I. INTRODUCTION

Terrorism has had a lasting impact on economies and financial markets. The literature on this is rich (see, inter alia, Arin, Ciferri, and Spagnolo 2008; Brounen and Derwall 2010; Charles and Darne 2006; Chen and Siems 2004; Chesney, Reshetar, and Karaman 2011; Meierrieks and Gries 2013; Ratcliffe and Scholder 2015; Shahbaz 2013; Shahbaz et al. 2013). (1) These studies explore the effect of terrorism on economic growth and stock market performance, and the key message emanating from these studies is that terrorism negatively impacts economic growth and stock market performance. Surprisingly, nothing is known about how, if at all, terrorism affects exchange rates. The current study is a response to this research gap. We test, using intraday (10-minute) exchange rate data for a large number of countries, exactly how country-specific terrorist attacks affect exchange rate returns. Our study reveals three things previously unknown about the effect of terrorist attacks on exchange rate behavior. First, we show that terrorist attacks lead to both appreciation and depreciation of currencies. Second, we show that as information on terrorist attacks becomes stale, its effect on exchange rate behavior weakens but persists, suggesting that the bulk of the effect of terrorist attacks is felt on the day of attacks. Third, we show, consistent with exchange rate theories, that for some countries the effect of terrorist attacks on exchange rate reverses with time while for other currencies the effect is persistent. On the whole, these results imply that terrorist attacks influence exchange rate behavior.

Our approaches to addressing the proposed research question follows two steps. In the first step, we construct a unique data set of 10-minute exchange rate returns and terrorist attacks for 21 countries. In the second step, we run time-series regression models to test the hypothesis that terrorist attacks influence exchange rate returns.

Our approaches and findings add to the literature in two ways. First, there is a literature that models the high-frequency response of exchange rates to macroeconomic announcements. These studies show that: (a) unexpectedly strong macroeconomic announcements lead to appreciation of the U.S. dollar in the short term (see Faust et al. 2007), (b) two-thirds of the total effect of macroeconomic news on the DM/$ exchange rate are transmitted via order flow (Evans and Lyons 2008), and (c) macroeconomic announcements influence exchange rate volatility (Andersen and Bollerslev 1998; Ederington and Lee 1993; Melvin and Yin 2000). There is one thing common with this literature and our study--most of these studies, like our own, are based on intraday exchange rate behavior. However, there is also a key difference. This literature almost exclusively focuses on the effects of macroeconomic announcements on exchange rate behavior. We introduce into this story a different type of unexpected event, namely terrorism, and propose the hypothesis that terrorism as an unexpected event moves exchange rate returns. Two theories motivate our research question. The first one relates to the overshooting hypothesis which owes to the work of Dornbusch (1976). The tenet of this hypothesis has roots in the idea that an unexpected monetary contraction (which terrorist attacks lead to) paves the way for an immediate appreciation of the currency. Over time, though, the currency is expected to depreciate, thus converging to its long-run equilibrium. This type of behavior is also consistent with the uncovered interest rate parity. The key point here is that the instantaneous response of the exchange rate to the terrorist attack will be greater than its long-run response, thus mimicking an exchange rate overshooting type behavior. The second theory is motivated by the exchange rate undershooting hypothesis also known as the bandwagon effect, which has the opposite effects to the overshooting hypothesis (see, inter alia, Dornbusch 1976; Frenkel and Rodriguez 1982; Papell 1985). In this case, a monetary contraction due to bad news instigates currency appreciation, the magnitude of which is less than the currency's long-term appreciation.

Second, our study connects with studies that show that terrorism negatively affects stock market performance (see, inter alia, Arin, Ciferri, and Spagnolo 2008; Brounen and Derwall 2010; Chen and Siems 2004; Chesney, Reshetar, and Karaman 2011). None of these studies shows whether the effect of terrorist attacks goes beyond stock markets to the exchange rate market. Our study by showing that terrorism affects exchange rate behavior complements the extant literature on terrorism's effect on the stock market and, as a result, joins this group of studies in establishing a strong empirical relation between terrorism and financial markets more broadly.

We organize the remainder of the paper as follows. In the next section, we discuss the data set. The main findings are presented in Section III, and the final section summarizes the key messages emerging from our paper.

II. OVERVIEW OF DATA

We have two sets of data. The first data set is on the exchange rate. We have 10-minute exchange rate data for 21 countries, namely, Australia (AUD), Canada (CAD), Chile (CLP), China (CNY), Colombia (COP), Croatia (HRK), India (INR), Indonesia (IDR), Israel (ILS), Japan (JPY), Jordan (JOD), Mexico (MXN), Pakistan (PKR), the Philippines (PHP), South Africa (ZAR), Sri Lanka (LKR), Switzerland (CHF), Thailand (THB), Tunisia (TND), Turkey (TRY), and the United Kingdom (GBP). For each of these countries, the currency is quoted (a direct quote) vis-a-vis the U.S. dollar (USD), such that an increase (decrease) in the rate reflects an appreciation (depreciation) of the local currency against the USD. The data are sourced from Thomson Reuters Tick History and cover the period from January 1, 1996 (12:00 a.m.--local time) to December 31, 2014 (11:50p.m.--local time). In terms of the sample size, we have close to a million observations per country. (2)

We calculate returns (R,) of these currencies vis-a-vis the USD using the mid-quote price at the beginning and end of each 10-minute frequency: E[P.sub.t] = ([P.sup.OB.sub.t] + [P.sup.OA.sub.t] + [P.sup.CB.sub.t] + [P.sup.CA.sub.t])/4, where E[P.sub.t] is the estimated mid-quote price at time t and [P.sup.OB.sub.t] [P.sup.OA.sub.t][P.sup.CB.sub.t], and [P.sup.CA.sub.t] are the prices at the open bid, open ask, close bid, and close ask, respectively. Consequently, the return is: [R.sub.t] = log(E[P.sub.t]/E[P.sub.t-1]) x 100.

In this way, we obtain 10-minute exchange rate returns, which are plotted in Figure 1. (3) A summary of these data appears in Table 1. Several interesting statistical features are reflected in these summarized outputs. Reading Figure 1, for example, suggests that the 10-minute exchange rate behavior is heterogeneous: (a) some currencies (AUD, CAD, CHF, CNY, GBP, JOD, and JPY) are relatively stable whereas others are extremely volatile (COP, PHP, PKR, TND, TRY, and ZAR); (b) some currencies experience strong patterns of volatility clustering (IDR, PHP, PKR, THB, and TRY); and (c) many cases of structural breaks are evident in most currencies.

Over the 1996-2014 sample period, we notice that 17 of 21 currencies had depreciated against the USD; the exceptions are AUD, CAD, CHF, and CNY, which apparently, as depicted in Figure 1, were among the most stable currencies. The largest annualized depreciation is experienced by the Turkish Lira (13.94%), followed by the ZAR (6.31%). Another three currencies (COP, IDR, and PKR) depreciated by at least 4% per annum against the USD over the sample period. The most volatile currencies, based on the standard deviation of exchange rate returns, turn out to be THB, IDR, and TND. In 9 of 21 currencies, skewness is positive, suggesting that the chances of further appreciation of these currencies are high. For the majority of currencies, however, skewness is negative, implying that chances of further depreciation are high. The large kurtosis statistic is symptomatic of fat tails--a feature of high frequency exchange rate returns (see Westerfield 1977). The Jarque-Bera test of the null hypothesis of a normal distribution is strongly rejected for all exchange rate returns; the augmented Dickey and Fuller (1979) test, examining the null hypothesis of a unit root, reveals that all exchange rate returns are stationary.

A test of heteroskedasticity is implemented on residuals from a 12th order autoregressive model of exchange rate returns. Specifically, we utilize the ARCH Lagrange multiplier test which has the null hypothesis that there is no ARCH. The results reported in the last column of Table 1 reveal strong evidence that exchange rate returns are characterized by heteroscedasticity; the null hypothesis is rejected with a p value close to zero. The first-order autoregressive coefficient, which we use to judge persistency in returns, is reported in the second last column. Mixed evidence of persistency, which falls in the -0.39% to 0.38% range, is discovered. The implication is that past returns help explain current period returns for most countries. More specifically, we discover that for 16 of 21 currencies at least 10% of current period returns are explained by the previous period's returns.

Next is the terrorism data. The data on the number of terrorist attacks begin in 1994 and are available on a consistent basis for only 40 countries. We obtain these data from the Global Terrorism Database. (4) However, because high frequency exchange rate data are unavailable for many countries, when these data are matched with countries for which terrorist attack data are available we end up with only 21 countries with time-series data from 1996 to 2014. Since terrorist attack data are daily and not all days have terrorist attacks, on the day of the attack for every 10-minute frequency we repeat the number of attacks. We also create a dummy variable that takes a value of one in every 10-minute interval on the day of a terrorist attack and zero for every 10-minute interval on the day when there is no terrorist attack.

A summary of the data on the number of terrorist attacks is presented in Figure 2. All summary information is obtained over the sample period 1996-2014. Panel A plots the total number of attacks by country which sees Pakistan lead the list of 21 countries with a total of 10,134 attacks, followed by India (6,576 attacks), the Philippines (3,032 attacks), Colombia (2,915 attacks), and Thailand (2,890 attacks). The countries with the most number of attacks naturally have the highest average--thus, Pakistan has the highest average followed by India. Panel B plots the mean number of daily terrorist attacks by country. Pakistan records the highest number of daily attacks, followed by India, Colombia, the Philippines, and Thailand.

III. MAIN RESULTS

This section reports and discusses the main results that are in response to our proposed hypothesis: that terrorist attacks influence exchange rate behavior. The results are divided into four parts. In the first part, we provide a cursory (univariate) analysis of the behavior of exchange rate returns in response to terrorist attacks. The way we achieve this is through averaging exchange rate returns on days of terrorist attacks and comparing them with those on nonterrorist attack days. In the second part of our analysis, we estimate time-series regression models aimed at unraveling the contemporaneous effect of terrorist attacks on exchange rate returns followed by a test of whether terrorist attacks can actually predict exchange rate returns. In the third part of our analysis, we test and identify the time period over which the effect of terrorist attacks on exchange rate returns lasts. The final part of the article undertakes additional analysis aimed at (a) establishing the robustness of our findings and (b) testing if data frequency influences the results.

A. Univariate Analysis

Table 2 documents mean exchange rate returns on terrorist attack days (column 2) and nonterrorist attack days (column 3) together with the standard deviation of returns. Three main messages are contained here. First, for 19 of 21 currencies the mean returns are higher (in absolute values) on terrorist attack days compared to nonterrorist attack days. The exceptions are IDR and PHP. Second, the largest difference between terrorist and nonterrorist attack day mean returns is found for CHF, HRK, and ZAR.

Third, 18 of 21 (16/21) currencies experienced a depreciation in terrorist (nonterrorist) attack periods; however, the magnitude of depreciation is much stronger in the terrorist attack period for around 14 of those currencies. Three currencies (GBP, CHF, and CAD) experienced a depreciation in the terrorist attack period and an appreciation in the nonterrorist attack period whereas AUD and CNY experienced appreciation in both time periods.

On the whole, the main implication emanating from these univariate statistics on mean exchange rate returns on terrorist and nonterrorist attack days is that there is a strong pattern in mean returns that seems to be influenced by terrorist attacks. We explore this empirical possibility further through regression analysis next.

B. Regression Analysis

This section explores the contemporaneous effect of terrorist attacks (TERt) on exchange rate returns (E[R.sub.t]) based on the following GARCH(1.1) model with a t distribution (5):

(1) E[R.sub.t] = [alpha] + [[beta].sub.1][TER.sub.t] + [[beta].sub.2]E[R.sub.t-1] + [[beta].sub.3]E[R.sup.2.sub.t] + [[epsilon].sub.t].

We define the conditional variance of return as: [h.sub.t] = Var (E[R.sub.t]|[[OMEGA].sub.t-1]) = E [[(E[R.sub.t] - [[beta].sub.1][TER.sub.t] - [[beta].sub.2]E[R.sub.t-1] - [[beta].sub.3]E[R.sup.2.sub.t]).sup.2]|[[OMEGA].sub.t-1]) where [Q.sub.t-1] denotes the set of all information available at time t - 1.

We use two proxies for terrorist attacks: (a) a dummy variable that takes a value of one every 10-minute interval on a day on which a terrorist attack takes place and a value of zero every 10-minute interval on days when there are no terrorist attacks; and (b) the number of terrorist attacks, such that if, for example, five attacks took place on day t then in every 10-minute interval on that day we take a value of five. The terrorism variables in (a) and (b) relate to domestic terrorist attacks only. We do not consider in Equation (1) terrorist attacks in other countries. We return to this issue later in robustness tests (see Section III.E). Equation (1) is estimated for each country in our sample. The results from the dummy variable regression model and the terrorist number-based regression model are reported in columns 2 and 5 of Table 3, respectively. The results reveal the following. The dummy variable model tells us that in 14 of 21 currencies the effect of terrorism is a depreciation of the local currency vis-a-vis the USD and the range of effect is 0.02%-32.6% per annum. By comparison, ten currencies experience an appreciation, ranging from 0.007% to 155.1% per annum. Similarly, when we consider the number of terrorist attacks as a measure of terrorism, 14 of 21 (7/21) currencies experience a depreciation (appreciation). The magnitudes of depreciation and appreciation fall in the [0.001%, 22.8%] and [0.0003%, 80.4%] per annum range, respectively.

Table 4 reports results from a regression model that aims at establishing how exchange rates are affected (a) on the day of terrorist attacks and (b) 1-2 days after the attacks. The exchange rate model augmented with the contemporaneous effect takes the following form:

(1) E[R.sub.t] = [alpha] + [[beta].sub.1][TER.sub.t] + [[beta].sub.2]E[R.sub.t-144] + [[beta].sub.3][TER.sub.t-288] + [[epsilon].sub.t].

There are several important results contained in Table 4. First, notice how the effect of terrorist attacks on exchange rate changes as terrorist attacks become stale. Terrorist attacks lagged a day lead to appreciation of 13 currencies and depreciation of 8 currencies. Second, when the effect is tested with attacks occurring two days ago, 12 (9) currencies experience an appreciation (depreciation). The implication of these results is that for all currencies the effect of terrorist attacks persists, that is, past attacks influence exchange rates in the current period. The way to interpret these asymmetric responses of exchange rates to terrorist attacks is from a behavioral perspective. Terrorist attacks, and their effects, determine market confidence. If, following an attack, markets are worried about the future of a country's economy, then investors will sell that country's currency leading to a fall in the value of that country's currency. However, if the predicted repercussions of the attack are deemed not serious enough to dent investor confidence then investors (both existing and new) will continue to invest in the country regardless of the attack. This will increase the value of the currency, hence an appreciation. The story that emerges from our data suggests that exchange rate traders/investors across the 21 countries are not homogenous in how they perceive the economy following a terrorist attack. There are multiple reasons as to why investors are heterogeneous. These reasons range from different macroeconomic conditions (including credit risk profiles) to different institutional features in each of the countries. Therefore, the manner in which investors behave in each country following an attack is heterogeneous thus displaying the asymmetric behavior we observe in our results.

The second message emerging from these results is that the initial reaction of exchange rate to terrorist attacks is either completely reversed, partially reversed, or that it persists 2 days following the attacks. In the case of ten currencies, the effect persists, and for seven (four) currencies there is complete (partial) reversal in exchange rates. There are two ways to interpret these results. First, on the evidence that the effect persists, the idea has roots in the investor under-reaction hypothesis (see Hong and Stein 1999). Investors in this situation may gradually change their beliefs about a shock and therefore prices will continue to adjust as long as investors are updating their beliefs about the potential impact of a shock. Second, the evidence that the effect of terrorist attacks reverses in time has roots in theories developed in financial economics in particular with respect to investor overreaction (see Barberis, Shleifer, and Vishny 1998). Based on this, we argue that if investors do indeed overreact to terrorist attacks then they will correct this behavior resulting in the reversal of the effect on exchange rate.

C. ADDITIONAL RESULTS

Do the Effects of Terrorist Attacks Persist? In this section, we present results on the joint null hypothesis that the coefficients on lagged terrorist attacks are zero. In other words, we run the following regression model:

(2) E[R.sub.t] = [alpha] + [[gamma].sub.1] [TER.sub.t] + [n.summation over (i=1)] [[beta].sub.i][TER.sub.t-i] + [[epsilon].sub.t].

In the first predictive regression model, we set n = 144, and in the second model we set n = 288, representing effects of terrorist attacks after 1 and 2 days, respectively. The implications from these models can be traced to two hypotheses, namely, the overshooting and undershooting hypotheses. Consider the work of Dornbusch (1976), for example, who proposed the overshooting hypothesis. This hypothesis has roots in the role played by unexpected monetary shocks as key drivers of short-run exchange rate movements. (6) The immediate effect of an unexpected monetary contraction (expansion) is an appreciation (depreciation) of the dollar. Over time, the dollar is expected to depreciate (appreciate) as it returns to its long-run equilibrium level, consistent with the uncovered interest rate parity. (7) Exchange rate overshooting results when the instantaneous response of the exchange rate to the monetary shock is greater than its long-run response.

The opposing hypothesis is that of exchange rate undershooting. This occurs when following a contractionary (expansionary) monetary shock, or bad (good) news, the exchange rate appreciates (depreciates) by less than its long-run appreciation (depreciation) (see, inter alia, Dornbusch 1976; Frenkel and Rodriguez 1982; Papell 1985). Related to the undershooting hypothesis is the evidence of what some studies refer to as the bandwagon effect. (8)

The results are reported in Table 5. To investigate the prevalence of the hypotheses, we simply compare the size of the contemporaneous effect (column 2) against the corresponding effect on exchange rate after 1 day (column 3) and 2 days (column 4). We observe that the instantaneous effect of terrorism is significant for 15 of 21 countries' currencies. We also notice that the exchange rate is still affected by terrorist attacks that took place 1 day and 2 days ago. The 1-day effect of terrorist attacks is statistically different from zero for 11 of 21 countries' currencies while the effect of terrorist attacks 2 days ago is statistically different from zero for 11 of 21 countries' currencies. The message here is that, consistent with our earlier observations, the effect of terrorist attacks on exchange rate behavior weakens as the information on terrorist attacks gets outdated, although the effect persists even 2 days after the terrorist attack.

Second, we observe weak evidence of overshooting when testing the effect of terrorist attacks a day after the attacks. Only in the case of CLP, LKR, TRY, and ZAR, the instantaneous effect of terrorism is an appreciation of the currency and its size effect is bigger than the net effect (depreciation) summed over the day, suggesting exchange rate overshooting. When testing the effect of terrorist attacks over the past 2 days, we find that the evidence of overshooting completely disappears for only LKR but persists for CLP, TRY, and ZAR.

However, exchange rate undershooting or bandwagon effect is found in the case of COP, JPY, PHP, PKR. and THB when we consider the effect of terrorist attacks a day ago. In all these cases, the instantaneous appreciations are less than the long-run appreciations.

From this, we learn that terrorism, like unexpected macroeconomic news or announcements, can cause over- or under-shooting in the short term. Further similar to unexpected macroeconomic news, evidence of over- or under-shooting is thin but there is ample evidence that the effects of terrorism on the exchange rates for some countries are persistent two days following the attacks.

Do Fatalities from Terrorist Attacks Also Impact Exchange Rate Behavior? In this section, we consider studying the role of fatalities (resulting from terrorist attacks) on exchange rate behavior. Like with terrorist attack-based dummies, we now create a dummy variable capturing the effect of fatalities from terrorist attacks. On the day of the attack when there are fatalities, then for every 10-minute frequency on that day we repeat the number of fatalities. We also create a dummy variable that takes a value of one in every 10-minute interval on the day of a terrorist attack when there are fatalities and zero for every 10-minute interval on the day when there are no fatalities. In this way, we end up with two proxies capturing the effect of fatalities from terrorist attacks. The results are reported in Tables 6 and 7. Except for Japan, for which there are insufficient data on fatalities, and for TND, all currencies are impacted by terrorist attack-induced fatalities. Nine currencies experience a depreciation while another nine currencies experience an appreciation. Fatalities resulting a day ago lead to appreciation (depreciation) of nine (eight) currencies. The results are broadly consistent when we use the number of fatalities: at time t - 1, eight currencies experience appreciation and another eight experience depreciation.

The main message here is that regardless of whether we use terrorist attacks or fatalities resulting from terrorist attacks, exchange rates respond to both these events in a similar manner; that is, some currencies undergo a depreciation while others experience an appreciation.

D. Out-of-Sample Test

So far our empirical analysis has focused on in-sample evidence of the role of terrorist attacks in influencing exchange rate behavior. An influential literature in exchange rate economics when testing for exchange rate predictability has almost exclusively relied on out-of-sample predictability tests. This literature has been motivated by the pioneering work of Meese and Rogoff (1983). We feel that to provide completeness to our research question of whether or not terrorist attacks influence exchange rate returns in addition to the in-sample analysis undertaken in previous sections, an out-of-sample evaluation is necessary. This section, therefore, dwells on out-of-sample tests, where we compare the terrorist attack-based exchange rate model (M-T), which amounts to setting [beta] j = [[beta].sub.3] = 0 in Equation (1), with a constant-only exchange rate model (MC), which amounts to setting [[beta].sub.1] = [[beta].sub.2] = [[beta].sub.3] = 0 in Equation (1). Following Rapach, Strauss, and Zhou (2010) and Narayan, Narayan, and Sharma (2013), we utilize a recursive window approach; that is, we estimate the predictive regression model for the in-sample period [t.sub.0] to t (50% of the sample) and forecast exchange rate returns for the period t + 1. The model is then reestimated for the period [t.sub.0] to t + 1 and forecasts are generated for the period t + 2, stopping only at the last data point of the sample. Multiple out-of-sample test statistics--namely, the root-mean-squared error of M-T relative to M-C (RRMSE), the out-of-sample R-squared (O[R.sup.2]) which is computed as one minus the mean-square forecast error from the M-T relative to M-C (see Campbell and Thompson 2008), and the Clark and West (2007) mean-square forecast error adjusted test statistic (MSFE-adjusted)--are generated.

The results appear in Table 8. We begin with evidence obtained from RRMSE: the RRMSE <1 for 12 of 21 countries. However, the O[R.sup.2] > 0 for 12 of 21 countries, and the MSFE-adjusted test statistic reveals that in 11 of those cases the null hypothesis that O[R.sup.2] = 0 is rejected in favor of O[R.sup.2] > 0. This implies that the terrorist attack-based exchange rate model beats the constant returns model in those 11 countries. In terms of consistency, we notice that all three metrics support the terrorist attack-based model over a constant returns-based model in 11 of 21 countries. These countries are Australia, Canada, Colombia, Hong Kong, India, Jordan, Japan, Mexico, Thailand, Turkey, and South Africa.

E. Robustness Tests

In this section, we attempt to establish the robustness of our results along four lines. We believe that, given the data-intensive nature of our research question, a skeptic can rightfully question both our empirical specification and choice of data frequency. We, therefore, make an attempt to address these issues in this section. We begin by first looking specifically at the predictive regression model specification. While our empirical framework is consistent with models used to study predictability of stock returns (see, e.g., Garcia 2013; Tetlock 2007), the rather obvious question that arises is: are our results sensitive to the use of the contemporaneous variable? To address this issue, we reestimate all predictive regression models without the contemporaneous variable. The results are reported in Table 9. There are three observations to be made from these results. First, it is clear that with or without the contemporaneous variable, the effect of terrorist attacks on exchange rate behavior holds, thus rendering our results robust. The second thing we notice is that the magnitude of effect on exchange rate is larger with the contemporaneous variable, suggesting that contemporaneous effects should not be ignored. Finally, we observe that the effect from the 1-day lagged predictive regression model most closely mirrors the results from with and without the contemporaneous variable. This again goes to show the importance of modeling the contemporaneous effect, particularly when the information on terrorist attacks becomes stale--that is, when we utilize 1- and 2-day lags.

The second issue we have in mind relates to data frequency. By using 10-minute data, we managed to depict more closely the behavior of the exchange rate. A pertinent question in this regard is whether different data frequencies matter for the effects of terrorism on exchange rate behavior. We reestimate all predictive regression models with and without the contemporaneous variable using hourly data. We begin with a simple linear regression of the number of terrorist attacks on hourly exchange rate returns. The slope coefficient together with the p value examining the null hypothesis that the slope coefficient is zero is reported in column 2 of Table 10. In 17 of 21 currencies, terrorist attack has a slope coefficient that is statistically different from zero, suggesting that terrorist attacks matter for exchange rate behavior even when using hourly data. We notice that ten currencies experience an appreciation while seven currencies experience a depreciation from terrorist attacks. In columns 3 and 4, we report, respectively, the joint null hypothesis that 12 lags of terrorist attacks are statistically zero, both with and without the contemporaneous variable. The results of the effect of each of the 12 hourly lags of terrorist attacks on exchange rate are available upon request. (9)

The third issue regards the out-of-sample statistics. While we addressed the robustness of the results in terms of the number of currencies for which the terrorist attack-based exchange rate model beats the constant returns model, the one issue that remains is the choice of the in-sample period in generating forecasts. Previously, we used a 50% in-sample period to generate recursive forecasts. Is out-of-sample evidence that favors the terrorism-based exchange rate model sensitive to the choice of the in-sample period? We answer this by considering in-sample periods of 25% and 75%. In other words, in generating out-of-sample statistics, we consider both a short in-sample period and a long in-sample period. The results are reported in Table 11. The main conclusion we draw from these results is that the in-sample period does matter for the out-of-sample evidence of predictability of exchange rate returns, although the effect is not dramatic to change our main findings. For example, when using a small in-sample period of 25%, we find that in 15 of 21 countries the terrorism-based model is preferred over a constant returns model. By comparison, when we consider a long in-sample period of 75% in ten countries the terrorism-based model beats the constant returns model.

The final issue is about whether terrorist attacks in the United States influence currencies of other countries. To test this hypothesis, we simply augment Equation (1) with a U.S. terrorist attack dummy variable--which takes a value of one when there is a terrorist attack in the United States and zero otherwise. The regression is then estimated for each of the 21 countries in our sample. The results are presented in Table 12. The first point of note is that in 18 of 21 countries, exchange rate is affected by a terrorist attack in the United States. Seven currencies undergo depreciation while 11 currencies experience appreciation. Evidence of predictability at time t - 1 and t - 2 is still strong, with 17 and 16 countries, respectively, having predictable exchange rates.

IV. CONCLUDING REMARKS

In this study, we examine the effects of terrorist attacks on exchange rate behavior. This is a unique question, unaddressed thus far by the literature. The uniqueness of the research question is matched by a unique data set that includes 10minute exchange rate returns and terrorist attacks. Constrained by data analysis our hypothesis test is based on 21 countries. Three new findings are unraveled. First, we show that terrorist attacks lead to both an appreciation and depreciation of currencies. Second, we show that as information on terrorist attacks becomes stale, its effect on exchange rate weakens but persists. This suggests that while the bulk of the effect of terrorist attacks is felt on the day of the attacks, attacks that took place even 2 days ago still influence exchange rate behavior. Third, we show, consistent with exchange rate theories, that for some countries the effect of terrorist attack on exchange rate reverses with time while for some currencies the effect is persistent. The main message of our paper, and therefore its key contribution, is that terrorist attacks influence exchange rate behavior. This evidence is built on an empirical analysis that considers 21 currencies and is robust to (a) different empirical specifications, (b) different data frequencies, and (c) both in-sample and out-of-sample tests.

ABBREVIATIONS

ABD: Andersen, Bollerslev, and Diebold

O[R.sup.2]: Out-of-Sample [R.sup.2]

RRMSE: Root-Mean-Squared Error of M-T Relative to M-C

UIP: Uncovered Interest Parity

USD: U.S. Dollar

REFERENCES

Andersen, T., and T. Bollerslev. "Deutsche Mark-Dollar Volatility: Intraday Activity Patterns, Macroeconomic Announcements, and Longer Run Dependencies." Journal of Finance, 53, 1998. 219-65.

Andersen, T. G., T. Bollerslev, and F. X. Diebold. "Roughing It Up: Including Jump Components in the Measurement, Modeling and Forecasting of Return Volatility." Review of Economics and Statistics, 89, 2007, 701-20.

Arin, K. P., D. Ciferri, and N. Spagnolo. "The Price of Terror: The Effects of Terrorism on Stock Market Returns and Volatility." Economics Letters, 1010, 2008, 164-67.

Barberis, N., A. Shleifer, and R. Vishny. "A Model of Investor Sentiment." Journal of Financial Economics, 49, 1998, 307-43.

Bjornland, H. C. "Monetary Policy and Exchange Rate Overshooting: Dornbusch Was Right After All." Journal of International Economics, 79, 2009. 64-77.

Bonser-Neal, C., V. V. Roley, and G. H. Sellon Jr. "Monetary Policy Actions, Intervention, and Exchange Rates: A Re-examination of the Empirical Relationships Using Federal Funds Rate Target Data." Journal of Business, 71, 1998, 147-77.

Brounen, D., and J. Derwall. "The Impact of Terrorist Attacks on International Stock Markets." European Financial Management, 16, 2010, 585-98.

Campbell, J. Y., and S. B. Thompson. "Predicting the Equity Premium Out of Sample: Can Anything Beat the Historical Average?" Review of Financial Studies, 21, 2008, 1509-31.

Charles, A., and O. Darne. "Large Shocks and the September 11th Terrorist Attacks on International Stock Markets." Economic Modelling, 23, 2006, 683-98.

Chen, A. H., and T. F. Siems. "The Effects of Terrorism on Global Capital Markets." European Journal of Political Economy, 20, 2004, 349-66.

Chesney, M., G. Reshetar, and M. Karaman. "The Impact of Terrorism on Financial Markets: An Empirical Study." Journal of Banking & Finance, 35, 2011, 253-67.

Chevallier, J. "Detecting Instability in the Volatility of Carbon Prices." Energy Economics, 33, 2011, 99-110.

Clark, T. E., and D. W. West. "Approximately Normal Test for Equal Predictive Accuracy in Nested Models." Journal of Econometrics, 138, 2007, 291-311.

Dickey, D. A., and D. W. Fuller. "Distribution of the Estimators for Autoregressive Time Series with a Unit Root." Journal of the American Statistical Association, 74, 1979, 427-31.

Dornbusch, R. "Expectations and Exchange Rate Dynamics." Journal of Political Economy, 84(6), 1976, 1161-76.

Ederington, L. H., and J. H. Lee. "How Markets Process Information: News Releases and Volatility." Journal of Finance, 48(4), 1993, 1161-91.

Eichenbaum, M., and C. L. Evans. "Some Empirical Evidence on the Effects of Shocks to Monetary Policy on Exchange Rates." Quarterly Journal of Economics, 110, 1995, 975-1009.

Evans, M. D. D., and R. K. Lyons. "How Is Macro News Transmitted to Exchange Rate?" Journal of Financial Economics, 88, 2008, 26-50.

Fama, E. F., and K. R. French. "Business Conditions and Expected Returns on Stocks and Bonds." Journal of Financial Economics, 25, 1989, 23-49.

Faust, J., J. H. Rogers, S.-Y. B. Wang, and J. H. Wright. "The High-Frequency Response of Exchange Rates and Interest Rates to Macroeconomic Announcements." Journal of Monetary Economics, 54, 2007, 1051-68.

Frenkel, J., and C. A. Rodriguez. "Exchange Rate Dynamics and the Overshooting Hypothesis." International Monetary Fund Staff Papers, 29, 1982, 1-29.

Garcia, D. "Sentiment during Recessions." Journal of Finance, 68(3), 2013, 1267-300.

Gnabo, J. Y., L. Hvozdyk, and J. Lahaye. "System-Wide Tail Comovements: A Bootstrap Test for Cojump Identification on the S&P 500, US Bonds and Currencies." Journal of International Money and Finance, 48, 2014, 147-74.

Hong, H., and J. C. Stein. "A Unified Theory of Underreaction, Momentum Trading and Overreaction in Asset Markets." Journal of Finance, 54, 1999, 2143-84.

Kalyvitis, S., and A. Michaelides. "New Evidence on the Effects of US Monetary Policy on Exchange Rates." Economics Letters, 71, 2001, 255-63.

Kim, S., and N. Roubini. "Exchange Rate Anomalies in the Industrial Countries: A Solution with a Structural VAR Approach." Journal of Monetary Economics, 45, 2000, 561-86.

Meese, R. A., and K. Rogoff. "Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?" Journal of International Economics, 77, 1983, 167-80.

Meierrieks, D., and T. Gries. "Causality between Terrorism and Economic Growth." Journal of Peace Research, 50, 2013, 91-104.

Melvin, M., and X. Yin. "Public Information Arrival, Exchange Rate Volatility and Quote Frequency." The Economic Journal, 110, 2000, 644-61.

Narayan, P. K., S. Narayan, and S. Sharma. "An Analysis of Commodity Markets: What Gain for Investors?" Journal of Banking & Finance, 37, 2013, 3878-89.

Nasir, M., F. U. Rehman, and M. Orakzai. "Exploring the Nexus: Foreign Aid, War on Terror and Conflict in Pakistan." Economic Modelling, 29, 2012, 1137-45.

Palandri, A. "Do Negative and Positive Equity Returns Share the Same Volatility Dynamics?" Journal of Banking & Finance, 58, 2015, 486-505.

Papell, D. H. "Activist Monetary Policy, Imperfect Capital Mobility, and the Overshooting Hypothesis." Journal of International Economics, 18, 1985, 219-240.

Rapach, D. E., J. K. Strauss, and G. Zhou. "Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy." Review of Financial Studies, 23, 2010, 821-62.

Ratcliffe, A., and S. H. K. Scholder. "The London Bombings and Racial Prejudice: Evidence from the Housing and Labour Market." Economic Inquiry, 53, 2015, 276-93.

Raza, S. A., and S. T. Jawaid. "Terrorism and Tourism: A Conjunction and Ramification in Pakistan." Economic Modelling, 33, 2013, 65-70.

Roetheli, T. "Bandwagon Effects and Run Patterns in Exchange Rates." Journal of International Financial Markets Institutions & Money, 12, 2002, 157-66.

Shahbaz, M. "Linkages between Inflation, Economic Growth and Terrorism in Pakistan." Economic Modelling, 32, 2013, 496-506.

Shahbaz, M.. M. S. Shabbir, M. N. Malik, and M. E. Wolters. "An Analysis of a Causal Relationship between Economic Growth and Terrorism in Pakistan." Economic Modelling, 35, 2013, 21-29.

Tetlock, P. C. "Giving Content to Investor Sentiments: The Role of Media in the Stock Market." Journal of Finance, 62, 2007, 1139-68.

Westerfield, J. M. "An Examination of Foreign Exchange Risk under Fixed and Floating Rate Regimes." Journal of International Economics,!, 1977, 181-200.

(1.) The effect of terrorism on tourism and foreign aid have been studied by Raza and Jawaid (2013) and Nasir, Rehman, and Orakzai (2012).

(2.) In some cases, we have missing exchange rate data at the 10-minute frequency. Our approach here is to use the linear interpolation technique to deal with missing values.

(3.) We use 10-minute data motivated by the discussions in the study by Andersen, Bollerslev, and Diebold (ABD) (2007). The key message of ABD is that one needs to strike a balance between using high frequency data and being free of market microstructure contamination, which if left unaddressed will bias variance estimates. Given this, ABD recommend not using too high a data frequency (such as 1- or 5-minute). We, therefore, avoid these frequencies and consider 10-minute data. There are other studies, like ours, which ignore 1- and 5-minute data modeling for the same reasons as alluded to by ABD (see, e.g., Chevallier 2011; Gnabo. Hvozdyk. and Lahaye 2014; Palandri 2015).

(4.) http://www.start.umd.edu/gtd/

(5.) The GARCH(1.1) specification is chosen because higher order GARCH models do not influence the main conclusions.

(6.) Dornbusch (1976) achieves this by assuming a sluggish price adjustment in the goods market relative to exchange rate and asset markets and the uncovered interest parity (UIP). The UIP condition states that nominal interest rate differentials reflect movements in exchange rate. Hence, taking the case of the Australian dollar per USD we can state the UIP condition as: (E[[e.sub.t+1]-[e.sub.t]/([e.sub.t])]= [r.sub.AUS] - [r.sub.US]. Here [r.sub.AUS] and [r.sub.US] are interest rates on the Australian dollar deposits and the USD deposits at time, t, respectively; e is the exchange rate (Australian dollars per USD or (AUD/USD)); and E ([e.sub.t+1]) is the expected exchange rate at time t + 1 conditional on information available on time t. The UIP suggests that a relatively high interest rate on the Australian dollar deposit will lead to a depreciation of the currency.

(7.) Empirical evidence of exchange rate overshooting is still scarce. Evidence of exchange rate overshooting, for example, has been found in studies by Eichenbaum and Evans (1995), Bonser-Neal, Roley, and Sellon (1998), Kalyvitis and Michaelides (2001), Kim and Roubini (2000), and Bjornland (2009).

(8.) For empirical evidence of exchange rate undershooting, see Dornbusch (1976), Frenkel and Rodriguez (1982), and Papell (1985). More recent studies, such as Roetheli (2002), use an autoregressive framework to show evidence of persistence in monthly average exchange rate returns and suggest a bandwagon effect.

(9.) We do not tabulate these results here because they span many pages and do not add much value to what we have reported already.

Narayan: Alfred Deakin Professor, Centre for Financial Econometrics, Deakin Business School, Deakin University, Melbourne, Vic. 3125, Australia. Phone +61 3 9244 6180, Fax +61 3 9244 6034, E-mail narayan@deakin.edu.au

Narayan: Associate Professor, Department of Economics, Finance, and Marketing, Royal Melbourne Institute of Technology University, Melbourne, Vic. 3000, Australia. Phone +61 3 9925 5890, E-mail seema.narayan@rmit.edu.au

Khademalomoom: Econometrician, Centre for Financial Econometrics, Deakin Business School, Deakin University, Melbourne, Vic. 3125, Australia. Phone +61 3 9651 1647, E-mail siroos.khademalomoom@dtf.vic.gov.au Phan: Senior Lecturer, School of Business, Monash University, Bandar Sunway, 47500, Malaysia. E-mail dinh.phan@monash.edu

doi:10.1111/ecin.12447

Online early publication March 25, 2017

Caption: FIGURE 1 A Plot of Spot Exchange Rate Returns

Caption: FIGURE 2 Mean and Maximum Number of Terrorist Attacks by Country

TABLE 1
Descriptive Statistics

(A) Descriptive Statistics of Currency Returns

         Mean      SD     Skew.     Kurt.

AUD      0.743    0.046   -0.030    42.096
CAD      0.743    0.030   0.166     34.498
CHF      1.958    0.037   0.153     34.189
CLP     -1.456    0.037   -1.063   303.843
CNY      1.400    0.006   3.491    1788.217
COP     -4.480    0.072   -0.515   850.191
GBP     -0.222    0.031   -0.100    31.480
HRK     -1.348    0.069   0.025     37.713
IDR     -5.537    0.174   -1.264   358.325
ILS     -1.005    0.041   -0.290    52.513
INR     -2.286    0.042   -5.140   229.365
JOD     -0.203    0.018   -0.678   125.423
JPY     -0.559    0.037   0.283     35.145
LKR     -3.506    0.028   0.485    662.757
MXN     -3.537    0.048   -0.180   575.399
PHP     -2.645    0.082   -1.083   1744.969
PKR     -4.291    0.049   -3.822   1199.281
THB     -1.338    0.349   0.155    449.532
TND     -0.064    0.116   0.385    120.309
TRY     -13.944   0.085   2.788    2515.329
ZAR     -6.305    0.068   -1.240   913.704

         JB       ADF      AR(1)    Hetero.

AUD     <.001   -371.259   0.304     <.001
CAD     <.001   -311.837   0.366     <.001
CHF     <.001   -420.734   0.346     <.001
CLP     <.001   -248.302   0.183     <.001
CNY     <.001   -299.836   0.014     <.001
COP     <.001   -197.359   -0.145    <.001
GBP     <.001   -392.512   0.376     <.001
HRK     <.001   -146.021   -0.259    <.001
IDR     <.001   -100.909   -0.195    <.001
ILS     <.001   -149.291   -0.144    <.001
INR     <.001   -121.213   -0.006    <.001
JOD     <.001   -116.090   -0.302    <.001
JPY     <.001   -392.609   0.364     <.001
LKR     <.001   -121.021   -0.111    <.001
MXN     <.001   -221.244   0.084     <.001
PHP     <.001   -100.985   -0.158    <.001
PKR     <.001   -119.701   -0.092    <.001
THB     <.001   -155.355   -0.275    <.001
TND     <.001   -225.993   -0.389    <.001
TRY     <.001   -94.566    -0.071    <.001
ZAR     <.001   -355.472   0.136     <.001

(B) Descriptive Statistics of Terrorist Attacks

                  Mean     SD      JB       ADF

Australia         0.005   0.077   <.001   -54.545
Canada            0.005   0.073   <.001   -83.689
Chile             0.012   0.127   <.001   -70.024
China             0.028   0.600   <.001   -83.071
Colombia          0.420   1.199   <.001   -8.315
Croatia           0.003   0.059   <.001   -11.505
India             0.948   1.563   <.001   -9.312
Indonesia         0.091   0.748   <.001   -54.535
Israel            0.155   0.717   <.001   -10.452
Japan             0.006   0.106   <.001   -33.177
Jordan            0.004   0.071   <.001   -83.540
Mexico            0.036   0.280   <.001   -8.669
Pakistan          1.460   2.589   <.001   -3.362
Philippines       0.437   0.954   <.001   -4.871
South Africa      0.022   0.290   <.001   -37.768
Sri Lanka         0.162   0.492   <.001   -8.397
Switzerland       0.003   0.063   <.001   -80.559
Thailand          0.416   1.200   <.001   -13.296
Tunisia           0.009   0.110   <.001   -15.905
Turkey            0.127   0.463   <.001   -14.163
United Kingdom    0.138   0.454   <.001   -21.724

Notes: Panel A reports the annualized mean of exchange rate
returns, its standard deviation (SD), skewness and kurtosis, a
Jarque-Bera test of nonnormality of returns, a stationarity (ADF)
test, autocorrelation (AR(1)) coefficient of returns, and a test
for heteroskedasticity in returns. The evaluation period covers the
10-minute data from January 1, 1996 to December 1, 2014. The null
hypothesis of normality is based on the p values from the
Jarque-Bera test. The results (t statistics) from the ADF unit
test, which examines the null hypothesis of a unit root against the
alternative of no unit root is implemented using a model with an
intercept but no time trend. Panel B reports the mean of the daily
terrorist attacks for each country and its standard deviation,
followed by the Jarque-Bera normality test and the ADF unit root
test.

TABLE 2
Mean Exchange Rate Returns on Days of
Terrorist Attacks and Nonterrorist Attacks

Exchange    Mean Returns on      Mean Returns on
Rate        Terrorist Attack   Nonterrorist Attack
               Days (SD)            Days (SD)

AUD          26.298(10.982)      0.631 (10.534)
CAD          -9.535 (9.517)       0.795 (6.837)
CHF         -99.965 (9.105)       2.240 (8.387)
CLP         -17.584 (8.635)       -1.297(8.381)
CNY           2.394(1.166)        1.384(1.395)
COP         -11.073(18.362)      -2.281 (15.716)
GBP          -5.639 (6.699)       0.452 (7.036)
HRK         -50.598(13.231)      -1.200(15.716)
IDR          5.690 (30.568)      -6.254 (40.427)
ILS          -3.009 (9.234)      -0.789 (9.372)
INR          -4.224(10.362)      -0.692 (8.908)
JOD          -2.502 (5.395)      -0.195 (4.134)
JPY         -27.067 (9.036)      -0.420 (8.556)
LKR          -7.279 (5.951)      -2.978 (6.563)
MXN         -29.836(12.915)      -2.804(10.942)
PHP          -1.589(16.364)      -3.014(19.446)
PKR         -4.988 (10.243)      -3.773 (11.935)
THB         -2.133 (92.869)      -1.112(76.007)
TND         -6.152 (27.780)      -0.016(26.635)
TRY         -16.968(11.186)     -13.636 (20.235)
ZAR         -52.135 (11.841)     -5.537(15.732)

TABLE 3
Contemporaneous Effects of Terrorist Attacks on
Bilateral Exchange Rate Returns

         Terrorist      p Value   Annualized
       Attacks and FX             Effect (%)

AUD       3.30E-05       0.00        1.734
CAD      -1.01E-04       0.00       -5.309
CHF      -5.33E-04       0.00      -28.014
CLP      -1.93E-04       0.00      -10.144
CNY       2.70E-05       0.00        1.419
COP      -4.08E-07       0.00       -0.021
GBP      -1.13E-06       0.00       -0.059
HRK      -1.17E-04       0.00       -6.150
IDR       1.35E-07       0.00        0.007
ILS      -9.58E-06       0.00       -0.504
INR       2.38E-07       0.00        0.013
JOD       2.95E-03       0.00      155.052
JPY      -4.25E-04       0.00      -22.338
LKR      -2.06E-05       0.00       -1.083
MXN      -5.31E-05       0.00       -2.791
PHP      -1.55E-06       0.00       -0.081
PKR       1.73E-07       0.00        0.009
THB       6.65E-06       0.00        0.350
TND      -4.33E-05       0.00       -2.276
TRY      -3.22E-06       0.00       -0.169
ZAR      -6.21E-04       0.00      -32.640

       No. of Terrorist   p Value   Annualized
        Attacks and FX              Effect (%)

AUD        2.28E-04        0.00       11.984
CAD       -5.91E-05        0.00       -3.106
CHF       -4.34E-04        0.00      -22.811
CLP       -5.28E-05        0.00       -2.775
CNY       -1.37E-08        0.00       -0.001
COP       -2.70E-08        0.00       -0.001
GBP       -5.43E-08        0.00       -0.003
HRK        2.35E-04        0.00       12.352
IDR       -1.19E-07        0.00       -0.006
ILS        1.86E-07        0.00        0.010
INR        6.03E-09        0.00       0.0003
JOD        1.53E-03        0.00       80.417
JPY       -3.07E-04        0.00      -16.136
LKR       -3.16E-08        0.00       -0.002
MXN        5.03E-08        0.00        0.003
PHP        1.33E-06        0.00        0.070
PKR       -2.08E-07        0.00       -0.011
THB       -3.23E-08        0.00       -0.002
TND       -4.36E-05        0.00       -2.292
TRY       -2.21E-06        0.00       -0.116
ZAR       -4.54E-06        0.00       -0.239

Notes: The regression model has the following form [ER.sub.t] =
[alpha] + [[beta].sub.1] [TER.sub.t] + [[beta].sub.2][ER.sub.t-1] +
[[beta].sub.3][ER.sup.2.sub.t] + [[epsilon].sub.t] and is estimated
using a GARCH(1,1) model with a Student's t distribution. In this
regression, ER represents exchange rate returns and TER proxies for
terrorist attacks. Two proxies are used. Results reported in column
2, for instance, are based on a dummy variable that takes a value
of one every 10-minute interval on days on which a terrorist attack
takes place and a value of zero every 10-minute interval on days
when there are no terrorist attacks. Results reported in column 5
proxy terrorism with the actual number of terrorist attacks, such
that if, for example, five attacks took place on day t then in
every 10-minute interval on that day we take a value of five.
Columns 4 and 7 report the annualized effect of terrorist attacks
associated with each of the two proxies for terrorism. The
regression model controls for autocorrelation (with a one-period
lagged dependent variable) and volatility of exchange rate returns
(with a squared exchange rate return variable) directly in the mean
equation.

TABLE 4
Contemporaneous and Lagged Effects of
Terrorist Attacks on Exchange Rate Returns

       Contemporaneous       [t - 1 day]        [t- 2 days]
            Effect

AUD      0.53 (0.00)         0.21 (0.02)        0.24 (0.00)
CAD      0.02 (0.00)         0.01 (0.00)        -0.01 (0.00)
CHF      -0.61 (0.00)        0.42 (0.07)        0.32 (0.23)
CLP      2.59 (0.00)        -2.36 (0.00)        -4.33 (0.00)
CNY      -0.01 (0.02)       -0.03 (0.00)        -0.10(0.00)
COP      -0.10(0.00)        -0.07 (0.00)        -0.11 (0.00)
GBP    -3.53E-05 (0.00)   -4.63E-08 (0.00)    -4.75E-08 (0.00)
HRK      -0.08 (0.00)        0.02 (0.00)        -0.11 (0.00)
IDR      4.45 (0.00)         2.59 (0.00)        -1.91 (0.00)
ILS       0.14(0.00)         0.08 (0.00)        0.02 (0.00)
INR    1.75E-04 (0.00)     1.61E-04 (0.00)    1.53E-04 (0.00)
JOD      -0.48 (0.00)       -0.21 (0.00)        -0.14(0.00)
JPY      -0.52 (0.00)       -0.02 (0.00)        0.49 (0.00)
LKR      0.03 (0.02)         0.04 (0.00)        0.02 (0.18)
MXN    2.09E-04 (0.00)      -0.01 (0.00)         0.10(0.00)
PHP    -2.63E-03 (0.00)   - 1.81E-03 (0.00)   -3.26E-03 (0.00)
PKR    -8.85E-05 (0.00)    5.73E-05 (0.00)    7.23E-05 (0.00)
THB      0.01 (0.07)         0.03 (0.00)        0.04 (0.00)
TND       0.12(0.00)         0.07 (0.00)        0.09 (0.00)
TRY       0.16(0.00)         0.28 (0.00)         0.14(0.00)
ZAR      1.53 (0.00)         1.39 (0.00)        0.32 (0.00)

Notes: This table reports results based on a regression model that
aims at establishing the predictive ability of terrorist attacks,
using information on past terrorist attacks, that is, attacks that
took place 1 and 2 days ago. The exchange rate model augmented with
the contemporaneous effect takes the following form:

TABLE 5
Joint Null Hypothesis That the Coefficients on
Lagged Terrorist Attacks Are Zero

       Coefficient   Sum of 144   Sum of 288
        at Time t     Lags = 0     Lags = 0

AUD       0.67         -2.75         3.70
         (0.00)        (0.00)       (0.00)
CAD       -0.09         0.61        -0.58
         (0.86)        (0.89)       (0.90)
CHF       -1.08         0.01        -0.97
         (0.10)        (0.99)       (0.37)
CLP       1.88         -0.62         0.53
         (0.00)        (0.87)       (0.93)
CNY       -0.01        -0.05        -0.48
         (0.00)        (0.00)       (0.00)
COP       2.03         26.60         4.54
         (0.00)        (0.01)       (0.04)
GBP       0.05         -2.09        -2.42
         (0.54)        (0.00)       (0.00)
HRK       -1.17         3.25         7.52
         (0.51)        (0.87)       (0.68)
IDR       -0.24        -4.19        -2.23
         (0.37)        (0.09)       (0.41)
ILS       -1.71        92.48        98.06
         (0.00)        (0.00)       (0.00)
INR       0.16         -0.31        -2.36
         (0.00)        (0.28)       (0.00)
JOD       1.32         -4.36        -4.18
         (0.08)        (0.49)       (0.28)
JPY       0.16          5.65         1.07
         (0.46)        (0.00)       (0.36)
LKR       1.17         -1.16        -4.16
         (0.00)        (0.01)       (0.00)
MXN       -1.28         3.37         4.68
         (0.00)        (0.14)       (0.02)
PHP       0.76          0.08         4.00
         (0.00)        (0.72)       (0.00)
PKR       1.27          3.31         3.82
         (0.00)        (0.00)       (0.00)
THB       31.35        24.45        31.75
         (0.00)        (0.00)       (0.00)
TND       -1.28        -0.60         0.75
         (0.23)        (0.95)       (0.94)
TRY       11.32        -0.57         0.05
         (0.00)        (0.97)       (0.99)
ZAR       1.75         -1.33        -0.11
         (0.00)        (0.00)       (0.81)

Notes: Column 2 shows the contemporaneous effect of
terrorist attacks at time t. Columns 3 and 4 represent effects
of terrorist attacks after 1 and 2 days, respectively. All
coefficients are multiplied by 1,000 for ease of presentation,
p Values are presented in parentheses.

TABLE 6
Contemporaneous and Lagged Effects of
Terrorist Attacks--Induced Fatalities (Dummy)
on Exchange Rate Returns

          Terrorist      [t - 1 day]    [t - 2 days]
       Attacks Lead to
       Fatality and FX
         at Time [t]

AUD      0.36 (0.53)     0.86 (0.09)    -1.57 (0.00)
CAD     -1.04 (0.00)     1.03 (0.00)    0.49 (0.10)
CHF     -5.10 (0.00)     -2.22 (0.00)   -0.88 (0.07)
CLP     -4.71 (0.00)     -7.53 (0.00)   14.26 (0.00)
CNY     -0.37 (0.00)     -0.13 (0.00)   -0.22 (0.00)
COP     -0.34 (0.00)     -0.58 (0.00)   1.02 (0.00)
GBP      0.20 (0.08)     -0.08 (0.50)   0.26 (0.02)
HRK      0.62 (0.00)     -1.08 (0.00)   1.19 (0.00)
IDR    -117.81 (0.00)    130.24 (0.00)  16.05 (0.00)
ILS      0.09 (0.02)     0.19 (0.00)    0.20 (0.00)
INR      2.18 (0.00)     2.62 (0.00)    2.90 (0.00)
JOD      2.25 (0.00)     -0.03 (0.87)   3.82 (0.00)
JPY          N/A             N/A            N/A
LKR     -1.32 (0.00)     -0.34 (0.00)   -1.94 (0.00)
MXN      1.57 (0.00)     0.09 (0.00)    -4.62 (0.00)
PHP     -5.49 (0.00)     -4.01 (0.00)   2.57 (0.00)
PKR     -2.64 (0.00)     1.99 (0.00)    0.01 (0.64)
THB      0.91 (0.00)     3.73 (0.00)    1.17 (0.00)
TND     -0.10 (0.82)     -0.53 (0.23)   1.73 (0.00)
TRY      1.14 (0.00)     -0.64 (0.00)   -0.62 (0.00)
ZAR      3.68 (0.00)     4.50 (0.00)    6.76 (0.00)

Notes: This table reports results based on a regression model that
aims at establishing the predictive ability of terrorist
attack--induced fatalities, using information on past 2 days of
fatalities. The fatalities variable in this model is a dummy
variable that takes a value of one in all 10-minutes in a day if
that day had fatalities and zero otherwise. The exchange rate model
augmented with the contemporaneous effect takes the following form:

[ER.sub.t] = [alpha] + [[beta].sub.1] [TER.sub.t] +
[[beta].sub.2][TER.sub.t-144] + [[beta].sub.3][TER.sub.t-288] +
[[epsilon].sub.t].

The regression is estimated using a GARCH(1,1) model using a
Student's t distribution. All coefficients are multiplied by 1,000
for ease of presentation, p Values testing the null hypothesis that
the slope coefficient is zero are reported in parentheses.

TABLE 7
Contemporaneous and Lagged Effects of
Terrorist Attacks-Induced Fatalities (Number of
Fatalities) on Exchange Rate Returns

       No. of Terrorist     [t - 1 day]        [t - 2 days]
       Attacks Lead to
       Fatality and FX
         at Time [t]

AUD      0.45 (0.12)        0.43 (0.09)        -0.60 (0.00)
CAD      -0.41 (0.01)        0.43(0.01)         0.20(0.18)
CHF      -0.54 (0.00)       -0.30 (0.01)       -0.14(0.00)
CLP      -7.56 (0.00)       -5.94 (0.00)       8.27 (0.00)
CNY    -1.22E-03 (0.00)   -1.23E-03 (0.00)   -4.59E-04 (0.26)
COP      -0.09 (0.00)       -0.39 (0.00)       0.26 (0.00)
GBP      -0.01 (0.41)       -0.01 (0.52)       -0.18(0.00)
HRK      0.84 (0.00)        -0.04 (0.87)       3.30 (0.00)
IDR      -8.69 (0.00)       18.61 (0.00)       -2.05 (0.00)
ILS      -0.02 (0.00)       -0.02 (0.00)       0.03 (0.00)
INR      -0.02 (0.00)        0.14(0.00)        -0.02 (0.00)
JOD      -0.06 (0.00)       -0.01 (0.11)       -0.03 (0.00)
JPY          N/A                N/A                N/A
LKR      -0.07 (0.00)       0.06 (0.02)        -0.02 (0.00)
MXN      -0.03 (0.00)       -0.03 (0.00)       0.23 (0.00)
PHP      -0.12(0.00)        -0.05 (0.00)       -0.06 (0.00)
PKR      -0.02 (0.00)       -0.02 (0.00)       -0.06 (0.00)
THB      0.47 (0.00)         1.02(0.00)        -0.77 (0.00)
TND      -0.10(0.07)        -0.05 (0.33)       0.02 (0.64)
TRY      0.06 (0.00)        0.02 (0.00)        0.03 (0.00)
ZAR      -1.03(0.00)        0.95 (0.00)         1.24(0.00)

Notes: This table reports results based on a regression model that
aims at establishing the predictive ability of terrorist
attack--induced fatalities, using information on past 2 days of
fatalities. The fatalities variable in this model is a dummy
variable that takes a value of one in all 10-minutes in a day if
that day had fatalities and zero otherwise. The exchange rate model
augmented with the contemporaneous effect takes the following form:

[ER.sub.t] = [alpha] + [[beta].sub.1] [TER.sub.t] +
[[beta].sub.2][TER.sub.t-144] + [[beta].sub.3][TER.sub.t-288] +
[[epsilon].sub.t].

The regression is estimated using a GARCH(1,1) model using a
Student's t distribution. All coefficients are multiplied by 1,000
for ease of presentation, p Values testing the null hypothesis that
the slope coefficient is zero are reported in parentheses.

TABLE 8
Out-of-Sample Forecasting Results

Country       RRMSE    [OR.sup.2]   p Value
Currencies                (%)

AUD           0.9991     0.0018      0.00
CAD           0.9999     0.0001      0.00
CHF           1.0002    -0.0003      1.00
CLP           1.0000     0.0000      0.00
CNY           1.0001    -0.0002      1.00
COP           0.9972     0.0055      0.00
GBP           1.0002    -0.0004      0.97
HRK           0.9976     0.0048      0.00
IDR           1.0066    -0.0133      1.00
ILS           0.9752     0.0489      1.00
INR           0.9985     0.0030      0.00
JOD           0.9998     0.0004      0.00
JPY           0.9994     0.0013      0.00
LKR           1.0069    -0.0139      0.00
MXN           0.9992     0.0016      0.00
PHP           1.0017    -0.0035      1.00
PKR           1.0072    -0.0145      0.00
THB           0.9955     0.0089      0.00
TND           1.0002    -0.0004      1.00
TRY           0.9798     0.0400      0.00
ZAR           0.9996     0.0007      0.00

Notes: This table reports the out-of-sample (10-minute) forecast
performance results for terrorism-based model against the benchmark
historical mean model based. Forecasts are based on using 50% of
the in-sample period for recursive forecasting for the remainder of
the sample of data. Three forecast evaluation metrics, namely,
RRMSE, Fama and French (1989) [OR.sup.2], and the p value of Clark
and West (2007) MSFE-adjusted statistic are generated.

TABLE 9
Effect of Terrorist Attacks on Exchange Rate
Returns without the Contemporaneous Effect

            [t - 1 day]        [t - 2 days]

AUD         0.20 (0.09)         0.19 (0.07)
CAD         0.03 (0.00)        -0.11 (0.00)
CHF         0.40 (0.09)         0.52 (0.05)
CLP        -1.94 (0.00)        -4.38 (0.00)
CNY        -0.03 (0.00)        -0.12 (0.00)
COP        -0.70 (0.00)        -0.04 (0.00)
GBP     2.I3E-05 (0.03)     1.26E-05 (0.29)
HRK         0.29 (0.00)         0.31 (0.00)
IDR        -3.09 (0.00)         1.55 (0.00)
ILS        -0.03 (0.00)         0.08 (0.00)
INR    -1.50E-06 (0.00)     5.37E-05 (0.00)
JOD    -4.43E-03 (0.00)         0.03 (0.00)
JPY        -0.04 (0.00)         0.05 (0.00)
LKR        -0.03 (0.00)        -0.02 (0.00)
MXN     9.67E-04 (0.00)     3.56E-05 (0.00)
PHP         0.01 (0.42)         0.05 (0.00)
PKR    -5.57E-05 (0.00)    -2.51E-08 (0.00)
THB        -0.53 (0.00)        -0.54 (0.00)
TND        -0.07 (0.00)        -0.06 (0.00)
TRY         0.25 (0.00)     1.18E-04 (0.99)
ZAR         1.37 (0.00)         0.27 (0.00)

Notes: This table reports results based on a regression model that
aims at establishing the predictive ability of terrorist attacks,
using information on past terrorist attacks, that is, attacks that
took place 1 and 2 days ago. The model takes the following form:

[ER.sub.t] = [alpha] + [[beta].sub.1][TER.sub.t-144] +
[[beta].sub.2][TER.sub.t-288] + [[epsilon].sub.t].

The regression is estimated using a GARCh (1,1) model using a
Student's t distribution. All coefficients are multiplied
by 100 for ease of presentation. P Values testing the null
hypothesis that the slope coefficient is zero are reported in
parenthesis.

TABLE 10
Robustness Test Based on Hourly Data

        Terror Attack       Sum of 12         Sum of 12
       Contemporaneous      Lags = 0          Lags = 0
                              (With           (Without
                         Contemporaneous   Contemporaneous
                            Variable)         Variable)

AUD      5.34 (0.00)      -0.50 (0.91)       0.80 (0.00)
CAD     -10.50 (0.18)      1.10(0.01)        0.30 (0.00)
CHF     -1.34 (0.73)       7.20 (0.42)       -0.10(0.54)
CLP     77.73 (0.00)       2.20 (0.99)       0.00 (0.99)
CNY      -0.19(0.00)      -3.00 (0.00)       0.10(0.02)
COP     -0.02 (0.00)      -15.70 (0.00)     -0.30 (0.00)
GBP     -0.48 (0.00)      -5.80 (0.00)       0.10(0.37)
HRK      -2.18(0.64)      13.70 (0.49)       9.40 (0.12)
1DR     22.29 (0.00)      -0.80 (0.00)       0.20 (0.00)
ILS      0.08 (0.00)       1.80 (0.09)      -0.50 (0.00)
INR      0.05 (0.00)       0.02 (0.00)       0.00 (0.00)
JOD      -1.24(0.00)      -0.90 (0.00)      -0.20 (0.00)
JPY      8.16(0.00)       -16.70 (0.00)     -3.60 (0.06)
LKR      3.81 (0.00)      -4.00 (0.00)      -0.01 (0.10)
MXN     -0.79 (0.00)      155.30 (0.00)      0.01 (0.75)
PHP      0.04 (0.00)       0.20 (0.00)       0.10(0.00)
PKR     -0.01 (0.00)       0.03 (0.00)       0.00 (0.00)
THB      0.10(0.00)        0.30 (0.00)       0.04 (0.00)
TND      4.65 (0.00)       2.30 (0.00)       -0.40(0.11)
TRY     -0.79 (0.00)      -0.60 (0.97)       -3.00(0.14)
ZAR      0.79 (0.57)       0.30 (0.40)      -0.20 (0.20)

Notes: This table reports results from the same predictive
regression models as before; see Tables 4 and 7 for a description
of these models, but this time to establish the robustness of our
findings the model is applied to hourly exchange rate data. The
results in column 2 are for the contemporaneous effect of terrorist
attacks on exchange rate returns. Results in column 3 are those
summed-up over 12 hours from a model that includes the
contemporaneous terrorism variables while the last column has
corresponding results but from a model without the contemporaneous
variable. All coefficients are multiplied by 1,000 for ease of
interpretation, and p values are reported in parentheses.

TABLE 11
Out-of-Sample Forecasting Results

Country       RRMSE    [OR.sup.2] (%)   p Value
Currencies

(A) Out-of-sample results based on an in-sample
period of 25%

AUD           1.0000       0.0000        0.00
CAD           0.9999       0.0002        0.00
CHF           0.9882       0.0235        0.00
CLP           0.9952       0.0096        0.00
CNY           1.0000       0.0000        0.97
COP           0.9997       0.0007        0.00
GBP           1.0003      -0.0006        1.00
HRK           0.9999       0.0002        0.00
IDR           1.0039      -0.0078        0.00
ILS           0.9991       0.0018        0.00
INR           0.9977       0.0047        0.00
JOD           1.0117      -0.0234        0.00
JPY           0.9999       0.0001        0.00
LKR           1.0066      -0.0133        1.00
MXN           0.9980       0.0041        0.00
PHP           0.9997       0.0006        0.00
PKR           0.9978       0.0044        0.00
THB           0.9904       0.0191        0.00
TND           0.9998       0.0004        0.00
TRY           0.9988       0.0024        0.00
ZAR           0.9956       0.0088        0.00

(B) Out-of-sample results based on an in-sample
period of 75%

AUD           0.9977       0.0045        0.00
CAD           1.0000       0.0000        1.00
CHF           0.9955       0.0089        0.00
CLP           0.9999       0.0001        0.00
CNY           1.0029      -0.0058        1.00
COP           0.9676       0.0638        0.00
GBP           1.0001      -0.0002        1.00
HRK           0.9998       0.0004        0.00
IDR           1.0032      -0.0064        1.00
ILS           0.9985       0.0030        0.00
INR           1.0012      -0.0023        1.00
JOD           1.0003      -0.0005        1.00
JPY           1.0000       0.0000        1.00
LKR           0.9529       0.0919        0.00
MXN           1.0002      -0.0005        1.00
PHP           0.9666       0.0656        0.00
PKR           1.0252      -0.0510        0.00
THB           1.0122      -0.0246        0.00
TND           0.9976       0.0048        0.00
TRY           0.9988       0.0024        0.00
ZAR           1.0001      -0.0002        0.00

Notes: This table reports the out-of-sample (10-minute) forecast
performance results for terrorism-based model against the benchmark
historical mean model based. Forecasts are based on using 25%
(panel A) and 75% (panel B) as the in-sample period for recursive
forecasting for the remainder of the sample of data. Three forecast
evaluation metrics, namely, RRMSE, Fama and French (1989)
[OR.sup.2], and the p value of Clark and West (2007) MSFE-adjusted
statistic are generated.

TABLE 12
Results with U.S. Dummy for Terrorist Attacks

          US Dummy       No. of Terrorist
                        Attacks and FX at
                             Time [t]

AUD    -0.12 * (0.00)     0.42 * (0.00)
CAD    0.32 * (0.00)       0.01 (0.87)
CHF    -0.15 * (0.00)     -1.96 * (0.00)
CLP    -2.76 * (0.00)     0.34 * (0.00)
CNY    1.49 * (0.00)    -3.41E-03 * (0.02)
COP    1.69 * (0.00)       0.09* (0.00)
GBP     -0.04 (0.47)       0.01 (0.78)
HRK    0.017 * (0.01)      0.34 (0.10)
IDR     1.60 (0.26)      -32.94 * (0.00)
ILS    -0.38 * (0.00)     0.05 * (0.00)
INR    2.38 * (0.00)      -0.27 * (0.00)
JOD    -1.52 * (0.00)     -0.33 * (0.00)
JPY     0.06 (0.35)        0.05 (0.74)
LKR    -0.87 * (0.00)      0.03 (0.47)
MXN    0.76 * (0.00)      0.37 * (0.00)
PHP    0.32 * (0.00)      0.13 * (0.00)
PKR    0.82 * (0.00)      -0.06 * (0.00)
THB    -0.89 * (0.00)     -0.24 * (0.00)
TND    0.58 * (0.00)       0.11 (0.67)
TRY    0.79 * (0.00)      -0.03 * (0.01)
ZAR    1.40 * (0.00)      0.70 * (0.00)

        [t - 1 day]      [t- 2 days]

AUD    0.37 * (0.01)    0.50 * (0.00)
CAD     -0.01 (0.84)    -0.50 * (0.00)
CHF    0.88 * (0.00)    -0.57 * (0.00)
CLP    0.20 * (0.00)     -0.01 (0.79)
CNY    -0.05 * (0.00)   -0.11 * (0.00)
COP    0.11 * (0.00)    0.11 * (0.00)
GBP     -0.01 (0.74)     -0.02 (0.39)
HRK    0.40 * (0.03)    -1.76 * (0.00)
IDR    47.31 * (0.00)   -2.90 * (0.00)
ILS    0.03 * (0.00)    -0.05 * (0.00)
INR    -0.30 * (0.00)   -0.26 * (0.00)
JOD    0.71 * (0.00)    1.04 * (0.00)
JPY    0.01 * (0.98)     -0.11 (0.47)
LKR    -0.73 * (0.00)   -0.25 * (0.00)
MXN     -0.05 (0.14)    -0.53 * (0.00)
PHP    0.13 * (0.00)    -0.20 * (0.00)
PKR    -0.04 * (0.00)   0.04 * (0.00)
THB    -0.23 * (0.00)   -0.22 * (0.00)
TND     -0.20 (0.40)     0.27 (0.31)
TRY    0.15 * (0.00)     0.02 (0.14)
ZAR    0.68 * (0.00)    0.22 * (0.00)

Notes: This table reports results based on a regression model that
aims at establishing the predictive ability of terrorist attacks
but by augmenting the regression model with a U.S. terrorist attack
dummy: [TER.sup.USA.sub.t] is a dummy variable that takes one in
every 10-minute interval on a day there is a terrorist attack and
zero otherwise. The exchange rate model therefore is of the
following form:

[ER.sub.t] = [alpha] + [[beta].sub.1] [TER.sup.USA.sub.t] +
[[beta].sub.2][TER.sub.t] + [[beta].sub.3][TER.sub.t-144] +
[[beta].sub.4][TER.sub.t-2SS] + [[epsilon].sub.t].

The regression is estimated using a GARCHf 1.1) model using a
Student's t distribution. All coefficients are multiplied by 1,000
for ease of presentation, p Values testing the null hypothesis that
the slope coefficient is zero are reported in parentheses.

* denotes statistical significance at the 1% level.