Tax Incentives and Foreign Direct Investment Claudio Agostini and Soraphol Tulayasathien, Ph.D.1
July 2006
Abstract This paper presents an evidence of strategic tax setting behavior among jurisdictions and its effect on the effectiveness of tax incentive programs. While investors seem to consider tax incentives when deciding where to invest, the strategic interaction among jurisdictions in setting tax benefits makes it difficult for a jurisdiction to attract investors using only tax incentives. The paper argues that Thailand should emphasize on investment facilitation services in conjunction with the tax incentive packages currently in place. In 2005, the Board of Investment of Thailand approved tax incentive packages to 1,254 projects which amounted to 571.2 billion baht in investment. The packages are sizable: if these projects yield 10% returns and are subjected to 30% tax, the packages cost 17.1 billion baht in foregone tax revenue, roughly 6.9% of corporate income tax revenue.
1
Claudio Agostini: Alberto Hurtado, Chile. Soraphol Tulayasathien: Office of the Public Sector
Development Commission, Government of Thailand (
[email protected]).
If the tax incentive packages attract incremental investments and the investments, as intended, cause positive spillovers in term of job creation, technology transfer and activities in related industries, we can evaluate the effectiveness of the packages by performing a cost-benefit analysis. That is to compare the foregone tax revenue and the spillover benefits. However, if the tax incentive packages do not bring in incremental investment (e.g. the case where investors would have invested anyway with or without the incentives), the tax incentive packages are ineffective. The important question is then whether or not incremental investments are influenced by tax incentives. Empirical studies of such tax effects show mixed results. Several articles in the literature have found that taxes have no influence 2
on investment location at all; some few articles have found a consistent and robust negative effect, and a few others have found a positive effect.3 To make the story even richer, since investors shop around for the most attractive tax packages, the process of tax setting involves strategic interaction among countries. This strategic tax setting, also known as tax competition, 4 can influence the effectiveness of tax incentive packages. For example, if Thailand offers to increase tax benefits and Malaysia follows, the incremental investments
2
This is consistent to what many have pointed out that in choosing investment location taxes
are secondary concerns. That is, investors choose strategic locations first before bargaining for the most attractive tax packages.
3
Wasylenko (1997) reports a summary of econometric results of tax effects on business location.
Out of 6 studies that estimate an overall tax elasticity for investment in manufacturing, 3 studies report statistically significant tax elasticities, ranging from -1.1 to 0.54. Out of 7 studies that estimate a business tax elasticity (corporate income tax or property tax) for investment in manufacturing, 6 studies report statistically significant tax elasticities, ranging from -0.36 to -0.10.
4
See Case et al. (1993), Besley and Case (1995) and Rork (2000) for discussion and evidence of
the interaction between different jurisdictions’ policies.
into Thailand may not be as large. Therefore, neglecting the tax competition in the estimation of the tax effects may bias the tax elasticity estimated.
5
In this paper, we present results of econometric estimates of the tax effects taking into account the endogeneity problem of the tax variable that may arise from strategic tax setting among state governments. The paper utilizes econometric techniques from the Discrete Choice6 and Instrumental Variable to correct for this biased. 7
The Estimating Equation
In deciding where to invest, the investor compares several different characteristics of each location, one of which is the corporate tax rate. Let π*isj be the profit of investor i from source country s conditional on investing in location j. The investor will choose to invest in location j if and only if location j’s investment yields the greatest profits among all investment alternatives. That is, π*is = max{π*isn; n} where n is the set of investment locations, including the outside investment alternative. Adding up investment choices by all investors yields the aggregated demand function. Following the discrete-choice framework, we define the conditional profit function of investor i from source country s for investing in location j as: π*isj = -α tj + Xj β + ξj + εisj
5
In this paper, we will argue that the tax endogeneity will dampen the overall estimated tax
effect; consequently neglecting tax setting behavior may lead to insignificant estimates of the tax effect (even if the true effects are negative).
6 7
See McFadden (1974) and Nevo (2000) for discussion on the Discrete Choice literature. Parts of the analytic sections that follow are taken from a working paper of Claudio Agostini
and the author. For detailed analysis and results, please refer to Agostini and Tulayasathien (2001).
where X are observed location characteristics, ξ are unobserved location characteristics, t is the corporate tax rate and ε is the investor’s characteristics. In this model, each investors is assumed to invest in one location, the one that gives him the highest profits. An investor is defined as a vector of locationspecific shocks, (εisn). We can define the set of investor’s characteristics that lead to the choice of location j as: A sj = {(εisn)| π*isj > π*isn ∀n} Then, for a given set of parameters, we can predict the FDI share of each location, as a function of location characteristics, tax rates and unknown parameters. The assumption of the Logit demand model developed by McFadden is that the investor’s characteristics is i.i.d. and is distributed according to Type I Extreme Value Distribution. The assumption simplify the calculation of the FDI share of each location as follows: Ssj = exp(-α tj + Xj β + ξj) / [1 + ∑ n exp(-α tn + Xn β + ξn)] With the logistic assumption, the equation to be estimated is: ln(Ssj) – ln(Ss0) = -α tj + Xj β + ξj The tax elasticities of the FDI shares Ssj with respect to location k tax rate are: ηsj
= - α tj (1 – Ssj)
if j=k
= α tk Skj
otherwise (cross-location elasticity)
From the above elasticities formulae, we can see that the main problem of the Logit assumption is related with cross-location tax elasticities. The Logit
model restricts investors to substitute investment from one location towards other locations in proportion to FDI shares. This does not constitute a big problem for us because we are interested in the own elasticity for each sate and the mean elasticity for the whole sample. Tax Endogeneity and Instrumental Variable Technique The tax effect on FDI can be decomposed into two components: the direct effects and the indirect effects through observed (X) and unobserved (ξ) characteristics. dFDI/dt = ∂FDI/∂t + ∂FDI/∂X dX/dt+ ∂FDI/∂ξ dξ/dt Since in the estimation of the tax effects we control for observed characteristics on the right-hand-side of the estimating equation, the bias come from unobserved factors. If dξ/dt is different from zero, then estimate of the tax effects on FDI without considering the unobserved characteristics is biased. The direction of the bias will depend on the signs of ∂FDI/∂ξ and dξ/dt. If ξ is the tax abatements offered by other location, and there is tax competition, we would expect dξ/dt to be negative. In this example, ∂FDI/∂ξ will also be negative. Consequently, ignoring tax competition would bias the estimated tax elasticity upward (towards zero). One way to deal with the tax endogeneity is to use instrumental variables for the tax rate in the estimating equation. Natural candidates to be used for this purpose are the exogenous variables that affect tax setting but do not appear in the estimating equation: for example, other types of tax rates (e.g. personal and sales tax rates), total government expenditures and determinants of the budget process. In choosing the appropriate instrumental variables, we have to consider how exogenous these potential instruments are with respect to investors’ decisions.
In the case of state personal tax rates, we think that jurisdictions that want to have higher taxes will have higher tax rates for both corporate and personal incomes. If jurisdictions set their corporate tax rates much lower than the personal tax rate, some individuals may avoid high personal taxes by incorporating themselves. For this reason, we expect corporate and personal tax rates to be positively correlated. If personal taxes do not influence foreign investors’ decisions (that is, it is not in the estimating equation), then this might be a good instrumental variable. In our view, we believe that personal and corporate tax rates are set simultaneously, and therefore personal tax rates are not truly exogenous. In the case of total government expenditures, higher government expenditures are associated with higher taxes. However, since some of the expenditures might be considered by investors when deciding where to invest, expenditures may not be a suitable instrument. If we look at budget practices across U.S. states, we can observe a good variation among them. Particularly, there is a considerable variation in states’ revenue and expenditure limitations. These limitations constrain the annual growth of revenue or expenditures to either a fixed rate or a rate that is based on one or more of the following indices: inflation rate, population growth, growth of personal income and ratio of revenue to personal income. In addition, some states have a statutory or constitutional requirement to balance the budget. The two most common requirements are that the governor must submit a balanced budget and that the legislature must pass a balanced budget. We think that these budget practices are good instruments for the tax rate. The main concern is that the observed correlation between budget rules and fiscal outcomes reflects just a correlation of these two variables with an omitted third one, specifically citizens’ preferences for fiscal outcomes. We have two arguments to counter this concern. First, the empirical literature on state budgeting and fiscal policy support the hypothesis that budget rules matter and
many of these studies have controlled for some measure of voter preferences.
8
The second argument is that most of the constitutional balanced budget requirements were enacted a long time ago, some of them even in the nineteenth century. We do not think that rules enacted many years ago reflect the preferences of current state residents. Therefore, in this paper, we will use a set of dummy variables that reflect these different budget limitations as instrumental variables to correct for the tax endogeneity. Data FDI Data The Bureau of Economic Analysis (BEA) at the U.S. Department of Commerce publishes a benchmark survey of FDI in the U.S. roughly every five years. The surveys report FDI data by state and by source country for major 9
investing countries in the U.S. This paper uses information on the value of property, plant and equipment (PPE) owned by affiliates of foreign multinational companies in the U.S. for manufacturing industries by state and by source country for the years 1974, 1980, 1987, 1992 and 1997.
8
10
As the outside
See Crain and Miller (1990), Alt and Lowry (1994), Poterba (1994 and 1995), Bohn and Inman
(1996) and Reuben (1997)
9
These countries are Australia, Canada, France, Germany, Japan, the Netherlands, Switzerland
and the United Kingdom
10
The 1974 benchmark survey has information only for four countries (Canada, Japan,
Switzerland and the United Kingdom) and the 1980 survey is missing Australia. Therefore, we have 1,750 observations. Among these observations, 195 (11.1%) are suppressed by the BEA for confidentiality reasons, and 30 (1.7%) are reported as being less than $500,000 but no specific amount is provided. We deal with these data problems in three ways: first the two types of observations are excluded from the sample; second, we estimate an interval regression using the 30 observations that we know represent PPE of less $500,000; and third, we assume that all
investment alternative for investors, we use the total amount of investment in fixed capital in their own countries.
11
Tax Data With regard to measure of tax burden, the most common approaches are to use either the statutory tax rate or the ratio of revenue collected to income. The latter approach has the advantage that it captures both the nominal rate and the tax base, and in that sense, it might be a better measure of the tax burden. However, in this paper, we use the top marginal statutory corporate tax rate. The main reason is that for newly locating firms or expanding businesses, the statutory tax rate is appropriate as it captures the marginal rate of taxation on the investment or the user’s cost. Table 1 shows summary statistics of the data. The mean corporate tax rate is 6.6% with a range from 0% to 12.65%, which should provide good tax variation across sates in our estimation of the tax effects. Still, we have to take into account the fact that some states allow corporations to deduct their federal tax liabilities. In these cases, we multiply the state corporate tax rate by (1 – federal top statutory corporate tax rate). This variable is called Etaxrate in table 1 and is the one we use in our analyses. We also need to consider the source country tax regime. For this purpose, we use a dummy variable that distinguishes foreign-tax-credit and exemption regimes. Specifically, the variable Credit Country equals 1 if the source country grants its residences credits for foreign taxes paid and equals 0 otherwise.
missing observations represent PPE of less $500,000. The results are very similar across the three cases.
11
The outside investment alternative is very important in the estimation of the tax effect.
Without outside investment alternatives, a uniform increase in corporate tax rates of all jurisdictions would not result in changes in total investment.
Another important aspect of the states’ tax system is the use of different formulas to allocate the national profits of a firm among states. Given that we use property, plant and equipment variable, we will consider the weight given by each state to the property factor in the apportionment formula. The variable apportionment then, is the weight assigned to property in the apportionment formula of each state. Table 2 shows summary statistics for corporate tax rates by year. The mean and the median tax rate have been increasing over time, and the standard deviation has been decreasing since 1987. This might be a symptom of some degree of collusion among state governments in setting taxes. Table 3 shows the changes in state corporate tax rates over time. From 1974 to 1997, 29 states had a higher corporate tax rate in 1997 as compared to the one they had in 1974, 7 have a lower one, and 14 states have exactly the same one. If we split the period in two, we can see that more states increased their tax rates between 1974 and 1987 than between 1987 and 1997. The opposite is true in terms of decreases in tax rate. This table may suggest that the degree of collusion was higher between 1974 and 1987 and that the degree of competition has increased after 1987. Observable Characteristics There are some public goods provided by the state that help business activities. As a measure of this type of public goods, we use the variable Road per Area, which is calculated as the total number of miles of roads in each state divided by its area (measured in squared miles.) The variable Real Wage and Real Price of Energy capture the prices of other inputs in the production process. Real wages are measured in dollars per hour, deflated by CPI, and the real price of energy is an index published by the Bureau of Labor and Statistics, also deflated by CPI. Instrumental Variables
The variable Expenditure Limit and Revenue Limit are dummies that equal 1 if the state has a constitutional or statutory provision that constraints the annual growth of expenditures and revenues respectively and 0 otherwise. The variable Legislature equals 1 if the state legislature must pass a balanced budget and 0 otherwise. Result and Discussion Table 4 shows the Tobit regression of the tax equation. Model (1) uses the statutory corporate tax rate directly as a variable to capture tax effects, and models (2) to (5) use the set of budget limits dummies described before as 12
instrumental variables to control for the endogeneity of the tax rate.
Model (1)
shows a negative relationship between FDI share and state’s corporate tax rate, but the coefficient is not significant, and the elasticities implied are quite low. The tax elasticity, calculated at the mean and the median values, is -0.19 and 0.20 respectively. Model (2) has the same specification of Model (1) but corrects for the possible tax endogeneity using instrumental variables. Now, the coefficient of the state’s tax rates is -16.911. This implies a tax elasticity of -1.06 at the mean values and -1.11 at the median values. This result shows and upward bias that occurs when the tax effects on investment location are estimated without taking into account the tax endogeneity. The coefficients for the population and road per area variables are both positive, significant and with the expected signs. Total population in one state captures the level of business activity; therefore, it should have a positive impact on the amount of FDI the state receives. The number of miles of roads per area
12
For Model (1), standard errors were calculated using the Huber-White estimator. For Models
(2) to (5), in which we use instrumental variables, we calculated the standard errors using bootstrapping with 1,000 repetitions.
measures the level of a public good provided by the state that helps business activity; thus it should also have a positive impact on FDI. The real price of energy captures the cost of non-labor inputs to production; therefore the negative and significant coefficient is as expected. The coefficient for the wage variable is negative but not significant. A negative sign would be expected as higher wages increase the production costs and therefore decrease firms’ profits. One possible explanation for the non-significant finding is that the real wage variable we use is also capturing higher productivity of the labor force. Model (3) adds to Model (2) the credit country dummy, alone and interacted with the tax rate. The inclusion of these two variables attempts to control for the tax regime of the source country and its effects on the effective state tax rate faced by investors from these two different tax regimes. The coefficient for the tax rate is -18.08, which implies a tax elasticity of -1.14 at the mean values and -1.19 at the median values for investment form countries with exemption regime. As expected, the coefficient for the credit country dummy interacted with the tax rate is positive, showing that investors from countries using credit regimes are less sensitive to corporate income tax rates. Model (4) adds to Model (2) the property weight in the apportionment formula interacted with the tax rate. The coefficient of this variable is -22.97 and is statistically significant. The implication of this result is that is a sate increases the weight of the property factor in its apportionment formula, the share of total FDI received by the state would decrease around half percent on average, ceteris paribus. The coefficient of the corporate tax rate is statistically significant but its size is half as before (-8.73 as compared to -16.911 in Model (2)). These results show that the property factor in the apportionment formula has an important impact on the effective state corporate income tax rate that investors face and, therefore, on the fraction of FDI that state receive. However, in terms of the corporate income tax elasticity, controlling for the apportionment system does not change the results obtained with only the top statutory tax rate is used.
Model (5) controls for both the apportionment formula and the tax regime of the source country. The corporate tax rate coefficient is statistically significant and is higher than that of Model (4). However, the tax coefficient is half the size of those in Models (2) and (3). The effect of the property weight and the apportionment formula interacted with the tax rate is again negative and significant. The coefficient of the credit country dummy variable interacted with the state tax rate, as in Model (3), has the expected sign, but it is not significant. Table 5 shows the results of the first-stage regression corresponding to Model (2). The coefficients on the variable Revenue Limit and Expenditure Limit are both negative and statistically significant. This implies that state governments that face constraints to the annual growth of its expenditures and revenues set lower taxes than states that do not have these limits. The coefficient on the Legislature variable is positive and significant. This implies that states whose legislatures must enact a balanced budget have higher taxes. The F-test of joint significance of the three variables we used as instruments is 97.18, which strongly rejects the hypothesis of non-significance. Furthermore, every instrument by itself is highly significant. Conclusion In sum, our results show that without controlling for the tax endogeneity, the estimated tax effect is not significant. In contrast, when tax endogeneity is considered, the estimated tax effect is negative and statistically significant. Hence, we can conclude that tax incentives significantly influence investment decisions. However, with tax competition, the effectiveness of tax incentives is compromised. What is the policy implication of these findings? The simple answer is that while investors seem to consider tax incentives when deciding where to invest, the strategic interaction among jurisdictions in setting tax benefits make it
13
difficult for a jurisdiction to attract investors using only tax incentives.
Thailand
can choose whether she wants to follow the price leader strategy as many tax haven countries do or the differentiation leader strategy by focusing in investment facilitation.
14
In our opinion, we think both strategies can be used in
conjunction and their effectiveness must be considered altogether.
13
It should be noted that a caution must be taken when extending the findings using U.S. states
data to the context of international tax competition. Further empirical study on international tax competition is needed.
14
The Board of Investment of Thailand has initiated many investment facilitation program. For
example, the recently launched the BOI Unit for Industrial Linkage Development (BUILD) provides a channel of communication with the manufacturing sectors in Thailand.
Reference Agostini, C. and Tulayasathien, S. “Tax Effects on Investment Location: Evidence from Foreign Direct Investment in the U.S. States,” Office of Tax Policy Working Paper No. 2001-19, University of Michigan Business School. Alt, J. and Lowry, R. (1994), “Divided Government, Fiscal Institutions, and Budget Deficits: Evidence from the States,” American Political Science
Review 88, 811-828. Besley, T. and Case, A. (1995), “Incumbent Behavior: Vote Seeking, Tax-Setting and Yardstick Competition,” American Economic Review 85, 25-45. Bohn, H. and Inman, R. (1996), “Balanced-Budget Rules and Public Deficits: Evidence from the U.S. States,” Carnegie-Rochester Conference Series on
Public Policy 45, 13-76. Case, A., Hines, Jr., J. and Rosen, H. (1993), “Budget Spillovers and Fiscal Policy Interdependence: Evidence from the States,” Journal of Public Economics 52, 285-307. Crain, W. and Miller, III., J. (1990), “Budget Process and Spending Growth,”
William and Mary Law Review 31, 1021-1046. McFadden, D. (1974), “Conditional Logit Analysis of Qualitative Choice Behavior,” In Zarembka, P. (ed.), Frontier in Econometrics, New York: New York Academic Press. Nevo, A. (2000), “A Practitioner’s Guide to Estimation Random Coefficients Logit Models of Demand,” Journal of Economics and Management Strategy 9, 513-548. Poterba, J. (1994), “State Responses to Fiscal Crises: The Effects of Budgetary Institutions and Politics,” Journal of Political Economy 102, 799-821. ---. (1995), “Balanced Budget Rules and Fiscal Policy: Evidence form the States,”
National Tax Journal 48, 329-337. Rueben, K. (1997), “Tax Limitations and Government Growth: The Effect of State Tax and Expenditure Limits on State and Local Government,”
Ph.D. Dissertation, MIT Department of Economics. Rork, J. (2000), “Neighboring Influences in State Tax Determination,” Proceedings of the 93rd Annual conference on Taxation, Santa Fe, NM, National Tax Association, 291-294. Wasylenko, M. (1997), “Taxation and Economic Development: The State of the Economic Literature,” New England Economic Review, 36-52.
Table 1: Summary Statistics Variable Real PPE ($ millions) State Corporate Tax Rate Etaxrate Population (‘000s) Real Wage ($/hour) Road per Area (mile-1) Real Price of Energy Revenue Limit Expenditure Limit Legislature Credit Country Apportionment
Obs. 1,555 1,750 1,750 1,750 1,750 1,750 1,750 1,750 1,750 1,750 1,750 1,750
Mean 1,119.4 6.63% 6.31% 4,874.4 13.5 1.6 7.4 0.089 0.334 0.489 0.286 0.288
St. Dev. 2,170.2 2.87% 2.81% 5,254.1 1.8 1.2 2.5 0.285 0.472 0.500 0.452 0.084
Min. 0 0 0 344 10.0 0.1 2.8 0 0 0 0 0
Max. 35,419.8 12.65% 12.65% 32,182.1 21.8 6.1 16.7 1 1 1 1 0.333
Table 2: State Tax Rate by Year Year 1974 1980 1987 1992 1997
Mean 5.94% 6.22% 6.76% 7.10% 7.05%
St. Dev. 2.84% 2.95% 3.03% 2.93% 2.71%
Median 6% 6% 7% 7.81% 7.6%
Min. 0 0 0 0 0
Max. 12% 12% 12% 12.65% 12%
Table 3: Tax Rate Variation Increases Mean Increased Reductions Mean Reduction No Change
1974-1997 29 2.25% 7 1.55% 14
1974-1987 27 2.13% 5 2.36% 18
1987-1997 15 1.16% 11 0.73% 24
Table 4: FDI Equation Dependent Variable: log(Ssjt) – log(Ssot) Corporate Tax Rate
(1) TOBIT -3.020 (4.708)
(2) TOBIT-IV -16.911 (4.627)
Tax Rate x Credit Country
(3) TOBIT-IV -18.079 (4.803) 5.619 (8.394)
Tax Rate x Apportionment
-22.973 (8.780)
Credit Country Population x 103 Real Wage Real Energy Price Road per Area Constant Wald Chi2 Log Likelihood Observations ε
0.175 (0.020) -0.0260 (0.0510) -0.1166 (0.0342) 0.3261 (0.1516) -6.329 (0.8326) 106.91 -3,131.11 1,555 -0.1906 (0.2972)
0.183 (0.008) -0.0287 (0.0292) -0.1067 (0.0215) 0.4580 (0.0720) -5.753 (0.4631) 117.16 -3,121.01 1,555 -1.0676 (0.2953)
εexempt εcredit Numbers in parenthesis are standard errors. All regressions include year dummies.
Table 5: IV First Stage Regression Corporate Tax Rate Revenue Limit
(4) TOBIT-IV -8.725 (4.459)
OLS -0.0271 (0.0018) Expenditure Limit -0.0037 (0.0013) Legislature 0.0046 (0.0015) Constant 0.0290 (0.0051) Adjusted R2 0.3735 F(all) 121.72 F(IVs) 97.18 Observations 1,555 Numbers in parenthesis are standard errors
-0.584 (0.563) 0.182 (0.009) -0.0262 (0.0298) -0.1070 (0.0225) 0.4482 (0.0684) -5.620 (0.4760) 96.86 -3,125.44 1,555 -1.1413 (0.2975) -0.7866 (0.5152)
0.179 (0.009) -0.0324 (0.0287) -0.0995 (0.0232) 0.4317 (0.0721) -5.792 (0.4797) 122.92 -3,125.59 1,555 -0.9631 (0.2718)
(5) TOBIT-IV -9.959 (4.633) 5.489 (8.072) -22.702 (9.110) -0.575 (0.546) 0.178 (0.009) -0.0299 (0.0284) -0.0998 (0.0230) 0.4223 (0.0688) -5.665 (0.4589) 126.12 -3,122.61 1,555 -1.0361 (0.3018) -0.6896 (0.5370)
