A multitude of studies have estimated the effect of taxes on country-level economic growth. With so many studies working over a limited number of datasets, one would think that something resembling a consensus would have arisen. Not so. In their review of tax reforms, Kneller and Misch (2017, page 165) conclude ‘… at least the direction of the short-run and long-run growth effects can be predicted with a reasonable degree of certainty, but there is disagreement with respect to the magnitude.’

One approach that policymakers can take given this state of affairs is to select one or a few studies that, in their judgment, provide ‘best estimates’. This requires that one has confidence in the reliability of those estimates. However, without confirmation from replication studies (still very rare in economics), this can be a risky strategy. As noted by Larry Summers in the context of the infamous Reinhart and Rogoff data error, ‘no important policy conclusion should ever be based on a single statistical result.’

Our approach is to pool the estimates in the tax effects literature. We synthesize 979 estimates from 49 studies published between 1993 and 2020 that estimated the effect of taxes on economic growth in OECD countries. Our headline result is that a 10% increase in taxes is associated with a decrease in annual GDP growth of approximately -0.2% when bundled as part of a ‘TaxNegative’ tax-spending-surplus combination. The same tax increase is associated with an increase in annual GDP growth of approximately 0.2% when part of a ‘TaxPositive’ fiscal policy package. In the space below, we define what we mean by ‘TaxNegative’ and ‘TaxPositive’ fiscal policies and provide some detail about how we arrived at these estimates.

**Problem: Estimates of Tax Effects from Different Studies Measure Different Things**

Studies estimating the effect of taxes on economic growth typically estimate a regression equation having the following general specification:

*(1) Economic Growth = α _{0 }+ α_{1}TaxRate + Control Variables + error,*

where *α _{1 }*is the effect of taxes on economic growth and the tax rate is commonly measured by the ratio of taxes to income. The main problem with trying to compare estimates from different studies is that they generally measure different things. No change in taxes takes place in isolation. It is accompanied by a change in one or more other fiscal categories: other revenues, expenditures, and/or the budget surplus. This gives rise to the following identity:

*(2) (Taxes/Income) + (OtherRevenues/Income) – (Expenditures/Income) = (Surplus/Income).*

To avoid perfect multicollinearity, one or more of these variables must be omitted from a regression specification. However, the interpretation of *α _{1}* in Equation (1) will differ depending on which government budget constraint variables are omitted from the control variables.

If (*Expenditures/Income*) is omitted but other revenues and surplus are included, then *α _{1 }*measures the effect of increases in taxes “holding constant” other revenues and surplus. As a result, the estimated tax effect incorporates the growth effects of associated increases in expenditures. Alternatively, if (

*OtherRevenues/Income*) is omitted and expenditures and surplus are held constant, then the estimated tax effect picks up changes in the composition of revenues, mixing in the effects of a greater reliance on taxes accompanied by a lesser reliance on other revenue sources. Accordingly, estimates from two different studies using the exact same data could estimate very different tax effects, even opposite-signed estimates, depending on the specification of the regression equation.

**How We Synthesize Tax Estimates from Different Studies**

Studies differ greatly with respect to the tax-spending-surplus specifications they use in their regressions. There is no such thing as a standard empirical specification. This greatly complicates the task of synthesizing estimated tax effects from different studies. We address this problem by combining two taxonomies of fiscal policies; one from Kneller, Bleaney and Gemmell (1999), and one from Gemmell, Kneller and Sanz (2009).

‘TaxNegative’ fiscal policies are defined as policies where an increase in the tax rate is predicted to produce negative economic growth. Gemmell, Kneller and Sanz (2009) identify two corresponding tax-spending-surplus combinations. The first is an increase in distortionary taxes to fund unproductive expenditures. The second is an increase in distortionary taxes accompanied by a decrease in non-distortionary taxes.

‘TaxPositive’ fiscal policies are policies where an increase in the tax rate is predicted to produce positive economic growth. According to Gemmell, Kneller and Sanz (2009), there are three tax-spending-surplus combinations that belong in this category: (i) An increase in non-distortionary taxes to fund productive expenditures, (ii) An increase in non-distortionary taxes accompanied by a decrease in distortionary taxes, and (iii) an increase in non-distortionary taxes to decrease the deficit/increase the surplus. Every other combination of taxes-spending-surplus has ‘ambiguous’ growth effects because the individual components conflict and it is unclear which are stronger.

For each of the 979 estimated tax effects in our sample, we identify both the operative tax type and the omitted fiscal categories implied by the respective regression specification. Tax types and expenditures are classified as non-distortionary/distortionary, productive/unproductive, or other according to Kneller, Bleaney and Gemmell (1999). We then use the taxonomy from Gemmell, Kneller and Sanz (2009) to code the respective tax-spending-surplus combinations as belonging to one of the three fiscal policy categories. Finally, we convert all estimated tax effects to represent the effect on annual GDP growth of a one-percentage point increase in the tax rate.

**How We Obtain Our Estimates**

Meta-analysis provides several methods for synthesizing estimated tax effects across studies. Essentially, these consist of using Weighted Least Squares to estimate variations of the following regression specification:

*Estimated Tax Effect _{i} = β_{0} + β_{1}TaxNegative_{i} + β_{2}TaxPositive_{i} + Control Variables_{i} + error_{i},*

where the dependent variable is the *ith* estimated tax effect gleaned from the economic growth literature. *(β _{0}* +

*β*is the estimated growth effect of a one-percentage point increase in taxes associated with a

_{1})*TaxNegative*fiscal policy; with

*(β*+

_{0}*β*estimating the corresponding effect for a one-percentage point tax increase related to a

_{2})*TaxPositive*fiscal policy. We use four sets of weights to alternatively account for different precisions in the respective tax estimates, unexplained heterogeneity across tax estimates, and the number of estimates per study. This results in four separate estimates of

*(β*+

_{0}*β*and

_{1})*(β*+

_{0}*β*.

_{2})We also employ a procedure for identifying and correcting ‘publication bias’. Publication bias occurs when researchers and journals prefer not to report estimates that are statistically insignificant or have the ‘wrong sign’. This can cause the estimates reported in the literature to overstate the true effect. Our subsequent analysis finds statistically significant evidence of negative publication bias, meaning that negative tax estimates are favoured in publication. However, the size of the bias is relatively small and economically unimportant.

Our main results are reported in Panel C of Table 4 in our paper. We can use those estimates to do back-of-the-envelope calculations to determine their respective economic significance. We start by noting that tax burden typically ranges between 25-45% for the countries and time period covered in our sample. A 10% increase in the tax burden would thus approximately equal a 3.5 percentage point increase in taxes. If it were associated with a *TaxNegative *fiscal policy, our analysis indicates that an increase of this size would lower annual GDP growth anywhere from -0.070 to -0.392 percent, on average. If it were associated with a *TaxPositive* fiscal policy, we estimate it would increase annual GDP growth between 0.126 and 0.325 percent, on average. This is the basis of our headline result that a 10% increase in taxes is associated with a decrease in annual GDP growth of approximately -0.2% when bundled as part of a ‘TaxNegative’ tax-spending-surplus combination; and an increase of approximately 0.2% when part of a ‘TaxPositive’ fiscal policy package. Compared to an average annual growth rate of 2.50 percent for the countries in our sample, these tax effects can be considered modest in size, but not insubstantial.

This article is based on Alinaghi, N. & Reed, W. R. (2020). *Taxes and Economic Growth in OECD Countries: A Meta-Analysis*. Department of Economics and Finance, College of Business and Economics, University of Canterbury.

For those who were curious like me, the paper itself gives good examples of distortionary/non-distorionary and productive/unproductive:

“Taxes on labor and capital are commonly classified as “distortionary”, while taxes on consumption are considered relatively“non-distortionary”. Likewise with expenditures: expenditures on health and public infrastructure are generally regarded as relatively “productive”, while income transfers such as welfare and social security are generally regarded as relatively “unproductive”. ”

Based on this, a negative GDP effect would be expected to occur from the increase in taxes on individuals (capital gains or normal income), which was then used to boost welfare programs. Alternatively, using an increase on GST to fund infrastructure investments would be expected to increase GDP. Interesting given the regressive nature of consumption taxes and progressive nature of welfare and income taxes, this paper seems to indicate that directly helping the poor hurts the overall economy.