Reexamining how markets respond to irregular regime changes

Stable capital flows are important to financial stability in emerging markets, which are particularly exposed to shifts in the availability of foreign capital. ^{Footnote 9} High country risk reduces capital flows, which in turn has been shown to reduce domestic output growth. ^{Footnote 10} Conventional theory and cross-country empirical evidence suggests that political instability—as measured by events such as protests, strikes, assassinations, number of regime changes, and ongoing elections—depresses investment. ^{Footnote 11} However, if market returns reflect investor expectations about future economic growth and expected returns, there are reasons to believe such hypotheses may be overly simplistic.

First, a change in regime may be seen as a positive event for future economic performance if the current regime is regarded as anti-business or anti-global. Regime changes that attempt to overthrow anti-business autocrats may be expected to result in positive abnormal returns as investors see little risk of a replacement government “worse” than the status quo. For example, there is an active debate on whether “good coups” that lead to democratization or economic liberalization can be good for economic growth. Alesina et al. (Reference Alesina, Özler, Roubini and Swagel1996) argue that the negative effects of uncertainty dominate the positive effects of coups staged by pro growth factions, while Alesina and Perotti (Reference Alesina and Perotti1996) submit that “unconstitutional” regime changes such as coups are worse for economic growth than other types of political instability. Londregan and Poole (Reference Londregan and Poole1990) find that coups have no impact on economic growth and hypothesize that this may be because there is a bimodal distribution of coups, some of which enhance growth and others which restrict it.

Most research suggests that “good coups” are not the norm, ^{Footnote 12} but some evidence suggests that they have become more frequent ^{Footnote 13} and could have positive economic effects. Meyersson (Reference Meyersson2016) finds that coup attempts that occur in democratic regimes are associated with negative GDP growth, while those that occur in autocratic regimes are associated with positive GDP growth. In another study examining the same outcome variable as our own—financial returns—Girardi and Bowles (Reference Girardi and Bowles2018) show that Salvador Allende’s socialist government was associated with declines in stock prices while the coup that replaced him boosted them.

Previous research therefore suggests that on average coups should not be expected to lead to increased economic development or positive market returns, but that positive returns associated with “good coups” could occasionally be present. We shed light on this theory as our event study approach also allows us to examine “good coups”—coups that lead to democratization or economic liberalization—separately from others. As we are not limited to a single case study, we can provide both a cross-country and within-case analysis.

Second, it is not necessarily the case that even a sudden, extra-legal, or unexpected regime change will result in higher levels of instability. Not all regime changes are equivalent—resignations, assassinations, and coups are all examples of sudden regime changes and have been used as proxies for instability in previous research. However, resignations often occur when an ineffective leader steps down. By contrast, assassinations are not always related to the effectiveness of a leader, can occur seemingly at random, and the successor to the assassinated leader may be unclear. Coups can occur in the name of democracy or autocracy, and in general signal relative weakness in the current political system. These differences are often overlooked in past empirical strategies, which have proxied for instability in terms of the number of coups, ^{Footnote 14} assassinations or revolutions, ^{Footnote 15} or have combined various metrics into single indices. ^{Footnote 16,Footnote 17} In another cross-country regression study, Jong-A-Pin (Reference Jong-A-Pin2009) disaggregates instability into four main metrics, one of which is regime changes, but does not subset below this metric. By contrast—recognizing the different degrees of instability that these events may reflect—we estimate effects separately for coups, assassinations, resignations, and protests. Our event study approach affords us this freedom without encountering the statistical power issues that would arise in a regression setting.

Third, the theoretical link through which political events are expected to have an impact on daily stock prices is changes in the broader macroeconomic environment that influence the expected stream of future profits of existing publicly-held firms. Examples of expected changes in the macroeconomic environment that can be influenced by sudden political events include: shifts in monetary policy, changes in trade or FDI openness, shifts in regulatory policy, changes in the share of output going to capital, transfer and convertibility restrictions, degree of resource nationalism, and heightened risk of capital expropriation or breach of contracts. Not all shocks to the political system will result in similar shifts in these macroeconomic variables depending on the actors involved, their ideological preferences, and their expected ability to enact policy changes.

Different types of regime changes may therefore have disparate effects on investment, just as they may differentially effect overall economic growth. Taken together, the theory and research above suggests that coups should on average lead to negative returns, but sometimes lead to positive returns when the coup’s instigators are clearly more pro market than the regime they replace. Assassinations should have a neutral or negative impact as they increase uncertainty, but institutional responses to assassinations vary by country. By contrast, resignations may be viewed positively on average, as they typically signal the departure of an ineffective leader and may in fact signal increased stability. Using immediate, day-of shocks to stock prices as our outcome variable, we investigate the hypothesis that investors will positively update their perceptions of future economic performance when a sudden political event clearly shifts a country in a pro-democratic or pro-business direction, and negatively update when an event merely exacerbates instability or uncertainty.

Our empirical findings validate these predictions. We find that the type of political event and its expected impact on economic policy determines the direction of investment. Events expected to lead to more stable governance, economic liberalization, or democratization (such as willful resignations and coups that overthrow protectionist or leftist autocrats) are associated with positive returns, while those that consolidate authoritarian rule (e.g., military coups), exacerbate poor economic policies, or merely increase policy uncertainty (such as assassinations) have the opposite effect. Of course, changes in stock prices reflect investor’s expected changes in the macroeconomic environment, and these expectations may differ from realized future economic performance. Nevertheless, it is clear from our results that even sudden and extra-legal regime changes can be viewed as a “positive” event by investors under the right circumstances.

Data

Financial data are from the Global Financial Data database, which includes the longest available daily time series of stock prices. We collect data on national equity indices and two global equity indices, the S&P/IFC Emerging Market Investable Composite and the Morgan Stanley Capital International (MSCI) World Index. The S&P/IFC index includes securities from emerging markets while the MSCI index includes securities from developed markets only. We collect national stock index data on every country in which there was a coup or coup attempt, an assassination or failed assassination, or a forced resignation and for which daily financial data is available. ^{Footnote 18} The longest available daily time series for these stock indices are listed in Table 1.

Table 1.

List of stock indices

Political data are primarily drawn from the Center for Systemic Peace’s (CSP) Polity IV Coup d’etat dataset and Coup d’etat Events handbook. The Coup d’etat dataset includes the date of (1) successful coups, (2) attempted coups, (3) plotted coups and (4) alleged coup plots. We focus on successful coups because it is difficult to classify failed coups, and the choice of dataset and classification methodology would directly alter our findings. Needler (Reference Needler1966, p. 617) has gone so far as to say that “the categories of coups that were aborted, suppressed, or abandoned melt into each other and into a host of other non-coup phenomena so as to defy accounting,” the Center for Systemic Peace states that it is “confident that [its] list of successful coups is comprehensive” but does not extend this confidence to attempted or failed coups, and Powell and Thyne (Reference Powell and Thyne2011) state that it is “difficult to identify more ambiguous forms of coup activity, such as coup failures, plots, and rumors.” With this caveat in mind, we nevertheless additionally construct a database of failed coups from the Colpus dataset ^{Footnote 19} in order to test for robustness to inclusion of ambiguous coup cases, which can be found in Table A.6 in the appendix.

The Coup d’etat Events handbook also provides a list of (1) auto-coups ^{Footnote 20} , (2) the ouster of leadership by foreign forces, (3) the ouster of leadership by rebel forces, (4) assassinations of the executive and (5) resignations of the executive due to poor performance and/or loss of authority. Daily financial data is available for countries in categories 4 and 5, so we supplement the coups with assassinations and resignations to form a list of “irregular” regime changes. The resignations are those in which the ruling executive was coerced to resign due to poor performance, public discontent and popular demonstrations. Note that the Polity IV definition of “poor performance” is not synonymous with poor economic performance, and in practice the reasons cited for resignation across events are: loss in conflict/war, anti-authoritarian protest, corruption scandals, Supreme Court ruling against unconstitutional actions, contested elections, and abuse of power.

We further supplement the CSP data with leadership data from Archigos Version 4.1, which allows us to identify additional cases in which a “leader lost power through irregular means.” Irregular transfers of power are those in which leaders do not leave office “in a manner prescribed by either explicit rules or established conventions.” Nearly all removals by irregular means result from the threat or use of force (e.g., coups, revolts and assassinations).

A list of the political events in our dataset is shown in Table 2. Coups tend to have the largest impact on the level of democratization as a number of countries have subsequently transitioned from democracies to anocracies or autocracies. On the other hand, neither assassinations or resignations have much of an impact on the level of democratization.

Table 2.

Regime changes

Notes: The Polity score is used to classify political outcomes as follows: autocracy = $ - 10 \le score \le - 6$ , anocracy = $ - 5 \le score \le 5$ , and democracy = $6 \le score \le 10$ .

The discussion above shows that there is considerable debate about classification of regime changes. We recognize that some readers may feel that certain events are missing based on their own substantive knowledge. However, we choose to rely on commonly used third-party classifications in order to minimize the possibility that our results are driven by our own subjective classifications. The one exception is that we separately study the failed coup attempt in Venezuela in April 2002 in which the President of Venezuela, Hugo Chavez, was removed from office for two days because it provides a natural test of the impact of the seemingly successful removal of a left-wing populist with a pro-business regime, and the ensuing reinstatement of a left-wing populist.

To gain perspective on the relationship between irregular regime changes and the stock market, Figure 1 plots the absolute value of daily stock returns averaged across all events. Returns are for 200 trading days before and after regime changes. The absolute value of returns on the event day (trading day 0) are significantly larger than on any other day. In addition, the magnitude of returns begins increasing just before the event day and remains high for a short period after. This suggests that financial volatility increases during the days surrounding regime changes, although it does not provide any evidence on mean returns.

Figure 1.

Absolute value of daily returns.

Impact of Political Instability on Stock Returns tests these results more formally. It will first formally estimate the amount of volatility surrounding all irregular regime changes, then analyze coups, assassinations and resignations separately and determine what impact they have on mean returns.

Estimation strategy

We estimate the effect of irregular regime changes on financial flows using an interrupted time series—also known as event study—approach, and examine the impact of political events on stock returns using daily financial data. Additional explanation of the theoretical logic of why mean daily stock market returns would be influenced by politically unstable events is warranted. Event studies rely on the theoretical assumption that investors anticipate the effect of sudden events (e.g., corporate announcements, regulatory policy changes, political events, etc.) on future profitability of a firm or firms in a market, and importantly, that these anticipated changes in profitability are immediately reflected in stock market prices. The approach therefore implicitly relies on the efficient market hypothesis, or the idea that in an efficient market, stocks instantly incorporate publicly available information into their prices. ^{Footnote 21} The immediate nature of this effect allows us to use the magnitude and direction of sudden fluctuations in granular daily data as a proxy for investors’ expected valuation of the event in terms of future profits. This in turn allows us to examine how global financial actors value different kinds of politically unstable events—for example, pro-vs-anti-business coups.

The focus on immediate fluctuations in daily financial data implies that we are examining the effect of “one shot” regime changes akin to a sudden day-of shock. Our estimates therefore capture the effect of these sudden, unexpected shocks on stock prices on the same day of the shock. Importantly, this implies that we are not estimating the effect of any larger underlying political conditions surrounding the event, but rather only the immediate event itself independent of other surrounding conditions. Further, as this approach provides an estimate at the exact time of the event, confounding events would need to occur on the same day, and do so for a large portion of all of our independently tested events in order to influence our estimates. We recognize that broader definitions of political instability may encompass political volatility over a wider range of time. In an attempt to address this concern, we also examine the impact of periods of protest and failed coup attempts on financial returns in the appendix.

The key identifying assumption behind our interrupted time-series approach is that the events are unpredicted in an immediate temporal sense. In other words, for an event to be plausibly exogenous, the exact timing of the event must be unknown to investors. If information about the event has leaked prior to occurrence of the event, a portion of the change in investment will already be priced into the index during the estimation window.

Volatility

Before examining the magnitude and direction of returns, we first confirm that our sample of events exhibits the increases in volatility suggested by previous literature. Since stock volatility is not directly observable, one must decide how to best estimate volatility. Our estimates are obtained from a generalized autoregressive conditional heteroskedasticity (GARCH) model estimated using 1000 pre-event days, the event day and 1000 post-event days. As in Jensen and Schmith (Reference Jensen and Schmith2005) and Leblang and Mukherjee (Reference Leblang and Mukherjee2005), we use the GARCH (1,1) specification. In particular, for national stock index $i$ ,

$${R_{it}} = {\mu _i} + {\epsilon _{it}},{\rm{\;\;\;\;}}{\epsilon _{it}}\sim{\cal N}\left( {0,\sigma _{it}^2} \right),$$

where ${\mu _i}$ is a constant and,

$$\sigma _{it}^2 = {\gamma _i} + {\alpha _i}\epsilon _{i,t - 1}^2 + {\beta _i}\sigma _{i,t - 1}^2.$$

The key parameter of interest is the conditional variance, $\sigma _{it}^2$ . The one-period-ahead volatility forecasts, ${\sigma _{it}}$ , are larger when $\epsilon _{i,t - 1}^2$ and $\sigma _{i,t - 1}^2$ are larger. In other words, the model predicts that large shocks will be followed by other large shocks. As robustness checks, we also estimate volatility using exponential GARCH, threshold GARCH, fractionally integrated GARCH, and GARCH(1,1) accounting for structural breaks in the time-series data in the appendix.

Abnormal returns

We now turn to the magnitude and direction of the effect of irregular regime changes on stock returns. We follow the standard event study methodology as presented by, among others, MacKinlay (Reference MacKinlay1997) and Campbell, Lo and MacKinlay (Reference Campbell, Lo and MacKinlay1997). Normal performance is measured with a constant mean return model,

(1) $${R_{it}} = {\mu _i} + {\epsilon _{it}},$$

where ${R_{it}}$ is the logged return of national stock index $i$ on trading day $t$ and ${\epsilon _{it}}$ is the error term. We calculate abnormal returns (ARs), in an “event window” surrounding the date of each coup, ${R_{i\tau }} = {R_{i\tau }} - {\hat \mu _i}$ , where $\tau $ is a date in the event window, and ${\hat \mu _i}$ is estimated in an “estimation window” preceding the event window with Equation 1. We use a 41 day event window (i.e., 20 pre-event trading days, the event day, and 20 post-event trading days). The estimation window is the 250 trading days prior to the start of the event window. The abnormal returns are then used to generate cumulative abnormal returns (CARs) between event day ${\tau _1}$ and event day ${\tau _2}$ : $CAR\left( {{\tau _1},{\tau _2}} \right) = \mathop \sum \nolimits_{\tau = {\tau _1}}^{{\tau _2}} A{R_{i\tau }}$ .

A constant mean return model is used instead of a market model in order to maximize the number of observations and because our unit of analysis is country-wide market indices rather than firms. ^{Footnote 22} To address concerns regarding use of a constant mean return model, we also create a synthetic control portfolio for each event and compare observed returns to the synthetic returns as a robustness check.

We define the event date as the first trading day in which the market could have reacted to news of the event. For example, during the October 12, 1999 coup d’etat in Pakistan led by General Pervez Musharraf, the army announced that Prime Minister Nawaz Sharif had been dismissed after market hours at 10:15 pm. We code October 14th, the day in which the market re-opened, as the event day. When events occurred on weekends, we change the event date to the following Monday.

$\left( {0,\tau - 1} \right)$ is used to denote the $\tau $ -day period beginning with the event day and $\left( { - 1,\tau } \right)$ to denote the negative $\tau $ -day period beginning with the day prior to the event day. In other words, for cumulative abnormal returns prior to the event date, we aggregate backwards starting at the day of the event. For example, $CAR( - 1, - 2$ ) is the sum of the abnormal returns on event date $ - 1$ and event day $ - 2$ . We present results for the sum of abnormal returns over the post-event windows of the event date only $(0,0$ ), the event date plus 6 days $(0,6$ ), and the event date plus 19 days $(0,19$ ). In addition, we present results for the pre-event windows $( - 1, - 7$ ) and $( - 1, - 20$ ).

As we hypothesize that different types of regime changes will have disparate effects on markets, we report abnormal returns separately for coups, assassinations and resignations. Standard errors and p-values are calculated using asymptotic t-statistics as in MacKinlay (Reference MacKinlay1997). ^{Footnote 23}

Impact of political instability on stock returns

Volatility

Figure 2 shows the mean volatility ( $\overline {{\sigma _t}} $ ) estimates from the GARCH (1,1) model across all irregular regime changes for 250 trading days prior to and 250 days after each event. ^{Footnote 24} As expected, the volatility estimates stay between a narrow range at nearly all dates except those surrounding the regime change. Volatility appears to increase slowly just before the regime change, albeit not to a degree out of line with previous fluctuations in volatility. This may suggest that investors sometimes have information about the events before they occur. Nonetheless, there is still an enormous volatility jump on the day of the regime change. Volatility then decreases to normal levels within a month of the event.

Figure 2.

Mean of volatility estimates from GARCH(1,1) models.

Abnormal returns

Coups

Table 3 shows abnormal returns for national stock indices both preceding and following coup d’etat. Table 3 contains all coups presented in Table 2 with the exception of the Argentinian coup of March 24, 1976. The March 24, 1976 Argentinian coup is excluded from our analysis because the stock market remained closed from March 24 to April 5, 1976, or a period of twelve days. ^{Footnote 25}

Table 3.

Abnormal returns following coups

Notes: Standard errors are in parentheses. “Days to rebound” is the number of trading days following a negative stock return for the national stock index to return to pre-event level (it is calculated if the price decreases on the event day, not if the event day abnormal return is negative). Returns are inflation adjusted.

The average coup has a −2.1 percent event day AR. Event day ARs for the 1970 coup in Argentina, the 1991 coup in Thailand, the 1992 coup in Peru, and the 1999 coup in Pakistan are all negative and statistically different than zero. Moreover, all of these cases except Thailand have negative post-event CARs and pre-event CARs that are statistically indistinguishable from zero. In all of these cases, the coup either overthrew a democratically elected government or changed governance from one military ruler to another. The initial negative reaction followed by additional post-event negativity is consistent with the expected market reaction from a successful authoritarian coup followed by post-event consolidation of power.

The only events with positive ARs are the 1971 coup in Argentina and the 2002 coup in Nepal. These results provide evidence that coups do not necessarily lead to negative abnormal returns. While the 1971 Argentinian coup did result in another military leader, it did so while calling for free and democratic elections and replaced a government that had adopted extreme protectionist economic policies. In fact, by 1973 Argentina had transitioned to a democracy. ^{Footnote 26} The 2002 coup in Nepal resulted in a monarchical restoration, but occurred after the country’s prime minister postponed general elections, itself a democratically subversive action.

Assassinations

The results in Table 4 are produced from analyses identical to those in Table 3 but for assassinations rather than coups. Like the majority of coups, there is evidence that assassinations decrease stock prices. The mean event day abnormal return is negative and statistically different than zero. However, the result is driven by five events: the shooting of US President William McKinley on September 6, 1901; the assassination of US President John F. Kennedy on November 22, 1963; the assassination of Indian Prime Minister Indira Gandhi on October 31, 1984; the suicide bombing that killed Sri Lankan president Ranasinghe Premadasa on May 1, 1993; and the assassination of Israeli Prime Minister Yitzhak Rabin on the evening of November 4, 1995.

Table 4.

Abnormal returns following assassinations

Notes: Standard errors are in parentheses. “Days to rebound” is the number of trading days following a negative stock return for the national stock index to return to pre-event level (it is calculated if the price decreases on the event day, not if the event day abnormal return is negative). Returns are inflation adjusted.

These results are consistent with our hypothesis that the nature of the political event and its expected impact on policy matters, and that assassinations should have a negative effect as they occur seemingly at random and increase uncertainty. While the mean effect of assassinations is negative, it is smaller in magnitude than for coups. Unlike a coup, an assassination may not necessarily be expected to cause immediate change in economic policy, particularly in the presence of an institutionalized line of succession. As such, we would expect CARs to be negative due to increased instability and uncertainty, but smaller in magnitude to a coup or resignation due to greater expectations of policy inertia.

There is no evidence of post- or pre-event CARs in almost any of the assassinations. This is consistent with expectations as assassinations are typically not predictable. As with coups, the number of days that it took the stock market to rebound to pre-event levels is fairly low. ^{Footnote 27}

Resignations

In contrast to coups and assassinations, abnormal returns following resignations are large and positive (see Table 5). The mean event day abnormal return is over 4 percent and the positive returns are persistent and grow larger over time (mean 20-day CAR $ \approx $ 12 percent). Furthermore, event day ARs are only negative and statistically significant at even the 10 percent level in two out of the fifteen resignations (Pakistan on April 19, 1993 and Tunisia on June 31, 2011).

Table 5.

Abnormal returns following resignations

Notes: Standard errors are in parentheses. “Days to rebound” is the number of trading days following a negative stock return for the national stock index to return to pre-event level (it is calculated if the price decreases on the event day, not if the event day abnormal return is negative). Returns are inflation adjusted.

These results are again consistent with our hypothesis that different events will have disparate effects, and that resignations may lead to positive returns. The positive event day abnormal return following resignations is not surprising as resignations typically occur because of poor performance and/or loss of authority. Among our sample of events, leaders were ousted following loss in conflict/war, anti-authoritarian protest, corruption scandals, Supreme Court ruling against unconstitutional actions, contested elections, and abuse of power.

For example, consider Ferdinand Marcos’ resignation from office as President of the Philippines in February 1986. Prior to his resignation, the Philippine regime was known for rampant corruption, crony capitalism, extreme inequality, high unemployment, failed import substitution industrialization policy, and oligarchic control of the economy. ^{Footnote 28} In fact, the Philippines was the least preferred site for foreign investment amongst East Asian capitalist economies and possessed one of the worst capital investment to economic output ratios in Asia. ^{Footnote 29} Marcos held a snap presidential election on February 7, 1986, in which he declared victory despite overwhelming evidence of electoral fraud. Public protests ensued, and two weeks later the military withdrew its support of the Marcos regime. ^{Footnote 30} Marcos was replaced by his electoral opponent, Corazon Aquino, who had run on a platform of economic liberalization and elimination of crony capitalism. ^{Footnote 31} This event was associated with an approximately 13 percent positive event day AR.

By contrast, the largest negative event in our sample (−3 percent) is the 1993 resignation of President Ghulam Ishaq Khan and Prime Minister Nawaz Sharif in Pakistan. The resignations occurred after months of political infighting when the army demanded the President and Prime Minister resign and call for new elections. An interim prime minister was installed, but uncertainty about Pakistan’s political and economic future remained high prior to the next round of elections.

The resignations studied in this paper are those in which leaders left office because of poor performance, public discontent and popular protests. It is therefore not unreasonable to expect the political actions preceding the resignations to have similarly large effects on financial markets. ^{Footnote 32} To examine this, we explore all resignations that were driven by significant popular demonstrations, riots, non-violent civil resistance and other forms of public discontent in Public Protests in the appendix. ^{Footnote 33} When taking directionality into account, it appears that public protests have no effect on stock returns. However, this occurs because some political movements increase stock prices while others decrease them, and the absolute value of stock returns are approximately 1.5 percent higher during public protests (see Table A.2). ^{Footnote 34}

Exploring possible mechanisms

While markets may generally dislike political instability, the immediate effect of regime changes on markets may not always be unpredictable. For example, investors may generally value democracy if it is perceived to provide stronger property rights and lower susceptibility to capital appropriation. ^{Footnote 35} In addition, when the successor in an irregular regime change is clear, investors may have strong priors about the effect of the new leader on economic and/or market performance. Two possible mechanisms that could be driving the differences may therefore be: (1) whether the regime change is associated with an authoritarian or democratic shift in governance, and (2) whether a new leader is clearly more pro- or anti-business than their predecessor.

We first attempt to explore these mechanisms by aggregating all of the events in our sample by whether they resulted in a shift in an authoritarian or democratic direction. ^{Footnote 36} We find suggestive evidence that regime changes associated with authoritarian shifts are on average perceived negatively by investors (see Table 6), while those that move governance in a democratic direction are on average perceived positively (see Table 7). We refer to this evidence as suggestive despite statistical significance due to the small sample of cases which fit these criteria, particularly with regard to democratic shifts. We observe ten cases associated with an authoritarian shift in governance. Only two of these events result in positive CARs, and neither are significantly different from zero. Five cases resulted in a democratic shift in governance. Three of these five cases result in positive CARs. Of the two negative CARs, one is not significantly different from zero, and the other is associated with a forced market closure that lasted 17 days. However, the majority of the positive returns from democratic regime changes come from the 13 percent positive CAR associated with the resignation of Ferdinand Marcos in the Philippines in 1986.

Table 6.

Abnormal returns following authoritarian regime changes

Notes: Standard errors are in parentheses. “Days to rebound” is the number of trading days following a negative stock return for the national stock index to return to pre-event level (it is calculated if the price decreases on the event day, not if the event day abnormal return is negative). Returns are inflation adjusted.

Table 7.

Abnormal returns following democratic regime changes

Notes: Standard errors are in parentheses. “Days to rebound” is the number of trading days following a negative stock return for the national stock index to return to pre-event level (it is calculated if the price decreases on the event day, not if the event day abnormal return is negative). Returns are inflation adjusted.

A similar analysis is not possible for shifts from pro- or anti-business leaders, as examples of such clear and plausibly exogenous shifts in leader economic ideology do not exist in our sample. ^{Footnote 37} We therefore look outside of the sample above and conduct an in-depth case study of a seemingly pro-business and anti-socialist coup followed by the reinstatement of a socialist government: the 2002 failed coup against Hugo Chavez in Venezuela.

The ultimately failed Venezuelan coup against Hugo Chavez replaced a left-wing populist government with a new pro-business president, and therefore provides a natural test of the effects of both pro-business and anti-business regime changes separately from simple uncertainty because investors reacted to an expected regime change twice: first, when Chavez was ousted, and second, when he was reinstated. On the evening of April 11, 2002, coup plotters removed Chavez from office and later detained him. Pedro Carmona, a Venezuelan economist and business leader, was named the transitional President of Venezuela. Two days later, on April 13, 2002, a popular uprising led to Chavez’s reinstatement as president. This provides an estimate of the market’s valuation of a transition from the Chavez regime to the Carmona regime and its valuation of a transition from the Carmona regime back to the Chavez regime. By extension, it provides an estimate of the impact of a shift from a left-wing populist government to a pro-business regime in an emerging market.

Figure 3 provides graphical evidence on the effect of the coup attempt. The top panel shows CARs for the 10 days prior to and following the event, along with 95% confidence intervals. The daily ARs and corresponding confidence intervals are displayed in the bottom panel. The abnormal return on April 12, the first trading day in which investors could react to the coup, was +10 percent. The market reacted in the opposite direction to Chavez’s reinstatement as president: the abnormal return on the next trading day, April 15, 2002, was −8 percent.

Figure 3.

Abnormal returns surrounding the 2002 Venezuelan coup attempt.

The results in Figure 3 are particularly striking given the discrepancy between the ARs on event days 0 and 1 and all other days. Consistent with our earlier findings that coups tend to have pre- and post-event CARs that are statistically indistinguishable from zero, the only days on which the 2002 Venezuelan coup d’etat attempt ARs are statistically different from zero is on event days 0 and 1 after the coup attempt. The almost 0 percent 10-day CAR preceding the coup makes this an ideal case as it implies that investors were completely unaware of the coup plot, increasing our confidence that the abnormal returns capture the true effect of the Chavez to Carmona and Carmona to Chavez regime changes on stock returns. This failed coup therefore demonstrates a large positive market reaction to the attempted overthrow of a socialist leader, and an equally large negative reaction to his reinstatement. More generally, the large magnitudes and precision of these effects suggest that investors primarily value transition to a pro-business government regime, regardless of how the regime change is achieved.

Robustness

There are some potential concerns with the results in the Coups, Assassinations, and Resignations sections. First, the abnormal returns could have been driven by factors unrelated to the regime changes. Second, the true effects of regime changes on firm value may be underestimated if investors had a-priori information. Third, the reported means are based on small sample sizes so confidence intervals based on normally distributed abnormal returns may be inappropriate. Finally, for coups, it may also be useful to compare successful coups to unsuccessful events (rather than no coup), as the latter may adjust for selection into unstable events.

We explore these concerns in the following ways. First, we reestimate mean CARs on a set of time-shifted placebo dates, with means computed across all events for each type of regime change. We shift event dates surrounding the actual event date backwards and forwards in increments of five days (−20, 15, 10, 5, 0, 5, 10, 15, and 20 days). In addition, we extend the forward shifted event dates to one year (110, 195, 285, and 365 days) to capture dates that are likely to be completely unaffected by the regime change. The general intuition is that we should not observe significant abnormal returns when performing an identical test on dates where no intervention (i.e., a politically unstable event) occurred. Observing such effects would call the research design and modeling assumptions into question, and raise concerns that the abnormal returns were caused by factors other than the regime changes.

Figure 4 compares mean CARs estimated using the actual event date (Figure 4a) to CARs estimated with the event date shifted 1-year (365 days) into the future (Figure 4b). Figure 4a—which reproduces the tabular results presented in the Coups, Assassinations, and Resignations sections—shows that assassinations and coups are associated with negative event day ARs while resignations are associated with positive event day ARs. In contrast, there are no discernible abnormal returns in Figure 4b. The event day ARs are considerably smaller in magnitude and are not statistically different from zero for either coups or resignations. Moreover, while CARs estimated using the actual event date for resignations trend upwards following the event day, there is no consistent trend in the placebo analysis. Figure 4 therefore suggests that the main results are not merely an artifact of the data.

Figure 4.

Mean cumulative abnormal returns by type of regime change.

To ensure that results of the placebo test based on 1-year did not occur by chance, Figure A.14 plots event-day abnormal against the number of days shifted. There are a few instances in which there are statistically significant (at the 5 percent level) ARs in the same direction as those on the actual event day, but they are always smaller in magnitude than the ARs estimated using the actual event date. Most of the statistically significant placebo estimates also occur when dates are shifted within the post-event window (days 0–20), a period during which stock returns remain volatile and, in the case of resignations, there is a consistent upward trend in the CARs. When dates are shifted forward further ( $ \ge $ 110 days), the event day AR is only statistically significant in one case (day 365 for assassinations). Overall, these results reinforce the main results: the ARs on the actual event date capture most of effect of the regime change, although effects can sometimes persist in the short event window following the event date.

These figures also provide evidence on the extent to which regime changes appear to be unexpected. For instance, the CARs prior to assassinations and coups presented in Figure 4a tend to be close to zero, suggesting that investors were unaware that a negative event was likely to occur in the coming days. For resignations, CARs trend downward in the 10 days prior to the event; however, if investors were aware that a resignation were about to occur one would expect the pre-event CARs to be positive given the positive CARs observed in the post-event window. By contrast, a clear break exists between the negative pre-event trend and positive event day jump and post-event trend. For coups, there is suggestive evidence of a negative pre-trend in the run up to the event. However, the overall 7-day and 20-day pre-trends are both insignificant at conventional levels. There is also a much larger jump in negative CARs on the day of the event. This therefore implies that even if investors had some expectation of a negative event occurring, they were unaware of its timing, and at worst a portion of the negative change in investment was already priced into the day-of shift. There is thus little evidence that the CARs in the post-event window are not capturing most of the effect caused by the regime changes.

Second, we create a synthetic control portfolio for each event based on the techniques introduced in Abadie and Gardeazabal (Reference Abadie and Gardeazabal2003) and Abadie, Diamond and Hainmueller (Reference Abadie, Diamond and Hainmueller2010). This approach creates a “synthetic” time series of daily financial returns based on the returns in a basket of countries which did not experience tumultuous political events on the day of the event. We are then able to compare the returns in the country which experienced the unexpected regime change with the actual returns of a reference market—in this case all countries that did not experience an unexpected regime change. ^{Footnote 38} Each country is given a weight which represents its influence in the synthetic control portfolio. The weight is chosen so that the daily returns and the variance of the daily returns of the control portfolio and the event country are most similar in the estimation window. ^{Footnote 39} The set of possible countries in the control portfolio consists of all countries listed in Table 1. In contrast with the large and significant abnormal returns in the actual countries of study, the returns in the synthetic control portfolio are never significantly different from zero for coups, assassinations, or resignations (or when aggregated together).

Non-parametric statistical techniques that are free from distributional assumptions are used to address concerns about inferences from small sample sizes. We employ the sign and the rank tests, which are based on the sign and the rank of the event day ARs respectively. ^{Footnote 40} Both tests are less influenced by departures from normality than statistics based on traditional t-tests such as those reported earlier in this paper. Table 8 compares event day ARs as well as “abnormal absolute returns” between the event country and the synthetic control portfolio using the non-parametric methods discussed above. The “abnormal absolute returns” are abnormal returns for the absolute value of stock returns. This is done to combine events since resignations tend to increase returns while assassinations and coups tend to decrease them. The idea that the absolute value of returns might increase during irregular regime changes is similar to the finding that volatility increases and is consistent with Figure 1.

Table 8.

Non-parametric tests of the impact of regime changes

Notes: Estimates for assassinations do not include the assassination of US president William McKinley in 1901 because no control portfolios are available.

As shown in Table 8, the mean event day abnormal returns for coups, assassinations and resignations are all statistically different from zero at the 1 percent level using the rank test statistic and the abnormal returns for coups and assassinations are significant at at least the 10 percent level using the sign test. In addition, abnormal absolute returns for all events are statistically significant at at least the 5 percent level using both the rank and sign tests. On the other hand, the event day abnormal returns for the control portfolio are never statistically different from zero at even the 10 percent level using the rank or sign tests. Finally, the difference in means between the regime change country and the control portfolio are statistically different from zero for coups (1 percent level), assassinations (10 percent level), resignations (5 percent level), and all events combined (1 percent level) when using two-sided p-values from the Wilcoxon rank test. ^{Footnote 41} In sum, the synthetic control and small sample tests suggest that the main results are not a result of deviations from normality or confounding world events.

Finally, we supplement our list of successful coups with a list of failed coups from the Colpus data set ^{Footnote 42} and show robustness to including and excluding ambiguous cases by reestimating our models while also including failed coups. This analysis demonstrates that including failed coups reduces the magnitude of the negative effect of coups overall, but that the estimates remain negative and statistically significant (see Table A.7). This result is clarified by analyzing failed coups independently of successful coups, which reveals a slightly negative but insignificant effect of failed coups on returns, suggesting that ambiguous cases and plots tend not to have the same level of directed impact on average as successful regime changes (see Table A.6).