Economic uncertainty and structural reforms: Evidence from stock market volatility

Bonfiglioli, Alessandra; Crinò, Rosario; Gancia, Gino Alessandro Article Economic uncertainty and structural reforms: Evidence from stock market volatility Quantitative Economics Provided in Cooperation with: The Econometric Society Suggested Citation: Bonfiglioli, Alessandra; Crinò, Rosario; Gancia, Gino Alessandro (2022) : Economic uncertainty and structural reforms: Evidence from stock market volatility, Quantitative Economics, ISSN 1759-7331, The Econometric Society, New Haven, CT, Vol. 13, Iss. 2, pp. 467-504, https://doi.org/10.3982/QE1551 This Version is available at: https://hdl.handle.net/10419/296280 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc/4.0/ Quantitative Economics 13 (2022), 467–504 1759-7331/20220467 Economic uncertainty and structural reforms: Evidence from stock market volatility Alessandra Bonfiglioli School of Economics and Finance, Queen Mary University of London and CEPR Rosario Crinò Department of Economics, University of Bergamo, CEPR, and CESifo Gino Gancia School of Economics and Finance, Queen Mary University of London and CEPR Does economic uncertainty promote the implementation of structural reforms? We answer this question using one of the most exhaustive cross-country panel data sets on reforms in six major areas and measuring economic uncertainty with stock market volatility. To identify causality, we exploit exogenous differential variation in countries’ exposure to foreign volatility shocks due to predetermined and time-invariant bilateral characteristics. Across all specifications, we find that stock market volatility has a positive and significant effect on the adoption of reforms. This result is robust to the inclusion of a large number of controls, such as political variables, economic variables, crisis indicators, and a host of country, reform and time fixed effects, as well as across various approaches for accommodating heterogeneous trends and contemporaneous shocks. Overall, this evidence suggests that times of market turmoil, which are characterized by a high degree of uncertainty, may facilitate the implementation of reforms that would otherwise not pass. Keywords. Liberalizations, stock market volatility, reforms, uncertainty. JEL classification. E02, E60, L51. 1. Introduction The Great Recession has been accompanied by an enormous increase in macroeconomic volatility, which has stimulated a new literature on how the resulting increase Alessandra Bonfiglioli: [email protected] Rosario Crinò: [email protected] Gino Gancia: [email protected] We thank three anonymous referees, Alberto Alesina, Nicholas Bloom, Allan Drazen, Jeffry Frieden, Laura Ogliari, Maria Petrova, Giacomo Ponzetto, Romain Ranciere, Kevin Sheedy, David Strömberg, Guido Tabellini, Aaron Tornell, Fabrizio Zilibotti, and seminar participants at University of Porto, UPF, the NBER Summer Institute (Political Economy), the workshop on the Political Economy of Tax Reforms (the Hague), the workshop on the impact of uncertainty shocks on the global economy (UCL) and the 2019 EEA Annual Meeting for comments. We acknowledge financial support from the Spanish Ministry of Economy and Competitiveness (ECO2014-55555-P and ECO2014-59805-P), and the Spanish State Research Agency MDM-2016-0684 under the Maria de Maeztu Unit of Excellence Programme. ©2022 The Authors. Licensed under the Creative Commons Attribution-NonCommercial License 4.0. Available at http://qeconomics.org.https://doi.org/10.3982/QE1551 468 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) in uncertainty impacts economic activity and especially investment decisions (see, e.g., Bloom (2009,2014)). Despite the growing attention of both economists and policy makers to the topic, little effort has been devoted to studying the effect of economic uncertainty on public choices. Such an omission is unfortunate, because the financial crisis, besides marking the end of a period of market stability, has also exposed the urge for structural reforms. The literature has studied extensively the effect of crises as a possible stimulus or obstacle to reforms. However, economic uncertainty per se has received scant attention. In this paper, we test the hypothesis that economic uncertainty, as captured by stock market volatility, has a causal effect on the adoption of structural reforms. In theory, the effect could be ambiguous. For instance, as in the case of private investment, uncertainty may make politicians more cautious. On the other hand, the literature has found cases in which it can promote reforms. For instance, uncertainty may divert attention and give an opportunity to implement policies that would otherwise not pass.1Iden- tifying the effect of economic uncertainty is therefore an open empirical question. Using one of the most exhaustive cross-country panel data sets on reforms together with recent measures of macroeconomic uncertainty, as captured by realized stock-market volatility, this paper finds that economic uncertainty facilitates the implementation of structural reforms. Following Giuliano, Mishra, and Spilimbergo (2013), we define a reform as an increase in deregulation indices available in six sectors: domestic financial sector, capital account, product markets, agriculture, trade, and current account transactions.2Our measure of economic uncertainty is taken from the rapidly-expanding literature (e.g., Bloom (2014)) that proxies for it with the volatility of stock market returns, built whenever possible from daily data. The idea behind this measure is that when the stock market is more volatile, macroeconomic performance is harder to predict.3The main advantage of this measure is that it is widely available and has been shown to be highly correlated with other proxies for macroeconomic uncertainty.4Henceforth, we refer to this measure as economic volatility or simply volatility. The resulting data set spans 6 sectors of reform in 56 countries with yearly observations over the 1973–2006 period. We start the analysis by showing that volatility is positively and significantly correlated with the adoption of reforms, and that this finding is robust to the inclusion of a large set of controls such as political variables, economic variables, crisis indicators, country-sector fixed effects, sector-time fixed effects, and country-specific linear trends. These estimates inform us about conditional correlations, but do not have a causal interpretation. To identify the effect of economic volatility on reforms, we develop an Instrumental Variables (IV) strategy that builds on two premises. The first is the well-known 1We discuss more in detail the predictions of existing models and the available evidence in Section 2. 2One advantage of focusing on structural reforms rather than fiscal reforms is that they are not affected by automatic stabilizers, which react directly to fluctuations in income. 3The theoretical underpinning behind this measure is that macroeconomic variables affect both expected cash flows accruing to stockholders and discount rates. Hence, when macroeconomic performance is harder to predict, the stock market becomes more volatile. Consistent with this view, stock market volatility is significantly related to the dispersion of economic forecasts (e.g., Arnold and Vrugt (2008)). 4We also compare the results obtained with alternative measures of economic uncertainty. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 469 result that stock markets are correlated across countries, so volatility shocks originating in one economy tend to spread to other countries. The second premise is that the interdependence between stock markets is stronger among countries that are more integrated with one another. Building on these insights, we construct an instrument for a country’s volatility by interacting the volatility of foreign countries with a measure of bilateral integration entirely based on predetermined geographical and historical characteristics. Identification is thus driven by the differential effect that foreign volatility shocks have on countries that differ in their exogenous exposure to these shocks. Since volatility abroad may be correlated with other characteristics that could directly influence reforms in neighboring countries, we also control for possible policy spillovers from foreign reforms, macroeconomic conditions, and interest rates. The IV regressions confirm that an increase in stock market volatility promotes the adoption of structural reforms. Next, we conduct an extensive sensitivity analysis and address potential remaining threats to identification. First, since foreign volatility is more likely to be exogenous for larger foreign countries, we show that our results continue to hold if we restrict the instrument to the largest economies or if we estimate the effect of volatility on small countries only. Second, we show that the results are largely insensitive to alternative definitions of reforms, such as focusing on large reforms or using changes in the liberalization indices over longer time windows, and to various ways of computing stock market volatility. Third, we find that the results do not crucially depend on any specific subset of countries or different reform areas. Fourth, we use various strategies to accommodate differential trends and contemporaneous shocks within country-sector pairs. Fifth, we implement a falsification test showing that current reforms are not explained by future realizations of volatility, suggesting that our results are driven by period-specific volatility shocks rather than by secular trends in reforms that antedate an increase in volatility. Finally, we study how a potential violation of the exclusion restriction would affect the statistical significance of our coefficient of interest. We find that even substantial relaxations of the exclusion restriction would leave inference informative about the effect of volatility on reforms. While the main goal of this paper is to establish the causal effect of economic volatility on reforms, in the Appendix of the Online Supplementary Material (Bonfiglioli, Crinò, and Gancia (2022)), we investigate further aspects of the relationship in light of the existing theories. One prominent view is that reforms may be triggered by economic crises, which may in turn vary systematically with volatility. Although we always control for various measures of economic activity in our regressions, we show that our results hold if we add other proxies for economic growth, and are driven neither by the presence of IMF programs nor by countries with a frequent occurrence of crises. We also show how our results compare to recent evidence on the determinants of reforms, such as Abiad and Mody (2005), Mian,Sufi,andTrebbi(2014), Ranciere and Tornell (2015), and Giuliano, Mishra, and Spilimbergo (2013). Next, we show that volatility promotes liberalizations, but not their reversals, and that it has no effect on noneconomic reforms. We also consider alternative measures of economic uncertainty used in the literature. While not conclusive, we argue that these findings are consistent with the hypothesis 470 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) that economic uncertainty may help diverting attention from the economic costs of liberalizations. The results in this paper are important in at least two respects. First, they establish a new empirical fact that may contribute to better understanding the nature of the political resistance to reforms. Second, from a policy perspective, our results suggest that times of market turmoil, which are characterized by a high degree of uncertainty, may provide an opportunity to implement liberalizations that are needed but perceived as unpopular. The remainder of the paper is organized as follows. Section 2reviews the existing theoretical and empirical literature on reforms and uncertainty. Section 3presents the data and shows some descriptive evidence. Section 4discusses our empirical approach and identification strategy. Section 5presents the main results. Section 6contains an extensive sensitivity analysis to assess the robustness of our main evidence to the use of alternative instruments, estimation samples, variables definitions, and specifications. Section 7discusses possible remaining threats to identification. Section 8concludes. In Appendix SD.2, we provide additional results on the relationship between economic volatility and reforms. 2. Economic uncertainty and reforms:Alook at the literature The literature on the political economy of reforms is vast and summarizing it goes beyond the scope of this section.5Rather, we briefly discuss some of the main theoretical channels through which economic uncertainty may affect the incentives to implement reforms and then review the existing empirical evidence. 2.1 Theory The term “reform” usually refers to a major change in policy, and common examples of structural reforms are liberalizations of markets for goods or services and changes in the regulatory environment. Even when considered welfare improving, reforms are often difficult to implement because of the unequal distribution of their costs and benefits. The costs may arise from relative price changes, implying adjustment costs, transitional unemployment and redistribution of income between different agents in the society. Frequently, the time profile is also troubling, with costs being paid up-front and benefits accruing with time (see Tommasi and Velasco (1996), for a more extensive discussion). In the absence of efficient compensation and incentive schemes, the conflict of interest between winners and losers or between voters and policy makers can lead to institutional inertia. The literature has studied how economic conditions, especially crises, may affect the resistance to reform. Since economic crises and uncertainty are often correlated, it is imperative to distinguish between them. We therefore start by studying the predictions of some of the leading approaches regarding crises, and then discuss how uncertainty may have an independent effect. 5See Tommasi and Velasco (1996) and Drazen (2000) for some surveys. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 471 Negative economic shocks may trigger reforms through various channels. Directly, an economic crisis may signal the need for reforms (see, for instance, Drazen (2000)and Ranciere and Tornell (2015)). Indirectly, it may increase the cost of waiting and hence help resolving any delay due to a war of attrition between political parties (see, for instance, Alesina and Drazen (1991)). On the other hand, an economic crisis may also reduce the likelihood of reforms. For instance, it may increase polarization, thereby weakening ruling coalitions, or it may trigger a backlash against liberal policies (see, for instance, Mian, Sufi, and Trebbi (2014)andBuera, Monge-Naranjo, and Primiceri (2011)). It has also been argued that a sovereign crisis can distort the country’s incentives, since the economic benefits of reforms may go largely to foreign creditors (see, for instance, Krugman (1988)andMuller, Storesletten, and Zilibotti (2019)). Focusing instead on economic uncertainty, there are various reasons why it may block or delay the adoption of reforms. As in models of private investment, uncertainty may increase the option value of waiting, especially when considering decisions with long-term consequences. In the influential paper by Fernandez and Rodrik (1991), uncertainty regarding the distribution of gains and losses of a policy change may lead to astatus quo bias. Economic uncertainty may amplify this bias if it makes it harder to predict who will benefit or lose from a reform. On the other hand, uncertainty can also facilitate economic reforms. In Alesina and Cukierman (1990), uncertainty can act as a smoke screen that allows politicians more freedom over their actions. In this way, uncertainty might promote any policy change. Moreover, there are instances in which this mechanism alleviates agency problems.6In particular, Bonfiglioli and Gancia (2013) show that economic uncertainty can promote the adoption of policies with short-run costs and future benefits, which seems a plausible description of liberalizations. If the upfront costs are more visible than the future benefits, politicians are subject to a myopic bias against reforms. By making the reelection probability depend more on luck than on political actions, higher economic uncertainty lowers this bias.7 In sum, the theoretical literature suggests that the relationship between economic uncertainty and reforms may largely be an empirical question and that an important challenge is to control for the independent effect of economic crises. 2.2 Evidence There is a large literature on the empirical determinants of reforms. Although many papers have studied how various economic conditions affect the likelihood of the adoption of reforms, the role of uncertainty has so far received little attention. After reviewing the 6For example, Dewatripont, Jewitt, and Tirole (1999) and Holmström (1999) present examples in which the agent works harder in order to prove his worth if the principal receives a coarser signal on performance or has less precise information about the agent’s type. More uncertainty can also lower the incentive for pandering (Maskin and Tirole (2004)) or conformism (Prat (2005)). 7Similar to Rogoff (1990), the model requires that citizens cannot fully separate the cost of a reform from the effect of the competence of the politician undertaking it. We show these results in Appendix SA, where we present a simplified version of the model. 472 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) experiences of developing countries with market-oriented policies, Tommasi and Velasco (1996) argue that there is a broad consensus in favor of the hypothesis that crises facilitate economic reforms. However, systematic evidence is still scarce. A recent paper by Ranciere and Tornell (2015) shows that trade liberalization, as measured by the Sachs and Warner (1995) index, tends to follow periods of severe crises. On the other hand, Mian,Sufi,andTrebbi(2014)andAbiad and Mody (2005) show that banking crises hinder the adoption of financial reforms.8 Most of the existing evidence focuses on the adoption of stabilization plans aimed at reducing inflation, government deficit, and the black market premium (see, among others, Alesina and Ardagna (1998), Drazen and Easterly (2001), and Hamann and Prati (2002)). This literature shows that stabilization plans are more likely to be put in place during periods when inflation, deficit, and black market premium are particularly high. Moreover, Alesina, Ardagna, and Trebbi (2006) provide evidence from a large panel of countries that fiscal reforms are more likely to occur during times of inflationary and budgetary crisis, when new governments take office and when governments are “strong.” Although crises and volatility are typically correlated, there is almost no evidence on the relationship between volatility and reforms. The only exception is Bonfiglioli and Gancia (2013), who find preliminary evidence that economic uncertainty, measured by the standard deviation of the output gap, is positively correlated with deficit stabilization in a panel of 20 OECD countries observed between 1975 and 2000. However, their analysis is limited to a restricted sample, one indicator of reform only, and provides no evidence on causality. Other political variables that have been found to be associated with more reforms include the presence of left-wing governments (e.g., Alesina, Ardagna, and Trebbi (2006) and Bonfiglioli and Gancia (2013)) and democracy (e.g., Giavazzi and Tabellini (2005) and Giuliano, Mishra, and Spilimbergo (2013)). We contribute to this literature by identifying the effect of economic volatility using a relatively new and extensive data set on structural reforms and controlling for the economic and political variables usually considered in previous work. 3. Data and descriptive evidence In this section, we present the data, with special emphasis on the indicators of structural reforms and on our proxy for economic uncertainty, and show some preliminary descriptive evidence on the relationship between the two variables. 3.1 Measuring structural reforms and economic uncertainty We base the empirical analysis on two recent data sets, which provide useful information for measuring structural reforms and economic uncertainty. For structural reforms, 8Other papers (see Broz, Duru, and Frieden (2016) and Forbes and Klein (2015)) show that governments often react to balance-of-payment crises by imposing restrictions to capital flows and trade, and that this may depend on the visibility of the costs of such policies. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 473 we rely on data that were collected and codified by the Research Department of the IMF, and consist of regulation indices for six sectors covering three areas of reform. In particular, these indices are available for the domestic financial sector and the external capital account (financial sectors reforms), for trade and the current account (foreignoriented reforms), and for product markets and agriculture (product market reforms). These measures are available for 150 countries with annual observations between 1960 and 2006. The indices of regulation, from which we derive our measures of reforms, are constructed as means or sums of a series of subindices, which are aimed at capturing the extent of regulation of a sector in different respects. We source our data from Prati, Onorato, and Papageorgiou (2013)where,asinGiuliano, Mishra, and Spilimbergo (2013), all indices take on values between 0 and 1, with 1 corresponding to the minimum degree of regulation. Since the values of these indices increase with the degree of deregulation, hereafter, we refer to them as liberalization indices. A structural reform in a sector is then measured as the annual change in its liberalization index. In Appendix SB, we provide a description of the liberalization index for each sector, along with the other variables used in the analysis. Here, we report some of the aspects that are taken into account when compiling the indices, and refer to Ostry, Prati, and Spilimbergo (2009)formore details. The index for domestic finance takes into account restrictions imposed to banks in setting interest rates, amounts and conditions on credit and in opening branches; the presence of government ownership of banks; and the quality of bank supervision. It also assesses the policies put in place to develop stock, bond, and security markets and to encourage access of foreign actors in these markets. The capital account index captures the degree of control and restrictions imposed to residents and nonresidents when borrowing or lending across the border, and to firms doing Foreign Direct Investment in the country. The index for trade is based on actual, or imputed, average tariff rates and captures the degree of restrictions applied to imports. It takes on value 0 if tariffs are above 60%. The current account index measures restrictions imposed on the proceeds from international transactions (both imports and exports) in goods and services that may be visible and invisible (e.g., finance). It therefore captures additional regulations to trade. The index for product markets focuses on the electricity and telecom sectors, and assesses to what extent these sectors are competitive and free of the direct control of the government. For instance, it contains subindices taking into account the extent of privatization, the regulatory power of the government, and the degree of competition in the electricity wholesale market and in the local telecom services. Finally, the index for agriculture captures the degree of government regulation in the market for the main agricultural export commodities of the country (e.g., wheat, soybeans, and cotton for the US or coffee and sugar for Brazil). Liberalizations in all these sectors are widely considered to be beneficial by economists, since they are believed to improve efficiency and promote economic growth. Yet, these reforms often find harsh resistance. For instance, opponents of financial deregulation argue that it may induce excessive risk taking and may lead to a costly restructuring 474 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) of the banking system. Among the downsides of trade liberalizations, reallocations and job losses are often mentioned. Privatizations are often blamed to have a regressive distributive impact and to lead to job losses and lower wages. In all cases, the potential costs are often more visible than the expected benefits for society at large, which often take the form of future economic growth. Our measure of uncertainty, aimed at capturing ex ante uncertainty about macroeconomic outcomes, is based on stock market volatility, which reflects the variability in investors’ expectations over the sales of firms. This indicator is commonly used in the literature (see Bloom (2014) for a survey) and it is often computed as implied volatility in option prices (VIX). Given the limited availability of VIX for many countries and over an extended time period, we follow Baker and Bloom (2013)andBloom (2014)andusethe volatility of daily stock market returns as our measure of uncertainty.9In particular, we use the data compiled by Baker and Bloom (2013), which cover a sample of 60 countries with daily observations on stock market indices from 1970 to 2013. The series we use is computed as the standard deviation of daily returns on the stock market index over nonoverlapping quarters. For better cross-country comparability, stock market indices are taken from the same source, the Global Financial Database. In case daily data are not available (for seven countries in the early 1980s and 1990s), weekly or monthly observations are used instead. In the analysis, we take annual averages of quarterly volatility observations. More details on the construction of this variable is provided in Baker and Bloom (2013). After merging the two data sets, we are left with a sample of 56 developed, emerging, and developing countries (see Appendix Table S1) with annual observations between 1973 and 2006, and data on structural reforms in 6 sectors. This means that, after excluding missing data, our data set contains an unbalanced panel of up to about 6700 observations. 3.2 Reforms and economic volatility: A preliminary glance at the data Table 1reports summary statistics on the liberalization indices for the six sectors. The statistics are computed across all country-year pairs in our sample, using between 1043 and 1169 country-year observations depending on the index. The liberalization indices are equal to 0.79 on average for the trade and current account sectors; in the other sectors, the indices vary from 0.28 for product markets to 0.73 for capital account, agriculture and domestic finance being in between with an average index of 0.58 and 0.67, respectively. The liberalization indices also vary substantially across country-year pairs, with standard deviations ranging from 0.2 (trade) to 0.35 (agriculture). Finally, the pairwise correlations between the six liberalization indices, shown in the right-hand panel of Table 1, are all positive and statistically significant at the 1% level. Table 2reports statistics on the occurrence and size of reforms in our sample. For each sector, columns (1)–(6) report the number and fraction of all country-year pairs 9As shown by Arnold and Vrugt (2008), US stock market volatility is significantly related to the dispersion of economic forecasts from participants in the Survey of Professional Forecasters, which is a frequently used alternative indicator of fundamental macroeconomic uncertainty. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 481 the concern that the instrument could be correlated with the error term due to adjustments in country-specific conditions or differences in the deterministic evolution of reforms across countries. Despite the large set of controls included in equation (1), two threats remain to our identification strategy. First, foreign volatility could be correlated with other characteristics of foreign countries that influence country c’s reforms directly rather than through its own volatility. To account for this, the vector of covariates Xs,c,t−1includes four spillover variables controlling for key characteristics of foreign countries that could have a direct influence on country c’s reforms. These characteristics are: (i) reforms implemented in each sector, as country cmay choose to adopt some of the reforms enacted abroad, for example, by imitation (see, for instance, Buera, Monge-Naranjo, and Primiceri (2011)); (ii) real per-capita GDP and (iii) inflation, as foreign countries’ business cycle could affect reforms in country cby influencing its own economic conditions; and (iv) interest rates, as lower returns on financial assets abroad may induce capital inflows into country c, stimulating reforms therein (see, e.g., Abiad and Mody (2005)and Bartolini and Drazen (1997)). We construct each spillover variable analogously to the instrument in equation (2), replacing volj,t−1with one of these four characteristics. A related concern is that reforms taking place in country ccould influence volatility in foreign countries, making the instrument endogenous to country c’s reforms. Conceivably, this is more likely to happen when cis a large economy and jis a small country. Accordingly, in Section 6.1 and 6.2, we show that our results continue to hold if we restrict the instrument to the largest economies and estimate equation (1) on small countries only. The second threat to identification is that, in specific countries and sectors, some underlying trend or contemporaneous shock might remain that influences the adoption of reforms and is correlated with the instrument. This concern is largely mitigated by the sector-year fixed effects and country-specific linear trends that are included in all specifications. However, these controls may still leave room to shocks hitting specific countries within sectors and to nonlinear trends. In Section 7, we use various approaches to more flexibly account for underlying trends and contemporaneous shocks, and find that they are unlikely to drive our main results. We also implement a falsification test showing that current reforms are not explained by future realizations of volatility. This further suggests that our results are driven by period-specific volatility shocks rather than by secular trends in reforms that antedate an increase in volatility. Overall, the controls included in the specification and the extensive sensitivity analysis indicate that our results are robust to the most plausible confounders. Yet, this may not dispel all possible concerns with a violation of the exclusion restriction. In Section 7, we therefore implement a complementary exercise, by studying how inference about the parameter of interest β2would change under various degrees of violation of the exclusion restriction, using an approach developed by Conley, Hansen, and Rossi (2012). Given the strong predictive power of the instrument at the first stage, we find that inference would remain informative about the causal effect of volatility even in the presence of substantial violations of the exclusion restriction. 482 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) 5. Main results The OLS estimates of β2are reported in Table 3.Tohaveasenseofhowthecorrelation between volatility and reforms is influenced by other covariates, we start with a benchmark specification that does not include any control variable, and then progressively add controls until we reach the most complete version of equation (1). Due to the long list of control variables, the full set of coefficient estimates is reported in Appendix Table S2. Column (1) refers to a simple regression of ref s,c,ton volc,t−1, including only country-sector and sector-year fixed effects. The positive and very precise estimate of β2 implies that higher stock market volatility is associated with reforms leading to deregulations. The point estimate remains stable when we add country-specific linear trends in column (2). In column (3), we further include the initial level of the liberalization index. As shown in Appendix Table S2, this variable enters with a negative and significant coefficient, suggesting that countries and sectors that start highly regulated tend to undergo stronger liberalizations. The negative autoregressive coefficient is consistent with previous findings in the empirical literature on reforms, and lends support to the view that deregulations tend to be enacted when they are needed the most. Yet, the correlation between volatility and reforms is robust to controlling for the initial level of the liberalization index. Table 3. Baseline estimates, OLS. (1) (2) (3) (4) (5) (6) (7) vol 0.743 0.657 0.363 0.747 0.965 0.947 0.941 [0.115][ 0.105][ 0.103][ 0.212][ 0.257][ 0.263][ 0.259] Country-Sector FE yes yes yes yes yes yes yes Sector-Year FE yes yes yes yes yes yes yes Country-Specific Linear Trends no yes yes yes yes yes yes Initial Liberalization Index no no yes yes yes yes yes Economic and Financial Controls no no no yes yes yes yes Development Controls no no no no yes yes yes PoliticalControls nononononoyesyes Controls for Spillovers no no no no no no yes Observations 6725 6725 6725 6381 5833 5703 5703 R-squared 0.10 0.12 0.19 0.21 0.23 0.23 0.23 Note: The regressions are estimated on pooled data across countries, sectors of reform and years. The dependent variable is the annual change in the liberalization index for a sector within country c.Vol is the one-year lag of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in a year. Initial Liberalization Index is the one-year lag of the liberalization index. Economic and Financial Controls are: average stock market returns; inflation; four dummies for the occurrence of a recession or a banking, currency and sovereign crisis in a given year; and four dummies for the occurrence of a recession or a banking, currency and sovereign crisis over the previous three years. Development Controls are: log real per-capita GDP; a dummy for countries that are OECD members; and a dummy for countries that are EU members two years later. Political Controls are: the polity2 index; a dummy for countries in which the party leading the government has a left-wing orientation with respect to economic policy; a dummy for countries with presidential political systems; and a dummy for years in which a legislative and/or an executive election takes place in country c.Controls for Spillovers are four variables defined as arithmetic averages of (i) the change in the liberalization index for a sector in country j= c, (ii) log real per-capita GDP in country j= c, (iii) inflation in country j= c, and (iv) the interest (lending) rate in country j= c,multiplied by the log inverse bilateral distance from country c. See Appendix Table S2 for the full list of estimated coefficients on the control variables. All regressors enter with a one-year lag, except for EU membership, which enters with a two-year lead. The standard errors, reported in square brackets, are corrected for clustering at the country level. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 483 The remaining columns of Table 3extend the specification by adding other timevarying controls. In column (4), we include several proxies for countries’ economic and financial conditions. In particular, we add four dummies capturing the existence of a crisis in a country: a dummy equal to 1 in the presence of a recession, defined as a negative growth rate of real per-capita GDP; two dummies indicating the onset of a banking and a currency crisis, respectively, as coded by Laeven and Valencia (2012); and a dummy equal to 1 if the country declared default on its sovereign debt. We also include four equivalent dummies taking on value 1 if a crisis of a certain type occurred over the previous 3 years, in order to account for the long-term consequences of crises besides their short-run impact. Moreover, we control for inflation, as reforms may follow changes in macroeconomic conditions even in the absence of a crisis, and for average stock market returns, given that the first and second moments of stock market returns may be correlated even after accounting for macroeconomic conditions.16 As shown in Appendix Table S2, while reforms are negatively correlated with inflation, they are not correlated with the first moment of stock market returns. The results also show that reforms are correlated with crises, with coefficient signs varying across types of crises and between the short and the long run. We go back to these results in Appendix SD.2, where we compare our findings on volatility to those of other works studying alternative determinants of reforms. In any case, adding these controls does not overturn the positive correlation between reforms and volatility, which now is possibly even stronger. In column (5), we add a number of development indicators to account for the fact that countries at different stages of development may have different incentives to adopt reforms. In particular, we include the log of real per-capita GDP and a dummy equal to 1 if a country is an OECD member in a given year. To take into account that the prospective accession to the European Union (EU) may provide additional incentives to adopt reforms, we also include a dummy that takes on value 1 at time tif a country is a member of the EU two years later (i.e., at time t+2).17 These controls enter with small coefficients and have little bearing on the estimate of β2. In column (6), we add various political controls. Giuliano, Mishra, and Spilimbergo (2013) find that democracy leads to reforms. Hence, we account for a country’s degree of democracy using the polity2 index sourced from the Polity IV database. The ideology of the ruling party may also affect the adoption of reforms, as pointed out by Muller, Storesletten, and Zilibotti (2016), among others. Therefore, we also control for a dummy taking on value 1 if the party leading the government has a left-wing orientation with respect to economic policy, as coded by the World Bank in the Database on Political Institutions (DPI). Presidential systems are argued to be better suited to overcome the resistance of small interest groups and hence to adopt more reforms (see, for instance, Persson and Tabellini (2002)). Thus, we also include a dummy equal to 1 if the political 16Like volatility, all controls enter the specification with a 1-year lag. Hence, the dummies for the existence of crises over the previous 3 years are equal to 1 if a crisis of a certain type occurred between t−2 and t−4. We have also experimented with dummies for crises over the previous 5 years, obtaining similar results (available upon request). 17We find that the effect of joining the EU is strongest 2 years before accession. However, the results are not very sensitive to changing this time window. 484 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) system is coded as presidential according to the DPI. Finally, we control for a dummy equal to 1 in years in which a legislative and/or an executive election takes place, as recorded by the DPI. None of these controls enters with a statistically significant coefficient, and the correlation between structural reforms and volatility is accordingly unchanged. Finally, in column (7), we include the four spillover variables described in the previous section. These variables control for the possibility that country c’s reforms are correlated with foreign countries’ reforms or macroeconomic variables, such as percapita GDP, inflation, and interest rates.18 Appendix Table S2 shows a positive and statistically significant coefficient on the proxy for reform spillovers, consistent with recent studies finding that imitation and catching up across countries play an important role in the implementation of structural reforms (see, e.g., Abiad and Mody (2005), Buera, Monge-Naranjo, and Primiceri (2011), Giuliano, Mishra, and Spilimbergo (2013)). The other spillover variables enter instead with small and imprecisely estimated coefficients. The estimate of β2is slightly reduced when controlling for the spillover variables, but remains positive, very precisely estimated and in the same ballpark as the previous estimates. Column (7) is our preferred specification and we henceforth refer to it as the baseline. The R2associated with this specification is 0.23, implying that a substantial fraction of the variation in structural reforms remains unexplained despite the large set of fixed effects and covariates. This result partly reflects the nature of structural reforms: in some sectors, these reforms do not occur very frequently, so the liberalization indices for these sectors exhibit sporadic changes on a yearly basis. As shown in the next section, when this feature of the data is taken into account by measuring reforms over a longer time window, the explanatory power of the regressors substantially increases. More generally, the moderate R2reported in column (7) confirms the view that the implementation of structural reforms is a complex phenomenon, which depends on many factors and is thus hard to explain. In this respect, the existing theories reviewed in Section 2.1 suggest that volatility can facilitate reforms although it does not necessarily act as a direct determinant. To dig deeper into the timing of the relationship between volatility and reforms, Figure 3reports point estimates and 95% confidence intervals of β2obtained by separately estimating the baseline specification using different lags and leads of volatility. The results show that reforms are positively and significantly correlated with past volatility over a period of 5 years, which corresponds to the typical lifetime of a legislature. Within this time frame, the correlation is stronger for shorter lags of volatility and reaches its maximum at the first lag. Consistent with this evidence, we use the first lag of volatility in our main regressions. At the same time, the significant coefficients on longer lags of volatility are consistent with the fact that some reforms take more than 1 year to be completed. In Section 6.3, we provide more evidence on this point using alternative specifications, which relate changes in the liberalization indexes to volatility over longer time 18To construct the proxy for interest rate spillovers, we use data on lending rates, which are available for the largest number of countries and years from the World Bank Development Indicators. We have also experimented with deposit rates and government bold yields (sourced from FRED, St. Louis FED) obtaining similar results (available upon request). Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 485 Figure 3. Timing of the relationship between volatility and reforms. Note:Eachcoefficientis obtained by separately estimating the specification in column (7) of Table 3using a different lag or lead of volatility, as indicated on the horizontal axis. All regressions are estimated using OLS. The confidence intervals are based on standard errors corrected for clustering at the country level and refer to the 95 per cent significance level. horizons. Conversely, Figure 3shows that reforms are uncorrelated with future volatility, suggesting that governments are largely insensitive to expectations about uncertainty when deciding upon a reform. While this finding also suggests that reforms do not seem to be a driver of volatility, we now turn to 2SLS regressions to identify a causal effect of volatility on reforms. The main results for the baseline 2SLS specification are reported in Table 4,which shows the coefficients on vol_shockc,t−1from both the first-stage regression (column 1) and the reduced-form regression (column 2), as well as the coefficient on volc,t−1from the second-stage regression (column 3). Appendix Table S3 contains the complete list of coefficient estimates on all the right-hand side variables from these three regressions. The first-stage coefficient on vol_shockc,t−1is positive, as expected, and also large and highly statistically significant, with a point estimate of 0.632 and a standard error of 0.069. This underscores the strong predictive power of the instrument at explaining differences in stock market volatility across countries.19 The reduced-form coefficient on vol_shockc,t−1is also positive and very precisely estimated, with a point estimate of 1.298 and a standard error of 0.371. These numbers imply that countries that are more exposed to foreign volatility shocks tend to implement larger reforms. Under the exclusion restriction that vol_shockc,t−1affects ref s,c,tonly through volc,t−1, the second-stage coefficient yields the causal effect of volatility on reforms. 19The Kleibergen–Paap F-statistic is equal to 82.9, and thus exceeds the value of 10 normally considered as a rule-of-thumb threshold for instrument relevance. 486 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) Table 4. Baseline estimates, 2SLS. (1) (2) (3) First Stage Reduced Form Second Stage vol 2.055 [0.545] vol_shock 0.632 1.298 [0.069] [0.371] Country-Sector FE yes yes yes Sector-Year FE yes yes yes Country-Specific Linear Trends yes yes yes Control Variables yes yes yes Kleibergen–Paap F-stat. 82.9 Observations 5703 5703 5703 R-squared 0.86 0.23 0.23 Note: The regressions are estimated on pooled data across countries, sectors of reform and years. Vol is the one-year lag of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in ayear.Vol_shock is the arithmetic average of the one-year lag of stock market volatility in all countries j= cwith non-missing observations, multiplied by the log inverse bilateral distance from country c. The dependent variable is vol in column (1) and the annual change in the liberalization index for a sector within country cin columns (2) and (3). Control variables are those included in column (7) of Table 3. See Appendix Table S3 for the full list of estimated coefficients on the control variables. The standard errors, reported in square brackets, are corrected for clustering at the country level. This coefficient corresponds to the ratio between the reduced-form and the first-stage coefficients, and is thus equal to 2.055 (s.e. 0.545). The larger size of β2when using 2SLS suggests a downward bias in the effect of volatility on reforms detected by OLS. In terms of magnitude, the point estimate of β2reported in Table 4implies that an increase in volc,t−1by one interquartile range (0.007), which is also roughly equal to one standard deviation in our sample, would lead to an increase in the dependent variable ref s,c,t by 1.4%, which is a reform of approximately the average size in our data. Overall, these numbers imply that the effect of volatility on reforms is not only statistically significant but also quantitatively sizable. 6. Robustness checks In this section, we perform an extensive sensitivity analysis to assess the robustness of the main results to the use of alternative instruments, estimation samples, variables definitions and specifications. 6.1 Alternative instruments As mentioned in Section 4, our baseline instrument treats countries equally, independent of the size of their economies and the capitalization of their stock markets. To study the implications of this formulation for our results, in columns (1) and (2) of Table 5,we reconstruct the instrument to allow for differences in stock market capitalization and GDP across countries, by multiplying the log inverse distance of country cfrom country Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 487 Table 5. Alternative formulations of the baseline instrument. (1) (2) (3) (4) Stock Market Volatility Interacted With: Inverse Distance and SMC Inverse Distance and GDP Inverse Distance Inverse Distance 2nd Stage Regression vol 2.045 2.055 1.644 1.973 [0.546][ 0.539][ 0.849][ 0.587] Observations 5703 5703 5703 5703 R-squared 0.23 0.23 0.23 0.23 1st Stage Regression vol_shock 0.025 0.025 0.374 0.24 [0.003][ 0.003][ 0.057][ 0.032] Kleibergen–Paap F-stat. 89.6 84.4 42.3 54.9 Country-Sector FE yes yes yes yes Sector-Year FE yes yes yes yes Country-Specific Linear Trends yes yes yes yes Control Variables yes yes yes yes Countries Included in Instrument All All Top-5 SMC Top-1 SMC in each Continent Note: The regressions are estimated on pooled data across countries, sectors of reform and years. The dependent variable is the annual change in the liberalization index for a sector within country c.Vol is the one-year lag of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in a year. To construct the instrument, vol_shock, the one-year lag of stock market volatility in country j= cis multiplied by: the log inverse distance of country jfrom country cand the log stock market capitalization of country jin 2006 (column 1); the log inverse distance of country jfrom country cand the log GDP of country jin 1973 (column 2); the log inverse distance of country jfrom country c(columns 3–4). The resulting products are averaged across all countries j= cbelonging to the set indicated in the last row of the table. Control variables are those included in column (7) of Table 3. The standard errors, reported in square brackets, are corrected for clustering at the country level. jin equation (2) by the log of either variable in country j.20 In both cases, the coefficient β2is very close to the baseline estimate. Foreign countries’ volatility could respond to reforms undertaken in country c,especially when foreign countries are relatively small. One way to address this concern is to exclude small countries from the construction of the instrument. Accordingly, in columns (3) and (4), we reconstruct the instrument by restricting the set of foreign countries jin equations (2) to, respectively, the top five economies by stock market capitalization and the economy with the highest stock market capitalization in each continent.21 This approach also represents a complementary way of allowing for differences in size, 20We source data on stock market capitalization and GDP from the World Development Indicators. We use data on GDP for 1973, the first sample year, and on stock market capitalization for 2006, the first year with complete data coverage for all countries in our sample. 21The top five countries in terms of stock market capitalization in 2006 are the US, Japan, the UK, France, and Germany. By continent, the top economies by stock market capitalization are the US (North America), Argentina (South America), Japan (Asia), the UK (Europe), and Australia (Oceania). 488 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) Table 6. Alternative instruments. (1) (2) (3) (4) (5) Stock Market Volatility Interacted With: Predicted Trade Predicted Trade and SMC Predicted Trade and GDP Predicted Trade Predicted Trade 2nd Stage Regression vol 1.984 1.99 1.99 1.751 1.768 [0.575][ 0.570][ 0.569][ 0.857][ 0.776] Observations 5703 5703 5703 5703 5703 R-squared 0.23 0.23 0.23 0.23 0.23 1st Stage Regression vol_shock 0.57 0.022 0.023 0.35 0.20 [0.066][ 0.003][ 0.003][ 0.057][ 0.035] Kleibergen–Paap F-stat. 73.2 77.1 75.0 36.5 32.7 Country-Sector FE yes yes yes yes yes Sector-Year FE yes yes yes yes yes Country-Specific Linear Trends yes yes yes yes yes Control Variables yes yes yes yes yes Countries Included in Instrument All All All Top-5 SMC Top-1 SMC in each Continent Note: The regressions are estimated on pooled data across countries, sectors of reform and years. The dependent variable is the annual change in the liberalization index for a sector within country c.Vol is the one-year lag of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in a year. To construct the instrument, vol_shock, the one-year lag of stock market volatility in country j= cis multiplied by: the log predicted bilateral trade between country jand country c(columns 1 and 4–5); the log predicted bilateral trade between country jand country cand the log stock market capitalization of country jin 2006 (column 2); the log predicted bilateral trade between country j and country cand the log GDP of country jin 1973 (column 3). The resulting products are averaged across all countries j= c belonging to the set indicated in the last row of the table. The log predicted bilateral trade is constructed using the estimated coefficients from a gravity-type regression of log bilateral trade in 1972 on origin country fixed effects, destination country fixed effects, log distance, a dummy for the existence of a common border, a dummy equal to 1 if both countries are landlocked, and four dummies for common religion, common legal origin, common language and a colonial relationship. Control variables are those included in column (7) of Table 3. The standard errors, reported in square brackets, are corrected for clustering at the country level. as only the top countries are included in the instrument. In all cases, the coefficient β2 is positive, precisely estimated, and in the same ballpark as the baseline estimate. We now turn to the role of geographical distance as a proxy for economic integration. As discussed in Section 4, while distance is arguably exogenous, it is not the only measure of exposure to foreign volatility. As an alternative, one could construct the instrument using direct measures of market integration, such as international trade. These measures, however, would be endogenous, as they respond to structural reforms undertaken, e.g., in the trade or current account sectors. Building on Frankel and Romer (1999), we therefore proxy for market integration between any two countries using the component of their bilateral trade that is explained by predetermined (geographical and historical) bilateral characteristics, while netting out origin- and destination-specific factors that could have a direct impact on reforms (see Appendix SC for details). With this predicted trade variable, ˆ Tc,j, in hand, we then construct an alternative instrument Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 489 by setting ln Intc,j=ln ˆ Tc,jin eq. (2). While exploiting a direct measure of market integration, this instrument requires stronger functional-form assumptions and is subject to stricter identification conditions than the baseline instrument, as none of the bilateral characteristics must be correlated with unobservable determinants of reforms. This notwithstanding, the estimates of β2reported in Table 6are close in size to their counterparts obtained with the baseline instrument and shown in Tables 4and 5. 6.2 Alternative samples As previously mentioned, our identifying assumption would be endangered if the political debate over reforms taking place in country cinfluenced volatility in foreign countries. In this respect, we have shown that our main results continue to hold when restricting the construction of the instrument to the largest countries, whose volatility is less likely to be influenced by events occurring abroad. Here, we perform a complementary exercise and reestimate the baseline specification in equation (1) after excluding large countries from the estimation sample. Focusing on small countries makes it less likely that their domestic reforms could have an influence on foreign countries’ volatility. In column (1) of Table 7, we start by dropping the US, the largest country in our sample by both GDP and stock market capitalization. In columns (2) and (3), we instead exclude all countries whose stock market capitalization and GDP, respectively, are at least as high as the sample median. The coefficient β2is positive and very precisely estimated regardless of how we restrict the sample to exclude large countries. A related concern is that, due to structural change, some countries could have actively engaged in reforms and experienced sustained volatility over the sample period. One example are Central and Eastern European (CEE) countries during the transition from a communist to a market-oriented economy. More generally, reforms could have especially taken place in developing countries, which also tended to be relatively more volatile. To account for these facts, in columns (4) and (5), we reestimate the baseline specification after excluding CEE countries and less-developed countries (LDC), respectively, from the estimation sample. The effect of volatility on reforms continue to hold equally strong in both subsamples. Finally, to allay the concern that the effect of volatility could be confounded by common trends inducing some groups of countries to adopt reforms in different waves, we estimate our baseline specification on the split samples of advanced versus non-advanced economies, as classified by the IMF, and of OECD members versus nonmember states. The similar coefficients reported in columns (6)–(7) and (8)–(9), respectively, suggest that even if more advanced countries may have concentrated their reforming efforts in different time periods than the less advanced ones, this has no bearing on the effects of volatility on reforms. 6.3 Alternative variables definitions and specifications We now consider alternative variables definitions and specifications. In Table 8,weuse alternative ways of constructing the main variables. In column (1), we recompute the key explanatory variable after excluding the quarter of maximum volatility for country 490 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) Table 7. Alternative samples. (1) (2) (3) (4) (5) (6) (7) (8) (9) No US No Large Countries (SMC) No Large Countries (GDP) No CEE No LDC Advanced Countries Other Countries OECD Members Non-OECD Members 2nd Stage Regression vol 2.087 1.539 2.124 2.20 2.37 2.366 2.097 3.115 2.792 [0.545][ 0.507][ 0.379][ 0.554][ 0.559][ 1.457][ 0.864][ 1.677][ 0.582] Observations 5535 2036 2205 5493 4959 3270 2433 3224 2479 R-squared 0.23 0.33 0.32 0.22 0.23 0.27 0.28 0.27 0.27 1st Stage Regression vol_shock 0.628 0.974 0.972 0.607 0.60 0.70 0.50 0.62 0.50 [0.071][ 0.108][ 0.118][ 0.068][ 0.074][ 0.066][ 0.087][ 0.064][ 0.103] Kleibergen–Paap F-stat. 79.2 81.8 67.5 80.4 64.0 113.6 32.4 92.6 23.7 Country-Sector FE yes yes yes yes yes yes yes yes yes Sector-Year FE yes yes yes yes yes yes yes yes yes Country-Specific Linear Trends yes yes yes yes yes yes yes yes yes Control Variables yes yes yes yes yes yes yes yes yes Note: The regressions are estimated on pooled data across countries, sectors of reform and years. The dependent variable is the annual change in the liberalization index for a sector within country c.Vol is the one-year lag of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in a year. Vol_shock is the arithmetic average of the one-year lag of stock market volatility in all countries j= cwith non-missing observations, multiplied by the log inverse bilateral distance from country c. Control variables are those included in column (7) of Table 3. CEE are Central and Eastern European countries; LDC are low-income and lower-middle income countries according to the World Bank classification; advanced and other countries are defined according to the International Monetary Fund classification. In column (2), large countries are defined as those whose stock market capitalization in 2006 is above the sample median; in column (3), large countries in terms of GDP in 2006 are defined accordingly. The standard errors, reported in square brackets, are corrected for clustering at the country level. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 497 Table 11. Threats to identification: Underlying trends. (1) (2) (3) (4) (5) (6) (7) (8) 2nd Stage Regression vol 1.8 2.212 2.389 2.04 1.79 2.29 2.26 [0.431][ 0.537][ 0.515][ 0.491][ 0.474][ 0.562][ 0.705] future_vol −0.421 [0.441] Observations 5703 5703 5703 5703 5703 5703 5703 5700 R-squared 0.24 0.23 0.23 0.24 0.24 0.23 0.23 0.23 1st Stage Regression vol_shock 0.683 0.631 0.657 0.645 0.64 0.584 0.548 [0.065][ 0.072][ 0.074][ 0.063][ 0.067][ 0.077][ 0.086] future_vol_shock 0.918 [0.110] Kleibergen–Paap F-stat. 109.7 77.6 77.9 104.0 92.3 57.0 40.2 70.0 Country-Sector FE yes yes yes yes yes yes yes yes Sector-Year FE yes yes yes yes yes yes yes yes Country-Specific Linear Trends yes yes yes yes yes yes yes yes Control Variables yes yes yes yes yes yes yes yes Controls for Trends: by initial aggr. ref. by initial GDP by initial inflation by initial degree of democr. by initial political orient. by initial OECD member. by develop. group – Note: The regressions are estimated on pooled data across countries, sectors of reform and years. The dependent variable is the annual change in the liberalization index for a sector within country c.Vol and future_vol are, respectively, the one-year lag and the one-year lead of stock market volatility in country c, computed as the arithmetic mean of all quarterly volatility observations for the country in a year. Vol_shock and future_vol_shock are the arithmetic averages of the one-year lag and one-year lead, respectively, of stock market volatility in all countries j= cwith non-missing observations, multiplied by the log inverse bilateral distance from country c. Control variables are those included in column (7) of Table 3.The regressions in columns (1)-(7) also include a full set of interactions between the year dummies and: the first-year value of the change in the aggregate liberalization index of country c (column 1); the first-year value of real per-capita GDP of country c(column 2); the first-year value of inflation in country c(column 3); the first-year value of the polity2 index of country c (column 4); the first-year value of a dummy for left-wing orientation of the government in country c(column 5); the first-year value of a dummy for membership of the OECD by country c (column 6); and a dummy for whether country cis an advanced economy (column 7). The standard errors, reported in square brackets, are corrected for clustering at the country level. 498 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) study how strong a violation of the exclusion restriction would have to be for inference about β2to become uninformative about the causal effect of volatility on reforms. If we found that inference remains informative even for sizable violations of the exclusion restriction, this would further raise confidence in our baseline results, suggesting that they are not crucially driven by confounding factors. To briefly illustrate the idea behind the approach of Conley, Hansen, and Rossi (2012) using our set-up, consider the following version of equation (1): ref s,c,t=ηs,c+ηs,t+β1libs,c,t−1+β2volc,t−1+β3Xs,c,t−1+β4,ct +λvol_shockc,t−1+s,c,t, where λis a parameter measuring the size of the violation of the exclusion restriction. The results presented so far are based on the standard IV assumption that λ=0. However, if the exclusion restriction is violated, so that λ= 0, inference about β2can still be performed, provided that alternative priors can be formed about λand conditional on the assumed values of this parameter. This can be done by estimating the following specification: (ref s,c,t−λvol_shockc,t−1)=ηs,c+ηs,t+β1libs,c,t−1+β2volc,t−1+β3Xs,c,t−1 +β4,ct+s,c,t with 2SLS, using vol_shockc,t−1as an instrument for volc,t−1. By varying the prior about λ, we can assess how inference about β2would be influenced by different degrees of violation of the exclusion restriction. We can also study how strong a violation would have to be for inference to become completely uninformative about the causal effect of uncertainty on reforms. Conley, Hansen, and Rossi (2012) emphasize that, because the sensitivity of the 2SLS estimator to violations of the exclusion restriction is inversely related to the strength of the instrument, the same value of λimplies a smaller decrease in the precision of the estimate of β2(compared to the case in which λ=0) the stronger is the first-stage relationship. We implement the above approach for several values of λand compare the resulting inference about β2. To this purpose, we set λto be a function of a parameter δthat we progressively raise so as to generate increasingly stronger violations of the exclusion restriction. In particular, δ=0 will correspond to the benchmark case in which the exclusion restriction is satisfied; δ=x>0 will correspond instead to a violation of the exclusion restriction such that a change in vol_shockc,t−1by one interquartile range has a direct effect on ref s,c,tequal to the effect of a change in volc,t−1by xinterquartile ranges. We increase δby intervals of 0.01 starting from 0. For each resulting value of λ, we estimate the confidence interval of β2for both the lower and the upper end of the support [−λ,λ]and compute the final confidence interval of β2as the union of the two confidence intervals.26 26In particular, λ≡2.055 ×δ/6, where 2.055 is the baseline 2SLS estimate of β2(see column 3 of Table 4) and the interquartile range of volc,t−1relative to vol_shockc,t−1is approximately one-sixth. Besides this “union of confidence intervals” approach, Conley, Hansen, and Rossi (2012) discuss other strategies that use Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 499 Figure 4. Sensitivity of inference about the effect of volatility to violations of the exclusion restriction. Note: The figure plots 90 per cent confidence intervals around the baseline coefficient on vol (obtained using the specification in column 3 of Table 4and indicated with a red line in the graph) for different priors about a potential violation of the exclusion restriction. Priors are described by the parameter delta reported on the horizontal axis: delta equal to zero implies that the exclusion restriction is satisfied; delta equal to x>0 corresponds to a violation of the exclusion restriction such that a change in the instrument vol_shock by 1 interquartile range has a direct effect on the dependent variable ref equal to the effect of a change in vol by xinterquartile ranges. The confidence intervals are based on standard errors corrected for clustering at the country level. The results are shown in Figure 4, which plots the 90% confidence interval of β2cor- responding to different values of δ.Whenδ=0, the confidence interval is [1.176, 2.907]. As δdeparts from this benchmark, the confidence interval progressively widens. However, thanks also to the strong predictive power of the instrument at the first stage, the decrease in precision proceeds very slowly and the confidence interval of β2starts including zero only when δ>2.14. Hence, for our parameter of interest to become statistically not significant, and thus uninformative about the causal impact of volatility on reforms, the direct effect of vol_shockc,t−1on ref s,c,twouldhavetobemorethantwice more prior information about λ, for example, using priors also on the distribution of λwithin the support or applying Bayesian techniques that use priors over all model parameters and assumptions about the error distribution. Compared to just specifying the support of λ, these alternative approaches impose additional parametric restrictions, and thus yield narrower confidence intervals around the treatment parameter. Accordingly, these approaches are advisable when researchers have additional information allowing them to confidently give a higher or lower likelihood to specific types of violation of the exclusion restriction in a certain application. Lacking this additional prior information, we remain agnostic about the distribution of λwithin a given support, so as to obtain the most conservative inference about β2. 500 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) as large as the effect of a commensurate exogenous change in volc,t−1. We conclude that even substantial, and likely implausible, relaxations of the exclusion restriction would leave inference informative about the effect of volatility on reforms. 8. Conclusions How does economic uncertainty affect the adoption of structural reforms? This paper is the first to answer this question empirically. Using an exhaustive panel data set on structural reforms and widely-used data on stock market volatility, we have shown that economic volatility is positively correlated with liberalizations in six sectors of the economy. This positive correlation is robust to the inclusion of a large set of fixed effects and a wide host of controls for political institutions as well as economic and financial crises. To identify causality, we have used an instrument that exploits exogenous differential variation in countries’ exposure to foreign volatility shocks stemming from predetermined and time-invariant bilateral characteristics. Our results have important implications. First, they suggest that times of market turmoil, which are characterized by a high degree of uncertainty, may facilitate the implementation of reforms that would otherwise not pass. Second, one hypothesis consistent with our findings is that economic volatility may alleviate electoral concerns when implementing unpopular reforms, an interpretation that can be rationalized by agency models with asymmetric information.27 If confirmed, this would suggest that promoting transparency, guaranteeing media independence and educating voters could help making reforms more politically viable. We believe that providing more evidence in this direction would be a desirable avenue for future research. References Abiad, Abdul and Ashoka Mody (2005), “Financial reform: What shakes it? What shapes it?” American Economic Review, 95, 66–88. [469,472,481,484] Alesina, Alberto and Silvia Ardagna (1998), “Tales of fiscal adjustment.” Economic Policy, 27, 489–545. [472] Alesina, Alberto, Silvia Ardagna, and Francesco Trebbi (2006), “Who adjusts and when? On the political economy of reforms.” IMF Staff Papers, 53, 1–29. [472] Alesina, Alberto and Alex Cukierman (1990), “The politics of ambiguity.” Quarterly Journal of Economics, 105, 829–850. [471] Alesina, Alberto and Allan Drazen (1991), “Why are stabilizations delayed?” American Economic Review, 81, 1170–1188. [471] 27See Bonfiglioli and Gancia (2013). Additional evidence in support of agency models is provided by Shi and Svensson (2006), who show that political budget cycles take place mainly in countries where voters cannot effectively monitor fiscal policies, and by Brender and Drazen (2008), who show that high growth increases the re-election probability especially in less developed countries. Media scrutiny has been found to improve the incentives of politicians (e.g., Snyder and Strömberg (2010)), but its effect on reforms has not been studied extensively. Ponzetto (2011) shows that more information promotes trade liberalization. Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 501 Arnold, Ivo and Evert Vrugt (2008), “Fundamental uncertainty and stock market volatility.” Applied Financial Economics, 18, 1425–1440. [468,474] Aviat, Antonin and Nicolas Coeurdacier (2007), “The geography of trade in goods and asset holdings.” Journal of International Economics, 71, 22–51. [480] Baker, Scott and Nicholas Bloom (2013), “Does uncertainty drive business cycles? Using disasters as natural experiments.” NBER Working Paper 19475, https://www.nber.org/ papers/w19475.[474,480] Bartolini, Leonardo and Alan Drazen (1997), “Capital account liberalization as a signal.” American Economic Review, 87, 138–154. [481] Bloom, Nicholas (2009), “The impact of uncertainty shocks.” Econometrica, 77, 623–685. [468] Bloom, Nicholas (2014), “Fluctuations in uncertainty.” Journal of Economic Perspectives, 28, 153–176. [468,474] Bonfiglioli, Alessandra, Rosario Crinò, and Gino Gancia (2022), “Supplement to ‘Economic uncertainty and structural reforms: Evidence from stock market volatility’.” Quantitative Economics Supplemental Material, 13, https://doi.org/10.3982/QE1551. [469] Bonfiglioli, Alessandra and Carlo Favero (2005), “Explaining co-movements between stock markets: The case of US and Germany.” Journal of International Money and Finance, 24, 1299–1316. [479] Bonfiglioli, Alessandra and Gino Gancia (2013), “Uncertainty, electoral incentives and political myopia.” Economic Journal, 123, 373–400. [471,472,500] Bonfiglioli, Alessandra and Gino Gancia (2015), “Economic uncertainty and structural reforms.” Economics Working Papers 1494, Department of Economics and Business, Universitat Pompeu Fabra, https://econ-papers.upf.edu/papers/1494.pdf.[480] Brender, Adi and Allan Drazen (2008), “How do budget deficits and economic growth affect re-election prospects? Evidence from a large panel of countries.” American Economic Review, 98, 2203–2220. [500] Broz, Lawrence, Maya Duru, and Jeffry Frieden (2016), “Policy responses to balance-of- payments crises: The role of elections.” Open Economies Review, 27, 207–227. [472] Buera, Francisco, Alex Monge-Naranjo, and Giorgio Primiceri (2011), “Learning the wealth of nations.” Econometrica, 79, 1–45. [471,481,484] Conley, Timothy, Christian Hansen, and Peter Rossi (2012), “Plausibly exogenous.” Review of Economics and Statistics, 94, 260–272. [481,496,498] Dewatripont, Mathias, Ian Jewitt, and Jean Tirole (1999), “The economics of career concerns, part II: Application to missions and accountability of government agencies.” Review of Economic Studies, 66, 199–217. [471] 502 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) Diebold, Francis, and Kamil Yilmaz, eds (2015), Financial and Macroeconomic Connectedness: A Network Approach to Measurement and Monitoring, Oxford University Press, Oxford, UK. [479,480] Drazen, Allan, ed (2000), Political Economy in Macroeconomics, Princeton University Press, Princeton, USA. [470,471] Drazen, Allan and William Easterly (2001), “Do crises induce reform? Simple empirical tests of conventional wisdom.” Economics and Politics, 13, 129–157. [472] Fernandez, Raquel and Dani Rodrik (1991), “Resistance to reform: Status quo bias in the presence of individual-specific uncertainty.” American Economic Review, 81, 1146–1155. [471] Forbes, Kristin and Michael Klein (2015), “Pick your poison: The choices and consequences of policy responses to crises.” IMF Economic Review, 63, 197–237. [472] Forbes, Kristin and Roberto Rigobon (2002), “No contagion, only interdependence: Measuring stock market comovements.” Journal of Finance, 57, 2223–2261. [479] Frankel, Jeffrey and David Romer (1999), “Does trade cause growth?” American Economic Review, 89, 379–399. [488] Giavazzi, Francesco and Guido Tabellini (2005), “Economic and political liberalizations.” Journal of Monetary Economics, 52, 1297–1330. [472] Giuliano, Paola, Prachi Mishra, and Antonio Spilimbergo (2013), “Democracy and reforms: Evidence from a new dataset.” American Economic Journal: Macroeconomics,5, 179–204. [468,469,472,473,478,483,484] Goldberg, Pinelopi, Amit Khandelwal, Nina Pavcnik, and Petia Topalova (2010), “Imported intermediate inputs and domestic product growth: Evidence from India.” Quarterly Journal of Economics, 125, 1727–1767. [496] Hamann, Javier and Alessandro Prati (2002), “Why do many stabilizations dail? The importance of luck, timing and political institution.” IMF Working Paper 02/228, https: //www.elibrary.imf.org/view/journals/001/2002/228/001.2002.issue-228-en.xml.[472] Holmström, Bengt (1999), “Managerial incentive problems: A dynamic perspective.” Review of Economic Studies, 66, 169–182. [471] King, Mervyn and Sushil Wadhwani (1990), “Transmission of volatility between stock markets.” Review of Financial Studies, 3, 5–33. [479] Krugman, Paul (1988), “Financing vs. forgiving a debt overhang.” Journal of Development Economics, 29, 253–268. [471] Laeven, Luc and Fabian Valencia (2012), “Systemic banking crises database: An update.” IMF Working Paper 12/163, https://www.imf.org/external/pubs/ft/wp/2012/wp12163. pdf.[483] Maskin, Eric and Jean Tirole (2004), “The politician and the judge: Accountability in government.” American Economic Review, 94, 1034–1054. [471] Quantitative Economics 13 (2022) Economic uncertainty and structural reforms 503 Mian, Atif, Amir Sufi, and Francesco Trebbi (2014), “Resolving debt overhang: Political constraints in the aftermath of financial crises.” American Economic Journal: Macroeconomics, 6, 1–28. [469,471,472] Muller, Andreas, Kjetil Storesletten, and Fabrizio Zilibotti (2016), “The political color of fiscal responsibility.” Journal of the European Economic Association, 14, 252–302. [483] Muller, Andreas, Kjetil Storesletten, and Fabrizio Zilibotti (2019), “Sovereign debt and structural reforms.” American Economic Review, 109, 4220–4259. [471] Okawa, Yohei and Eric van Wincoop (2012), “Gravity in international finance.” Journal of International Economics, 87, 205–215. [480] Ostry, Jonathan, Alessandro Prati, and Antonio Spilimbergo (2009), “Structural reforms and economic performance in advanced and developing countries.” IMF Occasional Papers, https://www.imf.org/en/Publications/Occasional-Papers/ Issues/2016/12/31/Structural-Reforms-and-Economic-Performance-in-Advanced- and-Developing-Countries-22594.[473] Persson, Torsten, and Guido Tabellini, eds (2002), Political Economics: Explaining Economics Policy, MIT Press, Cambridge MA, USA. [483] Ponzetto, Giacomo (2011), “Heterogeneous information and trade policy.” Unpublished Manuscript, Universitat Pompeu Fabra, https://repositori.upf.edu/bitstream/handle/ 10230/19900/1296.pdf?sequence=1&isAllowed=y.[500] Portes, Richard and Helene Rey (2005), “The determinants of cross-border equity flows.” Journal of International Economics, 65, 269–296. [480] Prat, Andrea (2005), “The wrong kind of transparency.” American Economic Review, 95, 862–877. [471] Prati, Alessandro, Massimiliano Onorato, and Chris Papageorgiou (2013), “Which reforms work and under what institutional environment? Evidence from a new data set on structural reforms.” Review of Economics and Statistics, 95, 946–968. [473,476] Ranciere, Romain and Aaron Tornell (2015), “Why do reforms occur in crises times?” Unpublished Manuscript, University of California Los Angeles, http://www.econ.ucla. edu/people/papers/Tornell/Tornell702.pdf.[469,471,472] Rogoff, Kenneth (1990), “Equilibrium political budget cycles.” American Economic Review, 80, 21–36. [471] Sachs, Jeffrey and Andrew Warner (1995), “Economic reform and the process of global integration.” Brookings Papers on Economic Activity, 1, 1–118. [472] Shi, Min and Jakob Svensson (2006), “Political budget cycles: Do they differ across countries and why?” Journal of Public Economics, 90, 1367–1389. [500] Snyder, James and David Strömberg (2010), “Press coverage and political accountability.” Journal of Political Economy, 118, 355–408. [500] 504 Bonfiglioli, Crinò, and Gancia Quantitative Economics 13 (2022) Tommasi, Mariano and Andres Velasco (1996), “Where are we in the political economy of reforms?” Journal of Policy Reforms, 1, 187–238. [470,472] Co-editor Tao Zha handled this manuscript. Manuscript received 10 February, 2020; final version accepted 30 October, 2021; available online 3 December, 2021.