Fiscal Policy, Business Cycles and Economic Stabilisation: Evidence from Industrialised and Developing Countries*

Transcription

1 FISCAL STUDIES, vol. 28, no. 4, pp (2007) Fiscal Policy, Business Cycles and Economic Stabilisation: Evidence from Industrialised and Developing Countries* YOUNG LEE and TAEYOON SUNG Hanyang University, Korea Yonsei University, Korea Abstract This paper empirically investigates the responsiveness of fiscal policy to business cycles and the effectiveness of fiscal policy in reducing economic fluctuations. From regressions on the responsiveness of fiscal policy to business cycles, we find that the government s current expenditures and subsidies & transfers move counter-cyclically, whereas taxes and capital expenditures move pro-cyclically. Using economic fluctuations in neighbouring countries as an instrumental variable, we show that ordinary least squares (OLS) estimates understate the responsiveness of fiscal policy to economic fluctuations. We also find that fiscal policy responds asymmetrically over economic fluctuations. In investigating the effectiveness of fiscal policy in reducing economic fluctuations, we mitigate omitted variable bias by adding four important factors military expenditures, oil production, economic fluctuations in neighbouring countries and fiscal policy responsiveness to business cycles. The results of effectiveness regressions are consistent with the responsiveness regressions, *Submitted November The authors would like to thank Woocheol Kim, Seokhoon Kang and seminar participants at IIPF 2005 Congress, Hanyang University, KAIST Graduate School of Management, KDI School of Public Policy and Management, Sogang University and the Annual Conference of the Korean Association of Public Finance and Economics for comments on their previous draft. This work was supported by the research fund of Hanyang University (HY-2007-I). JEL classification numbers: E6, H6.. Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA

2 438 Fiscal Studies highlighting the importance of current expenditures, especially subsidies and transfers, in responding to business cycles and stabilising the economy. I. Introduction Using a comprehensive data-set of 94 countries, this paper empirically examines how fiscal policy responds to economic fluctuations and how effectively fiscal policy has reduced them. From responsiveness regressions, we find that the government s current expenditures and subsidies & transfers move counter-cyclically, while taxes and capital expenditures move procyclically. A comparison of the OECD with the non-oecd reveals that government expenditures respond much more counter-cyclically in the OECD than they do in the non-oecd. We also observe that fiscal policy responds asymmetrically over economic fluctuations, thus implying that government debt can grow over business cycles. From effectiveness regressions, we confirm that economic fluctuations are smaller in larger economies with bigger governments. We also find that military expenditures, oil production and economic fluctuations in neighbouring countries are positively associated with economic fluctuations and that the responsiveness of fiscal policy to business cycles is negatively associated with economic fluctuations. The three main contributions of this paper are as follows. First, in investigating fiscal policy responsiveness to business cycles, we conduct an instrumental variable (IV) estimation to correct the downward bias of OLS regressions, which is associated with the reverse causality effect of fiscal policy on economic fluctuations. 1 Lane (2003) also conducted IV estimation, but it was not the main focus of his study. 2 As an instrumental variable, we use the weighted average of economic fluctuations in each country s neighbouring economies, weighting by the inverse of the distance between the two countries. The validation of the instrumental variable is based on the fact that a country s economic fluctuations are highly related to those of neighbouring economies but the fluctuations in neighbouring economies are exogenous to the country s fiscal policy. Our IV results indicate that OLS estimates significantly underestimate the responsiveness of fiscal policy to business cycles. 1 An expansionary fiscal policy response to a recession, for example, would lead to an increase in output, thus implying that the reverse causality effect of fiscal policy to economic fluctuations would bias the actual responsiveness of fiscal policy downward. 2 Lane does not report IV estimation results in his paper, but indicates that the IV results are available on his website. Our paper is significantly different from Lane (2003) in the following respects. First, our analysis uses a data-set of 94 countries, which is much more comprehensive than Lane s of 22 OECD countries. Second, our IV has stronger explanatory power than the IV used in Lane. Third, we conduct our analysis on revenues as well as government expenditures. Fourth, our analysis on the effectiveness of fiscal policy, which Lane did not analyse, provides additional insight into the responsiveness regressions.

3 Fiscal policy, business cycles and economic stabilisation 439 Second, we use a more comprehensive data-set than the existing literature, consisting of both 22 OECD and 72 non-oecd countries. Using a more comprehensive data-set enables us to utilise larger cross-country variations than previous studies and to examine systematically the differences between developed and developing countries. Third, we also examine the effectiveness of fiscal policy and confirm the consistency of empirical results between the responsiveness and effectiveness regressions. In investigating the effectiveness of fiscal policy in reducing economic fluctuations, we mitigate omitted variable bias by adding four notable factors affecting the degree of a country s economic fluctuations military expenditures, oil production, economic fluctuations in neighbouring countries and the responsiveness of fiscal policy to business cycles. Specifically, the first two factors military expenditures and oil production are found to have a very strong explanatory power for economic fluctuations in the non-oecd. We also find that in the OECD, the responsiveness of fiscal policy is even more important in explaining economic fluctuations than the sheer size of the government. The rest of the paper is organised as follows. Section II reviews previous studies, Section III describes empirical specifications and Section IV explains the data. Note that in Sections II to IV, we discuss literature, specification and data for both the responsiveness and the effectiveness regressions. Section V reports regression results for the responsiveness of fiscal policy to business cycles. Section VI reports regression results for the effectiveness of fiscal policy in stabilising the economy. Section VII concludes the paper with a summary and discussion of policy implications. II. Literature There is a large body of empirical work on the relationships between fiscal policy, business cycles and stabilisation. 3 Below, we review previous studies on how fiscal policy responds to economic fluctuations and then review previous studies on how effectively fiscal policy has reduced them. 1. Responsiveness of the fiscal position to business cycles Our regression results on the responsiveness of fiscal policy in the OECD qualitatively confirm the counter-cyclical responsiveness of fiscal policy reported in previous studies (Fiorito, 1997; Sorensen, Wu and Yosha, 2001; 3 See, among many others, Fiorito and Kollintzas (1994), Gavin et al. (1996), Fiorito (1997), Stein, Talvi and Grisanti (1998), Agenor, McDermott and Prasad (1999), van den Noord (2000), Talvi and Vegh (2000), Sorensen, Wu and Yosha (2001), Fatas and Mihov (2001a and 2001b) and Lane (2003).

4 440 Fiscal Studies Lane, 2003). 4,5 However, in terms of magnitude, our IV estimators suggest a much stronger responsiveness of fiscal policy than those obtained from OLS regressions. As for the non-oecd, most previous studies have focused on Latin America, reporting pro-cyclical patterns in fiscal policy (Gavin et al., 1996; Gavin and Perotti, 1997; Stein, Talvi and Grisanti, 1998). However, previous studies using data for other non-oecd countries report a diversity of patterns. For example, while Talvi and Vegh (2000) show a pro-cyclical pattern in a wider sample of developing countries, Agenor, McDermott and Prasad (1999) report a counter-cyclical pattern in some non-oecd countries Effectiveness of fiscal policy in stabilising the economy Our empirical investigation of the effectiveness of fiscal policy in stabilising the economy mitigates the omitted variable bias in previous studies. Several papers have already examined the relationship between government size and output volatility. Notwithstanding the theoretical ambiguity mentioned in Gali (1994) about the effect of government size on output volatility, many studies provide empirical evidence that government size has a negative relationship with output volatility (Gali, 1994; van den Noord, 2000; Fatas and Mihov, 2001a and 2001b). 7 Rodrik (1998) takes this stabilisation role of 4 By examining the stylised facts of government finance in the G7, Fiorito (1997) suggests that government deficits are counter-cyclical. Sorensen, Wu and Yosha (2001) study the cyclical properties of US state and local government fiscal policy. According to their study, the budget surpluses (deficits) of both are cyclical (counter-cyclical) over short- and medium-term horizons. In a sample of the OECD, Lane (2003) suggests that current government spending tends to be mildly counter-cyclical, while the government consumption component of current spending is pro-cyclical, thus implying that the countercyclical behaviour of current government spending emanates from the behaviour of government transfers and/or debt interest payments. 5 As an exception to the fiscal policy pattern, Fiorito and Kollintzas (1994), using data for G7 countries, report that government consumption does not show a clear pattern in the G7 countries, whereas consumption and investment are pro-cyclical. 6 In Talvi and Vegh (2000), the total sample consists of six G7 countries, 14 other industrial economies and 36 developing countries not restricted to Latin America. They suggest that, as measured by correlations, the fiscal policy variables have pro-cyclical patterns with output in developing countries. However, according to Agenor, McDermott and Prasad (1999), who document the stylised features of macroeconomic fluctuations for 12 developing countries, the fiscal impulse (defined as the ratio of government spending to government revenues) is negatively correlated with business cycles. They also suggest that government expenditure is counter-cyclical, while government revenues are acyclical in some countries but counter-cyclical in others. 7 Using data for 22 members of the OECD, Gali (1994) provides empirical evidence indicating the presence of a negative relationship between output variability and the tax/gdp ratio. Van den Noord (2000) shows that, in the OECD, the larger the share of government expenditure in domestic output, the greater the sensitivity of the fiscal position to fluctuations in economic activity, thereby dampening cyclical fluctuations. Fatas and Mihov (2001a) report a strong negative correlation between government size and output volatility, both for the OECD and for US states. Fatas and Mihov (2001b) also present evidence that large governments reduce the volatility of output. They show that this result is robust to the introduction of controls such as openness, GDP, GDP per capita and average growth.

5 Fiscal policy, business cycles and economic stabilisation 441 the government sector as given and suggests that more open economies tend to have bigger governments. 8 III. Specification We need two specifications one for testing the responsiveness of fiscal policy to business cycles and the other for testing the effectiveness of fiscal policy in stabilising the economy. 1. Responsiveness of the fiscal position to business cycles In measuring the responsiveness of fiscal policy to business cycles, it is important to note that fiscal policy and GDP can both have a time trend. If we do not de-trend fiscal variables and GDP, their overall trend may generate a spurious correlation between them. To extract the cyclical component of the time-series variables, we employ two alternative detrending procedures: (i) Hodrick and Prescott filtering (hereafter, HP filtering ) and (ii) first-differencing. HP filtering is a widely used way to detrend, but it is not an entirely satisfactory approach to de-trending. Cogley and Nason (1995) show that HP filtering can generate business-cycle dynamics (namely, the Slutsky effect ) even if none are present in the original data. Furthermore, Ehlgen (1998) reports that not only when the conditions for HP filter optimality are not satisfied, but even when the HP filter is optimal, its application can lead to distortions. Ashley and Verbrugge (2006) also show that the de-trending of stochastically trended data through application of the HP filter yields notably mis-sized hypothesis tests. Considering these potential weaknesses of the HP filter pointed out in the literature, we also employ an alternative method of de-trending namely, first-differencing. 9 As suggested in Solon, Barsky and Parker (1994) and Shin (2000), we assume that the trends in fiscal positions and GDP are second-order polynomial functions of time t. 10 Assuming further that the relationship between the de-trended fiscal position and the de-trended (log of) GDP, i.e. the business cycle, is linear, we can derive equation (1): (1) 2 2 ln( Zit ) ( at 2 + at 1 + a0) = α0 + α 1 ln( GDPit ) ( bt 2 + bt 1 + b0) 8 Alesina and Wacziarg (1998) provide some evidence of a positive relationship between openness and the size of government transfers, which is consistent with Rodrik s argument concerning the stabilising role of governments in open economies. They also cast some doubt on the direct link between openness and the share of government consumption. 9 In the empirical analysis, we find that the results from HP filtering and from first-differencing are broadly consistent. However, the first-differencing results are more robust than HP filtering to variations in specifications and sample. 10 Using a third- or fourth-order polynomial function delivers very similar empirical results.

6 442 Fiscal Studies where Z it is the fiscal position of country i at time t. Taking the first difference and adding country fixed effects and year effects, we can derive equation (2): (2) Δ ln( Zit ) = α0 + α1δ ln( GDPit ) + α2t + γ jd j + δsts + εit j 1 s 1 where D j is the country dummy of country j and T s is the year dummy of year s. 11 Equation (2) is the main specification we use to estimate the responsiveness of the fiscal position to business cycles a responsiveness regression specification. We use the log of expenditures, or the log of tax revenues, as a key fiscal position variable. In this log log specification, α 1 is the elasticity of the fiscal position to business cycles. Note that we de-trend fiscal variables and GDP by including second-order polynomial functions of time t; thus trends in the fiscal variables and GDP are already controlled for. When α 1 is greater than 1, the fiscal variable increases more than proportionally in a boom. This implies that the ratio of the fiscal variable to GDP increases in a boom. Our working definition of pro-cyclical fiscal policy is that the ratio of the fiscal variable to GDP increases (decreases) in a boom (recession). To stabilise the economy, taxes need to move pro-cyclically, indicated by α 1 greater than 1, while government expenditures need to move counter-cyclically, indicated by α 1 less than A key issue that has not been addressed sufficiently in the previous empirical literature is the reverse causal relationship of fiscal policy to the business cycle. To address this endogeneity issue, we adopt an IV estimation by using as instruments the weighted average GDP growth rates in neighbouring countries, weighting by the inverse of the distance between the two countries. 13 While GDP growth rates of neighbouring countries are not much influenced by the country s fiscal policy, the correlation between GDP growth rates in neighbouring countries and the country s GDP growth rates is remarkably high in the data. 14 This suggests that the weighted average of 11 This regression specification in differences is robust to non-stationarity problems. 12 We also examine the fiscal policy response by using the ratio of expenditures, taxes or budget surpluses to GDP as a fiscal position variable instead of the log of them. In this specification, the responsiveness can be examined by the sign of α 1. The regression results for the ratios are very consistent with those for the logs. We focus on the regressions with the logs instead of the ratios in order to make our results comparable to those of previous studies. The detailed results for the ratio regressions are also available from the authors. 13 Distance measures between two countries come from CEPII Centre d Etudes Prospectives et d Informations Internationales ( Geodesic distances are calculated following the great circle formula, which uses the latitudes and longitudes of the most important cities/agglomerations in terms of population. 14 Like Lane (2003), we also tried using trade volume as a weight. However, we decided to use the inverse of distances instead of trade volume as the weight because inverse-of-distance-weighted GDP

7 Fiscal policy, business cycles and economic stabilisation 443 GDP growth rates in neighbouring countries is an exogenous factor affecting the growth rate of a country without much being influenced by the country s fiscal policy, thus satisfying the characteristics of a good instrument. Although using many valid instrumental variables has the potential to improve efficiency, it also makes the usual inference procedures inaccurate (Hansen, Hausman and Newey, 2007). Furthermore, when the instruments are weak, the usual large-sample approximations provide a misleading basis for inference, in the sense that the estimator is biased towards the OLS estimator. Particularly, this occurs in our study. Thus we decided to use this variable as the only IV without adding other weak IVs. 2. Effectiveness of fiscal policy in stabilising the economy To empirically investigate the effectiveness of fiscal policy in stabilising an economy, we use the standard specification used in the existing literature and simply add four additional factors. As in previous studies, such as Gali (1994) and Fatas and Mihov (2001a and 2001b), we begin by regressing the standard deviation of GDP growth rates on the log of GDP (lgdp), GDP growth rates (ΔlGDP), the log of GDP per capita (lgdppc), government size (Gov/GDP) and trade openness (Open). The rationales for including these variables as independent variables are as follows. First, larger countries may experience smaller economic fluctuations because they have a more diverse set of economic activities, including types of industries, consumers and technology. For example, the world economy consisting of the most diverse set of economic activity shows smaller fluctuations than each individual country does. Second, economic fluctuations measured by the standard deviation of GDP growth rates would be larger when the average GDP growth rate is larger. Thus we need to control for average growth rates. Third, an argument could be made about the possibility that richer economies are less volatile for example, due to more developed financial systems. Fourth, a larger government can be associated with smaller economic fluctuations because the government sector is more stable than the private sector and can conduct fiscal and monetary policy tools to stabilise the economy. The effects of government size on economic stabilisation have been the focus of the literature (for example, Fatas and Mihov (2001a)). Finally, trade openness can increase economic fluctuations because it makes the country more vulnerable to external forces (Rodrik, 1998). Besides the above independent variables, the four factors we additionally consider are the ratio of military expenditures to GDP (Mil), the ratio of oil growth has much stronger explanatory power with the GDP growth rates of each country (R-squared = 0.366) than the trade-weighted GDP growth rate (R-squared = 0.040). Furthermore, trade volume itself is an endogenous variable whereas geographical distance is exogenous.

8 444 Fiscal Studies production to GDP (Oil), economic fluctuations in neighbouring countries (SD) and the responsiveness of fiscal policy to business cycles (Rsp). Military expenditures, a component of government expenditure, might destabilise rather than stabilise the economy because they are determined by exogenous factors and are not adjusted to stabilise the economy. Furthermore, high military expenditure might be a symptom of an unstable economy. Oil production might be associated with larger economic fluctuations because of the instability of oil prices. Countries surrounded by unstable economies face larger external shocks. Finally, the responsiveness of the fiscal position to business cycles can reduce volatility. The extent to which taxes and expenditure are adjusted to stabilise the economy can affect economic fluctuations. The measure of the responsiveness of the fiscal position to the business cycle we use is the estimated elasticity of taxes to the business cycle for each country using equation (2). 15 The resulting specification for the effectiveness of fiscal policy in stabilising an economy is (3) StDev( Δ lgdp) = β + β lgdp + β Δ lgdp + β lgdppc i 0 1 i 2 i 3 i + β ( Gov / GDP ) + β Open + β Mil + β Oil + β SD + β Rsp + ε. 4 i i 5 i 6 i 7 i 8 i 9 i i IV. Data and sample Our data come from the World Bank s World Development Indicators (WDI). The only exception is the ratio of oil production to GDP, which comes from OPEC s Annual Statistical Bulletin. 16 Our main sample for the responsiveness regressions consists of 1,982 observations covering 94 countries and 27 years from 1972 to The period is the same for both the OECD and the non-oecd. Each of the countries in our data-set has more than five observations and complete data regarding total tax revenues, total government expenditures and real GDP. 1. Responsiveness of the fiscal position to business cycles Table 1 reports summary statistics for the responsiveness regressions. The average of the first difference in the log of real GDP, which can also be 15 In addition to the estimated elasticity of taxes to the business cycle, we use the estimated elasticity of government expenditures to the business cycle for each country using equation (2). The two estimated elasticities give similar results. 16 Fiscal data from other data sources, including OECD Economic Outlook, the Office of Tax Policy Research (OTPR) World Tax Database and IMF s Government Finance Statistics, are also used for a robustness check. The regression results are very similar when these fiscal data from other sources are used. The results of using other data sources are available from the authors. WDI data are used mainly because of their broader country coverage.

9 Fiscal policy, business cycles and economic stabilisation 445 interpreted as the average GDP growth rate, is 3.49 per cent. The weighted average of GDP growth rates in neighbouring countries, weighting by the reciprocal of bilateral distance, is 3.17 per cent. Compared with the non- OECD countries, the OECD countries have lower growth rates, not only in terms of their own GDP growth rates but also in terms of the GDP growth rates of neighbouring countries. This pattern is due to the fact that OECD members tend to be located close to other members. The averages of the first differences in the log of tax revenues and in the log of expenditures are 4.32 per cent and 4.47 per cent respectively, suggesting that tax revenues and expenditures increase, on average, more rapidly than GDP. The fact that expenditures have higher growth rates than tax revenues indicates deterioration in fiscal positions. Among the major components of expenditures and tax revenues, subsidies & transfers and commodity taxes increase most rapidly. TABLE 1 Summary statistics for responsiveness regressions All OECD Non- OECD n Mean Min. Max. Mean Mean (st. dev.) (st. dev.) (st. dev.) First difference in GDP (%), ΔlGDP 1, (5.00) (2.52) 3.78 (5.65) GDP growth rates of other countries (%), ΔlGDP j i 1, (2.27) (1.83) 3.33 (2.40) First difference in log of total expenditures (%) 1, (13.49) (5.58) 4.59 (15.51) First difference in log of current expenditures (%) 1, (13.11) (5.58) 4.55 (15.13) First difference in log of capital expenditures (%) 1, (28.99) (16.68) 3.18 (32.70) First difference in log of subsidies & transfers (%) 1, (28.02) (11.04) 5.06 (32.65) First difference in log of tax revenues (%) 1, (15.57) (5.65) 4.50 (18.00) First difference in log of income tax revenues (%) 1, (20.57) (10.87) 5.38 (23.34) First difference in log of commodity tax revenues (%) 1, (17.32) (7.92) 6.14 (19.81) First difference in log of social security tax revenues (%) 1, (23.64) (23.19) 6.16 (23.87) Source: World Bank, World Development Indicators.

10 446 Fiscal Studies 2. Effectiveness of fiscal policy in stabilising the economy Summary statistics for the effectiveness regressions are reported in Table 2. Note that, in the effectiveness regressions, we utilise cross-sectional variations of 94 countries. The standard deviation of the GDP growth rates over the sample period is, on average, 4.50, with a much lower figure for the OECD (2.40) than for the non-oecd (5.11). The average tax ratio is slightly less than 20 per cent and the average ratio of total expenditures to GDP is 28 per cent. The estimated average responsiveness of expenditure to economic fluctuations is In particular, the responsiveness is estimated to be 0.52 in the OECD and 0.33 in the non-oecd, suggesting that OECD government expenditures move much more counter-cyclically to business cycles than non-oecd government expenditures. In contrast, taxes show cyclical TABLE 2 Summary statistics for effectiveness regressions All (n = 94) OECD (n = 22) Mean (st. dev.) Min. Max. Mean (st. dev.) St. dev. of GDP growth rates (2.26) (0.63) St. dev. of average GDP growth rates in other countries (0.82) (0.40) Total expenditures / GDP (%) (11.36) (9.52) Tax revenue / GDP (%) (9.19) (8.68) Estimated responsiveness of expenditure to business cycle (3.66) (1.10) Estimated responsiveness of taxes to business cycle (3.14) (0.95) Trade / GDP (%) (46.89) (41.01) GDP growth rates (%) (2.15) (0.86) GDP (1995 US$ million) 194, ,677, ,067 (704,421) (1,348,822) GDP per capita (1995 US$) 6, ,939 20,138 (9,066) (8,534) Military expenditures / GDP (%) (4.10) (1.18) Oil production / GDP (%) (12.70) (3.05) Sources: World Bank, World Development Indicators; OPEC, Annual Statistical Bulletin Non- OECD (n = 72) Mean (st. dev.) 5.11 (2.13) 2.12 (0.91) (11.35) (7.63) 0.33 (4.12) 1.15 (3.56) (48.65) 3.71 (2.38) 39,135 (85,065) 2,742 (3,781) 3.86 (4.60) 6.17 (14.22)

11 Fiscal policy, business cycles and economic stabilisation 447 behaviour in response to economic fluctuations, with an estimated average responsiveness of 1.14 (1.09 in the OECD and 1.15 in the non-oecd). The average ratio of trade to GDP is about 70 per cent. The ratio of oil production to GDP is, on average, 5 per cent, with a very large standard deviation, and the ratio of military expenditures to GDP is about 3.5 per cent. V. Responsiveness regression results In this section, we report the results of regressions on the responsiveness of fiscal policy to economic fluctuations. Table 3 reports the regression results for total expenditures and tax revenues using two different methods of detrending. Note that in all regressions, we add country and time dummies to control for country fixed effects and worldwide year effects. Simple OLS results without any de-trending are reported in columns 1 and 6, which show that expenditures (tax revenues) move counter-cyclically (pro-cyclically). 17 These OLS estimation results without any de-trending may bias fiscal responsiveness toward 1, due to the time trend of the fiscal variables and GDP. As expected, de-trending leads to a stronger cyclical response of fiscal policy. In columns 2 and 7, we report that total expenditures (tax revenues) show a slightly more counter-cyclical (pro-cyclical) response after detrending with the HP filter. De-trending with first-differencing in columns 4 and 9 appears to correct the bias more strongly. In columns 3, 5, 8 and 10, we report IV estimation results. Note that our instrumental variable the neighbouring countries GDP growth rates (ΔlGDP j i,t ) is strongly correlated with the country s GDP growth rate (ΔlGDP it ) with a correlation coefficient of 0.61 (p-value = 0.00), suggesting the validity of our instrumental variable. 18 Fiscal variables are estimated to be much more responsive to economic fluctuations in IV estimations than in OLS estimations. In the regressions for government expenditures, the IV estimation coefficient for the HP-filtered log of GDP, 0.60 (0.62 for the firstdifferencing of GDP), is much smaller than the OLS coefficient for the HPfiltered log of GDP, 0.85 (0.72 for the first-differencing of GDP). In the regression for tax revenues, the IV estimation coefficient for the HP-filtered log of GDP, 1.34 (1.39 for the first-differencing of GDP), is much larger than the OLS coefficient for the HP-filtered log of GDP, 1.18 (1.28 for the first-differencing of GDP). Our results indicate that OLS estimates underestimate the responsiveness of fiscal policy since they include the impact of the fiscal variables on economic fluctuations. 17 Note that in column 1 is less than 1, while in column 6 is greater than We do not conduct the overidentifying restriction test since we consider a single instrument.

12 TABLE 3 Responsiveness of the fiscal position to business cycles Dep. var.: Log of total government expenditures Log of total tax revenues De-trending: None HP filtering First-differencing None HP filtering First-differencing OLS OLS IV OLS IV OLS OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) lgdp (0.025) ** (0.032) ** HP lgdp (0.048) ** (0.080) ** (0.056) ** (0.093) ** ΔlGDP (0.066) ** (0.101) ** (0.073) ** (0.113) ** Trend (0.002) ** (0.001) * (0.001) * (0.138) (0.138) (0.003) (0.002) ** (0.002) ** (0.154) (0.154) Constant (0.475) (0.060) + (0.060) + (6.033) (6.039) (0.618) ** (0.070) (0.070) (6.742) (6.748) No. of obs. 1,982 1,982 1,982 1,982 1,982 1,982 1,982 1,982 1,982 1,982 R-squared F test of the null hypothesis that the estimated coefficient of GDP equals 1 F-value Prob > F Notes: Country dummies and year dummies are included in all regressions but not reported. Standard errors are given in parentheses. ξ significantly different from 1 at 10 per cent; significantly different from 1 at 5 per cent; significantly different from 1 at 1 per cent. + significant at 10 per cent; * significant at 5 per cent; ** significant at 1 per cent

13 Fiscal policy, business cycles and economic stabilisation 449 Our main results in columns 5 and 10, which are the IV estimation results for the whole sample, imply that total expenditures increase by 0.62 per cent and tax revenues increase by 1.39 per cent when there is a 1 per cent increase in GDP growth rates. 19 Table 4 examines the results of the responsiveness regression for subsamples. Columns 2, 6, 10 and 14 report the results for the subsample without the five largest countries: the US, Japan, France, the UK and Italy. The exogeneity of economic fluctuations in other countries, and thus the strength of IV, may be weak for large countries because economic fluctuations in these large countries can significantly affect economic conditions in adjacent countries. However, the estimation results indicate that dropping the five largest countries does not affect the main results. Table 4 also reports the IV estimation results separately for the OECD and the non-oecd. In the regressions for total expenditures, the estimated coefficient for the HP-filtered log of GDP (the first-differenced log of GDP) for the OECD is 0.35 (0.05), which is much smaller than 1, compared with 0.64 (0.64) for the non-oecd. We also estimate the GDP elasticity of government expenditures for each country, as reported in Tables A1 and A2 in the appendix. The country-by-country regression results also confirm that expenditures respond much more counter-cyclically in the OECD than in the non-oecd. Within the OECD, expenditures in European welfare states generally move more counter-cyclically, while expenditures in the US, Japan, Canada and members with lower incomes, such as Portugal, Turkey and Greece, move less counter-cyclically. This finding indicates the importance of social security in the counter-cyclical response of expenditures. Taxes are found to be more pro-cyclical in the non-oecd than in the OECD. In an attempt to at least partly explain this somewhat puzzling finding, we discuss factors affecting the responsiveness of taxes to the business cycle. This responsiveness can be affected by the progressivity of Notes to Table 4: Country dummies and year dummies are included in all regressions but not reported. The five countries dropped in columns 2, 6, 10 and 14 are the US, Japan, France, the UK and Italy. Standard errors are given in parentheses. ξ significantly different from 1 at 10 per cent; significantly different from 1 at 5 per cent; significantly different from 1 at 1 per cent. + significant at 10 per cent; * significant at 5 per cent; ** significant at 1 per cent. 19 To consider the time persistency of fiscal policy, we also conduct regressions in which the lagged difference in fiscal variables is added as an independent variable. The regression results do not change substantially when the lagged values of fiscal variables are added. The detailed results are available from the authors.

14 TABLE 4 Responsiveness of the fiscal position: different samples, IV Dep. var.: HP-filtered log of total expenditures HP-filtered log of total revenues All Excl. five largest OECD Non-OECD All Excl. five largest OECD Non-OECD (1) (2) (3) (4) (5) (6) (7) (8) HP-filtered lgdp (0.080) ** (0.083) ** (0.202) (0.094) ** (0.093) ** (0.097) ** (0.188) ** (0.111) ** No. of obs. 1,982 1, ,428 1,982 1, ,428 R-squared F test of the null hypothesis that the estimated coefficient of HP-filtered lgdp equals 1 F value Prob > F Dep. var.: First-differenced log of total expenditures First-differenced log of total revenues All Excl. five largest OECD Non-OECD All Excl. five largest OECD Non-OECD (9) (10) (11) (12) (13) (14) (15) (16) ΔlGDP (0.101) ** (0.104) ** (0.228) (0.119) ** (0.113) ** (0.116) ** (0.226) ** (0.133) ** No. of obs. 1,982 1, ,428 1,982 1, ,428 R-squared F test of the null hypothesis that the estimated coefficient of ΔlGDP equals 1 F value Prob > F

15 Fiscal policy, business cycles and economic stabilisation 451 the tax system, tax-reporting behaviour, institutions, tax composition and the size of the unofficial economy. First, progressive tax systems can lead to a stronger pro-cyclical movement of tax revenues. Members of the OECD generally have more progressive tax systems than non-oecd countries, implying that tax revenues move more pro-cyclically in the OECD. Second, when taxpayers smooth their reported income, tax revenues move less procyclically. Lee (2005) presents a model in which taxpayers smooth their reported income out of fear that a large change in reported income over time could invoke a tax audit. Taxpayers in the OECD may have a stronger incentive to smooth their reported income because of a more advanced tax administration system that can trace over-time changes in reported income. Therefore tax revenues would move less pro-cyclically in the OECD. Third, we conjecture that economies without a strong legal-political-institutional infrastructure may respond to a positive productivity shock by increasing taxes more than proportionally (for example, see Tornell and Lane (1999) for this voracity effect ). Therefore taxes move less pro-cyclically in the OECD than in the non-oecd. Taking these factors into account, it is ambiguous whether the OECD or the non-oecd has more pro-cyclical tax revenue. Table 5 disaggregates expenditures and taxes into their main components. 20 Current expenditures, especially in the OECD, show a strong counter-cyclical response, while capital expenditures are estimated to be procyclical. Subsidies and transfers are estimated to be slightly more counter- Current exp. TABLE 5 Responsiveness of the fiscal position, by component, IV Dep. var. is first-differenced log of: Capital exp. Subsidies & transfers Total Subsidies & transfers Dep. var. is first differenced log of: Income Commodity taxes taxes Social security taxes All (0.106) ** (0.234) ** (0.239) (0.111) ** (0.165) ** (0.141) ** (0.311) ** OECD (0.228) (0.715) (0.474) (0.290) (0.449) ** (0.327) ** (1.253) Non- OECD (0.126) ** (0.270) ** (0.286) (0.130) ** (0.192) ** (0.166) ** (0.328) ** Notes: Country dummies and year dummies are included in all regressions but not reported. Standard errors are given in parentheses. ξ significantly different from 1 at 10 per cent; significantly different from 1 at 5 per cent; significantly different from 1 at 1 per cent. + significant at 10 per cent; * significant at 5 per cent; ** significant at 1 per cent. 20 Responsiveness regression results by components using the HP-filtered variables are qualitatively similar to those using the first-differenced variables reported in Table 5.

16 452 Fiscal Studies TABLE 6 Responsiveness of expenditure and tax revenue: test of asymmetry, IV All OECD Non-OECD Dep. var.: Exp. Tax Exp. Tax Exp. Tax Negative ΔlGDP it (0.863) (0.938) * ξ (1.705) (1.678) (1.037) (1.113) Positive ΔlGDP it (0.395) ** (0.429) * (0.585) (0.576) (0.491) ** (0.527) * No. of observations 1,982 1, ,428 1,428 R-squared Test of (Negative ΔlGDP it Positive ΔlGDP it ) = 0 Difference p-value Notes: The constant, time trend, country dummies and year dummies are included in all regressions but not reported. Standard errors are given in parentheses. ξ significantly different from 1 at 10 per cent; significantly different from 1 at 5 per cent; significantly different from 1 at 1 per cent. + significant at 10 per cent; * significant at 5 per cent; ** significant at 1 per cent. cyclical than the other parts of expenditure. As shown in Table 5, among the three main components of taxes, income tax revenues are estimated to respond most pro-cyclically. In the literature on the relationship between fiscal policy and the business cycle, an important issue has been whether or not fiscal policy responds symmetrically over business cycles, since this has attendant implications for the issue of fiscal soundness. 21 If the government takes a strong expansionary fiscal position during recessions, while not tightening the fiscal position sufficiently during booms, government debt would grow over business cycles. 22 Table 6 reports whether or not government expenditures and taxes respond differently depending on whether GDP growth rates are positive or negative. 23 We find that government expenditures show a strongly countercyclical response during recessions, as reported in Alesina and Perotti 21 Sorensen and Yosha (2002) report on the asymmetry of state fiscal policy using data for US states. Alesina and Perotti (1995) report that loose fiscal policies are the result of sharp increases in government expenditures, but tight policies are carried out through increases in taxes rather than through reductions in expenditures. Alesina and Perotti (1996) also suggest that fiscal adjustments that rely primarily on spending cuts on transfers and the government wage bill have a better chance of being successful. 22 In particular, if there is political pressure to increase government expenditures or decrease taxes during recessions, there is a tendency for fiscal policy to take a strong expansionary position during recessions. However, it is politically difficult to take a strong contractionary fiscal position during booms by increasing taxes or curtailing expenditures. 23 We simply assume that asymmetric responses occur based on the sign of real GDP growth rates. Negative GDP growth rates occur in 356 (18 per cent) of the 1,982 observations in the total sample.

17 Fiscal policy, business cycles and economic stabilisation 453 (1995), but respond pro-cyclically during booms. The hypothesis that the response of expenditures to business cycles is symmetric can be rejected at the 5 per cent, 16 per cent or 6 per cent p-value, respectively, for the whole sample, the OECD and the non-oecd. Expenditures during booms are estimated to increase only proportionally in the OECD, but progressively in the non-oecd, thus indicating that governments in developing countries have a tendency to spend away additional revenues during booms. On the other hand, expenditures during recessions are estimated to increase more strongly in the OECD, suggesting the active role of expenditures (probably for social security) in stabilising OECD economies during recessions. Our results suggest that fiscal policy takes a strong expansionary position during recessions, but does not tighten the fiscal position during booms, which can cause a serious problem for the long-run sustainability of government debt. VI. Effectiveness regression results In this section, we analyse the effectiveness of fiscal policy in stabilising the economy. As mentioned earlier, we begin by regressing the standard deviation of GDP growth rates on a set of independent variables used in the previous literature. Then we add four additional variables that were not considered in the previous studies military expenditures, oil production, economic fluctuations in neighbouring countries and the responsiveness of fiscal policy to business cycles. In analysing the effectiveness of fiscal policy, our paper examines both revenues and expenditures. Without controlling for our four new variables, column 1 of Table 7 shows that the ratio of total expenditures to GDP is not associated with economic fluctuations. 24 This could happen because we use total expenditure while ignoring the possibility that certain components of expenditure, such as military expenditures, destabilise rather than stabilise the economy. When military expenditures, oil production and the standard deviation of GDP growth rates in other countries are included in the estimation, as shown in column 2, the ratio of expenditures to GDP is estimated to be significantly negatively associated with economic fluctuations, with an estimated coefficient of (standard error: 0.016), significant at 5 per cent. 25 When the estimated responsiveness of taxes to business cycles is included in the specification, government expenditure remains significant. The 24 This insignificant association between total expenditure and economic fluctuations appears to be the reason why previous studies focus on tax revenue. 25 When we add only the ratio of military expenditures to GDP without oil production and economic fluctuations in other countries, the ratio of government expenditures to GDP remains significantly negatively associated with the standard deviation in the GDP growth rate, with an estimated coefficient of (standard error: 0.019), significant at 5 per cent.

18 454 Fiscal Studies TABLE 7 Stability of the economy and government size: all countries, OLS Dependent variable: Standard deviation of GDP growth rate All All All All (1) (2) (3) (4) Government expenditures / GDP (0.022) (0.016) * (0.016) * (0.016) * Military expenditures / GDP (0.043) ** (0.044) ** (0.043) ** Oil production / GDP (0.013) ** (0.014) ** (0.014) ** St. dev. of GDP growth rates of other countries (0.209) (0.234) (0.231) Responsiveness of taxes to business cycles (0.057) Responsiveness of expenditures to business cycles (0.045) Log of GDP (0.173) ** (0.122) ** (0.129) ** (0.123) ** GDP growth rates (0.106) (0.075) (0.076) (0.075) Log of GDP per capita (0.243) (0.170) (0.178) (0.171) Trade volume / GDP (0.007) (0.005) (0.005) (0.005) Constant (3.108) ** (2.289) ** (2.409) ** (2.305) ** No. of observations R-squared Continues opposite responsiveness of taxes to business cycles is estimated for each country by running IV regressions using the specification in column 10 of Table 3. For the whole sample of 94 countries, as shown in column 3 of Table 7, the responsiveness takes the expected negative sign but is not statistically significant. When the estimated responsiveness of expenditures to business cycles is included in the specification instead of the estimated responsiveness of taxes (column 4), we observe that the responsiveness takes a positive sign but is not statistically significant. However, in separate regressions for the OECD and the non-oecd, as shown in columns 5 and 6, the responsiveness of expenditures to business cycles becomes statistically significant in the OECD but is insignificant in the non-oecd. This result is not surprising, since the estimated values for the expenditure responsiveness to business cycles in the OECD countries themselves are mostly significant, while those

19 Fiscal policy, business cycles and economic stabilisation 455 TABLE 7 continued Dependent variable: Standard deviation of GDP growth rate OECD Non-OECD All All (5) (6) (7) (8) Government expenditures / GDP (0.009) ** (0.021) Total tax revenue / GDP (0.025) ** (0.020) * Military expenditures / GDP (0.080) (0.052) ** (0.040) ** Oil production / GDP (0.024) * (0.016) ** (0.014) ** St. dev. of GDP growth rates of other countries (0.164) * (0.276) (0.210) Responsiveness of expenditures to business cycles (0.080) ** (0.051) Log of GDP (0.095) (0.149) * (0.158) ** (0.121) ** GDP growth rates (0.118) (0.086) (0.098) (0.074) Log of GDP per capita (0.223) (0.219) (0.228) (0.174) Trade volume / GDP (0.004) (0.006) (0.006) (0.005) Constant (2.196) * (3.018) ** (2.814) ** (2.253) ** No. of observations R-squared Notes: Standard errors are given in parentheses. + significant at 10 per cent; * significant at 5 per cent; ** significant at 1 per cent. in the non-oecd are not (as shown in Tables A1 and A2 in the appendix). This significance in the responsiveness of fiscal policy indicates that the responsiveness of fiscal policy is as important as the sheer size of the government in explaining economic fluctuations in the OECD. 26 The last two columns of Table 7 use the tax ratio instead of the ratio of expenditures to GDP. Unlike expenditures, the tax ratio is significant even when military expenditures are not controlled for. This is not surprising once we recognise that military expenditures are often financed by non-tax methods, such as government bonds or seigniorage. When military 26 One should note that the number of observations in regressions for the OECD is only 22, thus implying that the degree of freedom in column 5 of Table 7 is only 12. In the most parsimonious specification, with only the responsiveness and the ratio of expenditures to GDP, the responsiveness remains significant, with an estimated coefficient of (t = 2.96), while the ratio of expenditures to GDP becomes insignificant, with an estimated coefficient of (t = 1.38).

20 456 Fiscal Studies expenditures, oil production, economic fluctuations in other countries and the responsiveness of the fiscal position to business cycles are added, the estimated coefficient of the tax ratio, 0.042, has a similar value to that of expenditures, The resulting significant tax effect itself does not necessarily imply that taxes alone are very effective in stabilising the economy. Rather, the tax ratio works as a proxy for the degree to which government responds to business cycles and stabilises the economy, through means such as the development of progressive tax systems and flexible social welfare systems. Among other determinants of economic fluctuations, we find that larger economies in terms of (log of) GDP have smaller economic fluctuations. However, in the OECD, the sheer size of the country turns out to be insignificant. We also find that economic fluctuations in other countries take a positive sign, as predicted, in the OECD. Countries with larger oil production and military expenditures are found to have larger economic fluctuations. Separate regressions for the OECD and the non-oecd in columns 5 and 6 indicate that oil production significantly increases economic fluctuations in the non-oecd while decreasing them in the OECD. This suggests that the non-oecd fails to utilise unstable oil revenues for reducing economic fluctuations, while the OECD succeeds in doing so. Military expenditures significantly increase economic fluctuations, particularly in the non-oecd. Table 8 reports regression results for each component of taxes and expenditures. Row 1 confirms the tendency that OECD countries, with larger government expenditures, experience less variation in GDP growth rates. Row 2 uses the ratio of total expenditures minus military expenditures to show that the expenditures net of military expenditures are negatively associated with economic fluctuations. Rows 3 and 5 show that countries with larger current expenditures and subsidies & transfers tend to have less variation in their GDP growth rate, which is consistent with the finding in Section V that current expenditures and subsidies & transfers move more counter-cyclically than capital expenditures and other expenditures. Rows 7 to 11 report the results for the tax revenue side. We find that income taxes show a negative, but not statistically significant, relationship with economic fluctuations. Commodity taxes are significantly negatively correlated with economic fluctuations in the total sample and in the OECD. Social security taxes in the OECD are also significantly negatively associated with economic fluctuations, suggesting again that social welfare systems may work as stabilisers in the OECD. When we examine the results of both the revenue and expenditure sides in the effectiveness regression, this indicates the importance of expenditures, such as social security and current expenditures, in responding to business cycles and stabilising the economy.