Austerity, Inequality, and Private Debt Overhang

Austerity, Inequality, and Private Debt Overhang By Mathias Klein a and Roland Winkler b a TU Dortmund University, Department of Economics, Vogelpothsweg 87, 44221 Dortmund, Germany; e-mail: mathias.klein@tu-dortmund.de b TU Dortmund University; e-mail: roland.winkler@tu-dortmund.de Preliminary and incomplete Abstract Using panel data of 17 OECD countries for 1980-2011, we find that the distributional consequences of fiscal consolidations depend significantly on the level of private indebtedness. Austerity leads to a strong and persistent increase in income inequality in periods of private debt overhang. By contrast, there are no discernible distributional effects when private debt is low. Controlling for the state of the business cycle does not affect our results. Keywords: austerity, fiscal policy, inequality, private debt JEL classifications: E62, E64, D63

1 Introduction A recent literature has shown that private debt matters for the impact of fiscal policy. Theoretically, Kaplan and Violante (2014) demonstrate that the aggregate effects of government spending expansions are amplified by the level of private indebtedness. Bernardini and Peersman (2015) provide empirical support for this hypothesis. Using historical U.S. data, they show that the government spending multiplier is considerably larger in periods of private debt overhang. Klein (2016) presents multi-country evidence that the effects of fiscal consolidations on aggregate macroeconomic variables are elevated when private debt is high. Another strand of literature investigates how austerity impacts on inequality (see, e.g., Agnello and Sousa 2014 or Ball, Furceri, Leigh, and Loungani 2013). All of these studies, however, ignore the potentially important role of the credit cycle in shaping the distributional effects of austerity. To the best of our knowledge, ours is the first study that fills this gap by providing empirical evidence on whether private debt overhang influences the distributional consequences of fiscal consolidations. Our results can help inform policy debates about when is the right time for austerity. Based on a panel of 17 OECD countries for 1980-2011, we estimate state-dependent responses of income inequality to fiscal consolidations, identified using the narrative approach of Devries, Guajardo, Leigh, and Pescatori (2011). We differentiate between high and low private debt states, defined as positive and negative deviations of the private debt-to-gdp ratio from trend. We find strong and significant differences in the distributional consequences of fiscal consolidations across the credit cycle. Austerity leads to a severe and significant increase in income inequality when private debt is high. Contrary, fiscal consolidations are associated with no significant distributional consequences when private debt is low. In additional evaluations, we control for fiscal foresight, provide evidence that it matters whether fiscal consolidations are achieved by cutting spending or raising taxes, and explore channels through which our results can be rationalized. Moreover, we rule out that our results are driven by the state of the business cycle. Inequality significantly increases in periods of private debt overhang, irrespective of whether the economy is experiencing a boom or a slump. Likewise, in booms and slumps, austerity has no discernible distributional consequences when private debt is low. Notably, once we control for private debt overhang, we find no significant differences across states of the business cycle. 2

2 Data and Econometric Method We estimate state-dependent impulse responses to fiscal consolidations using local projections as proposed by Jordà (2005). We use annual data of 17 OECD countries for 1980-2011. 1 The data set comprises the Gini coefficient for disposable income as our measure of income inequality, a narrative measure of fiscal consolidation episodes, the cyclically-adjusted primary balance relative to GDP (CAPB), real GPD, the employment rate, the GDP deflator, and an indicator for private indebtedness. Our fiscal consolidation series is the narrative measure of Devries, Guajardo, Leigh, and Pescatori (2011), available for 1980-2009, which we extend for the years 2010-2011. The series contains only those changes in the primary balance to GDP ratio that are motivated by a desire to reduce the budget deficit. Private indebtedness is measured by the private debt-to-gdp ratio. To differentiate between high- and low-debt states, we filter the debt-to-gdp ratio by country-specific HP trends (with smoothing parameter λ = 100). 2 High (low) private debt states are defined as periods with positive (negative) deviations of debt-to-gdp ratios from trend. For each horizon k = 0,..., 4, we estimate the following regression model: Y i,t+k Y i,t 1 = I i,t 1 [β H,k D i,t + ψ H,k X i,t 1 ] + (1 I i,t 1 ) [β L,k D i,t + ψ L,k X i,t 1 ] + α i,k + η t,k + ɛ i,t+k, (1) where Y i,t+k Y i,t 1 is the change in the variable of interest at horizon k (e.g. income inequality), D i,t is the narrative consolidation series, α i,k are country fixed effects, η t,k capture time fixed effects, and X i,t 1 is a vector of control variables (real GDP growth and the change in the respective variable of interest). The dummy variable I i,t 1 captures the state {H, L} of the economy prior to the shock, where I i,t 1 = 1 if private debt is high. Given our specification, β H,k provides the response of Y i,t+k Y i,t 1 to the consolidation shock at time t in high private debt states, whereas β L,k provides the response in low private debt states. To facilitate the interpretation of the results, we normalize responses so that CAPB rises by one percentage point on impact (for details, see Appendix). 3 Results Figure 1 displays large and significant differences in the distributional consequences of fiscal consolidations across the credit cycle. In periods of high private debt, fiscal consolidations lead to a strong and long-lasting rise in income inequality. Four years after the consolidation, 1 Data definitions, data sources, methodological details, and additional results can be found in the Appendix. 2 Our main results are robust to different values of λ, see Appendix. 3

Figure 1: Effects of Fiscal Consolidation on Income Inequality Notes: First two columns display changes in income inequality after fiscal consolidations. Shaded areas indicate 90% confidence bands based on robust standard errors clustered by country. Last column shows the estimated difference between high-debt and low-debt responses. Dots indicate statistically significant differences at 10% level. the Gini coefficient increases by 2 percentage points. By contrast, when private debt is low, fiscal consolidations are followed by hardly any change in income inequality. Additionally, for most periods, the responses differ significantly between high-debt and low-debt states (see right column of Figure 1). Jordà and Taylor (2016) question the exogeneity of the narrative consolidation measure by showing that it has a predictable component. To take account of anticipation effects, we regress D i,t on a set of possible predictors (real GDP growth, change in CAPB, growth rate in GDP deflator) and use the residuum of this regression as a purified measure of consolidation shocks. Table 1 shows that our results are robust to this alternative identification scheme. The upper block of Table 1 displays the change in the Gini coefficient one year after the fiscal consolidation (for the alternative identification and the baseline case). For the sake of brevity, we now and henceforth report responses for horizon k = 1 (our main results are robust across alternative forecast horizons). Table 1: Effects of Fiscal Consolidations (one year after shock) Dependent Variable High Debt Low Debt Difference Gini (baseline) 0.748 0.078 0.826 (0.400) (0.087) Gini (alternative identification) 1.131 0.076 1.207 (0.562) (0.084) Employment Rate 0.921 0.013 0.909 (0.417) (0.127) GDP Deflator 0.492 0.103 0.389 (0.292) (0.141) Notes: Point estimates and robust standard errors clustered by country.,, : 16%, 10%, 5% significance. 4

Table 2: Spending-based vs. Tax-based Consolidations (one year after shock) Spending-based Tax-based Dependent Variable High Debt Low Debt Difference High Debt Low Debt Difference Gini 1.277 0.167 1.440 0.105 0.028 0.134 (0.548) (0.132) (0.743) (0.096) Employment Rate 1.644 0.044 1.687 0.156 0.213 0.058 (0.564) (0.110) (0.518) (0.182) GDP Deflator 1.058 0.166 0.892 0.901 0.096 0.805 (0.623) (0.142) (1.103) (0.176) Notes: Point estimates and robust standard errors clustered by country.,, : 16%, 10%, 5% significance. Two additional estimation results help understand what may explain the debt-dependent distributional consequences of fiscal consolidations (see lower block of Table 1). First, employment falls significantly more strongly in high-debt states than in low-debt states. Together with the facts that i) labor earnings are the primary source of income for most households and ii) employment losses fall disproportionately upon low-income groups (see, e.g., Jefferson 2008), this can explain why income inequality rises more strongly in periods of private debt overhang. Second, when debt is high, we observe a strong and significant fall in the aggregate price level. An unexpected deflation redistributes resources from borrowers to savers and thus generates a rise in income inequality (to the extent that borrowers are generally at the lower part of the income distribution). In low-debt states, by contrast, there are no discernible price effects. Ball, Furceri, Leigh, and Loungani (2013) and Agnello and Sousa (2014) find that spendingbased consolidations lead to a stronger increase in inequality than tax-based adjustments. Table 2 shows that composition also matters for the debt-dependent effects of fiscal consolidations. Tax-based consolidations do not significantly impact on inequality, irrespective of the state of the debt cycle. By contrast, spending-based consolidations are associated with rising income inequality only when implemented in high-debt states. Thus, debt-dependent distributional effects of fiscal consolidations are mainly driven by spending-based consolidation programs. Concerning the channels through which fiscal consolidations may trigger debt-dependent distributional consequences, we find that employment differences across states are significant only for spending-based consolidations. Furthermore, spending-based consolidations in highdebt states lead to a significant decline in the price level, thereby redistributing resources from borrowers to savers. Tax-based consolidations, by contrast, tend to be inflationary, although the price responses are estimated to be insignificant. 5

Jordà and Taylor (2016) and Agnello and Sousa (2014) find that the aggregate and distributional effects of fiscal consolidations are amplified in periods of economic slack. To investigate the role of the business cycle for our results, we now differentiate between booms and slumps following Jordà and Taylor (2016). Booms (slumps) are identified as positive (negative) deviations of log real GDP from country-specific HP trends (λ = 100). Table 3 shows that our results appear in both states of the business cycle. If private debt is high, inequality significantly increases, irrespective of the state of the business cycle. Likewise, if private debt is low, consolidations do not impact significantly on income inequality, neither in booms nor in slumps. Notably, the respective point estimates within debt states are rather similar. This implies that the distributional consequences do not differ significantly between booms and slumps, once we control for private debt overhang. Table 3: Consolidation Effects (k = 1), controlling for the Business Cycle High Debt Low Debt Dependent Variable Slump Boom Difference Slump Boom Difference Gini 0.715 0.686 0.029 0.116 0.123 0.239 (0.399) (0.416) (0.110) (0.194) Notes: Point estimates and robust standard errors clustered by country.,, : 16%, 10%, 5% significance. 4 Conclusion This paper has revealed important nonlinear effects of austerity. In particular, we have shown that the distributional consequences of fiscal consolidations vary considerably over the credit cycle. Our evidence implies that policy makers that are concerned about inequality should implement austerity measures when private debt is low. Conversely, our results suggest adverse distributional consequences of the large-scale austerity programs undertaken in the aftermath of the financial crisis where most countries were confronted with significant private debt overhang. References Agnello, L. and R. M. Sousa (2014). How Does Fiscal Consolidation Impact on Income Inequality? Review of Income and Wealth 60, 702 726. Alesina, A. and S. Ardagna (2010). Large Changes in Fiscal Policy: Taxes versus Spending. In Tax Policy and the Economy, Volume 24, NBER Chapters, pp. 35 68. National Bureau of Economic Research, Inc. 6

Auerbach, A. J. and Y. Gorodnichenko (2012). Measuring the Output Responses to Fiscal Policy. American Economic Journal: Economic Policy 4, 1 27. Ball, L. M., D. Furceri, D. Leigh, and P. Loungani (2013). The Distributional Effects of Fiscal Consolidation. IMF Working Papers 13/151. Bernardini, M. and G. Peersman (2015). Private Debt Overhang And The Government Spending Multiplier: Evidence For The United States. Ghent Working Papers 15/901. Dell Erba, S., T. Mattina, and A. Roitman (2015). Pressure or prudence? Tales of market pressure and fiscal adjustment. Journal of International Money and Finance 51, 196 213. Devries, P., J. Guajardo, D. Leigh, and A. Pescatori (2011). A New Action-Based Dataset of Fiscal Consolidation. IMF Working Papers 11/128. Guajardo, J., D. Leigh, and A. Pescatori (2014). Expansionary Austerity? Evidence. Journal of the European Economic Association 12, 949 968. International Jefferson, P. N. (2008). Educational attainment and the cyclical sensitivity of employment. Journal of Business and Economic Statistics 26, 526 535. Jordà, O. (2005). Estimation and Inference of Impulse Responses by Local Projections. American Economic Review 95, 161 182. Jordà, O. and A. M. Taylor (2016). The Time for Austerity: Estimating the Average Treatment Effect of Fiscal Policy. Economic Journal 126, 219 255. Kaplan, G. and G. L. Violante (2014). A Model of the Consumption Response to Fiscal Stimulus Payments. Econometrica 82, 1199 1239. Klein, M. (2016). Austerity and Private Debt. Conference Paper, Annual Meeting of the German Economic Association 2016 Session: Panel Macroeconometrics, No. C01-V2. Ramey, V. A. and S. Zubairy (2014). Government Spending Multipliers in Good Times and in Bad: Evidence from U.S. Historical Data. Technical report. Ravn, M. O. and H. Uhlig (2002). On adjusting the Hodrick-Prescott filter for the frequency of observations. The Review of Economics and Statistics 84 (2), 371 375. 7

Appendix This appendix includes all data definitions and sources, explains the estimation procedure in more detail and reports the results of additional estimation results and robustness checks mentioned in the text. Data Definitions and Sources The baseline sample covers the period 1980-2011 and the countries Australia, Austria, Belgium, Canada, Germany, Denmark, Ireland, Spain, Portugal, France, Finland, United Kingdom, Italy, Japan, the Netherlands, Sweden and the United States. Table A1: Data Definitions and Sources Variable Definition Source GDP, real Gross domestic product, constant prices, OECD base year OECD GDP, nominal CAPB Narrative fiscal consolidation measure Income inequality Employment rate GDP deflator Total credit to private sector Private debt-to-gdp ratio Gross domestic product, current prices, current PPPs, in US Dollar Cyclically-adjusted primary balance relative to GDP Changes in fiscal policy motivated by a desire to reduce the budget deficit and not by responding to prospective economic conditions Gini coefficient for disposable income Civilian employment as % population (15-64 years old) Gross domestic product, deflator, index, hundreds, base year 2010 End-of-year credit to private non-financial sector from all sectors, market value, in US Dollar, Adjusted for breaks Total credit to private sector divided by GDP, nominal OECD Alesina and Ardagna (2010), for 2010, 2011 OECD series used Devries, Guajardo, Leigh, and Pescatori (2011) and extended for the years 2010, 2011 Standardized World Income Inequality Database OECD OECD Bank for International Settlements Own calculation

Extension of the Narrative Measure In extending the narrative consolidation measure, we follow Dell Erba, Mattina, and Roitman (2015) who provide data of the consolidation measure for the years 2010 and 2011. The extension of the dataset is based on the following two OECD reports: Restoring Public Finances, 2011 and Restoring Public Finances, 2012 Update. These reports outline the economic situation, fiscal consolidation strategy and major consolidation measures for each of the OECD member countries. The country notes in each report lay out each government s rationale for pursuing fiscal adjustment and are used to identify consolidation periods that were motivated by a desire for deficit reduction. Table A2: Narrative Fiscal Shock, 2010-2011 (% GDP) Country 2010 2011 Size Composition Size Composition Australia 0.00 0.00 Austria 0.00 0.90 Spending-based Belgium 0.40 Spending-based 0.40 Tax-based Canada 0.00 0.10 Spending-based Germany 0.00 0.50 Denmark 0.00 0.90 Spending-based Finland 0.20 Spending-based 0.30 Spending-based Spain 2.70 2.20 France 0.00 1.10 Tax-based Ireland 2.70 Spending-based 4.00 Spending-based United Kingdom 0.60 Spending-based 1.20 Spending-based Italy 0.00 0.90 Spending-based Japan 0.00 0.00 Portugal 2.30 Tax-based 3.40 Tax-based Netherlands 0.00 0.30 Spending-based Sweden 0.00 0.40 Spending-based United States 0.00 0.00 Notes: Following Devries, Guajardo, Leigh, and Pescatori (2011), we define consolidations as tax-based and spending-based if the budgetary impact of tax hikes and spending cuts, respectively, is greater than half the total impact. ii

Estimation Procedure Normalization of Impulse Responses. Following Guajardo, Leigh, and Pescatori (2014), we normalize the impulse responses such that the cyclically-adjusted primary balance relative to GDP (CAP B) rises on impact by one percentage point. This is achieved by applying the following two-step approach. In a first step, we estimate for each horizon k = 0,..., 4 CAP B i,t+k CAP B i,t 1 = I i,t 1 [β CH,k D i,t + ψ CH,k X i,t 1 ] + (1 I i,t 1 ) [β CL,k D i,t + ψ CL,k X i,t 1 ] + α i,k + η t,k + ɛ i,t+k, (A.1) where CAP B i,t+k CAP B i,t 1 is the change in the cyclically-adjusted primary balance relative to GDP at horizon k. Given our specification, β CH,k indicates the response of CAP B to the consolidation shock in high private debt states whereas β CL,k shows the effect in low private debt states. In the second step, we estimate for each horizon k = 0,..., 4 Y i,t+k Y i,t 1 = I i,t 1 [β H,k D i,t + ψ H,k X i,t 1 ] + (1 I i,t 1 ) [β L,k D i,t + ψ L,k X i,t 1 ] + α i,k + η t,k + ɛ i,t+k, (A.2) where Y i,t+k Y i,t 1 is the change in the variable of interest (e.g. income inequality) at horizon k. The normalization is now achieved by dividing the high-debt and low-debt response, β H,k, β L,k by the respective impact response of CAP B, β CH,0, β CL,0. Thus, the rescaled responses reported in the paper imply that in both debt states CAP B rises by 1 percentage point on impact. Test for Significant Differences. We test for every variable of interest and at each year of the forecast horizon whether the responses in high-debt and low-debt states are significantly different. Formally, for each horizon k, we test the following null hypothesis: H 0 : β H,k β CH,0 = β L,k β CL,0. (A.3) This hypothesis is be tested with a standard F-test. A similar approach is applied by Ramey and Zubairy (2014) to test whether government spending multipliers statistically differ across the business cycle. iii

Robustness Alternative Definition of Debt States. We define high (low) private debt states as positive (negative) deviations of private debt-to-gdp ratios from (country-specific) HP trends. For our benchmark estimation, we set the smoothing parameter λ equal to 100, which corresponds to the usual value used for annual observations. Table A3 shows that our results are robust to alternative values for λ. As lower threshold we set λ equal to 6.25 as suggested by Ravn and Uhlig (2002). As upper threshold we choose a smoothing parameter of 1000, implying a relatively smooth trend. Irrespective of the specific value assigned to λ, we find strong differences in distributional consequences of fiscal consolidation over the credit cycle. Table A3: Alternative Debt States Definition (one year after shock) Dependent Variable High Debt Low Debt Difference Baseline (λ = 100) 0.748 0.078 0.826 (0.400) (0.087) λ = 6.25 0.503 0.019 0.522 (0.326) (0.104) λ = 1000 0.232 0.096 0.328 (0.160) (0.139) Notes: Point estimates and robust standard errors clustered by country.,, : 16%, 10%, 5% significance. Alternative Identification. Jordà and Taylor (2016) question the exogeneity of the narrative measure. They show that the Devries, Guajardo, Leigh, and Pescatori (2011) series has a predictable component. Therefore, our estimates could be biased when using the narrative measure as indicator for exogenous consolidation shocks. To take account of possible anticipation effects, we combine the approach suggested by Jordà and Taylor (2016) with the forecast error-approach proposed by Auerbach and Gorodnichenko (2012). 1 The procedure consists of two steps. First, we regress the narrative consolidation measure, D i,t, on a set of control variables which possibly include information that help predict the outcome variable (real GDP growth, change in CAPB, change in the GDP deflator). The residuals of this regression measure the unpredictable component of fiscal consolidations. In a second step, the residuals are used as proxy for exogenous austerity innovations in the estimation of Equations (A.1) and (A.2). 1 Auerbach and Gorodnichenko (2012) use the unpredictable component of government spending as proxy for exogenous variations in fiscal expenditures. iv

Controlling for the State of the Business Cycle. We estimate the following two specifications separately for low and high private debt states: CAP B i,t+k CAP B i,t 1 = I C,i,t 1 [ψ CC,k (L)X i,t 1 + β CC,k D i,t ] + I D,i,t 1 [ψ CD,k (L)X i,t 1 + β CD,k D i,t ] + I E,i,t 1 [ψ CE,k (L)X i,t 1 + β CE,k D i,t ] + α i,k + η t,k + ɛ i,t+k, (A.4) Y i,t+k Y i,t 1 = I C,i,t 1 [ψ C,k (L)X i,t 1 + β C,k D i,t ] + I D,i,t 1 [ψ D,k (L)X i,t 1 + β D,k D i,t ] + I E,i,t 1 [ψ E,k (L)X i,t 1 + β E,k D i,t ] + α i,k + η t,k + ɛ i,t+k. (A.5) I C,i,t and I D,i,t now indicate the state of the business cycle of the respective private debt states. In the estimation for high private debt states, I C,i,t measures periods of high private debt that coincide with periods of economic contractions, whereas I D,i,t indicates periods of high private debt that are also characterized by economic expansions. I E,i,t is then a dummy variable for being in the opposing private debt state (low private debt) irrespective of the state of the business cycle. β CC,k, β C,k and β CD,k, β D,k then provide the state-dependent responses in slumps and booms within the high-debt regime, respectively. Analogously, in the estimation for low private debt states, I C,i,t (I D,i,t ) measures periods of low private debt that coincide with periods of economic slumps (booms) and I E,i,t indicates periods of high private debt. As described before, we normalize all responses such that CAP B rises by 1 percentage point on impact. v