Political Conflict and Foreign Portfolio Investment: Evidence from North Korean Attacks

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Political Conflict and Foreign Portfolio Investment: Evidence from North Korean Attacks Jeffrey R. Gerlach and Youngsuk Yook 2016-037 Please cite this paper as: Gerlach, Jeffrey R. and Youngsuk Yook (2016). “Political Conflict and Foreign Portfolio Investment: Evidence from North Korean Attacks,” Finance and Economics Discussion Series 2016-037. Washington: Board of Governors of the Federal Reserve System, http://dx.doi.org/10.17016/FEDS.2016.037. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Political Conflict and Foreign Portfolio Investment: Evidence from North Korean Attacks JEFFREY R. GERLACH ∗ Federal ReserveBank ofRichmond YOUNGSUK YOOK† Federal ReserveBoard ofGovernors April2016 ABSTRACT We examine the response of foreign (i.e., non-SouthKorean)investorsto escalating political conflict and its impact on the South Korean stock market surrounding13 North Korean military attacksbetween1999and 2010. Usingdomestic(i.e., SouthKorean)institutionsandindividuals asbenchmarks,weevaluatethetradingbehaviorandperformanceofforeigninvestors. Following attacks, foreignersincrease their holdingsof Korean stocksand buy more sharesof riskystocks. Performanceresultsshowforeignersmaintaintheirpre-attacklevelofperformancewhiledomestic individuals,whomaketheoverwhelmingmajorityofdomestictrades,performworse. Inaddition, domesticinstitutionsimprovetheirperformance.Overall,theresultsareconsistentwiththepredictionsbasedonthebenefitsofinternationaldiversification. Unlikedomesticindividuals,foreigners trademoresharesthanusualanddeviatefromtheirgeneralstrategyofpositivefeedbacktrading. Keywords: Politicalconflict;Foreignportfolioinvestment;NorthKoreanattacks FederalReserveBankofRichmond;e-mail:jeffrey.gerlach@rich.frb.org;phone: +17043582517. ∗ †Federal Reserve Board of Governors, 20th Street and Constitution Avenue NW, Washington, DC 20551; e-mail: youngsuk.yook@frb.gov; phone: +1 202 475 6324. The views expressed in this article are those of the authors and not necessarilyoftheFederalReserveSystem. WethankseminarparticipantsattheFinancialManagementAssociationAnnual Meeting,HongKongBaptistUniversity,KAIST,andSeoulNationalUniversityforprovidingusefulcomments.

1. Introduction Emerging market economies command increasingly greater weight in foreign investment portfolios, makingforeigninvestorsmoresusceptibletotherisksassociatedwithpoliticallyoreconomicallyfragileregimes. Politicalconflictsarecommoninmanyareasoftheworld,including largepartsofAfrica, Asia,andtheMiddleEast. Accordingly, agrowingliteraturehasexaminedtheeconomiceffectofwar, terror, and more generally, political instability, providing insight into the impact of political conflicts ontheeconomyandfinancialmarkets.1 However,previousstudiestendtofocusontheaggregateeffect and do not distinguish the roles played by foreign and domestic investors. A host country’s political risk may have different implications for foreign investors than for domestic investors, leading them to respond differently. Thus, the role foreign investors play in the local equity market subsequent to political conflictscanbeverydifferentfromwhatisimpliedbytheeffectsaggregated overallinvestor types. The literature on international capital flows suggests that foreign and domestic investors can be motivated by different factors (e.g., Forbes and Warnock, 2012; Rothenberg and Warnock, 2011). A decomposition into foreign and domestic investors not only provides insight into foreign portfolio investment decisions but also has policy implications in that a different policy response may be required depending on whether an observed pattern is driven by foreigners or domestic investors. Our study attempts to fill this gap in the literature by analyzing the behavior of foreign (i.e., non-South Korean) investors separately from that of domestic (i.e., South Korean) investors surrounding events thatescalate geopolitical riskontheKoreanpeninsula. SincetheKoreanWarceased in1953withoutapeacetreaty,2 NorthKoreahasintermittently initiatedmilitaryconflictssuchasborderfightsandnavalbattles. Theseabruptattacksraiseconcernsabout thepossibilityofanall-outwarbetweenthetwoKoreas. Increasedinstabilityintheregioncandiscouragecorporateinvestmentanddomesticconsumption,whichinturncanhurtequitymarkets. According totheWallStreetJournal,politicalinstabilityisoneofthereasonswhyKoreanstockstradeatthelowestvaluation amongAsia’smajormarkets –abouteleven timesexpected earnings in2010.3 However, 1SeeSection2forareviewofrelevantliterature. 2TheKoreanWarstartedwhenNorthKoreainvadedSouthKoreain1950. Fightingceasedin1953withanarmisticethat restoredtheborderbetweenthetwoKoreasnearthe38thParallelandcreatedtheKoreanDemilitarizedZone(DMZ),abuffer zonebetweenthetwoKoreas.Technically,thetwoKoreasarestillatwar. 3The Wall Street Journal Online, Mohammed Hadi and James Simms, “As ties go south, Korean investors shrug,” 24 November2010. 1

North Korean risk is not limited to domestic investors in the Korean markets. The reclusive state also poses a potentially serious threat to neighboring countries such as Japan and Taiwan. Furthermore, today’s interconnected economies and well-diversified foreign portfolio investment imply that North Korea poses real political risk to many investors around the world. Indeed, Nomura Securities ranked theinter-Korean tensionastheworld’s5thmostseriousgeopolitical riskin2012.4 TheconflictbetweenSouthKoreaandNorthKoreapresentsauniquesettingtostudytheimpactof politicalriskonstockmarkets. First,thetimingoftheNorthKoreanattacksislargelyexogenous from theperspectiveofinvestors. AswediscussinSection3.3,theattacksaretosomeextentpredictable,but thereisstillalotofuncertaintyabouttheirexacttimingandmagnitude. Unlikesomecasesofincreased politicalriskinwhicheconomicfactorsmayhavecontributed tothepoliticalconflict,theattacksseem to be driven mostly by internal political processes in North Korea. Given that the country is insulated fromtherestoftheworld,itishighlyunlikely thetimingoftheattacksisinfluenced bydevelopments intheSouthKoreanstockmarket. Second,politicaleventsaroundtheworldtendtobeuniqueinnature and one-off developments, making itdifficult toobtain reliable results based onanextended period of time-series data. North Korean attacks are different in that they are recurring events with 13 attacks during the sample period of 12 years. Third, unlike manypolitical crises in which the precise starting date is hard to pinpoint (e.g., Willard, Guinnane, and Rosen (1996) and Zussman, Zussman, Nielsen (2008)), thesemilitaryattacksareobservable; thustheexacttimingofattackscanbeidentified. Our investigation uses a novel data set. The Korean stock market makes publicly available daily transactions data aggregated by three investor types: foreigners, domestic institutions, and domestic individuals. The breakdown into different investor types allows us to analyze the difference in difference by evaluating foreigners’ behavior before and after the attacks relative to that of domestic investors. Because foreign investors tend to be large institutions, having two sets of benchmarks, domesticinstitutionsanddomesticindividuals,alsohelpspindownwhethertheobservedtradingpatterns of foreigners are attributable to foreign characteristics or institutional characteristics. In addition, the Korean stock market is an attractive testing ground for several reasons. First, the market imposes no restrictions onforeign ownership during oursampleperiod of1999through 2010. Second, themarket has a high level of foreign participation with foreign ownership representing 32.9% of total market 4NomuraSecuritiesInternationalInc.,2011,GlobalFXOutlook2012. 2

capitalization in the KOSPImarket, South Korea’s main stock exchange in 2010 year-end. Third, the Korean stock market is large and liquid, with its annual turnover the 9th highest in the world and its total market capitalization the 17th highest in 2011. In sum, wehave a unique setting for testing how foreigninvestors respond tounexpected andrepeating politicalconflicts. Since our analysis is primarily focused on the response of foreign investors, it requires a sample of stocks with nontrivial foreign ownership. While the average foreign ownership stake in Korean stocks is high, the size of foreign ownership varies substantially across stocks. Kang, Lee, and Park (2010),forexample,documentthathalfofthestockslistedontheKoreaStockExchangehaveforeign ownership oflessthan1%forthe2000-2004 period. Assuch, anequal treatment ofallKoreanstocks cannot provide an accurate assessment of the behavior of international investors. We consider stocks that were included in the Morgan Stanley Capital International (MSCI) Korea index in 2010 because having greater representation inthe MSCIIndex tends to drive investment byforeigners (Ferreira and Matos, 2008). Excluding the stocks with missing transactions information results in the final sample of 53 Korean stocks. The 53 sample stocks, which constitute more than half of KOSPI’s total market capitalization, closely track the market around the attacks. The KOSPI and sample stock average daily returns are –0.78% and –0.85%, respectively, on attack days. In dollar terms, KOSPI’s market capitalization drops byanaverage of$4billion, roughly 0.5%ofthecountry’s GDP,ondaysofNorth Korean attacks during our sample period. This is substantial considering that it represents a drop in one day. Not surprisingly, regression analysis reveals that the sample stock returns become positively correlated withthesizeofstocksfollowingattacks, suggesting flighttosafety. Weusethisuniquesettingtounderstandwhatdrivesforeigninvestors’tradingstrategiesandperformanceduringattacks. Thehomebiasliteraturesuggeststhatlocalinvestorsareendowedwithsuperior information about companies located in close geograpahic proximity, and that this information asymmetryleads toabias intheir investment portfolios (e.g.,Brennan andCao,1997; Gehrig, 1993; Coval and Moskowitz, 1999, 2001). Kang and Stulz (1997) further show that home bias may manifest in a foreign country in the form of higher holdings of large firms by foreign investors than suggested by market portfolios. That is, domestic investors may have more information about the companies headquartered in Korea, allowing them to evaluate the effect of political risks on operations and profits of these firms better. We then would expect foreigners to reduce the size of their Korean portfolios 3

following attacks, and to shift their portfolio weights among Korean stocks toward larger firmswhere the information asymmetry problem is less severe. Alternatively, foreigners may demand more risky stocks to realize the well-documented benefits of international diversification.5 Foreigners are better positionedtobeartheriskassociatedwithanescalatinggeopoliticalconflictbecauseKoreanstocksare likelytohaverelatively smallweights intheirinternational investment portfolios. Domesticinvestors’ portfolios, ontheotherhand, arelikelytobehighly concentrated onKoreanassetssuchashouses and human capital. Thus, the hypothesis implies that foreigners buy more Korean stocks on net after the North Korean attacks. Also, they are likely to receive a premium for bearing additional political risk according tothestandard risk-return tradeoff. We document three main findings. First, the trading volume analysis shows that foreign investors increase their holdings ofthe sample stocks after theattacks whiledomestic individuals, whoaccount for over 80% of domestic trades, withdraw from the market, and domestic institutions hold a middle ground. Foreignersarealsotheonlynetsellersofhighexport-to-salesstocksonattackdays. Firmsthat haveasubstantialshareofrevenuescomingfromoverseasarelikelytobelessaffectedbylocalpolitical conflicts. Thus, these patterns can be interpreted asforeigners generally assuming more political risk, consistent withtheinternational diversification hypothesis. Furthermore, foreigninvestors becomenet buyers of high book-to-market stocks after the attacks. The willingness of foreigners to buy value stocksisconsistent withtheviewthatforeignersarebetterabletobearincreasesinpoliticalrisktothe extentthatbook-to-market ratiosareassociated withriskfactors(e.g.,FamaandFrench,1992,1993). Second, we examine whether foreigners outperform domestic traders following attacks. We evaluate performance by measuring the buy ratio, which is defined as the fraction of future winners (the stocks with the highest future returns) and future losers that an investor group buys on a given day. Essentially, we evaluate each investor group’s ability to pick winners and avoid losers. We find that foreigners’ performance neither improves nor deteriorates while domestic individuals perform worse anddomesticinstitutionsimprovetheirperformancesignificantlyfollowingattacks. Onabroaderlevel, thisisconsistentwiththeinternationaldiversificationhypothesisinthatforeigners,whobearadditional risk following attacks, perform better than an average domestic trader (note that domestic individuals makeuptheoverwhelming majority ofdomestic trading volume). Onamoregranular level, however, 5SeeSection4.2formoredetail. 4

the breakdown into domestic institutions and individuals produces mixed results. Because foreigners aremostlyinstitutional investors, comparing foreignerswithdomesticinstitutions inparticularmaybe morerelevantwhenitcomestoevaluating theirrelativeinformation advantage. Domesticinstitutions’ superior performance suggests thatdomestic institutions aremoreinformed thanforeigners, providing supportforthehomebiashypothesis. Finally, we examine whether foreign investors tend to destabilize domestic equity markets followinganincreaseingeopoliticalrisk. Weconsiderthewell-documentedstrategyofpositive-feedback tradingintheinternationalfinanceliterature,whichreferstobuyingpastwinnersandsellingpastlosers. Thisstrategycancontribute toshort-term pricedestabilization because negativepost-attack marketreturnswouldinducepositivefeedbacktraderstosellmoreshares,whichinturnputsdownwardpressure on prices, destabilizing the market further in the short run (De Long, Shleifer, Summers, and Waldmann, 1990). If foreigners engage in positive feedback trading following the attacks, their trading wouldcontribute todestabilization bymagnifying theinitialpricedecreasescausedbytheattacks. We findthatforeigninvestorsgenerally employapositivefeedback strategyintheKoreanmarket,butthat they do not on the days of North Korean attacks. Combined with the earlier evidence that foreigners’ total and net trading volume increases following attacks, the results suggest foreigners are unlikely to destabilize themarket. Overall, these attack-day changes in foreigners’ trading pattern and performance are consistent with the international diversification hypothesis where foreigners update their risk assessment of Korean stocks upon attacks and trade to rebalance their portfolios accordingly. These patterns are also consistent with unsophisticated domestic individuals overreacting to the attacks and foreign investors trading to take advantage of the response of domestic individuals. Our results are robust to various sensitivitychecksincludingcomparingsubsamples,examiningdifferentinvestmenthorizonsforfuture returns,replacingtherawreturnswithmarket-adjustedreturns,andmakingexchangerateadjustments. Furthermore, thedocumented effects arestronger formoresevereattacks proxied bylowerattack-day marketreturns,suggestingthatconfoundingeffectsarenotlikelydrivingtheresults. However,wenote acaveatthatourstudydoesnotaddress possible short-sale activities around attacks, whichmayaffect theattack-day tradingstrategies. 5

The rest of the paper proceeds as follows. Section 2 reviews the literature and discusses our contribution. Section 3 provides institutional details on the Korean stock market, summary statistics of our sample stocks, and a description of the North Korean attacks. Section 4 examines the changes in tradingpatternsofthethreeinvestorgroupsaroundattacksandSection5analyzestheperformanceresults. Section 6 examines whether foreigners’ post-attack trading activities contribute to destabilizing themarket. Section7providesvariousrobustness checks, andSection8concludes. 2. Literature Review Thisstudyaddstothegrowingliteratureontheeconomicconsequences ofpoliticalconflicts. Eckstein andTsiddon(2004)modeltheeffectofterrorontheeconomyandfindthatterrorleadstoloweroutput and welfare in equilibrium. McCandless (1996) presents a model in which military events during the U.S. Civil War are shown to be important in describing the movements of the prices of money in both the northern and southern states. Martin, Mayer, and Thoenig (2008) analyze theoretically and empirically therelationship betweenmilitaryconflictsandtrade. Empirically,manyresearchershaveimplementedevent-studymethodologiestoinvestigatetheeconomic consequences of military or terrorist attacks. For instance, Eldor and Melnick (2004) analyze how stock and foreign exchange markets react to terror that occurs inIsrael. Abadie and Gardeazabal (2003)estimatetheeconomiccostsoftheterroristconflictintheBasqueCountryinSpain. Bram,Orr, andRapaport(2002)estimatethecostoftheSeptember11attackonNewYorkCity. Nordhaus(2002) andDavis,Murphy,andTopel(2009)estimatethecostsofthewarwithIraq. AmihudandWohl(2004) and Rigobon and Sack (2005) examine the impact of the war with Iraq on financial variables such as stock prices, bond spreads, oil prices, and exchange rates. Fisman, Hamao, and Wang (2013) analyze stockmarketresponses toepisodesofhostility betweenChinaandJapan. More broadly, Chen and Siems (2004), Chesney, Reshetar, Karaman (2010), and Karolyi and Martell (2010) document the effect of various terrorist and military attacks on global capital markets. Using a panel of 177 countries, Blomberg, Hess and Orphanides (2004) compare the macroeconomic consequences of terrorism with those of other types of conflict such as external wars or internal con- 6

flict. Glick and Taylor (2010) study the effects of war on bilateral trade. Karolyi (2006) provides an overviewoftheconsequences ofterrorism forfinancialmarkets. Severalpapers haveexaminedtheKoreanstockmarketresponse toNorthKoreanmilitaryactions, or more broadly, to news about North Korean developments. Ahn, Jeon, and Chay (2010) find that the Korean stock market reacts significantly to news related to the inter-Korean relations. Lee (2006) focuses on the news about North Korea’s nuclear weapons and finds similar results. Pak et al. (2015) examinetheKoreanstockslistedontheNewYorkStockExchangeanddocumentthatnewsrelatedto North Korea have a significant impact on the volatility of these stocks. Nam (2002) and Kim (2011), ontheotherhand,documentinsignificant priceresponses tomajorNorthKoreandevelopments. Whilethesestudiesfocusontheaggregateeffectsofpoliticalconflicts,wefocusonthedifferences in the response across different investor types. Foreign and domestic investors can be motivated by different factors, thus responding differently to political conflicts. A decomposition of foreign and domestic investors sheds light on how geopolitical risk affects foreign portfolio investment decisions. Thedisaggregate level analysis also has policy implications inthat adifferent policy response maybe required depending on whether an observed pattern is driven by foreign or domestic investors. Our study highlights that foreign investors perceive North Korean risk differently from domestic investors and respond differently. Our work complements Kim and Jung (2014), who analyze intraday trading and short-sale activities. They document a negative post-attack return on the Korean stock market as wellasasignificantincreaseinshort-sale activitiesbysomeforeigninvestorsinthedaysleadingupto theattacks. 3. Data Description 3.1. The KoreanStock Market TheKoreanstockmarketoffersseveraladvantagesasanempiricalsettingforthestudyofinternational equity investment. First, themarketmakes available daily transactions andownership dataaggregated by three investor types – foreigners, domestic institutions, and domestic individuals. Thus, the data allow us to evaluate the response of foreign investors to attacks using domestic investors as bench- 7

marks. A further distinction between domestic institutions and domestic individuals helps us understand whetherobserved foreign trading patterns stem fromforeign attributes orinstitutional attributes. Given that foreign investors tend to be large institutions,6 their trading patterns in the Korean market maymanifestbothforeign andinstitutional traits. Second,theKoreanmarketimposesnorestrictionsonforeignownershipforoursampleperiod. The Korean market wasfirstopen to foreign investors inJanuary 1992 withcertain ownership restrictions. Initially foreign investors were not allowed to own more than 10% of a stock collectively and 3% individually. Theceilingsoncollectiveandindividualforeignownershipweregraduallyincreasedover timetoreach100%inMay1998,whenallownershiprestrictions forforeigninvestorswereremoved.7 Oursampleperiodbeginsin1999toavoidapossiblebiasstemmingfromthetime-varyingrestrictions onforeignownershipbetween1992and1998. Third, the Korean market shows active foreign participation. Foreign participation has rapidly increased overtime,andasofyear-end 2010, thenumberofregistered foreign traders was31,060 and theirownership stakesaccounted for32.9%ofthetotalmarketcapitalization intheKOSPImarket. Finally,theKoreanmarketisliquidandsizable. Korea’sstockmarketconsists oftwomarkets: the KOSPImarketasthemainstockexchange andtheKOSDAQmarketasavenueforsmallandmedium sized enterprises. According to the Korea Exchange (2010), the two stock markets combined offered 1,806 listed equity issues with total market capitalization of KRW 1,240 trillion ($1,093 billion) at the end of 2010.8 The KOSPI market comprised the majority of the total market capitalization with KRW 1,142 trillion, an amount approximately 97% of the country’s GDP,with 777 listed companies. KOSPI’s average daily trading volume was 381 million shares with an average daily trading value of KRW5.6trillion. AccordingtotheWorldFederationofExchanges,theKOSPIandKOSDAQmarkets combined areranked 17th intermsofmarketcapitalization and9thintermsoftrading volume among memberexchanges (KoreaExchange,2011). 6SeeChoe,Kho,andStulz(1999)andKimandWei(2002),amongothers. Inparticular,aproprietarydatasetemployed byKimandWei(2002)revealsthatthemajorityofforeigninvestorsintheKoreanstockmarketareindeedinstitutions. 7TheexceptionswerekeyregulatedindustriessuchasutilitiesandtelecommunicationsinwhichtheKoreangovernment hadownershipstakesofover30%. 8Thedollarvaluewasobtainedusingtheexchangerateattheendof2010. 8

3.2. Sample Firms While foreign participation in the Korean market is substantial, the magnitude of foreign ownership varies considerably across firms. Kang, Lee, and Park(2010), forexample, document that themedian foreign ownership for the stocks listed on the Korea Stock Exchange is less than 1% while the mean is 11.5% for the 2000–2004 period. As such, an equal treatment of all Korean stocks cannot provide an accurate assessment of the behavior of international investors. Rather, our analysis requires a set of stocks with nontrivial foreign ownership. Thus, we consider the stocks that were included in the Morgan Stanley Capital International (MSCI)Korea index at the end of 2010. We restrict our sample to the stocks for which market data are available for the entire sample period of 1999 through 2010. These steps produce 53 stocks for our sample. We expect our sample stocks to represent well foreign tradingactivitiesintheKoreanmarketasMSCIisaleadingproviderofinternationalequitybenchmarks that are widely used by institutional investors.9 Having greater representation in the MSCI World Index is documented to drive investment by foreigners (Ferreira and Matos, 2008). For our sample stocks, we obtain daily transactions and ownership data compiled by DataGuidePro and aggregated by three investor types – foreigners, domestic institutions, and domestic individuals.10 Our sample period begins in 1999, which ensures that our analysis does not suffer from a possible bias from the time-varying restrictions on foreign ownership that were present between 1992 and 1998. Extending the sample period to the date of financial liberalization (January 1992) would add only 7 data points to our sample while exposing the test results to possible bias due to the effect of time-varying foreign ownershiplimitsoninvestmentdecisions. Table 1 reports descriptive statistics by year for the 53 firms in our sample. The second column shows that average daily returns vary between -.18% and .25% from year to year. The next three columns report daily turnover averaged and summed over the sample firms each year as well as the fraction of the total constituted by foreign investors. The sum of the daily turnover across the 53 samplestockswasKRW2.8trillionin2010,approximately50%ofKOSPI’stotalturnoverofKRW5.8 trillion inthe sameyear. Thefinalthree columns present theyear-end market capitalizations averaged 9Ferreira,Massa,andMatos(2010)notethat90%ofinternationalinstitutionalequityassetsarebenchmarkedtoMSCI indicesaccordingtosurveyssuchastheThomsonExtelPan-EuropeansurveyandtheGlobalEquitiesStudy. Hau,Massa, and Peress (2010) also show that the rebalancing of the MSCI Global Equity Index announced in December 2000 had a substantialinfluenceonsubsequentportfoliochoicesofinternationalequityflows. 10Governmentandmunicipalinvestorsareexcludedfromtheanalysis. 9

and summed over the sample firms as well as the fraction of market capitalization owned by foreign investors. The sum of market capitalization over the sample stocks was 608 trillion at 2010 yearend, constituting roughly 53% of KOSPI’s total market capitalization. The 53 sample firms clearly overrepresenttheKOSPImarketintermsofmarketcapitalization andturnover,consideringthatatotal of 777 companies are listed in theKOSPImarket. Giventhat large and liquid firmstend tosuffer less from information frictions, information asymmetry between foreign and domestic investors is likely to be smaller in our sample, which is comprised of relatively larger stocks. Thus, if anything, our sample biases us against finding a meaningful difference between foreigners and locals, providing moreconservative results. Next, the foreign ownership statistics show that the average foreign ownership of the 53 sample firms was 23% in 1999, the first year after the removal of foreign ownership restrictions. From that point forward, foreign ownership gradually increased to a peak of 46% in 2004. At the end of 2010, foreigners in aggregate owned 37% of the total market capitalization of the 53 stocks while owning only 33% of the KOSPI market.11 That implies that foreigners are slightly overrepresented in the sample stocks relative to the KOSPImarket. Finally, the comparison of foreign ownership stakes and turnover reveals that foreigners are not frequent traders relative to domestic investors. Thefraction of turnover constituted by foreign investors is smaller than the fraction of their ownership. For example, theforeignerownershipstakewas37%atyear-end2010,buttherateofturnoverinthatyearwas23%. 3.3. Description oftheNorth KoreanAttacks For our study, we consider North Korean attacks against South Korea over the period from 1999 to 2010. We define an attack as an event such as a border fight or naval battle in which North Korea exhibits military aggression against South Korea, resulting in the exchange of fire between the two Koreas. We also include nuclear tests, which, while not direct attacks on South Korean soil, carry an equally, if not more threatening message. Note, though, that we do not include verbal threats from NorthKoreainouranalysisbecause theyareveryfrequent, andrarelycredible. 11KOSPImarketstatisticswereobtainedfromKoreaExchange(2010). 10

We describe the North Korean attacks as “largely” exogenous because, although it is difficult to predict the timing and magnitude of the attacks, there is evidence that they are to a certain degree predictable. Kim, Kang, and Lee(2016) show that thetone of foreign newscoverage, particularly the British press, and dates around the birthdays of North Korean top leaders are significant predictors of the attacks. The pseudo-r2 values in their models of the timing of North Korean attacks are in the range of 5%-6%. The relatively low pseudo-r2 values are consistent with our characterization of the attacks as largely exogenous to market participants in South Korea. Nuclear attacks in particular tend to be preceded by somewhat informed guesses by analysts and officials.12 We address this concern by repeating the analysis excluding the two nuclear tests and find that the substantive results do not change. Forexample, theattackontheSouthKoreanislandofYeonpyeong in2010, whichkilled four peopleandleftavillageinflames,cameasacompletesurprisetoSouthKoreans. Thecommuniststate had not attacked civilians directly since fighting during the Korean War, which ceased more than 50 yearsearlier.13 NorthKorea’sunpredictable behaviorishighlightedinarecentFinancialTimesarticle, which argues that investors should be more concerned with North Korea than Iran mainly because NorthKorea’sgovernmentisuniquely unpredictable whereasIran’stheocracy isanopenbook14. We obtained the attack information from several sources. We first searched for articles in Factiva thatmentionNorthKoreanmilitaryaggression. Theidentifiedattackswerecross-examined withnews articles to pinpoint the exact local time of the incidents. These steps produced a total of 13 attacks for the sample period. We double-checked the validity of our data by examining the Congressional Research Service (2007) report prepared for the U.S.Congress, which lists North Korean provocative actionsthrough 2007. Weconfirmedthattheattacksidentifiedinthereportareidentical tooursample of attacks for the 1999–2007 period. Table 2 provides descriptions of the attacks as well as summary statisticsaboutthestockmarketperformanceonthedaysoftheattacks. Asexpected,themarketsuffers a sizable loss on average when attacks occur. The attack-day mean KOSPI return is -0.89% over the previous one-year mean returns, which translates into the loss of $4 billion of market capitalization on average on the days of attacks. The attack-day KOSPI returns fluctuate considerably likely due to 12Mullen,Jethro,“FivethingstoknowaboutNorthKorea’splannednucleartest,”CNN,6February2013. 13Shin,Hae-in,“N.K.commits221provocationssince1953,”KoreaHerald,5January2011. Notethatthe221provocationsincludenotonlyphysicalattacksbutalsoverbalthreats. 14FinancialTimes,IanBremmer,“WorrymoreaboutNorthKoreathanaboutIran,”12April2012. 11

the variation in the magnitude of attacks, from as low as -2.63% (June 15, 1999) to as high as 1.62% (October29,2010). The53samplefirmsexhibitasimilarpatternofdailyreturns. IfthepoormarketperformanceonattackdaysreflectspoliticalriskgeneratedbytheNorthKorean attacks,thenthemagnitudeoftheattack-daymarketperformanceshouldvarywiththeperceiveddegree of political risk. We utilize news coverage of attacks as a measure of perceived political risk. We use Factiva to search three major news sources, the Wall Street Journal, the Financial Times, and the New York Times, for a period of seven days after the attacks to identify articles covering those attacks. Table 2 reports the results for the two subsamples of attacks sorted according to the news article coverage. When the attacks receive more media attention (i.e., the article counts are higher than or equal to the median value), the market performance is worse with an average KOSPI return of -1.18% over the previous year. In contrast, the average KOSPI return is -0.56% over the previous yearwhentheattacksreceivelessmediaattention. The53samplestocksexhibitthesamepattern. The pattern in the data is consistent with our premise that the stock market performance reflects political risksassociated withinter-Korean conflicts. 4. Trading Patterns and Political Risk 4.1. Stock Returns Around Attacks Wefirstexaminethecross-sectionofstockreturnsaroundtheattacks. Areallstocksaffecteduniformly or are some stocks more vulnerable to geopolitical risk than others? We investigate this question by regressingattack-daystockreturnsagainstasetoffirmcharacteristics. Thevariablesrepresenting firm characteristics include size (log of market capitalization), leverage (the ratio of total liabilities to total assets minus book value of equity plus market value of equity), book-to-market ratio (book value of equitydividedbythemarketvalueofequity),returnonassets(netincomedividedbytotalassets),beta (estimatedusingweeklystockreturndataoverthepreviousone-yearperiod),andindustryindicators.15 Alsoincludedareforeignownershipstakespriortoattacksandtheirnettradingvolumeonattackdays. Dataonfirmcharacteristics areobtained fromtheKIS-Valuedatabase. 15Theobservationsareclassifiedintothreeindustrygroups: manufacturing(49%ofthesamplestocks),financialservices (19%),andothers(32%). 12

Wealsoconsidermeasurestoproxyforafirm’spoliticalsensitivity. First,export/sales (theratioof export revenues to total sales) may capture a firm’s exposure to political risk because firms that have a substantial part of revenues coming from overseas are likely to be less affected by domestic events. Second, firmsoperating intheKaesongIndustrial Region,aspecial economiczoneinNorthKoreaset upbythetwoKoreastopromote economiccooperation, maybeaffected morebythetension between the two Koreas. Third, firms in the defense industry may be more sensitive to military conflicts. We construct Kaesong and Defense indicator variables according to the classifications provided by Ahn, Jeon,andChay(2010)andKimandJung(2014). ThefirsttwocolumnsofTable3showtheresultsusingallstock/dayobservationsonattackdaysand on pre-attack days (five trading days preceding the attacks), respectively.16 The final column reports the differences in coefficients between the first two regressions. The results show that size (market capitalization) is not associated with returns in the pre-attack period, but become strongly correlated withreturns onattack days. Thischange isstatistically significant atthe1% level. Therelatively high demandforlargecapstocks followingattacks issuggestive offlighttosafety inresponse toescalating political conflicttotheextentthatsizecaptures riskfactors(e.g.,Banz,1981; FamaandFrench,1992, 1993). Investors also shy away from financial industry stocks following attacks although the change is only marginally significant. Foreigners’ net trading volume is positively related to returns on both attackandpre-attack days. 4.2. AnalysisofTrading Volume Wenextexaminetradingvolumeofeachinvestorgrouptoseewhetherthedemandforlessriskystocks documentedintheprevioussubsectionisdrivenbyaparticularinvestorgroup. Thehomebiasliterature suggests that information asymmetry leads foreigners to hold a disproportionately smaller fraction of foreign stocks (e.g., Gehrig, 1993; Brennan and Cao, 1997; Ahearne, Griever, Warnock, 2004; Portes and Rey, 2005).17 Furthermore, the information disadvantage relative to domestic investors may lead foreigners to hold a bigger fraction of large and liquid firms in their foreign portfolios than indicated 16Table3reportstheresultsusingrawreturns. TheresultsareessentiallyunchangedwhenweuseseveraldifferentmeasuresofexcessreturnsadjustedbyrollingwindowsofmeandailyKospireturnsrangingfrom10to90days. 17Otherexplanationsofhomebiassuchasdirectbarrierstointernationalinvestmentareunlikelytoplayanimportantrole inourwithin-countrysettingconcerningashortwindowsurroundingattacks. 13

bymarket portfolios (Kang andStulz (1997)). Wehypothesize thatifdomestic investors areendowed withsuperior information aboutthecompanies headquartered inKorea,foreigners arelikely toreduce thesizeoftheirKoreanportfoliosfollowingattacksand,withinKoreanportfolios,shiftweightstoward largerfirmswheretheinformationasymmetryproblemislesssevere. NorthKoreanactionshavedrawn much attention from the international community over time, including extensive international media coverage of the country. Thus, foreign investors may have the same access to information regarding the timing of North Korean attacks as domestic investors. However, domestic investors may have advantages inevaluating theeffectoftheattacksoncompanies operating locally. Onthe other hand, foreigners maydemand morerisky stocks to realize the well-documented benefits of international diversification (e.g., Grubel, 1968; Levy and Sarnat, 1970; Solnik, 1974; Grauer andHakansson,1987;DeSantisandGerard,1997). Theinternationaldiversificationhypothesisimplies that foreigners are better positioned to bear the risk associated with an escalating geopolitical conflict becauseKoreanstocksarelikelytohaverelativelysmallweightsintheirinternational investmentportfolios. Anexaminationofattack-daytradingvolumeshowsthatforeignersincreasethesizeoftheirportfoliosofthesamplestocksafterattacks,providingsupportfortheinternationaldiversificationhypothesis. PanelAofTable 4reports thethree investor types’ average attack-day trading volume aswellastheir average daily volume over the five trading days preceding attacks. On the days of attacks, foreigners are net buyers of the 53 sample stocks, buying 9.4% more while selling only 4.5% more than in the dayspreceding attacks. Theirnettototalvolumeratio(2.8%)isthehighestofthethreeinvestortypes, suggesting that foreigners are more willing to bear the additional risk of an increase in political risk. Domestic individuals become net sellers of the 53 sample stocks with the lowest net to total volume ratioof-0.5%. The total volume result reveals a similar pattern in which foreigners step up trading on the days ofattacks compared topre-attack dayswhiledomestic individuals withdraw from themarket. Foreign investors trade 7% more shares on the days of attacks relative to the previous five trading days. By contrast, domestic individuals’ trading volume declines sharply (25.5%). Domestic institutions show theleastchange, tradinglessbyamoderateamountof3.1%. 14

We next investigate foreigners’ portfolio rebalancing patterns within their Korean portfolios by regressing the net trading volume of each investor type against a set of firm characteristics. We run twosetsofregressions, oneforstock/dayobservations onattackdaysandtheotherforobservationson the five trading days preceding the attacks, and compare the coefficients of the two regressions. The firmcharacteristic variablesaredefinedasbefore. Theforeigninvestors’previousownershipstake(%) is included as acontrol variable because portfolio rebalancing decisions are influenced bythe level of priorportfolioholdings. Thedependentvariableisnettradingvolumetransformedusingthefollowing procedure: y=ln(Vol+p1+Vol2). (1) Thisvariation ofalogtransformation preserves thenegativevalues ofnetvolume (BusseandHefeker (2007)). PanelBofTable4presents theresults. Thefirsttwocolumns report results forforeigners on attackandpre-attackdays. Thesecondcolumnalsoreportswhetherthedifferencesbetweentheattackdayandpre-attack-day coefficients arestatistically significant. Thecoefficients onthebook-to-market ratio indicate that foreigners become net purchasers of value stocks (high book-to-market ratios) on attack days. This indicates a shift toward more risky stocks following attacks to the extent that high book-to-market ratios are associated with risk factors (e.g., Fama and French, 1992, 1993; Petkova and Zhang, 2005). While not statistically overwhelming (10%), this change is in stark contrast to the pattern exhibited by domestic individuals, who stop buying high book-to-market stocks on net after the attacks. This change exhibited by domestic individuals is statistically significant at the 5% level. As noted before, the comparison between foreigners and domestic individuals is important because domesticindividuals constitute theoverwhelmingmajorityofdomestictradingvolume. Also interesting is that, both on attack and pre-attack days, foreigners are net sellers of high export/sales stocks while domestic institutions are net buyers. Furthermore, foreigners are the only net sellersofgeographically diversifiedfirmsonattackdays. Firmsthathaveasubstantialpartofrevenues coming from overseas are likely to be less affected by local political conflicts. Thus, this pattern can be interpreted as foreigners generally assuming more political risk, consistent with the international diversification hypothesis. There issomeevidence that foreigners avoid Kaesong stocks on pre-attack days (10% significance) but this pattern goes away on attack days. Overall, the findings suggest that 15

foreigners departfromtheirtypicaltrading behavior inresponse totheinter-Korean conflictbychangingthesizeandweightsoftheirportfoliosofthe53samplestockstobearmorepoliticalrisk,providing supportfortheinternational diversification hypothesis. 5. Investor Performance and Political Risk This section examines changes in foreigners’ performance around attacks using domestic investors as benchmarks. The international diversification hypothesis suggests that foreigners should bear additionalriskfollowingattacksandthusreceiveapremiumwhiledomesticindividuals, whosellsharesto avoid political risk, should pay apremium according tothestandard risk-return tradeoff. Ontheother hand, thehomebiasliterature suggests thatdomestic investors wouldperform better following attacks due to an advantage in evaluating the effect of political risk on local firms. We evaluate the performance using a buy ratio, the fraction of future winners and losers an investor group buys on a given day,similartoGrinblattandKeloharju(2000). Performancecanbemeasuredinvariouswaysincluding comparisonsofportfolioreturnsandtransactioncosts.18 Ourchoiceofaperformancemeasureisbased ontwoconsiderations. First,sinceourempiricalsetting concerns ashortwindowsurrounding attacks, wefocus on the choice of stocks that investors buy and sell on attack days. That is, weare moreconcernedwiththerelativechangemadetotheportfoliosonattackdaysthanwiththeoverallcomposition ofthe portfolios, whichishighly correlated withthe investors’ prior ownership positions. Second, we compare an investor group’s choice among the cross-section of stocks across different points in time rather than stock returns perse. Because an attack has anegative effect onstock returns on average, a time-series comparison ofstock returns purchased before andafter attacks maynotaccurately capture aninvestor’s abilitytochoose stocks. We define the buy ratio as the number of shares of firm i an investor group buys on day t divided bythetotalnumberofthefirm’ssharesbought andsoldbythesameinvestor grouponthatday: SharesofFirmiPurchased onDayt BuyRatio = . (2) it SharesofFirmiPurchased onDayt+SharesofFirmiSoldonDayt 18Somestudiesexaminetransactioncostsbornebyinvestors(e.g.,Choe, Kho, andStulz,2005; Dvorak, 2005), butthis requiresaccesstointra-daytransactionsdatabyinvestortypes. 16

The5-day,20-day,and60-dayfuturereturnsarecalculatedforeachstockiandeachdaytbysumming the daily returns over the period from day t+1 to t+5, t+20, and t+60, respectively. For attack days, we select stock/day observations with the 100 highest and lowest returns (approximately the top 15% and bottom 15% ofthe 689 stock/day observations) for each of the 5-day, 20-day, and 60-day periods followingtheattacks. Werepeatthesameprocessfornon-attackdays,selectingthesameproportionof stock/dayobservationswiththehighestandlowestfuturereturns. Wethencalculatethemeanbuyratio for winners (stocks with the highest future returns) and losers (stocks with the lowest future returns) for each of the return windows on attack and non-attack days. The longest investment horizon tracks 60tradingdaysfollowingattacks,approximatelythreemonths,whichshouldbesufficientlylonggiven that the shock caused by attacks tends to be relatively short-lived. We evaluate the performance by the difference in buy ratios between the highest-return stocks (H) and the lowest-return stocks (L) chosen on a given day by each investor group, where the difference is denoted by H L. An average − buy ratio of 0.5 indicates that the performance is no better than that of a randomly selected portfolio. A buy ratio greater than 0.5 for a winner (H) or loser (L) portfolio indicates that an investor group bought disproportionately more stocks that were subsequently winners or losers. Thus, H L should − be significantly positive if an investor group systematically buys a larger fraction of winners and a smallerfraction oflosers. Table5presentsmeanbuyratiosfordifferentinvestmenthorizons. Forexample,PanelBdescribes theresults forthefuture return window of20trading days. ThefirstrowofPanelBshowsthe 20-day meancumulative returns forthehighest-return andlowest-return stocks averaged overattack daysand non-attackdays,respectively. Thenextthreerowsshowthemeanbuyratioforeachofthethreeinvestor groups. Thefirstsixcolumnsreportthemeanbuyratiosforthehighest-return andlowest-returnstocks as well as H L, the differences between the highest-return and lowest-return stocks for attack and − non-attack days, respectively. For non-attack days, the buy ratios are close to 0.5, which means none ofthethreegroupsbuysanunusuallylargenumberofsharesthatsubsequently haveextremelygoodor badperformance. Naturally,H Liscloseto0. Foreigners,forinstance,haveanH Lvalueof-0.01, − − whichindicatesthatforeignersbuyslightlymorefuturelosersthanfuturewinners. Ingeneral,allthree investor groupshavebuyratioscloseto0.5andH Ldifferences closeto0onnon-attack days. − 17

On attack days, however, there is a considerably more variation in the buy ratios. The first two columns of Panel B show that domestic institutions have a buy ratio of 0.58 for the winners and 0.44 forthelosers. Thatmeansthat,onattackdays,domesticinstitutionsbuyalargerfractionofstocksthat subsequently havehighreturnsandbuyasmallerfractionofstocksthatsubsequentlyhavelowreturns. Consequently,theirH Lvalueof0.14onattackdaysrepresentsasubstantialimprovement(0.11)over − the value of0.03 onnon-attack days. Thisimprovement, denoted by Attack(H L) NoAttack(H − − − L),isstatistically significantatthe1%level. Foreigninvestorshaveslightlyworseperformanceresults on attack days, with the buy ratio falling to 0.50 for the highest-return stocks and rising to 0.53 for the lowest-return stocks. Overall, their ability to choose winners and to avoid losers changes little as indicated by the Attack(H L) NoAttack(H L) value of -0.03, which is not significantly differ- − − − ent from zero. Finally, domestic individuals produce the worst results of all three groups in terms of their performance relative to non-attack days. The Attack(H L) NoAttack(H L) value is -0.04 − − − and statistically significant at the 1% level. The 5-day and 60-day investment horizons exhibit similar patterns, suggesting that the findings are not sensitive to the choice of investment horizons. Domestic institutions showimprovement inallthreereturn windows, twoofwhicharestatistically significant at the 1% level. Domestic individuals show deterioration in all three windows, two ofwhich are statistically significant atthe1%level. Foreigners holdthemiddle ground withstatistically insignificant and economically marginalchangesacrossallthreewindows. Takentogether,theresultsshowthattheforeigners’responsestoattackshaveverydifferentperformanceconsequences fromthoseofdomesticinvestors. Domesticindividuals, whoturnintonetsellers following attacks, perform worse following attacks. Foreigners, who are net buyers, do not perform worseasdomesticindividuals do,buttheydonotshowimprovement,either. Domesticinstitutions improve their performance following attacks. Onabroader level, this isconsistent withtheinternational diversification hypothesis inthatforeigners, whobearadditional riskfollowing attacks, perform better thananaveragedomestictrader(notethatdomesticindividualsmakeupalmost86%ofdomestictrades onattackdays). Theseresultsalsosuggestthatdomesticindividualsdonothaveaninformationadvantage over foreigners when it comes to evaluating the effect of North Korean attacks.19 Theresults are 19Thisisconsistentwiththeviewthatindividualinvestorsdonothavevalue-relevantinformationaboutthelocalstocks theytrade(SeasholesandZhu,2010) 18

alsoconsistentwiththeviewthatunsophisticated domesticindividualsoverreacttotheattacks,leading totheirunder-performance relativetoforeigninvestors. On a more granular level, however, because foreigners are mostly institutional investors, it is reasonable to compare foreigners with domestic institutions when it comes to evaluating their relative informationadvantage. Thefactthatdomesticinstitutions improvetheirperformancesignificantly following attacks while foreigners maintain the same level suggests that domestic institutions are more informed than foreigners, providing somesupport forthehomebiashypothesis. Wealsonote thatour results are not necessarily at odds withprevious studies documenting foreign investors outperforming domesticinstitutions(e.g.,GrinblattandKeloharju,2000)inthatwefocusonparticulareventspertaining to political conflicts whereas the previous studies examine the performance averaged over several yearsofdatawhichmaynotcontain eventsassociated withpoliticalconflicts. 6. Analysis of Feedback Trading Strategies We next investigate how the response of foreigners affects the market. As documented in Section 3, foreigners buy more shares on net following attacks and trade more than usual while domestic individuals withdraw from the market as indicated by the sharp reduction (25.5%) in their total trading volume. Having relatively smaller exposure to the geopolitical risks, foreigners seem well-positioned tocontribute tostabilizing themarketsbysharingtheriskswithdomesticindividuals whoseportfolios aremoreconcentrated inKoreanassetsthatincludenonfinancialassetssuchashousesandhumancapital. In this section, we provide additional evidence supporting this view by analyzing the changes in foreigners’ tradingstrategies aroundtheattacks. 6.1. UnivariateAnalysis We examine whether attacks influence foreigners’ trading strategies and, in particular, the strategy of positive-feedback trading, which refers to buying past winners and selling past losers. The positive association between net equity flows and returns is one of the stylized facts in international finance. Several within-country studies document a positive feedback strategy on the part of foreign investors 19

in Korea (Choe, Kho, and Stulz, 1999) and in Japan (Karolyi, 2002) surrounding the Asian financial crisis in the late 1990s, and in Finland (Grinblatt and Keloharju, 2000). In addition, cross-country studies document a positive correlation between international equity flows and contemporaneous or lagged stock returns, as evidence suggestive of apositive feedback ormomentum strategy (e.g., Bohn andTesar,1996;BrennanandCao,1997;Froot,O’Connell,andSeasholes, 2001).20 Positivefeedback traders can contribute to short-term price destabilization because negative post-attack market returns would induce positive feedback traders to sell more shares, which in turn puts downward pressure on prices, destabilizing the market further inthe short run (DeLong, Shleifer, Summers, and Waldmann, 1990). Thus, if international investors maintained their well-documented feedback trading strategy followingattacks, theycouldexertadestabilizing influenceontheKoreanequitymarket. We identify feedback trading patterns using the methodology of Choe, Kho, and Stulz (1999) to makeourresultscomparable. Weevaluatewhetheragiveninvestorgroupengagesinfeedbacktrading strategies by comparing its order imbalances on a given day with its previous-day returns. The order imbalance (OI) for each firm/trading day is defined as the net buy volume divided by daily volume averaged over the ten previous trading days, where net buy volume is the number of shares bought minusthenumberofsharessoldsuchthat: SharesofFirmiPurchasedonDayt–SharesofFirmiSoldonDayt OI = . (3) it AverageDailyVolumefromDayt-10through Dayt-1 Notethataverage daily volumeiscalculated excluding theattack dayt because theattack-day volume maybeaffectedbyanattack. Theorderimbalancesforattackandnon-attackdaysaresortedseparately foreachofthethreeinvestorgroups–foreigners,domesticinstitutions,anddomesticindividuals–into quintiles based on the previous-day returns.21 The order imbalances are then averaged across each quintile. PanelAofTable6showstheaverageorderimbalances forthequintileportfolios formedbasedon the previous-day returns. On non-attack days, foreigners and domestic institutions engage in positive 20Replacingbilateralflowdatawithportfoliopositionsdata,Curcuru,Thomas,Warnock,andWongswan(2011)document thatU.S.investorsarenotreturnchasersinforeignequitymarkets. 21Thefivegroupsarenotexactlythesameinsizebecauseinsomecasesthepreviousday’sreturnsatthecutoffpointsfor eachquintilehavemultipleobservations. Weadjustthecutoffpointssothatallobservationswiththesamereturnbelongto thesameportfolio. 20

feedback trading, buying stocks with high previous-day returns and selling stocks with low previousdayreturnsonnet. TheysellP1andP2stocks(theportfolios withthelowerprevious-day returns) and buy P4 and P5 stocks (the portfolios with the higher previous-day returns). Furthermore, their order imbalances monotonically rise from P1 to P5. Domestic individuals display the opposite pattern in whichstockswithlowprevious-day returnsareassociated withlargeandpositiveorderimbalances, or high net purchases, and vice versa. Since non-attack days constitute the vast majority of the sample period, wecanconclude that,foroursampleperiod, foreigners anddomesticinstitutions aregenerally positivefeedbacktraderswhiledomesticindividualsarenegativefeedbacktraders. Thisgeneralpattern is consistent with the findings of Choe, Kho, and Stulz (1999), who document the trading behavior of thethreeinvestor typesintheKoreanmarketsurrounding theAsianfinancialcrisisin1997. The patterns change on attack days, particularly among foreign investors. Foreigners no longer pursue positive-feedback strategies. Their order imbalances are now highest among the P2 and P3 portfolios with 0.044 and 0.068, respectively, rather than among P4 and P5 portfolios. Also notable is that foreigners seem to be buying more across the board as indicated by positive order imbalances in four of the five portfolios, consistent with an increase in foreigners’ net trading volume following attacks documented in Section 3. We further consider the possibility that investors base their trading strategies on longer frequencies. Previous studies have examined the feedback trading pattern in variousfrequencies according totheirempiricalsettings anddataavailability.22 PanelsBandCofTable6 reportaverageorderimbalancesoverthetwo-dayandfive-daywindows,respectively,forquintileportfolios formed based on the returns over the previous two-day periods. The results are not sensitive to changesinthelengthofwindows. Takentogether,thefindingssuggestthatforeigners’tradingactivities surrounding attacksarenotlikelytohaveadestabilizing effectontheKoreanequitymarket. Unlike foreigners, domestic individuals make little change to their usual trading strategy when attacks occur. The attack-day patterns are not monotonic and statistical significance is weaker due in parttothesmallersamplesizeonattackdays. Nonetheless, individuals continuetobuypastlosers(P1 and P2) and sell past winners (P4 and P5), and P5–P1 remains negative and significant across Panels 22Forexample,Choe,Kho,andStulz(1999)examinedailyfrequency,Lakonishok,Shleifer,andVishny(1992)andWermers(1999)quarterly,andNofsingerandSias(1999)annual. Inmanycases,dataavailabilitytendstodictatethechoiceof frequency. Inourempiricalsetting,aone-daywindowshouldsufficetocapturethechangeintradingpatternssurrounding NorthKoreanattacks. 21

A, B, and C. Domestic institutions deviate somewhat from the monotonically increasing pattern of non-attack daysbuttheycontinue tosellstocks inthetwolowest-return quintiles (P1andP2)andbuy fromthetwohighest-return quintiles(P4andP5)acrossPanelsA,B,andC.However,P5–P1becomes insignificant in Panels B and C, suggesting domestic institutions deviate somewhat from their usual positivefeedback tradingstrategies inlongerwindows(2-dayand5-daywindows). 6.2. RegressionAnalysis Next, we examine the feedback trading strategy in a multivariate setting, controlling for lagged stock returns,marketreturns,andUSD-KRWexchangeratechanges. Theinclusionofmarketreturnsallows forthepossibilitythatinvestorsmayemployafeedbacktradingstrategybasedonmarketreturnsrather than individual stock returns. Exchange rate changes are included because foreign investors generally repatriate theirincomeandcapital.23 Theregression specification isasfollows: OI = b +b Attack +b r +b r Attack +b rm +b rm Attack +b s it 0 1 t 2 i,t 1 3 i,t 1 t 4 t 1 5 t 1 t 6 t 1 − − − − − +b s Attack +b Kaesong +b KaesongAttack +b Defense 7 t 1 t 8 i 9 i t 10 i − +b DefenseAttack +e , (4) 11 i t it where OI is the order imbalance of firm i on day t, r is the stock return of firm i on day t, rm is it it t the return on the KOSPI composite index on day t, and s is the USD-KRW exchange rate on day t. t Attack is set to one on attack days and zero otherwise, Kaesong indicates whether the company has production facilities in the Kaesong Industrial Region, and Defense measures whether the company supplies materialstotheDepartmentofDefense. Table 7 reports the results of the regressions for the three investor groups. On non-attack days, foreigners’orderimbalancesarepositivelyrelatedtolaggedstockreturns,r ,but,onattackdays,the t 1 − coefficient on lagged returns drops by -0.600 as shown by the interaction term (r Attack ). Also, t 1 t − · 23WeusetheUSD-KRWexchangeratebecauseUSinvestorsconstitutethelargestfractionofforeigninvestors. Foreign investors intending toinvest directly intheKorean securities market must register withthe Financial Supervisory Service andobtainanInvestmentRegistrationCertificate,andtheKoreaExchange(2010)reportsthenumberofregisteredforeign investors by nationality at each year-end from 2000 through 2010. Among foreign investors, US nationals constitute the largestfractionofforeigninvestors–between34%to39%dependingontheyear. Japanesenationalsareadistantsecond, comprising8%to10%ofthenumberofregisteredforeigninvestors. 22

testsoflinearcombinations inthelasttworowsshowthattheattack-day coefficientofthestockreturn (r + r Attack )of0.180isstatistically insignificant, indicating thatforeigners nolonger pursue t 1 t 1 t − − · theirusualstrategyofpositivefeedbacktradingaftertheattacks. Domesticinvestors,ontheotherhand, maintain their usual trading strategies on attack days. The lagged stock return variable for domestic individuals has a negative and significant coefficient both on attack and non-attack days though the coefficient is less negative on attack days (-0.810 on attack days as opposed to -1.616 on non-attack days), suggesting they engage in a contrarian strategy on attack and non-attack days alike. Similarly, thelagged stockreturn variable fordomestic institutions hasapositive andsignificant coefficient both onattackandnon-attack days. In sum, after controlling for market returns and exchange rate changes, the feedback trading patternsofthethreeinvestortypesdocumentedintheunivariatetestsinTable6remainunchanged. Results are also similar when contemporaneous values of returns and exchange rate variables are included to controlfortheeffectsofintra-day feedback trading(unreported). Thesechangesinforeigners’ trading patternsseemconsistent withtheinternational diversification hypothesis whereforeigners updatetheir riskassessmentofKoreanstocksuponattacksandtrademoretorebalancetheirportfoliosaccordingly. These patterns are also consistent with unsophisticated domestic individuals overreacting to the attackswhileforeigninvestorstradetotakeadvantageoftheresponse ofdomesticindividuals following attacks. 7. Robustness Checks and Additional Analysis 7.1. Magnitude ofAttacks We conduct various robustness checks on the observed feedback trading patterns and performance results. First, Table 2 indicates that some attacks have a more detrimental impact on the market than others. Ifattackscausedforeignerstodeviatefromtheirusualtradingpatternandperformance,thenwe wouldexpectmoresevereattackstobeassociatedwithgreaterdeviationsfromthegeneralpattern. We test this conjecture by repeating the analysis of order imbalances and mean buy ratios on subsamples ofattacks withdifferent magnitudes. Theattack-day sampleissortedintotwosubgroups based onthe 23

severity of attacks, where the magnitude of attacks is proxied by the market performance on a given attackdayrelativetothemedianmarketreturnacrossallattackdays. Panel A of Table 8 calculates mean buy ratios for the two subsamples. The results show that the subsample of more severe attacks displays the same pattern as the attack-day pattern found in the full sample whilethe subsample oflesssevere attacks does notexhibit anystatistically significant pattern. The findings are consistent with the view that the patterns in buy ratios documented in Table 5 are indeeddrivenbytheattacks. Next,PanelBofTable8reportsorderimbalanceresultsforthesubsample analysis. As predicted, the more severe attacks (represented by the lower-than-median KOSPI return subsample) areassociated withgreaterdeviations fromthefeedback strategyonthepartofforeigners. On the days of more severe attacks, foreigners no longer buy past winners (P5). In fact, they sell the highest-return stocks(P5)morethantheysellthelowest-return stocks(P1),asindicated bytheirorder imbalances of -0.025 for P5 and of -0.018 for P1. On the days of less severe attacks, on the other hand, foreign investors continue tobuy past winners withanorder imbalance of0.033 and tosell past losers withanorder imbalance of-0.007, although themagnitude oftheimbalances issmallerthanon non-attack days. Fordomestic investors, thedifference in order imbalances between thedays ofmore severe attacks and the days of less severe attacks is much smaller. Similar to the full sample results, the trading strategy of domestic investors appears to be less affected by attacks. Alternatively, weuse thenewscoveragedescribedinSection2.3asameasureofthemagnitudeofanattack. Thesubsample analysis usingarticlecountsproduces similarresults(unreported). 7.2. Additional Robustness Tests We next consider the effect of changes in exchange rates on the portfolio choices of foreign investors because many foreign investors measure their returns in dollars. Table 9 shows buy ratios calculated based onthedollar returns. Theoverall results change little, indicating thatexchange rate fluctuations do not drive thedocumented patterns ofportfolio choices madeby the three investor groups on attack days. Wealsoexaminewhethertheresultsaredrivenbyasubsampleofacertaintimeperiod. Totestthis possibility, wesorttheattack-day sampleintotwosubsamples according tothechronological orderof 24

the attacks. Table 10 reports the results. Overall, the statistical significance in the subsample analysis is lower than in the full sample analysis due to the reduction in the sample size. However, the order imbalances of foreign investors are similar across the different sub-periods. The differences in their order imbalances between past winners and past losers (P5 – P1) are 0.023 for the earlier period and 0.018forthelaterperiod,respectively. Insum,theforeigninvestors’attack-daytradingpatternsremain fairlyconsistent acrossthesampleanddonotseemtobedrivenbytradingpatternsinaparticularsubperiod. Finally, we note that some attacks have overlapping 60-day return windows due to short intervals between attacks. Werecalculate buy ratios excluding the attacks with overlapping windows, and confirmthattheresultsremainunchanged (unreported). 7.3. DiscussionofAlternativeEvents One could consider specific non-political events to evaluate whether the documented pattern is driven by the political nature of the events. Forexample, natural disasters such as earthquakes and typhoons are exogenous events in the sense that they are not caused by the decisions of political leaders in the countries wheretheyoccur. According totheCenterforResearchontheEpidemiology ofDisasters,24 South Korea had 35 incidences of natural disasters during our sample period. However, the impact of thesedisasteristoosmalltodrawameaningfulinference. Theaveragedamageofthenaturaldisasters isonly$313million. WhilewecannotdirectlymeasurethedollarvalueofdamagesincurredbyNorth Korean attacks, we can take a hint from the stock market reaction. The market capitalization drops by about $4 billion on average on the days of attacks. One could also consider the Asian financial crisis in 1997. However, it is empirically challenging to establish a causal relationship because of the endogenous nature of the crisis development. Furthermore, the crisis was a one-off event so it is difficulttogeneralize thepatterninthedata. Theconsideration ofalternative eventsreinforces thefact thattheNorthKoreanattackspresentaunique settingtostudytheeffectofincreasesinpolitical risk. 24Seehttp://www.emdat.be/database. 25

8. Conclusion We examine foreign investors’ trading patterns and performance surrounding 13 North Korean military attacks against South Korea between 1999 and 2010. We document three main findings. First, following attacks, foreigners increase their holdings of the sample Korean stocks and hold more risky stocksproxied byhighbook-to-market ratios. Second, performance resultsshowthatforeigners maintaintheirpre-attacklevelofperformancewhiledomesticindividuals, whoconstitutetheoverwhelming majority of domestic trading volume, perform much worse following attacks. Domestic institutions improvetheirperformance. Third,foreigners’attack-daytradingactivitiesareunlikelytohaveadestabilizing effectontheKoreanequity market. Foreigners stepintobuy shares, primarily from domestic individuals, in the wake of the North Korean attacks. Foreigners trade more shares than usual while domestic individuals trade substantially less. Also, foreigners do not engage in their usual strategy of positive-feedback trading onattackdays. Overall, these results are consistent with the view that foreigners are better positioned to bear the risk associated with an escalating geopolitical conflict. Another non-mutually exclusive explanation is that foreigners trade to take advantage of unsophisticated domestic individuals overreacting to the attacks. The results also highlight that foreign investors perceive North Korean risk differently from domestic investors. Foreigners’ responses topolitical conflict and their impactonthelocal market are very different from those of domestic investors, highlighting the importance of a separate analysis for foreignanddomesticinvestors. 26

References Abadie,A.,Gardeazabal, J.,2003. Theeconomiccostsofconflict: acasestudyoftheBasquecountry. AmericanEconomicReview93,113–132. Ahearne, A., Griever, W., Warnock, F., 2004. Information Costs and Home Bias: An Analysis of U.S.HoldingsofForeignEquities. JournalofInternational Economics62,313–336. Ahn, H. J., S. P. Jeon, and J. B. Chay. 2010. The impact of news about relations between North andSouthKoreaonthestockmarket. KoreanInstitute ofFinance16(2),200–238. Amihud, Y., Wohl, A., 2004. Political news and stock prices: the case of Saddam Hussein contracts. JournalofBankingandFinance28,1185–1200. Banz, R., 1981. The relationship between return and market value of common stocks. Journal of FinancialEconomics9,3–15. Blomberg,S.,Hess,G.,Orphanides, A.,2004. TheMacroeconomic Consequences ofTerrorism. JournalofMonetaryEconomics51,1007–1032. Bohn, H., Tesar, L., 1996. U.S. Equity Investment in Foreign Markets: Portfolio Rebalancing or ReturnsChasing? AmericanEconomicReview86,77–81. Bram, J., Orr, J., Rapaport, C., 2002. Measuring the effects of the September 11 Attack on New YorkCity. EconomicPolicyReview,FederalReserveBankofNewYork. Brennan, M., Cao, H., 1997. International portfolio investment flows. Journal of Finance 52, 1851– 1880. Busse, M.,Hefeker, C.,2007, Politicalrisk, institutions andforeign direct investment. European JournalofPoliticalEconomy23,397–415. Chen, A., Siems, T, 2004, The effects of terrorism on global capital markets. European Journal of PoliticalEconomy20,349–366. Chesney, M., Reshetar, G., Karaman, M., 2010. The Impact of Terrorism on Financial Markets: An EmpiricalStudy. WorkingPaper Choe, H., Kho, B., Stulz, R., 1999. Do foreign investors destabilize stock markets? The Korean Experience in1997. JournalofFinancialEconomics54,227–264. Choe, H., Kho, B., Stulz, R., 2005. Do domestic investors have an edge? The trading experience offoreign investors inKorea. ReviewofFinancialStudies18(3),795–829. Congressional Research Service (U.S.), 2007. CRS Report for Congress: North Korean Provocative Actions. 1950–2007. 27

Coval,J.andT.Moskowitz. 1999. HomeBiasatHome: LocalEquityPreference inDomesticPortfolios. Journal ofFinance54,2045-73. Coval,J.andT.Moskowitz. 2001. TheGeography ofInvestment: Informed TradingandAssetPrices. JournalofPoliticalEconomy109,811–841. Curcuru, S., Thomas, C., Warnock, F., Wongswan, J., 2011. U.S. international equity investment and pastandprospective returns. AmericanEconomicReview101,3440–3455. Davis, S., Murphy, K., Topel, R., 2009. War in Iraq versus Containment, Published in Guns and Butter: TheEconomic Causes and Consequences of Conflict, Gregory D.Hess, ed., Cambridge, MA: MITPress,2009. De Long, J., Shleifer, A., Summers, L., Waldmann, R., 1990. Positive feedback investment strategiesanddestabilizing rationalspeculators. JournalofFinance45,379–395. De Santis, G., Gerard, B., 1997. International asset pricing and portfolio diversification with timevaryingrisk. Journal ofFinance52,1881–1912. Dvorak, T., 2005. Do domestic investors have an information advantage? Evidence from Indonesia. JournalofFinance60,817–839. Eckstein, Z., Tsiddon, D., 2004. Macroeconomic Consequences of Terror: Theory and the Case of Israel. JournalofMonetaryEconomics51,971–1002. Eldor, R., Melnick, R., 2004. Financial markets and terrorism. European Journal of Political Economy20,367–386. Fama, E., French, K., 1992. The cross-section of expected stock returns. Journal of Finance 47, 427–465. Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of FinancialEconomics33,3–56. Ferreira, M., Massa, M., Matos, P., 2010. Shareholders at the gate? Institutional investors and crossbordermergersandacquisitions. ReviewofFinancialStudies23,601–644. Ferreira, M., Matos, P., 2008. The Colors of investors’ money: The role of institutional investors aroundtheworld? JournalofFinancialEconomics88,499–533. FinancialSupervisory Service,2000. ForeignPortfolioInvestment 1999(InKorean). Financial Times. 2012. Worry more about North Korea than about Iran (by Ian Bremmer). April 12. Fisman, R, Hamao, Y., Wang, Y., 2013. The impact of interstate frictions on economic exchange: Evidencefromshocks toSino-Japanese relations, WorkingPaper. 28

Forbes, K., Warnock, F., 2012. Capital flow waves: Surges, stops, flight, and retrenchment. JournalofInternational Economics88,235–251. Froot, K.,O’Connell, P.,Seasholes, M.,2001. ThePortfolio FlowsofInternational Investors. Journal ofFinancial Economics59,151–193. Gehrig, T., 1993. An information based explanation of the domestic bias in international equity investment. TheScandinavian JournalofEconomics21,7–109. Glick, R., Taylor, A., 2010. Collateral Damage: Trade Disruption and the Economic Impact of War. ReviewofEconomicsandStatistics92,102–127. Grauer, R.,Hakansson, N.,1987. Gainsfrominternational diversification: 1968–85 returns onportfoliosofstocksandbonds. JournalofFinance42,721–741. Grinblatt, M., Keloharju, M., 2000. The investment behavior and performance of various investor types: AstudyofFinland’suniquedataset. Journal ofFinancialEconomics55,43–67. Grubel,H.,1968. Internationally diversifiedportfolios. AmericanEconomicReview58,1299–1314. Hau, H., Massa, M., Peress, J., 2010. Do demand curves for currencies slope down? Evidence from theMSCIglobalindexchange. ReviewofFinancialStudies23,1681–1717. Kang, J., Stulz, R., 1997. Why is there a home bias? An analysis of foreign portfolio equity ownershipinJapan. Journal ofFinancialEconomics46,3–28. Kang, H., Lee, D., Park, K., 2010. Does the difference in valuation between domestic and foreign investors help explain their distinct holdings of domestic stocks? Journal of Banking and Finance 34, 2886–2896. Karolyi, G., 2002. Did the Asian financial crisis scare foreign investors out of Japan? Pacific-Basin FinanceJournal10,411–442. Karolyi, G., 2006. The consequences of terrorism for financial markets: what do we know? WorkingPaper,OhioStateUniversity. Karolyi, G., Martell, R., 2010. Terrorism and the stock market. International Review of Applied FinancialIssuesandEconomics2,285-314. Kim, C. 2011. Inter-Korean relations and Korea Discount. Journal of Peace and Unification Studies 3(1),219-252. Kim, Y. H. and H. Jung. 2014. Investor trading behavior around the time of geopolitical risk events: evidence fromSouthKorea. BankofKoreaWorkingPaper. Kim, Y. H., H. G. Kang, and J. K. Lee. 2016. Can big data forecast North Korean military aggres- 29

sion? WorkingPaper. Kim, W., Wei, W., 2002. Foreign portfolio investors before and during a crisis. Journal of InternationalEconomics56,77–96. KoreaExchange, 2010. KRXfactbook2010. https://global.krx.co.kr/board/GLB0205020100/bbs#. Korea Exchange, 2011. Introduction to trading at KRX stock market. https://global.krx.co.kr/board/ GLB0205020100/bbs#. Martin, P., Mayer, T., Thoenig, M., 2008. Make Trade Not War? Review of Economic Studies 75, 865–900. Lakonishok, J., Shleifer, A., Vishny, R., 1992. The impact of institutional trading on stock prices. JournalofFinancialEconomics32,23–44. Lee, K. 2006. The Impact of News about North Korea on the Stock and Foreign Exchange Markets. TheJournalofNorthEastAsianEconomicStudies18(1),61–90(inKorean). Levy,H.,Samat,M.,1970. International diversification ofinvestment portfolios. AmericanEconomic Review50,668–675. McCandless, G., 1996. Money, expectations, and the US Civil War. American Economic Review 86,661–671. Nam, S. 2002. The North-South Korean Relation and Country Risk: The Impact of North-South KoreanRelationonStockMarketIndex. TheKoreanPoliticalScienceAssociation (inKorean) Nofsinger, J., Sias, R., 1999. Herding and feedback trading by institutional and individual investors. JournalofFinance54,2263–2295. Nordhaus, W., 2002. The economic consequences of a war with Iraq. In: War with Iraq: Costs, Consequences, andAlternatives. AmericanAcademyofArtsandSciences. Pak, Y., Y. Kim, M. Song, and Y. Kim. 2015. Shock Waves of Political Risk on the Stock Market: TheCaseofKoreanCompaniesintheU.S.DevelopmentandSociety44(1),143–165. Petkova, R., Zhang, L., 2005. Is value riskier than growth? Journal of Financial Economics 78, 187– 202. Portes, R., Rey, H., 2005. The determinants of cross-border equity flows. Journal of International Economics65,269–296. Rigobon, R., Sack, B., 2005. The effects of war risk on US financial markets. Journal of Banking andFinance29,1769–1789. Rothenberg, A., Warnock, F., 2011. Sudden flight and true sudden stops. Review of International 30

Economics19,509-524. Seasholes, M., Zhu, N., 2010. Individual Investors and Local Bias, Journal of Finance 65, 1987– 2011. Solnik, B.,1974. Whynotdiversify internationally rather thandomestically? Financial AnalystsJournal30,48–54. Wermers, R., 1999. Mutual fund herding and the impact on stock prices. Journal of Finance 54, 581–622. Willard,K.,Guinnane, T.,Rosen,H.,1996. Turningpointsinthecivilwar: viewsfromtheGreenback Market. AmericanEconomicReview86,1001–18. Zussman, A., Zussman, N., Nielsen, M., 2008. Asset perspectives on the Israeli-Palestinian conflict. Economica75,84–115. 31

Table1 SummaryStatistics Thistableprovidesdescriptivestatisticsforoursampleof53stocksbyyearfrom1999through2010. Thesecondcolumn reports the daily returns averaged over the 53 sample firms each year. The next three columns report the daily turnover averaged and summed over the sample firms as well as the fraction of total turnover accounted for by foreign investors. Turnover (unit: KRW billions) is defined as the number of shares traded multiplied by the price at which the shares are traded. Thefinalthreecolumnspresenttheyear-endmarketcapitalizationsaveragedandsummedoverthesamplefirmsas wellasthefractionoftotalmarketcapitalizationownedbyforeigninvestors. Year MeanDaily DailyTurnover(unit:KRWbillions) Year-endMarketCap(unit:KRWbillions) Return(%) Mean Sum Foreigners(%) Mean Sum Foreigners(%) 1999 0.24 28.0 1,485.1 7.87 4,439.3 235,284.9 22.53 2000 -0.18 24.1 1,278.5 14.24 2,351.6 124,634.0 32.34 2001 0.25 19.2 1,018.3 14.65 3,067.0 162,549.8 40.25 2002 0.01 27.2 1,443.1 15.67 2,988.5 158,389.7 41.11 2003 0.19 20.8 1,102.3 19.57 4,035.6 213,886.8 44.80 2004 0.13 21.4 1,132.1 27.14 4,439.2 235,278.2 46.21 2005 0.25 27.4 1,454.8 25.41 6,583.9 348,944.4 43.06 2006 0.10 33.1 1,755.8 29.14 7,091.1 375,830.2 40.21 2007 0.23 50.3 2,668.2 29.00 9,304.1 493,116.1 34.92 2008 -0.15 49.2 2,607.8 29.56 5,902.1 312,813.1 31.78 2009 0.22 51.6 2,732.2 20.82 9,167.0 485,849.9 37.07 2010 0.12 52.8 2,796.7 23.29 11,464.8 607,636.9 37.45 32

Table2 StockMarketPerformanceAroundNorthKoreanAttacks This table describes the North Korean attacks on South Korea and South Korea’s stock market performance on the days of attacks. Thetablelistsall13attacksconsidered inthestudyandthenatureoftheseattacks. Thethirdcolumn reports attack-dayreturnsoverthepreviousone-year returnsaveragedoverthe53samplestocks. Iftheattackoccurredwhenthe stockmarketwasclosed,thefirsttradingdayaftertheattackisused.ThelasttwocolumnsreporttheKOSPI(SouthKorea’s mainstockexchange)attack-dayreturnsoverthepreviousone-yearreturnsaswellasthechangeintheKOSPI’stotalmarket capitalizationfollowingattacks. ThemarketcapitalizationisreportedinU.S.dollarsandreflectstheexchangerateonthe dayofagivenattack. Thefinalsixrowssummarizetheresultsforthetwosubsamplesofattackssortedonthenewsarticle coverage aswell asfor the full sample. Wesearch theWall StreetJournal, Financial Times, and the New York Timesin Factivatoidentifythearticlescoveringtheattacksforthedurationofsevendaysfollowingeachattack.Thesubsamplewith articlecountslowerthan(higherthanorequalto)themedianincludestheattacksthatreceiveless(more)mediaattentions. SampleFirms KOSPI MeanDaily Daily ChangeinMarket DateofAttack NatureofAttack Return(%) Return(%) Cap($millions) June15,1999 NavalBattle -3.19 -2.63 -4,464 November27,2001 BorderFight -0.59 -0.76 -1,311 June29,2002 NavalBattle -0.26 0.36 1,250 July17,2003 BorderFight -2.46 -2.37 -6,251 October9,2006 NuclearTest -3.33 -2.46 -16,675 August6,2007 BorderFight -1.29 -1.32 -12,031 May25,2009 NuclearTest 0.55 -0.13 -1,254 November10,2009 NavalBattle -0.03 0.20 2,446 January27,2010 ArtilleryBattle -0.64 -0.87 -5,564 March26,2010 S.KoreanNavalShipSunk -0.82 -0.48 -2,470 May20,2010 AnnouncementontheShip* -1.99 -1.90 -13,722 October29,2010 BorderFight 1.83 1.62 15,656 November23,2010 ArtilleryBattle -0.99 -0.87 -7,393 SubsamplesSortedonNewsArticleCounts ArticleCounts<Median -0.57 -0.56 -1,375 ArticleCounts Median -1.40 -1.18 -6,219 ≥ All13attacks Mean -1.02 -0.89 -3,983 t-stat (-8.95) (-2.58) N 689 13 *AlthoughtherewasspeculationthatNorthKoreawasinvolvedinthesinkingoftheSouthKoreannavalship onMarch26,2010,theofficialinvestigationresultsannouncedonMay20,2010confirmedthatNorthKoreais responsibleforsinkingtheship. 33

Table3 Attack-dayReturnRegressions Thistablepresentsaregressionanalysisofreturnsonattackdaysandpre-attackdays(fivetradingdaysprecedingtheattacks). The daily returnsof the 53 sample stocks are regressed against a set of firmcharacteristics. Firmcharacteristic variables include size (log of market capitalization), leverage ratio (the ratio of total liabilities to total assets minus book value of equityplusmarketvalueofequity),book-to-marketratio(bookvalueofequitydividedbythemarketvalueofequity),return onassets(netincomedividedbytotalassets),beta(whichisestimatedusingweeklyreturnsdataoverthepreviousone-year period), export/sales (the ratio of export revenues to total sales), and industry indicators. The observations are classified into three industry groups: manufacturing (49% of the sample stocks), financial services (19%), and others (32%). Also includedareKaesongandDefenseindicators,foreignerownershipstakesattheendoftheprevioustradingday,andforeign investors’nettradingvolumeonattackdays. KaesongissettooneifthecompanyhasproductionfacilitiesintheKaesong IndustrialRegion,andDefenseissettooneifthecompanysuppliesmaterialstotheDepartmentofDefense.Thefirstcolumn reportsresultsusingallstock/dayobservationsonattackdays.Thenextcolumnpresentsregressionresultsusingallstock/day observationsonpre-attackdays. Thefinalcolumnreportsthedifferencesincoefficientsbetweenthefirsttworegressions. Thecorrespondingt-statisticsarereportedinparentheses. AttackDays Pre-AttackDays Differences Constant -5.167*** 0.860 -6.028*** (-3.09) (0.96) (-3.18) Ln(MarketCap) 0.321*** -0.042 0.363*** (2.94) (-0.72) (2.93) Leverage 0.006 0.000 0.006 (1.17) (0.44) (1.15) Book-to-Market -0.154 -0.066 -0.088 (-1.04) (-0.87) (-0.53) ReturnonAssets -0.426 -0.354 -0.072 (-0.54) (-0.84) (-0.08) Beta -0.210 -0.035 -0.175 (-0.73) (-0.23) (-0.53) Manufacturing 0.566* 0.081 0.485 (1.96) (0.52) (1.48) Financial -0.794** 0.050 -0.844* (-2.06) (0.24) (-1.93) Export/Sales -0.756 0.052 -0.808 (-1.59) (0.20) (-1.50) Kaesong -0.086 0.078 -0.164 (-0.29) (0.48) (-0.48) Defense 0.052 0.131 -0.080 (0.13) (0.64) (-0.18) ForeignerStake 0.004 -0.004 0.007 (0.45) (-0.83) (0.79) ForeignerNetVolume 0.044*** 0.063*** -0.019* (4.37) (11.70) (-1.65) Adj.R2 0.064 0.036 N 689 3,445 *,**,and***indicatesignificanceatthe10,5,and1percent. 34

Table4 TradingActivityAroundtheNorthKoreanAttacks PanelAreportsthedailytradingvolumeaggregatedoverthe53samplestocksandaveragedoverthe13attackdays.Trading volumeisreportedforthedaysofattacksandthefiveprecedingtradingdaysbythreeinvestortypes: foreigners,domestic institutions,anddomesticindividuals. Tradingvolume(unit: shares)measuresthenumberofsharestradedbyeachinvestor type.Net/Total(%)istheratioofnetvolumetototalvolume,wherenetvolumeisthedifferencebetweenbuyvolumeandsell volumeandtotalvolumeisthesumofbuyvolumeandsellvolume. Thepercentagechangeintradingvolumecomparesthe attack-daytradingvolumetodailytradingvolumeaveragedoverthepreviousfivetradingdays.PanelBpresentsaregression analysisofnettradingvolumebyeachofthethreeinvestorgroups. Thetradingvolumeofthe53samplestocksisregressed againstasetoffirmcharacteristicvariablesforstock/dayobservationsonattackdaysandonpre-attackdaysseparately,where pre-attackdaysaredefinedasfivetradingdaysprecedingtheattacks. Thedependentvariableisdefinedasavariationoflog transformation of net trading volume as follows: ln(Vol+√1+Vol2). Thet-statisticsare reported in parentheses. Also reportedarethet-statisticsfor thedifferencesincoefficientsbetween attack-daysandpre-attackdays. Firmcharacteristic variablesincludesize(logofmarketcapitalization),leverageratio(theratiooftotalliabilitiestototalassetsminusbookvalue ofequityplusmarket valueofequity), book-to-market ratio(bookvalueofequitydividedbythemarket valueofequity), returnonassets(netincomedividedbytotalassets), beta(whichisestimatedusingweeklyreturnsdataovertheprevious one-yearperiod),andexport/sales(theratioofexportrevenuestototalsales). AlsoincludedaretheKaesongandDefense indicators,foreignerownershipstakeattheendoftheprevioustradingday,andindustryindicators(manufacturing,financial services,andothers). PanelA:DailyTradingVolumeAveragedOver13AttackDays Buy Sell Total Net/Total(%) Foreigners Fivepreviousdays 152,238 150,809 303,047 0.5% Attackdays 166,570 157,624 324,194 2.8% Changes(%) 9.4% 4.5% 7.0% DomesticInstitutions Fivepreviousdays 203,895 223,845 427,740 -4.7% Attackdays 209,874 204,690 414,564 1.3% Changes(%) 2.9% -8.6% -3.1% DomesticIndividuals Fivepreviousdays 1,748,307 1,726,667 3,474,974 0.6% Attackdays 1,288,546 1,301,223 2,589,769 -0.5% Changes(%) -26.3% -24.6% -25.5% 35

PanelB:NetTradingVolumeRegressions Foreigners D.Individuals D.Institutions Attack Pre-Attack Attack Pre-Attack Attack Pre-Attack Constant -1.743 3.059 0.242 -7.362** 0.772 6.028** (-0.27) (1.08) (0.04) (-2.47) (0.12) (2.06) Ln(MarketCap) -0.003 -0.113 -0.001 0.543*** -0.183 -0.456** (-0.01) (-0.61) (-0.003) (2.78) (-0.43) (-2.38) Leverage -0.024 -0.001 -0.012 0.001 0.029 0.001 (-1.26) (-0.98) (-0.620) (0.95) (1.48) (0.93) Book-to-Market 1.062* 0.050c -0.937 0.398b 0.897 -0.814***,a (1.89) (0.21) (-1.582) (1.56) (1.56) (-3.26) ReturnonAssets 1.002 -2.231* -6.799** -0.949c 5.673* 4.254*** (0.33) (-1.66) (-2.153) (-0.67) (1.85) (3.08) Beta 1.442 0.047 0.347 0.090 -1.784 -0.854* (1.31) (0.10) (0.30) (0.17) (-1.58) (-1.68) ForeignerStake 0.002 0.002 -0.031 -0.025* 0.035 0.002 (0.06) (0.18) (-0.97) (-1.76) (1.12) (0.17) Manufacturing 1.846* -0.190c 0.648 0.808 -0.740 -0.247 (1.68) (-0.38) (0.56) (1.56) (-0.66) (-0.49) Financial -1.845 -1.546** 1.799 1.251* 1.771 0.320 (-1.26) (-2.35) (1.17) (1.81) (1.18) (0.47) Export/Sales -5.114*** -1.965** 0.136 -1.960** 5.575*** 3.414*** (-2.83) (-2.43) (0.07) (-2.31) (3.02) (4.10) Kaesung 0.388 -0.840* 0.800 0.701 -1.166 0.374 (0.34) (-1.65) (0.67) (1.31) (-1.00) (0.71) Defense 1.708 0.077 0.540 -0.190 0.367 -0.403 (1.17) (0.12) (0.35) (-0.28) (0.25) (-0.60) Adj.R2 0.007 0.001 0.004 0.003 0.017 0.012 N 689 3,445 689 3,445 689 3,445 *,**,and***indicatesignificanceatthe10,5,and1percentlevels,respectively.a,b,andcdenotesignificant differencesbetweenattackandpre-attackdaysatthe1,5,and10percentlevels,respectively. 36

Table5 BuyRatios Thistablereportsthebuyratiosofthewinnersandlosersselectedbyeachinvestortypeonattackdaysandnon-attackdays, respectively.Thebuyratioisdefinedasfollows: SharesofFirmiPurchasedonDayt BuyRatioit= . SharesofFirmiPurchasedonDayt+SharesofFirmiSoldonDayt Thestock/dayobservationsaresortedbysubsequentreturnsover5-day,20-day,and60-daywindows,wherethefuturereturns arethesumofthedailyreturnsforeachstockstartingfromdayt+1. Forattackdays,weselectthe100highestandlowest returns,whichisapproximatelythetopandbottom15%ofallobservations(=100/689attack-dayobservations),foreachof the5-day,20-day,and60-dayperiods.Fornon-attackdays,weselectthesamefractionofobservationsandassigntheminto thehighest andlowestreturncategories. Wethencalculatethemeanbuyratioforeachgroupofinvestorsforeachofthe returnwindowsonattackandnon-attackdays.Thefirsttwocolumnsreport,foreachinvestortype,themeanbuyratiosofthe 100highest-returnstocksandofthe100lowest-returnstocksonattackdays. Thenextcolumnshowsthedifferenceinbuy ratiosbetweenthehighest-returnstocksandthelowest-returnstocks,whichisdenotedbyH–L.Thenextthreecolumnsreport themeanbuyratiosandtheirdifferences(H–L)fornon-attackdays.Thefinaltwocolumnsreportthedifferencebetweenthe H–LvalueonattackdaysandtheH–Lvalueonnon-attackdaysforeachoftheinvestorgroups,whichisdenotedbyAttack (H–L)–NoAttack(H–L).Thecorrespondingt-statisticsarereportedinparentheses. PanelA:5-DayFutureReturns AttackDays NoAttackDays Attack(H–L)– Highest Lowest H–L Highest Lowest H–L NoAttack(H–L) AverageReturn(%) 13.64 -6.93 13.09 -11.45 MeanBuyRatios Foreigner 0.51 0.46 0.05 0.51 0.51 0.00 0.03 (0.72) Institution 0.55 0.50 0.05 0.50 0.47 0.03 0.02 (0.70) Individual 0.46 0.50 -0.04 0.50 0.49 0.01 -0.04*** (-2.62) PanelB:20-DayFutureReturns AttackDays NoAttackDays Attack(H–L)– Highest Lowest H–L Highest Lowest H–L NoAttack(H–L) AverageReturn(%) 26.84 -11.06 25.99 -21.21 MeanBuyRatio Foreigner 0.50 0.53 -0.03 0.51 0.52 -0.01 -0.03 (-0.56) Institution 0.58 0.44 0.14 0.50 0.47 0.03 0.11*** (3.27) Individual 0.47 0.51 -0.04 0.49 0.49 0.00 -0.04*** (-2.84) PanelC:60-DayFutureReturns AttackDays NoAttackDays Attack(H–L)– Highest Lowest H–L Highest Lowest H–L NoAttack(H–L) AverageReturn(%) 46.97 -18.53 50.15 -33.82 MeanBuyRatio Foreigner 0.52 0.51 0.01 0.52 0.53 -0.01 0.01 (0.25) Institution 0.56 0.41 0.15 0.49 0.48 0.01 0.15*** (4.32) Individual 0.48 0.49 -0.01 0.49 0.49 0.00 -0.01 (-0.78) *,**,and***indicatesignificanceatthe10,5,and1percentlevels,respectively. 37

Table6 OrderImbalancesofLaggedReturnPortfolios PanelAshowsorderimbalancesforquintileportfoliosformedbasedonstockreturnsontheprevioustradingday.Theorder imbalanceforeachfirm/tradingdayisdefinedasthenetbuyvolumedividedbyaveragedailyvolumeoverthepreviousten tradingdays,wherenetbuyvolumeisthenumberofsharesboughtminusthenumberofsharessold: SharesofFirmiPurchasedonDayt-SharesofFirmiSoldonDayt OIit= . AverageDailyVolumefromDayt-10throughDayt-1 Theorderimbalancesforattackandnon-attackdaysaresortedseparatelyforeachofthethreeinvestorgroups–foreigners, domesticinstitutions,anddomesticindividuals–intoquintilesbasedontheprevious-dayreturns. Theorderimbalancesare thenaveragedacrosseachquintile. Thecorrespondingt-statisticsarereportedinparentheses. PanelBrepeatstheanalysis usinga2-daywindow. Theaverageorderimbalancesarecalculatedoveratwo-dayperiodforthequintileportfoliosformed based on stock returns over the previous two trading days. Panel C reports the average order imbalances over a five-day windowforthequintileportfoliosformedbasedonstockreturnsoverthepreviousfivetradingdays. PanelA:1-DayWindow Prior-Day OrderImbalances ReturnPortfolios Foreigners Individuals Institutions Non-AttackDays(numberofobservaitons: 156,668) P1(lowest) -0.041 (-7.18) 0.105 (4.56) -0.060 (-3.58) P2 -0.015 (-8.15) 0.032 (21.04) -0.019 (-11.48) P3 0.003 (2.24) -0.010 (-6.41) 0.004 (2.60) P4 0.024 (15.52) -0.046 (-18.81) 0.017 (9.67) P5(highest) 0.044 (26.32) -0.081 (-36.62) 0.032 (19.74) P5–P1 0.085 (14.18) -0.186 (-8.05) 0.092 (5.49) AttackDays(numberofobservaitons: 669) P1(lowest) -0.011 (-0.65) 0.021 (1.17) -0.015 (-0.90) P2 0.044 (1.84) 0.014 (0.68) -0.056 (-2.40) P3 0.068 (1.33) -0.042 (-1.82) -0.027 (-0.71) P4 0.012 (0.62) -0.054 (-3.06) 0.038 (2.15) P5(highest) 0.009 (0.74) -0.045 (-3.05) 0.029 (2.30) P5–P1 0.020 (0.97) -0.066 (-2.88) 0.044 (2.15) 38

PanelB:2-DayWindow Prior-Day OrderImbalances ReturnPortfolios Foreigners D.Individuals D.Institutions Non-AttackDays P1(lowest) -0.046 (-3.37) 0.132 (2.52) -0.082 (-2.20) P2 -0.008 (-3.02) 0.015 (4.21) -0.013 (-5.15) P3 0.014 (5.15) -0.006 (-0.61) -0.012 (-1.19) P4 0.033 (11.69) -0.050 (-16.17) 0.013 (4.12) P5(highest) 0.034 (12.72) -0.064 (-17.11) 0.020 (7.19) P5–P1 0.080 (5.73) -0.197 (-3.74) 0.102 (2.73) AttackDays P1(lowest) 0.000 (0.00) 0.058 (1.64) -0.050 (-1.04) P2 -0.019 (-0.77) 0.064 (2.22) -0.040 (-1.45) P3 0.098 (3.34) -0.095 (-3.13) -0.004 (-0.15) P4 -0.002 (-0.07) -0.036 (-1.09) 0.036 (0.94) P5(highest) 0.023 (0.87) -0.077 (-3.06) 0.028 (0.79) P5–P1 0.023 (0.34) -0.135 (-3.04) 0.077 (1.27) PanelC:5-DayWindow Prior-Day OrderImbalances ReturnPortfolios Foreigners D.Individuals D.Institutions Non-AttackDays P1(lowest) -0.130 (-2.50) 0.297 (2.46) -0.194 (-2.66) P2 -0.008 (-1.50) 0.030 (4.33) -0.023 (-4.28) P3 0.045 (8.13) -0.046 (-7.09) -0.009 (-1.75) P4 0.070 (12.68) -0.110 (-20.34) 0.024 (4.46) P5(highest) 0.043 (7.93) -0.097 (-20.19) 0.043 (9.05) P5–P1 0.173 (3.31) -0.394 (-3.26) 0.237 (3.24) AttackDays P1(lowest) -0.022 (-0.34) 0.082 (1.30) -0.063 (-1.24) P2 0.109 (0.98) 0.087 (1.27) -0.210 (-2.27) P3 0.112 (1.85) 0.010 (0.16) -0.122 (-2.10) P4 0.041 (0.60) -0.096 (-1.59) 0.040 (0.75) P5(highest) 0.076 (1.28) -0.133 (-2.05) 0.017 (0.26) P5–P1 0.097 (1.11) -0.215 (-2.38) 0.081 (0.96) 39

Table7 OrderImbalancesRegressions Thistablepresentstheregressionresultsoftheattack-dayfeedbacktradingstrategyofthethreeinvestortypesrelativetothat ofnon-attackdays.Theregressionspecificationisasfollows: OI it = b 0+b 1 Attackt+b 2 r i,t 1+b 3 r i,t 1 Attackt+b 4 rm t 1+b 5 rm t 1 Attackt+b 6 s t 1+b 7 s t 1 Attackt − − − − − − +b 8 Kaesong i+b 9 Kaesong i Attackt+b 10 Defense i+b 11 Defense i Attackt+e it , whereOIit istheorderimbalanceoffirmiondayt,rit isthestockreturnoffirmiondayt,rmt isthereturnontheKOSPI compositeindexondaydayt,andst istheUSD-KRWexchangerateondaydayt. Attackissettooneonattackdaysand zerootherwise.KaesongissettooneifthecompanyhasproductionfacilitiesintheKaesongIndustrialRegion,andDefense issettooneifthecompanysuppliesmaterialstotheDepartmentofDefense. Thecorrespondingt-statisticsarereportedin parentheses. Foreigners D.Individuals D.Institutions Constant 0.427*** -0.558*** -0.120* (5.86) (-7.25) (-1.70) Attackt 1.349 -1.254 -0.141 (1.21) (-1.07) (-0.13) r 0.780*** -1.616*** 0.760*** t 1 − (37.61) (-73.83) (37.88) r t 1 Attackt -0.600* 0.806** -0.155 − · (-1.84) (2.34) (-0.49) rm -0.059 1.354*** -1.151*** t 1 − (-1.46) (31.90) (-29.57) rm t 1 Attackt -1.160* 0.684 0.100 − · (-1.95) (1.09) (0.17) s -0.266*** 0.387*** -0.151* t 1 − (-3.04) (4.19) (-1.78) s t 1 Attackt -7.302*** 1.144 4.049*** − · (-4.56) (0.68) (2.61) Kaesongt -0.116 0.079 0.042 (-0.77) (0.50) (0.29) Kaesongt Attackt 4.738** -1.426 -2.936 · (2.08) (-0.59) (-1.33) Defenset -0.372* -0.061 -0.091 (-1.93) (-0.30) (-0.49) Defenset Attackt 0.154 -0.385 1.416 · (0.05) (-0.13) (0.50) Adj.R2 0.071 0.034 0.010 N 157,410 157,410 157,410 TestsforLinearCombinationsofCoefficients r t 1 + r t 1 Attackt 0.180 -0.810** 0.605* − − · t-statistic 0.55 -2.35 1.92 40

Table8 MagnitudeofAttacks Thistableexaminessubsamplesofattackswithdifferentmagnitudes. Theattack-daysampleissortedintotwosubsamples accordingtotheattack-dayKOSPIreturns.Anattackisconsideredtobemore(less)severeiftheattack-daymarketreturnis lower(higher)thantheaverageattack-daymarketreturn. PanelAreportsmeanbuyratiosforthetwosubsamples. Foreach subsample,weselectthesamefractionofobservationsforthehighestandlowestreturncategoriesandcalculatebuyratios asdescribedpreviously. [Attack(H–L)–NoAttack(H–L)]measuresthedifferencebetweentheH–Lvalueonattackdays andtheH–Lvalueonnon-attackdays,whereH–Listhedifferenceinbuyratiosbetweenthehighest-returnstocksandthe lowest-returnstocks.PanelBexaminesorderimbalancesseparatelyformoresevereattacksandlesssevereattacks.Foreach ofthesubsamples,orderimbalancesarecalculatedforquintileportfoliosformedbasedontheprevious-dayreturnforeach ofthethreeinvestorgroups.Thecorrespondingt-statisticsarereportedinparentheses. PanelA:MeanBuyRatios InvestorType Attack(H–L)–NoAttack(H–L) 5-DayFutureReturns 20-DayFutureReturns 60-DayFutureReturns LessSevereAttacks:Higher-Than-MedianKOSPIReturnSubsample Foreigner 0.10 (1.47) -0.02 (-0.30) -0.02 (-0.29) Institution -0.04 (-0.80) 0.01 (0.27) 0.02 (0.35) Individual 0.00 (-0.22) -0.03 (-1.44) -0.01 (-0.26) MoreSevereAttacks:Lower-Than-MedianKOSPIReturnSubsample Foreigner 0.00 (-0.04) 0.14* (1.86) 0.08 (1.14) Institution -0.02 (-0.38) 0.12** (2.52) 0.12** (2.55) Individual -0.03 (-1.31) -0.06*** (-3.13) -0.03 (-1.46) *,**,and***indicatesignificanceatthe10,5,and1percentlevels,respectively. PanelB:OrderImbalances Prior-Day OrderImbalances ReturnPortfolios Foreigners D.Individuals D.Institutions LessSevereAttacks:Higher-than-medianKOSPIreturnsubsample P1(lowest) -0.007 (-0.30) 0.028 (1.21) -0.025 (-1.02) P2 0.034 (1.29) 0.042 (1.58) -0.077 (-3.09) P3 0.021 (1.08) -0.035 (-1.33) 0.014 (0.60) P4 0.021 (0.86) -0.040 (-1.92) 0.010 (0.41) P5(highest) 0.033 (2.30) -0.040 (-2.22) 0.003 (0.16) P5–P1 0.039 (1.48) -0.068 (-2.32) 0.027 (0.93) MoreSevereAttacks:Lower-than-medianKOSPIreturnsubsample P1(lowest) -0.018 (-0.72) 0.018 (0.69) -0.010 (-0.50) P2 0.064 (1.45) -0.023 (-0.72) -0.035 (-0.85) P3 0.135 (1.18) -0.071 (-1.83) -0.064 (-0.76) P4 0.006 (0.16) -0.077 (-2.10) 0.065 (2.41) P5(highest) -0.025 (-1.32) -0.039 (-1.67) 0.059 (3.53) P5–P1 -0.007 (-0.24) -0.057 (-1.62) 0.069 (2.61) 41

Table9 ExchangeRateAdjustedBuyRatios Thistablereportsmeanbuyratiosadjustedbyexchangeratechanges. ThestockreturnsadjustedbyUSD-KRWexchange ratechangesareusedtocalculatethemeanbuyratios.Thecorrespondingt-statisticsarereportedinparentheses. InvestorType Attack(H–L)–NoAttack(H–L) 5-DayFutureReturns 20-DayFutureReturns 60-DayFutureReturns Foreigner 0.06 (1.34) -0.04 (-0.76) 0.00 (0.02) D.Institution 0.03 (0.81) 0.10*** (3.11) 0.13*** (3.99) D.Individual -0.05*** (-3.21) -0.05*** (-3.13) 0.00 (-0.21) *,**,and***indicatesignificanceatthe10,5,and1percentlevels,respectively. 42

Table10 OrderImbalancesandChronologyofAttacks Thistable compares order imbalances for the earlier period and those for thelater period. Thesample issorted into two subsamples according to the chronological order of the attacks. For each of the two subsamples, order imbalances are calculatedforquintileportfoliosformedbasedontheprevious-dayreturnforeachofthethreeinvestorgroups. PanelA:OrderImbalancesintheearlier-periodsubsample Prior-Day OrderImbalances ReturnPortfolios Foreigners D.Individuals D.Institutions Non-AttackDays P1(lowest) -0.047 (-6.36) 0.107 (3.62) -0.058 (-2.67) P2 -0.020 (-8.66) 0.028 (15.83) -0.010 (-5.33) P3 0.001 (0.72) -0.012 (-6.49) 0.008 (4.46) P4 0.029 (15.03) -0.046 (-14.50) 0.013 (6.02) P5(highest) 0.050 (25.24) -0.076 (-27.53) 0.022 (11.12) P5–P1 0.096 (12.58) -0.183 (-6.10) 0.079 (3.64) AttackDays P1(lowest) -0.013 (-0.46) -0.014 (-0.52) 0.018 (0.74) P2 0.039 (1.29) -0.053 (-1.54) 0.006 (0.20) P3 0.179 (1.43) -0.110 (-2.57) -0.079 (-0.87) P4 0.047 (1.40) -0.093 (-3.12) 0.042 (1.81) P5(highest) 0.010 (0.60) -0.028 (-1.59) 0.012 (0.87) P5–P1 0.023 (0.73) -0.014 (-0.45) -0.006 (-0.23) PanelB:OrderImbalancesinthelater-periodsubsample Non-AttackDays P1(lowest) -0.022 (-7.18) 0.097 (28.67) -0.069 (-21.77) P2 -0.001 (-0.38) 0.044 (14.43) -0.042 (-15.15) P3 0.009 (3.73) -0.004 (-1.55) -0.008 (-3.03) P4 0.012 (4.72) -0.046 (-15.70) 0.027 (9.79) P5(highest) 0.024 (8.18) -0.099 (-33.28) 0.067 (23.75) P5–P1 0.046 (10.84) -0.196 (-43.63) 0.135 (32.13) AttackDays P1(lowest) -0.009 (-0.46) 0.053 (2.37) -0.045 (-2.13) P2 0.048 (1.33) 0.069 (2.96) -0.108 (-3.06) P3 -0.003 (-0.16) 0.001 (0.03) 0.007 (0.32) P4 -0.014 (-0.61) -0.025 (-1.20) 0.035 (1.36) P5(highest) 0.008 (0.43) -0.067 (-2.68) 0.050 (2.25) P5–P1 0.018 (0.63) -0.120 (-3.58) 0.095 (3.10) 43