The Economic Consequences of National Security Threats: The Case of the Korean Peninsula

ABSTRACT This paper examines the impact of national security threats on a nation’s economic growth and fiscal policy based on a case study of the Korean peninsula. I construct four measures of provocations using a newly-assembled list of North Korean provocative events going back to 1960. The results show that the overall impact of North Korean provocations on South Korea’s short-run economic growth is negligible. Since inter-Korean relations have gone through four phases, this paper also estimates the impact of provocations over each subperiod. Provocations had a significant impact on South Korea’s economic growth during 1960-1970 and 1992-1997 when inter-Korean tensions were high, but the effects took on different signs. While provocations decreased South Korea’s economic growth during 1992-1997, it had a positive impact on South Korea’s macroeconomy before 1970. This paper provides evidence that the effect of national security threats may vary with the responses from the government and political factors such as the relation between the targeted country and the country that inflicts the threat.


Introduction
National security constitutes an important part of a government's budget.Global spending on national defense totaled around 1.7 trillion dollars in 2017, which is equivalent to 2.2 percent of global gross domestic product (GDP) (Man et al. 2018).Government agencies such as the National Security Council in the United States systematically monitor the safety of the nation, and global think tanks such as the Institute for Economics and Peace are dedicated to developing global threat measures.Despite the large amount of resources devoted to defending national security, the macroeconomic consequences of national security threats are not clear.While there is evidence that national security threats increase uncertainty and lower output (Caldara and Iacoviello 2022), some argue that an economy can grow faster in the presence of threats if government expenditure rises (Aizenman and Glick 2006).The purpose of this paper is to systematically investigate the impact of national security threats on a nation's economic growth and fiscal policy using the Korean peninsula as a case study.
The Korean peninsula is a suitable candidate for the study of national security threats because North Korea's continuous provocations since the division of Korea constitute an important threat to South Korea's national security.Previous research has mostly focused on the response of South Korean financial markets to these North Korean threats, with somewhat mixed results. 1 For example, Kim and Roland (2014) perform an event analysis and find no significant impact of North Korean This paper is also related to the literature that studies the effect of collective violence such as wars and terrorist attacks (Ahern 2018;Brück, Llussá, and Tavares 2011;Gaibulloev and Sandler 2009;Blomberg S, Gregory D, and Orphanides 2004;Eckstein and Tsiddon 2004;Tavares 2004;Abadie and Gardeazabal 2003).While many of these papers estimate the effect of violence using a cross-country framework, I carry out a time series analysis of the interstate threats on the Korean peninsula.My results share some similarities with Brück, Llussá, and Tavares (2011) and Ahern (2018) in the sense that national security threats may have a positive effect on the economy.Brück, Llussá, and Tavares (2011) show that terrorist attacks increase entrepreneurial activities using a panel of 43 countries, and Ahern (2018) finds that terrorism leads to an increase in economic output by studying European regions which are not directly affected by terrorist attacks.This paper suggests that national security threats may promote economic growth but the impact of threats is contingent on the responses from the government.
The rest of the paper is organized as follows.Section 2 describes the data used to construct the provocation measures and relevant economic variables.Section 3 establishes the statistical exogeneity of North Korean provocations.Section 4 discusses the estimation strategy and presents the main results.Section 5 considers a number of robustness checks.Section 6 concludes.

Data
This section provides a description of the data used in this paper.South Korean macroeconomic series come from the Bank of Korea.This paper uses data on South Korea's GDP, components of GDP, and military spending.These macroeconomic series begin in 1960 and are available at a quarterly frequency. 2All data are seasonally adjusted and are deflated with their own 2010 deflators.The growth rates are computed as the first differences of natural logarithms.
Information about North Korean provocations comes from two sources: the Congressional Research Service (CRS) and the Center for Preventive Action (CPA).The CRS conducts research for the United States Congress.The CRS report includes North Korean provocations against South Korea, the United States, and Japan between June 1950 and March 2007.This paper selects those 148 events from the CRS list that are between North and South Korea only. 3The CPA belongs to the Council on Foreign Relations, a think tank specializing in U.S. foreign policy and international affairs.The CPA report describes military clashes on the Korean peninsula between 1955 and 2010.Among the military clashes in the CPA list, 163 events are provocations initiated by North Korea. 4Although both reports provide information about North Korean provocations, there is a slight difference in the choice of events.The CRS report (Fischer 2007) includes a variety of events such as hijacking, kidnapping, infiltration of spies, and tests of missile or nuclear technology.The CPA report (Wittels 2010) mainly focuses on military conflicts, and it provides a more detailed coverage of the military incidents between the two Koreas than the CRS report does.Given this distinction, events from both reports are included in the sample of this paper. 5The sample period is restricted to be between 1960q1 and 2006q4, the period when we have events from both CRS and CPA reports.
Table 1 provides a summary of North Korean Provocations.The majority of provocations are military conflicts, i.e. exchange of fire between South and North Korean soldiers.The second largest group consists of missile and nuclear threats.Despite the sanctions against its missile and nuclear program, North Korea has made considerable progress in its missile and nuclear technology and relies on it to boost its military capability.North Korean agents also attempted to hijack South Korean airlines and kidnap South Koreans living in different countries.This kind of event makes up 7% of total provocations.Assassinations and bombings are the most lethal provocations in terms of the number of casualties.For example, a foiled assassination attempt on the South Korean president in January 1968 caused 100 deaths and 69 injuries (Fischer 2007).North Korea also sent spies to South Korea to collect intelligence information, and 3% of provocations are related to such spy activities.Furthermore, North Korea used verbal threats to intimidate South Korea's government and media outlets.In July 2000, North Korea lashed out at South Korean daily newspaper Choson Ilbo for publishing stories related to the death of the North Korean president.Finally, any provocations that do not belong to any of the categories are classified as 'others'.
Table 2 classifies North Korean provocations according to the time of occurrence.North Korea has maintained a steady stream of provocations over the last five decades.The number of provocations seems to have risen in recent years as North Korea tests its missile and nuclear device more frequently.However, the recent provocations were less deadly than the provocations in the 1960s.While the provocations in the 1960s caused a total of more than 800 casualties, the early 21 st century had only 35 casualties related to North Korean provocations. 6 With this information, I create four variables for North Korean provocations.PROV-D, a dummy variable, indicates whether a given quarter contains a provocation or not.PROV-N equals the number of provocations in a given quarter.PROV-L intends to capture the severity of provocations based on the number of casualties.Specifically, PROV-L takes on four values, ranging from zero to three.Zero means no provocations in a quarter, and one indicates that there were provocations but no casualties.PROV-L takes on the value two if the total number of casualties from provocations in a given quarter is less than or equal to five.PROV-L is three if the number of casualties exceeds five.It is possible that nuclear and missile tests are perceived by South Korea as serious threats despite its lack of casualties.Therefore, the fourth measure PROV-L' accounts for this possibility by assigning the highest severity level, three, to the incidents which are nuclear or missile tests.
Figure 1 plots these provocation variables over the sample period.The gray bars indicate Korean recession periods calculated with the Bry-Boschan algorithm. 7I also labeled the x-axis with North Korean leaders corresponding to the period when they had power.Figure 1 suggests that there is substantial difference in the number and the severity level of provocations from one quarter to another.The frequency of provocations does not seem to be affected by the political regimes of North Korea.

Exogeneity Tests
Identifying the causal effect of provocations on South Korea depends on the condition that North Korean provocations are exogenous to events in South Korea.If North Korean provocations are not exogenous, changes in an economic series of South Korea will affect the provocative incidents, which may in turn affect South Korea's economic conditions.Are North Korean provocations exogenous?There is some evidence that North Korea provokes strategically, but it is debatable whether or not these provocations are related to the economy of South Korea.The study by Michishita (2010) states that the underlying causes of North Korean provocations include both external events on the Korean peninsula and internal conditions within North Korea.According to Michishita, North Korea has attempted to use military action to hamper U.S. reconnaissance activities and overthrow the South Korean government.These objectives motivated events such as the failed assassination attempt of President Park Chung-hee in January 1968 and the damaging of a U.S. reconnaissance plane in April 1969.Michishita also suggests that North Korean leaders may have used provocations to strengthen their political power.For example, North Korean President Kim Il-sung used the tension created by military conflicts near the border to justify his 'military-first' policy (Michishita 2010, 31).President Kim also PROV-N represents the number of provocations in a given quarter.PROV-L proxies the severity level of provocations.PROV-L is zero if there are no provocations in a given quarter.PROV-L is one if there were provocations but no casualties.PROV-L has value two if the total number of casualties from provocations in a quarter is less than or equal to five.PROV-L is three if the number of casualties exceeds five.PROV-L' is the same as PROV-L, except that PROV-L' is three if the number of casualties exceeds five or the incident is a nuclear or missile threat.The gray shadings are South Korea's recession periods.
appeared to initiate the Pueblo incident in January 1968 to purge his political rivals and consolidate his position in the military establishment.Lastly, Michishita mentions that North Korea may have hoped to benefit economically from some of the provocations.The two countries' confrontations in the Yellow Sea have succeeded in deterring South Korean fishermen from fishing near the North Korean coasts.As a result, North Korea is able to reap larger gains from fishing in bigger areas.
Since the historical evidence on the exogeneity of North Korean provocations is mixed, I use statistical methods to determine whether North Korean provocations are exogenous to the South Korean macroeconomy and other macroeconomic developments on the southern part of the peninsula.Two tests are conducted.First, I examine whether North Korea provokes in response to the economic conditions of South Korea.I focus on the series of GDP growth, government spending growth, military spending growth, and a dummy variable for South Korean recession periods. 8If North Korea becomes hostile to its neighbour when it feels threatened by South Korea's economic success or military strength, one of those series would have predictive power for the provocations.Second, I investigate whether North Korean provocations are reactions to news about South Korea's economy with the forward-looking variable non-durable consumption growth. 9If North Korea reacts to news on South Korea's economy, we should see a change in non-durable consumption growth before the provocations, and non-durable consumption growth should predict the latter.
To test the exogeneity of North Korean provocations, I estimate a univariate regression where I regress the provocation variable on its own lags and the lags of economic variables.I use a separate equation for each economic variable, and the lag length is selected by the Hannan and Quinn (HQ) information criterion. 10If economic variables do not predict North Korean provocations, the coefficients on the lagged economic variables should be jointly zero.I test the null hypothesis that the lagged economic variables do not enter the equation of provocations using an F test.
Table 3 summarizes the results from these exogeneity tests using the provocation dummy variable, PROV-D. 11The first column lists a series of null hypotheses, such as 'GDP growth does not predict North Korean provocations'.The second column presents the p-values from the F test.A low p-value rejects the null hypothesis and implies that the relevant variables have predictive power for North Korean provocations.The last column lists the lag length chosen for each regression.
Except for the case with non-durable consumption growth, the exogeneity tests fail to reject the null hypothesis that the selected economic series do not predict the provocation dummy variable.The p-values for South Korea's GDP, military spending, and the recession dummy are quite high.The p-value for government spending growth is lower, but it is still higher than the 10% significance level.
Non-durable consumption growth appears to predict North Korean provocations based on the exogeneity tests.To gain a deeper understanding on the predictive power of non-durable consumption growth, I conduct the exogeneity tests with non-durable consumption growth using each category of provocations.Table 4 presents the p-values from the exogeneity tests for the seven categories of provocations discussed in Section 2. The results suggest that non-durable consumption growth has predictive power for three out of the seven categories.The groups of provocations that are not predicted by non-durable consumption growth are hijacking and kidnapping, assassinations and bombings, spy activities, and intimidations.
Since the data for non-durable consumption starts in 1970q1 and the sample period is from 1960q1 to 2006q4, I cannot directly control for the non-durable consumption growth in the regression when estimating the effect of provocations.In order to test whether this potential endogeneity problem with non-durable consumption growth affects the results of the estimation, I conduct two tests as robustness checks.First, I run the baseline regression using the categories of provocations which are not predicted by the non-durable consumption growth.Second, I restrict the sample period to 1970q1-2006q4 and control for the non-durable consumption growth in the estimation.These robustness checks suggest that the predictive power of non-durable consumption growth on provocations has a limited effect on the regression results.

Baseline Results
This section examines the effect of North Korean provocations on the quarterly growth rates of South Korea's GDP and various components of GDP using the baseline provocation dummy variable, PROV-D.The section is divided into three parts.First, I examine the overall effect of provocations on the economy of South Korea.Second, I consider the possibility that the effect of provocations varies with inter-Korean relations and estimate the effect over four subperiods.Third, I study possible channels through which provocations affect South Korea's economy.

I begin by estimating a regression of the following form:
where Y is the natural logarithm of the real South Korea's economic variable.PROV-D is the provocation dummy variable with up to N lags included.I set N to one to estimate the contemporaneous and once-lagged effect of provocations.The controls include up to M lagged values of the dependent variable, and the lag value M is chosen according to the HQ information criterion. 12It is well known that South Korea has experienced the IMF crisis at the end of 1997 which severely affected South Korea's macroeconomy.Therefore, a dummy for IMF crisis is also included in the regression.Table 5 presents the estimated effects of North Korean provocations on the quarterly growth rates of South Korea's GDP and components of GDP.Provocations have no contemporaneous and lagged effects on the growth rates of GDP, private consumption, exports, and imports.However, provocations significantly decrease the growth rate of investment.The results also suggest that government spending increases in the quarter after provocations.
Overall, Table 5 implies that the effect of provocations on South Korea's economic growth is negligible when estimated over the entire sample.Although investment decreases and government spending increases after provocations, the overall effect of provocations on South Korea's GDP growth is not statistically different from zero.

Subsample Analysis
The previous analysis assumes a constant effect of provocations over time, but there is reason to believe that North Korean provocations had a varied impact on South Korea's economy over the years.Historical evidence suggests that the two Koreas have gradually moved into a more peaceful coexistence, but North Korea's continual attempts on its nuclear weapon program hamper the progress (Agov 2016).This subsection considers a regression specification which allows for the possibility that the effect of provocations varies with inter-Korean relations.
A careful study of South Korea's history suggests that inter-Korean relations have gone through four phases between 1960 and 2006 (Agov 2016).The first period lasts until 1970.This period witnessed severe clashes on the Korean peninsula.North Korea plotted to destabilize South Korea with their infiltrations and continual assassination attempts on the South Korean president.The most famous case is the Blue House raid where North Korean commandos were discovered within hundreds of meters from the presidential mansion, the Blue House (Michishita 2010).For the next two decades 1971-1991, the inter-Korean relation oscillated between conflict and reconciliation.The two Koreas started drawing plans for unification, but they were abandoned as anticommunism aroused in South Korea.The inter-Korean tension escalated between 1992 and 1997.South Korea became increasingly concerned with North Korea's pursuit of nuclear weapons, and the relations deteriorated further when North Korea refused to comply with the international denuclearization treaty.The rise of South Korean President Kim Dae-jung in 1998 ushered in a period of warming relations which lasted until 2007.He promoted reconciliation with the North and went to North Korea for the first Inter-Korean Summit in 2000.The two Koreas made  1960-1970, 1971-1991, 1992-1997, and 1998-2006.Regression equation ( 1) is therefore modified so that the effects of provocations are indexed by period.The modified equation is: Compared to the equations ( 1), (2) replaces PROV-D t by four PROV-D t 1 �t�t 2 terms with t 1 and t 2 being specific years between 1960 and 2006.The variable PROV-D t1�t�t2 is defined as PROV-D The time indicator takes on unity if t is between t 1 and t 2 , and zero otherwise.The lag value N is still one, and I use the same controls as before in the regression. 13 Table 6 presents the subsample estimates of the effect of provocations on the same set of economic variables as Table 5.The results suggest that South Korea's economic growth was most affected by North Korean provocations during 1960-1970 and 1992-1997, the two phases when inter-Korean tensions were high.However, the effects of provocations during these two phases took on different signs.During 1960During -1970, provocations increased the growth rates of GDP,  : 1960-1970, 1971-1991, 1992-1997, and 1998-2006.The regression follows equation (2) which includes lagged economic variables, a dummy for IMF crisis, and the interaction terms between PROV-D and the time indicators.The coefficients are estimated by the OLS method.Values of heteroscedasticity robust t-statistics are provided in parentheses.The lag length in the regressions follows the optimal lag length used in Table 5.The last row shows the unconditional standard deviation of the dependent variables.*** significant at 1%. ** significant at 5%. * significant at 10%.
imports, and government spending.By contrast, provocations had a negative lagged effect on GDP, investment, and imports between 1992 and 1997.Provocations did not affect GDP growth rate in the other periods, but some components of GDP still responded to provocations.During 1971-1991, investment and exports decreased, and government spending increased after provocations.In the latest period of the sample 1998-2006, provocations led to a decrease in investment and exports.In summary, Table 6 shows that the effect of provocations on South Korea's economic growth is concentrated in the periods when inter-Korean tensions were high, namely in the 1960-1970and the 1992-1997periods. Between 1960and 1970, provocations led to an increase in the quarterly growth rates of GDP.The positive effect seemed to wear off over time, and the effect of provocations on GDP growth rate was significantly negative during 1992-1997.

Channels
The empirical evidence suggests that North Korean provocations had a positive impact on South Korea's economic growth rate in the 1960s and a negative impact between 1992 and 1997.In this subsection, I investigate the potential channels through which provocations affect South Korea's economic growth.

Effect of Provocations in the 1960-1970
The literature has suggested that the threats to national security can affect the economic growth in the following three ways.First, the threats might increase the uncertainty of the economy.This increase in uncertainty may delay investment which, in turn, can affect the GDP growth.Second, the attacks might destroy infrastructure and interrupt the production.Third, government spending or military spending might increase in response to the attacks.Among these explanations, only increased military spending is likely to cause an increase in the GDP growth rate.Therefore, I investigate whether the positive effect of North Korean provocations on South Korea's GDP growth between 1960 and 1970 is driven by changes in South Korea's military spending.
Table 7 shows the estimated effect of North Korean provocations on the quarterly growth rate of South Korea's military spending.Column (1) shows the effect estimated over the entire sample period and column (2) shows the effect estimated over the four subsample periods.I find that overall, provocations do not have a significant effect on the growth rate of South Korea's military spending.However, the subsample analysis suggests that provocations led to an immediate increase in South Korea's military spending between 1960 and 1970.This result provides support to the explanation that the increase in South Korea's GDP growth during 1960-1970 following North Korean provocations is driven by an increase in South Korea's military spending. 14 To further test the role of military spending in the effect of North Korean provocations on South Korea's GDP growth, I examine the impact of provocations on quarterly GDP growth rate while controlling for the growth rates of military spending.Table 8 shows that the effect of provocations on South Korea's GDP growth rate before 1970 is not statistically different from zero once the growth rate of government spending or military spending is controlled for.Overall, the results in Table 8 suggest that military spending plays an important role in explaining the positive effect of provocations on the GDP growth rate before 1970. 15

Effect of Provocations in the 1992-1997
During the period of 1992-1997, North Korean provocations had a significant negative impact on South Korea's GDP growth rate but did not affect South Korea's military spending.The impact of provocations during 1992-1997 may be explained by firms' investment response to uncertainty in South Korea.Assuming that the investment is irreversible, higher uncertainty caused by North Korean provocations will encourage firms to postpone investment and therefore decreases the GDP growth rate in South Korea.The empirical evidence presented in Table 6 seems to support this explanation.During 1992-1997, the results show that the growth rate of investment in South Korea decreased significantly following the onset of North Korean provocations.
However, investment can be affected by investors' confidence and expectations of economy rather than uncertainty.To test whether North Korean provocations indeed increased uncertainty between 1992 and 1997, I estimate the effect of provocations on uncertainty in South Korea where uncertainty is measured by the volatility of stock price index.Figure 2 plots the volatility of the Korea Composite Stock Price Index (KOSPI).I computed the monthly standard deviation of the daily returns on the KOSPI and converted the series to a quarterly frequency by simple averaging.The figure shows that the volatility of stock price index fluctuates from quarter to quarter, and it increased substantially when the IMF crisis happened at the end of 1997.
To test whether provocations lead to higher uncertainty, I replicate equation ( 2) but using the volatility of KOSPI as a dependent variable.The estimated effect of provocations on volatility over different subperiods are shown in Table 9. Column (1) uses the baseline measure of provocations which takes the value one if a given quarter contains a provocation.The point estimates of the effect  : 1960-1970, 1971-1991, 1992-1997, and 1998-2006.The coefficients are estimated by the OLS method.Values of heteroscedasticity robust t-statistics are provided in parentheses.The lag length in column ( 1) is chosen optimally by the HQ information criterion and used also in column (2).The last row shows the unconditional standard deviation of the dependent variables.*** significant at 1%. ** significant at 5%. * significant at 10%.Notes: Dependent variables are the first differences of natural logarithms of real GDP.Column (1) repeats the baseline result in the first column of Table 6.Column (2) estimates: and G is the natural logarithm of real government spending.Column (3) estimates the same equation as column (2), except that G is replaced by the natural logarithm of real military spending.The coefficients are estimated by the OLS method, and the controls include lagged dependent variables and a dummy for IMF crisis.Values of heteroscedasticity robust t-statistics are provided in parentheses.The lag value M in column (1), chosen optimally by the HQ information criterion, is applied to all regressions.*** significant at 1%. ** significant at 5%. * significant at 10%. of provocations during 1992-1997 are positive, but they are not statistically different from zero.Column (2) restricts the set of provocations to those with at least one casualty.The results show that the provocations with casualties increased uncertainty significantly during 1992-1997.
Overall, the empirical evidence seems consistent with the explanation that North Korean provocations decreased the growth rate of South Korea during 1992-1997 by increasing uncertainty and lowering investment in the economy.

Response of Military Spending to North Korean Provocations
One may wonder why the response of military spending to North Korean provocations was only significant in the 1960s.This subsection offers one potential explanation to this result.South Korea's main military objective is to deter an attack by North Korea.The pressure to build up military strength would be stronger when the size of the economy is smaller than that of the enemy state.As Figure 3 shows, North Korea had a similar GDP per capita as South Korea during the 1960s, and it held a slight technological edge over South Korea by controlling heavy and chemical industries from the aftermath of the Korean War (Lee 1995).It was also widely believed that North Korea had a more powerful armed forces than South Korea at that time (Choi 1985).Therefore, it is reasonable that South Korean government significantly increased the military spending to strengthen its military power after North Korean provocations in the 1960s.
By the early 1970s, South Korea had surpassed North Korea in terms of economic growth, and its GDP per capita became 16 times larger than that of North Korea at the end of 2006.The economic development allows South Korea to modernize military equipment, and South Korea's military is now seen as having qualitative and technological advantages over that of North Korea (Suh 2004).As a result, South Korean government in recent periods are likely to dismiss North Korean provocations as it holds a dominant position over North Korea. 16  Table 9.Estimated effect of provocations on the stock market volatility in South Korea. (1) ( Notes: Dependent variables are the monthly standard deviation of the daily returns on the KOSPI averaged at the quarterly frequency.The data of KOSPI begins in 1975.Column (1) estimates the effect of provocations using the baseline provocation measure PROV-D.Column (2) restricts the set of provocations to those with at least one casualty.The coefficients are estimated by the OLS method, and the regressions control for the lagged dependent variables and a dummy for IMF crisis.The lag value M in column ( 1) is chosen optimally by the HQ information criterion and used also in column (2).Values of heteroscedasticity robust t-statistics are provided in parentheses.The last row shows the unconditional standard deviation of the dependent variables.*** significant at 1%. ** significant at 5%. * significant at 10%.

Robustness
This section considers a series of robustness checks to the baseline results.First, I show that the baseline results are robust to alternative measures of North Korean provocations.Second, I show that we obtain similar results if we focus only on major types of North Korean provocations such as hijacking, kidnapping, assassinations and bombings.Finally, I show that the baseline results are qualitatively the same even after controlling for lagged non-durable consumption growth.
Alternative Measures of Provocations -In the baseline regression, the measure of North Korean provocations is a dummy variable, which takes the value one if a given quarter contains at least one provocation.However, the potential concern with using the dummy variable is that it does not capture the severity and the rate of occurrence of North Korean provocations, which have not been constant over the years.Therefore, in this robustness check, I replace PROV-D, the dummy measure of North Korean provocations, with PROV-N, PROV-L, PROV-L' which are the three alternative measures of North Korean provocations constructed based on the number of provocations and the number of casualties resulting from the provocations.
In the online appendix Table B.1, I present the estimated effect of provocations on GDP growth rate using four different measures of provocations.The estimates are qualitatively the same no matter which measure of provocations is used in the regression.The impact of provocations on South Korea's economic growth is always positive and statistically significant during 1960-1970.The estimates of the lagged effect of provocations between 1992 and 1997 are all negative, although they are not statistically significant when PROV-L and PROV-L' are used.
I also examine the impact of provocations using different provocation measures while controlling for the growth rate of military spending (see online appendix Table B.2). Except for the regression with PROV-N, provocations do not significantly affect the growth rate of GDP between 1960 and 1970 when military spending is controlled for.Unlike PROV-L, the results using PROV-N may be driven by small provocations as they count each provocation equally regardless of its severity.When I modify PROV-N to include only provocations with at least three casualties, the results are again consistent with the baseline findings.To sum up, the baseline results are robust to alternative measures of North Korean provocations.Categories of Provocations -The exogeneity tests fail to rule out the possibility that provocations are reactions to South Korea's economic news.As a result, the estimates of the effect of provocations may be biased due to the effect of news in South Korea.Fortunately, certain categories of provocations do not suffer from this problem.Therefore, I conduct robustness checks to the baseline results by focusing on the categories of provocations which have events before 1970 and pass the exogeneity tests with non-durable consumption growth rate.
In the online appendix Table B.3, I show the effect of three types of North Korean provocations on South Korea's GDP growth with and without controlling for military spending growth.The first category includes events where North Korea attempted to hijack South Korean airlines and kidnap South Koreans living in different countries.The second category includes events that are related to assassinations and bombings.Events in the third category include situations where North Korea used verbal threats to intimidate South Korean government.Regardless of the categories, provocations always had a positive effect on South Korea's GDP growth rate between 1960 and 1970.The estimates of the lagged effect of provocations between 1992 and 1997 are negative when we look at events related to hijacking, kidnapping and intimidation, but the estimate is only significant in the regression with hijacking and kidnapping.
After controlling for military spending growth, the effect of provocations becomes statistically insignificant for the category of hijacking and kidnapping.The effect of the other two categories of provocations in the 1960s are still positive after controlling for military spending growth, but the estimates become significantly smaller in all regressions.Overall, there is no evidence that the baseline results are driven by the potential endogeneity of North Korean provocations.The results are qualitatively the same when I focus on the categories of provocations that are not predicted by the forward-looking variable non-durable consumption growth.
Additional Controls in the Regressions -In this set of results, I modify the baseline regression to include lags of non-durable consumption growth rate as additional controls.Since the data of nondurable consumption is only available from 1970 onwards, I restrict the sample to be 1971-2006 and estimate the effect of provocations for the subperiods : 1971-1991, 1992-1997, and 1998-2006.The results are presented in the online appendix Table B.4.I find that the estimated effects of provocations since 1971 are almost the same with and without controlling for the non-durable consumption growth rate.North Korea provocations still had a significant negative impact on South Korea's GDP growth in the period of 1992-1997, and the effect during other periods are not statistically different from zero.These findings alleviate the concern that the baseline results are biased due to the potential endogeneity of North Korean provocations.

Conclusion
Countries spend a significant amount of resources on defending national security, but the macroeconomic consequences of national security threats are not clear.This paper examines how national security threats impact a nation's short-run economic growth and fiscal policy based on a case study of the Korean peninsula.
This paper shows that the effect of national security threats needs not be constant.The impact of North Korean provocations on South Korea's economic growth varies considerably over four subperiods, which are categorized by various degrees of adversarialness in inter-Korean relations.The results also suggest that the effect of national security threats can depend on domestic economic conditions as well as the responses from the government.From 1960 to 1970, South Korea's military spending rose significantly after provocations, which may potentially explain why provocations had a positive effect on South Korea's GDP growth rate during that period.From 1992 to 1997, the negative effect of provocations on South Korea's GDP growth rate may be attributable to the decreasing investment and rising uncertainty.Policies that design to reduce the cost of national security threats need to take into consideration that the effect of national security threats can depend on the responses from the government and political factors such as the relation between the targeted country and the country that inflicts the threat.

Notes
1.One exception is Lee, So, and Ha (2018)  Korean army troops made an unauthorized entry into the Demilitarized Zone. 5.There are 58 events reported in both CRS and CPA reports.Hence, the final list includes 253 events of North Korean provocations.6.The casualties include people who are dead, injured, and captured during the incident.7. South Korea's recession periods are calculated with the Bry-Boschan algorithm using the real GDP level.Official recession dates, as the NBER provides them in the United States, are not available.8.The recession dummy takes on the value one during the recession periods indicated by the Bry-Boschan algorithm, and zero otherwise.9.The data for South Korea's real non-durable private consumption is from the Bank of Korea, and has been available since 1970.The quarterly growth rate is computed as the first difference of the natural logarithm.10.Liew (2004) finds that, with relatively large samples (120 or more observations), the HQ information criterion surpasses the Akaike's information criterion, the Schwarz information criterion, and the final prediction error criterion in correctly identifying the true lag length.11.The results of the exogeneity tests using the other provocation measures are presented in the online appendix (see Table A.1). 12.In the baseline regressions, the lag value M for each economic variable is chosen optimally by the HQ information criterion.I also tried applying the lag value M for the GDP growth, chosen optimally by the HQ information criterion, to all regressions.The estimated coefficients using alternative lag values are very similar to the baseline results.13.The lag value of the dependent variables M is set to be the same as the value used in estimation with the entire sample.I also run the regressions using the lag value chosen optimally according to the HQ information criterion.The results are qualitatively the same.Smith 1968).15.The Sobel test is conducted to test whether the indirect effects of North Korean provocations via military spending of South Korea are statistically significant.I follow Sobel (1982) and estimate the coefficients and t-statistics of the indirect effects.The results suggest that the indirect effects of North Korean provocations via military spending in the 1960s were positive and statistically different from zero.A detailed description of how I conduct the Sobel test is summarized in the online appendix (see Appendix C). 16.One may also argue that South Korean government in recent years is in favor of a policy that pushes for peace with North Korea.Therefore, South Korea may be reluctant to increase military spending after provocations as it may exacerbate military tension between the two Koreas.

Figure 1 .
Figure1.Provocation variables, 1960q1-2006q4.PROV-D is the provocation dummy variable, and it takes the value one or zero.PROV-N represents the number of provocations in a given quarter.PROV-L proxies the severity level of provocations.PROV-L is zero if there are no provocations in a given quarter.PROV-L is one if there were provocations but no casualties.PROV-L has value two if the total number of casualties from provocations in a quarter is less than or equal to five.PROV-L is three if the number of casualties exceeds five.PROV-L' is the same as PROV-L, except that PROV-L' is three if the number of casualties exceeds five or the incident is a nuclear or missile threat.The gray shadings are South Korea's recession periods.

Figure 2 .
Figure 2. Volatility of stock price index for South Korea, 1975-2006.The KOSPI index is sourced from the Korea Stock Exchange and downloaded via Datastream.The data for the KOSPI is available since 1975.The figure plots the monthly standard deviation of the daily returns on the KOSPI.The series is converted to a quarterly frequency by simple averaging.

Figure 3 .
Figure 3. GDP per capita, South Korea vs North Korea, 1960-2006.The data is sourced from the Maddison Project via its website.The figure plots the yearly GDP per capita for South Korea and North Korea between 1960 and 2006.The solid line represents South Korea, and the dash-dotted line represents North Korea.
Provocations are sourced from the Congressional Research Service reports and the Center for Preventive Actions.

Table 1 .
Categories of North Korean provocations.
0 : Non-durable consumption growth ⇏ PROV-D 0.00 3 Notes: All series, except non-durable consumption which starts in 1970q1, are from 1960q1 to 2006q4.I regress the provocation dummy variable on its own lags and the lags of macroeconomic variables.I estimate a separate regression for each macroeconomic variable.The lag length is chosen by the HQ information criterion.I apply the F test to examine whether the coefficients on the lagged macroeconomic variables are jointly zero.P-values from the F test are reported.

Table 4 .
Predictive power of non-durable consumption growth for each category of provocations.The non-durable consumption growth ranges from 1970q1 to 2006q4.For each category of provocations, I regress the corresponding provocation dummy variable on its own lags and the lagged non-durable consumption growth.The lag length is chosen by the HQ information criterion.I apply the F test to test whether the coefficients on the lagged non-durable consumption growth are jointly zero.P-values associated with the F test are reported in the table.

Table 5 .
Overall effect of provocations on the first differences of natural logarithms of South Korean economic variables.Dependent variables are the first differences of natural logarithms of real GDP, real private consumption, real investment, real exports, real imports, and real government spending.The coefficients are estimated by ordinary least squares (OLS) regressions which include lagged economic variables, a dummy for IMF crisis, contemporaneous and lagged PROV-D.The first row shows the estimated immediate impact of provocations.The third row shows the estimated impact of provocations one quarter after their occurrences.Values of heteroscedasticity robust t-statistics are provided in parentheses.The lag value M is chosen optimally for each regression according to the HQ information criterion.The last row shows the unconditional standard deviation of the dependent variables.***significant at 1%. ** significant at 5%. * significant at 10%.significant progress in their economic cooperation, and the Kaesong Industrial Complex was established in 2005 where South Korean businesses could manufacture goods with labor from North Korea.As a result, I divide the sample into four periods:

Table 6 .
Subsample estimates of the effect of provocations on the first differences of natural logarithms of South Korean economic variables.Dependent variables are the first differences of natural logarithms of real GDP, real private consumption, real investment, real exports, real imports, and real government spending.The table presents the estimated effect of provocations over four periods

Table 7 .
Subsample estimates of the effect of provocations on the first differences of natural logarithms of South Korean military spending.
Notes: Dependent variables are the first differences of natural logarithms of real military spending.Column (1) presents the estimated effect of provocations over the entire sample period.Column (2) presents the estimated effect of provocations over four periods

Table 8 .
Effect of provocations controlling for government/military spending.
which use North Korea-related events between 2003 and 2017 as instruments to study the macroeconomic impact of uncertainty shocks.2. Data on South Korea's GDP, components of GDP, and industrial GDP is downloaded from the Bank of Korea's Economic Statistics System.Accessible online at http://ecos.bok.or.kr/flex/EasySearch_e.jsp.Before 2000, South Korea had no data on government investment expenditure.Therefore, the government spending series in this paper is a consumption series.Data on military spending is also sourced from the Bank of Korea but downloaded via Datastream.The series is titled South Korea's public administration and defense spending series (KOGDPPUBD).3. The following 12 events in the CRS list are thus excluded from the sample: 1970MarchNorth Korea provided sanctuary to a Japanese radical group; 1997MarchJapan's daily newspaper disclosed that North Korea abducted a Japanese in the mid-1970s; 1999MarchNorth Korean spy ships entered Japanese territorial waters; 2000MarchNorth Korea insisted to provide shelter to a Japanese radical group; 2000OctoberNorth Korea and the United States issued a joint statement agreeing to oppose all forms of terrorism; 2001SeptemberNorth Korea admitted of kidnapping Japanese in the 1970s and the 1980s; 2001Decembera North Korean spy ship was sunk by the Japan's coast guard; 2002MarchThe Guinea-Bissau navy discovered a North Korean ship carrying machine guns and ammunition; 2002OctoberA North Korean ship en route to Yemen was found carrying Scud missiles; 2003JanuaryA North Korean diplomat was quoted as saying that North Korea would not attack South Korea if it was attacked by the United States; 2004DecemberNorth Korea failed to return the remains of a Japanese girl kidnapped by North Korean agents; 2005SeptemberU.S. treasury charged Banco Delta Asia for drug trafficking activities with North Korea.4. The following 7 events in the CPA list are excluded from the sample because they were initiated by South Korea: 1966OctoberA South Korean attack team penetrated North Korean territory to mount a raid; 1967NovemberSouth Korean infiltrators blew up North Korean division headquarters; 1975FebruaryA South Korean destroyer sunk a North Korean fishing vessel loitering near the Northern Limit Line; 1978AprilSouth Korean patrol boat attacked a North Korean boat; 1978JulySouth Korean patrol boat attacked a North Korean boat; 2005JuneA South Korean soldier detonated a grenade near a South Korean encampment; 2006JulySouth 14. Articles published in South Korean newspapers show that South Korea's military spending indeed increased immediately after North Korean provocations between 1960 and 1970, consistent with the military spending channel proposed in this paper.For instance, following the attempted assassination of South Korean president by North Korean commandos on 21 January 1968, the Chosun Ilbo reported a special meeting between South Korea's major and minor parties which took place on January 28 ('Parties issued a joint statement, calling for special U.S. support and a summit if necessary 1968).During the six-hour meeting, the parties issued a joint statement to request immediate monetary support from the United States to modernize military equipment.The appeal from South Korea for an immediate increase in military aid also appeared in American newspapers.The Bulletin, Lodi News -Sentinel and The Shreveport Times reported that President Johnson had asked Congress for an emergency appropriation of $100 million to strengthen South Korea's defense system (LBJ requests aid money for Koreans 1968; LBJ seeks Military Spending Boost for Korea 1968;