2.1. Political shocks

Pastor and Veronesi (Reference Pastor and Veronesi2013) develop a general equilibrium model in which stock prices respond to political signals. They assume that governments care about the welfare of the average investor as well as other objectives. The latter are conceived as political costs, or benefits, associated with any policy change. For example, the government may need to burn political capital to implement a certain policy. Alternatively, a new policy may allow the government to curry favor with a given constituency.

To capture the difficulty that investors face in predicting policy outcomes, the authors assume that political costs/benefits are randomly determined. Therefore, although government actions are not fully predictable, investors can rely on a host of political signals (such as appointments, dismissals, official pronouncements, etc.) to update their beliefs about future policy decisions. Given this learning process, political shocks can be conceived as the differences between the political signals and their expectations. As such, these shocks shape investors' beliefs about which government policy is likely to be adopted in the future. The results Pastor and Veronesi (Reference Pastor and Veronesi2013) indicate that the effect of political shocks on investors' expected returns is bigger when the sensitivity of marginal utility to political costs/benefits is larger, when the uncertainty about the political costs/benefits is larger, as well as when the accuracy of the political signals is larger.

Investors typically rely on publicly available information to forecast what the government will do in the future. Although the universe of “political signals” is enormous, most studies have focused on how markets react to military and political crises (Amihud, and Wohl, Reference Amihud and Wohl2004; Rigobon and Sack, Reference Rigobon and Sack2005; Zussman et al., Reference Zussman, Zussman and Nielsen2008; Berkman et al., Reference Berkman, Jacobsen and Lee2011), terrorist attacks (Zussman and Zussman, Reference Zussman and Zussman2006; Arin et al., Reference Arin, Ciferri and Spagnolo2008; Berrebi and Klor, Reference Berrebi and Klor2008; Brounrn and Derwall, Reference Brounrn and Derwall2010; Chesney et al., Reference Chesney, Reshetar and Karaman2011), as well as revolutions, coup d'états and leaders' deaths while in office (Jones and Olken, Reference Jones and Olken2005; Dube et al., Reference Dube, Kaplan and Naidu2011; Dasgupta and Ziblatt, Reference Dasgupta and Ziblatt2015; Girardi and Bowles, Reference Girardi and Bowles2018; Incerti and Incerti, Reference Incerti and Incerti2019; Baker et al., Reference Baker, Bloom and Terry2020).Footnote ²

In addition, a related stream of literature on political events and financial volatility focuses on “normal” politics in democracies (Bernhard and Leblang, Reference Bernhard and Leblang2006; Leblang and Satyanath, Reference Leblang and Satyanath2008). For example, the effect of national elections on asset prices has received a considerable amount of attention. These studies show that the winning candidate's partisanship (left versus right), as well as the predictability of the electoral outcome can affect stock-market returns (Pantzalis et al., Reference Pantzalis, Stangeland and Turtle2000; Leblang and Mukherjee, Reference Leblang and Mukherjee2004; Bialkowski et al., Reference Bialkowski, Gottschalk and Wisniewski2008; Boutchkova et al., Reference Boutchkova, Doshi, Durnev and Molchanov2012; Sattler, Reference Sattler2013; Kelly et al., Reference Kelly, Pastor and Veronesi2016; Carnahan and Saiegh, Reference Carnahan and Saiegh2020).

Following the existing literature, as well as the analysis in Pastor and Veronesi (Reference Pastor and Veronesi2013), we consider a collection of events that exhibit sufficient variation to identify the effect of political uncertainty and signal precision on market volatility. Specifically, we focus on the following policy-shifting events: changes in a country's economic stewardship; changes in a country's administration; national elections; wars; and terrorist attacks. Some of these events provide more precise political signals than others (e.g., coups versus successions that take place according to constitutional rules); some events are more unpredictable than others (e.g., contested versus uncontested elections); and, some events should affect all firms (i.e., they cannot be diversified away), whereas others are usually a source of idiosyncratic rather than systematic risk (e.g., changes in economic stewardship versus terrorist attacks).

3.1 Data and measurement

We combine daily prices from the Buenos Aires Stock Exchange (Bolsa de Comercio de Buenos Aires) with detailed information on major political shocks in Argentina. Due to data availability on stock-market prices, we focus on the period between 1967 and 2020. We examine eight types of policy-shifting events: changes in the country's economic stewardship (with and without a change in administration); changes in the country's administration (including coup d'états, resignations, and deaths while in office, as well as planned successions); national elections; wars; and terrorist attacks. Our starting point is Keesing's Record of World Events. This source provides a detailed, accurate, and neutral account of all major political events around the world.Footnote ⁵

Our data include the date and type of event. Assigning each event to the appropriate trading day is crucial for our analysis. We rely on media coverage to identify each event's exact announcement date. Our sources are the two major Argentinian newspapers, La Nación, and Clarín, as well as two English-language publications, The New York Times, and the British daily The Times. Table 1 reports the number of events in our sample by event type, as well as the dates of their first and last occurrence. There are 125 distinctive events in our sample.Footnote ⁶

Table 1. List of events

^a There is only one incident involving an international war. We list the beginning of the war as the first occurrence, and its end as the last occurrence.

Note that the numbers in Table 1 do not “add up” because some types of events are not mutually exclusive: for example, we count 62 instances where a person in charge of managing Argentina's economy changed, including 47 cases where these officials changed without a modification in the country's administration. But, we only count 21 cases where a change in the composition of the government took place. The reason is that there are eight cases where the leaders of a new government decided to keep the previous administration's economic stewardship.

We consider that a change in the country's economic stewardship occurs whenever the economic minister (Ministro de Economía) and/or the Central Bank governor leaves office. Although other high-ranking officials also manage important elements of the country's macroeconomic policies, these policy-makers are subordinate to our selected economic stewards, and tend to follow their bosses in leaving office. Therefore, we decided to exclude these underlings from our analysis. We also excluded from the analysis public officials (usually another cabinet member) who may serve as acting, interim, economic ministers for a just a few days until a new one is appointed. Even when we restrict our focus to the country's top economic policy-makers, there is an abundance of cases in our sample. A total of 39 different people occupied the office of economic minister; and 34 different persons served as Central Bank governors in the 53 years between 1967 and 2019.Footnote ⁷

Regarding changes in the country's administration, we consider that an event of this type occurs whenever a new administration is inaugurated, even if the incumbent president was reelected.Footnote ⁸ Not all successions, although, take place in accordance with constitutional rules. In addition to presidential turnover due to elections, other irregular, as well as unexpected, changes in the country's top leadership also occur in our sample. These include coup d'états, a presidential death, as well as a few resignations. Of the 21 changes in the country's administration, nine of them happened in an irregular way during the period under study: one coup d'état against a constitutionally elected civilian administration; three coups within military, unconstitutional, administrations; four resignations; and, one president died in office (Juan Domingo Peron, in 1974). As with economic ministers, we excluded from the analysis public officials who served as acting, interim, presidents for a just a few days until a new one was appointed. In total, 20 different persons occupied the presidency; for an average tenure of roughly 32 months.Footnote ⁹

In the case of national elections, we include both presidential as well as legislative (midterm elections). We also consider the country's peculiar brand of primaries as a significant electoral event. Introduced in 2009, the simultaneous and mandatory open primaries (Primarias Abiertas Simultáneas y Obligatorias, PASO) serve in practice as a nationally representative poll revealing the share of votes that candidates are expected to receive in the actual presidential elections.Footnote ¹⁰ A total of 28 national elections were held in the period between 1967 and 2020. Most of these electoral contests (26) took place, however, after 1982.

Unlike national elections, coding terrorist acts is not a straightforward task. Following Berrebi and Klor (2010), we define a terrorist attack as a premeditated, politically motivated act of violence perpetrated against noncombatant targets by domestic or international clandestine groups, usually intended to influence an audience. In addition, we adopt the following criteria to identify the most salient, publicly known, fatal attacks against non-combatants that occurred on Argentine soil between 1967 and 2020. First, we consider unclassified information based on Argentine media coverage. Second, we only include in our sample terrorist attacks that received front-page coverage by Clarín, one of the country's most prominent newspapers.Footnote ¹¹

Based on these criteria, a total of 29 terrorist acts exist in our sample. Most of them (26) occurred in the 1970s, when the country faced a surge in political violence. The remaining three include the 1989 attack on La Tablada barracks, as well as the 1992 suicide bombing of the Israeli embassy in Buenos Aires, and the 1994 suicide bombing of the Argentine Israelite Mutual Association (Asociación Mutual Israelita Argentina, AMIA) building in Buenos Aires. These terrorist acts varied in the identity of the perpetrators as well as in the number of confirmed fatal casualties (ranging from 1 to 86). Yet, they all had significant shock value, especially those involving prominent political figures, such as the assassinations of former Argentine president Pedro Eugenio Aramburu; former interior minister Arturo Mor Roig; the leader of Argentina's national trade union federation, José Ignacio Rucci; and national legislator Rodolfo Ortega Peña.

Finally, another major and extremely unusual policy-shifting event, the war between Argentina and the United Kingdom over the Falklands/Malvinas Islands, also took place during the period under study. The conflict began on 2 April 1982, when Argentina occupied the islands by surprise, and ended on 14 June 1982, when the commander of the Argentine forces (General Mario Benjamin Menéndez) surrendered to British Major General Jeremy Moore. Unlike the other types of events discussed above, this was a single instance. As such, we give this event a special statistical treatment in the analyses presented below.

To examine how the Argentine stock market reacted to these different types of events, we examine the General Index of the Buenos Aires Stock Exchange (Bolsa de Comercio de Buenos Aires). The index, which represents the prices of the market's most important stocks, is based on the quotes of ordinary shares, and adjusted to reflect the quotes' bids and withdrawals, as well as changes in firms' social capital. It includes 15 sector indices for food, banks, insurance, beverages, trade and imports, construction, finance, raw materials, chemicals, manufacturing, metals, paper and printing, utilities, textiles, and miscellaneous. We construct daily returns using data for the period between 2nd January 1967 and 30th March 2020.Footnote ¹²

As it is standard in finance, we need to control for global conditions. As our benchmark index, we use the S&P 500, one of the best representations of the US stock market. A capitalization-weighted index, the S&P 500 measures the performance of 500 large companies listed on stock exchanges in the United States.Footnote ¹³ Note that the Buenos Aires Stock Exchange index is measured in local consumption units whereas the return on the S&P index is measured in US consumption units. Between 1967 and 2020, large real exchange rate depreciations took place in Argentina. In addition, the Argentine economy has a long history of high inflation. During the period under study, the average annual inflation rate was roughly 190 percent (with a historical maximum of 3079 percent in 1989).

To address these issues, we measure the daily returns of the Buenos Aires Stock Exchange's General Index in local consumption units both in nominal as well as in real terms. We also calculate real returns expressed in US consumption units. Our results are quite similar irrespective of how market returns are measured.

Following the standard practice in the literature, we use logarithmic returns to calculate both indices' daily cumulative returns:

$$r_t = \ln \left({\displaystyle{{I_t} \over {I_{t-1}}}} \right), \;$$

where ln is the natural logarithm operator, and I represents the total return from holding the market index between the periods t and t − 1 (which correspond, in this case, to two consecutive trading days).

Determining how government policies affect aggregate corporate value is a non-trivial task. Some actions affect the value of firms directly, and others indirectly through general equilibrium channels. In addition, most government interventions tend to have distributive consequences across different sectors of the economy. For example, a policy change that increases the relative prices of tradables in terms of non-tradables should raise the value of firms in the former sector and reduce it in the latter. Therefore, as Cruces and Garcia-Cicco (Reference Cruces and Garcia-Cicco2012) note, the actual effect of the policy change on the aggregate stock-market index would depend on its composition.

Our main interest is, thus, not on returns per se, but rather on the second moment of the return distribution. Larger swings in price are riskier than smaller swings in price; and more frequent price changes are riskier than less frequent price changes. It is, thus, common for investors to use variance (or, its square root, the standard deviation) as a measure of how far a security's return deviates from its average during a given period. We model the joint process governing stock-market returns and the variance of those returns using a generalized autoregressive conditional heteroskedasticity (GARCH) framework. This statistical approach seeks to replicate how traders predict the conditional volatility of returns, as it incorporates in each period the most recent forecast error as well the previous period's forecasted variance.

We adopt the following model of the return-generating process:

$$r_t^{{\rm BA}} = \alpha + \beta r_t^{{\rm SP}} + \varepsilon _t$$

$$\varepsilon _t\sim N( 0, \;\;h_t^2 ) $$

$$h_t^2 = \gamma _0 + \gamma _1\varepsilon _{t-1}^2 + \gamma _2h_{t-1}^2 $$

where $r_t^{{\rm BA}}$ and $r_t^{{\rm SP}}$ are vectors of returns for the Buenos Aires index and S&P 500, respectively; $h_t^2$ is the conditional variance of ɛ_t; and γ ₀, γ ₁, and γ ₂ are the coefficients of a GARCH(1,1) specification.Footnote ¹⁴

Following the approach in Dubofsky (Reference Dubofsky1991) and Clayton et al. (Reference Clayton, Hartzell and Rosenberg2005), we use the variance forecast from the GARCH(1,1) model to calculate our key variable of interest, the volatility ratio (VR). We compute it as the natural logarithm of the ratio of the estimated variance at time t to the estimated variance at time t − 1. This measure is equivalent to the percentage increase of the conditional variance relative to the previous day, allowing us to consider how volatility changes, not its level. By looking at changes, not levels, we can control for differences in the conditional variance in different time periods.Footnote ¹⁵

3.2 Identification strategy

Financial markets quickly incorporate new information that is publicly available into asset prices (Fama, Reference Fama1970). That happens because markets are forward-looking; meaning that an asset's price does not represent the value of its past performance. Instead, the assets' current price represents the value investors assign to its future performance. When potential news indicates that a security may be worth more in the future, investors will buy it up to the point where the asset is no longer undervalued. The result is that asset prices immediately reflect current expectations of future value. Therefore, only a truly unexpected event would trigger portfolio rebalancing and lead to an increase in short-term volatility, as investors update their beliefs and search for new asset prices.

In addition, to credibly claim that the market responded to a specific event, it is imperative to show that such reaction is not driven by other confounders. For example, the release of another piece of unexpected news, such as a change in global real interest rates, a devaluation of the domestic currency, or some other market-related information disclosure. The potential presence of contaminating news is particularly problematic when the exact date of the event of interest is uncertain and/or when the window around the event is too broad (Dyckman et al., Reference Dyckman, Philbrick and Stephan1984). But, when the precise timing of the event under consideration is known, the adoption of a very narrow window around its announcement should allow one to minimize the presence of contaminating information.

To make sure that our analysis is free from contaminating information, we adopt the period of three trading days centered on each political shock as our event window. The days are identified as −1, 0, 1, with day zero denoting the trading day immediately before the news of a domestic policy-shifting event breaks. Day one is defined as the first day the Argentine stock market could respond to the news of the event.Footnote ¹⁶ Day minus one is the trading day preceding the day before the event became publicly known. An examination of all the stories that received front-page coverage by the Argentine newspaper Clarín in each of these three-day windows reveals that no other major political, macroeconomic, or stock-market-related news were concurrently released. We are, thus, confident that our event window around each event is sufficiently narrow and free from contaminating information to produce unbiased estimates of the impact of political shocks on financial volatility.

Once we establish its appropriate three-day window, we can calculate the pre-event volatility ratio for each political shock ν in our sample as:

$$VR_\nu ^\kappa = \ln \left({\displaystyle{{h_{t = 0}^2 } \over {h_{t = {-}1}^2 }}} \right); \;$$

and its post-event volatility ratio as:

$$VR_\nu ^\tau = \ln \left({\displaystyle{{h_{t = 1}^2 } \over {h_{t = 0}^2 }}} \right).$$

The pre-event volatility, $VR_\nu ^\kappa , \;\;$ indicates the extent of contaminating information in the three-day window, and/or investors' anticipation of possible news. The post-event volatility, $VR_\nu ^\tau , \;\;$ on the other hand, allows us to gauge the market's ability to quickly and fully absorb the information flowing to it. The following three hypotheses illustrate what our measure means, and how to interpret it:

Under the null hypothesis H ₀, asset prices are not affected by the event ν. According to $H_0^{\prime}$, the market reaction: (a) is driven by extraneous news released on the eve of the event ν; or (b) indicates that the event ν did not take investors by surprise. Therefore, only after observing the combination of $VR_\nu ^\tau$, and $VR_\nu ^\kappa$ postulated in H _A, we can conclusively assert that asset prices responded to the event ν. We empirically test these hypotheses in the following section. But, before reporting on the results of our analyses, we need to discuss our estimation details, as well as our inferential approach.

3.3 Estimation and inference

To examine the effect of political shocks on financial volatility we use an event study approach. Its component steps are well known. Based on an estimation window prior to the analyzed event, the method estimates what the normal stock returns should be at the day of the event and several days prior and after the event (i.e., during the event window).

In studies using daily data, an estimation window going from day −250 to day −30 relative to the event date is often (somewhat arbitrarily) chosen. As Aktas et al. (Reference Aktas, de Bodt and Cousin2007) note, this conventional choice is not free of complications. If unrelated events are present during the chosen estimation window, the estimation of the return-generating process parameters may be biased.Footnote ¹⁷ To address this issue, we adopt the whole period between 1 January 1967 and 30 March 2020 as our estimation window. In conjunction with our focus on changes in volatility rather than levels, and the identification strategy delineated above, using the whole sample to estimate the long-run average daily variance of the Buenos Aires Stock Exchange's General Index reduces the potential effect of contaminating events.

Our sample contains multiple volatility ratios-event type combinations, and it is desirable to aggregate results into a single hypothesis test. Specifically, our goal is to compare how different kinds of policy-shifting shocks affect asset prices, rather than the effect of a single, idiosyncratic, event on stock-market volatility. We group all the events into different categories c ∈ E as outlined in Table 1, except for war, which is a singleton. Then, we calculate an equally weighted average of pre- and post-event volatility for each one of them: $\overline {VR_{\nu = c}^\kappa }$, and $\overline {VR_{\nu = c}^\tau }$.

It is reasonable to expect that aggregating individual events into categories could create cross-sectional variation in each group's volatility ratios (Harrington and Shrider, Reference Harrington and Shrider2007). Therefore, we do not use these averages to test our main hypotheses. Instead, we use a test statistic. In the case of the post-event volatility, it takes the form of:

$$t_c^\tau = \displaystyle{{\overline {VR_{\nu = c}^\tau } -VR_0} \over {s.e.\,\,( {\overline {VR_{\nu = c}^\tau } } ) }}, \;$$

where VR ₀ is the expected volatility ratio under the null. Therefore, for example, in the case of post-event volatility, we want to reject $\overline {VR_{\nu = c}^\tau } = 0$; therefore, we should set VR ₀ = 0. The test statistic for the pre-event volatility, $t_c^\kappa$, can be calculated in a similar way.

The next step is to compute the observed significance levels for these test statistics. If the returns were normally distributed, theoretical values could be used. The non-normality of daily returns, however, is well documented in the literature. As it turns out, the residuals of our return-generating model—from which the values of our key measure of interest are calculated—are not very close to a normal distribution. In addition, the coefficients of our GARCH specification indicate that the variance process mean reverts slowly. Therefore, in our calculations of the volatility ratios, the denominator is larger than the numerator more frequently than the other way around. As a result, despite our use of the log transformation, the distribution of the volatility ratios exhibits considerable skewness.

To address this issue, and make certain that we do not make incorrect inferences, we evaluate the statistical significance of our test statistics following the approach in Białkowski et al. (Reference Bialkowski, Gottschalk and Wisniewski2008). Specifically, we rely on bootstrapped p-values generated using the following iterative procedure:

1. (1) Let n _c be the measure of a set composed by type-c events, where n _c is positive and integer-valued; and $\sum\nolimits_{c = 1}^E {n_c = N}$, where N is the total number of distinctive events in our sample. To simulate the conditions under which the null hypothesis is true, we restrict our attention to observations corresponding to days in which the events listed in Table 1 did not take place. Then, we randomly draw with replacement from this sample n _c observations, to match the number of events in each category c ∈ E.

2. (2) For each category c ∈ E, we compute the average volatility ratio $\overline {VR_c}$ and its associated test statistic, $t_c^\eta$ with VR ₀ = 0, for the randomly drawn sample of n _c, using these observations' volatility ratios calculated from our GARCH(1,1) specification's estimated conditional variances.

3. (3) We repeat steps (1) and (2) 10,000 times and sort the collection of test statistics $t_c^\eta$ in the ascending order to obtain the empirical distribution.

The next step is to evaluate the statistical significance of our test statistics. Suppose that one wants to reject $\overline {VR_{\nu = c}^\kappa } > 0$. In this case, the p-value can be defined as the number of bootstrapped test statistics $t_c^\eta$ whose values exceed the value of the pre-event test statistic, $t_c^\kappa$, divided by the number of replications (in this case, 10,000). Using a two-tailed test, one should reject $\overline {VR_{\nu = c}^\kappa } > 0$ at the 95 percent confidence level if there are 250 or more test statistics $t_c^\eta$ with values that are higher than the value of $t_c^\kappa$. Following this logic, we calculate ${\rm Pr}( {t_c^\eta > t_c^\kappa } )$ and ${\rm Pr}( {t_c^\eta > t_c^\tau } )$ to evaluate the combination of pre- and post-event volatilities postulated in H _A.

4.1 Robustness tests

We generated the results presented in Table 2 using our three-day estimation period as delineated above. Note that we identified day zero as the trading day immediately before an event became publicly known, and day one as the first trading day when the stock market could respond to the news of the event. Pairs of consecutive trading days, however, do not always correspond to two consecutive calendar dates due to weekends and holiday closures. For example, if the announcement of the economic minister's resignation is made public on a Friday after the close of trading, then the first-trading day when the stock market can respond to the news would be Monday, rather than Saturday. Similarly, when news of an event is released right before a holiday closing, the next trading day (i.e., post-holiday) takes place in a non-consecutive calendar date.

Although some market closings are preannounced and coincide with general holidays (such as Christmas, New Year, and other national celebrations), some other ones pertain exclusively to financial institutions. These so-called bank holidays are usually emergency shutdowns declared by the authorities to prevent financial disasters. These closures can be, thus, driven by the same factors that trigger some of the events in our sample. For instance, the Buenos Aires exchange remained closed for a period of 12 days following the 1976 coup (from Wednesday 24th March to Monday 5th April). In addition, it is not uncommon to publicize major policy-shifting events around weekend and holiday closings. The announcement of a change in the country's economic stewardship is a case in point. Of the 47 cases in our sample where such a change occurred within an existing administration, 22 took place after the close of trading on the eve of a weekend/holiday, or during the weekend/holiday day(s).

Considering all the events in our sample, 14 of them took place during a three-day window with non-consecutive calendar days due to holiday closings. In the case of weekends, five government inaugurations, and all the elections in our sample were held on Sundays. Therefore, despite having a three-day window involving non-consecutive days, these events took place immediately before the first calendar day when the stock market could respond to the news of the event. If we exclude these cases, a total of 45 events in our sample took place during a three-day window with non-consecutive calendar days due to weekends. One could still argue that any three-day window that straddles a weekend should be counted as one involving non-consecutive calendar days (i.e., only events that take place on a Tuesday, Wednesday, or Thursday conform to a three-day window with consecutive calendar days). Under this definition, a total of 77 events in our sample possess a three-day window involving non-consecutive days. If we also account for holiday closings, then as many as 91 of the events in our sample took place during three-day windows involving non-consecutive calendar days.

As mentioned above, we reviewed every story that received front-page coverage by the newspaper Clarín in each of our three-day event windows to make sure that our estimates are free of contamination from other shocks. Nonetheless, the existence of non-consecutive calendar days within each window raises another possible identification threat. As French and Roll (Reference French and Roll1986) note, asset returns can display a difference in volatility between trading and non-trading periods due to dissimilarities in the flow of information. In addition to this closed-market effect, the abnormality of post-weekend price returns is also a well-documented fact in the financial literature (Thaler, Reference Thaler1987). It is, thus, possible that our results may be susceptible to these holiday and weekend effects.

To address these issues, we replicate the analysis presented in Table 2 excluding events that took place during a three-day window with non-consecutive calendar days. We present our findings in Table 3, panels A through C. To avoid clutter, only the results for returns expressed in US consumption units are shown. We obtain very similar findings when returns are measured in local consumption units both in nominal as well as in real terms. As mentioned above, some of the political events listed in Table 1 took place mostly around weekends/holidays. To account for this fact, we only report the average pre- and post-event volatility ratios for categories of events with five observations or more.

Table 3. Robustness tests

An examination of the results in Table 3 indicates that irrespective of these alternative specifications of our three-day window and sample sizes, the effect of political shocks on financial volatility remains robust and statistically significant. Therefore, we are confident that our findings are not driven by weekend/holiday effects.

As a final robustness check on our results, we examine the effect of political shocks on financial volatility using intraday price movements. As Andersen and Bollerslev (Reference Andersen and Bollerslev1998) note, a test for volatility responses to specific events should account for both the inter-daily volatility process as well as the intraday pattern. The GARCH(1,1) specification that we use treats volatility as being constant over the trading day. As such, our estimates may constitute a noisy measure of the underlying latent volatility. For example, suppose that asset prices fluctuate wildly after an unexpected policy-shifting event becomes publicly known; but nonetheless, end up close to the previous day's closing price by the end of the trading day. If this were the case, then our inter-day estimate would falsely signal a low volatility state.

The intraday data only cover the period between 14th March 1994 and 30th March 2020. In addition, high-frequency returns, covering a short time span (such as five-minute intervals) are not available. Instead, we only have information on daily opening, closing, high, and low prices. Nevertheless, we can use these intraday price movements to evaluate the robustness of our previous findings. To measure volatility, we use the true range (TR), a well-known technical analysis indicator.Footnote ¹⁹ Following Forman (Reference Forman2006), we normalize the TR to make meaningful comparisons over time. The normalized TR (NTR) at time t is calculated as follows:

$$NTR_t = [ {\max ( {high_t, \;close_{t-1}} ) -{\rm min}( {low_t-close_{t-1}} ) } ] /close_{t-1}.$$

A low NTR value indicates a period with small ranges (quiet days). In contrast, a large NTR value indicates increased volatility in the market. Using these NTR values, we calculate the pre- and post-volatility ratios following the procedures laid out in Section 3.2. We also follow the same steps delineated in Section 3.3, restricting our attention to observations corresponding to days in which the events listed in Table 1 did not take place, and setting VR ₀ = 0 to calculate our t-statistics.

Panel D of Table 3 shows the average values of the pre- and post-event volatility ratios across different political events calculated using intraday price movements. For comparability, only the results for returns expressed in US consumption units are shown. Measuring returns in local consumption units in either nominal or real terms yields very similar results. Because the intraday data do not cover the period between 2nd January 1967 and 13th March 1994, some of the political events listed in Table 1 cannot be examined (e.g., only two irregular government changes, and a single terrorist act occurred during this time frame). Therefore, we only report the average pre- and post-event volatility ratios for categories of events with five observations or more.

The results are analogous to those reported in Table 2. The only difference is that trading days immediately following a change in government do not exhibit any increases in volatility. Note that only two irregular government changes took place during the period 1994 and 2020; therefore, most of the observations in this category of events correspond to planned successions (seven out of nine). As discussed above, investors typically priced in to the market the policy changes associated with this type of government change. Therefore, if anything, the analysis based on the intraday price movements reinforces our main findings regarding the effects political shocks on financial volatility.Footnote ²⁰