Tax Smoothing Hypothesis : A Turkish Case

We tested the tax smoothing hypothesis for Turkey using annual data for the period of 1949-2010. Although our preliminary estimation results imply the existence of the weak form of tax smoothing for Turkey, further tests indicate the violation of exogeneity of permanent government spending, which is a requirement for the tax smoothing hypothesis to hold. Our causality tests indicate that permanent government spending is not exogenous due to the causality running from lagged tax rates to permanent government spending. Therefore, we conclude that our results provide evidence against the tax smoothing hypothesis. Our results are important because the existence of random-walk behavior of the tax rates alone or some preliminary regressions do not guarantee the existence of tax smoothing.

The concept of tax smoothing suggested in a seminal article by Robert J. Barro (1979) and extended to more general settings by Robert E. Lucas Jr. and Nancy L. Stokey (1983) has become one of the most important concepts with substantial policy implications in public finance.Tax smoothing is especially important in developing countries with relatively scarce resources.Implementation of certain policies to smooth taxes avoids excessive deadweight losses on the shoulders of taxpayers.Therefore, policy makers could achieve a more efficient allocation of resources and promote growth in a better, and most importantly, in a healthier way.As mentioned by Philip Arestis (2011) and Michael Bergman (2011), the quality of fiscal policy is especially important for countries that are in turbulent times.An efficient taxation behavior in good times can be a lifesaver when times are troublesome.Moreover, the latest global financial crisis triggered the need to re-focus on the fiscal side of the economy due to the fact that monetary policy is insufficient to combat economic recession.Thus, the issue of taxation and fiscal policy practices in both developed and developing countries become even more crucial.Recently, several studies such as Guglielmo D 'Amico, Giuseppe Di Biase, andRaimondo Manca (2013), Jesús Ferreiro, Carmen Gómez, andFelipe Serrano (2013), and Jérôme Creel, Paul Hubert, and Francesco Saraceno (2014) investigated different aspects of fiscal policy.
When there is no uncertainty, optimal tax rates are persistent under certain assumptions.Conversely, in the case of a stochastic economy with an incomplete bond market, tax rates should follow a random walk generated by a Martingale process.In this perspective, the tax smoothing idea also has a strong and basic logic.As ex-PANOECONOMICUS, 2014, 4, pp.487-501 pected, distortions and burdens due to taxation rise disproportionately with tax rates.Governments, therefore, can curtail tax distortions through smoothing of tax rates rather than adjusting them based on the budget requirements.Thus, the idea of tax smoothing implies that budget deficits or surpluses can be used to keep tax rates relatively stable.At this point, the availability and use of debt instruments plays a crucial role to smooth taxes and shape taxation policies.In addition, unlike temporary changes, governments should adjust tax rates in response to permanent increases in government spending.According to the tax smoothing hypothesis, tax rates should only be changed in the case of unpredicted shocks and, thus, predictable changes cannot occur in the tax rates.
In this paper, we tested the tax smoothing hypothesis for Turkey by applying more direct methods for the first time.Our focus is on the existence of tax smoothing, which is shown to hold seemingly in our initial analysis for the Turkish case.However, more in-depth analysis clarifies the results.Our results are important because like other developing economies with relatively insufficient resources, Turkey has no luxury to trifle away its resources by inappropriate taxation policies.Thus, we believe that our results will lead policy makers to take corrective actions.
The next section of the paper presents literature survey.Section 2 outlays the empirical methodology and Section 3 presents empirical results.The last section concludes the paper with some policy implications.

Literature Review
A large number of studies have examined the tax smoothing hypothesis and have very diverse results.Most of these studies analyzed the hypothesis for developed countries using different subgroups of countries and time periods.The very first studies of tax smoothing for developed countries belong to Barro (1979Barro ( , 1981)).In this preliminary stage of the literature, Barro argued that tax rate should be constant in a stable economy under certainty where government expenditure is exogenously given and an initial amount of debt exists.On the other hand, the expected value of tax rate is equal to the current rate in the case of uncertainty.The policy makers' duty is to adjust tax rates to minimize deadweight loss burden with respect to borrowing constraints.Barro (1979Barro ( , 1981) ) found support for the existence of tax smoothing hypothesis in the US.In an important contribution, Atish R. Ghosh (1995) argued that there are at least two reasons for going beyond the random-walk tests of tax smoothing models.First, it is often difficult to reject the null hypothesis of a random walk for many economic time series.Secondly, tax smoothing is not the only explanation for tax rate changes being unpredictable.Ghosh (1995) used vector autoregression method to test the tax smoothing hypothesis for the US and Canada, and the result showed that the hypothesis holds both for the US and Canada.Apostolos Serletis and Richard G. Schorn (1999) applied vector autoregression for Canada, France, the US, and the UK, and pointed out that even though revenue smoothing was not the case for these countries, tax smoothing behavior exists in these economies.Mark C. Strazicich (2002) used a panel of industrialized countries to check the existence of tax smoothing by means of panel unit-root tests.The results of the study supported tax smoothing in the context of Barro (1979).Johan Adler (2006) extended the theoreti-PANOECONOMICUS, 2014, 4, pp.487-501 cal framework in Ghosh (1995) and utilized vector autoregression on a Swedish dataset.Adler (2006) found empirical results that both support tax smoothing in Sweden and explained about 60% of variability in Swedish central government surpluses.
Recently, Gerhard Reitschuler (2010) tested the tax smoothing hypothesis for the 15 EU-member countries by means of vector autoregression.The basic estimation results of Reitschuler (2010) supported the hypothesis for Germany, Greece, the Netherlands, and Portugal.On the other hand, his more complicated estimation results pointed out that the tax smoothing hypothesis holds only for the cases of Germany and the Netherlands before the Maastricht fiscal rule, but it is rejected for all countries in the dataset after the Maastricht fiscal rule takes effect.In another study, Reitschuler (2011) investigated the existence of tax smoothing for the new member states of the EU.In this study, the hypothesis was found to be valid for the Czech Republic, Hungary, Lithuania, Poland, and Romania and it was also shown that the tax smoothing behavior of these countries was not affected by the Maastricht fiscal rule.In another recent study, Ananda Jayawickrama and Tilak Abeysinghe (2013) used a direct method to test the existence of tax smoothing for Australia, Canada, Italy, Japan, the Netherlands, New Zealand, the US and the UK.They also classified the forms of tax smoothing into "no tax smoothing", "weak form", and "strong form" for these countries.Their results are in favor of the weak form of tax smoothing for all countries they analyzed.
There are also many studies reporting the contradictory results on the existence of tax smoothing for developed countries.Chaipat Sahasakul (1986) used the US data for the period of 1937-1982 and showed that tax rates respond not only to permanent increases in government spending but they also respond to temporary defense purchases, general price level, and a time trend.These results invalidate the tax smoothing hypothesis for the said period.Bharat Trehan and Carl E. Walsh (1988) used co-integration among government expenditures inclusive of interest, tax receipts, and seignorage for the US data covering the period between 1890 and 1986.They found that even though intertemporal budget balance exists, the tax smoothing hypothesis does not hold.Henning Bohn (1990) showed that tax smoothing cannot be rejected on the basis of time path of taxes for post-war US data.However, the results of this study pointed out that tax smoothing can be rejected by means of an extensive analysis that includes the fact that some security returns are correlated with tax rates.Chao-Hsi Huang and Kenneth S. Lin (1993) examined the relationship between budget deficit and growth rates of national income and government expenditure and they applied vector autoregression to the US data covering the period of 1929-1988.They rejected the tax smoothing hypothesis for the full period.However, contradictory to the results on the full period, they failed to reject the existence of tax smoothing after 1947.Nilss Olekalns (1997) tested the tax smoothing hypothesis for the case of Australia using a dataset that covered between 1964 and 1995.Olekalns (1997) presented results against tax smoothing because the Australian fiscal policy was too volatile.Recently, Reitschuler (2011) rejected the tax smoothing hypothesis for Bulgaria, Cyprus, Estonia, Latvia, Malta, Slovak Republic and Slovenia.
There are relatively fewer studies investigating the tax smoothing hypothesis for developing countries.Nearly all of these studies are country-specific case studies.Paul Cashin, Olekalns, and Ratna Sahay (1998) tested Barro's tax smoothing model on India for the period between 1951 and 1997 and showed that the tax smoothing hypothesis holds for central government, but this was not the case for regional governments.Cashin, Nadeem Haque, and Olekalns (1999)  There are even fewer studies providing evidence against the tax smoothing hypothesis in developing countries.Cashin, Haque, and Olekalns (1999) used unitroot test on a dataset covering between 1964 and 1997 for Sri Lanka.Their results showed that Sri Lanka's fiscal behavior was not in line with the idea of tax smoothing.Fabiana Rocha (2001) applied vector autoregression to Brazilian data for the period between 1970 and 1994.His results lent evidence on the rejection of the tax smoothing hypothesis for the full sample.Recently, Mesut Karakas, Taner Turan, and Halit Yanikkaya (2014), using Turkish data with VAR estimation method, reported that although the results of preliminary analyses lent evidence in favor of the hypothesis, further in-depth econometric analysis failed to support the existence of tax smoothing.
The existence of tax smoothing is tested mostly with indirect empirical methods.Tests on the tax smoothing hypothesis usually investigate the random-walk behavior of tax rates.However, this does not guarantee the existence of tax smoothing because, as in our study, the existence of random-walk behavior of tax rates does not necessarily mean that there is actually tax smoothing.Thus, commonly used tests in related literature are less likely to give robust and consistent results.Recently, Jayawickrama and Abeysinghe (2013) proposed a much better and direct way to investigate the tax smoothing hypothesis.They argued that if tax smoothing holds, then future tax rates should co-integrate with current permanent government expenditure rates even though tax rates and government expenditure rates are random walks.In our paper, we followed this direct methodology for the first time to test the tax smoothing hypothesis in Turkey using the period of 1949-2010.Our analysis also went a step further from the usual indirect determination of tax smoothing and classified tax smoothing into strong form, weak form, or no tax smoothing as alternative cases.Our initial results indicate the existence of the weak form of tax smoothing for Turkey.However, because an assumption of the exogeneity of permanent government spending is violated due to the causality running from lagged tax rates to permanent government spending, we can conclude safely that our estimation results lend evidence against the tax smoothing hypothesis for Turkey.

Methodology and Data
Jayawickrama and Abeysinghe (2013) tested the existence of tax smoothing hypothesis and its forms (weak or strong) via four main equations using average tax rate ( ) (calculated as total tax revenue to GDP ratio) and government-spending rate ( ) (calculated as government spending to GDP ratio).They first decomposed government-spending rates by means of Kalman (KF) and Beveridge-Nelson (BN) filters into two parts: permanent government-spending rate ( ) and transitory governmentspending rate ( ).Afterwards as in almost all works related to tax smoothing, the random-walk behavior of tax rate is determined by an ADF (Augmented Dickey-Fuller) test: (1) As tax rate is assumed to have a Martingale property, the unit-root component of government-spending rate (permanent government-spending rate) can be extracted and used in the co-integration test between tax rate and permanent governmentspending rate:

=
(Beveridge-Nelson filtered series), or = (Kalman filtered series). (2) The existence of tax smoothing and its form are also checked directly by means of an error correction model (ECM) as in below: ( Where m t is the growth rate of M1 definition of money supply and u t is the unemployment rate.
If α 1 = 0, then the tax smoothing hypothesis does not hold.If -1 < α 1 < 0, the weak form of tax smoothing hypothesis holds.When α 1 = -1, the strong form of tax smoothing holds.In the case where the tax smoothing hypothesis is satisfied, λ 1 generally equals to 0.
Lastly, checking for causality is needed.Permanent government-spending rate is assumed exogenous, but there can be spurious or true feedback effects.To determine such a problem, we run the regression in Equation ( 4): If a government-spending rate is exogenous, the coefficients of right-hand side variables are usually insignificant.When spurious feedback effects exist, α 2 is significant and takes negative values.However, in the case of true feedback effects, α 2 is significant and takes positive values.The existence of true feedback effects points out that even though the tax smoothing hypothesis is supported throughout Equations (1-3), there is actually no tax smoothing in the economy.

Empirical Results
Our study examined both the existence and the form of tax smoothing for Turkey within the period of 1949-2010.Data on government expenditures and tax revenues came from the Turkish Ministry of Finance; data on M1 (money supply) were from the Turkish Statistical Institute; and data on unemployment were from Tuncer Bulutay (1995) until 1980, and afterwards, from the Turkish Ministry of Development.We also obtained data on GDP from the Ministry of Development to calculate tax and spending shares in GDP.The tax rate and actual government-spending rate are depicted in Figure 1.Both series seem to follow a form of random walk.
Source: Authors.Actual government-spending rates and filtered government-spending rates are given in Figure 2. Actual government-spending rate and the Beveridge-Nelson unitroot smooth series of government-spending rate show close correspondence, but the Kalman smooth series of government-spending rate diverges from these series during the economic crises at the late 1980s and 1990s.Nevertheless, it can be easily observed that fluctuations in government-spending rates stem from the long-term shifts in government-spending rates.The co-integration results based on Equation (2) between tax rates and permanent government expenditure for BN and KF series are given in Table 3.The residual-based ADF tests and KPSS test support the co-integration.The degree of cointegration measured by Rho is not very strong but permissible.Most interestingly, β coefficients estimated from BN and KF series are very similar.AR and ARCH tests show autocorrelation and heteroscedasticity, but these do not create problems except in the case of strong tax smoothing hypothesis.
To test the tax smoothing hypothesis, the ECM in Equation ( 3) is applied to BN and KF series.Table 4 presents the OLS results.Except for the estimated coefficients on the transitory part of government-spending rate, the estimated coefficients in the two models are almost identical.The difference in the coefficients on transitory government spending stems mainly from the methodological approaches that BN and KF use.The growth rate of M1 money does not have any effect on changes in tax rates probably due to the fact that the excessive indirect taxation in Turkey extinguishes the effects of any monetary expansion.The social welfare expenditures in Turkey have a higher share in government budget.Our results indicate that higher unemployment rates raise the tax rates to cope with social welfare issues.As noted in Jayawickrama and Abeysinghe (2013), however, the presence of these transitory effects does not necessarily violate the tax smoothing hypothesis.The statistically significant estimate on λ 1 (the coefficient on the permanent government spending) implies the lack of a one-to-one relationship between the tax rate and the permanent government expenditure rate in the long-run.If we had longer and better data on tax rates, we could actually get this one-to-one relationship.The statistically significant estimate on α 1 indicates the presence of co-integration between tax rates and permanent government expenditures.Because our results show that α 1 is between -1 and zero, the weak form of tax smoothing hypothesis seems to hold for Turkey.PANOECONOMICUS, 2014, 4, pp. 487-501 Although our analyses so far provide significant evidence for the weak form of tax smoothing, the exogeneity of permanent government expenditures is a crucial issue for the Turkish case.We performed a causality test as in Equation ( 4); the test results are in Table 5.The estimated coefficients on independent variables are insignificant except for M1 and they have similar signs.Generally, we expect all variables to be insignificant for the exogeneity of permanent government spending.Most importantly, as α 2 is statistically significant and positive, it gives rise to true positive feedback effects in our case.Thus, based on our preliminary results as similar to most of the related literature, the tax smoothing hypothesis initially seems to hold for Turkey; however, it does not stand when scrutinized closely.Similar to the many studies undertaken in the tax smoothing literature, the period of 1949-2010 is quite long and there can be structural breaks that may affect the results of our analyses1 .To cope with this, Bai-Perron multiple structural break tests were applied to Equation (2).The results of sequentially determined structural break tests for co-integration equation with BN filtered series are given in Table 6.The Bai-Perron tests determined only one structural break date for the co-integration equation in 1979, the year when there was an oil crisis.Source: Authors' estimations.
Table 7 presents the results of Bai-Perron multiple structural break tests for co-integration equation with Kalman filtered series.There were four structural breaks at 1965, 1980, 1991, and 2000.These structural break dates correspond mainly to domestic economic crises except for 1980.In addition, the 1979-1980 oil crisis was captured as a structural break date for the co-integration equation with Kalman smooth series.1965,1980,1991,2000 Notes: Critical values are based on significance at 5% level.
Note that we have only one structural break when we employed the Beveridge-Nelson unit-root smoothed series.Figure 2 shows actual government-spending rate and the Beveridge-Nelson unit-root smooth series of government-spending rate show close correspondence.On the other hand, the Kalman smooth series of government-spending rate shows divergent behavior at times of economic crises in the late 1980s and 1990s.In such a case, it is proper to expect less structural breaks for an equation with the Beveridge-Nelson unit-root smooth series.
After determining the structural break dates, we utilized structural dummy variables (sd) for Equations (3) and (4).Using structural dummy variables, we tested whether structural breaks have important effects on our analyses.The results of ECM with structural dummies in Equation (3) are given in Table 8.The coefficients on structural dummy variables are insignificant for the series.Coefficients on the other variables are similar with the results reported in Table 4.We can then conclude safely that the existence of structural breaks does not alter our conclusion of the weak form of tax smoothing hypothesis for Turkey.To test the exogeneity of permanent government expenditure within the structural-breaks context, we performed the causality test with structural break dummies as in Equation (4).The test results are reported in Table 9.The coefficients on structural dummy variables are not significant for the cases of BN and Kalman filtered series.Coefficients on the other variables are very similar compared with previous results in Table 5. Causality tests with structural dummies also support the existence of true positive feedback effects because α 2 is statistically significant and positive.
Overall, our results with structural breaks indicate that even though structural breaks exist in the time period we considered, estimation results with structural breaks do not alter our main results in any significant way.
Our empirical results for Turkey, which is an emerging economy, are consistent with that of Rocha (2001), presenting evidence against tax smoothing for Brazil, while they differ from that of Cashin, Olekalns, and Sahay (1998) for India, Cashin, Haque, and Olekalns (1999) for Pakistan, Kurniawan (2011) for Indonesia, and Pasten and Cover (2011) for Chile, which find evidence for the tax smoothing hypothesis in some emerging economies.Thus, we can say that at best, empirical results on tax smoothing are mixed for these economies.Although there is a lack of studies on tax smoothing for those countries to reach a consensus on the issue, we can still put forward some explanations for these divergent results.For example, Cashin, Olekalns, and Sahay (1998) and Cashin, Haque, and Olekalns (1999) discussed possible reasons that can lead to deviations from tax smoothing.They argued that the inability of developing countries to collect revenue from conventional ways might be a possi-PANOECONOMICUS, 2014, 4, pp.487-501 ble reason.Rocha (2001) suggested that political economy factors would better explain the observed deviations from tax smoothing.One can persuasively argue that because a well-functioning public finance system and effective institutional structure can result in implementing tax smoothing easier, tax smoothing is thus expected to be achieved in developed countries rather than in developing ones.However, empirical studies show that there is evidence both for and against tax smoothing in both developing and developed countries.Therefore, we may conclude that empirical results do not lend any conclusive evidence on the proposition that compared with developing countries; developed countries are more successful in achieving tax smoothing.

Conclusion
Our paper examined whether the tax smoothing hypothesis holds for Turkey using a more direct testing method that is recently developed.Our initial estimation results imply the existence of the weak form of tax smoothing.Further tests on causality, however, indicate that permanent government spending is not exogenous due to the causality running from lagged tax rates to permanent government spending.Thus, we can safely conclude that our results lend evidence against the tax smoothing hypothesis.Given the frequent economic and financial crises experienced throughout the period considered, the failure to find evidence for tax smoothing for Turkey is not surprising.Our estimation results are important because the existence of random-PANOECONOMICUS, 2014, 4, pp.487-501 walk behavior of the tax rate alone or some preliminary regressions do not guarantee tax smoothing.
Whether or not a country achieves tax smoothing has significant fiscal policy implications.If implemented appropriately, fiscal policies consistent with the tax smoothing idea can avoid potential welfare losses for the overall public by minimizing deadweight losses stemming from erratic changes in tax rates.Most importantly, given that they have relatively fewer resources; developing countries do not have the luxury to waste their resources due to the suboptimal taxation policies.Because tax smoothing has different policy implications depending on whether a change in government spending is permanent or temporary, governments should be more focused on making a distinction regarding this issue.In practice, however, making a distinction between permanent and temporary government spending would be challenging but not impossible.For example, a priori, governments can classify governmentspending items while preparing or drafting next year's budget proposal.It may then be easier for policymakers to differentiate between permanent and temporary parts of government spending.
Therefore, in the case of temporary change in spending, it should be dealt mainly with public borrowing.For example, after the massive earthquake occurred in 1999 in Turkey, which is temporary in nature, the Turkish government introduced some new or additional taxes to finance extra government spending.The Turkish government should have financed the additional spending with budget deficits to comply with tax smoothing.Taxpayers in most developing countries such as Turkey suffer from the distortionary effects of sporadic changes in tax rates.To mitigate such effects, changes in tax rates should also be implemented in accordance with the changes in permanent government spending.
However, in addition to the sound fiscal institutions and prudent fiscal rules, government credibility and sound fiscal stance can also be the key on the implementation of tax smoothing.Governments with weak public finances are frequently exposed to pressures on the spending side, and they are prone to respond to higher spending by raising tax rates without taking the source of spending into account.Moreover, in cases where government credibility is missing due to the continuously increasing public-debt stock, markets may react to more government borrowing by charging higher interest rates and prevent government from attaining a tax smoothing policy.
applied unit-root tests on a dataset for Pakistan which covered between 1956 and 1995.The results support the existence of tax smoothing in Pakistan.Recently, Rudi Kurniawan (2011) used a vector autoregression methodology on a dataset between 1969 and 2008 for Indonesia.The results of this study lent support for tax smoothing in Indonesian economy.In another recent work, Roberto Pasten and James P. Cover (2011) also utilized vector autoregression on Chilean dataset for the period between 1973 and 2002.They provided strong evidence for the existence of tax smoothing when royalty revenues of the government from the mining sector are assumed not totally controlled by the administration.

Figure 1
Figure1Tax and Government-Spending Rates (% on Vertical Axis)

Figure 2
Figure 2 Actual, BN Unit-Root, and Kalman Smooth Series of Expenditure Rates (% on Vertical Axis)

Table 1
Test for a Random Walk in Tax Rates

Dependent variable Constant τ t-1 Δτ t-1 Δτ t-2 AR(2) F ARCH(2) F
The coefficient t-values are in parentheses and diagnostic p-values are in square brackets.The critical value for ADF test under the null of a driftless walk at the 5% level is -2.92.

Table 2
ADF Test for a Unit-Root in Government Expenditure Rates Source: Authors' estimations.

Table 3
Test for Co-Integration between Tax Rate and Permanent Expenditure Rate (Dependent Variable: ) Source: Authors' estimations.

Table 8
Further Tests on Tax Smoothing with sd (Dependent Variable: Δτ t ) Source: Authors' estimations.