Ideology and Tax. America. Ernesto Stein Lorena Caro. Regional Seminar on Fiscal Policy Santiago, Chile, March 2013

Size: px

Start display at page:

Download "Ideology and Tax. America. Ernesto Stein Lorena Caro. Regional Seminar on Fiscal Policy Santiago, Chile, March 2013"

Eunice Ellis
5 years ago
Views:

1 Ideology and Tax Revenues in Latin America Ernesto Stein Lorena Caro Regional Seminar on Fiscal Policy Santiago, Chile, March 2013

2 Motivation Shift to the left is probably the most important political development in Latin America in the last 15 years election of Chavez in Venezuela just the beginning, followed by Lagos, Lula, Kirchner, Vazquez, Morales, Correa Shift was pretty robust. Most countries that switched to the left stayed on the left (Chile, Guatemala being recent exceptions)

3 Shift to the left Timeline: Leftist Presidents in America Latina Venezuela - Chavez Venezuela - Chavez Argentina - N. Kichner Brazil - Lula Brazil - Lula Nicaragua - Ortega Venezuela - Chavez Paraguay - Lugo Guatemala - Colom Dominican Republic - Bolivia - Morales Mejia Ecuador - Correa Chile - Lagos Uruguay - Vasquez Chile - Bachelet Argentina - C. Kichner Source: Debs and Helmke (2008)

4 Shift to the left The Shift to the Left in Latin America Leftist Presidents,% Total Presidents Year Source: Author's calculation based on data from Debs and Helmke (2008)

5 Motivation Lots of work focused on documenting shift and exploring factors behind it (see Levitsky and Roberts, 2011). Very little work on the policy impact of this shift. In this paper, we focus on the impact of the shift to the left on one specific aspect of economic policy: fiscal revenues. Fiscal revenues is a good variable to look at. Larger government commonly associated with the left; lower taxes and limited spending is usually associated with the right. This is expected to be the case in particular for tax bases that fall predominantly on the rich, such as income taxes. We will look at impact on overall tax revenues, but also on some specific tax bases: income taxes, VAT and labor taxes.

6 Tax revenues in Latin America are comparatively low Total Tax Revenues Total Tax x Revenue, %GDP Year OECD Eastern Europe Asia Latin America Africa Middle East Source: IMF(2011) and BID-CIAT(2012) Number of countries by region: OECD 32, Eastern Europe 6, Asia 12, Latin America 20, Africa 21, Middle East 6

7 even after controlling for development, informality, sectorial structure of the economy and age composition Tax Pressure Gap between Eastern Europe 3.5 OECD 2.6 Africa.45 Asia.03 Latin America -1.4 Middle East Source: Author's calculation based on data from IMF(2011), BID-CIAT(2012) Controling for: GDP per capita, Level of informality, sectoral structure of the economy, age composition and rents from natural resources

8 But in recent years, revenues in Latin America have been catching up Changes in average Tax Revenues Average Tax Revenue(07-09) - Average Tax Revenue(91-93) Latin America 3.67 Asia 1.35 Middle East.672 Africa.407 OECD Source: Author's calculation based on data from IMF, BID-CIAT. Controling for: GDP per capita, Level of informality, sectoral structure of the economy age composition and rents from natural resources

9 Motivation Could it be that the shift in ideology observed in the region may be partly responsible for the increase in tax revenues? We will explore this empirically, using IDB-CIAT revenue data. We use data on ideology from expert surveys (Debs and Helmke, 2008, Murillo et al, 2008) as well as elite surveys of legislators (PELA). We find that the left is associated with higher revenues, in particular with regards to income tax revenues.

10 Relevant literature Literature on impact of partisanship/ ideology on macro outcomes goes back to Hibbs (1977). He found that leftleaning governments in developed countries had higher inflation, lower unemployment. Unemployment decreased with Democratic (US) and Labour (UK) governments, increased with Republican and Conservative governments Since early work of Hibbs, many authors have looked at impact of ideology on fiscal outcomes. While most have focused on debt, deficits and expenditures, some have focused on tax revenues, mostly in developed countries

11 Ideology and taxes: evidence from developed countries Author(s) Unit of Observation Methodology Results Besley and Case (2003) United States: States Multivariate regression, controlling for fixed effects and year fixed effects. Goverments headed by democracts are associated with higher taxes and spending than republican ones Angelopoulos, Economides and Kammas (2009) Petterson- Libdom(2008) Ferreira and Gyourko (2009) Migueis (2010) OECD Countries Fixed Effects and GMM Sweden: Municipalities Regression Discontinuity Design United States: Cities Regression Discontinuity Design Portugal: Municipalities Regression Discontinuity Design Left wing governments rely more on capital relative to labor income taxation Left governments are associated with higher taxes and spending than their right counterparts There is no significant impact of partisanship on taxation Left wing municipalities are more likely to adopt a municipal corporate tax

12 Ideology and taxes: evidence from Latin America Literature on ideology and tax policy very recent, very sparse. Sakurai and Menezes-Filho (2010) look at electoral cycles and partisan cycles in panel of Brazilian municipalities. They find impact of ideology on spending but not on tax collection. Machado and Stein (2012) also look at Brazilian municipalities, using RDD. They find some evidence that the left collects more revenue than the right from business taxes, but less revenue from property taxes. Hallerberg and Scartascini (2011) look at determinants of different types of tax reform. They find that left leaning governments are more likely to implement tax reforms that result in increased taxes, and, in particular, in income tax revenues.

13 Ideology and taxes: evidence from Latin America Hart (2010) uses expert survey data on ideology, looks at impact on taxation in a context in which tax policies are constrained by globalization. Finds that tax revenues are higher for right wing governments compared to their left leaning counterparts because they are more willing to rely on regressive consumption taxes. In comparison to Hart (2010), we use wider set of countries (17 rather than 9), wider period coverage (through 2009 rather than 2006), wider variety of ideology variables (not just expert surveys but also elite legislator surveys), as well as a broader set of methodologies.

14 Revenue data The taxation data taken from IDB/CIAT dataset (used for the 2013 IDB flagship report on taxation). Available for 21 countries in Latin America between , disaggregated by revenue source. Tax revenue data corresponds to the general government level. Comparable across countries and time (results robust to using central government data) High quality data validated by country governments. We use data on (all in proportion to GDP): Total tax revenues excluding social security taxes. Income tax revenues. Revenues from VAT. Revenues from social security taxes.

15 Ideology data expert surveys We use data from Debs and Helmke (2008), who build on earlier efforts by Coppedge (1997) and others. Covers 18 countries in Latin America, from 1990 through Governments coded on 5 point scale (1-left, 2-CL, 3-C, 4-CR, 5-R). We follow these authors in using a dummy for governments on the left. We also used a dummy for L+CL. For robustness, we use data from Murillo et al (2008) also from expert surveys. They have a more narrow view of what is left. We also combine both sources, and build a dummy for governments that are coded as L or CL in both surveys. Potential problem with expert surveys: tends to be retrospective (an expert today coding a past administration). Thus, we cannot rule out possibility that experts code as left (ex-post) a government because it raised certain taxes.

16 Ideology data elite survey of legislators The second ideology measure based on Parliamentary Elites of Latin America survey conducted by the University of Salamanca, available for 18 countries. Survey asks deputies about ideological position of main parties on a 1 (left) to 10 (right) scale. It also asks about ideology of key politicians (including the president). We used the average placement of the president s ideology as our ideology measure. Advantage: these are not retrospective. For example, survey corresponding to legislature in Argentina conducted in May/June Disadvantage: lower coverage, as some countries have only one or two waves of the survey.

17 Descriptive Statistics Variable Number of observation Mean Standard Deviation Min Max Revenues Total Tax Revenues (% GDP) VAT (% GDP) Income Tax (% GDP) Social Security Taxes (% GDP) Ideology Left dummy (D&H) President's ideology (PELA) Left + Center Left (D&H) Left (Murillo) Left + Center Left (Murillo) D&H + Murillo Controls GDP per capita (in logs) Openness (in logs) Self Employment Natural Resources Rents (% GDP) Share of the population under 15 and over 65 year

18 Methodology: Fixed Effects model Baseline model: This model allows us to identify the impact of ideology on taxation from within country variation across time. We include year dummies to avoid spurious correlation caused by simultaneous increase in leftist governments and revenues over time. Std. errors clustered by administration. Additional controls: openness (WEO); informality (WDI); age composition of the population (WDI); natural resource rents as a share of GDP (WDI).

19 Fixed Effects Results (baseline) Dependent Variable Total Tax Revenue/GDP (1) Total Tax Revenue/GDP (2) VAT/GDP (3) VAT/GDP (4) Income Tax/GDP (5) Income Tax/GDP (6) Social Security/GDP (7) Social Security/GDP (8) Left 1.860** ** [0.590] [0.199] [0.363] [0.200] President's Ideology ** ** [0.142] [0.0395] [0.138] [0.0475] Log GDP per Capita [2.395] [2.657] [1.099] [1.090] [1.304] [1.874] [1.254] [1.722] Constant [20.45] [23.81] [9.401] [9.550] [11.13] [17.20] [10.69] [15.22] Time Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Observations Number of countries Adjusted R-squared Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

20 Fixed Effects Results (with additional controls) Dependent Variable Total Tax Total Tax Revenue/GDP Revenue/GDP (1) (2) VAT/GDP (3) VAT/GDP (4) Income Tax/GDP (5) Income Tax/GDP (6) Social Social Security/GDP Security/GDP (7) (8) Left 1.440** ** [0.498] [0.214] [0.185] [0.198] President's Ideology ** ** [0.126] [0.0456] [0.0794] [0.0489] Log GDP per Capita ** 2.795* [2.656] [2.667] [1.259] [1.231] [0.893] [1.229] [1.228] [1.422] Constant [23.84] [24.63] [10.72] [10.78] [8.686] [12.96] [11.90] [12.92] Time Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Observations Number of countries Adjusted R-squared Controlling for Openness, self employment, the ratio of the population under 15 and over 65 years old and natural resources rents as GDP percentage Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

21 Fixed Effects Results: Robustness Checks Left Left + Center Left (D&H) Left (Murillo) Left + Center Total Tax Income Social VAT/GDP Revenue/GDP Tax/GDP Security/GDP 1.860** ** [0.590] [0.199] [0.363] [0.200] ** * [0.471] [0.160] [0.292] [0.158] 3.317* * [1.468] [0.314] [1.435] [0.434] * ** [0.559] [0.183] [0.344] [0.179] Left (Murillo) Left + Center Left (D&H * * ** Murillo) [0.580] [0.190] [0.370] [0.188] Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

22 Endogeneity concerns An important concern has to do with endogeneity. For example, a third factor, omitted from the model, may be explaining both ideology and the level of taxation. Most obvious candidate: preferences of the population, which are not observed More liberal preferences of the population Increase in demand for redistribution Increase in votes for left leaning candidates Higher taxes Greater odds that left will gain office If omitted variable varies through time, fixed effect model may attribute to government ideology tax increases that are due to changing preferences.

23 Exploiting the temporal pattern of taxation around shifts in ideology If we think that preferences shift gradually but government ideology jumps discretely, temporal pattern of taxation around changes in government ideology may provide some clues. The idea is to follow revenues in left-bound countries within an 8-year window centered around shift in government ideology, and check whether there is a jump in revenues around the arrival of the left. If revenues jump on impact, or gradually right after the shift, we can attribute the revenue increase to the shift in government ideology. If revenues rise gradually before the change in government ideology, the increase is more likely due to shift in preferences. Preferences for redistribution may also jump discretely. But, one would have to explain why such change happened exactly at the same time as the change in government ideology.

24 Exploiting the temporal pattern of taxation Model, = = 4 + ( ), +, In this exercise, left takes a value of 1 for the whole period for countries that at some point adopt a leftist government. We create a series of dummies Period j with j going from -4 to 4, to indicate the period before or after the arrival of the left. We interact these period dummies with our redefined left dummy For example, β -3 is the coefficient corresponding to all left-bound countries in the year 3 before the arrival of the left. β 2 is the coefficient corresponding to all left-bound countries during the second period of their first leftist administration. The key to the interpretation of the results is not the significance of the β coefficients per se, but rather the difference in these coefficients within the window, before and after the event.

25 Exploiting the temporal pattern of taxation VARI ABLES Total Tax Revenue/GDP (1) VAT/GDP (2) Income Tax/GDP (3) Social Security/GDP (4) Left* I * [0.429] [0.154] [0.231] [0.232] Left* I [0.475] [0.173] [0.281] [0.214] Left* I [0.50 9] [0.262] [0.297] [0.231] Left* I [0.594] [0.275] [0.311] [0.281] Left* I [0.671] [0.236] [0.481] [0.254] Left* I * * [0.669] [0.198] [0.418] [0.226] Left* I ** 0.526* 1.438* [0.680] [0.214] [0.549] [0.264] Left* I * [1.0 48] [0.192] [0.655] [0.226] Log GDP per capi ta [2.10 6] [1.0 94] [1.250] [1.228] Constant [17.99] [9.358] [10.67] [10.48] Observati ons Adj usted R-squared Country Fi xed Effects Yes Yes Yes Yes Ti me Year Dummi es Yes Yes Yes Yes Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

26 Results for total tax revenue Total Tax Revenue Coe efficient Time

27 Results for different tax bases Income Taxes VAT Coefficient Coefficient Time Time Social Security Taxes Coefficient Time

28 Testing differences between means Total Tax Revenue/GDP VAT/GDP Income Tax/GDP Social Security Taxes/GDP -3 to -1 vs 2 to Left (D&H) 4.94* * to -1 vs 2 to Left and Center Left (D&H) to -1 vs 2 to Left (Murillo) 25.3** ** 7.59** to -1 vs 2 to Left and Center Left (Murillo) Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1 Row 1. Difference between means. 2. F-value. 3. p-value

29 Conclusions Over the last 15 years, Latin America has been experiencing a shift to the left in government ideology, and an increase in tax revenues. In this paper, we study the link between these two trends. We find that total tax revenues are between 1.4 and 1.9 percentage points of GDP higher under governments from the left, compared to others. Thus, our results suggest that ideology is partly responsible for increase in revenues in Latin America. We find a substantial impact of ideology on income tax revenues, which increase between 0.7 and 1.3 percentage points of GDP under left-leaning governments, but no clear impact on VAT or social security taxes. The fact that revenues increase on impact following the shift in government ideology suggests that the jump can be attributed to changes in government ideology, rather than changes in preferences or other factors that tend to change more gradually.

30 Ideology and Tax Revenues in Latin America Ernesto Stein Lorena Caro Regional Seminar on Fiscal Policy Santiago, Chile, March 2013

31 Difference in Differences For the purposes of this exercise, we now focus on government changes, rather than shifts in ideology These changes may fit within 4 different categories Transitions from No Left to No Left (31) From No Left to Left (9) From Left to No Left (1) From Left to Left (3) We only focus on the first two types, leaving out the other transitions For each one of the transitions considered, we look at the two years before the arrival of the new president, as well as the first two of the new government.

32 Difference in Differences Model where: Left takes a value of 1 within the 4-year window centered around a transition to the left. For example, left=1 for Brazil from 2001 to 2004, even before the arrival of Lula in Post is equal to 1 once a new presidential period begins. For example, post=1 for Brazil for 2003 and We interact post with our redefined left dummy. A positive and significant coefficient for the interaction term is consistent with increased revenues for left-leaning governments

33 Results: Differences in Differences VARIABLES Total Tax Revenue/GDP (1) Total Tax Revenue/GDP (2) VAT/GDP (5) VAT/GDP (6) Income Tax/GDP (7) Income Tax/GDP (8) Social Security/GDP (9) Social Security/GDP (10) Left [0.667] [0.617] [0.381] [0.351] [0.360] [0.368] [0.273] [0.307] Post [0.331] [0.251] [0.184] [0.162] [0.164] [0.123] [0.123] [0.127] Left* Post 1.603* 1.542** [0.707] [0.558] [0.309] [0.268] [0.546] [0.406] [0.289] [0.275] Log GDP per capita [3.093] [2.769] [1.547] [1.579] [2.041] [1.431] [1.658] [1.675] Constant [26.33] [28.67] [13.18] [13.96] [17.35] [15.85] [14.10] [17.67] Observations Adjusted R-squared Aditional Controls Yes Yes Yes Yes Yes Yes Yes Yes Country Fixed Effects No Yes No Yes No Yes No Yes Time Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Controlling for Openness, self employment, the ratio of the population under 15 and over 65 years old and natural resources rents as GDP percentage Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

34 Placebo regressions Ourβ j coefficients compare revenues in left-bound countries in period j before/after the arrival of the left to revenues in these countries outside the event window and revenues in non-left-bound countries at any time. So in period 1 (first period of left administration following ideology shift), we compare revenues in countries with new administration vs. countries that are at different points in the political cycle. Could our results be capturing the impact of the political cycle? Placebo regressions: we center the events around the last election prior to ideology shift in left-bound countries. If the increase in revenues persists, it may be due to the political cycle effect, rather than ideology.

35 Placebo regressions VARI ABLES Total Tax Revenue/GDP (1) VAT/GDP (2) Income Tax/GDP (3) Social Security/GDP (4) Left* I * [0.652] [0.310] [0.342] [0.353] Left* I * [0.611] [0.237] [0.286] [0.30 2] Left* I [0.672] [0.349] [0.324] [0.248] Left* I [0.653] [0.341] [0.336] [0.237] Left* I ** [0.574] [0.321] [0.315] [0.190] Left* I [0.467] [0.201] [0.280] [0.240] Left* I [0.599] [0.20 7] [0.335] [0.198] Left* I [0.599] [0.280] [0.420] [0.271] Log GDP per capita [2.353] [1.0 44] [1.781] [1.269] Constant 42.68* [20.13] [8.931] [15.19] [10.82] Observati ons Adj usted R-squared Country Fi xed Effects Yes Yes Yes Yes Ti me Year Dummi es Yes Yes Yes Yes Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1

36 Placebo results: no change in tax revenue around last government change before ideology shift Total Tax Revenue Coefficie ent Time

37 Placebo results: no change in tax revenue around last government change before ideology shift Income Taxes VAT Coefficient Time Coefficient Time Social Security Taxes Coefficient Time

38 Placebo results: testing differences between means Total Tax Revenue/GDP VAT/GDP Income Tax/GDP Social Security Taxes/GDP Left (D&H) -3 to -1 vs 2 to Cluster standard errors in brackets ** p<0.01, * p<0.05, + p<0.1 Row 1. Difference between means. 2. F-value. 3. p-value

On Minimum Wage Determination

On Minimum Wage Determination Tito Boeri Università Bocconi, LSE and fondazione RODOLFO DEBENEDETTI March 15, 2014 T. Boeri (Università Bocconi) On Minimum Wage Determination March 15, 2014 1 / 1 Motivations