Tax policy and economic growth: does it really matter?

Abstract

We study the relationship between per capita GDP, overall tax revenue and tax composition. We find that there is some evidence of a negative and statistically significant relationship between tax revenue and economic growth, while there are no robust relationships between revenue-neutral tax shifts and economic growth. The results hold in different samples of OECD countries for different time periods. We also identify different relationships between the short run and long run. Overall, our results cast doubts on the potential growth enhancing effects of a shift from direct to indirect taxation, with paramount consequences on tax policy.

1 Introduction

In a time of tight public finance conditions, often because of the high level of sovereign debts, there is a large and expanding debate on the effects of taxation on the level and growth of GDP. Scholars and policy makers alike—for a given level of fiscal pressure—are interested in whether the tax mix, i.e., the distribution of overall tax revenue across different tax sources, has a separate influence on the growth performance of a country. The ‘common wisdom’ on the relationship between tax mix and growth is that a shift from direct to indirect taxation is associated with higher GDP growth. The main international organizations that deal with economic matters, i.e., the IMF and the OECD, claim that high taxes on labor are detrimental to economic growth, and a shift from direct to indirect taxes has a growth enhancing effect. The European Commission adopted this view back to 1993. Since then, the tax shift has remained at the center of the European Commission’s agenda and it has pervaded its policy recommendations to member states: ‘tax should be designed to be more growth-friendly, for instance by shifting the tax burden away from labor on to tax bases linked to consumption, property and combating pollution’ (European Commission 2013).

This paper provides new empirical evidence on the relationship between taxation and tax shift and growth, which challenges this ‘common wisdom.’ While we are able to identify in some cases a negative and significant relationship between the overall tax revenue and growth, we do not find any significant and robust relationships between various types of revenue-neutral tax shift and growth.

On the theoretical side, scholars suggest to cut tax rates and broaden tax bases in order to reduce economic distortions while keeping a constant level of fiscal pressure. Consumption taxes allow to precisely follow these prescriptions: consumption is a broader tax base than labor income because households’ consumption choices can be financed by using other sources of income than wages. Moreover, reducing labor taxes should enhance economic growth by increasing labor supply and demand, and by stimulating investment. Thus, the recommendations on the tax shift from direct to indirect taxation are essentially based on the higher economic efficiency of consumption taxes with respect to income taxes.¹ Similarly, a shift from income to wealth taxation could lead to positive efficiency gains, given that income is more elastic to taxes than wealth. Also, the traditional public finance argument posits that taxing income-producing assets is equivalent to taxing the ‘normal’ income arising from those assets, so that extra income due to extra effort is effectively taxed at a zero rate, thus enhancing that effort (Einaudi 1924).

Even if this is not explicitly discussed within the traditional Public Finance theory, growth is the dynamic counterpart of efficiency at a macroeconomic level and has been for decades the main focus and goal of government action.²

What do the data say about the relationships between the tax mix and economic efficiency (growth)? The most influential empirical contribution on this issue is a paper by Arnold et al. (2011). For a sample of 21 OECD countries over the 1971–2004 period, they show that tax revenue is negatively and significantly associated with per capita GDP, while a shift from direct to indirect taxation is positively and significantly correlated with per capita GDP. More precisely, they identify a ‘tax and growth ranking’ according to which the most harmful taxes for economic growth are: corporate taxes, personal income taxes, consumption taxes and finally property taxes. This ranking has been adopted by international organizations, such as IMF, OECD and the European Commission, to offer policy recommendations to countries. This analysis has been recently confirmed by Acosta-Ormaechea and Yoo (2012) using an enlarged sample of countries (69, including non-OECD countries) and an extended time period (up to 2009).³ Focusing on a smaller set of countries, Xing (2012) challenges the robustness of these results, suggesting that they are less general than what is claimed by previous studies, as they depend on the set of countries and the time period under investigation.⁴ Indeed, some European countries have partly aligned with the Commission’s prescriptions (Estonia, Italy, Spain), while others have recently introduced the tax shift objective in their reforms programs (Belgium, Finland, France, Germany, Lithuania, the Netherlands). However, the evidence on the implementation of this tax shift is still limited (see Mathé et al. 2015).

Fig. 1

Trend in tax composition and GDP—34 OECD countries in the time period 1995–2014. Notes The figure shows the relationship between the rate of growth of the ratio between income taxes and the sum of consumption and income taxes (horizontal axis) and GDP growth (vertical axis). Our elaboration on OECD data

Figure 1 shows the relationship between the tax shift, computed as the rate of growth of the ratio between income taxes and the sum of consumption and income taxes (horizontal axis) and GDP growth (vertical axis) in 34 OECD countries in the time period 1995–2014. Figure 1 suggests that—in line with the conclusions by Mathé et al. (2015)—there is no unconditional evidence in favor of the idea that a shift from direct to indirect taxation has had a growth enhancing effect in the last 50 years. In fact, to date those policy recommendations on efficient tax shifts have been only very partially implemented by developed countries.⁵

The limited correlation between the tax shift from direct to indirect taxes and economic growth may be to some extent explained by a low elasticity of labor supply: if lower taxes on labor income do not have a strong positive impact on labor supply, we expect to observe a rather small—if not null—effect on economic growth. Indeed, a large recent literature has shown that the income tax elasticities of labor supply are quite small (see Saez et al. 2012 for a review). More specifically, Jäntti et al. (2015) have discussed the discrepancy between more traditional macroeconometric studies, which provide evidence of high labor supply elasticities, and the more recent microeconometric studies, which instead find low labor supply elasticities. In their exercise, where macrovariables are aggregated on the basis of microdata for several countries, there is no strong support for macrolevel elasticities to be higher than microlevel (low) elasticities.

Another argument that could help reconcile theory and evidence is the cost of transition, which is often neglected in the literature on the efficiency effects of tax shifts. In fact, tax reforms are always costly, both in terms of political constraints and in terms of their administrative burden. The so-called status quo bias (see Castanheira et al. 2012) is a crucial well-studied political obstacle to tax reforms, even when they are theoretically efficient, independently from equity concerns. Moreover, on the administrative side, complex tax systems are not easily modified, and, when they are, the costs of transition may be substantial (see Winer et al. 2015) and thus negatively affect economic growth. Taking this into account, even tax reforms which are expected to be theoretically efficient may turn out not to be implemented and, if they are, they could show no significant relationships with economic growth.

This paper estimates these relationships between tax burden, tax mix and per capita GDP, controlling for the standard growth determinants (fixed capital formation, human capital, population growth). Following the previous literature, we adopt the panel version of an error correction model (ECM) specification, i.e., the pooled mean group estimator (Pesaran et al. 1999). We face a trade-off in the choice of our sample: data for the largest set of 34 OECD countries are only available for the 1995–2014 period, while data for a narrower sample of 23 OECD countries are available for a longer time period, i.e., from 1971 to 2014. We therefore show results for both choices.

First, focusing on a sample of 23 OECD countries in the time period from 1971 to 2014, we find a significant long-run negative correlation between the revenue-neutral shift from indirect to direct taxes and GDP growth. However, in the short run, this correlation becomes positive even when we separately look at consumption and property taxes. On the other hand, the long-run correlation between tax revenue and GDP growth is only significant in one specification over five. Different from all previous studies, we integrate the standard ECM models with more conservative estimates of the standard errors, i.e., we cluster them at the country level (Bertrand et al. 2004; Cameron et al. 2008). When doing so, none of these long-run correlations is significant.

Second, we consider an extended set of countries in the available years: the full—and balanced—sample of 34 OECD countries during the 1995–2014 period. In this case, neither the correlation of per capita GDP with tax revenue nor the ones with the revenue-neutral tax shifts are found to be statistically significant. This holds mainly for the long run, with and without clustered standard errors. The only noticeable exceptions to this lack of significant results are that—when not clustering the standard errors—a shift from consumption to corporate taxation of income is positively and significantly correlated with GDP per capita in the long run and that a shift from income to property taxes is negatively associated with per capita GDP in the short run.

To understand why our results differ from those of previous contributions, we also replicate the analysis by Arnold et al. (2011) for the same 21 countries and the same time period 1971–2004, with and without clustering the standard errors. While we can confirm some of their results with ‘unclustered’ standard errors, those same results are no longer significant when clustering the standard errors. This lack of robustness should suggest more caution in deducing policy recommendations from the relationships emphasized by Arnold et al. (2011), on which in fact the OECD view is mainly based.

Finally, we restrict ourselves to the sample of OECD countries belonging to the Eurozone, during the 1995–2014 period. Results are similar to what obtained for the sample of 34 OECD countries in the same investigated time period.⁶

Overall, this paper challenges the validity of the relationship between tax mix and growth when a significant number of years after the 2008 economic and financial crisis and different sets of countries are considered. The lack of robust findings seems to be particularly relevant from a policy perspective, since robust findings should be the basis for sound advice on growth enhancing fiscal policies, as the one provided by international organizations such as the IMF and the OECD. It is also the case that the shift from direct to indirect taxes, which has been so far implemented only by few European countries, is now attracting the attention of non-European countries as well. A prominent example is the USA, where President Trump has recently proposed a large reduction in personal and corporate income taxes, with the Republican majority in Congress initially pushing for a ‘Border Adjustment Tax’ to (partially) achieve revenue neutrality.

The paper is organized as follows: Sect. 2 presents the data and the empirical strategy, while Sect. 3 displays our results. Section 4 concludes.

2 Data and empirical strategy

We collect economic and fiscal data on OECD countries for the period from 1971 to 2014.⁷

Economic variables include GDP at constant 2010 prices per head of population aged 15–64 years, the investment rate (proxied by the ratio of gross fixed capital formation to real GDP), the stock of human capital (proxied by the average number of years of schooling of the population aged between 15 and 64 years) and the growth rate of the working age population. All variables come from OECD but the human capital variable, which is taken from the Barro and Lee (2013) dataset, recently updated up to 2016.

Fiscal data refer to total government (central and local) and are taken from OECD revenue statistics. We organize these data as follows: income taxes include categories 1000 (taxes on income, profits and capital gains), 2000 (social security contributions) and 3000 (taxes on payroll and workforce); personal income taxes include categories 1100 (taxes on income, profits and capital gains of individuals), 2000 and 3000.⁸ Consumption and property taxes include categories 4000 (taxes on property), 5000 (taxes on good and services) and 6000 (other taxes); consumption taxes in turn include categories 5000 and 6000. Recurrent taxes on immovable property include categories 4100 (recurrent taxes on immovable property) and 4600 (other recurrent taxes on property), while other property taxes include categories from 4200 to 4500 (recurrent taxes on net wealth; estate inheritance and gift taxes; taxes on financial and capital transactions; non-recurrent taxes on property). The overall tax burden is computed as the ratio of total tax revenue to GDP, while all the other fiscal variables are expressed as shares over total tax revenue. This allows to estimate the relationship between revenue-neutral tax shifts and GDP per capita. All data are available at yearly frequency.

Table 1 Descriptive statistics for the sample of 23 OECD countries in the period 1971–2014

Table 2 Descriptive statistics for the sample of 34 OECD countries in the period 1995–2014

Table 3 Descriptive statistics for the sample of 21 OECD countries in the period 1971–2004

Table 4 Descriptive statistics for the sample of Eurozone countries in the period 1995–2014

Tables 1, 2, 3 and 4 show the summary statistics for the four different samples of countries and years that we consider in our analysis.

The methodology used to estimate the relationship between the distribution of overall tax revenue across different tax sources, and the economic performance of a country is based on macrogrowth regressions,⁹ where, under the hypothesis of a constant return to scale technology, aggregate production at time t is assumed to be a Cobb-Douglas type, which is a function of physical capital (K), human capital (H) and labor (L) as follows:

$$\begin{aligned} Y(t) = K(t)^{\alpha }H(t)^{\beta }\left[ A(t)L(t)\right] ^{(1-\alpha -\beta )} \end{aligned}$$

(1)

The parameters $\alpha$ and $\beta$ are the partial elasticity of output with respect to physical and human capital, while the variable A(t) incorporates technological progress, the growth-related effects of institutions and public policies (see Cellini 1997) and is assumed to be labor augmenting. Equation (1) can thus be easily rewritten as an expression for the steady-state output in intensive form as follows:

$$\begin{aligned} \varDelta \hbox {log} Y_{i,t}= & {} a_{o,i} +\phi _{i}log Y_{i,t-1}+ a_{1,i}log K_{i,t}+ a_{2,i}log H_{i,t}+ a_{3,i}n_{i,t}\nonumber \\&+\sum a_{j,i}{} \textit{FISCAL}_{i,t}+ F_{i}(t)\nonumber \\&+\,b_{1,i}\varDelta log K_{i,t}+b_{2,i}\varDelta log H_{i,t}+b_{3,i}\varDelta n_{i,t} \nonumber \\&+ \sum b_{j,i}\varDelta \textit{FISCAL}_{i,t}+\epsilon _{i,t} \end{aligned}$$

(2)

where, for each country i and in each year t, Y is real GDP per head of population, K is the investment rate, H is the stock of human capital, and n is the growth rate of the working age population. $\textit{FISCAL}$ is a vector of tax variables: the overall tax burden over GDP, and the shares over total tax revenue of (i) income taxes, (ii) personal income taxes, (iii) corporate income taxes, (iv) consumption and property taxes (taken together), (v) consumption taxes, (vii) property taxes, (viii) recurrent taxes on immovable property and (ix) other property taxes. The parameters $a_0$ are a set of country fixed effects, and $F_{i}(t)$ is a set of 5-year dummies, whose coefficients are country specific.

The terms with coefficients starting with ‘a’ and ‘b’ capture long-term and short-term dynamics, respectively, while $\epsilon _{i,t}$ is the error term.¹⁰

Equation (2) is estimated by means of a panel error correction model (ECM), where estimation results are computed with the pooled mean group (PMG) estimator proposed by Pesaran et al. (1999), which allows short-run coefficients, the speed of adjustment and error variances to differ across countries, but imposes coefficients on long-run slopes that are common across countries. Cross-sectional dependence is also taken into account, given that the countries analyzed are characterized by a high degree of economic integration.¹¹

Notice that—for each sample of countries and years under consideration—Eq. (2) is estimated twice: first with conventional standard errors and then with standard errors that are clustered at the country level. The use of clustered standard errors allows us not to inflate the precision of our estimates because of potential within-group correlation of the error terms (Bertrand et al. 2004). Notice that the choice of clustering the standard errors at the country level is not the most conservative one, as one could argue that within-cluster correlation of the error terms might also arise at a wider geographical level, e.g., at the continental level.¹²

We perform several sets of estimates. First, we estimate Eq. (2) for 23 OECD countries in the time period 1971–2014. Second, we enlarge the set of countries up to all the 34 OECD countries. In this case, since for some countries OECD fiscal data are only available from 1995 we obtain a balanced panel by focusing on the 1995–2014 period. Third, we estimate Eq. (2) for 21 OECD countries in the 1971–2004 period to replicate Arnold et al. (2011). Finally, we perform our analysis on a reduced set of more homogeneous countries, i.e., OECD countries that are current members of the Eurozone.¹³

3 Empirical results

Results are presented in Tables 5, 6, 7, 8, 9 and 10. Odd-numbered tables show results with conventional standard errors, while in even-numbered tables we cluster the standard errors at the country level.¹⁴ Since the PMG estimator allows for country-specific short-run dynamics, there is a difference in the estimated conventional and clustered standard errors only for the long-run coefficients. Hence, in odd-numbered tables we display both long-run and short-run results, while in even-numbered tables we only show the long-run ones.

3.1 23 OECD countries, from 1971 to 2014

Tables 5 and 6 show the estimation results of Eq. (2) for 23 OECD countries in the time period 1971–2014. In Table 5 only in one specification out of five the negative long-run correlation between the tax burden and per capita GDP is found to be (mildly) significant. We can confirm the positive long-run correlation of per capita GDP with a shift from income taxation to consumption and property taxation. However, when disentangling personal and corporate income taxes, only a revenue-neutral shift to the former is negatively associated with GDP. Moving to the short run, the tax burden is negatively and significantly associated with GDP in all specifications, while correlations with the revenue-neutral tax shifts run contrary to what is found in the long run: a shift from consumption and property taxation to income taxation is positively and significantly correlated with GDP, while a shift from direct to indirect taxes (both consumption and property) is negatively and significantly correlated with GDP. In Table 6 we find that, when clustering standard errors at the country level, none of the long-run correlations is significant.¹⁵

Table 5 Taxation and GDP per capita: 23 OECD countries, 1971–2014

Table 6 Taxation and GDP per capita: 23 OECD countries, 1971–2014, clustered standard errors

Table 7 Taxation and GDP per capita: all 34 OECD countries, 1995–2014

Table 8 Taxation and GDP per capita: 34 OECD countries, 1995–2014, clustered standard errors

Table 9 Taxation and GDP per capita: 21 OECD countries, 1971–2004 (Arnold et al. 2011 sample)

Table 10 Taxation and GDP per capita: 21 OECD countries, 1971–2004, clustered standard errors (Arnold et al. 2011 sample)

3.2 34 OECD countries, from 1995 to 2014

Tables 7 and 8 show results for an expanded dataset composed of all the 34 OECD countries, for the period 1995–2014. We find that the tax burden is not significantly associated with economic growth in the long run, and that it is negatively associated with economic growth in the short run only in two out of five specifications. A shift from income to consumption and property taxes is not significantly related to per capita GDP in the long run, while the association between these two variables in the short run turns out to be negative and statistically significant, as already found in Table 5. Interestingly, a revenue-neutral shift from consumption and property taxes to corporate taxation of income is positively and significantly correlated with GDP per capita in the long run. Table 8 shows our results when we cluster the standard errors at the country level. In this case, we do not find any significant relationship.¹⁶ As a robustness check, to exclude that the lack of significant relationships is driven by an outlier country, we rerun our baseline regression on the set of 34 OECD countries by dropping one country at a time. Results, reported in Tables A1–A5 in the Supplementary material, are reassuring against this potential concern: the relationship between tax revenue and economic growth and the relationship between a revenue-neutral tax shift and economic growth are never statistically significant.

3.3 A critical assessment of Arnold et al. (2011)

Table 9 replicates the estimates contained in Arnold et al. (2011) for the same sample of 21 countries and the same time period 1971–2004.

In line with what they found, in the long run we identify a negative and significant long-run correlation between the tax burden and per capita GDP, and a positive and significant long-run correlation between a revenue-neutral shift from income to consumption and property taxes and growth.¹⁷ Different from Arnold et al. (2011)—who find a negative and significant correlation—in our replication exercise a shift to corporate income taxation is not significantly correlated with per capita GDP. Interestingly, in the short run the tax burden is negatively and significantly correlated with per capita GDP only in three specifications out of five, while the only significant tax shift is the one toward taxes on immovable property, which in fact is negatively correlated with per capita GDP. In Table 10, the only result robust to clustering is the negative relationship between overall tax revenue and per capita GDP in the long run.¹⁸

3.4 Eurozone countries, from 1995 to 2014

Finally, since the tax shift has been mainly implemented in European countries, Tables 11 and 12 are focused on the OECD countries that are current members of the Eurozone, again for the period 1995–2014. In Table 11, we find a negative and significant correlation between the overall tax burden and per capita GDP only in one specification out of five in the long run and in two specifications out of five in the short run. Regarding the tax mix, the shift from consumption and property taxation to income is not statistically significant in the long run: in fact, it is a combination between a negative and significant correlation of per capita GDP with personal income taxation and a positive and significant correlation with corporate taxation. In the short run, the only significant correlation we detect is the negative one between a shift from income to property taxes and GDP, that is driven by immovable property taxes, similar to what is found in Tables 5 and 7.¹⁹ In Table 12, when we cluster the standard errors at the country level, the only significant result is that a revenue-neutral shift away from personal income taxes is related to higher economic growth.²⁰

Table 11 Taxation and GDP per capita: Eurozone countries, 1995–2014

Table 12 Taxation and GDP per capita: Eurozone countries, 1995–2014, clustered standard errors

3.5 Addressing the role of business cycle effects

Finally, we try to address potential endogeneity concerns regarding tax variables by purging them from business cycle effects (Arnold 2008, paragraphs 40 and 41). More precisely, we regress the tax variables against the first two lags of the output gap, including those lags both in their linear and in their quadratic forms. The obtained residuals are then used in the second-stage regressions instead of the tax variables themselves. To note, when doing so our sample sizes get reduced with respect to those of our main regressions (Tables 5–12), since the output gap data is only available starting from 1987. Thus, we rerun our main regressions on these restricted samples, in order to have the appropriate benchmark for comparisons with these second-stage regressions. Overall, we find that long-run results of our main regressions are comparable to those obtained in these second-stage regressions only when clustering the standard errors. On the other hand, short-run results are broadly comparable. Thus, this exercise casts some further doubts on the relevance of the tax shift from direct to indirect taxes as a policy to enhance economic growth in the long run.²¹

4 Conclusions

This paper develops a data-driven analysis of the relationship between tax burden, revenue-neutral tax shifts and economic growth. In the long run, the overall tax pressure is negatively and statistically significantly related to per capita GDP, and the revenue-neutral tax shift from direct to indirect taxes is positively and statistically significantly related to economic growth only when we consider the sample of 21 OECD countries for the period 1971–2004 as in Arnold et al. (2011). However, when we adopt more cautious estimates of the standard errors, i.e., when we cluster them at the country level, only the result on the negative relationship between the overall tax burden and per capita GDP still holds. When we extend the time period under investigation, consider the full sample of OECD countries or focus only on current members of the Eurozone neither the long-run relationship between tax burden and economic growth nor the relationships between revenue-neutral tax shifts and economic growth are statistically significant. Overall, our analysis provides a comprehensive assessment of these relationships which is robust to the choice of the investigated sample of countries and years. Our study also suggests that the inconsistent findings that appear in previous empirical contributions in the macroliterature may be due to the different—mainly arbitrary—choices about the sets of countries and years that different authors decide to include in their analyses. International organizations such as the IMF and the OECD, if they want to make sound policy recommendations—e.g., to move from direct to indirect taxation—need more robust analyses than the ones provided by Arnold et al. (2011).

More fundamentally, one could also argue that cross-country regressions based on macrodata, as in Arnold et al. (2011), even when using fixed effects may not be sufficiently informative to infer policy implications. In fact, the identification strategy behind this approach could be plagued by first-order consistency issues such as omitted variable bias and reverse causality; moreover, the specification itself is based on strong theoretical assumptions (e.g., a Cobb-Douglas production function); thus, there are cautious limits to what we can learn from the results of these types of analyses, over and above the lack of robust results we have shown here. Alternative empirical approaches are instead based on quasi-experimental within-country microeconometric data (Saez et al. 2012) and might represent a more robust guidance to tax policy. As already mentioned, these studies find a small labor supply elasticity to income tax, thus suggesting that shifting taxes from income to other tax bases might not have significant growth enhancing effects, in a direction that is consistent with our robust and cautious macroeconometric approach.

Still, to the extent that one trusts this type of macroeconometric exercise, we find some interesting differences in the relationships between tax variables and growth between the short run and the long run. In the short run, the negative and statistically significant relationship between tax burden and economic growth seems to be more robust. At the same time, regardless of the sample of countries and the time period under investigation, we find a negative and significant correlation between a shift from income taxes to recurrent taxes on immovable property and economic growth, which runs contrary to the standard ‘OECD view’ and the recommendations by both IMF and the European Commission. Short-run correlations might be very important when assessing the political feasibility of economic reforms, because voters are more likely to care about short—rather than long—run ‘effects’ (Castanheira et al. 2012). This is particularly true in times of economic instability, as reflected by the growing consensus obtained by populist parties which base their platforms on short-term protectionist policies and on ‘pandering’ to what people want (Canes-Wrone et al. 2001). From a theoretical viewpoint, the differences we find in the short-run and long-run correlations between economic growth and taxation might be connected with the differential salience of taxes along different time horizons (Chetty et al. 2009). This is a promising avenue for future research.