Strategic corporate tax interaction among
developing countries
Derck St¨
abler
June, 2017
Abstract
1
Introduction
Taxes are an ancient concept and basing fiscal reforms on adjacent states can be traced back to the dawn of globalization. In the Histories, Herodotus docu-mented the manner in which the Greeks reviewed the tax systems of the Persians and Egyptians and set taxes dependent on foreign developments. Employing tax rates as strategic instruments has since manifested a perpetual position on every political agenda. Fiscal reaction literature, pioneered by Zodrow and Mieszkowski (1986) and Wilson (1986), has a rich theoretical foundation but empirical literature is relatively scarce. This can largely be attributed to the difficulty of acquiring accurate data. Using corporate tax revenue as a per-centage of GDP, a backward-looking measure, Altshuler and Goodspeed (2002) were one of the first to estimate a fiscal reaction function, where the tax in-teraction variable is the weighted average of other countries’ tax rates. The common theoretical consensus prevails that an increase in tax rate in a certain country reduces the capital employed there and increases capital in all other countries. Following this rationale, theory dictates that countries compete over capital investments by setting tax rates in relation to other countries. In this case, a backward-looking tax measure does not adequately capture the role of taxes in firms’ location decisions. Devereux and Griffith (2003) argue that the impact of tax on location choice is reflected by the post-tax net present value of an investment project. Hence, they have formulated a methodology to compute an effective average tax rate (EATR), using the framework of Jorgenson (1963), by constructing a hypothetical investment project that captures all tax effects. The EATR thus reflects the ratio of the present value of taxes to the present value of profits and takes the following form:
EAT R = R* − R
p/(1 + r) (1)
Where p is the profit rate, r is the real interest rate and R* and R represent the present discounted value (PDV) of the economic rent in the absence- and presence of taxes, respectively. Klemm and Van Parys (2012) have extended the two-period methodology of Devereux and Griffith (2003) to an infinite in-vestment horizon. Their modifications thus include a capital depreciation rate δ and take the following form:
EAT R = R*t− Rt
p/(r + δ) (2)
effective average tax rate under the most generous special regime (EATRs). This
measure, constructed by Abbas and Klemm (2013) for 50 developing countries, also captures the influence of time-bound tax incentives and complicated special regimes.
The majority of the empirical literature studying tax competition on a na-tional level, summarized in table 1, uses the corporate income tax rate, which is the law-based statutory tax rate (STR). Numerous studies have documented the Bertrand like ”race to the bottom” in statutory tax rates, stemming from fierce tax competition that started in the 1980’s. The large gap between statutory tax rates and effective tax rates is prima facie evidence rendering the STR an inadequate tax measure. However, the STR is not sensitive to strong assump-tions related to an investment project and as Chen et al. (2013) specify, it is a highly visible tax measure that signals policy intent. Devereux et al. (2008) were the first to develop an extensive theoretical model of tax competition and empirically show how developed countries compete over statutory and effective tax rates. This development induced various studies to employ this methodol-ogy to shed new light on tax competition between different samples of European and OECD countries. Klemm and Van Parys (2012) have extended the analysis to developing countries and found evidence for strategic interaction in tax holi-days, investment allowances, and the STR in 40 Latin American, African, and Caribbean countries. Moreover, Chen et al. (2013) have documented significant interaction in statutory tax rates for 14 Asian and Pacific countries. Finally, Suzuki (2015) constructed EMTR and EATR measures for 12 Asian countries, following the methodology of Klemm and Van Parys (2012), and also provided significant evidence of tax competition over effective tax rates.
independent variables of notable explanatory value is their lack of discussion of extending the model to include spatially lagged independent variables. This possibly stems from the fact that IV-estimators are unable to estimate models with spatially lagged independent variables, since they use these variables as instruments. Perhaps Allers and Elhorst (2005) were correct in suggesting that the ease with which the IV-estimator can be implemented is the reason for its popularity.
A feature that differed greatly across studies is the selection of spatial weights. The standard binary contiguity matrix is often rejected ex ante on the grounds that tax competition is likely to go beyond first order neighbors. Some studies use uniform or group matrices, where all countries or a specific selection are considered via equal weights. Other studies use inverse distance matrices, either based on euclidean distance or on a population based distance measure to better capture the effective distance used for bilateral trade. A fi-nal class of spatial weights captures the size of the interfi-nal market in terms of population, GDP or FDI. The economic intuition of these weights is straight-forward; weighting by groups implies some countries matter more than others and weighting by distance implies that proximate countries matter more than distant countries, while weighting by internal market size implies that larger countries matter more than small ones.
in this paper are based on a measure of market potential, that among others, captures the aforementioned relevant factors.
A maximum likelihood estimation shows significant strategic interaction among 30 developing countries across 3 continents1. The tax measures ap-plied are the STR, EMTR, EATR and the EATRs, which are constructed and
provided by Abbas and Klemm (2013). This research thus contributes to the scarce literature on tax competition in developing countries. Moreover, this paper distinguishes itself by showing the improved performance of the market potential weight, by providing results that are comparable across continents due to application of a homogeneous approach, by shedding new light on the manner of interaction by extending the model to include spatially lagged inde-pendent variables, and by showing that tax competition is the most severe over the EATRs, which is the most complete tax measure, and has not been used to
study tax competition prior to this research.
The rest of the paper is organized as follows. In section 2, the theoretical considerations are discussed. Section 3 presents the data and the empirical strategy. Section 4 reviews the results and the final section concludes.
2
Theory
Theoretical tax competition literature is predominantly based on a model pro-posed by Zodrow and Mieszkowski (1986) and Wilson (1986) (ZMW). The basic principle of the workhorse ZMW model is a world economy where the investment opportunities of n-countries are defined by an increasing and strictly concave product-of-capital function. Firms are price takers in a market where each coun-try sets a per-unit tax rate. Allowing capital to flow freely between countries, the model yields an equilibrium market outcome. This implies that taxes are strategic complements, such that an increase in a tax rate induces an outflow of capital to other countries. This capital flow affects the country specific marginal product of capital and continues until the countries are back in the equilibrium condition. In game theoretic terms, the countries are assumed to return to a Nash equilibrium in every period. As extensively covered in Keen and Konrad (2012), the ZMW model has shown to be very versatile and is the foundation for numerous fruitful extensions. However, construing a version of the ZMW model to realistically portray strategic interaction in tax rates among developing countries would be a heroic task. A straightforward limitation is that the model satisfies an Inada condition that ensures that the taxes set in equilibrium are strictly positive. Although this is a sensible assumption for developed countries, the sample used in this paper contains effective tax rates that are negative in occasional periods as a result of aggressive special regimes. Moreover, the model needs to capture multiple tax bases and include an infinite planning horizon due to the character of the specified tax rates. Finally, accurately describing firm behavior also proves to be an intricate challenge. Firms can engage in various sorts of tax evading behavior, such as transfer pricing or financial restructuring.
Consequently, this paper does not present such an elaborate model, but rather discusses some theoretical considerations.
In this study, a necessary degree of symmetry is assumed such that countries react in a similar way to tax developments abroad. The ZMW model formally assumes identical countries, perfectly symmetric in production opportunities. Although not identical, all countries is this sample display a similar degree of development, which shares them in the same class of emerging economies. In re-cent years however, China has manifested a strong position in the world economy and is thus not competing over capital investments in a similar fashion to this sample period. Hence, according to the calculations of Suzuki (2015), China’s EATR has shown an upward trend since 2007. This could imply that additional developments may have violated the symmetry assumption such that China no longer competes under the same terms. In sum, countries need to be in a similar stage of development to interact in the same class of competition. An intuitive addition to this assumption is the necessity of some form of geographical prox-imity. The symmetry condition implies that the constituted Nash equilibrium is Pareto inefficient, as all countries would benefit from a slight increase in tax rates. An intriguing point to assess, is whether the strategic complementary nature of tax rates necessarily implies the downward trend found in practice, especially as the development of emerging economies progresses. However, due to the sample period, this is beyond the scope of this research.
3
Data and Methods
3.1
Data
The tax measures used in the paper are constructed by Abbas and Klemm (2013) from the annual worldwide corporate tax guides published by Price Wa-terhouse Coopers and Ernst and Young over 1996–20072. The authors make
standard assumptions regarding the macroeconomic conditions over the sample period3, assume a rate of return of 20 percent and calculate tax incentives with broad applicability, such as the manufacturing or export sector. As specified in Abbas and Klemm (2013), incentives not considered include investment in backward areas, research and development investments or production with high technology content. The special regimes considered are tax holidays (periods of tax exemption), reduced tax rates (temporary or indefinite reductions), ad-ditional investment allowances (allowances on top of depreciation) and special economic zones (a combination of the aforementioned incentives and other non-quantifiable advantages). The original data set covered 50 countries, but by excluding Europe from this analysis and as a result of several missing years or specific tax rates the sample has been reduced to 30 countries in Latin America, Asia and Africa, covering 1996-2007.
Abbas and Klemm (2013) use this data set to describe corporate tax devel-opments in emerging economies and report the following stylized facts:
• The present discounted value of depreciation allowances normalized by the tax rate (PDV depr.)4, which represents the tax base for projects that do not qualify for special regimes, has remained relatively stable over the sample period but has been narrowed (increased) in Africa.
• Statutory tax rates show a small downward trend, except for Europe where the rates have dropped significantly.
• The EATR and EMTR show a small downward trend, except for Africa where the EMTR shows significant decreases due to the narrowing of the tax base.
• Lowering tax rates negatively impacts revenues in the short term, but revenues have held up over the sample period, even better than developed countries. This relationship is weakened in the presence of special regimes and in Africa the effect does not hold up.
• Corporate tax revenue makes up a larger share total tax income in devel-oping countries as opposed to developed countries. This is characterized by a large dependence on revenue from a few large companies.
2Background information regarding the derivation of the effective tax rates is available in
the paper of Devereux and Griffith (2003).
The remaining variables, all of which are novel to fiscal reaction functions, are a measure for government fractionalization, market potential and a state fragility index. Government fractionalization (GovFrac) is defined as the prob-ability that two deputies picked at random from among the parties in govern-ment will be of different parties, by definition ranging from 0 to 1. The variable is taken from the Database of Political Institutions (DPI), which is compiled by the World Bank. As can be seen in the descriptive statistics, this measure differs greatly across countries and years. Considering the existence of presi-dential regimes in this sample, it is not uncommon to find a level of government fractionalization of zero. This implies that the policymaker finds no political resistance regarding decision making from within the government. This allows countries to react to foreign tax developments at a much faster pace than de-veloped countries. To illustrate the contrast, the mean level of government fractionalization across the sample is 0.203, while the mean for the Netherlands across the same years is 0.608.
The state fragility index (SFI), a measure constructed by the Center for Sys-temic Peace, holds a range of 0-25. This index is computed by the sum of points of 8 different measures that score a country’s legitimacy and effectiveness re-garding politics, economics and security. Factors taken into account, among others, are vulnerability to political violence, regime stability and human capi-tal development. Considering that this is a measure of state fragility, an increase in the index would thus imply a more fragile state.
The measure of market potential is constructed by Mayer (2009)5, using the
calculation method of Redding and Venables (2004). The former author refers to this measure as ’real market potential’, where the latter call it ’market ac-cess’. To derive a value for market potential, Mayer (2009) estimates a wage equation, with the main goal of explaining the level of factor incomes by a weighted sum of expenditures of all countries in the world, where the weights are bilateral trade costs. To effectively do so, Mayer (2009) incorporates a broad range of real world rigidities that affect this relation, which include a measure for institutional quality, skill measures, having a common language, having former colonial links, and dummy variables for regional trade agreement membership, currency union membership, GATT/WTO membership, primary resource endowments, a measure of property rights protection and various fea-tures of physical geography. Another important factor taken into account is the size of the internal market. As previously touched upon, some studies use GDP or FDI based weights to capture the potential of the internal market. In the measure for market potential of Davies and Voget (2008), the size of the internal market is reflected by the level of domestic consumption. However, when con-sidering a small country such as Singapore, these measures still seem to paint a rather odd picture. The internal market will be appear to be considerably small, while goods produced in Singapore can easily be sold on the Malaysian market, located on the same island. Mayer (2009) therefore considers regions to be circular, such that all consumers falling within the specified radius are
included in the internal market, again taking into account other rigidities. The market potential data is available for 1996-2003, where the value of market potential used in this paper is composed by taking the natural loga-rithm6 of the mean over these years. Qualitatively speaking, the values evolved very little over this period, which makes sense considering the determinants are relatively time-invariant or very slow in adjustment. The 2015 SWOT analy-sis of Nganga and Maruyama (2015), using analytic hierarchy process methods (AHP), provides a market potential ordering of Sub-Saharan African countries consistent with the ordering found in the data provided by Mayer (2009). This finding supports the validity of the market potential measure used in this pa-per and emphasizes the stability of the relative values of market potential. An interesting feature of the AHP approach is the facility with which the inputs of market potential can be alternatively weighted to reflect importance. A pos-sible extension of this methodology is using the integrated approach of AHP methods and Bayesian analysis, proposed by Mimovi´c et al. (2015). This allows the weighted pairwise decision making process to be evaluated over different expected states of nature. An investor will thus be able to incorporate differ-ent, possible future country developments into one forward-looking measure of market potential. This is especially relevant considering capital investments generally span over multiple years and that developments are inherent to the very nature of developing countries. However, this is beyond the scope of this research.
A low level of state fragility and a high level of market potential are both favorable for firms’ investment decisions. Where market potential captures the economic gain of investing, state fragility portrays the general investment cli-mate. This latter measure is not very relevant for developed counties, as the general investment climate is invariably excellent. For developing countries on the other hand, this measure is essential, as there are numerous countries that show a high (low) level of market potential, while having a low (high) level of state fragility. This accentuates the importance of taking both measures into account.
3.2
Empirical specification
Estimation of a tax reaction function using the spatial lag model takes the following form:
Yit= ρWYit+ Xitβ + µi+ Tr + t (3)
Where Yit is a N × 1 vector of consisting of one observation of a specified tax
rate for country i (i =1,...,N ) at time t (t =1,...,T ), ρ represents the spatial autoregressive coefficient, W is a N × N spatial weight matrix, Xit is a N × K
vector of exogenous explanatory variables with associated coefficient β contained
6Note that using the natural logarithm is consistent with related literature, where this
in a K × 1 vector, µi is vector of spatial fixed effects, Tr denotes a time trend,
and tis a vector of independently and identically distributed error terms, with
zero mean and variance σ2. Country fixed effects are included to control for country-specific, time-invariant variables. Generally, one would be inclined to include time-dummies to control for time-specific, country-invariant variables. However, this is not feasible given the nature of the data, where an evident case in point is the time-invariant measure for market potential. Hence, a time trend is included to account for unobserved factors varying over time as far as possible. The inclusion of country fixed effects is standard practice within the fiscal reaction literature, while opting for a time trend is common, but debated as opposed to omitting a time trend. As touched upon in the theoretical section, this specification does not include a time lagged dependent variable, as a result of the assumption that countries can adjust to foreign tax setting with sufficient agility.
An additional tax reaction function estimated in this paper is the spatial Durbin model (SDM), an extension of the spatial lag model, which takes the following form:
Yit= ρWYit+ Xitβ + µi+ θWXit+ t (4)
Where the added term WXit is a set of spatially lagged independent
ables, with associated coefficient θ. This allows the weighted explanatory vari-ables government fractionalization and state fragility to affect tax rates in spa-tially related countries.
3.3
Spatial weights
The paper employs three different weight specifications7, namely an inverse
dis-tance matrix, a population based inverse disdis-tance matrix and a matrix based on market potential. The first weight is mainly used to replicate other papers, as this euclidean distance based weight is most often employed in relevant litera-ture. The second weight measures the distance between the largest cities of the two countries under consideration, where the inter-city distances are weighted by the share of the city in the overall population of that country. The third weight uses the natural logarithm of the previously described measure of mar-ket potential. The intuition of this weight is that a country takes the location profitability of other countries into account when setting their own tax rate, where a country with a higher level of market potential matters more as it has a larger effect on capital flows when altering tax rates.
Common practice in previous tax competition literature is to row-normalize all weights. For example, Klemm and Van Parys (2012) row-normalize inverse distance weights and Davies and Voget (2008) row-normalize their market po-tential weight. However, as explained by Kelejian and Prucha (2010), row-normalizing a matrix uses a different row-normalizing factor for the elements of each row. To illustrate with an example, in a matrix containing values for market potential all diagonal elements are zero, meaning each row sum is different due to the exclusion of a different country’s market potential. Therefore, there is no re-scaling factor that would lead to a specification that is equivalent to the un-normalized weights matrix. So, in order to preserve the economic interpretation, such that the mutual proportions between the elements of the matrix remain unchanged, all elements per matrix are divided by their largest characteristic root.
4
Results
4.1
Latin America
The estimation results in table 6 show that all values of the spatial autoreggre-sive coefficient ρ are positive and highly significant, where the positive coefficient confirms the common consensus that corporate tax rates are strategic comple-ments. This outcome indicates strong tax competition across all tax measures. A large wedge between statutory and effective tax rates implies that compet-ing over statutory tax rates is merely engagcompet-ing in a beauty contest to attract capital, as it poorly reflects the actual tax rates paid by firms. Contrary to this intuition, column (1) indicates that the statutory tax rate remains an im-portant instrument in tax competition, as it is the most visible tax rate and unambiguously signals policy intent. When comparing column (2) and column (3), one can deduce from the magnitude of the coefficients and their corre-sponding t-statistics that the strategic interaction over the EATR is stronger
and more significant than over the EMTR. This result reflects that countries compete more severely over attracting an investment project than the size of the project. Column (4) shows that competition over the EATRs is decidedly the
strongest and most significant, where the notable increase in R2indicates that this tax measure provides the best fit for the model. This outcome signifies the importance of using a tax measure that incorporates special regimes, as Latin American countries aggressively use these instruments to compete over capital investments. Column (5) shows this regression result is robust to the exclusion of a time trend, although including a time trend is preferred. Finally, column (6) shows estimates without controlling for own market potential to provide co-efficients that can be compared to the spatial Durbin model, where own market potential is excluded as lagged independent variable for obvious reasons.
Column (1) of table 6 shows that government fractionalization and the state fragility index have a significant, positive effect on the statutory tax rate. The former finding is in line with the theoretical prediction that increased govern-ment opposition inhibits policymakers in competitive tax setting and is thus associated with increased tax rates. The latter result is consistent with the no-tion that total tax income in developing countries is more dependent on revenue from corporate tax rates, whereby a more fragile state would be even more re-liant on this source of income and thus opt for a higher STR. Finally, column (4) and (5) indicate that a higher level of market potential negatively affects the EATRs. This coefficient implies that a country with a higher level of market
potential will cause a larger inflow of investment when reducing tax rates and thus be able to set lower tax rates while maintaining revenues. It thus makes sense that this result is significant for the tax rate that most accurately reflects the effect on the inflow of investment, as it best portrays the tax rates actually paid by firms.
The SDM estimates in table 7 show that the effects of the state fragility index and its spatially lagged counterpart are significant at a 1 percent level for all tax rates, where the aforementioned interpretation of this effect holds for all tax measures. Moreover, the spatially lagged index of state fragility shows significant, negative coefficients. This result can be interpreted that an increase in state fragility of spatially related countries increases the relative attractiveness of the country in question, allowing it to slightly reduce tax rates. Equivalently, the significant value of the spatially lagged measure of government fractionalization in column (1) signals other countries’ agility in setting tax rates, where an increase is associated with a slight decrease in tax rates, because it boosts the profitability of setting lower tax rates, due to the increased reaction time of the competitors. A likelihood ratio test (22.84) comparing column (4) of table 7 and column (6) of table 6 statistically justifies the extension to the spatial Durbin model at a 1 percent significance level. The value of ρ in column (4) of table 7 indicates that a 10 percent decrease in the spatially lagged EATRsis
associated with a 7.97 percent decrease of the EATRs, while column (1) indicates
4.2
Asia
The results for Asia are somewhat comparable to the results found for Latin America. In the spatial lag estimates of table 8, the degree of competition increases when comparing the effective tax rates with the statutory tax rate, where competition is the most severe over the EATRs. In contrast to the Latin
American estimates, the R2decreases substantially. Additionally, the coefficient
for ρ in column (2) is remarkably higher, implying that the size of the investment plays a larger role in tax competition among Asian countries. The effect of government fractionalization is not significant, while coefficients of the state fragility index are significant across all tax measures, with the exception of the EATRs. The magnitude of the positive coefficients of the SFI is similar to
the coefficients found in Latin America. Finally, own market potential has a positive, significant effect on the statutory tax rate. Vital to take into account in the interpretation of the effect, is the lower degree of symmetry exhibited in terms of market potential, which is reflected in the standard deviation of 1.562 as opposed to 0.0535 in Latin America. This could indicate that some countries are in a further stage of development in which market potential is sufficiently large that countries no longer find it optimal to heavily compete over capital investments. An illustration of this is the previously mentioned example of China raising effective tax rates after 2007.
The SDM estimates in table 9 are rejected at a 10 percent significance level for all tax measures by likelihood ratio tests in favor of the spatial lag model. Moreover, they show the quizzical result that government fractionalizaton and its spatially lagged counterpart show unexpected signs. Apart from the large difference in market potential, the diversity in political systems further amplifies the asymmetry, where Pakistan and Indonesia even changed systems during the sample period. This is a major difference with the Latin American countries, that all have a presidential system. Finally, the Asian sample has a lower degree of geographical proximity and does not cover data on possibly relevant countries. In sum, the Asian sample is less symmetric and is therefore less in line with the theoretical setup of the model specification. This limitation implies that this sample could suffer from biased estimates due to omitted countries or omitted political variables that better describe tax setting behavior. To conclude, the spatial lag model does an adequate job in quantifying strategic interaction in corporate tax rates, but a model specification using other independent variables could possibly prove the spatial Durbin model to be of increased explanatory value.
4.3
Africa
stan-dard deviation. Additionally, Egypt and Morocco are notable outliers in location and other specified descriptives. As a result, the strategic interaction among tax rates is very low compared to other continents, and the coefficients of govern-ment fractionalization shows opposite signs for the EATR and the EATRs. All
things considered, this sample does not provide a representative depiction of reality. In spite of this conclusion, this result does highlight the importance of assembling an accurate data set in accordance with theoretical assumptions. This is currently not possible because of the low availability of African data. An additional issue is that the EATRshad dropped to 0 for most Sub-Saharan
countries due to the high levels of state fragility. This entails that countries offer other non-quantifiable advantages, such as lighter regulation. Capturing the firm advantages and government benefits in a data set is very difficult, if not impossible.
4.4
Comparison with results of previous research
Table 11 shows regression outcomes for Latin America based on an inverse dis-tance spatial weights matrix. The results may be compared with the estimates found by Klemm and Van Parys (2012) in their study of tax competition in Latin America. When the authors consider uniform weight matrices within the same region and include a time trend, they find a value of ρ of 0.437 for statutory tax rates, 0.603 for tax holidays and 0.801 for investment allowances, significant at a 1 percent, 10 percent and 1 percent level, respectively. Recall that tax holidays and investment allowances are both incorporated in the EATRs, which
shows a coefficient of 0.743. However, when Klemm and Van Parys (2012) opt for an inverse distance matrix, ρ reduces to 0.173 for statutory tax rates, 0.312 for tax holidays and investment allowances no longer show a significant effect. This could indicate that this set of countries takes all other Latin American countries into account when setting tax rates. Another possible explanation is that row-normalizing the inverse distance matrix misspecifies the model. This issue does not apply to the uniform weight matrix, because row-normalizing this matrix or dividing all elements by their largest characteristic roots yields the same result.
When repeating this analysis using a population based weight matrix, cov-ered in table 12, the results show higher values for R2and higher log-likelihood
continents.
The results in table 8 are relatively comparable to other studies. Suzuki (2015) finds coefficients for ρ to be 0.477 for the EMTR and 0.458 for the EATR, for the period 1991-2012. The estimates found in this paper are slightly higher, which could be the result of the usage of an inverse distance matrix, which has previously shown to provide lower coefficients; the row-normalization of this matrix which could affect the coefficients; the inclusion of Japan which has the highest EMTR and EATR by a large margin of all countries over the whole sample period and thus inherently violates the symmetry assumption; or finally, due to the decreased accuracy of the used GMM-estimator which has produced a coefficient larger than 1 in this paper. Chen et al. (2013) found evidence of competition in statutory tax rates using a uniform weight matrix over the period 1980-2007. Their ρ coefficient range is 0.679-0.857, which differed greatly in terms of significance, but were generally higher than the effect found in this paper. This is presumably the case due to the longer time period that also captures the fierce tax competition in the 1980’s. Moreover, their sample exclusively contains the most developed countries, including Australia and New Zealand, which show a much larger downward trend in statutory tax rates.
5
Conclusion
5.1
Conclusion
This paper provides statistically significant evidence of tax competition across all tax measures, where applying the market potential based spatial weight consistently provides the best model specification. This is compelling evidence that countries strongly take their competitors’ location profitability into account when setting their own tax rates. The STR and EATRs have proven to be of
most explanatory value, where the former is the most visible tax rate and the latter best describes the true tax rates paid by firms. Competition over the EATRsis the most severe, which illustrates the importance of including the
ef-fects of special regimes in the tax measure. In Latin America, the spatial Durbin model is a significant extension of the spatial lag model. For this subsample, this confirms the notion that, when taking market potential into account, govern-ment fractionalization and state fragility act as a signalling device of tax setting abilities to other countries. An increase in state fragility raises own tax rates, while an increase in state fragility of spatially related countries decreases own tax rates. A similar effect takes place regarding government fractionalization for statutory tax rates. Although the spatial lag model is preferred in Asia, a model specification using other independent variables could also prove the spa-tial Durbin model to be of increased explanatory value. The analysis of African countries shows evidence of tax competition, but the effects are unusually small and considering the abundance of econometric issues, is unlikely to reflect a realistic estimate.
in corporate tax rates among developing countries. Regardless of what motiva-tion the reader finds most persuasive, setting corporate tax rates in relamotiva-tion to other countries remains a highly debated policy topic in all areas of the world. It is intriguing to consider whether the strategic complementary nature of tax rates documented in this paper will cause tax competition to persist, further reducing the EATRs, or whether it can cause a reversal in the downward trend,
as the development of emerging economies progresses. This paper has illus-trated the manner and the degree of strategic interaction and contributes to the literature in the following ways. It provides estimates of a unique data set comparable across different continents and tax measures, including the novel EATRs. It emphasizes the importance of a theory-motivated weighting scheme
and shows a significant extension to the spatial Durbin model. There are sev-eral interesting avenues for future research. The market potential based spatial weights can be improved by including an expectational component, as touched upon in section 3. The spatial Durbin approach opens the door for further re-search to uncover other signals relevant for tax setting. Moreover, future studies can examine the role of effective tax rates for firms engaging in research and development or production with a high technology content, considering the role of increasing importance this sector is establishing in developing countries.
References
Abbas, A. and Klemm, A. (2013). A partial race to the bottom: corporate tax developments in emerging and developing economies. International Tax and Public Finance, 20:596–617.
Allers, M. and Elhorst, P. (2005). Tax mimicking and yardstick competition among local governments in the netherlands. International Tax and Public Finance, 12:493–513.
Altshuler, R. and Goodspeed, T. (2002). Follow the leader? evidence on euro-pean and u.s. tax competition. Working Paper.
Altshuler, R. and Goodspeed, T. (2015). Follow the leader? evidence on euro-pean and u.s. tax competition. Public Finance Review, 43(4):485–504. Cassette, A. and Paty, S. (2008). Tax competition among eastern and western
european countries: With whom do countries compete? Economics Systems, 22:307–325.
Chen, Y., Regis, P., and Huang, W. (2013). Do asia and pacific countries compete in corporate tax rates? Journal of the Asia Pacific Economy, 19:25– 51.
Davies, R. and Voget, J. (2008). Tax competition in an expanding european union.
Devereux, M. and Griffith, R. (2003). Evaluating tax policy for location deci-sions. International Tax and Public Finance, 10:107–126.
Devereux, M., Lockwood, B., and Redoano, M. (2008). Do countries compete over corporate tax rates. Journal of Public Economics, 92:1210–1235. Jorgenson, D. (1963). Capital theory and investment behavior. American
Eco-nomic Review, 53:247–259.
Keen, M. and Konrad, K. (2012). International tax competition and coordina-tion. Working Paper.
Kelejian, H. and Prucha, I. (2010). Specification and estimation of spatial autoregressive models with autoregressive and heteroskedastic disturbances. Journal of Econometrics, 157:53–67.
Klemm, A. and Van Parys, S. (2012). Empirical evidence on the effects of tax incentives. International Tax and Public Finance, 19:393–422.
Mayer, T. (2009). Market potential and development. CEPII, 24:1–27.
Mimovi´c, P., Stankovi´c, J., and Mili´c, V. J. (2015). Decision-making under uncertainty – the integrated approach of the ahp and bayesian analysis. Eco-nomic Research-Ekonomska Istraˇzivanja, 28:868–878.
Nganga, P. S. and Maruyama, Y. (2015). Market attractiveness evaluation of sub-saharan africa, applying swot analysis and ahp methods. Journal of Economics and Economic Education Research, 16:1–18.
Overesch, M. and Rincke, J. (2009). Competition from low-wage countries and the decline of corporate tax rates: Evidence from european integration. The World Economy.
Overesch, M. and Rincke, J. (2011). What drives corporate tax rates down? a reassessment of globalization, tax competition, and dynamic adjustment to shocks. The Scandanavia Journal of Economics, 113(3):579–602.
Overesch, M., Rincke, J., and Heinemann, F. (2010). Rate-cutting tax reforms and corporate tax competition in europe. Economics Politics, 22:498–518. Redding, S. and Venables, A. (2004). Economic geography and international
inequality. Journal of Internation Economics, 62:53–82.
Ruiz, F. and Gerard, M. (2008). Is there evidence of strategic corporate tax interaction among eu countries? Working Paper.
Wilson, J. (1986). A theory of interregional tax competition. Journal of Urban Economics, 19:296–315.
Zodrow, G. and Mieszkowski, P. (1986). Pigou, tiebout, property taxation, and the underprovision of local public goods. Journal of Urban Economics, 19:356–370.
Table 2: Descriptive statistics for Latin America
Variable mean std. dev. min max obs source
ST R 0.289 0.071 0 0.35 120 Abbas and Klemm (2013) EM T R 0.264 0.071 0 .359 120 Abbas and Klemm (2013) EAT R 0.273 0.064 2.31e-08 .365 120 Abbas and Klemm (2013) EAT Rs 0.225 0.124 -0.019 .365 120 Abbas and Klemm (2013)
PDV depr. 0.174 0.053 0 0.339 120 Abbas and Klemm (2013) Market potential 18.489 .0535 17.389 19.361 120 CEPII
GovFrac 0.250 0.289 0 0.818 120 DPI
SFI 6.983 4.448 1 15 120 SFI
Table 3: Descriptive statistics for Asia
Variable mean std. dev. min max obs source
ST R 0.302 0.051 0.18 0.45 132 Abbas and Klemm (2013) EM T R 0.198 0.089 0.054 0.349 132 Abbas and Klemm (2013) EAT R 0.242 0.058 0.113 0.360 132 Abbas and Klemm (2013) EAT Rs 0.093 0.127 -0.113 0.438 132 Abbas and Klemm (2013)
PDV depr. 0.221 0.059 0.095 0.380 132 Abbas and Klemm (2013) Market potential 20.163 1.562 17.950 23.007 132 CEPII
GovFrac 0.202 0.254 0 .721 132 DPI
SFI 8.606 5.189 0 18 132 SFI
Table 4: Descriptive statistics for Africa
Variable mean std. dev. min max obs source
ST R 0.297 0.087 0.15 0.42 108 Abbas and Klemm (2013) EM T R 0.093 0.174 -0.362 0.344 108 Abbas and Klemm (2013) EAT R 0.194 0.086 0.057 0.362 108 Abbas and Klemm (2013) EAT Rs 0.086 0.102 -0.081 0.322 108 Abbas and Klemm (2013)
PDV depr. 0.256 0.089 0.105 0.388 108 Abbas and Klemm (2013) Market potential 17.50 1.490 15.64 19.50 108 CEPII
GovFrac 0.157 0.275 0 .774 108 DPI
Table 5: Overview of countries Latin America Asia Africa
Argentina China Botswana Brazil Indonesia Egypt
Chile Korea Kenya Columbia Malaysia Morocco Costa Rica Pakistan Mauritius
Equador Philippines Namibia Panama Singapore Uganda Paraguay Sri Lanka South-Africa
Peru Taiwan Zambia Uruguay Thailand
Vietnam
Table 6: Estimation results of the spatial lag model for Latin America, using a market potential weight
Variable (1) (2) (3) (4) (5) (6) W*ST R 0.648*** (9.29) W*EM T R 0.469*** (4.37) W*EAT R 0.596*** (7.29) W*EAT Rs 0.841*** 0.831*** 0.823*** (25.55) (24.09) (22.24) Market potential 0.004 0.016 0.008 -0.036** -0.036** (0.33) (1.12) (0.66) (-2.38) (-2.35) GovFrac 0.041** 0.003 0.02 0.031 0.036 0.009 (1.92) (0.11) (1.08) (1.30) (1.50) (0.41) SFI 0.003** 0.001 0.002 -0.000 0.000 0.002 (1.87) (0.85) (1.35) (-0.02) (0.06) (1.45) Time trend 0.001 0.000 0.001 -0.003 No No (0.62) (0.21) (0.50) (1.60) Log-likelihood 172.45 162.00 179.20 151.51 150.20 146.96 R2 0.39 0.23 0.33 0.75 0.74 0.73 Observations 120 120 120 120 120 120
Table 7: Estimation results of the spatial Durbin model for Latin America, using a market potential weight
Variable (1) (2) (3) (4) W*ST R 0.638*** (8.49) W*EM T R 0.463*** (4.23) W*EAT R 0.575*** (6.56) W*EAT Rs 0.797*** (18.61) GovFrac 0.057*** 0.008 0.030 -0.015 (2.60) (0.32) (1.41) (-0.58) SFI 0.007*** 0.004** 0.005*** 0.009*** (4.19) (2.17) (3.18) (4.61) W*GovFrac -0.064** -0.009 -0.033 0.015 (-2.06) (-0.24) (-1.09) (0.41) W*SFI -0.008*** -0.007*** -0.007*** -0.012*** (-4.05) (-2.64) (-3.38) (-4.75) Log-likelihood 182.43 165.06 185.33 158.38 R2 0.48 0.27 0.39 0.77 Observations 120 120 120 120 T-statistics are in parentheses
Table 9: Estimation results of the spatial Durbin model for Asia, using a market potential weight Variable (1) (2) (3) (4) W*ST R 0.655*** (9.18) W*EM T R 0.655*** (9.18) W*EAT R 0.613*** (7.69) W*EAT Rs 0.659*** (9.34) GovFrac -0.032*** -0.017 -0.019 0.037 (-2.84) (-0.56) (-1.05) (0.86) SFI 0.009*** 0.005*** 0.006*** -0.001 (14.93) (2.79) (6.10) (-0.60) W*GovFrac 0.046*** -0.034 -0.011 0.035 (2.59) (-0.71) (-0.39) (0.50) W*SFI -0.006*** -0.003 -0.004*** 0.001 (-6.51) (-1.36) (-2.83) (0.21) Log-likelihood 298.93 166.94 236.94 119.56 R2 0.77 0.45 0.54 0.45 Observations 132 132 132 132 T-statistics are in parentheses
Table 10: Estimation results of the spatial lag model for Africa, using a market potential weight Variable (1) (2) (3) (4) W*ST R 0.004*** (8.67) W*EM T R 0.006*** (28.40) W*EAT R 0.006*** (25.49) W*EAT Rs 0.006*** (28.32) Market potential 0.0039 0.008 0.003 0.017*** (0.88) (0.89) (0.57) (3.29) GovFrac 0.022 0.068 0.046* -0.043* (0.89) (1.316) (1.769) (-1.657) SFI 0.006*** -0.001 0.002 0.001 (3.88) (-0.52) (1.55) (0.94) Time trend -0.004* -0.010*** -0.005** -0.002 (-1.88) (-2.70) (-2.47) (-1.19) Log-likelihood 145.74 65.73 140.11 135.63 R2 0.53 0.51 0.49 0.61 Observations 108 108 108 108 T-statistics are in parentheses
Table 11: Estimation of the spatial lag model for Latin America, using an inverse distance weight Variable (1) (2) (3) (4) W*ST R 0.483*** (5.28) W*EM T R 0.346*** (3.15) W*EAT R 0.443*** (4.52) W*EAT Rs 0.743*** (14.60) Market potential 0.006 0.022 0.013 -0.044 (0.41) (1.47) (0.96) (2.39) GovFrac 0.0359 -0.003 0.017 0.044 (1.541) (-0.107) (0.771) (1.50) SFI 0.002 0.001 0.002 -0.002 (1.53) (0.73) (1.10) (-0.86) Time trend 0.001 0.000 0.001 -0.002 (0.54) (0.16) (0.43) (-0.99) Log-likelihood 162.21 158.84 172.27 127.01 R2 0.26 0.19 0.23 0.62 Observations 120 120 120 120 T-statistics are in parentheses
Table 12: Estimation results of the spatial lag model for Latin America, using a population based inverse distance weight
Variable (1) (2) (3) (4) W*ST R 0.561*** (6.85) W*EM T R 0.398*** (3.69) W*EAT R 0.508*** (5.55) W*EAT Rs 0.791*** (18.52) Market potential 0.006 0.021 0.011 -0.037*** (0.47) (1.33) (0.88) (-2.25) GovFrac 0.034 -0.003 0.016 0.031 (1.59) (-0.12) (0.77) (1.17) SFI 0.003*** 0.001 0.002 -0.001 (1.82) (0.79) (1.28) (10.41) Time trend 0.001 0.000 0.001 -0.003 0.47) (0.13) (0.35) (-1.63) Log-likelihood 168.36 161.19 176.69 139.73 R2 0.34 0.22 0.31 0.70 Observations 120 120 120 120 T-statistics are in parentheses
