An Event Study of Swedish Banks Stock Price Reactions to the Baltic Crisis

Transcription

1 STOCKHOLM SCHOOL OF ECONOMICS Master Thess n Internatonal Economcs and Fnance Sprng 2009 An Event Study of Swedsh Banks Stock Prce Reactons to the Baltc Crss Chrstoffer Carlborg 21420@student.hhs.se Davd Dembrower 21435@student.hhs.se Abstract The ongong fnancal crss n the Baltc States has rased concerns about the Swedsh bankng system. For ths purpose we examne how the four largest Swedsh banks stock returns are affected by economc events from the Baltc States, usng an event parameter approach. Wth our event study methodology we can test f the market dscrmnates between banks wth dfferent Baltc loan exposure or f contagon effects are present. For the events that had an mpact on the Swedsh banks we fnd evdence that the market only partally separated between the banks accordng to ther exposure. Key Words: Fnancal Crss, Baltc States, Swedsh Banks, Event Study Tutor: Rchard Frberg Examnator: Mats Lundahl Dscussants: Claes Hermodsson (21429) and Eml Kozlowsk (21428) Presentaton: 11 June 2009 Locaton: Stockholm School of Economcs, Room KAW Tme: 13:15-15:00 We would lke to express our grattude towards our tutor Rchard Frberg for hs help and advce.

2 Table of Contents 1 INTRODUCTION LITERATURE REVIEW DATABASE AND METHODOLOGY DATA SET AND DESCRIPTION OF THE EVENTS EVENT ANALYSIS METHODOLOGY BANK STOCK PRICE RESPONSES TO THE EVENTS FROM THE BALTIC STATES THE EXPOSURE LEVEL TO THE BALTIC STATES IS INCLUDED IN THE MODEL PRESENTATION OF THE RESULTS RESULTS FOR THE EVENT COEFFICIENTS IN EQUATION SYSTEM (1) TESTING HYPOTHESIS H 1 AND H RESULTS FOR THE EVENT COEFFICIENTS IN EQUATION SYSTEM (2) TESTING HYPOTHESIS H ANALYSIS ROBUSTNESS OF OUR FINDINGS CONCLUSIONS REFERENCES...25

3 1 Introducton The ongong fnancal crss has been one of the most dscussed economc topcs durng the past two years. It s among the most severe economc downturns snce the Great Depresson n the 1930s and economes worldwde have been affected to some extent. Sweden and the Swedsh banks are no excepton to be affected by the markets changng condtons and appette for rsk. The largest concerns are drected to the banks operatons n the Baltc States. The three Baltc States - Estona, Latva and Lthuana - are n a deep economc crss. They have experenced a journey from hgh stable GDP growth wth almost balanced budgets to a stuaton wth contractng economes. The countres have suffered from large current account defcts and hgh nflaton. The combnaton of a fxed exchange rate to the euro and hgh nflaton reduced the compettveness for the countres and prced them out of the market. Furthermore, the countres have experenced a boomng real estate market wth a large part of the loans gven n euro and fnanced by foregn owned banks. In 2007 the bubble n the Baltc s burst, leadng to a slowdown n the banks credt grant polces. Housng prces declned together wth prvate consumpton and nvestments, reducng the growth n GDP and rased unemployment (Åslund 2009). The economc development n the Baltc s s of partcular nterest for the large Swedsh banks, snce they are among the most exposed to the regon. From year the outstandng loans towards the Baltc s has ncreased from 30 to approxmately 425 bllon SEK (Forsberg 2009). Ths has rased concerns among nvestors regardng credt losses and mplcatons for the fnancal system. The purpose wth ths paper s to examne f potentally mportant economc events from the Baltc States affect the stock prces of Sweden s four major banks. Our man focus s to examne whether the reactons are n proporton to the banks respectve exposure levels or f contagon effects are present. 1

4 In the Swedsh fnancal system, contagon and systematc rsk s of partcular nterest for at least two reasons. At frst, the bankng sector s hghly concentrated and the four largest banks are closely nter-related. Secondly, they are nvolved n nternatonal bankng and especally SEB and Swedbank are sgnfcantly affected by the troublesome stuaton n the Baltc States. We chose to concentrate on the relatonshp between Swedsh banks and the crss n the Baltc regon, snce ths s one of the man concerns for the Swedsh bankng system n the ongong fnancal crss. In addton, the meda coverage of the most exposed banks, SEB and Swedbank, has been extensve. To our knowledge, the mpact of events from the Baltc crss has not yet been examned on Swedsh data. Our ambton s to ncrease the knowledge of how Swedsh banks nteract n ths fnancal crss. The subsequent parts of the paper are organzed n the followng way. Secton 2 gves an overvew of event studes n general and on smlar studes regardng banks reactons to fnancal crses. Secton 3 provdes a descrpton of the events selected and ntroduce the methodology for the event analyss. In secton 4 we present the results from the event study and n secton 5 we analyze our fndngs. In Secton 6 we dscuss the robustness of the results and secton 7 concludes. 2

5 2 Lterature revew Economsts are often faced wth the problem of measurng the effect of an economc event on the value of a frm. At a frst glance ths seems to be a complcated task however, a measurement can be created relatvely easy wth an event study, usng fnancal market data. The basc statstcal framework s based on measurements of securtes abnormal returns and cumulatve abnormal returns around the tme of an event (MacKnlay 1997). There exst no unque structure to perform event studes, but several papers and academc lterature proposes an outlne that s useful to follow. For example, Campbell et al. (1997) descrbes a general set-up whch s helpful when conductng an event study. In a frst step, the event(s) of nterest s defned and the perod over whch the frms returns wll be examned s determned. Secondly, the selecton crtera for the ncluson of a gven frm have to be determned. Ths may nvolve restrctons such as data avalablty or membershp of a specfc ndustry et cetera. Next, the mpact of an event s measured wth abnormal returns, whch s the actual ex post return of the securty of nterest mnus the normal return over the event wndow. The normal return s defned as the return that would be expected n absence of the event. In the next step the parameters n the model are estmated usng a subset of data known as the estmaton wndow, where the most common s to use days pror the selected event(s). When the normal returns have been estmated, abnormal returns are calculated and the sgnfcance of the event(s) mpact s tested. The frst study that appled an event study was Dolley (1933) and a substantal number of papers have snce then been publshed (MacKnlay 1997). As descrbed n Bowman (1983) the roots for modern event studes can be found n the papers by Ball and Brown (1968), where prce reactons to the unantcpated component of annual accountng earnngs was nvestgated and n Fama, Fsher, Jensen and Roll (FFJR) (1969) who examned f the stock market was effcent wth respect to stock splts. 3

6 Typcal examples where modern event studes are applcable nclude merger and acqustons, stock splts, earnng announcements, events durng fnancal crses and announcements of macro economc varables. Khotar and Warner (2006) descrbe n ther study that, over tme, at least two man refnements have been made n the event study methodology. Frst, there has been a change from monthly to daly securty return data. Ths change has made t possble to obtan more precse measurement of abnormal returns. Implcatons and the use of daly data are dealt wth n Brown and Warner (1985). Second, the methods used to measure abnormal returns and to test for sgnfcance have become more sophstcated. The exstng event study lterature on the reacton of bank stock prces to fnancal crses has been done prmarly on US data (Crouzlle et al. 2006). One of the most central questons rased n these papers s f contagon effects are present or f the market s able to dscrmnate between banks wth low as well as hgh rsk exposure. For example, Klc et al. (2000) examned how events of the Mexcan peso crss affected US banks returns and found, on average, evdence for contagon effects. Mathur and Sundaram (1997) analyzed how events of the Brazlan debt crss affected US banks and concluded that the market was able to dfferentate between banks wth dfferent exposure. Cornell and Shapro (1986) nvestgated how US banks stock prces reacted to dfferent nternatonal debt related news tems and concluded that the market reacted ratonally. There have also been studes based on European data, and these have been predomnantly addressed to the Asan and Russan fnancal crses n the late 90s. Crouzlle et al. (2006) examned how a selecton of European banks was affected by events from both the Asan and Russan fnancal crses and found ndcaton of possble contagon effects. Rme (2003) performed an event study for Swss banks reacton to the Russan collapse and could not draw any general conclusons regardng the reacton of the market, snce the results were mxed. 4

7 The models used n the mentoned papers above dffer somewhat to tradtonal event studes such as n FFJR (1969). The reason for ths s that event studes examnng fnancal crses often use the same events for all frms as argued n Bnder (1985). Ths poses the problems of event and ndustry clusterng as argued n Henderson (1990). Event clusterng refers to events occurrng at or near the same tme whlst ndustry clusterng refers to events concentrated n the same ndustry. The clusterng problem s solved by usng multvarate regresson models wth dummy varables for the event dates as descrbed n Bnder (1985). 5

8 3 Database and Methodology 3.1 Data set and descrpton of the events The data sample covers the four largest Swedsh banks Nordea, Svenska Handelsbanken ( SHB ), Swedbank and Skandnavska Ensklda Banken ( SEB ) whose equty s traded on the Swedsh stock exchange. Swedbank s the most exposed bank to the Baltc regon followed by SEB, Nordea and SHB. The approxmate exposure fgures are shown n Table 1 below. We defne the Baltc exposure for each bank as the book value of Baltc loans over book value of total loans. Table 1: The Baltc exposure for each bank Date Swedbank SEB Nordea SHB % 11% 2% 0,1% % 11% 3% 0,1% Source: Annual reports ( ). 1 Daly bank stock prces and daly stock ndces, for the sample perod January 2007 to the end of March 2009, are taken from Thomson Datastream Internatonal and are adjusted for corporate actons, such as stock splts. Furthermore, n our study we have chosen to treat the Baltc States as one regon. Durng the ongong fnancal crss, a large number of economc events have had an effect on the fnancal markets n the Baltc States. Our goal s as stated before to examne f a selecton of these events affects the Swedsh banks stock prces and f the responses are proportonal to ther exposure. 1 Note: The Baltc exposure for SHB year 2007 s receved from contact wth ts nvestor relatons department, snce t s not avalable n the annual report. The fgures for SEB nclude repurchase agreements and off-balance sheet credt exposure. 6

9 A lst of events was determned by combnng the Reuters and Bloomberg s news archves wth surveys and press releases from the Swedsh central bank ( the Rksbank ), the Internatonal Monetary Fund, the European Investment Bank and the central banks for the Baltc States. Table 2 below summarzes the selected events. Table 2: Economc events from the Baltc States Date Event Descrpton E Q1 GDP growth eases to 6,4% y/y n Lthuana E Q1 GDP growth tumbles to 3,6% y/y n Latva E Q1 GDP growth slows to 0,4% y/y n Estona E4 IMF offcals: The Baltc states could follow Icelandc scenaro E5 Moody's report that Estonan bank credt outlook s negatve E6 Natonalzaton of Parex bank n Latva E7 Moody's report that Lthuanan bank credt outlook s negatve E8 Largest drop n Estonan GDP y/y rate n 14 years E9 IMF and Latva agrees on rescue package E10 Ant-government rots n Latva E11 Lthuana launches economc stmulus plan E12 GDP collapse n Estona down 9,4% y/y E13 Latvan government falls, PM resgned E14 The Rksbank lends to the Estonan central bank E15 EIB lends 1,15b to support SME's n Lthuana. Source: Bloomberg and Reuters archves. The most commonly known events n our lst are the natonalzaton of Parex Bank, the agreement of the rescue package between IMF and Latva and the fall of the Latvan government due to the economc crss. In addton to these, we nclude other economcal and poltcal events from all the Baltc States that mght have an mpact on the valuaton of Swedsh banks exposed to the regon. Our hypothess regardng the selected events s that they wll affect some of the banks valuaton, and that SEB and Swedbank wll be the most affected, snce they are the two most exposed to the Baltc s. We expect that the Baltc States GDP reports for the frst quarter 2008 and the GDP collapse n Estona n the fourth quarter, to have a negatve mpact on the Swedsh banks stock prces wth exposure to the regon. The GDP collapse n Estona was the largest drop n the country s hstory and the GDP estmates for the frst quarter confrmed a slowdown. 7

10 The announcement by Domnque Strauss-Kahn, managng drector of the IMF, stated that there s a rsk for banks n the Baltc States to follow the Icelandc scenaro wth natonalzed banks due to ncreased exposure to strugglng property markets. Ths may negatvely nfluence the market and affect the Swedsh banks. Lkewse, we expect a negatve mpact of the reports from Moody s, confrmng that the ongong economc slowdown wll affect the asset qualty of Lthuanan and Estonan banks. Regardng the natonalzaton of Parex Bank, we antcpate that ths government nterventon wll have a negatve effect. However, we make no pror assumptons about the sgns of the effects regardng the nternatonal agreements and poltcal events. 3.2 Event analyss methodology Bank stock prce responses to the events from the Baltc States At frst we apply an event parameter approach based on the market model to test f the events from the Baltc States ndeed resulted n prce responses for the Swedsh banks. Ths s done rrespectve of the ndvdual bank s exposure to the Baltc States. Clusterng effects and ndustry nduced correlaton of returns may be a problem, snce all frms n our sample are from the same ndustry and affected smultaneously by the economc events of nterest. To deal wth these potental problems, we estmate a multvarate regresson based on Zellner s (1962) seemngly unrelated regresson (SUR) developed by Bnder (1985) and used for nstance by Smrlock and Kaufold (1987), Unal et al. (1993) and Rme (2003). Ths allows for correlaton between resduals when a system of equatons s estmated smultaneously. The multvarate regresson model s estmated as follows: 8

11 R R R R Swed, t SEB, t Nord, t SHB, t R R R 2 R 4 3 M, t M, t M, t M, t n 1 n 1 1 n n 1 D 1, 2, 3, 4, D 1, t D. 2, t D. 3, t 4, t (1) Where R, = the return on the stock of bank j on day t. j t, = regresson parameters for bank j. j j R, = the return of the OMX Stockholm market ndex M t D = dummy varable equal to one on the event day and zero otherwse. j, = captures the effect of event for bank j. j,t = regresson resdual for bank j on day t. The above lnear equaton system (1) s estmated for all banks over the tme nterval startng and endng The coeffcent of nterest s γ j,, whch measures the prce response (abnormal return) of the specfed event. The sgn of the coeffcent determnes whether the event had a negatve or postve mpact on the stock prces. The t-value of the dummy coeffcent s used to assess the sgnfcance of the abnormal returns. To examne the magntude and sgnfcance of the prce responses the events had on the Swedsh banks, the followng two hypotheses are tested usng Wald-tests. Hypothess 1 (H 1 ): The abnormal return for each bank on the event day equals zero. (γ 1 = γ 2 = γ 3 = γ 4 = 0) Smlar test framework s used n Rme (2003) to test the equalty of the dummy coeffcents across the estmated equatons. 9

12 If H 1 s not rejected, then t s problematc to further examne the relatonshp between market response and the exposure level for the banks. Investors mght have the knowledge about the banks exposure, but due to that there s no market response, there wll be no relatonshp between stock prce response and exposure level. Alternatvely, the event mght have too small mpact on the stock prce relatve to nose n the model, that a detecton of a stock prce response s not possble. Smlar hypothess and argumentaton s used n Smrlock and Kaufold (1987). A related hypothess to H 1 s whether the prce response s equal across banks. Hypothess 2 (H 2 ): The abnormal returns are equal across all banks on the event day. (γ 1 = γ 2 = γ 3 = γ 4 ) If H 2 s not rejected, gven the exposure level of the banks, there s evdence that the market dd not dstngush among the exposure levels of the Swedsh banks (Unal et al. 1993). 10

13 The exposure levels to the Baltc States are ncluded n the model If H 2 s rejected, t s mportant to examne whether the observed dfference s n proporton to exposure for each bank or not. The task s to test whether nvestors rewarded or punshed the banks n proporton to the bank specfc Baltc exposure varable. Ths s examned n a SUR framework, followng Smrlock and Kaufold (1987) and Unal et al. (1993), the followng system of equaton s estmated. (2) The varables n ths system (2) are the same as n equaton system (1), except that the dummy varables are weghted by the level of exposure to the Baltc States. The EXP varable s the Baltc exposure for each bank, calculated as the book value of Baltc States loans over book value of total loans for each bank at the end of year 2007 and The banks exposure levels at the end of each year are used for events n the upcomng year. The system s estmated over the same perod as n equaton system (1). n t t M t SHB n t t M t Nord n t t M t SEB n t t M t Swed EXP D R R EXP D R R EXP D R R EXP D R R 1 4, 4 4,, 4 4, 1 3, 3 3,, 3 3, 1 2, 2 2,, 2 2, 1 1, 1 1,, 1 1,

14 We are now able to test f the stock prce response s proportonal to exposure, as n Smrlock and Kaufold (1987), we test the followng hypothess. Hypothess 3 (H 3 ): The event parameters j are equal across all the four banks. ( 1 = 2 = 3 = 4 ) If H 3 s not rejected, there s support for the ratonal-prcng hypothess descrbed n Bruner and Smms (1987). Under ths hypothess, economc events from the Baltc States would affect the banks n proporton to ther exposure. An alternatve hypothess focuses on contagon effects. Under ths hypothess, the market would not be able to dstngush among the exposure levels of dfferent banks. Instead the economc event would generate contagon effects, affectng the Swedsh bank system as a whole (Karafath et al. 1991). Thus, the prce response to the specfed event s not proportonal to exposure across banks. 12

15 4 Presentaton of the results In ths secton we estmate the daly abnormal returns related to the events selected usng the event parameter approach descrbed n equaton system (1). In these regressons we do not consder the banks Baltc exposure level. The purpose of ths frst stage s to examne f the Baltc events selected concded wth abnormal returns for the banks. 4.1 Results for the event coeffcents n equaton system (1) Table 3 presents the results from the event parameter regressons for each event and bank. The frst column (γ) presents the abnormal returns for each bank and event and the second column represents the t-values. The events that had a sgnfcant mpact on the banks are marked wth one or two stars (*) dependng on the sgnfcance level. Table 3: Results from the estmaton of equaton system (1) (**) Sgnfcant on the 5% level (*) Sgnfcant on the 10% level Swedbank SEB Nordea SHB Date Event γ t-value γ t-value γ t-value γ t-value E ** E E * E * E E ** ** E ** E ** E E E ** E ** * E * E ** E **

16 From Table 3 we can observe that fve events for Swedbank and sx events for SEB are sgnfcant, whlst when observng the results for the two other banks only two events for SHB and no events for Nordea are sgnfcant at the 5-10% sgnfcance level. Moreover, 11 events out of 15 are sgnfcant for at least one bank at the 10% level. 4.2 Testng hypothess H 1 and H 2 In our frst hypothess we test whether our dummy coeffcents for each event smultaneously are equal to zero for our dfferent banks or not. Wth our second hypothess we test whether the prce responses are equal across the banks. The results obtaned from these two tests are presented event by event n Table 4 below. Column three and four represents the F-statstc and the P-value for the two tests. Table 4: Results from the tests of H 1 and H 2 Hypothess 1 Hypothess 2 Date Event (γ1 = γ2 = γ3 = γ4 = 0) (γ1 = γ2 = γ3 = γ4) F-statstc P-value F-statstc P-value E * * E E E * E E ** * E ** ** E ** ** E E E ** ** E ** * E E ** ** E ** ** (**) Hypothess rejected on the 5% level (*) Hypothess rejected on the 10% level The results from the tests show that H 1 can be rejected for 8 events at the 10% level and 7 events at the 5% level. Ths mples that these events have sgnfcant mpact for at least one of the banks. As stated above, f H 1 s not rejected t wll be problematc to further examne the relatonshp between market response and the exposure level for the banks. 14

17 Therefore, we wll only focus on the events that can be rejected n H 1. These events are marked n bold. Furthermore, for all the events rejected n H 1, our results show that the hypothess of equalty among banks prce responses on the event day (H 2 ) can be rejected on at least the 10% level. Ths ndcates that the prce responses for these events were not unform across banks. 4.3 Results for the event coeffcents n equaton system (2) We next re-estmate the event parameter model and nclude the Baltc exposure level for each bank specfed as n equaton system (2). Ths s done snce hypothess 2 can be rejected for the events whch had a sgnfcant effect on the Swedsh banks. Therefore, we examne f the observed dfference s due to the exposure to the Baltc regon or not. Table 5 shows the results from the re-estmaton of the model. 2 Table 5: Results from the estmaton of equaton system (2) Swedbank SEB Nordea SHB Date Event t-value t-value t-value t-value E ** E E * E * E E ** ** E ** E ** E E E ** E ** * E * E ** E ** (**) Sgnfcant on the 5% level (*) Sgnfcant on the 10% level 2 The results from the estmatons of equaton system (1) and (2) are avalable from the authors on request. 15

18 The sgnfcance levels and sgns for the events are the same as for the regressons wthout the exposure levels taken nto account. The dfference s that the coeffcent j s approxmately equal to γ j /EXP j and s a measure of prce response of bank j per unt of exposure as descrbed n Smrlock and Kaufold (1987). 4.4 Testng hypothess H 3 Wth the thrd hypothess the equalty of prce responses adjusted for the banks exposure for each event s tested. The results from these tests are shown n Table 6 under headlne hypothess 3. 3 Table 6: Results from the test of H 3 (ncludng the results from tests of H 1 and H 2 ) Hypothess 1 Hypothess 2 Hypothess 3 Date Event (γ1 = γ2 = γ3 = γ4 = 0) (γ1 = γ2 = γ3 = γ4) (1 = 2 = 3 = 4) F-statstc P-value F-statstc P-value F-statstc P-value E * * E E E * E E ** * E ** ** ** E ** ** ** E E E ** ** ** E ** * * E * E ** ** E ** ** ** (**) Hypothess rejected on the 5% level (*) Hypothess rejected on the 10% level 3 Note: n the lterature, a second approach s used to test the equalty of prce response parameters adjusted for exposure. Ths s done n Cornell and Shapro (1986), Bruner and Smms (1987) and Musumec and Snkey (1990) usng an OLS framework. They estmate the followng equaton AR a EXP Where AR s the abnormal return on the event day and EXP s the exposure varable for bank. If the beta coeffcent s sgnfcant the authors argue that there s evdence for ratonal prcng. However, ths framework cannot be appled n our study, snce we have too few abnormal returns per event to obtan relable results. 16

19 The results for these tests show that 5 out of the 8 events rejected under hypothess 2, are also rejected under hypothess 3. These events are marked n bold. Table 7 summarzes the outcomes for the hypotheses that had an mpact on the Swedsh banks.e. rejected under hypothess 1. Table 7: Summarzed results for the events that had an mpact on the Swedsh banks Date Event Descrpton H1 H2 H E1 Q GDP growth eases to 6,4% y/y n Lthuana. * * E6 Natonalzaton of Parex bank n Latva. ** * E7 Moody's report that Lthuanan bank credt outlook s negatve. ** ** ** E8 Largest drop n Estonan GDP y/y rate n 14 years. ** ** ** E11 Lthuana launches economc stmulus plan. ** ** ** E12 GDP collapse n Estona down 9,4% y/y. ** * * E14 The Rksbank lends to the Estonan central bank. ** ** E15 EIB lends 1,15b to lthuana to support SME's. ** ** ** (**) Hypothess rejected on the 5% level (*) Hypothess rejected on the 10% level 17

20 5 Analyss In ths part we wll analyze the results from the prevous secton. Our man goal has been to examne f the lst of selected events form the Baltc States have had an mpact on Swedsh banks stock prces. If they had an mpact, the relatonshp between prce response and exposure level was further nvestgated. Eleven out of the ffteen events selected are sgnfcant for at least one bank at the 10% level. Ths fndng supports our hypothess that events from the Baltc States ndeed affect Swedsh banks. However, the dstrbuton of sgnfcant events among the banks s qute surprsng. Only one out of ffteen events s sgnfcant for SEB and Swedbank smultaneously. Our expectaton was that SEB and Swedbank would follow a more smlar pattern, snce these are the two most exposed banks to the Baltc regon. Furthermore, t s notable that Nordea s not affected by any event, whle SHB s sgnfcantly affected by two. If Nordea s used as a proxy, we fnd no reason to beleve that any other major events affect the Swedsh bankng system durng the selected events, snce t s the only bank wthout sgnfcant abnormal returns. We now focus on the events for whch we can reject the hypothess that the dummy coeffcents are smultaneously equal to zero (H 1 ). The rejecton of ths hypothess enables us to further examne the relatonshp between market response and exposure level. As argued n Smrlock and Kaufold (1987) t s mportant to note that the nablty to reject ths hypothess does not drectly mply that the market was unaware of the exposure level for each bank. Instead, the event may have too small mpact on the banks stock prces relatve to nose n our model and thus abnormal returns cannot be revealed. 18

21 The frst event where the dummy coeffcents are not smultaneously equal to zero s the GDP estmate for 2008-Q1 n Lthuana (E1). Ths mples that some banks exhbted negatve abnormal returns sgnfcantly dfferent than zero, ndcatng that the frst quarter GDP estmate for Lthuana was worse than expected. Further, hypothess 2 can also be rejected, whch means that the abnormal returns are not equal among the Swedsh banks. However, when we test f the abnormal returns are proportonal to the banks exposure level as specfed n Secton 4, the equalty of s across banks cannot be rejected. Ths suggests that the response s proportonal to the banks exposure levels.e. support for the ratonal prcng hypothess. The natonalzaton of Parex Bank (E6) s the only event that shows evdence of sgnfcant abnormal returns for both SEB and Swedbank. The event s a sgnal of a weakenng of the fnancal system n the Baltc s and had a clear negatve mpact for these two banks. The results from testng the two frst hypotheses confrm that the event had an effect and that the abnormal returns are not equal across all banks. In addton, nvestors were able to dscrmnate among the Swedsh banks exposure levels. Moody s report concernng the expected rse n problem loans n Lthuana (E7) causes abnormal returns and these are not equal across the examned banks. The dfference between ths event and the prevous ones s that the equalty of the exposure adjusted dummy coeffcents can be rejected. The result can be nterpreted as support for the hypothess of contagon effects. One reason why the market does not dstngush between the banks can be that the event has a sgnalng effect of lower asset qualty for the banks n the regon. Ths may be assocated wth a rse n credt losses and affect the Swedsh bankng system as a whole, snce t s hghly nter-connected. 19

22 Regardng the large drops n Estona s GDP (E8 and E12), both events have a sgnfcant negatve and unequal mpact on the banks. After adjustng for exposure, the hypothess of ratonal prcng can be rejected. Thus, these events have an mpact on the banks and the prce responses are not equal even after consderng exposure. Sgns of a hard-landng n the regon seem to affect the Swedsh banks rrespectve of ther exposure and suggest that the drops n GDP were larger than expected. The stmulus plan n Lthuana (E11) wth the am to dampen the economc downturn s the only event that had a sgnfcant postve effect for one of the banks. For ths event all three hypotheses can be rejected and as stated earler t means that the government nterventon ndeed yeld abnormal returns. However, these are not equal among banks even after the exposure level s consdered.e. support for the contagon effect hypothess. The agreement between Sweden and Estona, that the Rksbank wll support the Estonan central bank ( Eest Pank ) (E14), caused negatve abnormal returns for the four banks. For ths event both H 1 and H 2 can be rejected. However, when consderng exposure n the model H 3 cannot be rejected. Ths precautonary agreement can be vewed as an act of support n order to secure stablty n the fnancal system, by provdng lqudty under a currency board arrangement. The market dd however respond negatvely and the response suggests that t vews ths news as a sgn of nstablty n the fnancal system. Snce hypothess 3 s not rejected, the market was able to dscrmnate between the banks respectve exposure levels. Thus, the results from ths nternatonal agreement are n support for the ratonal prcng hypothess. The loan from EIB to support Lthuanan small and mddle szed frms (E15) follows the same pattern as for the agreement between the Rksbank and Eest Pank.e. a sgnfcant negatve market reacton for an act of support. The dfference between ths event and the prevous s that we have ndcatons of contagon effects rather than ratonal prcng. 20

23 We can now conclude that for all events that had an mpact on the Swedsh banks, the abnormal returns were unequal. However, after regardng the banks exposure levels the results are qute dfferent. Fve events gve support for the hypothess of contagon effects whlst three events support ratonal prcng. As argued n Rme (2003), these mxed results pont out that even f the market partally separated between banks accordng to ther Baltc exposure, we cannot exclude a contagon effect n the bankng system for some of the selected events. As the events affectng the Swedsh banks, show support for both the ratonal prcng and the contagon effect hypotheses, t s mportant to note that when rejectng the thrd hypothess we have sgns of contagon effects across banks. However, t s not certan. As argued n Unal et al. (1993), such a fndng propose that other factors than exposure to the Baltc regon could be nfluental n explanng the mpact of the event. Furthermore, our exposure varables are only approxmatons to the regon as a whole and do not capture country specfc exposure. Ths means that the banks exposure levels to each country are assumed to be the same. In realty ths s not true nstead the banks are exposed dfferently between the countres. Therefore, consderng the Baltc States as one regon may provde dfferent results than f each country would have been nvestgated separately. However, f conductng the study for each Baltc State, t would be dffcult to control for contagon effects between the countres. We beleve that ths would cause more problems and s the reason why we chose to treat the Baltc States as one regon. Notably, when observng the events wth no sgnfcant abnormal return, we fnd one surprsng result, the agreement between Latva and IMF of the rescue package. Ths agreement s one of the most wdely known events durng our perod of nterest and the absence of reacton for ths event can possbly be nterpreted as the market had already antcpated the rescue package and therefore no response can be observed on the event day. Smlar arguments regardng key events wthout market reactons are made by Rme (2003) for events durng the Russan crss. 21

24 6 Robustness of our fndngs The man purpose of an event study s to test for the exstence of an nformaton effect.e. mpact of an event on frm value. The relatvely smple framework has made the use of event studes very popular. However, the event methodology s obvously not perfect and several problems can arse. Wth ths sad, our results based on the method and models used rase concerns that are worth dscussng. Frst of all, the choce of events s of great mportance when conductng event studes. In the lterature regardng fnancal crses, the selecton of events are often based on surveys explanng the crss development and pontng out the man events. One reason why our study yelds mxed results, could be that t s based on an ongong crss and the selected events may not be perceved by the market as the most mportant ones. When the event(s) of nterest has been selected t s mportant to determne when t took place. The tmng of the event s of great mportance and n many cases t may seem obvous, however t s not. The ssue s not when an event took place, nstead when nvestors n the market could have antcpated t. In our study we have an example of ths ssue, n form of the agreement between the IMF and Latva. Usually ths problem s addressed by expandng the event wndow. However, even f there s a cost assocated wth a longer event wndow, n some cases t s worth bearng nstead of facng the rsk of mssng the effects of an event (MacKnlay 1997). Wth our specfcaton, usng a multvarate regresson wth some events close n tme, we beleve t s reasonable to have a short event wndow. We tred to mnmze the problem by controllng the tmng of the events.e. f an announcement occurred after the closure of the stock exchange we used the next day as the event day. Furthermore, another possble bas that can arse when conductng event studes s the non-tradng or non-synchronous tradng effect as dscussed n MacKnlay (1997). An example of ths s when usng closng prces to compute daly returns. In ths case we mplctly and ncorrectly assume that the returns are equally spaced at 24-hour ntervals. Moreover, we dsregard the possblty of prce adjustments durng the day when the event occurred and ths problem wll be more promnent f a larger event wndow s used. 22

25 In addton, as argued n Klc et al. (2000), results from event studes are senstve to the choce of estmaton perod. To control for ths n our event parameter models, we reestmate equaton system (1) and (2) wth shorter estmaton perod. Instead of usng data for 2007 we now estmate from the frst of January The results are reasonably smlar to the frst estmatons. The dfference s that the events whch had an mpact on the banks at the 10% level are no longer sgnfcant. Ths means that our prevous fndngs are relatvely robust rrespectve of the choce of estmaton perod. 4 The ssues presented above are meant to hghlght possble dffcultes assocated wth the event study methodology. The dscusson above should not be nterpreted as arguments for not usng the event study methodology. It s a smple and functonal desgn whch works under less than perfect condtons. However, to get relable results the framework used must be thoughtfully specfed. 4 The results from the re-estmaton of the models wth shorter estmaton perod are avalable from the authors on request. 23

26 7 Conclusons In ths paper, we have studed how events from the fnancal crss n the Baltc countres affect Swedsh banks valuaton and f the reactons were n proporton to the banks Baltc exposure. In a frst step, usng stock prce event analyss n a multvarate regresson framework, we tred to determne whch events that had an mpact for the banks. We can conclude that the majorty of the events had an mpact on the banks but the dstrbuton among them where surprsng. Only one event out of ffteen was smultaneously sgnfcant for the two most exposed banks, SEB and Swedbank. Furthermore, approxmately half of the selected events had a sgnfcant mpact on the banks as a group. Our nterpretaton s that the events wth no sgnfcant mpact were not a surprse for nvestors. In a second step, the hypothess of ratonal prcng and contagon effects were nvestgated by ncludng a frm specfc exposure varable n the model. Here we cannot draw any general conclusons, snce on ths ssue our results are mxed. Ths ponts out that the market only partally separated between the banks accordng to ther exposure. 24

27 8 References Ball, R. and Brown, P. (1968), An Emprcal Evaluaton of Accountng Income Numbers. Journal of Accountng Research, Bank of Estona, Bank of Latva, Bank of Lthuana, Bnder, J. (1985), On the Use of the Multvarate Regresson Model n Event Studes. Journal of Accountng Research, 23, Bloomberg s news archves. Bowman, R.G. (1983), Understandng and Conductng Event Studes. Journal of Busness Fnance & Accountng, 10, Brown, S. and Warner, J. (1985), Usng Daly Stock Returns: The Case of Event Studes. Journal of Fnancal Economcs, 14, Bruner, R.F. and Smms, J.M. (1987), The Internatonal Debt Crss and Bank Securty Returns n Journal of Money Cred and Bankng, 19(1), Campbell, J.Y., Lo, A.W., and MacKnlay, A.C. (1997), The Econometrcs of Fnancal Markets, Prnceton Unversty Press. Cornell, Bradford and Shapro, A.C. (1986), The Reacton of Bank Stock Prces to the Internatonal Debt Crss. Journal of Bankng and Fnance, 10,

28 Crouzlle, C., Lepett, L. and Taraz, A. (2006), Reacton of European Bank Stock Prces to Events of the Asan and Russan Fnancal Crses. Revue d'econome Poltque, Vol 116, European Investment Bank (EIB), Fama, E., Fsher, L., Jensen, M., and Roll, R. (1969), The adjustment of Stock Prces to New Informaton. Internatonal Economc Revew, Forsberg, B. (2009), Operaton Baltkum. Affärsvärlden, February 4. Handelsbanken (SHB), Annual Reports (2007 and 2008), Avalable [Onlne]: ankenseenglsh [ ] Henderson, G.V. (1990), Problems and Solutons n Conductng Event Studes. The Journal of Rsk and Insurance, Vol. 57, Internatonal Monetary Fund (IMF), Karafath, I., Mynatt, R. and Smth, K.L. (1991), The Braslan Default Announcement and the Contagon Effect Hypothess. Journal of Bankng and Fnance, 15, Khotar, S.P. and Warner, J.B. (2006), Econometrcs of Event Studes. Handbook of Corporate Fnance: Emprcal Corporate Fnance, Vol A, Ch 1. Klc, O., Hassan, M.K. and Tufte, D. (2000), Market Effecency, the Mexcan Peso Crss, and the US Bank Stock Returns An Applcaton of the Event Parameter Method. Global Fnance Journal, 11,

29 MacKnlay, A.C. (1997), Event Studes n Economcs and Fnance. Journal of Economc Lterature, Vol. 35, Mathur, I. and Sundaram, S. (1997), Reacton of Bank Stock Prces to the Multple Events of the Brazlan Debt Crss. Appled Fnancal Economcs, 7, Musumec, J.J. and Snkey J.F. (1990), The Internatonal Debt Crss, Investor Contagon, and Bank Securty Returns n 1987: The Brazlan Experence. Journal of Money, Credt, and Bankng, 22, Nordea, Annual Reports (2007 and 2008), Avalable [Onlne]: ml [ ] Norder, C. IR department Handelsbanken, chno07@handelsbanken.se, (2009), Exponerng Baltkum, [E-mal] Message to Dembrower, D. (21435@student.hhs.se). Sent Wednesday 22 Aprl 2009, 12:08. Avalable at: um.eml?cmd=open [ ] Reuters news archves. Rksbank, Rme, B. (2003), The Reacton of Swss Banks Stock Prces to the Russan Crss. Schwez Zetschrft für Volkswrtscaft und Statstk, Vol. 139 (1), Skandnavska Ensklda Banken (SEB), Annual Reports (2007 and 2008), Avalable [Onlne]: [ ] 27

30 Smrlock, M. and Kaufold, H. (1987), Bank Foregn Lendng, Mandatory Dsclosure Rules, and Reacton of Bank Stock Prces to the Mexcan Debt Crss. Journal of Busness, 60, Swedbank, Annual Reports (2007 and 2008), Avalable [Onlne]: [ ] Thomson Datastream Internatonal Unal, H., Demrgüc-Kunt, A., Leung, K.W. (1993), The Brady Plan, 1989 Mexcan Debt-Reducton Agreement, and Bank Stock Returns n Unted States and Japan. Journal of Money, Credt, and Bankng, Vol. 25, Zellner, A. (1962), An Effcent Method of Estmatng Seemngly Unrelated Regressons and Tests for Aggregaton Bas. Journal of the Amercan Statstcal Assocaton, Vol 57, Åslund, A. (2009), Baltc Protests and Fnancal Meltdowns. The Globalst, February 6. Avalable [onlne]: [ ] 28