Policy Research Working Paper 10214 Electoral Cycles and Public Spending during the Pandemic Michael Lokshin Aylén Rodriguez-Ferrari Iván Torre Europe and Central Asia Region Office of the Chief Economist October 2022 Policy Research Working Paper 10214 Abstract This paper uses a newly assembled data set on various types allocated to social assistance and income protection and the of social protection spending in 154 countries during the lower is the share allocated to job retention schemes. The COVID-19 pandemic in 2020 and 2021 to analyze the electoral cycle appears to have impacted the size of social effect of the electoral cycle on the size and composition assistance spending only in countries with high political of the social protection stimulus budget. The analysis competition. In this sense, countries with higher political shows that the longer is the time since the last election in competition experience stronger effects of political budget a country—and thus the sooner the next election date—the cycles. larger is the share of the social protection pandemic budget This paper is a product of the Office of the Chief Economist, Europe and Central Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mlokshin@worldbank.org, arodriguezferrar@worldbank.org, and itorre@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Electoral Cycles and Public Spending during the Pandemic Michael Lokshin, Aylén Rodriguez-Ferrari, and Iván Torre* JEL: H53, I38, O15 Keywords: Social protection, pandemic, governance, electoral cycle *The authors are with the World Bank. This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. 1. Introduction Electoral cycles are important drivers of social protection policies in developing and developed countries (de Haan and Klomp 2013). Before elections, incumbent politicians may adopt expansionary fiscal policies or change the composition of government expenditures to gain voters’ support, a phenomenon known as the political budget cycle (Brender and Drazen 2005). The impact of the political budget cycles on social policies depends on the maturity of democracy, the type of government, the quality of institutions, fiscal transparency, politicians’ time horizon and tenure, and other factors (Akhmedov and Zhuravskaya 2004). The analysis of political budget cycles is complicated by the fact that the timing of elections could, in some cases, be endogenous to other drivers of government spending, as governments may try to adjust the election schedule when faced with changing economic or political circumstances (James and Alihodzic 2020). Also, most cross-country studies on political budget cycles rely on highly aggregated measures of spending which may obscure the processes that occur at a more disaggregated level. This paper addresses these two main challenges -identification and disaggregation of spending- by studying the social protection response to the COVID-19 pandemic. The COVID-19 pandemic represents a unique opportunity to observe and test the political budget cycle hypothesis. In most countries, the shock of the pandemic was unexpected and required a massive expansion of social policies (Gentilini and others 2021). The pandemic affected countries at different points in their electoral cycles, providing thus exogenous, cross-country variation that allows us to estimate the cross-country relationship between the periods of the electoral cycle and the size and composition of government expenditures on social protection. Using two new data sets on the detailed expenditures on social protection in the pandemic stimulus packages and the timing of electoral cycles, this paper tests the presence of politically motivated cycles in the composition and size of pandemic stimulus budgets. In particular, it assesses whether the period of the electoral cycle in which a country was at the beginning of the pandemic affected its social protection response to the pandemic. The high level of disaggregation of government expenditures in our data set allows us to analyze the compositional changes in public spending, overcoming the weakness of many earlier studies 2 of political budget cycles, which relied on highly aggregated data on total expenditures and revenues, current and capital expenditures, or public deficits (Castro and Martins 2018; Akhmedov and Zhuravskaya 2004; de Haan and Klomp 2013). These new data distinguish expenditures on social assistance that involve direct transfers to citizens (such as cash and in-kind transfers) from expenditures on programs focused on job retention that are usually channeled through firms (such as wage subsidies or waivers or reductions in social insurance contributions). The results of our empirical analysis demonstrate that the electoral cycle affects the composition of the social protection response. The longer the time since the last election in a country—and thus the sooner the next election date—the larger the share of the social protection pandemic budget allocated to social assistance and income protection and the lower the share allocated to job retention schemes. In contrast, the electoral cycle has no impact on the overall size of the budget allocated to social assistance, except in countries with high political competition. In these countries, the electoral cycle appears to affect the size of the social assistance budget by crowding out the resources allocated to job protection programs. The maturity of democracy, the intensity of political competition, and the rule of law in a country all seem to affect how the electoral cycle influences the composition of the social protection stimulus budget. We identify a potential source of bias in our estimates – the postponement of elections due to COVID-19 restrictions. We address this by re-estimating our model, excluding countries that postponed elections. We find that this potential bias does not affect our main empirical results. This paper contributes to the literature on the effects of electoral cycles on public expenditure in several ways. To our knowledge, it is the first study to analyze the effect of political budget cycles on the size and composition of social protection expenditures in the COVID-19 stimulus. Our analysis shows that the predictions of the political budget cycles theory apply even during systemic shocks like a pandemic. Also, our study relies on cross-country, exogenous variation in the electoral cycle, a strategy that has not been used in the literature so far. The large and highly granular cross-country data sets on which the paper is based allow us to analyze compositional changes in the stimulus budget. The paper is organized as follows. The next section describes the literature on the causes and main drivers of political budget cycles. Section 3 discusses the sources of data and presents descriptive 3 statistics. Section 4 describes the empirical strategy used. Section 5 presents the main results. Section 6 summarizes the paper’s main findings. 2. Literature Review Nordhaus (1975), Lindbeck (1976), and Tufte (1978) were the first researchers to model government investment behavior when governments are constrained by political circumstances. Nordhaus (1975) introduced the concept of a political business cycle, describing a pattern of public expenditures that starts with relative austerity in the early years after an election and ends with a binge just before an election when transfers are increased, and government expenditures are redirected toward high-visibility projects. The simple explanation of this phenomenon is that incumbent politicians want to convince voters that they care about their interests and are effective in running the government. Voters are assumed to be myopic or suffer “fiscal illusion” (Alesina, Roubini, and Cohen 1997)—valuing expansionary fiscal policies and underestimating their costs in terms of inflation and the future tax burden (Rogoff 1990). Recent studies examine why rational, forward-looking voters are influenced by pre-election budget spending and do not instead punish politicians for fiscal manipulations. The model developed by Drazen (2008) assumes that incumbent politicians seek to both increase social welfare by providing public goods and get reelected. Voters are uncertain about the ability of incumbent politicians to deliver public goods with available resources and about politicians’ preferences among different voter groups. Competent politicians may be able to provide more public goods without significantly affecting future spending and taxes and signal their ability by undertaking more visible spending before elections. Incumbents may also express their political preferences by targeting spending toward particular groups of voters. Under these assumptions, the political budget cycle would exist even in the presence of rational and fiscally concerned voters. Extensive empirical analysis has been conducted on the effect of political competition on social policies. Akhmedov and Zhuravskaya (2004) find that the size and composition of government expenditures change during the political cycle in the Russian Federation. Expenditures increased nine months before the election date and went up again one month before the elections. Spending on education, culture, and health grew considerably during the two months before the elections and decreased sharply after the elections. Drazen and Eslava (2010) find in Colombia systematic 4 changes in the composition of government expenditures but not in size. They show that after elections, transfers to retirees and payments to temporary personnel were reallocated in favor of urban infrastructure projects such as housing, roads, schools, and water plants. As democracies mature, independent media and civil society develop, and voters become better informed, incumbent politicians lose their ability to alter the size of government expenditures to serve their electoral needs. Instead, they may change the budget composition to deliver policies preferred by their voters. Castro and Martins (2016) study Portugal and Galli and Rossi (2002) study Germany. In both countries, they find that spending on public services, social protection, and health care—categories that are more visible than others to voters and provide politicians with the best chances of winning elections—increases during election years. In democracies with weak state capacity, voters may perceive higher pre-election expenditures as a signal of the incumbent’s ability to provide public goods. Politicians are incentivized to increase budget expenditure on visible items such as welfare, child benefits, social insurance, public work, and similar programs to win elections (Drazen and Eslava 2010: Harding and Stasavage 2013: Mani and Shukand 2007). Sometimes, increasing programs’ expenditures is not enough to gain votes; policies should be attributed to the efforts of particular politicians to secure voters’ support during the elections (Bueno 2021). Cash transfers and conditional cash transfers (CCTs) have a direct and immediate impact on the well-being of targeted households. Politicians, therefore, use CCTs to sway voters’ choices. Empirical studies show that increasing spending on these highly visible programs was effective in inducing voters to support incumbent politicians in Brazil (Zucco 2013), Colombia (Baez and others 2012), Honduras (Galiani and others 2019), Mexico (De La O 2013), the Philippines (Labonne 2013), and Uruguay (Manacorda, Miguel, and Vigorito 2011). Labonne (2013) shows that in the Philippines, incumbent politicians benefit from CCT programs; they gained 26 percentage points more votes in municipalities with CCT programs than in municipalities without them. Manacorda, Miguel, and Vigorito show that Uruguay’s CCT program increased support for the incumbent by 13 percentage points. Several studies investigate the drivers of political budget cycles in cross-country settings. Alt and Lassen (2006) analyze a sample of 10 countries in the Organization for Economic Co-operation 5 and Development (OECD) in 1989–98. They show that the main drivers of political budget cycles were the levels of government transparency and political polarization. Using data from 32 European countries in 1990–2010, Enkelmann and Leibrecht (2013) find that political budget cycles are typical in newly democratized countries. In contrast, voters in mature democracies penalize politicians for increasing public expenditure to enhance their chances of re-election. Using a panel of 71 countries over 1972–2009, Brender and Drazen (2013) show that politicians’ tenure may be another driver of political budget cycles and argue that to win votes, experienced politicians are more likely to change the composition of government expenditures than newcomers. A new leader might need several years to make changes in the budget composition, probably because it takes time to understand legislative procedures and bureaucratic rigidities after taking office. Castro and Martins (2018) analyze government expenditures in 18 European countries in 1990– 2012. They show that governments increase public spending during election years regardless of the political party they belong to. Kyriacou, Okabe, and Roca‐Sagalés (2022) analyze electorate myopia and show that voters in a diverse sample of 67 countries prefer the immediate benefits to their consumption rather than longer-term fiscal sustainability, incentivizing politicians to implement opportunistic policies to meet voters’ demands. The literature on the political business cycles, discussed above, does not address, however, whether political budget cycles persist in the presence of systemic shocks like a pandemic, war, or natural disaster. To our knowledge, no prior studies have analyzed whether the maturity of democracy, political competition, and the rule of law in a country remain driving factors of political budget cycles in such contexts. Also, much of the cross-country literature has not exploited an exogenous variation in the national electoral cycle across countries, which would allow for clearer identification of the effects of the political budget cycle. The shock of the pandemic, which occurred almost simultaneously across the world and was exogenous to the electoral cycle of each country, provides such variation. Our data and analysis aim to fill this gap in the literature by providing cross-country evidence on the presence, or not, of a political budget cycle in the governments’ social protection response to the pandemic. 6 3. Data The data for our analysis come from two primary data sets. The first is a database compiled by the team that includes two of this paper’s authors (Demirgüç-Kunt, Lokshin, and Torre 2022). It provides estimates of the social protection measures the government implemented in response to the COVID-19 pandemic in 2020 and 2021 in 154 countries. The data come from the Global Database on Social Protection Responses to COVID-19 (Gentilini and others 2021), budget data from official documents (including IMF Article IV revisions and other international organizations’ related documents), government websites, and news sources. For inclusion in this data set, the key criterion is that social protection spending must have been part of a pandemic response package. In some cases, this spending was part of a new program, in others, it represented the expansion of an existing program. To the extent feasible, we estimated the budget corresponding to expanding existing programs (for example, the cost of extending the eligibility period of unemployment benefits). 1 Social spending programs can be broadly divided into income protection and job protection policies (Demirgüç-Kunt, Lokshin, and Torre 2022). Income protection policies include social assistance measures such as cash transfers, food, and other in-kind transfers; waivers and reductions of utility and financial obligations; some labor market policies, such as public works programs and activation measures; and social insurance policies, such as pensions, unemployment benefits, and out-of-work support. These policies seek to directly increase incomes of beneficiaries. Job protection policies include policies aimed at job retention, such as wage subsidies and other indirect subsidies, including waivers of or reductions in social insurance contributions. These policies represent a more indirect way of supporting individuals’ welfare by preserving their employment. The two main dependent variables in our analysis are the size of spending on social assistance and income protection measures and the share of this spending in the total social protection response 1 To validate our results, we compared our estimates against estimates by the International Monetary Fund (IMF) on total additional public spending (on social protection or other programs) implemented during the pandemic (IMF 2021). In no case did our estimates exceed the corresponding IMF figures. Our estimates are also consistent with those of Gentilini and others (2022). 7 budget. Because social assistance represents about three-quarters of income protection spending, we refer to this variable as the share of social assistance spending. The share of job protection measures in the total social protection budget is the complement of this variable. The pandemic response budget corresponds to governments’ COVID-19-related spending between March 2020 and June 2021. A second data set, compiled for this paper, provides information on national elections in 205 countries. We collected data on the date the authorities in place as of March 2020 were elected, the terms of office, the anticipated election date after March 2020, the date of legislative elections after March 2020, whether elections were postponed, and whether 2020 was an election year in each country. The information for this data set comes from the National Democratic Institute (NDI n.d.), the Association of World Election Bodies (AWEB n.d.), Freedom House (Freedom House n.d.), the International Institute for Democratic and Electoral Assistance (IDEA n.d), government websites, news sources, and Wikipedia. The main variable we use to measure the electoral cycle is the number of months that have elapsed since the last election as of March 2020. The theory of political budget cycles tends to focus on the electoral cycle for upcoming elections, which would suggest using the number of months until the next election. However, many political regimes allow the incumbent to alter the electoral schedule under extraordinary circumstances (James and Alihodzic 2020). Indeed, during the pandemic, 24 countries in our sample changed the national election schedule. The number of months until the next election could thus be endogenous to the development of the pandemic. On the other hand, the number of months elapsed since the last election is completely orthogonal to the start of the pandemic. Therefore, this measure provides us with an exogenous, cross-country variation in the period of the electoral cycle, which is key for the unbiased identification of our parameters of interest. 2 2 One of the problems in analyzing the relationship between electoral cycles and social budget allocations is the potential endogeneity of these processes. Competing political parties and the public may anticipate attempts by elected officials to influence the electorate through social spending and either try to counter them or push for measures even if they were not initially planned. Another issue could be the loss-aversion bias or the “long-term movements” of policies, in which risk-averse voters prefer the status quo and the behavior of politicians during the current election 8 We also use a set of control variables in our estimations. An important aspect to control for when analyzing electoral cycles is the degree of political competition. The political competition is likely to affect the incentives of incumbents with regard to public spending around election time (Block 2001). To measure political competition, we use the index produced by the Polity5 project (Center for Systemic Peace 2020). This index measures the degree of political competition in a country, independently of the political regime. It ranges from 1 (political competition is suppressed) to 10 (political competition is institutionalized and electoral). Other aspects of governance may also affect the relationship between electoral cycles and budget outcomes. We use three indicators from the World Bank’s Worldwide Governance Indicators (WGI) (World Bank 2022b). The voice and accountability index measures the responsiveness of governments to their citizens. This index is a composite of several indicators on democracy, electoral processes, accountability of public officials, rights, reliability of government budget documents, transparency in policy making, the freedoms citizens have, and citizens’ trust in their government. The second indicator provides a measure of government effectiveness across countries. This index includes measures of the quality, coverage, and citizen satisfaction with public goods and services (roads, public transport, electricity, health and education services, water and sanitation, and bureaucratic quality). The third indicator captures the extent to which citizens and firms trust and abide by the rule of law. It includes perceptions of the quality of contract enforcement, the protection of property rights, confidence in the police and the courts, and the likelihood of crime and violence. The three indexes are normalized to mean zero and a standard deviation of one in each year. Higher values correspond to more advanced democratic institutions in a country. Other variables used in the analysis are the level of real GDP per capita, from Eurostat and World Development Indicators (World Bank 2022a), and the size of the informal sector, sourced from the Informal Economy Database produced by the World Bank (Elgin and others 2021). Table 1 summarizes the descriptive statistics. cycle depends on the outcomes of previous elections (Alesina and Passarelli 2019). By using an exogenous variation in the electoral cycle, we avoid the bias driven by these effects. 9 4. Empirical Strategy To achieve desirable electoral outcomes, incumbent politicians might try to increase overall spending or skew the budget allocation toward visible policies that benefit their voters (de Haan and Klomp 2013). According to the political business cycle theory, we expect more funds to be allocated to such policies the closer the country is to the election date. We should see such patterns in spending on social assistance and other income protection programs—the policy instruments that are most visible to voters and that directly affect beneficiaries’ welfare (Castro and Martins 2018). Politicians reinforce budget spending strategically to obtain voters’ support as the electoral day approaches (Manacorda, Miguel, and Vigorito 2011). Abundant evidence demonstrates that program beneficiaries respond to these transfers with their votes (Drazen and Eslava 2010). Based on these patterns, we formulate two testable hypotheses. The first posits that the size of spending on social assistance and other income protection programs does not change the closer a country is to the next election. Rejection of this null hypothesis would be consistent with the notion that the electoral cycle drives the size of spending on social assistance. The second hypothesis postulates that the share of social assistance and other income protection programs within the overall spending on social protection does not change as the election approaches. Rejection of this null hypothesis would be consistent with the notion that the political business cycle drives the composition of social protection spending. The simplest cross-sectional empirical specification to test these hypotheses is: = + ℎ + + ( = 1, … , ; = , ), (1) 2020−21 ≡ ( = 1, … , ) (2) 2019 2020−21 ≡ 2020−21 ( = 1, … , ) (3) where is spending on social assistance and related income protection policies in country i during 2020-2021. The absolute size, , is calculated as spending expressed in percentage points of 2019 GDP. The relative share, , is calculated as the share of the stimulus expenditures on social protection allocated to social assistance and related income protection policies. The “absolute size” 10 specification tests the first hypothesis about the incumbent increasing social assistance spending during the months leading to the election. The “relative share” specification tests the second hypothesis about the incumbent shifting the composition of the social protection budget toward more visible social assistance programs as the next election draws near. The main variable capturing the electoral cycle (ℎ ) measures the number of months that have elapsed since the last elections. We expect coefficient β to be positive in both estimations, as the longer in the past the latest elections were, the closer the next election will be. is a vector of controls that consists of two blocs of variables. The first controls for country characteristics, such as the intensity of the COVID-19 health shock (expressed in total COVID-19-related deaths per million people) (Gentilini and others 2022), the country’s GDP per capita (Lokshin, Ravallion, and Torre 2022), and the share of the informal sector in the country’s GDP (ILO 2020). The second set of controls includes variables capturing different characteristics of governance in the country. is a standard innovation error term. We also want to allow for any (continuous) nonlinearity in the relation between the dependent variable(s) and the number of months since the last election. It is reasonable to assume that the effectiveness of expanding social protection programs to gain votes is increasing the closer are the elections. Such nonlinearity may be confounding in this context. For example, if the true relation is quadratic, but we do not include the squared term in our regression, the control variables will pick up the effect of this omitted variable. Relaxing the linearity assumptions in equation (1) yields: = (ℎ ) + + ( = 1, … , ; = , ), (4) where (. ) is some (data-determined) smooth nonparametric function and is now a vector of control variables. In this specification, we can estimate equation (2) as a partial linear regression (Yatchew 1998). 5. Results Table 2 presents estimations of equation (1) under different specifications to test the hypothesis about the effect of the electoral cycle on the composition of social protection expenditures. The first column of the top panel shows the baseline specification (specification 1 without controls), which regresses the shares of the social assistance and income protection budget in the total social 11 protection response budget on the number of months since the last election. The baseline specification shows that the share of the social protection pandemic response budget allocated to social assistance and income protection programs increases with the number of months since the last election by about 4 percentage points a year. The next column (specification 2) shows the same regression with added variables, to control for relevant country characteristics. The controls have little effect on the coefficient on the number of months since the last elections, and the coefficients on the control variables demonstrate the expected signs. Countries with a larger informal sector have a larger share of social assistance and income protection policies within their policy mix; wealthier countries seem to allocate proportionally less funds to such programs. Specification 3 in table 2 adds the voice and accountability index to control for the levels of democracy, trust in government, and overall responsiveness of governments to citizens. The coefficient on months since the last election remains significant, albeit attenuated. The coefficient on the voice and accountability index is negative and significant, suggesting, consistently with the literature, that greater government accountability and transparency may prevent incumbent politicians from manipulating the composition of the social protection budget to increase “visible” spending. Similar results emerge when we include the government effectiveness and the rule of law indicators. The effect of the political budget cycle is reduced in countries with more effective governments and stronger rule of law. 3 The second panel of table 2 presents the estimation of equation (1) using months until the next election to track the electoral cycle. The coefficient on this variable is insignificant in the baseline specification and significant only in specifications 2 and 3, although the sign and magnitude of the coefficients are consistent with those in the estimations shown in the top panel of this table. 4 All 3 We also tested a specification in which we included all three governance indicators. These indicators are strongly correlated (the pair-wise correlation coefficient is about 0.75), so the individual coefficients on these regressors are insignificant. However, the joint test demonstrates a highly significant total effect of the three indicators on the composition of the social protection stimulus budget. 4 The sign of the coefficient is the opposite of that of the months since the last election coefficient, but it is as expected, as the months to the next election is equal to the total number of months in the country’s electoral cycle less months since the last election. 12 other coefficients are also close to those in the months-since-the-last-election estimations. The weaker and less significant effect of the months-until-the-next-election variable could be explained by the fact that some countries postponed their elections because of the pandemic, so this measure is less precise than the months-since-the-last-election measure in reflecting the period of the electoral cycle (Asplund 2020). We test that hypothesis in table 4 below. Table 3 presents empirical tests of the hypothesis about the impact of the political budget cycle on the size of spending on social assistance and income protection during the pandemic. The overall explanatory power of these estimations is much lower than those shown in table 2. For example, the R2 of specification 3 in table 2 (0.486) is more than ten times higher than the R2 of this specification in table 3 (0.048). The coefficients on both measures of the electoral cycle (months since the last election and months until the next election) are insignificant for all specifications. These results indicate that the electoral cycle appears to have little effect on the absolute size of the social assistance budget. However, these aggregate results might mask the heterogeneity of the effects of the electoral cycle of public spending. We address this question in the discussion below. The coefficient on the months-since-the-last-election variable could be affected by a possible attenuation bias, originating from the fact that the variable may be a poor proxy of the period of the electoral cycle in countries where governments can postpone elections. This systematic error can be particularly sizable in the context of the COVID-19 pandemic, when some countries postponed elections. To partially address this bias, we re-estimate our empirical model by excluding the 24 countries that postponed elections in 2020. Table 4 presents the linear OLS estimations for that subsample. The coefficients on the months-since-the-last-election variable are significant and positive for all five specifications. 5 The magnitude of these coefficients and their statistical significance are stronger than those shown in the top panel of table 2, confirming the presence of the effect of the political budget cycle on social assistance spending during the pandemic. Moreover, excluding countries that postponed elections turns the coefficient on the months-until-the-next-election statistically significant and of the expected (negative) sign (bottom 5 The absolute magnitudes of these coefficients differ from those in Table 2 because of the difference in the sample size (24 countries are excluded from the estimations in Table 4). 13 panel of table 4), which, again, is consistent with the presence of a political budget cycle. We might speculate that the decision to postpone the elections in 2020 was at least partially motivated by the desire of incumbent politicians to influence the election results. Politicians may have strategically shifted the stimulus budget allocations to better prepare for the new elections. Removing countries in which the elections were delayed strengthens the effect of the political budget cycle on the composition of the social protection budget. The positive linear correlation between spending on social assistance and the number of months since the last election might mask a more complex, nonlinear relationship between these variables over the electoral cycle. Figure 1 shows the results of nonparametric estimation of that relationship, as in equation (4), for four specifications, corresponding to specifications 2, 3, 4, and 5 in table 1. 6 All estimations reveal the presence of relatively strong political budget cycles in social protection stimulus spending. The shares of spending on social assistance and income protection appear to be smaller in countries that recently held an election; these shares are larger in countries where elections took place longer time ago, reaching a maximum at about 46 months since the date of the last elections. This picture fits well with the four-year electoral cycle of many countries. Globally, the average duration of electoral cycles is between four and five years (Baturo and Elgie 2019). The amplitude of the estimated political budget cycle is relatively large. The share of social assistance increases from about 68 percent of the total social protection stimulus budget for countries in the first year after the last election to about 83 percent for countries in the fourth year—a 15 percentage point difference. We can formally test the nonlinearity of the relationship between the share of social assistance and periods of the electoral cycle by estimating a version of equation (1) in which we use the cubic polynomial of the variable on the month since the last election. Table 5 shows that for all specifications, the coefficients on the cubic polynomial have the correct signs (–/+/–), and in all specifications except specification 1, the coefficients on the polynomial are jointly and individually 6 These estimations were made using Stata routine PLREG (Lokshin 2006). Table A.1, in the appendix, shows the parametric parts of the PLREG estimations. 14 statistically significant. The magnitudes and signs of other coefficients are similar to those shown in the top panel of table 3. Next, we investigate the heterogeneity of the effects of the electoral cycle on public expenditures by the degree of political competition. Several recent studies demonstrate that political budget cycles are positively correlated with electoral competitiveness. Using data on 27 countries in the European Union for 1997–2008, Efthyvoulou (2012) shows that tighter electoral competition is correlated with a higher probability of observing a political budget cycle. Analysis of the composition of public expenditures in 69 developing countries from 1975 to 1990 demonstrates that in competitive elections, incumbent politicians are willing to incur the economic costs of changing the composition of public spending (Block 2001). Prichard (2018) and Vergne (2009) show similar results in a sample of 42 developing countries over the period 1975–2001. Eibl and Lynge-Mangueira (2017) show the mediating effect of electoral competitiveness on the political budget cycle on a sample of 112 countries from 1960 to 2006. To test the effect of political competition on public spending during the electoral cycle in the pandemic, we split countries in our sample into two groups based on the political competitiveness index. The 53 countries in the first group have an index of political competition with a value of 8 or less; the index for the 72 countries in the second group is greater than 8. A value of 8 in the political competition index represents a transitional situation between factional/restricted competition (values 7 and below) and institutionalized electoral competition (values of 9 and 10). 7 Figure 2 shows the semiparametric graphs of the relationships between the share of the social protection stimulus budget spent on cash and in-kind transfers and the number of months since the last election for these two groups of countries. The share of social assistance and income protection in the social protection stimulus budget in countries with a low political competition index (the left-hand panel of figure 2) seems to increase almost linearly from the early post-election month, peaking at about 97 percent of the total budget by the fourth year after the election. The political budget cycle is much more pronounced for 7 Our robustness checks indicate that the results that follow are not affected by including in either of the two groups countries with a value of 8. 15 countries with a high index of political competitiveness and is similar to the dynamics shown in figure 1. The share of social assistance in the social protection budget is particularly low for countries in the first months after the election. It bottoms out at around 18 months after the election, climbing to the highest point toward the end of the typical electoral cycle of 48 months. These results, providing evidence on the existence of different spending patterns depending on the degree of political competition, allow us to revisit the analysis of the effects of the political business cycle on the size of social assistance spending. As seen in the results in table 3, the electoral cycle appeared to have no significant association with the size of spending. Figure 3 shows the semiparametric graphs of the relationship between the size of social assistance and income protection spending and the number of months since the last election for countries with low or high political competition. The difference is striking: the size of social assistance spending increases strongly after elections in countries with low political competition (and decreases afterward), while it mildly decreases (and increases afterward) in countries with high political competition. These opposite trends could explain the lack of statistical significance of the months- since-last-election variable in the linear, aggregate analysis, shown in table 3. These results show that a political budget cycle of the expected pattern exists both in the size of social assistance spending and the composition of social protection spending in countries with more competitive political environments. In countries with lower political competition, there is no evident political budget cycle in the composition of spending, and there appears to be a somewhat counterintuitive cycle in the size of spending – suggesting that other political factors beyond the electoral cycle may be influencing spending patterns. The fact that spending on social assistance and income protection increases both in its size and in its relative share of social protection stimulus spending begs the question of whether it is crowding out spending on other social protection policies – namely, on job protection policies. Figure 4 shows the semiparametric graphs of the relationship between the size of job protection spending and the number of months since the last election for countries with low and high political competition. In countries with high political competition, the spending on job protection programs follows almost an opposite pattern to the one on social assistance and income protection programs (figure 3). The spending on job protection in countries with low political competition shows no 16 change over the electoral cycle. These patterns suggest that in countries with high political competition, the spending on social assistance and income protection appears to crowd out the spending on job protection programs over the electoral cycle. As elections approach, governments appear to reduce spending on job protection programs and increase it on social assistance and income protection policies. Once elections are behind, the spending on job protection increases, and spending on social assistance declines. 6. Conclusions In this paper, we use newly assembled data on various types of social protection spending across 154 countries during the COVID-19 pandemic in 2020 and 2021 to analyze the effect of the electoral cycle on the size and composition of the social protection pandemic stimulus budget. Electoral cycles appear to affect the size and composition of social programs in politically competitive environments, even during systemic shocks, such as the COVID-19 pandemic. Our results demonstrate that the preferences and motivations of politicians seeking re-election should be taken into account when designing social policies. To stay in power, incumbent politicians may feel the need to show immediate results by directing resources to highly visible policies, such as social assistance, and underinvesting in other policies, such as job protection schemes, which may be particularly important in severe economic downturns. A similar argument could apply to underinvestment in policies that bring significant longer-term benefits, such as health and education. Expanding social programs at certain periods of the electoral cycle to achieve political gains may undermine the policies’ effectiveness in protecting the most vulnerable population groups (Bueno 2021). Increasing the provision of public goods to gain votes could increase budget deficits and repayment costs, increasing inflation and the future tax burden. Reducing taxation to secure voter support before elections may decrease the quality of social protection programs and endanger their long-term sustainability (Prichard 2018). Better governance and promotion of the rule of law could mitigate these effects and result in more effective public policies and higher welfare. Our estimations show that the degree of political competition affects the relationship between the electoral cycle and the composition of the social protection policy mix implemented in response to the pandemic. The greater the political competitiveness, the larger the political budget cycle 17 effect. For countries with high political competitiveness, strengthening accountability to voters could incentivize long-run investments and promote welfare. 18 References Akhmedov, A. and E. Zhuravskaya (2004). “Opportunistic Political Cycles: Test in a Young Democracy Setting,” Quarterly Journal of Economics, 119(4): 1301-1338. Alesina, A. and F. Passarelli (2019). “Loss Aversion in Politics,” American Journal of Political Science, 63 (4): 936–947. Alesina, A., Roubini, N. and G. Cohen (1997). Political cycles and the macroeconomy. MIT Press. Alt, J. and D. Lassen (2006). “Transparency, Political Polarization, and Political Budget Cycles in OECD Countries.” American Journal of Political Science, 50(3): 530–50. Asplund, E. (2020). Global overview of Covid-19: Impact on elections, International IDEA. AWEB (n.d.). “World Election Calendar” Accessed July 7, 2022. http://www.aweb.org/eng/bbs/B0000007/list.do?menuNo=300052&option=all. Brender, A. and A. Drazen (2005). “Political Budget Cycles in New versus Established Democracies.” Journal of Monetary Economics 52, no. 7 (October 2005): 1271–95. Brender, A. and A. Drazen (2013). “Elections, Leaders, and the Composition of Government Spending.” Journal of Public Economics 97: 18–31. Baez, J., Camacho A., Conover E. and R. Zarate (2012). “Conditional Cash Transfers, Political Participation, and Voting Behavior,” Working Paper, World Bank, 36. Baturo, A. and R. Elgie (2019). “The Politics of Presidential Term Limits,” New York: Oxford University Press, 2019. Block, A. (2001). Elections, Electoral Competitiveness, and Political Budget Cycles in Developing Countries. Harvard University, 2001.78. Bueno, N. (2021). “The Timing of Public Policies: Political Budget Cycles and Credit Claiming.” American Journal of Political Science, https://doi.org/10.1111/ajps.12688. Castro, V. and R. Martins (2016). “Are There Political Cycles Hidden inside Government Expenditures?” Applied Economics Letters 23(1): 34–37. Castro, V. and R. Martins (2018). “Politically Driven Cycles in Fiscal Policy: In-Depth Analysis of the Functional Components of Government Expenditures.” European Journal of Political Economy, 55: 44–64. Center for Systemic Peace (2020). “Polity5: Political Regime Characteristics and Transitions 1800-2018”. Dataset available at: https://www.systemicpeace.org/inscrdata.html. 19 De La O, A. (2013). “Do Conditional Cash Transfers Affect Electoral Behavior? Evidence from a Randomized Experiment in Mexico.” American Journal of Political Science 57 (1): 1–14. Demirgüç-Kunt, A., Lokshin, M. and I. Torre (2022). “Protect Incomes or Protect Jobs? The Role of Social Policies in Post-pandemic Recovery.” World Bank Policy Research Working Papers, WPS10166. Drazen, A. and M. Eslava (2010). “Electoral Manipulation via Voter-Friendly Spending: Theory and Evidence.” Journal of Development Economics, 92: 39-52. Drazen, A. (2008). “Political budget cycles.” In P. Macmillan (Ed.), The new Palgrave dictionary of economics. :1– 10. Palgrave Macmillan. Eibl, F. and H. Lynge- Mangueira (2017). “Constraints, competition, and competitiveness: Explaining the non-linear effect of democratization on political budget cycles.” European Political Science Review, 9(4): 629– 656. Efthyvoulou, G. (2012). “Political Budget Cycles in the European Union and the Impact of Political Pressures.” Public Choice 153 (3–4): 295–327. Elgin, C., Kose, A., Ohnsorge, F. and S. Yu (2021). “Understanding Informality.” CERP Discussion Paper 16497, Centre for Economic Policy Research, London. Enkelmann, S, and L. Markus (2013). “Political Expenditure Cycles and Election Outcomes: Evidence from Disaggregation of Public Expenditures by Economic Functions.” Economics Letters 121 (1): 128–32. Freedom House. n.d. “Afghanistan: Freedom in the World 2022 Country Report.” Freedom House. Accessed July 7, 2022. https://freedomhouse.org/country/afghanistan/freedom- world/2022. Galli, E. and S. Rossi (2002). “Political Budget Cycles: The Case of the Western German Länder.” Public Choice 110: 283-303. Galiani, S., Hajj N., McEwan P., Ibarrarán P., and N. Krishnaswamy (2019). “Voter Response to Peak and End Transfers: Evidence from a Conditional Cash Transfer Experiment.” American Economic Journal: Economic Policy 11 (3): 232–60. Gentilini, U., Almenfi, M., Orton, I., and P. Dale (2021). “Social Protection and Jobs Responses to Covid-19.” Living Paper, World Bank. 20 ________. (2022). “Social Protection and Jobs Responses to Covid-19: A Real-Time Review of Country Measures” Living Paper, version16, World Bank. Haan, J., and J. Klomp (2013). “Conditional Political Budget Cycles: A Review of Recent Evidence.” Public Choice 157 (3–4): 387–410. Harding, R., and D. Stasavage (2013). “What Democracy Does (and Doesn’t Do) for Basic Services: School Fees, School Inputs, and African Elections.” Journal of Politics 76(1): 229–245. IDEA (n.d). “Global Overview of COVID-19: Impact on Elections” Accessed July 12, 2022. https://www.idea.int/news-media/multimedia-reports/global-overview-covid-19-impact- elections#-a-id-postponed-elections-due-to-covid-19-name-postponed-elections-due-to- covid-19-postponed-elections-due-to-covid-19-a-. International Labor Organization (ILO) (2020). “COVID-19 crisis and the informal economy Immediate responses and policy challenges.” ILO Brief, Geneva, Switserland International Monetary Fund (IMF) (2021). “Fiscal Monitor Database of Country Fiscal Measures in Response to the COVID-19 Pandemic,” Fiscal Affairs Department, International Monetary Fund, Washington DC. James, T. and S. Alihodzic (2020). “When Is It Democratic to Postpone an Election? Elections During Natural Disasters, COVID-19, and Emergency Situations.” Election Law Journal: Rules, Politics, and Policy, 19(3): 344-362 Kyriacou, A., Okabe, T. and O. Roca‐Sagalés (2022). “Conditional Political Budget Cycles: The Role of Time Preference.” Economics & Politics, 34(1): 67–91. Labonne, J. (2013). “The Local Electoral Impacts of Conditional Cash Transfers.” Journal of Development Economics 104: 73–88. Lindbeck, A. (1976). “Stabilization Policy in Open Economies with Endogenous Politicians,” American Economic Review, 66: 1-19. Lokshin, M. (2006). “Difference–Based Semiparametric Estimation of Partial Linear Regression Models,” Stata Journal 3: 377-383. Lokshin, M., Ravallion, M., and I. Torre (2022). “Is Social Protection a Luxury Good?” National Bureau of Economic Research, Working Paper Series #30484. Manacorda, M., Miguel, E. and A. Vigorito (2011). “Government Transfers and Political Support.” American Economic Journal: Applied Economics 3 (3): 1–28. 21 Mani, A. and S. Mukand (2007). “Democracy, visibility and public good provision.” Journal of Development Economics 83: 506–529 NDI (n.d). “Elections Calendar.” Text. Global Elections Calendar. Accessed July 7, 2022. https://www.ndi.org/elections-calendar. Nordhaus, W. (1975). “The Political Business Cycle,” Review of Economic Studies, 42(2): 169- 190. Prichard, W. (2018). “Electoral Competitiveness, Tax Bargaining and Political Incentives in Developing Countries: Evidence from Political Budget Cycles Affecting Taxation.” British Journal of Political Science, 48(2): 427–57. Rogoff, K. (1990). “Equilibrium Political Budget Cycles,” American Economic Review, 80(1): 21- 36. Tufte, E. (1978). Political Control of the Economy, Princeton: Princeton University Press. USAGov, n.d. “Presidential Election Process | USAGov.” Accessed July 4, 2022. https://www.usa.gov/election. Vergne, C. (2009). “Democracy, elections and allocation of public expenditures in developing countries.” European Journal of Political Economy, 25(1): 63– 77. World Bank (2021). World Development Indicators, World Bank, Washington DC, USA World Bank (2022b). Worldwide Governance Indicators, World Bank, Washington DC, USA Yatchew, A. (1998). “Nonparametric Regression Techniques in Economics,” Journal of Economic Literature 36: 669–721. Zucco, C. (2013). “When Payouts Pay Off: Conditional Cash Transfers and Voting Behavior in Brazil 2002–10.” American Journal of Political Science, 57 (4): 810-822. 22 Figure 1 Semiparametric estimation of the relationship between the number of months since the last election and spending on social assistance and income protection as a share of the total social protection stimulus budget 23 Figure 2 Semiparametric estimation of the relationship between the number of months since the last election and spending on social assistance and income protection as a share of the total social protection stimulus budget in countries with low and high levels of political competition 24 Figure 3 Semiparametric estimation of the relationship between the number of months since last election and spending on social assistance and income protection as a share in country’s 2019 GDP in countries with low and high levels of political competition 25 Figure 4 Semiparametric estimation of the relationship between the number of months since last election and spending on job protection policies as a share in country’s 2019 GDP in countries with low and high levels of political competition 26 Table 1 Descriptive statistics Variables Mean Standard Minimum Maximum Number of Deviation countries Share of social assistance and income 0.741 0.318 0.000 1.000 158 protection programs in SP budget Spending on social assistance and income 1.280 1.770 0.000 13.461 158 protection programs (% of the 2019 GDP) Spending on job protection programs (% of 0.722 1.242 0.000 6.477 158 the 2019 GDP) Months since last election as of March 2020 25.948 21.353 0.000 184.000 155 Months until next election as of March 2020 27.490 17.541 1.000 69.000 151 Log per capita 2019 GDP 9.458 1.171 6.806 11.703 154 Share of informal sector in GDP 28.165 11.200 7.967 63.404 141 Total COVID19-related deaths per million 0.870 0.983 0.003 5.992 157 (cumulative as of March 2020) Governance indicators (2019) Voice and accountability -0.058 0.951 -1.994 1.655 156 Government effectiveness -0.001 1.000 -2.279 2.221 156 Rule of law -0.041 0.988 -2.351 2.058 156 Index of political competition (2019) 7.071 2.979 1.000 10.000 140 27 Table 2 Effect of the electoral cycle on the distribution of the social protection stimulus budget (share of social assistance and income protection in social protection budget) Specification 1 Specification 2 Specification 3 Specification 4 Specification 5 Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Months since the last election 0.003 ** 0.002 0.003 ** 0.001 0.002 * 0.001 0.002 * 0.001 0.002 * 0.001 Total Covid death per 1000 -0.020 0.024 -0.016 0.023 -0.028 0.024 -0.034 0.023 Log per capita GDP -0.107 *** 0.026 -0.081 *** 0.029 -0.047 0.038 -0.043 0.033 Share of informal output 0.008*** 0.002 0.006** 0.003 0.006** 0.003 0.004 0.003 Governance indicators Voice and accountability -0.072** 0.034 Government effectiveness -0.098** 0.046 Rule of law -0.126*** 0.040 Constant 0.665 *** 0.045 1.501 *** 0.296 1.306 *** 0.307 1.012 *** 0.370 1.031 *** 0.322 R2 0.034 0.453 0.472 0.473 0.465 Number of countries 141 126 126 126 126 Months until the next election 0.000 0.002 -0.002* 0.001 -0.002* 0.001 -0.002 0.001 -0.002 0.001 Total Covid death per 1000 -0.028 0.024 -0.023 0.023 -0.037 0.023 -0.043 * 0.023 Log per capita GDP -0.118 *** 0.026 -0.087 *** 0.029 -0.052 0.038 -0.046 0.032 Share of informal output 0.008 *** 0.002 0.006 ** 0.003 0.005 ** 0.003 0.003 0.003 Governance indicators Voice and accountability -0.082** 0.034 Government effectiveness -0.106** 0.045 Rule of law -0.138*** 0.039 Constant 0.734*** 0.050 1.733*** 0.292 1.488*** 0.303 1.183*** 0.369 1.179*** 0.319 R2 0.001 0.461 0.486 0.485 0.513 Number of countries 139 125 125 125 125 Note: *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. Table 3 Effect of the electoral cycle on the size of the budget for in-kind and cash transferssocial assistance and income protection programs (share of the budget in country’s 2019 GDP) Specification 1 Specification 2 Specification 3 Specification 4 Specification 5 Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Months since the last election -0.010 0.009 -0.008 0.010 -0.008 0.010 -0.006 0.010 -0.007 0.010 Total Covid death per 1000 -0.137 0.173 -0.132 0.174 -0.081 0.173 -0.100 0.176 Log per capita GDP 0.317 0.193 0.345 0.215 -0.073 0.280 0.150 0.246 Share of informal output -0.001 0.018 -0.003 0.019 0.012 0.019 0.009 0.020 Governance indicators Voice and accountability -0.077 0.254 Government effectiveness 0.636* 0.335 Rule of law 0.325 0.299 Constant 1.574 *** 0.278 -1.359 2.166 -1.569 2.282 1.799 2.712 -0.144 2.437 R 2 0.008 0.047 0.048 0.075 0.057 Number of countries 141 126 126 126 126 Months until the next election -0.003 0.009 -0.000 0.009 -0.000 0.009 -0.001 0.009 -0.001 0.009 Total Covid death per 1000 -0.130 0.174 -0.127 0.176 -0.076 0.174 -0.092 0.177 Log per capita GDP 0.335 * 0.193 0.354 0.217 -0.076 0.282 0.149 0.249 Share of informal output -0.001 0.018 -0.002 0.019 0.013 0.019 0.010 0.020 Governance indicators Voice and accountability -0.051 0.255 Government effectiveness 0.661** 0.335 Rule of law 0.355 0.300 Constant 1.418*** 0.300 -1.719 2.158 -1.871 2.297 1.702 2.748 -0.295 2.468 R2 0.001 0.042 0.042 0.072 0.053 Number of countries 139 125 125 125 125 Note: *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. 29 Table 4 Effect of the electoral cycle on the distribution of the social protection stimulus budget in countries that did not postpone their elections during the pandemic (share of social assistance and income protection in the social protection budget) Specification 1 Specification 2 Specification 3 Specification 4 Specification 5 Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Months since last election 0.004 ** 0.002 0.003 ** 0.002 0.003 ** 0.002 0.003 ** 0.002 0.003 ** 0.002 Total Covid death per 1000 -0.028 0.027 -0.024 0.027 -0.036 0.027 -0.040 0.026 Log per capita GDP -0.100 *** 0.030 -0.081 ** 0.033 -0.047 0.043 -0.042 0.037 Share of informal output 0.008*** 0.003 0.007** 0.003 0.007** 0.003 0.005 0.003 Governance indicators Voice and accountability -0.057 0.039 Government effectiveness -0.086* 0.051 Rule of law -0.114*** 0.044 Constant 0.639 *** 0.048 1.413 *** 0.338 1.272 *** 0.350 0.970 ** 0.425 0.983 *** 0.368 R 2 0.051 0.441 0.452 0.456 0.476 Number of countries 117 107 107 107 107 Months until next election -0.000 0.002 -0.003** 0.001 -0.003** 0.001 -0.003** 0.001 -0.003** 0.001 Total Covid death per 1000 -0.039 0.026 -0.034 0.026 -0.047 * 0.026 -0.052 ** 0.026 Log per capita GDP -0.117 *** 0.030 -0.093 *** 0.032 -0.061 0.043 -0.051 0.036 Share of informal output 0.008 *** 0.003 0.006 ** 0.003 0.006 ** 0.003 0.004 0.003 Governance indicators Voice and accountability -0.069* 0.038 Government effectiveness -0.090* 0.050 Rule of law -0.128*** 0.043 Constant 0.744*** 0.057 1.747*** 0.338 1.564*** 0.350 1.270*** 0.427 1.239*** 0.368 R2 0.001 0.454 0.471 0.471 0.498 Number of countries 116 106 106 106 106 Note: *** indicates that the coefficient is significant at 1% level, ** - at 5% level, * - at 10% level. 30 Table 5 Cubic polynomial estimation of the effect of the electoral cycle on the size of the budget for social assistance and income protection Specification 1 Specification 2 Specification 3 Specification 4 Specification 5 Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Months since the last election -0.002 0.009 -0.017 * 0.010 -0.020 * 0.011 -0.017 0.011 -0.018 * 0.010 Months since the last election2 0.000 0.000 0.001* 0.000 0.001** 0.000 0.001* 0.000 0.001* 0.000 Months since the last election3 -0.000 0.000 -0.000* 0.000 -0.000** 0.000 -0.000* 0.000 -0.000* 0.000 Total Covid death per 1000 -0.021 0.023 -0.017 0.023 -0.029 0.023 -0.035 0.023 Log per capita GDP -0.109 *** 0.026 -0.080 *** 0.029 -0.050 0.038 -0.044 0.032 Share of informal output 0.008 *** 0.002 0.006 ** 0.003 0.006 ** 0.003 0.004 0.003 Governance indicators Voice and accountability -0.077** 0.034 Government effectiveness -0.096** 0.045 Rule of law -0.125*** 0.039 Constant 0.707*** 0.081 1.623*** 0.302 1.429*** 0.309 1.145*** 0.374 1.158*** 0.326 R2 0.037 0.469 0.491 0.489 0.511 F-test of joint significance 1.74 2.63* 2.56 * 2.30* 2.22* (p-value) (0.161) (0.053) (0.056) (0.081) (0.089) Number of countries 141 126 126 126 126 Note: F-test and p-value indicate the joint significance of coefficients on a cubic polynomial of months since the last election. *** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. 31 Appendix Table A.1 Nonparametric estimation of the effect of the electoral cycle on the share of social assistance and income protection programs in the social protection budget Specification 1 Specification 2 Specification 3 Specification 4 Specification 5 Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Coeff. Std. Error Total Covid deaths per 1000 n.a a -0.033 0.033 -0.032 0.031 -0.045 0.033 -0.054 * 0.032 Log per capita GDP -0.090 *** 0.035 -0.046 0.037 -0.016 0.049 -0.013 0.041 Share of informal output 0.011 *** 0.003 0.008 ** 0.003 0.008 ** 0.003 0.006 * 0.003 Voice and accountability index -0.130 *** 0.047 Government effectiveness index -0.125** 0.059 Rule of law -0.160*** 0.050 Pseudo-R 2 0.416 0.465 0.445 0.480 V-test -1.531 -1.093 -1.462 -1.280 Number of countries 126 126 126 126 Note: V-test tests hypothesis of nonlinearity of functional form of months since last election variable. *** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. a The semiparametric estimation by Yatchew (1998) requires at least one independent variable to be present in the linear part of the non-parametric equation. 32