The Arab Spring and the international defense industry

It is well documented in the previous literature that capital markets are not immune to geopolitical risks with adverse effects observed at both the aggregate and sector level.Footnote ⁷ Geopolitical risks are defined as the risks associated with wars, terroristic attacks, armed conflicts, violent protests, and political tension between and within states that affect the normal and peaceful course of international relations. Geopolitical risks capture both the risk that these events materialize and the development of new risks that result from the escalation of current events.Footnote ⁸ The economic intuition of geopolitical risks in driving capital markets is that armed or political contests create uncertainty over the economic costs. In particular, geopolitical events tend to serve as a learning mechanism for investors, with them re-assessing the risk component in their portfolios. When traders fear future geopolitical risks, they will become more risk-averse in the aftermath of a security shock and try to escape into less risky alternatives by investing in industries that are stable and strong. As a result, portfolios that are exposed to a great extent to geopolitical events will face a sudden and substantial increase in risk that cannot be diversified immediately. This will result in large movements in the market due to the reshuffling of portfolios and also panic selling by investors in their search for safer financial assets. This behavior will continue until the perception of investors about a stable future is restored again, and the fear and insecurity associated with geopolitical events have disappeared.Footnote ⁹ Thus, a negative collective belief about the possible course of action reduces the aggregate value of the stock market and, in particular, the market value of sectors that are the most vulnerable to the costs of conflict. For instance, the empirical results reported by Tanyeri (Reference Tanyeri2018) indicates that the abnormal stock market returns in several Middle-East countries decrease by around one-percentage-point after each major violent event during the Arab Spring period.Footnote ¹⁰

However, the equity return of defense-related companies might follow the opposite pattern compared to the aggregate stock market. In more detail, the equity returns of non-defense-related firms are determined by the economic uncertainty created by geopolitical events, while defense-related stocks are much more influenced by the expected change in demand for military goods by countries who are highly susceptible to geopolitical risks and events and also from those nations who plan to undertake military action against threats of such risks.Footnote ¹¹ Violent events dampen the activity of most economic sectors, but at the same time, may improve the perspectives of companies in the defense and security industries.Footnote ¹² The unstable climate originated by geopolitical incidents causes investors to expect increased dividends from the defense industry. For instance, the findings of Berrebi and Klor (Reference Berrebi and Klor2005) demonstrate that whereas terrorist attacks have a significant negative impact on the equity return of Israeli nondefense related companies, the effect of terrorism on defense and security-related firms is significantly positive. On a similar note, McDonald and Kendall (Reference McDonald and Kendall1994) find that during periods of extreme violence, including armed conflicts and terrorist attacks, the US defense sector experienced positive, abnormal returns.

Thus, based on the discussion provided above, one can argue that the Arab Spring creates a window of opportunity for international defense companies to expand their exports as the violent protests raise the demand for military goods. Arab governments might buy more military equipment to repress the violent protests or send a signal of power to the protesters and other political challengers in the attempt of the state leader to stay in power. However, when investors behave rationally, they will also take into account that the escalation of the protests and ending up in a violent conflict will increase the likelihood of more rigorous arms trade control policies or could even lead to arms embargoes to particular countries or regions.Footnote ¹³ This latter effect will again hamper military exports or at least make it more expensive as the costs of compliance, transportation, financing, and brokering may rise. Besides, the trade costs of dual-use goods that do not directly fall under the embargo but are also often produced by defense firms will rise.Footnote ¹⁴ Arms export control policies differed substantially among countries during the Arab Spring period. For instance, the United States and the United Kingdom revoked and withdrawn dozens of export licenses for equipment they feared could be used to quash civil unrest or cause a large number of civilian casualties. In contrast, many EU member states have considered the lion's share of arms export license applications for the Middle East as unproblematic. This latter issue suggests that the reaction of investors in the US and EU defense industry to the escalation of the conflict is likely to differ. It is well-known that changes in expectations about the future imposition of sanctions can lead to a change in investor behavior long before the sanction is really implemented or even agreed on. In particular, investors update their expectations about the economy much more frequently during periods of high news coverage than in periods of low news activity.Footnote ¹⁵ Thus, investors—in advance of the expected sanctions or arms embargoes—will rearrange their portfolios based on their own assessment of the likelihood of these measures. In turn, this assessment is built on (i) the severity of the underlying conflict and its consequences for human rights; (ii) the expression of earlier credible threats by EU and US government officials to impose or strengthen arms trade controls; and (iii) the importance of foreign relations with the Arab region. That is, the formal imposition of trade sanctions may come as no surprise to investors and should therefore not lead to any change in asset prices as predicted by the efficient market hypothesis.Footnote ¹⁶

Based on the discussion above, I formulate the two hypotheses that will be tested in the empirical section.

However, the complete effect of the Arab Spring on the stock return of the international defense companies goes beyond the influence on the expected cash flow accruing directly from the sales of military equipment to this region. Stock returns are also affected by systematic movements in the discount rates. A rise in the discount rate leads to a negative impact on the stock return of companies. For instance, there is some empirical evidence indicating that the Arab Spring has created some upward pressure on the oil prices as the uncertainty about the oil production in the Middle-East region rises. This increase is likely to be reflected in the future inflation rate and the production costs of firms.Footnote ¹⁷

Data and methodology

Data on political violence and abnormal equity return

One of the key challenges in analyzing the capital market impact of the violent protests during the Arab Spring is finding a suitable indicator for the intensity of the violent events. In my attempt, I follow the methodology proposed by Dreger et al. (Reference Dreger, Kholodilin, Ulbricht and Fidrmuc2016) and Hoffmann and Neuenkirch (Reference Hoffmann and Neuenkirch2017). This approach builds on the assumption that when a conflict intensifies and becomes more violent (i.e., creates more civil casualties or large-scale destruction), it will be more widely covered in the media. Thus, it is assumed that there is a positive relationship between the number of published news articles and the escalation of the conflict. For this purpose, I use information provided by the LexisNexis Academic database that collects the world's most reputable news. In particular, I consider all news items published in major international newspapers, including some specifically devoted to the defense industry (e.g., Aerospace Daily, Air Force Times, Aviation Week's Homeland Security and Defense, Avionics, Defense Daily, Defense News, Defense and Security, IAC Aerospace and Defense, Inside the Pentagon, etc.) and all the major world financial newspapers (Wall Street Journal, Financial Times, etc.). All news articles are collected containing a joint occurrence of the keywords “Arab Spring,” “Arab winter,” “conflict,” “violence,” “protest,” “demonstrations,” or “riots,” in any combination with the name of the countries that have been affected by the Arab Spring (Syria, Lebanon, Egypt, Libya, Tunisia, Yemen, Bahrain, Algeria, Jordan, Oman, Kuwait, and Morocco). In the next step, I count all entries for each day for a joint occurrence of the keywords between 1 January 2010 and 31 December 2014. The frequency of occurrence can be used as a proxy for the violence intensity of the Arab Spring. Or, put differently, as an indirect measure of the conflict's escalation level from an investor perspective. Since the combinations of search terms do not fully guarantee that a particular news item contains information about the Arab Spring, especially before January 2010, I set the composite news index to zero from 1 January 2010 through 18 December 2010.Footnote ¹⁸ In total, there are more than three thousand articles that were published in my period of analysis.

Figure 1 shows the density of violence-related news items during the Arab Spring (y-axis). The graph shows various peaks in the frequency that coincides with a number of major violent events that caused many casualties. The intensity of newspaper articles is clearly nonlinear distributed. After December 2012, the number of newspaper articles containing information about the Arab Spring drops quickly. This implicitly assumes the intensity of violence also to be lower after this month. Based on this data, I create my violence variable by computing a cumulative indicator of the number of newspaper articles. I apply a log plus one transformation to the indicator variable measuring the intensity of violence as the information content of an individual newspaper article diminishes with the total number.Footnote ¹⁹

Figure 1: Density of Arab Spring newspaper articles

The next step in conducting the event study is to define the sample of defense-related firms. In order to have a representative sample and to avoid any selection bias, I start with the various editions of the “World Top 100 Defense Firms” reported by the Defense News Media Group and the “Top 100 arms-producing and military services companies” published by the Stockholm International Peace Research Institute (SIPRI). These rankings are based on annual defense sales. This data is supplemented using the Google Finance list on defense-related stocks. In order to obtain a coherent sample for my study, I follow Capelle-Blancard Couderc (Reference Capelle-Blancard and Couder2008) by applying some exclusion criteria. As I am only interested in publicly listed companies, I drop out of the sample: (1) fully and partly state-owned firms; (2) family-owned firms; (3) firms with one dominant shareholder or with a low free float rate; and (4) firms with defense revenue below ten percent of their total revenue. In the next step, I collect the daily stock prices of the selected firm. This data is primarily taken from Thomson Datastream, Bloomberg, and Yahoo Finance. After dropping the firms for which no data was available, I was left with the seventy-five firms that constitute my sample. About two-thirds of the considered defense companies have their statutory headquarters located in the United States. In figure 2, I report the price index of the considered defense-related stocks in my period of analysis.Footnote ²⁰ Footnote ²¹

Figure 2: Weighted equity price development defense companies

In order to isolate stock market reactions to the conflict intensity during the Arab Spring, I control for market co-movement and exclude potentially confounding events. In particular, I explore the impact of the Arab Spring on the abnormal return in the international defense industry. In finance, the abnormal return is the difference between the actual return of a security and the expected return. Abnormal returns are typically triggered by events that influence the profitability of a firm, but have not yet been priced by the market. To compute this abnormal return, I apply a two-step approach. In the first step, I follow Liu et al. (Reference Liu, Shu and Wei2017) by estimating an asset pricing model that includes the risk-free rate, the market return, and the so-called Fama-French factors. In particular, the following empirical model is estimated using the OLS-FE estimator.

(1)$$R_{ijt}-R_{\,jt}^F = \beta _0 + \beta _1\lpar {R_{\,jt}^M -R_{\,jt}^F } \rpar + \beta _2SMB_{\,jt}^q + \beta _3HML_{\,jt}^q + \beta _4UMD_{\,jt}^q + e_{ijt}$$

Where R_ijt is the stock market return of defense company i at day tFootnote ²², R^F is the risk-free rate based on the ten year government bond rate in country j; R^M is the market return measured by the aggregate stock market return in the country j where the statutory headquarters of defense firm i is located. The variable SMB (small minus big) is the difference between the daily returns of the small and big firms’ portfolios; HML (high minus low) is the difference between the daily returns of high book-to-market and low book-to-market firms’ portfolios; and UMD (up minus down) is the momentum factor computed as the daily return differential between a portfolio of winners and a portfolio of losers. These latter Fama–French factors are taken for the EU and US market and obtained from the homepage of Kenneth French at Dartmouth College. The detailed results of the estimation of equation (1) are shown in table A2 in the appendix. Next, I use the residual e_ijt as my measure of the abnormal stock return of defense company i. In the second step, I use the abnormal return as my dependent variable in a heterogenous dynamic panel model.

Empirical model

Given the heterogeneity of the defense firms included in this study, the assumption that the data can be pooled might be questioned. The expected short-run benefits of expanding arms exports and long-run costs of more rigorous arms trade regulations are probably not shared evenly among the wide range of defense-related firms competing around the world.Footnote ²³ Standard dynamic panel models can produce inconsistent and misleading estimates if the sample is very heterogeneous.Footnote ²⁴ The econometric literature suggests two approaches to consistently estimate parameters in dynamic panels with considerable heterogeneity. First, under the so-called Mean Group (MG) estimator, an equation for each firm is estimated, and the distribution of the estimated coefficients across groups is examined. To be precise, this estimation method produces consistent estimates of the average of the parameters in heterogeneous panels, provided that group-specific parameters are independently distributed and the covariates are exogenous. However, it has also been shown that MG estimates will be inefficient if parameters are the same across groups, i.e., if the long-run slope homogeneity restriction holds.Footnote ²⁵ In this case, Pesaran et al. (Reference Pesaran, Shin and Smith1999) propose a maximum likelihood-based pooled mean group (PMG) estimator that combines pooling and averaging of the individual regression coefficients. In particular, this estimator allows the short-run coefficients and error variances to differ freely across groups, but the long-run coefficients are constrained to be the same. Not imposing equality of short-run slope coefficients allows the number of lags included to differ across groups.Footnote ²⁶ This approach has strong economic appeal, in that it is likely that investors in the long-run would hold firms to the same standards, but in the short-run, firm and market-specific factors most likely have dominant and differential effects. In particular, in the short-run, the opportunity to expand arms exports might be determined by the internal economies of scale of a particular firm, while in the long-run the arms trade control regulations by the government are enforced at the country level (US) or even on a supranational level (EU), creating a more homogenous effect on defense companies. To already anticipate the main results, for my purposes, the PMG estimator offers the best available compromise in the search for consistency and efficiency.

The relationship between the abnormal equity return of defense-related firms and the violent protests during the Arab Spring is estimated using the following autoregressive distributed lag (ARDL) (p, q₁, . . ., q_n) specification based on a dynamic panel between January 2010 and December 2014. This estimation technique allows separating the short and long-run dynamics and models the long-run convergence process, under the form of unique coefficients for the variables in the long-run and heterogeneous ones in the short-run. To be specific, as the short-run parameters vary across firms, I can use the following dynamic panel representation of the model

(2)$$AR_{it}^{} = {\rm \Phi }_i\bigg({AR_{it-k}^{} -\mu_i + \sum\nolimits_{\matrix{ j \,{ = 1} \cr } }^J {\delta_j{\bf x}_{it-k-1}^j } -\varphi \;protest_t} \bigg) + \sum\nolimits_{\,j = 1}^J {\gamma _{{\bi ji}}\Delta {\bf x}_{it-k-1}^j } + \theta _i{\rm \Delta }protest_t + \varepsilon _{it}$$

Where AR_it is the abnormal stock market return at day t of defense company i, the lagged abnormal return is included to control for autoregressive tendencies and serial correlation. The vector x^j is my set of control variables containing j elements. The terms between brackets represent the long-run relationship, including a firm-specific intercept μ_i and the long-run coefficients on the explanatory variables δ_j. These coefficients are restricted to be the same across firms to satisfy the long-run slope homogeneity condition, while in the short-run they may differ across firms (γ_ji). However, for a long-run relationship to exist, the error correction coefficient Φ_i, representing the speed of adjustment, must be significantly different from zero.

The variable protest represents my proxy composed above for the intensity of the violent protests during the Arab Spring. The average of a rolling window of two days is used for my news indicator since information may be released after the closing of the trading system. The parameters θ_i and φ can be interpreted as, respectively, the short and long-run abnormal return due to the expectations of the intensified violent protests. Under the efficient market hypothesis, security prices are assumed to reflect all public information and to adjust swiftly to the arrival of new public information. My first hypothesis presented above suggests that abnormal stock market return of defense-related firms are expected to increase in the short-run in response to the Arab Spring as the hoarding behavior of the parties involved creates a window of opportunity to expand arms exports (θ_i. > 0). The second hypothesis argues that, over the course of violent protests, investors become more concerned and might expect that foreign governments will take more rigorous arms control measures or even introduce an arms embargo against specific countries. These coercive trade measures will again reduce the trading volume or at least increase the trade costs. As a result, the firm profits will drop, and subsequently, the abnormal return of a defense firm will fall (φ < 0).

In the vector of control variables, I consider variables suggested by previous studies on explaining stock market returns and are required to avoid an omitted variable bias. First, the Arab Spring region is one of the most important regions for oil exports. The political violence might hamper the oil production, which increases the uncertainty about the number of barrels oil that will be traded on the world market in the near future. This would lead to upward pressure on the current oil price.Footnote ²⁷ In turn, this price uncertainty might create some spillover effects and may have important implications for monetary policy. For instance, investors might be concerned that the exchange rate starts to appreciate reducing the competitiveness of the US and EU defense industry. Consequently, there could be a danger that my results do not show the impact of the Arab Spring on the equity return of the international defense industry, but instead reflect the relationship between defense stocks and other financial assets or commodities. To address this issue, I include the daily log-change of the Brent spot oil price and the monthly log-change in the real exchange rate in my analysis. Additionally, the defense sector is a large buyer of base materials such as steel and aluminum. When the price of these raw materials increases, this will raise the production costs of a defense company. As a consequence, the future profits of a defense-firm are likely to fall. To capture this issue, I include the log-change in the monthly Base Metals Price Index reported by the IMF. Furthermore, I include the daily log-change of the trading volume of specific defense stock to capture its degree of liquidity. Finally, I add month and year fixed effects and a day-of-the-week dummy to capture trading day effects.

The PMG estimator involves the existence of a long-run equilibrium relationship. One crucial assumption is that the model cannot be consistently estimated when all the single variables have unit roots or are nonstationary of order one, unless the variables in the long-run relationship are co-integrated. Hence, I need to perform panel unit roots tests for all the variables and test whether a co-integrating equilibrium relationship between variables exists. If the time series mean-variance is constant, then it is said to be stationary over a period. The augmented Dickey-Fuller (ADF) test is used to test for the nonstationarity of the time series.Footnote ²⁸ The results in table A1 in the appendix present the p-values of panel ADF unit root tests for each variable applying the test suggested by Im et al. (Reference Im, Pesaran and Shin2003). The null hypothesis of this test is that all series are nonstationary. An advantage of this specific unit root test is that it allows for heterogeneous short-run dynamics. The lags included in the ADF regressions are selected on the basis of the Akaike Information Criterion (AIC). The results in the upper part of the table suggest to accept the null of unit roots for most of the variables in the level form at the five percent significance level, whereas the variables are mainly stationary in the first-difference form.

In the next step, I check for cointegration using the empirical tests suggested by Pedroni (Reference Pedroni1999), testing the null of no co-integration without imposing homogeneity of the co-integrating vector. The results are reported in the lower part of table A1 in the appendix. My findings suggest that the null hypothesis of no co-integration can be rejected most of the time. Hence, it may imply that there is a long-run steady, state relationship between all variables in my model after allowing for a firm-specific effect. Thus, PMG modeling is likely to be appropriate to use for my purpose.

Empirical findings

Baseline results

This section presents the estimation results on the relationship between the violent protests during the Arab Spring and the abnormal equity return of international defense companies. The optimal number of lags for each variable is determined using the Schwarz Bayesian Information Criterium (SBC).Footnote ²⁹ To obtain robust standard errors, I use the bootstrap estimator with one thousand replicators of the Newton-Raphson optimization algorithm to maximize the likelihood function. Using the bootstrap procedure reduces the error-in-error problem, as my dependent variable is based on the residual from a previous regression.

In column (1) of table 1, I report my baseline estimation results. To test for long-run homogeneity, I use the Hausman test based on the null of equivalence between the PMG and MG estimation.Footnote ³⁰ Rejecting the null implies rejecting the homogeneity of the cross-section's long-run coefficients.Footnote ³¹ The joint Hausman test statistic accepts the long-run slope homogeneity restriction at the 5 percent level. This implies that the consistency of the MG estimate is not warranted, and so the PMG estimate should be preferred.Footnote ³² The consistency and efficiency of the PMG estimates rely on two further conditions. First, the intensity of the violent protest should be identified as exogenous for international defense firms. This criterion is satisfied by arguing that individual defense firms are not able to influence the intensity and escalation of the violent protests and take it as given. The second criterion is that the residuals are serially uncorrelated. To capture this issue, I include firm-specific effects by allowing for an intercept for each firm. Besides, to control for cross-firm common factors, I demean the data on the explanatory variables using the corresponding cross-sectional means.

Table 1: Abnormal stock market return and the Arab Spring

Note: **/* Indicating significance levels of respectively 5 and 10 percent. The table shows short and long run results using the Pooled Mean Group estimator. Bootstrapped standard errors are shown between brackets. Estimated including the controls outlined in the main text.

Based on the PMG estimation results, I can conclude that the short- and long-run response of investors in the defense industry to the intensity of the violent protests during the Arab Spring differs. On the one hand, I find a significantly positive short-run effect of the number of newspaper articles containing information about the Arab Spring on the abnormal stock return of EU and US defense companies, but on the other hand, the abnormal return of these firms drop in the long-run at common confidence levels after the publication of protest-related news. One possible explanation is that investors recognized the intensification of the Arab Spring protests as a window of opportunity for defense firms in the short-run to expand exports. However, in the long-run, when the protests were escalating, investors became more concerned that the US and EU governments would start to strengthen their export control policies or even impose a complete ban on arms transfers to the Arab region. Both policies would have dampened arms exports again.

To provide an approximation of the overall impact of violence-related news items on the considered defense-related stocks, I multiply the cumulative abnormal return of the escalation indicator by the coefficients found in column 1. In the short-run, the Arab Spring increases the abnormal stock market return of a defense firm by 4.9 percentage points, while in the long-run, the abnormal equity returns drop by 2.7 percentage points. This latter finding indicates that the long-run effect compensates for a large part of the short-run positive return. The significant error correction term indicates that approximately 2 percent of the adjustment to the steady-state takes place each day. This means that about half of the news effect would disappear within a week. Thus, the abnormal equity return of the considered defense companies adjusts fairly quickly to its equilibrium level. This demonstrates that defense-related stocks have a rather efficient price discovery since most information is taken into account rather fast. To conclude, these findings provide some empirical support for both hypotheses outlined above.

The results found so far indicate that there is only a small economic effect of the Arab Spring on the stock market return of US and EU defense firms. One possible explanation is that the exposure to the Arab Spring of these firms is not immediatly clear. On the one hand, although the United States and European Union are important arms suppliers for the Arab region, the value of the arms trade to the countries that are directly subject to the Arab Spring is remarkably lower. Only about 15 percent of the total value of the military trade to these countries in the five years preceding the start of the Arab Spring comes from the US or an EU country. On the other hand, this latter figure is based on official statistics. It is well documented in the existing literature that during armed conflicts, the volume of re-exports and illicit transfer of arms increases substantially. This is especially the case when military sactions are imposed and firms try to avoid these restrictions as arms embargoes raise the price of military equipment on the black market in target countries.Footnote ³³ For instance, DellaVigna and La Ferrara (Reference DellaVigna and La Ferrara2010) find that the intensity of a conflict during an embargo period creates an upward pressure on the stock prices of weapon-making companies as many of them are trading illegally and violating the embargo. All of these practices are not directly reflected in the official figures reported by the defense firms. Besides, there might exist a contagious effect. When a particular Arab country is not directly subject to the Arab Spring, it may still influence the stock market return of a defense company indirectly, as investors may believe that in the near future, the civil protests will spread to other countries. As a reaction, investors might anticipate these expectations in advance.

To explore whether the reaction of investors in the international defense industry relies on the direct exposure to the Arab Spring, I include an interaction term between my violent protest measure and an indicator capturing the exposure of the defense sector of a country to the Arab Spring region. This country-specific exposure is based on the value of arms trade between country j and the countries that are subject to the Arab Spring as a share of the total value of the arms trade of a country j.Footnote ³⁴ This ratio is calculated as the average value over the five years preceding the Arab Spring to overcome any simultaneity concerns with my violent protest measure and to smooth out extreme values. The results in column 2 of table 1 again indicate that there is a general significant, positive short-run effect of the Arab Spring on the abnormal return, while in the long-run, the abnormal stock return drops.Footnote ³⁵ However, the interaction term is both in the short and long run statistically insignificant. This finding implies that the reaction of investors is not immediately based on the direct exposure to the Arab Spring, but relies probably more on the expections that the civil protests will spread to other countries in the region. An alternative explanation for this latter finding is that investors are maybe less well informed about the actual exposure of the EU and the US defense sector to the Arab Spring and assume that all arms supplying countries are equally exposed.

Firm-specific effects

It is highly questionable whether the Arab Spring has the same impact on the abnormal equity return of the broad range of defense-related firms competing around the world. For instance, the new trade theory stresses the importance of firms rather than sectors in understanding the challenges countries face in international trade.Footnote ³⁶ This latter argument suggests that the threats and opportunities offered by the Arab Spring rely, to a certain extent, on firm-specific characteristics. Consequently, the Arab Spring might have a significant impact on the abnormal equity return of one firm (or group of firms) and is insignificant in another. Thus, an important question to which I will turn now is whether firm-specific characteristics affect the impact of the violent protests during the Arab Spring on abnormal equity returns. In more detail, I estimate the following model.

(3)$$ \eqalign{ AR_{it}^{} = & {\rm \Phi }_i\bigg({AR_{it-k}^{} -\mu_i-\sum\nolimits_{\matrix{ j \,{ = 1} \cr } }^J {\delta_j{\bf x}_{it-k-1}^j } -\sum\nolimits_{\matrix{ m \,{ = 1} \cr } }^M {\varsigma_m{\bf Z}_{it-k-1}^m } } \cr& { -\varphi \;protest_{it} -p\lpar protest_t \times \sum\nolimits_{\matrix{ m \,{ = 1} \cr } }^M {{\bf Z}_{it-k-1}^m \rpar } } \right) + \sum\nolimits_{\,j \,= 1}^J {\gamma _{{\bi ji}}\Delta {\bf x}_{it-k-1}^j } \cr&+ \sum\nolimits_{\matrix{ m \, { = 1} \cr } }^M {\psi _m\Delta {\bf Z}_{it-k-1}^m } + \theta _i{\rm \Delta }protest_t + \nu _i\lpar protest_t \times \sum\nolimits_{\matrix{ m \,{ = 1} \cr } }^M {\Delta {\bf Z}_{it-k-1}^m } \bigg ) + \varepsilon _{it}$$

Where z^m is a vector containing m firm-specific characteristics represented by a series of dummies. The other variables have the same meaning as in equation (2). Moreover, I can test whether the impact of the violent protests differs between firms using the following long-run (LR) and short-run (SR) marginal effect.

(4)$${\bi LR}\colon {\rm \;}\displaystyle{{\partial AR} \over {\partial protest}} = \varphi + \rho {\bi z}_{}^{\bi m} \comma \;\,{\bi SR}\colon \displaystyle{{\partial AR} \over {\partial {\rm \Delta }protest}} = \theta _i + v_i{\bi z}_{}^{\bi m} $$

In the remaining of table 1, I report the regression results on the conditional impact of the violent protests on the abnormal equity return in the defense industry. First, the industrial base between the European Union and the United States substantially differs. The European defense industry is highly fragmented across countries and relies, to a great extent, on national protection, while the US industrial base is regarded to be much more competitive and dominated by economies of scale. Additionally, the stringency, implementation, compliance, and enforcement of export control policies differ significantly between the European Union and the United States. As already mentioned above, during the Arab Spring period, the United States revoked much of its export licenses to this region, while EU based firms were still able to export to the affected Arab countries. To explore whether the impact of news items containing information about violent protests on the abnormal return diverge between US and EU defense firms, I create an interaction term between the violence-related news indicator and a dummy taking the value one when a particular defense company has its statutory headquarters located in the United States. The results in column 3 of table 1 indicate that the magnitude of both the short- and long-run effects is significantly larger for US public listed defense companies. This finding suggests that investors in US firms reap more benefits in the short-run, when the intensity of the Arab Spring increased, but in the long-run, traders also expect that the US government will implement more stringent arms trade policies compared to other countries. One possible explanation for the difference in the magnitude of the effect is that US firms are likely to produce more strategic goods that can directly be used to repress the violent protests. This kind of equipment is likely to be most affected by the window of opportunity to expand exports as well as more subject to arms trade control. Alternatively, due to the political and legislative procedure and the national interest of the individual EU countries, it is much more difficult to reach a political agreement on common foreign policies such as the imposition of arms embargoes.

To elaborate some further on whether the impact of the Arab Spring on the abnormal equity return of defense companies relies on the type of goods a firm manufactures, I perform two sensitivity tests. First, I add an interaction term between the violence-news indicator and a dummy variable taking the value to one, when at least half of the firm revenues can be contributed to the sales of only military goods or services (based on information taken from Defense News and SIPRI).Footnote ³⁷ The results in column 4 indicate that firms that produce predominantly military goods benefit more from the hoarding effect in the short run compared to firms that produce mainly dual-use or civil goods. One explanation is that these former firms produce equipment that can readily be used for repressive purposes or within the escalation phase. In turn, in the long-run, the downward pressure on the equity prices caused by the violent protests is insignificant for firms that produce goods that also have a civil purpose as they are perhaps better able to diversify their business activities and effectively mitigate the adverse effects of arms embargoes as it is expected that only a small part of their products will be subject to an embargo. This view is strengthened some further when I apply a more detailed split on the types of military goods a firm produces by adding interactions with dummies that capture different product categories: (i) aircraft, ships, and vehicles; (ii) service and maintenance; (iii) electronics and communication; (iv) artillery and missiles; and (v) other strategic military goods (reference category). The information on this categorization is taken from the Defense Top 100, and a firm can appear in multiple categories as it produces different goods and services. The results in column 5 of table 1 indicate that especially the abnormal return of companies that produce goods that fall in the categories “aircraft, ships, and vehicles” or “artillery and missiles” are influenced by the violent protests. On the one hand, firms producing these types of military equipment benefit the most in the short-run from expanding military demand as the use of these types of goods can immediately be used for security purposes or provide a strong signal of power. On the other hand, in the long-run, firms manufacturing these kinds of goods suffer from the expected risk to be subject to coercive trade measures as these types of military goods may cause many civilian causalities. One important note regarding these latter results is that the sample size is reduced substantially due to the lack of detailed information about the military goods a particular firm produces. As a result, there might be a sample selection bias present as this data is only readily available for the largest defense companies.

The strength of the relationship between the intensity of violence and abnormal equity return might also alter during the course of the Arab Spring. The sample period used in my analysis covers the long period from the start of the Arab Spring in December 2010 to the end of April 2013, when the fragile peace was restored in many countries again. Splitting the period establishes a more consistent identification of the impact of the protests at different points in time, depending on the violence-intensity across the periods. Figure 1 clearly shows that there is a nonlinear time trend in the number of news articles containing information about the Arab Spring. Based on this figure, one can argue that the degree of violence peaks around the beginning of January 2012. To capture this issue in more detail, a post–major violence period dummy is created, taking the value one on the days after 1 January 2012, and zero before this date. I add this dummy variable in my empirical specification, together with the corresponding interaction term with my news indicator. The results in column 6 of table 1 demonstrate that news items published around the inception of the Arab Spring have a significantly larger positive effect in the short-run on the abnormal equity return. In contrast, the publication date leaves the long-run effect almost unaffected. One explanation is that after January 2012, the attention of investors declines as the media coverage is falling, or that the news content of the articles has become lower.

One alternative explanation of the latter finding is that the vast majority of the major violent incidents took place during the first half of my period of research. To determine whether my results are driven by the media coverage around these extreme violent events, I use the timelines prepared by two publications: Al-Jazeera and The Guardian. Combining these timelines leaves me with more than 150 significant incident days during the Arab Spring period. In the next step, I have classified the events in three broad groups and created count variables capturing (i) major (violent) protests and riots, (ii) political changes including government and regime change, and (iii) terror events, like suicide attacks and terrorist acts.Footnote ³⁸ To control for these major events, I add the count variables together with the interaction terms and the news indicator into the empirical estimation. Some of the events occur in tandem, for instance, major violent protests sometimes coincide with terror attacks. Simultaneous inclusion of the different events allows for isolation of the effects of each event. The results in column 7 of table 1 indicate, not surprisingly, that a lot of the variation in the abnormal return caused by the Arab Spring is explained by a number of major events. However, even after controlling for these events, the news indicator remains statistically significant at common confidence levels.

So far, I have explored the impact of the Arab Spring on the abnormal equity return of defense companies. Given the relative insignificance of the share of defense firms in these aggregate stock indices, it is expected that the Arab Spring should not affect these indices. However, when the stock market return is affected, focusing only on the abnormal return would underestimate the true effect due to endogeneity issues. To formally test the validity of the assumption underlying the previous results, I use the aggregate return on four major stock exchanges as my dependent variable (US S&P500; Germany DAX100; France CAC40; UK FTSE100). In table 2, I report the results of this placebo test. My expectations are mainly confirmed since, for any of the aggregate indices, there is a significant effect of my violence news indicator.

Table 2: Stock market return and idiosyncratic risk

Note: **/* Indicating significance levels of respectively 5 and 10 percent. The table shows short and long run results using the Pooled Mean Group estimator. Bootstrapped standard errors are shown between brackets. Estimated including the controls outlined in the main text.