the-can-down-the-road” policies. If states push their deficit problems of today into the following period, then they will have to actually deal with these problems in later periods. In these later periods, 41 theses states’ economic conditions may or may not have improved, and for states with structural deficits, the fiscal situation will only be worse. Once in this situation, the only options that these states have is to either raise taxes or drastically cut spending, as California managed to do in their 2012-2013 state budget. (Brown Jr., 2012, p. 2) Although, in the long run, this is precisely what California must do to address the state’s cyclical and structural deficits, in the short-run, this may cause much harm to the state as a whole. Chapter 7. Conclusion My study provides some hope that financial markets are providing discipline to states that incur unfunded pension liabilities. Investors consistently price in these unfunded liabilities in risk spreads, and in my regressions, pricing of unfunded pension liabilities appears to be more consistent than pricing of debt. The results suggest that, at least during non-crisis times, investors are treating unfunded liabilities as they would any other state debt. However, assuming the market accurately calculates the unfunded pension liabilities with a lower discount rate than the discount rate in state calculations, the true effect that unfunded pension liabilities have on spreads may be lower than my estimates. Despite the finding that the market does price state liabilities into state bond yield spreads, there are reasons for caution as well. As seen in Figures 5.2, debt, on average, has increased around half a percentage point since 1997, while unfunded liabilities have had a consistent upward trend that has only sky-rocketed during the financial crisis. Investors may be pricing in these liabilities, but the unfunded pension liability trend is clearly unsustainable. We may be able to rationalize the latest uptick in unfunded 42 liabilities during the financial crisis as a result of market fluctuations as opposed to state government irresponsibility, but, by that same logic, unfunded pension liabilities should have decreased during boom years, which was largely not the case. The upward trend is even more troubling if we take into account the large variance between states. Although most states have seen their unfunded pension liabilities increase since 2000, states with higher unfunded pension liabilities, as shown in Figure 5.3, are increasing their unfunded pension liabilities at a faster rate. These states have both higher unfunded liabilities and higher growing unfunded liabilities, and they will need to take a different course of action soon. Furthermore, instead of becoming more sensitive to these liabilities during the crisis, the results suggest that investors are becoming less sensitive to the liabilities. As explained in Chapter 6, this could potentially be due to investors understanding that during crisis times, states will reduce the unfunded liabilities. Although the 2010-2011 data that could to show this decrease was not available for verification at the time of this study, recent government reports on state legislation, as explained in Chapter 6, do describe states’ actions to dramatically decrease these liabilities. Investors may be aware that these liabilities are getting too high, but also know that states are aware as well and intend to act on the issue. Yet, why do states need a crisis to address the problem of rising pension liabilities? This question brings us back to the structural deficit issue in California and Illinois. Like unfunded pension liabilities in many states, California and Illinois’s structural deficits increased during boom years and then their deficit problems were exacerbated during the downturn when they were obligated to run an even higher deficit. 43 Both high deficit states and high unfunded pension liability states appear to have been compelled by the crisis to face their long-term problems. Only at a time when both these fiscal issues became dire do state governments appear to feel the need to respond. If problem states only correct themselves when they encounter a crisis, this may point to some flaws in the market discipline hypothesis. Investors do price in liabilities, but this might not serve to meaningfully discipline profligate borrowers until a crisis occurs. The nonlinearity of the credit supply curve may not provide enough incentive to curb these borrowers in the pre-crisis phase.10 The market may provide a disciplining effect only during a crisis, when the supply curve shifts and states end up facing a much steeper supply curve, or even a restriction in credit. However, it is during these crises that states should be allowed to accrue more debt. If high unfunded pension liability or deficit states are forced to deal with their problems during a crisis, this may lead to massive pension benefit and expenditure cuts that could severely hurt the citizens in these states during a time of financial distress. It would be far better if states were induced to deal with the structural imbalances before a crisis so that the needed adjustments are smaller and do not occur at a time of general economic distress. It may be that the discipline from the market has to improve in order to make states keep their fiscal conditions in shape. There should be better preventative measures to avoid continual increases in unfunded pension liabilities and large structural deficits during boom years. Furthermore, states should have greater accountability for pension funding shortfalls through greater transparency. The Public Employee Pension Transparency Act, a bill currently in Congress, attempts to make unfunded pension 10 Correspondingly, my nonlinearity test with squared terms did not find a supply curve that had an increasing slope, and although this may be due to specification problems, this may suggest that the supply curve is not obviously nonlinear. 44 liabilities more transparent to the public. This bill requires states to file an annual report with Treasury Department that presents the state’s unfunded pension liabilities and funding status using the fair market value of the pension fund’s assets and using both the current discount rate as well as the Treasuries discount rate to calculate the liabilities. This bill would also require a 20-year projection for annual contribution and would ban federal bailouts of any state liabilities. If a state did not comply, the state would no longer be eligible for federal tax-exempt bonds. This Act addresses crucial transparency issues with unfunded pension liabilities and incentivizes states to not overpromise benefits or underfund their pension funds. The bill also tries to prevent a moral hazard problem, in which states rely on the federal government to bail out state pension systems. Eliminating explicit and implicit federal guarantees would make market discipline more effective. The elimination of guarantees would cause lenders to become more sensitive to debt, deficits, and unfunded pension liabilities. However, in order to provide a credible alternative to a federal bailout, states should also be given a clear, well-defined procedure for bankruptcy. If investors know that their bond investments can be subject to bankruptcy proceedings, they will likely be more concerned about state liabilities, causing the credit supply curve to become steeper even during non-crisis times. By causing pain for both creditors and states through a more sensitive supply curve, the possibility of bankruptcy would provide a strong incentive for both parties to deal with both unfunded pension and debt problems before these problems become severe. A financial crisis may correct long-term problems, but if states could be incentivized to deal with their liability issues through a more gradual process, the correction would be much less painful. 45 References Alt, James E., and Robert C. Lowry. 1994. “Divided Government, Fiscal Institutions, and Budget Deficits: Evidence from the States.” American Political Science Review 88 (4) (December): 811–828. Barro, Josh. 2011. Unmasking Hidden Costs: Best Practices for Public Pension Transparency. Civil Report. Manhattan Institute for Policy Research. Accessed April 12, 2012. http://www.manhattan-institute.org/html/cr_63.htm. Bayoumi, Tamim, Morris Goldstein, and Geoffrey Woglom. 1995. “Do Credit Markets Discipline Sovereign Borrowers? Evidence from U.S. States.” Journal of Money, Credit and Banking 27 (4) (November): 1046-1059. Biggs, Andrew G. 2010. The Market Value of Public-Sector Pension Deficits. American Enterprise Institute for Public Policy Research. Accessed April 12, 2012. http://www.aei.org/article/economics/retirement/the-market-value-of-public-sectorpension-deficits/. Bloomberg Valuation of Municipal Yield Curves. June 30th, 1997 – June 30th, 2009, via Bloomberg LP, accessed November 21st, 2011. Brown, Edmund G. 2012. Governor’s Budget Summary 2012-13. The State of California. Brown, Jeffrey R., and David W. Wilcox. 2009. “Discounting State and Local Pension Liabilities.” American Economic Review 99 (April): 538-542. Capeci, John. 1994. “Local Fiscal Policies, Default Risk, and Municipal Borrowing Costs.” Journal of Public Economics 53, (1) (January): 73–89. Crain, W. M. 2000. “Volatile States: a Mean-Variance Analysis of American Political Economy.” Paper presented to the Annual Meeting of the American Economic Association, Boston, MA, January. Eaton, Jonathan, and Mark Gersovitz. 1981. “Debt with Potential Repudiation: Theoretical and Empirical Analysis.” Review of Economic Studies 48 (2) (April): 289-309. Eichengreen, Barry, and Tamim Bayoumi. 1994. “The Political Economy of Fiscal Restrictions: Implications for Europe from the United States.” European Economic Review 38, (3-4) (April): 783-791. Goldstein, Morris, and Geoffrey R. Woglom. 1991. “Market-Based Fiscal Discipline in Monetary Unions: Evidence from the U.S. Municipal Bond Market.” SSRN eLibrary (September). Accessed April 12, 2012. http://papers.ssrn.com/sol3/papers.cfm?abstract _id=885044. Lav, Iris J., and Elizabeth McNichol. 2011. Misunderstandings Regarding State Debt, Pensions, and Retiree Health Costs Create Unnecessary Alarm: Misconceptions Also Divert Attention from Needed Structural Reform. Center on Budget and Policy Priorities. Accessed April 12, 2012. http://www.cbpp.org/cms/index.cfm?fa=view&id=3372/ Lowry, Robert C, and James E Alt. 2001. “A Visible Hand? Bond Markets, Political Parties, Balanced Budget Laws, and State Government Debt.” Economics & Politics 13 (1) (March): 49-72. Lyons, Kimberly. 2010. 2010 State Debt Medians Report. Special Comment. New York: Moody’s Investors Services, Inc. 46 Metcalf, Gilbert E. 1993. “Federal Taxation and the Supply of State Debt.” Journal of Public Economics 51 (3) (July): 269–285. Munnell, Alicia H., Jean-Pierre Aubry, and Laura Quinby. 2010. The Impact of Public Pensions on State and Local Budgets. Center for Retirement Research at Boston College. ———. The Impact of Pensions on State Borrowing Costs. 2011. Center for State and Local Government Excellence. Murray, Matthew. 2011. Cyclical and Structural State Budget Deficits in California and the Intermountain West. Washington DC: The Brookings Institution and the Morrison Institute for Public Policy, Arizona State University. Accessed April 12, 2012. http://www.brookings.edu/papers/ 2011/0105_state_budgets.aspx. Novy-Marx, Robert, and Joshua D. Rauh. 2011a. “Fiscal Imbalances and Borrowing Costs: Evidence from State Investment Losses.” SSRN eLibrary (March). Accessed April 12, 2012. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1507800. ———. “Policy Options for State Pension Systems and Their Impact on Plan Liabilities.” 2011b. Journal of Pension Economics and Finance 10 (2) (April): 173-194. ———. “Public Pension Promises: How Big Are They and What Are They Worth?” 2010. SSRN eLibrary (October). Accessed April 12, 2012. http://papers.ssrn.com/sol3/papers. cfm?abstract_id=1352608. Office of the Controller, State of California. State of California Comprehensive Annual Financial Report For the Fiscal Year Ended June 30, 2009. Accessed April 12, 2012. http://www.sco.ca.gov/ard_state_cafr.html. ———. State of California Comprehensive Annual Financial Report For the Fiscal Year Ended June 30, 2008. Accessed April 12, 2012. http://www.sco.ca.gov/ard_state_cafr.html. Panizza, Ugo, Federico Sturzenegger, and Jeromin Zettelmeyer. 2009. “The Economics and Law of Sovereign Debt and Default.” Journal of Economic Literature 47 (3) (September): 651-698. Pension Liability, 1997-2008. The Trillion Dollar Gap: Underfunded State Retirement Systems and the Roads to Reform. Washington DC: The Pew Center on the States. http://www.pewcenteronthestates.org/ttw/trends_map_data_table.aspx?trendID=23& assessmentID=126&year=2008&mode=table. Poterba, James M., and Kim S. Rueben. 1999. “State Fiscal Institutions and the U.S. Municipal Bond Market.” In Fiscal Institutions and Fiscal Performance. 181-207. University of Chicago Press. ———. “Fiscal News, State Budget Rules, and Tax-Exempt Bond Yields.” 2001. Journal of Urban Economics 50 (3) (November): 537-562. Rauh, Joshua D. “Are State Public Pensions Sustainable? Why the Federal Government Should Worry About State Pension Liabilities.” SSRN eLibrary (May 2010). Accessed April 12, 2012. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1596679. Samuels, Nicholas. 2007. 2007 State Debt Medians. Special Comment. New York: Moody’s Investors Services, Inc. Snell, Ronald K. 2012. Pension and Retirement Plan Enactments in 2011 State Legislatures. National Conference of State Legislatures. 47 State Expenditure Report. 1998-2009. Washington DC: National Association of State Budget Officers. Accessed April 12, 2012. http://www.nasbo.org/publicationsdata/state-expenditure-report/archives. State of Illinois, Office of the Comptroller. Comprehensive Annual Financial Report, Fiscal Year Ended June 30, 2009. Accessed April 12, 2012. http://www.ioc.state.il.us/ ?LinkServID=30AB8324-1CC1-DE6E2F4874335C40BAE2&showmeta=0 ———. Comprehensive Annual Financial Report, Fiscal Year Ended June 30, 2008. Accessed April 12, 2012. http://www.ioc.state.il.us/ ?LinkServID=30AB83241CC1-DE6E-2F4874335C40BAE2&showmeta=0 Stiglitz, Joseph E., and Andrew Weiss. 1981. “Credit Rationing in Markets with Imperfect Information.” The American Economic Review 71, (3) (June): 393-410. United States Department of Commerce. “Bureau of Economic Analysis.” Accessed April 17, 2012. http://www.bea.gov/. United States Department of Labor. “U.S. Bureau of Labor Statistics.” Accessed April 17, 2012. http://www.bls.gov/. United States Department of the Treasury. “Daily Treasury Yield Curve Rates.” Accessed April 12, 2012. http://www.treasury.gov/resource-center/data-chart-center/interestrates/Pages/TextView.aspx?data=yieldAll. Urahn, Susan K. 2010. The Trillion Dollar Gap: Underfunded State Retirement Systems and the Roads to Reform. The Pew Center on the States. ———. 2010. Widening Gap: The Great Recession’s Impact on State Pension and Retiree Health Care Costs. Washington D.C.: The Pew Center on the States. 48