WPS7724 Policy Research Working Paper 7724 Do Fiscal Multipliers Depend on Fiscal Positions? Raju Huidrom M. Ayhan Kose Jamus J. Lim Franziska L. Ohnsorge Development Economics Development Prospects Group June 2016 Policy Research Working Paper 7724 Abstract This paper analyzes the relationship between fiscal multipli- fiscal position is strong, while it can be negative when the ers and fiscal positions of governments using an Interactive fiscal position is weak. Second, these effects are separate Panel Vector Auto Regression model and a large data-set and distinct from the impact of the business cycle on of advanced and developing economies. The methodol- the fiscal multiplier. Third, the state-dependent effects ogy permits tracing the endogenous relationship between of the fiscal position on multipliers is attributable to two fiscal multipliers and fiscal positions while maintaining factors: an interest rate channel through which higher enough degrees of freedom to draw sharp inferences. The borrowing costs, due to investors’ increased perception of paper reports three major results. First, the fiscal multipli- credit risks when stimulus is implemented from a weak ers depend on fiscal positions: the multipliers tend to be initial fiscal position, crowd out private investment; and larger when fiscal positions are strong (i.e. when govern- a Ricardian channel through which households reduce ment debt and deficits are low) than weak. For instance, consumption in anticipation of future fiscal adjustments. the long-run multiplier can be as large as unity when the This paper is a product of the Development Prospects Group, Development Economics Vice Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at rhuidrom@worldbank.org, akose@worldbank.org, jlim@worldbank.org, fohnsorge@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Do Fiscal Multipliers Depend on Fiscal Positions? Raju Huidrom, M. Ayhan Kose, Jamus J. Lim, and Franziska L. Ohnsorge Key Words: Fiscal multipliers, …scal position, state-dependency, Ricardian channel, interest rate channel, business cycle JEL Codes: E62, H50, H60 Huidrom: World Bank, Development Prospects Group; rhuidrom@worldbank.org. Kose: World Bank, Devel- opment Prospects Group; Brookings Institution; CAMA; CEPR; akose@worldbank.org. Lim: World Bank, De- velopment Prospects Group; jlim@worldbank.org; Ohnsorge: World Bank, Development Prospects Group; fohn- sorge@worldbank.org. Earlier drafts bene…ted from comments by Syud Amer Ahmed, Jean-Louis Arcand, Raphael Espinoza, Alejandro Izquierdo, Aart Kraay, Sergio Kurlat, Jaime Marquez, Rahul Mukherjee, Ugo Panizza, Assaf Razin, Luis Serven, Cedric Tille, Carlos Vegh, Guillermo Vuletin, Sebastian Weber, Kei-Mu Yi, Hakan Yilmazkuday, Charles Wyplosz, and seminar participants at the Graduate Institute-Geneva, the 2015 Midwest Macro Conference, and the 2016 Conference on Fiscal Policy and the Macroeconomy at Johns Hopkins SAIS. We thank Kiwako Sakamoto for excellent research assistance. Ethan Ilzetski kindly shared data. 1 Introduction During the Great Recession of 2008-09, many countries around the world - both advanced and developing - deployed …scal policy to support activity. As a result, government debt and de…cits increased in many countries, and they remain elevated (Huidrom, Kose, and Ohnsorge 2016). Against this backdrop of weak …scal positions, there has been a revival of interest in …scal policy as a macroeconomic stabilization tool. Yet, there is scant evidence regarding the extent to which …scal policy is e¤ective in stimulating the economy during times of weak …scal position. The objective of this paper is to …ll this gap in the literature. In particular, we ask: do …scal multipliers depend on …scal positions? The question we study follows the …nding in recent literature that an “average” …scal multi- plier which is assumed to apply universally is irrelevant and that multipliers depend on speci…c macroeconomic conditions. For instance, beginning with the work of Auerbach and Gorodnichenko (2012b), recent papers have established that multipliers tend to be larger during recessions than during expansions (Bachmann and Sims 2012; Candelon and Lieb 2013; Owyang, Ramey, and Zubairy 2013). The notion that …scal multipliers depend on the state of the business cycle is well grounded in theory. During recessions, the multiplier e¤ect from government spending can rise due to slack in labor markets, larger frictions in …nancial markets, and an increase in liquidity constrained agents.1 The literature has, thus far, o¤ered a convincing case that the phase of the business cycle should be regarded as a key conditioning state that may in‡ uence the e¢ cacy of …scal policy. Economic theory, however, does not limit the conditioning state to the phase of the business cycle alone. In fact, theory suggests that …scal position of the government, as distinct from the business cycle, can be another important determining factor for the size of …scal multipliers. This state-dependency of multipliers on …scal position can operate via two channels. First, a Ricardian channel: when a government with a weak …scal position implements a …scal stimulus, households expect tax increases sooner than in an economy with strong …scal position (Sutherland 1997; Perroti 1999). The perceived negative wealth e¤ect encourages households to cut consumption and save, thereby weakening the impact of the policy on output. Thus, the net e¤ects of …scal policy on output, the size of the …scal multiplier, may be negligible or even negative. Second, an interest rate channel: when the …scal position is weak, …scal stimulus can increase lenders’ perceptions of sovereign credit risk. This raises sovereign bond yields and hence, borrowing costs across the whole economy. This, in turn, crowds out private investment and consumption, reducing the size of the multiplier. Therefore, both channels suggest that …scal policy is less e¤ective when the …scal stimulus is implemented from a weak initial …scal position.2 1 These e¤ects may be further ampli…ed in the special case where monetary policy is also constrained by the zero lower bound (Christiano, Eichenbaum, and Rebelo 2011; Denes, Eggertsson, and Gilbukh 2013; Erceg and Linde 2014). In advanced economies, …scal policy has received much attention given the crisis-induced zero lower bound environment that has constrained conventional monetary policy (Blanchard et al. 2010 and 2013; Delong and Summers 2012). 2 Sutherland (1997) formalizes the Ricardian channel by postulating that there exists a debt threshold at which the government makes …scal adjustments, via increasing taxes, to remain solvent. Thus, households expect higher taxes to be more eminent when the government conducts an expansionary …scal policy from a high initial level of debt. In Perotti (1999), such expectations of higher taxes can also result in increased tax distortions which are an additional source of negative wealth e¤ects. With regard to the interest rate channel, Bi, Shen, and Yang (2014) theoretically establish that sovereign risk premia can increase nonlinearly as government indebtedness rises. Corsetti et al. (2013) highlight the interest rate is particularly relevant when monetary policy is constrained, for instance during a zero lower bound episode. 2 To estimate …scal multipliers that depend on the …scal position, we use an Interacted Panel Vector Autoregressive (IPVAR) model.3 The model is essentially an extension of an otherwise standard panel structural VAR (SVAR), with the distinction that the VAR coe¢ cients interact with (observable) state variables. Consequently, these coe¢ cients become time-varying, and evolve endogenously according to these states. This results in a framework where the VAR dynamics and hence, the …scal multipliers are conditional on the state variables which we take to be the …scal position. More importantly, since the state-dependency is captured by making use of the full sample, this nonlinear approach allows us to maintain enough degrees of freedom, thus allowing us to draw sharper inferences. This feature of the model is particularly useful when conditioning on multiple states of interest: a feature we exploit when we jointly condition on the …scal position and the phase of the business cycle. The latter exercise allows us to evaluate whether the …scal position is a unique state, di¤erent from the phase of the business cycle, which determines the size of the …scal multipliers. Applying our empirical methodology to a large dataset that covers 34 countries (19 advanced and 15 developing), at the quarterly frequency over the period 1980:1 – 2014:1, we empirically establish that the …scal position is a key conditioning state that determines the size of the …scal multipliers. In particular, estimated multipliers are systematically smaller when the …scal position is weak (i.e. government debt is low), and vice versa when it is strong. In addition, we show that the state-dependency of multipliers on the …scal position is independent of business cycle e¤ects. That is, while we …nd multipliers to be larger during recessions than expansions (consistent with Auerbach and Gorodnichenko 2012b), the weaker (stronger) multiplier e¤ect that derives from a weak (strong) …scal position applies even when the economy is experiencing a recession or an expansion. Furthermore, we provide empirical evidence that such state-dependent e¤ects operate through the two channels highlighted above. When the government conducts expansionary …scal policy during times of high debt, the private sector scales back on consumption in credible anticipation of future tax pressures due to the weak state of public …nances (Ricardian channel) and private investment is suppressed plausibly due to an increase in economy-wide interest rate as perceptions of heightened sovereign risk become stronger (interest rate channel). Some recent empirical studies have documented the importance of …scal positions for …scal multipliers. For instance, Ilzetzki, Mendoza, and Vegh (2013) include measures of …scal fragility in their analyses of multipliers. However, …scal considerations are not the centerpiece of their analysis, and so they apply only certain debt thresholds, as opposed to our more general stance that allows these thresholds to emerge naturally from the data. Using a similar IPVAR approach like ours, Nickel and Tudyka (2014) provide estimates of multipliers that depend on the …scal position for high- income European economies. However, they do not distinguish between the state of the business cycle and …scal position. There is, therefore, an indeterminacy over whether the state-dependency of the multipliers is uniquely attributable to the latter. Using a di¤erent econometric methodology than ours, Auerbach and Gorodnichenko (2012a) discuss the joint conditioning exercise and …nd that large government debt reduces the stimulative e¤ects of expansionary …scal policy even during 3 The model has been used in various areas of empirical macroeconomics: exchange rates (Towbin and Weber 2013); capital ‡ows (Sa, Towbin, and Wieladek 2014); and …scal policy (Nickel and Tudyka 2014). 3 recessions. But their identi…cation strategy requires data on government consumption forecast errors, which essentially limits their study to only OECD countries. Our paper makes three contributions. First, by clearly distinguishing between the state of the business cycle and the …scal position, we establish that the …scal position is a unique state that determines the size of the …scal multiplier. Second, we show the empirical relevance of the trans- mission mechanisms that underlay the state-dependent e¤ects due to …scal position: the Ricardian channel and the interest rate channel. Third, compared to previous studies, our sample includes a larger set of countries covering advanced and developing economies, thus providing a general result on the state-dependent e¤ects due to …scal position. The rest of the paper is organized as follows. Section 2 presents the econometric methodol- ogy. Here, we discuss the IPVAR model, the identi…cation strategy, and the database. We present estimates of state-dependent multipliers in Section 3. In Section 4, we discuss the transmission mechanisms that highlight the Ricardian and the interest rate channels. Section 5 discusses robust- ness exercises and Section 6 concludes. 2 Empirical Methodology 2.1 Econometric Model A standard panel structural VAR (SVAR) estimates a single set of parameters which then yields an “average” or unconditional multiplier. Our objective is to go beyond the unconditional multiplier, and investigate how multipliers can depend on speci…c macroeconomic conditions, in particular …scal position of governments. For that, we deploy the Interacted Panel Vector Autoregressive (IPVAR) model where the main innovation, with respect to a standard panel SVAR, is that the model coe¢ cients vary deterministically according to conditioning (state) variables. Thus, the IPVAR results in a framework where model dynamics and hence, estimated multipliers are conditional on the state variables. By choosing the conditioning variable to be a measure of …scal position in the IPVAR, we estimate multipliers that depend on …scal position. The IPVAR model, in its structural form, is represented by: 2 32 3 2 323 11 12 13 14 1 0 0 0 gcit l;it l;it l;it gcit l l;it 6 21 1 0 0 7 6 gdpit 7 XL 6 6 21 22 2376 7 6 gdpit l 7 24 6 0 ;it 76 7 = 6 l;it l;it l;it 74 l;it 7+Xit F +Uit ; 4 310;it 32 0;it 1 0 5 4 cait 5 l=1 4 31 32 33 5 cait l 5 34 l;it l;it l;it l;it 41 42 43 1 reerit 41 42 43 44 reerit l 0;it 0;it 0;it l;it l;it l;it l;it (1) where for a given country i in period t, gc represents real government consumption, gdp real gross domestic product (GDP), reer the real e¤ective exchange rate, and ca current account balance (as a share of GDP). We take government consumption as the …scal instrument and we track the e¤ects of …scal policy in terms of GDP. Separately, we check the robustness of our results by tracking …scal outcomes in terms of private consumption and private investment (Section 4). Real e¤ective exchange rate and the current account are included in the model to account for open economy features that characterize most of the countries in our sample. The matrix X captures additional controls, which include the time-invariant country …xed e¤ects, and U is a vector of uncorrelated, i:i:d: (structural) 4 shocks. The shock corresponding to government consumption is the …scal shock. Following Ilzetzki, Mendoza, and Vegh (2013), we set the lag length as L = 4.4 The impact matrix A0 (matrix of coe¢ cients on the left hand side of Equation (1)) is lower triangular. This along with the ordering of the variables in the VAR is related to our identi…cation scheme (discussed in detail in the next section). Both the impact matrix A0 and the coe¢ cient matrices Al , l = 1; : : : ; L (on the right-hand side of Equation (1)) comprises time-varying model coe¢ cients that, for any given entry in row j and column k , evolve deterministically according to: jk jk jk l;it = 1;l + 2;l f sit ; (2) where f s refers to the …scal position.5 Our baseline measure of the …scal position is the gov- ernment debt-to-GDP ratio. While the literature has used a variety of measures in this regard, our choice is in line with theoretical macro models, where government debt is the modal state variable.6 Since measures of …scal position are endogenous and move in tandem with the business cycle, we take lagged moving averages of all our …scal measures to control for business cycle e¤ects.7 Equations (1) and (2) jointly denote the IPVAR system. When the law of motion in Equation (2) is suppressed, the IPVAR reduces to a standard panel SVAR which we use to estimate the un- conditional multipliers. The latter serve as a baseline against which we compare the conditional multipliers from the IPVAR. The matrices Al , l = 0; : : : ; L determine the e¤ects of structural shocks on the dynamics of endogenous variables in the VAR system. By conditioning the law of motion of the coe¢ cients in these matrices on the …scal position, as in Equation (2), we are allowing those e¤ects to depend on the …scal position. This scheme allows us to calculate impulse responses and hence estimates of …scal multipliers conditional on a given level of …scal position.8 When estimating the VAR system, we make use of the full sample. This enables us to circumvent the degrees-of-freedom challenge that limits the ability of existing empirical models to account for joint conditioning on multiple states. As standard in the literature, we compute the cumulative …scal multiplier at horizon T as the discounted cumulative change in output until horizon T , as the discounted cumulative government consumption increases by one unit. That is, 4 We use the same lag length of 4 when we report results for speci…c country groups as well. Ilzetzki, Mendoza, and Vegh (2013) note that the optimal lag length in the VAR varies across country groups. Choosing the same lag length (that equals 4) ensures that di¤erences in the multipliers are not attributable to the lag structure of the VAR. 5 Including …scal position in the law of motion in Equation (2) is tantamount to having interaction terms with …scal position in the regressors of Equation (1). For this reason, we do not separately include …scal position as an endogenous variable in the IPVAR. 6 For instance, while Riera-Crichton, Vegh, and Vuletin (2014) condition multipliers on …scal balances, Auerbach and Gorodnichenko (2012a), Ilzetzki, Mendoza, and Vegh (2013), and Nickel and Tudyka (2014) condition on gov- ernment debt. For robustness, we present results when …scal balances are the conditioning variable. 7 In particular, we take the 5-quarter moving average of the …scal position, and then lag it by 2 quarters. Given the average length of the business cycle, this e¤ectively allows us to abstract from changes in the …scal state that may potentially be contaminated by cyclical movements. We allay any residual endogeneity concerns by jointly conditioning on the …scal position and the phase of the business cycle. 8 More precisely, the impulse response calculation assumes that the initial level of …scal position on which the impulses are conditioned prevails throughout the impulse horizon. In practice, …scal position can also respond to the …scal shock and its dynamics can have implications for …scal multipliers (see Ramey and Zubairy (2014) for a similar point). Since …scal position is not an endogenous variable in our IPVAR model, calculating impulse responses while taking into account the endogenous evolution of …scal position in not possible. 5 XT t (1 + r) gdpt t=0 M ultiplier (T ) = XT ; (3) t (1 + r) gct t=0 where r denotes the interest rate. We utilize the median short-term rate in the sample for this purpose which is about 7.4 percent. From (3), the impact multiplier is obtained by setting T = 0 and the long-run multiplier by setting T at an arbitrarily large number, which is taken to be T = 20 (5 years) in our exercise. At T = 20, impulse responses in our model by and large revert to their unconditional means, and so we take this to be representative of the long run. In addition, we speci…cally report multipliers corresponding to one-year (T = 4) and 2-year (T = 8) horizons, when the …scal multipliers typically peak. To calculate the …scal multiplier using the coe¢ cient estimates from the IPVAR, we …rst cumulate the discounted impulses of output and government consumption at di¤erent horizons and compute the ratio of the two impulses. That ratio is then multiplied by the average government consumption to GDP ratio in the sample to yield multipliers.9 2.2 Identi…cation and Estimation To identify …scal shocks, we rely on the standard recursive identi…cation scheme of Blanchard and Perotti (2002). The key timing assumption in this scheme is that discretionary …scal policy does not respond to macroeconomic conditions within the quarter.10 Such a timing assumption can be motivated by implementation lags typically associated with discretionary …scal policy. In the VAR model, this timing assumption is achieved by ordering government consumption …rst in Equation (1), before GDP. The timing assumption for the remaining variables in the VAR follows Ilzetzki, Mendoza, and Vegh (2013): the current account is ordered before the real e¤ective exchange rates. This ordering implies that GDP does not respond to the current account within one quarter, and that the current account does not respond within one quarter when the real e¤ective exchange rate moves. The precise ordering of the latter two variables is, however, immaterial for our main results. Of course, there are alternative identi…cation schemes used in the literature. For instance, Romer and Romer (2010) use a narrative approach to identify exogenous …scal shocks for the US. Auerbach and Gorodnichenko (2012a) proxy exogenous …scal shocks by forecast errors of government consumption for OECD countries. Due to data limitations, neither of these approaches is feasible for our sample that includes developing countries.11 The IPVAR system, comprising Equations (1) and (2), is estimated with ordinary least squares (OLS) applied separately to each equation.12 The estimated system yields model coe¢ cients that 9 This step to calculate multipliers from impulse responses follows Ilzetzki, Mendoza, and Vegh (2013). Since the conditional multipliers are estimated from the panel of countries, they re‡ ect an average estimate across those countries included in the panel. Thus, we use the average government consumption to GDP ratio in the sample to calculate the multipliers rather than country-speci…c government consumption to GDP ratios. 10 One caveat of the recursive identi…cation scheme is that the …scal shocks identi…ed using this scheme may be predicted by private forecasts (Ramey, 2011). Ilzetzki, Mendoza, and Vegh (2013), who use a similar identi…cation scheme and sample of countries like ours, provide evidence that this is unlikely the case. 11 There are alternative identi…cation schemes used in the literature. For instance, Romer and Romer (2010) use a narrative approach to identify exogenous …scal shocks for the US. Auerbach and Gorodnichenko (2012a) proxy exogenous …scal shocks by forecast errors of government consumption for OECD countries. Due to data limitations, these approaches are not feasible for our sample that includes developing countries. 12 Because the error terms are uncorrelated across equations by construction, estimating the IPVAR equation by equation does not result in loss of e¢ ciency. See Towbin and Weber (2013) for a discussion. 6 depend on the …scal position such that a given level of the …scal position maps out to a set of model coe¢ cients. For presenting the results, we evaluate model coe¢ cients at speci…c values of the …scal position which are taken to be the percentiles within the sample. Con…dence bands are calculated by bootstrapping over 300 iterations. We report median estimates, along with the 16 - 84 percent con…dence bands. 2.3 Database Our main database comprises an unbalanced panel that covers 34 countries (19 advanced and 15 developing), at the quarterly frequency over the period 1980:1 – 2014:1.13 Real government consumption and real GDP are based on the quarterly database in Ilzetzki, Mendoza, and Vegh (2013) which are extended until 2014:1 by splicing from the OECD and Haver Analytics. Real e¤ective exchange rates are from the narrow (wherever available) and broad indices of the BIS, and current account from the IMF’ s WEO database. The short-term rate used for discounting the multiplier is drawn mainly from the IMF’ s IFS database. For the robustness results, we augment this database to include quarterly real private consumption and private investment series. These are drawn from the OECD, Haver Analytics, and Eurostat. Additional details on the sources and de…nitions of all of these variables are provided in Table A2 in the Appendix. The government consumption and GDP series (as well as private consumption and private investment) are converted into logarithmic form, and detrended using a linear quadratic trend as in Ilzetzki, Mendoza, and Vegh (2013). The exchange rate is transformed into quarter-to-quarter growth rates, and the current account series is seasonally-adjusted using the X11 routine. All these series are detrended and demeaned on a country-by-country basis, which e¤ectively controls for country …xed e¤ects in the regressions. We also employ another database that is an unbalanced panel with the same cross sectional and time series coverage as before but at the annual frequency. This includes the conditioning variables that are not explicitly required for the identi…cation scheme to be valid in the VAR model but are necessary to estimate the interaction terms. These are government debt and …scal balances as percentage of GDP which are drawn from the IMF’ s WEO (October 2014) database; and government consumption-to-GDP ratios which we obtain from the World Bank’ s WDI database. 3 Results 3.1 Unconditional Multipliers To establish a benchmark, we …rst report estimates of the unconditional multiplier from a standard panel SVAR. For that, we suppress the law of motion for the coe¢ cients in Equation (2). This renders the coe¢ cient matrices Al in Equation (1) invariant across countries and time. Figure 1 presents the unconditional multipliers for the select horizons: on impact, 1 year, 2 years, and long run (5 years). Barring only a few periods in the impulse horizon, the unconditional impulse responses of output due to a positive …scal shock are either negative or insigni…cant.14 Indeed, 13 The list of countries is presented in Table A1 in the Appendix. Our developing-country coverage comprises primarily emerging and frontier market economies that have some ability to tap into international …nancial markets, which renders the …scal solvency risks that underpin our nonlinear crowding-out mechanisms relevant. We exclude low-income countries not only because of data reliability issues, but also because they primarily rely on concessional …nance for government expenditure, which would not re‡ ect the crowding-out mechanisms. 14 When we split our sample into advanced and developing economies, our estimates of the unconditional multiplier are very similar to the ones reported in Ilzetzki, Mendoza, and Vegh (2013). See Figure A2 in the Appendix. 7 across all horizons considered, the uncertainty surrounding these estimates is su¢ ciently large such that the multiplier is essentially statistically indistinguishable from zero. This echoes the often small and the wide range in the estimates of the …scal multipliers as reported in previous studies (see Batini and Weber (2014) for a survey). The unconditional impulse responses presented in Figure 2 corroborate the small and imprecise estimates of the e¤ects that …scal policy has on activity on an average. The main message we take away from above is that the unconditional multipliers can mask important state-dependencies as suggested by theory. The estimates of unconditional multipliers suggest that …scal policy, on average, has no stimulative e¤ects on the economy. However, as recent empirical work shows, …scal policy can be stimulative during speci…c times, for instance during recessions (Auerbach and Gorodnichenko, 2012b). Accordingly, we turn, in the following section, to our conditional multiplier estimates. 3.2 Fiscal Position-Dependent Multipliers Figure 3 presents the set of estimated …scal multipliers (on the vertical axis) that depend on government debt (on the horizontal axis) - our baseline measure of …scal position.15 The four panels correspond to the four horizons previously selected. The …gure shows that there is a systematic link between the size of the multiplier and the …scal position: the median value of the multiplier decreases monotonically in debt, for all horizons reported. That is, the estimated multipliers for all the horizons are positive and signi…cant for low levels of debt, but turn negative or insigni…cant when debt levels are high. For instance, the long run multiplier is close to unity when debt is low (strong …scal position), but is negative for high levels of debt (weak …scal position).16 The di¤erence in the estimated multipliers for low and high levels of debt is particularly signi…cant at longer horizons. Our empirical results therefore lend support to the theoretical insights of earlier studies which show that a weak …scal position can result in stronger crowding-out e¤ects, blunting the stimulative e¤ects of …scal policy (Sutherland 1997; Perroti 1999; Corsetti et al. 2013; Bi, Shen, and Yang 2014).17 Compared with the unconditional multipliers (Figure 1), the conditional multipliers paint a more nuanced picture of the e¤ects of …scal policy. For instance, at the 1-year horizon, the unconditional multiplier is small and insigni…cant. The estimated conditional multipliers at the same horizon (Figure 3) highlight that much of those small and insigni…cant average e¤ects actually re‡ ect episodes when …scal positions are weak. On the other hand, when the …scal position is strong, the conditional …scal multipliers are not only larger than the unconditional estimates but they are also statistically di¤erent from zero. To better grasp the economics underlying these results, it is useful to examine the conditional impulse responses associated with expansionary …scal policy. For the purpose of illustration, we consider impulse responses conditional on two levels of debt: one corresponding to the 10th per- centile in the sample (strong …scal position) and the other corresponds to the 90th percentile (weak 15 Figure A1 in the Appendix provides the distribution of government debt-to-GDP ratio in our sample. Table A3 provides the speci…c percentile values from the sample. 16 Our estimates suggest that the long-run …scal multiplier can be as low as -3 when the …scal position is weak. One way to reconcile such a magnitude is in terms of the private investment response: private investment declines signi…cantly in response to a positive …scal shock during times of weak …scal position (Figure 7). 17 The median multipliers for all horizons are presented in Figure A9 in the Appendix. Our headline result remains robust when we split the sample into advanced and developing economies. See Figure A3 in the Appendix. 8 …scal position). For comparability, the shock size in each case is normalized such that government consumption rises by 1 percentage point on impact. The conditional impulses are shown in Figure 4. While output increases on impact and remains signi…cantly positive for around 2 years when the …scal position is strong, such stimulative e¤ects dissipate after about a year with output falling signi…cantly below zero through till the end of the projection horizon.18 In the case of government consumption, the conditional impulses for both strong and weak …scal positions exhibit some per- sistence in response to the positive …scal shock. However, …scal expansion is more quickly unwound when the …scal position is strong than weak. In other words, relative to the strong …scal position, the government in fact spends more, especially during the initial periods, when …scal position is weak.19 Despite this, it is then quite remarkable that output falls more during times of weak …scal position. This is a result that reinforces our earlier point that a weak …scal position can blunt the stimulative e¤ects of expansionary …scal policy. 3.3 Distinguishing between Two States: Business Cycle and Fiscal Position Recent studies (e.g. Auerbach and Gorodnichenko 2012a) have established that …scal multipliers depend on the phase of the business cycle: they tend to be larger during recessions than expansions. To the extent that …scal position is endogenous and varies according to the business cycle, it is possible that our empirical exercise so far of conditioning only on debt is simply capturing business cycle e¤ects. Controlling for business cycle e¤ects is therefore important to establish that …scal position is a unique state that matters for the size of the …scal multipliers. In this section, we undertake a multi-pronged sequence of empirical exercises designed to demonstrate this. First, we tabulate a number of descriptive statistics to verify that there is little relationship between incidences of the two states. The top panel of Table 1 computes the relative frequency in which countries in our sample experience both a strong or weak …scal position state and a recession.20 The fact is that the two states rarely coincide: for the pooled sample, the concurrence of both states occurs around 2 percent of the time. Even for the category with the highest relative frequency - developing economies with a weak …scal state undergoing a recession - the coincidence of these states is very infrequent (at most 3 percent of the time).21 Second, we perform a number of formal tests that compares the distribution of …scal position (debt-to-GDP ratio) during recessions and expansions. These are reported in the bottom panel of Table 1. It is clear that any di¤erences - to the extent that they exist - are minimal: for instance, the average debt-to-GDP ratio in the expansionary state is 52 percent, compared to 54 percent 18 To allay any concerns that the choice of the 10th and 90th percentiles merely re‡ ects outliers, we report results for the 25th and the 75th percentiles as well (Figure A4 in the Appendix). Even though the di¤erences in the conditional impulse responses are admittedly not as sharp as before, they are statistically signi…cant in the relevant horizons so that our conclusion remains robust. 19 Nickel and Tudyka (2014) also report similar …ndings, although government consumption in their study is un- wound at longer horizons during times of high debt. 20 Like before, the strong …scal position corresponds to the 10th percentile of debt-to-GDP ratio in the sample while the weak …scal position corresponds to the 90th percentile. The recessionary state is de…ned as the period from peak to trough as determined by the Harding and Pagan (2002) business cycle dating algorithm. We discuss alternative approaches to date the business cycle in the robustness exercise in Section 5. 21 We check the relative frequencies at the country level as well. There are several countries for which there are no recessionary episodes either during periods of strong or weak …scal positions. Beyond those countries, incidences of recessions are generally lower when …scal position is strong. These results are available upon request. 9 during recessions. More formally, the t tests all fail to reject the null hypothesis of no di¤erence in means at the standard con…dence levels. In e¤ect, there is little evidence that the distributions of …scal position in our sample di¤er between recessionary and expansionary states. Our third approach is to estimate …scal multipliers conditional on the …scal position while explicitly controlling for business cycle e¤ects. For that, we replace Equation (2) by the following expression that jointly conditions the model coe¢ cients on both the …scal position and the business cycle state as follows: jk jk jk jk l;it = 1;l + 2;l f sit + 2;l bcit ; (4) where bc is an indicator variable that equals 1 for a recession and 0 for an expansion as de- termined by the Harding-Pagan (2002) dating algorithm. The IPVAR system now comprises of Equations (1) and (4). Figure 5 presents estimates of the multipliers for di¤erent …scal positions during recessions. Compared with the earlier result (Figure 3), which e¤ectively spans both phases of the business cycle, the magnitude of the multipliers during recessions (Figure 5) is larger for any given level of …scal position. For instance, the point estimate of the long-run multiplier for the strongest …scal position during recessions almost reaches 1.5, while it is less than 1 when conditioned only on the …scal position. This echoes the empirical literature that has argued that multipliers tend to be larger during recessions (Auerbach and Gorodnichenko 2012b; Bachmann and Sims 2012; Candelon and Lieb 2013). Our results show that multipliers remain dependent on …scal position even during recessions: estimated multipliers decline monotonically in debt for all horizons.22 One important corollary of this result is that the multiplier can be small even during recessions, if the …scal position is weak. This is especially the case in the longer-run, as the implications of a heavier debt burden on private demand ultimately play out. Where the …scal position is especially weak, the multiplier even turns signi…cantly negative. Our central result, therefore, nuances other …ndings that multipliers are larger during recessions than expansions. Conditioning our IPVAR only on the phase of the business cycle, we indeed obtain similar results reported in earlier studies (Figure 6) (e.g. Auerbach and Gorodnichenko 2012b).23 Yet, our results based on the joint conditioning show that, even during recessions, multipliers can be small and even negative if …scal position is weak. 4 Why Fiscal Positions Matter - Transmission Channels The key mechanism that could reduce multipliers when …scal positions are weak, especially in the long run, rests on private agents’ concerns about …scal sustainability when the government implements expansionary …scal policy. As mentioned earlier, this can operate via reductions in private consumption as households anticipate a larger tax burden in the future (the Ricardian channel), or via reductions in consumption and investment by investors facing an ever-greater 22 The state-dependency of …scal multipliers on the …scal position also holds during expansions (Figure A5). For a given level of government debt, the estimated multipliers are larger during recessions than expansions. 23 Despite the di¤erences in econometric approaches and sample, the precise magnitude of our multipliers during recessions and expansions is comparable with Auerbach and Gorodnichenko (2012b). For instance, their point estimate of the long-run multiplier (when government consumption is the …scal instrument) is around 1.47. The corresponding number from the IPVAR model is around 1.67. 10 borrowing costs (the interest rate channel). In this section, we attempt to assess the relative strength of these two channels. We …rst consider the Ricardian channel by augmenting the IPVAR system with private con- sumption, with the model coe¢ cients conditioned on …scal position. For this speci…cation, we order private consumption right after GDP, thus keeping intact the recursive identi…cation scheme of Blanchard and Perotti (2002). Ordering the current account and exchange rates last preserves a domestic macroeconomic bloc in the IPVAR. The conditional impulse responses of private con- sumption and output to the …scal shock, for both the strong and the weak …scal position, are presented in the left panel of Figure 7. As before, the strong and the weak …scal positions re- spectively correspond to the 10th and 90th percentile of debt-to-GDP ratio from our sample. We check the robustness of our results by choosing the 25th and 95th percentiles (Figure A6 in the Appendix). The results are unambiguous: when the …scal position is strong, private consumption rises fol- lowing the impact of the …scal shock, peaking around a year after the shock before returning to its initial level. On the other hand, when the …scal position is weak, private consumption falls precip- itously and remains depressed for around three years after the …scal shock. During these horizons, the di¤erence in the response of private consumption is also statistically signi…cant, judging from the non-overlapping con…dence bands. The divergence in private consumption responses across strong and weak …scal positions is consistent with the Ricardian channel outlined earlier where households reduce consumption in anticipation of more imminent …scal adjustments during times of high government debt (Sutherland 1997 and Perroti 1999).24 Our result on the divergence of private consumption paths provides a new dimension on the debate concerning how private consumption responds to …scal stimulus. Perroti (2005) …nds that private consumption rises in response to a positive …scal shock, while Ramey (2011) shows that private consumption actually declines –a di¤erence which is attributed to the speci…c identi…cation scheme used in these studies. Ilzetzki, Mendoza, and Vegh (2013) reconcile these two contrasting views in terms of monetary policy behavior and argue that once monetary policy is controlled for, …scal policy has expansionary e¤ects on private consumption. Our results, by explicitly showing how a weak …scal position undermines and reverses the response of private consumption, suggest an additional aspect that can help reconcile the con‡icting results found in the literature. For the interest rate channel, we would ideally introduce a proxy for sovereign risk, such as the yield spread, directly into our IPVAR system. However, this is precluded by the paucity of credible long-term rates, especially in developing countries, at the quarterly frequency. We thus proceed with our second-best option, which is to augment private investment into the IPVAR system. As in the case of private consumption, private investment is ordered after GDP but before the current account. Since private investment is particularly sensitive to borrowing costs, a reduction in private investment during times of weak …scal position is indicative of the interest rate channel. Figure 7 presents the conditional impulse responses of private investment for both weak and strong …scal positions.25 24 The estimates of the multipliers with this speci…cation are broadly in line with the baseline estimates. More importantly, our headline result that multipliers depend on …scal position holds when private consumption is included in the IPVAR. See Figure A7 in the Appendix. 25 The multipliers are presented in Figure A8 in the Appendix. Our main result that multipliers depend on …scal position generally holds. 11 The contrast between strong and weak …scal positions for the path of private investment is, again, striking. Investment rises signi…cantly when the …scal position is strong, peaking after around 6 quarters, but remaining sustained through at least 10 quarters. When the …scal position is weak, investment drops sharply after about a year, and never fully recovers, failing to revert even after 5 years. The di¤erence in the impulse responses across strong and weak …scal positions is also statistically signi…cant (barring the initial few quarters). These responses are qualitatively similar to those of private consumption but much larger in magnitude. This suggests that investor concerns about borrowing cost could be an additional channel for dampening the e¤ectiveness of …scal policy. 5 Robustness Exercises We consider three exercises to check the robustness of our headline …ndings: (a) an alternative measure of …scal position where we use …scal balances instead of government debt; (b) an alternative dating scheme of the business cycle similar to Auerbach and Gorodnichenko (2012b) to de…ne recessions as periods with a signi…cant probability of negative output growth; (c) estimating …scal position dependent multipliers while controlling for exchange rate regimes.26 For the last exercise, we estimate the IPVAR model by jointly conditioning the model coe¢ cients on both the …scal position and an exchange rate regime dummy. The law of motion of the model coe¢ cients then is: jk jk jk jk l;it = 1;l + 2;l f sit + 2;l erit ; (5) where er is an indicator variable that equals 1 for a …xed exchange rate regime and 0 for a ‡exible exchange rate regime.27 The measure of …scal position, f s, is taken to be the government debt-to-GDP ratio as in the baseline speci…cation. Table 2 presents the results. The top panel shows the range of estimates of the …scal multipliers for the strongest and weakest …scal positions which, like before, are taken to be …scal balances corresponding to the 10th (weak) and 90th (strong) percentiles from the sample. By and large, our baseline results are qualitatively similar when …scal balances, instead of debt-to-GDP ratios, are used to measure …scal positions. That is, the multipliers are systematically larger for high …scal balances (strong …scal position) than low …scal balances (i.e. weak …scal position). This is true regardless of the horizons considered and when jointly conditioned on the state of the business cycle. The middle panel of Table 2 presents the multipliers using the alternative de…nition of recessions. Our headline result – multipliers depend on the …scal position even during recessions - generally holds, especially at longer horizons. The bottom panel presents the …scal position- dependent multipliers for ‡ exible and …xed exhange rates. Consistent with the literature (e.g. Ilzetzki, Mendoza, and Vegh 2013), for a given …scal position, multipliers are larger in …xed than ‡ exible exchange rate regimes. That said, the state dependency on …scal position still holds: multipliers are larger when …scal position is strong than weak irrespective of the exchange rate regime. 26 exp( zit ) Following Auerbach and Gorodnichenko (2012b), we de…ne the indicator function, I (zit ) = 1+exp( zit ) , where zit is taken to be 7 quarter moving averages of quarter-to-quarter growth rates normalized to have a zero mean and a unit variance. Calibrating as 1:5 > 0, the indicator function pins down the probability of negative output growth. Recessions are then de…ned as periods where that probability exceeds a threshold, which in our implementation is taken to be 80 percent. 27 The exchange rate regime classi…cation follows Ilzetzki, Mendoza, and Vegh (2013) which is extended until 2014 using the IMF de-facto classi…cation of exchange rates. 12 6 Conclusion We document that …scal multipliers tend to be larger when the …scal position is stronger. For instance, our estimates suggest that the long run multiplier can be as big as unity when the …scal position is strong but it can turn negative when the …scal position is weak. A weak …scal position can undermine …scal multipliers even during recessions. Consistent with theoretical predictions, we provide empirical evidence suggesting that weak …scal positions are associated with smaller multipliers through both a Ricardian channel and an interest rate channel. Future work can usefully focus on two issues. First, while data limitations have precluded a deeper and more direct exploration of the interest rate channel, future research, perhaps with more comprehensive and representative data on yield spreads, can seek to improve our understanding of the interest rate channel. Second, …scal-monetary interactions can be studied using a similar empirical model like ours. In particular, one can evaluate whether monetary policy o¤ers a more e¤ective stabilization tool during times of weak …scal position. 13 References [1] Auerbach, A. J. and Y. Gorodnichenko (2012a). “Fiscal Multipliers in Recession and Expan- sion,” in Fiscal Policy after the Financial Crisis, ed. by A. Alesina and F. Giavazzi, Chicago, IL: University of Chicago Press, 63-98. [2] — — –(2012b). “Measuring the Output Responses to Fiscal Policy,”American Economic Jour- nal: Economic Policy, 4, 1-27. [3] Bachmann, R. and E. R. Sims (2012). “Con…dence and the Transmission of Government Spend- ing Shocks,” Journal of Monetary Economics, 59, 235-249. [4] Batini, N.L. and A. Weber (2014). “A Simple Method to Calculate Fiscal Multipliers,”Working Paper 14/93, International Monetary Fund, Washington DC. [5] Benhabib, J. and M. M. Spiegel (2000). “The Role of Financial Development in Growth and Investment,” Journal of Economic Growth, 5, 341-360. [6] Bi, H., W. Shen, and S.C. S. Yang (2014). “Fiscal Limits, External Debt, and Fiscal Policy in Developing Countries,” IMF Working Paper 14/49. [7] Blanchard, O., G. Dellariccia, and P. Mauro (2010): “Rethinking Macroeconomic Policy,” Journal of Money, Credit and Banking, 42, 199–215. [8] Blanchard, O., G. Dellariccia, and P. Mauro (2013): “Rethinking Macroeconomic Policy II: Getting Granular,” IMF Sta¤ Discussion Note 13/03. [9] Blanchard, O. and R. Perotti (2002). “An Empirical Characterization of the Dynamic E¤ects of Changes in Government Spending and Taxes on Output,”Quarterly Journal of Economics, 117, 1329-1368. [10] Booth, L., V. A. Aivazian, A. Demiruc-Kunt, and V. Maksimovic (2001). “Capital Structures in Developing Countries,” Journal of Finance, 56, 87-130. [11] Candelon, B. and L. Lieb (2013). “Fiscal Policy in Good and Bad Times,”Journal of Economic Dynamics and Control, 37, 2679-2694. [12] Christiano, L. J., M. S. Eichenbaum, and S. T. Rebelo (2011). “When Is the Government Spending Multiplier Large?” Journal of Political Economy, 119, 78-121. [13] Corsetti, G., K. Kuester, A. Meier, and G. J. Muller (2013). “Sovereign Risk, Fiscal Policy, and Macroeconomic Stability,” Economic Journal, 123, F99-F132. [14] Corsetti, G., A. Meier, and G. J. Muller (2012). “Fiscal Stimulus with Spending Reversals,” Review of Economics and Statistics, 94, 878-895. [15] Delong, B. and L.H. Summers (2012). “Fiscal Policy in a Depressed Economy,” Brookings Papers on Economic Activity. [16] Denes, M. R., G. B. Eggertsson, and S. Gilbukh (2013). “De…cits, Public Debt Dynamics and Tax and Spending Multipliers,” Economic Journal, 123, F133-F163. [17] Erceg, C. J. and J. Linde (2014). “Is There A Fiscal Free Lunch In A Liquidity Trap?”Journal of the European Economic Association, 12, 73-107. 14 [18] Harding, D. and A. R. Pagan (2002). “Dissecting the Cycle: A Methodological Investigation,” Journal of Monetary Economics, 49, 365-381. [19] Huidrom, R., A. Kose, and F. Ohnsorge (2016). “Having Space and Using It: Fiscal Policy Challenges in Developing Economies,” CEPR Working Paper forthcoming. [20] Ilzetzki, E. O., E. G. Mendoza, and C. A. Vegh (2013). “How Big (Small?) Are Fiscal Multi- pliers?” Journal of Monetary Economics, 60, 239-254. [21] Kraay, A. C. (2012). “How Large Is the Government Spending Multiplier? Evidence from World Bank Lending,” Quarterly Journal of Economics, 127, 829-887. [22] Mineshima, A., M. Poplawski-Ribeiro, and A. Weber (2014). “Size of Fiscal Multipliers,” in Post-Crisis Fiscal Policy, ed. by C. Cottarelli, P. Gerson, and A. Senjadji, Cambridge, MA: MIT Press, 315-372. [23] Nickel, C. and A. Tudyka (2014). “Fiscal Stimulus in Times of High Debt: Reconsidering Multipliers and Twin De…cits,” Journal of Money, Credit, and Banking, 46(7), 1313-1344. [24] Owyang, M. T., V. A. Ramey, and S. Zubairy (2013). “Are Government Spending Multi- pliers Greater during Periods of Slack? Evidence from Twentieth-Century Historical Data,” American Economic Review, 103, 129-134. [25] Perotti, R. (1999). “Fiscal Policy in Good Times and Bad,” Quarterly Journal of Economics, 114, 1399-1436. [26] — — –(2005). “Estimating the E¤ects of Fiscal Policy in OECD Countries,”CEPR Discussion Paper 4842. [27] Ramey, V.A. (2011). “Identifying Government Spending Shocks: It’s all in the Timing,”Quar- terly Journal of Economics 126, 1-50. [28] Ramey, V.A. and S. Zubairy (2014). “Government Spending Multipliers in Good Times and in Bad: Evidence from U.S. Historical Data,” NBER Working Paper No. 20719 [29] Riera-Crichton, D., C. A. Vegh, and G. J. Vuletin (2014). “Procyclical and Countercyclical Fiscal Multipliers: Evidence from OECD Countries,” Policy Research Working Paper 6993, NBER, Washington, DC. [30] Romer, C.D. and D. H. Romer (2010). “The Macroeconomic E¤ects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks,” American Economic Review, 100(3), 763-801. [31] Sa, F. G., P. Towbin, and T. W. Wieladek (2014). “Capital In‡ows, Financial Structure and Housing Booms,” Journal of the European Economic Association, 12, 522-546. [32] Sutherland, A. J. (1997). “Fiscal Crises and Aggregate Demand: Can High Public Debt Reverse the E¤ects of Fiscal Policy?” Journal of Public Economics, 65, 147-162. [33] Towbin, P. and S. Weber (2013). “Limits of Floating Exchange Rates: The Role of Foreign Currency Debt and Import Structure,” Journal of Development Economics, 101, 179-194. 15 Figure 1: Unconditional Multipliers 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 Impact 1 year 2 years Long run Note: The graph shows the unconditional fiscal multipliers for select horizons. These are based on estimates from the SVAR model of Ilzetzki, Mendoza, and Vegh (2013) that features with no interaction terms. Bars represent the median, and error bands are the 16-84 percent confidence bands. 16 Figure 2: Unconditional Impulse Responses 1.20 A. Government Consumption 0.08 B. GDP 1.00 0.06 0.04 0.80 0.02 0.60 0.00 0.40 -0.02 0.20 -0.04 0.00 -0.06 -0.20 -0.08 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Note: The graphs show the unconditional impulse responses (percentage points) to a positive shock to government consumption. These are based on estimates from the SVAR model of Ilzetzki, Mendoza, and Vegh (2013) that features no interaction terms. Solid lines represent the median, and dotted lines are the 16-84 percent confidence bands. 17 Figure 3: Fiscal Position-Dependent Multipliers A. On Impact B. 1 Year 0.3 0.7 0.6 0.2 0.5 0.4 0.1 0.3 0.2 0.0 0.1 -0.1 0.0 -0.1 -0.2 -0.2 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 C. 2 Years D. Long Run 1.0 2.0 0.8 0.6 1.0 0.4 0.0 0.2 0.0 -1.0 -0.2 -0.4 -2.0 -0.6 -3.0 -0.8 -1.0 -4.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 Note: The graphs show the conditional fiscal multipliers for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 18 Figure 4: Conditional Impulse Responses 1.20 A. Government Consumption 0.15 B. GDP 1.00 0.10 0.80 0.05 0.00 0.60 -0.05 0.40 -0.10 0.20 -0.15 0.00 -0.20 -0.20 -0.25 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Weak fiscal position Strong fiscal position Note: The graphs show the conditional impulse responses (percentage points) for the strong (blue) and the weak (red) fiscal positions. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position. The strong fiscal position corresponds to the 10th percentile of debt-to-GDP ratio from the sample, while the weak fiscal position corresponds to the 90 th percentile. Solid lines represent the median, and dotted bands a`re the 16-84 percent confidence bands. 19 Table 1: Comparison of Fiscal and Business Cycle States Full Sample Advanced Developing a Relative frequency Strong fiscal and recessionary state 2.2 2.4 1.8 Weak fiscal and recessionary state 2.1 2.4 3.0 Test of differences b [52.3, 54.0] [57.3, 57.9] [43.4, 44.6] In means 0.25 0.76 0.55 Note: The table shows the association (or lack thereof) between different fiscal positions and the recessionary state. a The top panel shows the relative frequency (percent of observations) of the strong fiscal position and the recessionary state, and that of weak fiscal position and the recessionary state. The frequencies are reported for the full sample and also for specific country groups: advanced and developing economies. The strong (weak) fiscal position corresponds to the 10th (90th) percentile of debt-to-GDP ratio in each sample. The bottom panel reports results that show the statistical significance of the difference of those relative frequencies. The recessionary state is determined by the Harding-Pagan (2002) business cycle dating algorithm. b The top entry shows the average debt-to-GDP ratio (in percent) during expansions (left) and recessions (right). The bottom entry shows the p-values of two-group t-test of difference in means with unequal variances. 20 Figure 5: Fiscal Position-Dependent Multipliers during Recessions A. On Impact B. 1 Year 0.7 1.6 0.6 1.4 0.5 1.2 1.0 0.4 0.8 0.3 0.6 0.2 0.4 0.1 0.2 0.0 0.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 C. 2 Years D. Long Run 2.5 2.5 2.0 2.0 1.5 1.5 1.0 0.5 1.0 0.0 0.5 -0.5 -1.0 0.0 -1.5 -0.5 -2.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 Note: The graphs show the conditional fiscal multipliers during recessions for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model, where model coefficients are jointly conditioned on fiscal position and the phase of the business cycle. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Recessions are determined by the Harding-Pagan (2002) business cycle dating algorithm. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 21 Figure 6: Fiscal Multipliers by Business Cycles only 4 Recessions Expansions 3 2 1 0 -1 -2 On Impact 1 year 2 years Long run Note: The graph shows the conditional fiscal multipliers during recessions at select horizons. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on the phase of the business cycle. Recessions are determined by the Harding-Pagan (2002) business cycle dating algorithm. Bars represent the median, and error bands are the 16-84 percent confidence bands. 22 Figure 7: Transmission Channels 0.15 A. Private Consumption 0.40 B. Private Investment 0.10 0.20 0.05 0.00 0.00 -0.05 -0.20 -0.10 -0.40 -0.15 -0.60 -0.20 -0.25 -0.80 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Weak fiscal position Strong fiscal position Note: The graphs show the conditional impulse responses (percentage points) of private consumption and private investment due to a positive shock to government consumption for the strong (blue) and the weak (red) fiscal positions. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position. The strong fiscal position corresponds to the 10th percentile of debt-to-GDP ratio from the sample, while the weak fiscal position corresponds to the 90 th percentile. Solid lines represent the median, and dotted bands are the 16- 84 percent confidence bands. 23 Table 2: Robustness Checks: Fiscal Multipliers Fiscal position only Recessions and fiscal position Impact 1 year 2 years Long run Impact 1 year 2 years Long run Alternative fiscal position Fiscal balances Strong [0.16, 0.31] [0.29, 0.63] [0.43, 1.10] [0.39, 1.38] [0.37, 0.57] [1.34, 2.00] [1.76, 2.66] [1.09, 2.35] Weak [-0.08, 0.08] [0.16, 0.54] [0.19, 0.76] [-0.05, 0.97] [-0.02, 0.24] [1.04, 1.78] [1.25, 2.16] [0.65, 1.74] Alternative business cycle dates Auberch and Gorodnichenko (2012b) Strong … … … … [0.54, 0.85] [1.23, 1.75] [1.13, 1.78] [0.61, 1.48] Weak … … … … [0.48, 0.80] [1.36, 1.96] [0.85, 1.52] [-0.56, 0.54] Flexible exchange rate and fiscal position Fixed exchange rate and fiscal position Controlling for exchange rate regime Strong [0.02, 0.23] [-0.06, 0.44] [-0.13, 0.62] [-0.37, 0.81] [0.72, 1.01] [2.11, 2.84] [2.26, 3.29] [1.44, 3.18] Weak [-0.14, 0.07] [-0.23, 0.15] [-0.63, -0.08] [-2.28, -0.90] [0.53, 0.90] [1.57, 2.32] [1.10, 2.20] [-7.81, -0.72] Note: The table presents estimates of fiscal multipliers from alternative specifications of the IPVAR model for the strong and the weak fiscal positions. The top panel presents the multipliers using an alternative measure of fiscal position. The middle panel considers an alternative business cycle dating scheme. The bottom panel presents estimates of fiscal position-dependent multipliers for flexible and fixed exchange rate regimes. Fiscal position is strong (weak) when government debt is high (low) or when fiscal balances are low (high). When fiscal position is measured in terms of government debt, the strong position corresponds to the 10th percentile and the weak position corresponds to the 90th percentile. When fiscal balances are taken as the measure of fiscal position, the strong position corresponds to the 90th percentile and the weak position corresponds to the 10th percentile. Numbers reported in square brackets are the 16-84 percent confidence range. 24 List of Figures and Tables in the Supplementary Appendix Figure A1: Distribution of Fiscal Position Figure A2: Unconditional Multipliers Figure A3: Fiscal Position-Dependent Multipliers by Country Groups Figure A4: Conditional Impulse Responses – Alternative Cut-offs Figure A5: Fiscal Position-Dependent Multipliers during Expansions Figure A6: Transmission Channels – Alternative Cut-offs Figure A7: Fiscal Position-Dependent Multipliers with Private Consumption Figure A8: Fiscal Position-Dependent Multipliers with Private Investment Figure A9: Fiscal Position-Dependent Multipliers – All Horizons Table A1: Country Coverage Table A2: Data Sources Table A3: Distribution of Fiscal Position 25 Supplementary Appendix Do Fiscal Multipliers Depend on Fiscal Positions? Raju Huidrom, M. Ayhan Kose, Jamus J. Lim, and Franziska L. Ohnsorge This appendix provides additional results to the main paper. 26 Table A1: Country Coverage Advanced Developing Country Period Country Period Australia 1980:1--2014:1 Argentina 1993:1--2014:1 Belgium 1991:1--2014:1 Bulgaria 1999:1--2014:1 Canada 1980:1--2014:1 Brazil 1995:1--2014:1 Germany 1991:1--2014:1 Chile 1989:1--2014:1 Denmark 1999:1--2014:1 Colombia 2000:1--2014:1 Spain 1995:1--2014:1 Czech Republic 1999:1--2014:1 Finland 1998:1--2014:1 Croatia 2000:1--2014:1 France 1980:1--2014:1 Hungary 1995:1--2014:1 United Kingdom 1980:1--2014:1 Israel 1999:1--2014:1 Iceland 1997:1--2014:1 Mexico 1991:1--2014:1 Italy 1999:1--2014:1 Poland 1999:1--2014:1 Lithuania 1995:1--2014:1 Romania 1998:1--2014:1 Netherlands 1988:1--2014:1 Slovak Republic 1999:1--2014:1 Norway 1996:1--2014:1 South Africa 1993:1--2014:1 Puerto Rico 1980:1--2014:1 Turkey 1998:1--2014:1 Slovenia 1995:1--2014:1 Sweden 1993:1--2014:1 United States 1980:1--2014:1 Note: The table shows the list of countries in the sample. Coverage corresponds to maximum temporal coverage for each country in the baseline specification of the IPVAR model. The coverage differs for specifications used in the robustness exercises. 27 Table A2: Data Sources Variable Definition Frequency Source Output Real gross domestic product (GDP) Quarterly Ilzetzki, Mendoza, and Vegh (2013), OECD, Haver Analytics Private consumption Real personal consumption expenditure Quarterly Ilzetzki, Mendoza, and Vegh (2013), OECD, Haver Analytics Private investment Real private gross fixed capital formation Quarterly Government consumption Real government consumption expenditure a Quarterly Ilzetzki, Mendoza, and Vegh (2013), OECD, Haver Analytics a Government investment Real government gross fixed capital formation Quarterly OECD, Haver Analytics, Eurostat b Real effective exchange rate Real effective exchange rate Quarterly Ilzetzki, Mendoza, and Vegh (2013), BIS Current account Current account as percent of GDP Quarterly Ilzetzki, Mendoza, and Vegh (2013), WEO Government debt General government debt as percent of GDP Annual WEO Fiscal balance Overall fiscal balance as percent of GDP Annual WEO Government consumption-to-GDP ratio Government consumption as percent of GDP Annual WDI Government investment-to-GDP ratio Government investment as percent of GDP Annual WDI Interest rate Short term nominal interest rate Quarterly Ilzetzki, Mendoza, and Vegh (2013) Note: The main source for the quarterly series is Ilzetzki, Mendoza, and Vegh (2013). This database which ends around 2008 is extended by splicing from different sources as mentioned in the table. a This refers to general government for most countries while for a few countries central government is taken. See Ilzetzki, Mendoza, and Vegh (2013). b The narrow index wherever available is taken while the remainder uses the broad index. Details are available upon request. 28 Figure A1: Distribution of Fiscal Position 25 20 Frequency (%) 15 10 5 0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 Debt-GDP Ratio Note: The graph shows the distribution of fiscal position, taken to be the annual government debt-to-GDP ratio, from the sample of advanced and developing economies during the period 1980-2014. Table A3: Distribution of Fiscal Position Percentile 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Debt-GDP Ratio 12.4 17.3 24.8 28.7 32.2 35.9 38.5 40.7 42.9 45.1 47.9 51.4 56.1 60.1 65.5 71.3 80.5 92.4 107.4 Note: The table shows the percentile values of fiscal position, taken to be annual government debt-to-GDP ratio, from the sample of advanced and developing economies during the period 1980-2014. 29 Figure A2: Unconditional Multipliers A. Advanced Economies B. Developing Economies 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 -0.2 -0.2 -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1.0 -1.0 Impact 1 year 2 years Long run Impact 1 year 2 years Long run Note: The graph shows the unconditional fiscal multipliers for select horizons. Panel A uses a sample of advanced economies only while Panel B uses only developing economies. These are based on estimates from the SVAR model of Ilzetzki, Mendoza, and Vegh (2013) that features with no interaction terms. Bars represent the median, and error bands are the 16-84 percent confidence bands. 30 Figure A3: Fiscal Position-Dependent Multipliers by Country Groups Advanced Economies A. On Impact B. Long Run 2.0 3.0 2.0 1.0 1.0 0.0 -1.0 0.0 -2.0 -3.0 -1.0 -4.0 -5.0 -2.0 -6.0 12 26 35 41 47 53 60 71 90 116 12 26 35 41 47 53 60 71 90 116 Developing Economies C. On Impact D. Long Run 0.4 2.0 1.5 0.3 1.0 0.5 0.2 0.0 -0.5 0.1 -1.0 -1.5 0.0 -2.0 -2.5 -0.1 -3.0 12 15 20 27 29 32 35 37 38 40 42 43 45 47 53 60 66 72 81 12 15 20 27 29 32 35 37 38 40 42 43 45 47 53 60 66 72 81 Note: The graphs show the conditional fiscal multipliers for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. The top (bottom) panel is based a sample of only advanced (developing) economies. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from each sample. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 31 Figure A4: Conditional Impulse Responses – Alternative Cut-offs 1.20 A. Government Consumption 0.10 B. GDP 1.00 0.05 0.80 0.60 0.00 0.40 -0.05 0.20 -0.10 0.00 -0.20 -0.15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Weak fiscal position Strong fiscal position Note: The graphs show the conditional impulse responses (percentage points) for the strong (blue) and the weak (red) fiscal positions. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position. The strong fiscal position corresponds to the 25th percentile of debt-to-GDP ratio from the sample, while the weak fiscal position corresponds to the 75 th percentile. Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 32 Figure A5: Fiscal Position-Dependent Multipliers during Expansions A. On Impact B. 1 Year 0.3 0.6 0.2 0.4 0.1 0.2 0.0 0.0 -0.1 -0.2 -0.2 -0.4 -0.3 -0.6 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 C. 2 Years D. Long Run 1.2 3.0 0.8 2.0 1.0 0.4 0.0 0.0 -1.0 -0.4 -2.0 -0.8 -3.0 -1.2 -4.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 Note: The graphs show the conditional fiscal multipliers during expansions for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model, where model coefficients are jointly conditioned on fiscal position and the phase of the business cycle. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Expansions are determined by the Harding-Pagan (2002) business cycle dating algorithm. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands 33 Figure A6: Transmission Channels – Alternative Cut-offs 0.10 A. Private Consumption 0.30 B. Private Investment 0.20 0.05 0.10 0.00 0.00 -0.05 -0.10 -0.20 -0.10 -0.30 -0.15 -0.40 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Weak fiscal position Strong fiscal position Note: The graphs show the conditional impulse responses (percentage points) of private consumption and private investment due to a positive shock to government consumption for the strong (blue) and the weak (red) fiscal positions. These are based on estimates from the IPVAR model, where model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position. The strong fiscal position corresponds to the 25th percentile of debt-to-GDP ratio from the sample, while the weak fiscal position corresponds to the 75th percentile. Solid lines represent the median, and dotted bands are the 16- 84 percent confidence bands. 34 Figure A7: Fiscal Position-Dependent Multipliers with Private Consumption A. On Impact B. 1 Year 0.3 0.6 0.2 0.5 0.2 0.4 0.1 0.1 0.3 0.0 0.2 -0.1 0.1 -0.1 -0.2 0.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 C. 2 Years D. Long Run 1.0 1.5 0.8 1.0 0.6 0.5 0.4 0.0 0.2 -0.5 0.0 -1.0 -0.2 -1.5 -0.4 -2.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 Note: The graphs show the conditional fiscal multipliers for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model that includes private consumption. The model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 35 Figure A8: Fiscal Position-Dependent Multipliers with Private Investment A. On Impact B. 1 Year 0.3 0.8 0.2 0.6 0.4 0.1 0.2 0.0 0.0 -0.1 -0.2 -0.2 -0.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 C. 2 Years D. Long Run 1.0 2.0 1.0 0.5 0.0 0.0 -1.0 -0.5 -2.0 -3.0 -1.0 -4.0 -1.5 -5.0 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 12.4 24.8 32.2 38.5 42.9 47.9 56.1 65.5 80.5 107.4 Note: The graphs show the conditional fiscal multipliers for different levels of fiscal position at select horizons. These are based on estimates from the IPVAR model that includes private investment. The model coefficients are conditioned only on fiscal position. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Fiscal position is strong (weak) when government debt is low (high). Solid lines represent the median, and dotted bands are the 16-84 percent confidence bands. 36 Figure A9: Fiscal Position-Dependent Multipliers – All Horizons A. Fiscal Position Only B. Fiscal Position and Recessions Note: The surf plots show the conditional fiscal multipliers for different levels of fiscal position and across all horizons. These are based on estimates from the IPVAR model. The left panel is when model coefficients are only conditioned on the fiscal position, and in the right panel they are jointly conditioned on the fiscal position and the phase of the business cycle. Government debt as a percentage of GDP is the measure of fiscal position and the values shown on the x-axis correspond to the 5th to 95th percentiles from the sample. Recessions are determined by the Harding-Pagan (2002) business cycle dating algorithm. Fiscal position is strong (weak) when government debt is low (high). Numbers shown are the median estimates of the multiplier 37