Measuring the impact of the Japanese Tōhoku earthquake on regional stock prices : using the BHAR and CAR methods to check for abnormal performance

(1)

Measuring the impact of the Japanese Tōhoku

Earthquake on

Regional Stock Prices

Using the BHAR and CAR methods to check for abnormal performance

UNIVERSIRY OF AMSTERDAM

AMSTERDAM BUSINESS SCHOOL

Bsc Finance & Organisation

Bachelor Thesis Finance

Author:

S.A.M. van der Heijden

Student number:

10968253

Thesis supervisor: Dr. J.J.G. Lemmen

Finish date:

26/06/2018

(2)

ii

PREFACE AND ACKNOWLEDGEMENTS

I would like to give many thanks to my supervisor Dr. J.J.G. Lemmen, who helped me a lot with

the design and quality of my research. Also, I would like to thank both my sisters and my parents

for helping me finalizing my thesis

Statement of originality

This document is written by student Sam van der Heijden, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

ABSTRACT

This paper examines the long-term impact of the 2011 Tōhoku earthquake on the returns of listed

companies with their main seat either the Fukushima, Iwate or Miyagi prefecture. These

companies form the regional portfolio. We find that there is significant abnormal performance

for the companies of the regional portfolio. These abnormal returns are negative for the first six

months, but become positive for the 4 consecutive years. These results were found using the

BHAR and CAR method. The normal return for these methods were found with the Fama and

French 3 Factor model.

Keywords:

(4)

iv

LIST OF TABLES

Table 1

Meta table of disaster studies

4 Table 2

Estimation period and event window

6 Table 3

Summary statistics

13 Table 4

BHAR and CAR results with and without outliers 16

Table 5

BHAR method results

18 Table 6

CAR method results

19 Table 7

Company abnormal returns per post-event window 24

(7)

LIST OF FIGURES

Figure 1

Small cap index & regional portfolio return

7 Figure 2

Returns per trading day

17 Figure 3

Outliers BHAR method

35

(8)

1

CHAPTER 1 Introduction

On the 11th_{of March 2011, Japan was hit by the most devastating earthquake in recorded Japanese} history, the Tōhoku earthquake. Even though Japan is relatively prone to earthquakes, no one expected an earthquake of this magnitude to hit. 16.000 people died as a results of the earthquake and several thousands of others were never found (Kazama & Noda). The damages of 210-309 billion US dollar were the largest ever caused by a single earthquake (em-dat). The destruction was mainly caused by the tsunami that hit the coast moments after the earthquake. The shores of the Japanese prefectures Fukushima, Iwate and Miyagi (further on referred to as the FIM prefectures) were hit most severely with more than 99% of the deaths occurring in these three prefectures. The situation in the prefecture of Fukushima worsened when the cooling system of the Fukushima nuclear reactor was not able to withstand the impact of the tsunami and had a nuclear meltdown. All these disasters occurred within hours and scarred the Japanese people for years.

Every day, somewhere in the world a natural disasters occurs (em-dat). Sometimes they have a strong destabilizing effect on the financial market of a country. Research has shown that global warming is increasing the frequency of different types of natural disasters (Francis & Vavrus, 2012). Because the frequency of natural disasters is increasing, it is becoming more important to have a good understanding of the effects they entail.

This thesis will use the Japanese Tōhoku earthquake as a case study. There are two reasons why this particular event is chosen. The first is because it is the most devastating natural disaster ever recorded (Kazama & Noda). Even though earthquakes might not increase because of global warming, they have a similar devastating and unpredictable character as other types of natural disasters, and it is therefore beneficial to gain more knowledge about the impact they entail. Another reason is Japan’s largest stock market, the Tokyo stock exchange. This stock market has more than 6.5 percent of the world's market capitalization. This is valuable for the data collection process. In addition, since Japan experiences several smaller earthquakes every year, there exists a lot of data about earthquakes as well. The thesis is divided in 6 chapters. This introduction will be followed by a summary of the existing literature and the debate around it. With regard to the discussed literature the hypotheses is stated. In chapter 3 the methodology for the research will be explained, together with the models to test the hypotheses. The ways of getting the data for these models are discussed in chapter 4. With the obtained data the models will be run. In chapter 5 the results will be discussed and interpreted. In the last chapter the conclusion is given.

(9)

CHAPTER 2 Literature review

In the following chapter the literature review will be given together with the resulting hypothesis.

2.1 Literature

Natural disasters occur often and can have great consequences for countries and their markets. Therefore, a lot has been written about them in the past decades. However, there is no clear consensus about the impact of a natural disaster. The debate about the impact of natural disasters is held on several areas such as economic growth, financial market consequences or insurance markets. It should be noted that the effect of a natural disaster on an economy is not the same for every country. Horwich (2000) argues that wealth is the critical factor that influences the response to a natural disaster, since wealthier countries are better equipped to manage the aftermath of a natural disaster. Tol and Leek (1999) found this correlation as well in the form of lower death tolls for high-income nations.

Firstly, the disagreement on the effect of a natural disaster on economic growth will be discussed. Nakamura et al. (2013) stated that consumption significantly decreases after natural disasters. Schumacher and Strobl (2011) were in line with this and concluded that economic growth halts for a while after a natural disaster but returns to normal levels when enough time has passed.

In contrast, Skidmore and Toya (2002) found that extreme weather events do not have a negative, but a neutral or positive effect on economic growth. They partly explained this with the fact that the destruction of buildings and property are not subtracted from the Gross Domestic Product, while the rebuilding costs are added. This results in an increase of GDP in the aftermath of a natural disaster. Skidmore and Toya also give the argumentation that the decrease in investments due to natural disaster hazards does not necessarily results in lower investments. The other part is that companies whose assets have been destroyed, get an opportunity to update capital stock, and are thus encouraged to adopt new technologies. Bloom (2013) agreed with them on this topic.

The second argument about economic growth has to do with the insurance market. Since Japan is prone to having earthquakes, the Japanese government forced insurance companies in 1966 to offer government-backed earthquake insurance. This insurance is mainly focused on residential properties to improve the livelihood of Japanese citizens after an earthquake occurs. Due to this insurance, consumption will decrease less after a natural disaster since consumers have to pay less rebuilding costs themselves. Which leaves more money for consumption. Several researchers focus on the effect of natural disasters on the insurance and found that this effect is significant (von Ungern-Sternberg (2010) and Yamori & Kobayashi (2002)). Yamori & Kobayashi even found that this effect is positive due to higher demand for earthquake insurance after a large-scale earthquake. Since the increase in demand outweighs the loss of capital to insured legal persons.

Lastly the impact of natural disasters on the financial markets. Worthington and Valadkhani (2004) observed significant abnormal returns after natural disasters on the Canadian stock market. However, these returns were both positive and negative, canceling each other out. Also, it was only

(10)

3 significant for a few days after the disaster. Wang and Kutan (2013) did not find significant changes in the Japanese and American stock markets after a natural disaster, nor did they find any cross-country influence on stock prices. Luo (2012) found the same insignificant result. Valizadeh et al. (2017) wrote about the Tōhoku earthquake as well. They found that all the Japanese market segments, except for oil and gas, had significant negative abnormal returns. This was done for a short and long term period. The also tested abnormal performance for other countries. Here they found significant results as well for both the short term and long term period.

The articles about the impact on the financial markets indicate that there is still uncertainty on what the impact of natural disasters on financial markets is. However, the discussed literature all looked at national stock returns to calculate pre- and post-event returns. Strobl (2011) found that while the impact of natural disasters on a national level is not significant, the impact on a regional level is. Bourdeau-Brien & Kryzanowski (2015) did similar research where they tested for volatility changes in U.S. stock markets. They found that the effects of natural disasters on regional stock price volatility were significant, and they were the largest during the two or three months after the peak of the news coverage. The main reason national and regional differences occur is because national stock indexes are naturally well hedged, and will therefore diversify most of the regional losses. For this reason this thesis will focus on the regional abnormal performance instead of national or cross-country.

From the research listed above, all relevant articles about event studies have been summarized in the table 1. With the help of this table the methodology will be set up.

(11)

Table 1

Meta table of disaster studies Author(s)

(publicaton year) Time period Country, Region Method Estimation period Results Shelor et al. (1992) 1980-1999 United

States, California

Event study

(market model) (-200,-1) CAR (0,14): +64,5% Yamori & Kobayashi

(2002) 1995 Japan, Hanshin Event study (portfolio approach)

(-160,-11) AR (-6.1): -2,7% Skidmore and Toya

(2002) 1960-1990 Cross-country Empirical analysis - - Worthington and

Valadkhani (2004) 1982-2002 Australia Event study (ARIMA) - AR (p-value = 0.000): -0,0043 von Ungern-Sternberg

(2010) - - Model of risk selection - Graphical proof Schumacher and

Strobl (2011) 1980-2004 Cross-country Utility function - Coëfficiënt (-3.3): 7.043 Strobl (2011) 1975-2005 United

States, Coastal areas

Hurricane

destruction proxy - Coëfficiënt (-2.52): -0.0451 Luo (2012) 2011 Japan Event study

(market model) (-60,-10) ACAR(-0.82401): -0.03332 Hood et al. (2013) 2008-2012 Japan Iterated feasible

generalized OLS method

(-606,-1) Coëfficiënt (pvalue = 0.009): -75.24

Bourdeau-Brien &

Kryzanowski (2015) 1990-2004 United States Event (ARMAX study -EGARCH model)

(-500,-1) CAV(p-vaue > 0.1): 0.89 Valizadeh et al. (2017) 2010-2011 Japan Event study

(CAR short term, BHAR long term)

(-250,104) CAR(p-value = 0.000): -0.139 BHAR(p-value = 0.000): -7.47

(12)

5 2.2 Hypotheses

The discussed literature will be taken into account for the hypothesis. This thesis will focus on the long term effect the Tōhoku earthquake had on the FIM prefectures. This will be done with the following two hypotheses.

Hypothesis 1: The Tōhoka earthquake had a significant impact on the regional economy of

Iwate, Miyagi and Fukushima.

Several articles state that while natural disasters might not have a significant impact on national stock indexes, they might on regional portfolios (Strobl (2011) and Bourdeau-Brien & Kryzanowski (2015)). Therefore, I believe that the results of the models will show significant abnormal performance of the regional companies after the Tōhoku earthquake.

Hypothesis 2: The Tōhoku earthquake has a long term positive effect on the stock returns of

listed companies with their main seat in the FIM prefectures.

As stated by Skidmore & Toya (2002) and Wang & Kutan (2013), natural disasters can have a neutral to positive effect on companies on the long run since they have an opportunity to adopt new technologies. If we take into account that according Bourdeau-Brien & Kryzanowski (2015) regional effects are stronger than national effects, the positive effect is believed to be even more significant. Therefore this thesis believes the Tōhoku earthquake will have a long-term positive effect on the companies from the FIM prefectures.

(13)

CHAPTER 3 Methodology

In this chapter, the methodology of the thesis will be presented. First the event horizon will be discussed and reasoned, then the regional portfolio will be brought forward. Lastly the calculation for the model will be given and explained.

3.1 Time frame

This study will focus on the long term effects of the Tōhoku earthquake on the Japanese companies from the FIM prefectures. This will be substantiated by two articles in particular. The first is of Downton and Pielke (2005). They found that it is hard to predict the true costs of damages in the first few days after a natural disaster. This is mainly due to the chaos of the disaster. The focus of post-disaster response is saving as many lives as possible. Only after search and rescue missions are concluded, parties are allowed to account for damages. The estimation of the total damages of the disaster are therefore subject to large estimation errors. Financial institutions can’t value the share prices of their assets correctly. This results in large positive or negative return.

The second article is about the decrease in consumption, that follows a natural disaster, will most likely unfold over several years (Constantinides, 2008). This is due to the remaining availability of products on the short-term due to warehouses stocked with finished products. However, there is a long-term decrease in available products due to the destruction of production facilities. This results in higher prices and lower consumption.

With regard to the these two articles, we can choose our estimation period and post-event window. Table 2 contains the chosen estimation and post event periods.

Table 2

Estimation period and event window

Horizon Estimation period (t,t) Post-event period Post-event period (t,t) 6 Months (-262,-2) 14/03/2011 – 13/09/2011 (0,131)

1 Year (-262,-2) 14/03/2011 – 14/03/2012 (0,263) 3 Years (-262,-2) 14/03/2011 – 14/03/2014 (0,785) 5 Years (-262,-2) 14/03/2011 – 14/03/2016 (0,1306)

The Tōhoku earthquake happened on Friday the 11th_{of March 2011. With an event study the event day} (t = 0) will be the first trading day after the event. The earthquake started 14 minutes before the end of the trading day, so investors did not have the time to deliberate and react until after the end of the trading day. Therefore Monday the 14th_{of March 2011 will be set as the event day. The horizons are chosen to} be 6 months, 1 year, 3 years and 5 years. Shorter time spans than 6 months have already been examined in the existing literature and will not be taken into account. Longer time spans than 5 years will most likely have too much noise due to external causes and will therefore not give results that can be

(14)

7 interpreted with any certainty.Since this thesis will focus on a long term effect, the estimation period will be set over the year before the event day (-262,-2). No variations will be done with the estimation period to exclude look-ahead bias. Since we use daily data, t will only count for trading days. So t = 1 is the first trading day after the event, t = 2 the second and so on.

3.2 Regional portfolio

Since the effect of the Tōhoku earthquake is more likely to be found at a regional than national level we have to conclude which areas were hit the most severely. According to Kazama and Noda (2012) the areas with the largest death count were Miyagi (9.512), Iwate (4.670) and Fukushima (1.605). Their numbers are significantly higher than the other prefectures, which had 66 casualties combined. Therefore, I have indicated the FIM prefectures as my regional economy, where the economy is defined as the combined market capitalization of the listed companies of the FIM prefectures. The companies that will consist the regional portfolio are all listed companies which have had their main seat in one of the FIM prefectures for the period of t = -262 to t = 1260. More about the regional data will be discussed in chapter 4.1.

Figure 1

Small cap index & regional portfolio return1

3.3 Model

The primary objective of this paper is to test if the Tōhoku earthquake had a significant long-term impact on the Japanese financial markets. While the discussed papers all focus on the effect of natural disasters on a national level, this paper will focus on the effect at a regional level. To test my hypotheses, I will look for significant abnormal returns after the event day with a long horizon event study. Since the event

1_{This figure shows the total return of both the S&P Japan small cap index and the average regional portfolio}

return. It is taken over a period consisting the estimation period and the total longest post-event period. 0,00% 50,00% 100,00% 150,00% 200,00% 250,00% 300,00% 350,00% 11-03-10 11-03-11 11-03-12 11-03-13 11-03-14 11-03-15 11-03-16 S&P Japan small cap index Regional portfolio

(15)

day is the same for every company, the abnormal returns will be correlated due to the clustering of events in time. In addition, the regional portfolio will not be rebalanced during the post-event period. For these two reasons the Buy-and-Hold Abnormal Return method (BHAR method) will be used in this thesis. A downside of the BHAR method is that it takes the abnormal return for t, and not for an average period. This can give a highly skewed distribution. However, according to the Central Limit Theorem, a n > 30 will be sufficient to counteract this problem. Another possibility for a long term event study is the Cumulative Abnormal Returns method (CAR method). This model does not take a single t from the entire post-event window, but takes the average of the period. Therefore the CAR method will serve to check for differences.

Another argument for using the BHAR and CAR methods is described by Valizadeh et al. (2017). With their analysis on the impact of the 2011 Tōhoku earthquake on 19 different market segments for both a short and long-run period, they used the BHAR and CAR methods as well.

3.3.1 Fama and French 3 factor model

To calculate abnormal performance with the BHAR and CAR methods, a normal return is necessary. A normal return is the return a company would have without the occurrence of the event. Since it is impossible to know the exacts performance without the event it is predicted, often with the help of regressions. The normal return in this thesis will be calculated using the Fama and French 3 Factor model (1993). This Ordinary Least Squares regression model uses the market return, a value factor and a growth factor to predict the post-event normal return. The value factor is slang for the High-Minus-Low factor. This factor measures the difference in return between a portfolio with firms with either a high book-to-value ratio or a low book-to-value ratio. The growth factor stands for the Small-Minus-Big factor, which compares a portfolio with either “large” or “small” firms. These factors are created by Fama and French to better the more simple CAPM model. With these factors the normal return can be calculated by regressing the 3 factors on the excess portfolio return for the entire estimation period (-262,-2). Another option is the market method. This method is used several times in the discussed literature. However, this model is better for short-term predictions about normal return than long-term predictions. Therefore, this thesis will stick with the Fama and French model, which is focuses on long-term predictions. The regression input for the Fama and French 3 Factor model looks as follows

(1) 𝑅

_&'

− 𝑅

_)'

= 𝛼

_&

+ 𝛽

_&

.𝑅

_/0'

− 𝑅

_)'

1 + 𝑠

_&

𝑆𝑀𝐵

_'

+ ℎ

_&

𝐻𝑀𝐿

_'

+ 𝜀

_&

𝑅&' = the return of company i at time t.

𝑅₎ = the risk free rate at time t. This will be a one-month risk free government bond. 𝛼_& = a constant term for company i.

(16)

9 𝑆𝑀𝐵_' = the Small-Minus-Big factor at time t

𝐻𝑀𝐿_' = the High-Minus-Low factor at time t 𝜀& = the error term of company i with mean zero

The above regression will be ran for every company i and every time t in the estimation period (-262, -2). The normal return post-event can then be calculated for every t using the obtained regressed coefficients. The formula for doing this is as follows.

(2) 𝑁𝑅

_&'

= 𝑅

)'

+ 𝛽<

&

.𝑅

/0'

− 𝑅

)'

1 + 𝑠̂

&

𝑆𝑀𝐵

'

+ ℎ>

&

𝐻𝑀𝐿

'

+ 𝜀

&

𝑁𝑅_&' = the normal return of company i at time t.

𝛽<_& = the regressed coefficient for predicting the excess market return 𝑠̂_& = the regressed coefficient of company i for predicting the SMB factor ℎ>_& = the regressed coefficient of company i for predicting the HML factor

This formula will be ran for every company i, every time t until the latest post-event window of table 2. When all the normal returns are obtained, they will be used for the BHAR method to calculate the abnormal performance.

3.3.2 Buy-and-Hold Abnormal Returns method

The BHAR method is widely used by investment banks and examines the long-term abnormal returns of stock prices after an event. The formula for the BHAR is as follows

(3)𝐵𝐻𝐴𝑅_& = A .1 + 𝑅&,'1 C

'DE − A (1 + 𝑁𝑅&,') C

'DE

𝐵𝐻𝐴𝑅_& = the total abnormal return of company i for the post-event window (𝑡_G, 𝑡_E,H,I,J) H = the end of the post-event window

The BHAR method calculates the total abnormal return for every company i from the event day until the end of the post-event window and subtracts the total normal return for the same period. Since there are four post-event windows, the BHAR calculation will be done once for every window and will therefore give four separate abnormal returns.

3.3.3 Cumulative Abnormal Returns method

(17)

(3)𝐶𝐴𝑅_& = L 𝐴𝑅_&,' C &DE = L(𝑅_&,'− 𝑁𝑅_&,') C &DE

𝐶𝐴𝑅_& = the cumulative abnormal returns for company i

3.3.4 Testing for significance

After the abnormal performance is calculated it has to be tested for significance. This will be done in the same for both the BHAR and the CAR2_{method. First the average BHAR value of the regional} portfolio must be calculated

(4)𝐴𝐵𝐻𝐴𝑅 = 1

𝑁L 𝐵𝐻𝐴𝑅&

N &DE

𝐴𝐵𝐻𝐴𝑅 = the average BHAR value of the regional portfolio

After the ABHAR has been calculated it will be used to calculate the standard deviation of the portfolio. This is done with the following formula.

(5)𝑠_PQ = R 1

𝑁 − 1L(𝐵𝐻𝐴𝑅& − 𝐴𝐵𝐻𝐴𝑅)H

N &DE

𝑠_PQ = the standard deviation of the portfolio

With both the ABHAR and the standard deviation we can test the null hypothesis of no abnormal performance with the following t-test.

(6) 𝑡 = √𝑁𝐴𝐵𝐻𝐴𝑅 𝑠_PQ

A two sided t-test will be done with a significance level of five percent, and 35 degrees of freedom. Since the table only gives the t-values for 30 or 40 degrees of freedom, we take 30 to be on the conservative side. From the t-statistics table we find the corresponding values of 𝑡_G.GHV,IG is ±2.042. This means we reject the null hypothesis of no abnormal returns if |𝑡| ≥ 𝑡G.GHV,IG.

2_{The following calculation are focused on the BHAR method. To convert this to the CAR method. The ABHAR}

(18)

11 Since the second hypothesis states a positive long term return, a one-sided t-test will be run as well. For this test the critical value will be 𝑡G.GV,IG, which correspond with a t-value of 1.697. Thus to

(19)

CHAPTER 4 Data

In this chapter the data collection process will explained and substantiated.

4.1 Regional portfolio

Firstly the regional portfolio will be discussed, beginning with daily versus monthly stock prices. As time progresses more data about stock prices becomes available. Here, the question arises if daily or monthly stock returns should be taken. As concluded by Morse in 1984, daily stock prices almost always give more accurate predictions than monthly ones. The only time monthly returns give better predictions is when news coverage is very low, this was not the case for the Tōhoku earthquake. Consequently, the prices that will be used to calculate the individual company return of the regional portfolio are the daily closing prices from the Tokyo Stock Exchange.

Secondly, the criteria for companies from the regional portfolio will be discussed. There were several criteria the companies of this portfolio had to fulfil. The first is that they had their main seat in one of the three FIM prefectures. This data is retrieved from the Japanese Exchange Group (http://www.jpx.co.jp/english/). However, they only post the current main seat of a company, and not their previous main seats. Therefore, all companies had to be checked for a shift in their main seat. No companies were found that changed their main seat to one the FIM prefectures during the post-event period. A possible limitation of this thesis is that no data was found on companies leaving the FIM prefectures during the post-event period. Therefore we do not know whether it happened or not. The second criterion is that the companies of the regional portfolio are all listed at the Tokyo stock Exchange, and have their stock prices available for both the estimation period and some part of the post-event window.

In table 3 the summary statistics of the regional portfolio companies are reported. From these companies the first 5 are from Iwate (Yakuodo - KitaNippon bank), the following 10 are from Fukushima (Joban Kaihatsu – Daito bank) and the last 21 are from Miyagi (Yurtec – Satoh).

(20)

13 Table 3

Descriptive Statistics

Variable Obs Mean Std.Dev. Min Max

YAKUODO 1567 404.597 298.914 100 1263.32 NCXX 1567 515.947 254.251 185.6 1435 BANKOFIWATE 1567 4272.131 782.106 2500 5860 TOHOKUBANK 1567 1422.534 164.003 960 1760 KITANIPPON~K 1567 2452.283 483.854 1612 3665 JOBANKAIHA~U 1567 2536.273 1301.663 960 6010 JOBANKOSAN 1567 1330.938 323.898 580 2130 HONEYSHOLD~S 1567 1085.708 167.622 685 1670 ASAKARIKEN 1567 853.241 710.453 450 7440 KOURAKUEN 1567 1299.325 153.307 992 1729 XEBIOHOLDI~S 1567 1917.08 246.03 1345 2666 TOHOBANK 1567 314.327 93.96 162 572 FUKUSHIMAI~S 1567 1200.163 746.663 325.5 2867 FUKUSHIMAB~K 1567 707.658 206.197 360 1050 FUKUSHIMAP~G 1567 330.588 47.812 240 450 DAITOBANK 1567 1006.311 454.138 440 2260 YURTEC 1567 493.591 253.22 236 1282 KARULA 1567 371.699 53.642 186 470 TOHOACETYL~E 1567 624.334 223.025 295 1255 TOSNET 1567 537.954 302.287 197 1639 KURAMOTO 1567 225.819 115.297 104 1039 TOYOKNIFE 1567 882.945 320.025 450 2120 YAMADAI 1567 1498.504 696.061 410 3875 KOHSOKU 1567 793.866 127.89 471 1070 FUJICORPOR~D 1567 527.549 148.837 292 823 KAMEI 1567 742.331 263.711 236 1408 BANK 1567 2373.985 653.022 1450 4130 FIDEAHOLDI~S 1567 206.615 33.218 128 295 SENKONLOGI~S 1567 648.702 59.462 530 791 TOHOKUELEC~R 1567 1254.848 405.164 468 1989 TOHOKUCHEM~L 1567 2608.807 576.481 1800 6850 TOHOKUSTEEL 1567 1060.943 213.439 698 1630 MAXVALUTOH~U 1567 855.877 233.763 467 1320 UEMATSUSHO~I 1567 250.188 53.436 136 342 YAMAYA 1567 1377.895 557.439 516.36 2901 SATOH 1567 918.412 101.565 770 1232

4.2 Survivorship bias

The survivorship bias arises when the poorly performing companies are removed from the database. In the case of this thesis, survivorshipbias could arise due to bankruptcy, merger or acquisition. Therefore, we have to include all listed companies of the FIM prefectures that went bankrupt after the Tōhoku earthquake up until t = 1260. However, after consulting the Japanese Exchange Group it was concluded that no listed companies in the FIM prefectures have filed for bankruptcy since 2011. The most likely possibility for this is that all low-performing companies that were on the edge of bankruptcy were bought out, therefore explaining the absence of bankruptcies. However, companies that were bought out cannot be included in the dataset for two reasons. First, the companies were either bought out by other companies of the FIM prefectures and are therefore already accounted for. Second, they would have been bought out by listed companies outside the FIM prefectures and still be in operation. In this case the bought-out companies would have a negative effect on the regression while in reality they are still

(21)

doing business in the FIM prefectures. Therefore we will assume that there won’t be a survivorshipbias in our output.

4.3 Fama and French factors

The factors for the Fama and French 3 factor model are already calculated for the Japanese market. These factors are accessible to the public on their website (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html#International). From their database the Fama/French 3 factors [daily] was chosen. This file contains the daily market return, SMB factor, HML factor and the risk free-rate. All of them were taken from January 2010 to April 2018. From this list the data was collected for the estimation period and the post-event window.

4.4 Market return

However, the market return will not be taken from the Fama/French database. Since t

he market return

is used

to account for national price changes in the Japanese stock market, the chosen market return has to be as in the same market segment as the average of the regional portfolio companies. To choose a proper index for the market return, market capitalization has to be taken into account. Smaller companies are more volatile to general market movements. Therefore a choice has to be made between a small cap index, mid cap index or large cap index. Where small cap is defined as market capitalization below 2 billion USD and mid cap between 2 and 10 billion USD. From the companies of the regional portfolio, 35 were small cap and 1 was mid cap. This means a small cap benchmark will probably give the most significant coefficients for predicting the market return in the post-estimation window. For certainty, the regression will be ran twice. Once with a small cap index and once with a general cap index. The small cap index will be the S&P Japan Small Cap Index, this index lists 250 companies from the Japanese small cap segment. The large cap index will be the Nikkei 225 with 225 of the top Japanese blue-chip companies.

(22)

15

CHAPTER 5 Results

In the following chapter the results from the regressions will be summarized and discussed. All the data from the Fama and French regressions are presented in the appendix B tables 1-36. The results from the BHAR method will be given in the tables in this chapter. Three parts will be presented in this chapter. Firstly the results are given, then they are discussed, and lastly the limitations of the thesis will be discussed.

5.1 Small cap vs Large cap market return

To be sure the Fama and French 3 factor model has been regressed with both a small cap index and a large cap index. From looking at the p-values of the regressed market return coefficients, the small cap index gave better predictions 12 cases, and worse predictions in 3 cases. Therefore the small cap index will be used as the market return.

5.2 Outliers

During the calculations of both the BHAR and the CAR method, tests were done to find outliers in the dataset. For both methods the time periods were put in a boxplot graph (appendix C). The outliers can be observed there. The companies who account for these outliers most often are Koban Kaihatsu, Kamei and Fukushima Electrical pwr. With this the question arises whether several outliers must me dropped from the database or not. The first thing to take into account is the reason why the outliers exist. I have not found any miscalculations and believe the data source is trustworthy. Therefore, I believe the outliers are measured correctly and there is no legitimate reason to delete them from the dataset for measurement flaws. Secondly, they affect both this thesis’ assumptions and results. Since the regressions are allowed to be run with these variables, there is no ground to cut them for this as well. Lastly, it should be tested if the significance of the test is solely due to the outliers. If this is the case it should be reported and taken into consideration when interpreting the results. After checking both methods, the 1 year periods are the only periods which have outliers and are significant. Therefore, the CAR and the BHAR will be run again for those period, but without the outliers. The outliers for the corresponding periods are NCXX, Joban Kaihatsu. Yamadai, Kamei and Yamaya. After deleting them from the dataset and doing the calculation again the results are put in table 4.

(23)

Table 4

BHAR and CAR results with and without outliers3

Model AR with outliers t-statistic with outliers AR without outliers t-statistic without outliers BHAR 0.189** 2.314 0.016 0.511 CAR 0.297*** 4.426 0.151*** 3.580

From this table we can observe that the abnormal returns of the BHAR method have lost their significance after deleting the outliers from the dataset (]𝑡_{^C_` E aPbc Q&'def' ef'g&Pch}] < 𝑡G.GHV,IG).

Therefore, the fact that the 1 year post-event window is significant partly due to the outliers will be taken into account when interpreting the results. In the following tables the results for the 1 year event window will be shown without the outliers. This is done from a conservative point of view.

5.3 Serial correlation

According to Fama (1998) serial correlation can be very impactful for long term studies. Serial correlation occurs when the error terms of a regression correlate with each other. If this is true the t-statistics will give inflated values. To test if there is significant positive or negative serial correlation a Durbin Watson test is performed for every company i. The value for not rejecting the null hypothesis of no negative or positive autocorrelation is dU < D < (4 – dL). The values dL and dU for (4, 260) are 1,633 and 1,715 respectively. After the Durbin-Watson D-statistic was calculated for every company with STATA, the results were that no company of the regional portfolio had a significant D-statistic. Therefore the null hypothesis of no serial correlation will not be rejected for any of the companies from the regional portfolio. With this finding we can conclude that serial correlation will not affect our t-statistics.

5.4 Analysis and discussion

In the following subchapters the analysis and discussion of the Fama and French 3 Factor model, BHAR method and CAR method are presented, analyzed and discussed.

5.4.1 Analysis and discussion Fama and French model

Firstly, the results of the Fama and French 3 factor model. The coefficients of the market return factor, SMB factor and HML factor are reported in the regressions outputs of appendix B. The normal returns calculated with the3 factor model are displayed in figure 1. This graphical view is given on how the normal return performed against the market return and regional portfolio return. As seen from the figure, the gap between the normal return with the two other increases substantially from November 2014 due

3_{This table represent the BHAR and CAR results with and without outliers. *, ** and ***, indicate statistical}

(24)

17 to an uplift in the Japanese stock markets. This is a possible limitation since it is harder to make solid interpretations about the variables which are to be explained.

Figure 2

Returns per trading day4

Out of the coefficient of the market return, 30 of the 36 companies have a significant coefficient for the market return on the excess return. However, from the SMB and HML factor, only 18 out of the 36 companies have one of the two factors as significant. This poses a serious limitation to the reliability of the normal returns. To gain certainty about the performance of the 3 factor coefficients, the Fama French model was compared with the standard CAPM model with a F-test (model 1: (𝑅&'− 𝑅)' = 𝛼&+

𝛽_&.𝑅/0'− 𝑅)'1 + 𝜀&) Model 2: (𝑅&'− 𝑅)' = 𝛼&+ 𝛽&.𝑅/0'− 𝑅)'1 + 𝑠&𝑆𝑀𝐵'+ ℎ&𝐻𝑀𝐿'+ 𝜀&)). The

results were that only 17 of the 36 companies had a significant F-test at 𝛼 = 10%. This concludes that less than half of the models were improved with the addition of the value and growth factor. However, the number of significant market return coefficients decreased with 4, and so did the level of significance of the remaining significant coefficients. Therefore, this thesis will stay with the Fama and French 3 factor model. However, further research should be done with different models to better predict the normal return for calculating the abnormal returns.

5.4.2 Analysis of the BHAR method

Secondly, the results from the Buy-and-Hold Abnormal Returns method. In table 4 all relevant results have been reported. It should however be noted that long-run analysis results should be interpreted cautiously as other co-founding events unrelated to the earthquake might have affected the abnormal returns. To test whether to reject the null hypothesis of no abnormal returns, the t-values are compared with the critical value of 𝑡G.GHV,IG = ±2.042. From this we reject the null hypothesis of no abnormal

4_{This figure gives a graphical illustration of the three main variables used in this thesis, namely the market}

return, average regional portfolio return and the normal return. This is taken over the entire post-event period (0,1306). 50% 70% 90% 110% 130% 150% 170% 190% 1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 Market Return Portfolio Return Normal Return

(25)

returns for the 6 month period, since |𝑡_k /el'dh| ≥ 𝑡_G.GHV,IG. The null hypotheses of the other three post-event windows are not rejected. Note that the 1 year period is significant if the outliers would have been included since ]𝑡_{E aPbc Q&'d ef'g&Pch}] ≥ 𝑡G.GHV,IG. With regard to the BHAR method we can therefore

state that hypothesis 1 is true. The BHAR method shows that the Tōhoku earthquake has an impact on the listed companies from the FIM prefectures. The second hypothesis about the long term effect of the Tōhoku earthquake being positive is not conclusive. The critical value of 𝑡G.GV,IG= 1.697 not crossed by

any of the longer term periods. Therefore, we cannot state with enough certainty that the second hypothesis is correct according to the BHAR method.

Table 5

BHAR method results5

Horizon Abnormal returns SD t-statistics P-value

6 Months -0.142*** 0.167 -5.096 0.000

1 Year 0.016 0.253 0.511 0.613

3 Years 0.061 0.685 0.531 0.599

5 Years 0.154 1.128 0.820 0.418

5.4.3 Analysis of the CAR method

Besides the BHAR method the CAR method was used as well. The same normal returns were used for the calculation. The results of the CAR method are displayed in table 5. Again the null hypotheses of no abnormal return will be tested for every post-event window. The critical value is the same as of the BHAR method, namely 𝑡_G.GHV,IG = ±2.042. With the found t-statistics we can reject the null hypothesis for 1 year, 3 years and 5 years periods since ]𝑡_{E aPbc,I aPbch,V aPbch}] ≥ 𝑡G.GHV,IG. The 6 months period is

no longer significant. With regard to the CAR method we can state that hypothesis 1 is true. The CAR method shows that the Tōhoku earthquake did have an impact on the listed companies of the FIM prefectures. The second hypothesis about the long term effect of the Tōhoku earthquake is regarded as true as well according to the CAR method. The 1-5 year windows all show positive signs and significant t-statistics at a one-sided 1% level. We can therefore conclude that according to the CAR method there is a long term positive effect of the Tōhoku earthquake on the regional portfolio.

5_{This table represent the BHAR method results. *, ** and ***, indicate statistical significance at the 10%, 5%}

(26)

19 Table 6

CAR method results6

Horizon Abnormal returns SD t-statistics P-value

6 Months -0.043 0.179 -1.433 0.161

1 Year _0.297*** _0.403 _4.426 0.000

3 Years _0.369*** _0.565 _3.920 0.000

5 Years 0.613*** 0.688 5.349 0.000

5.4.3 Discussion of the BHAR and CAR method

From the tables 4 and 5 we observe several similarities and differences. The similarities strengthen the model while the differences weaken it. For both methods the first hypothesis was found to be true. Therefore, with reasonable certainty we can state that the Tōhoku earthquake had a significant effect on listed companies with their main seat in one of the three FIM prefectures. Which is a logical result. The damages of the Tōhoku earthquake were estimated to be between 210 and 309 billion US dollar (em-dat). This enormous amount was mostly due to the destruction of building, plants infrastructure and other objects of value located in the FIM prefectures. Therefore it is a safe to assume the earthquake, tsunami and nuclear meltdown had a negative impact on the returns of the companies with their main seat in the FIM prefectures.

The second hypothesis is less clear since only the CAR method has the three longest horizons both positive and significant. However, there is still ground to state that there is a positive long term effect of the Tōhoku earthquake. This is since the BHAR method gives abnormal returns with positive signs for all the yearly periods. In addition, the 1 year period with outliers is significant as well. Therefore, the possibility of a positive long term effect of the Tōhoku earthquake on the regional portfolio will not be ruled out. Further research could strengthen the resulst obtained with the CAR method. The assumption of long-term positive abnormal returns is shared by Skidmore and Toya (2002). They stated that destruction of capital assets gives the opportunity to invest in better technologies and production solutions. Since expenditures are done to rebuild facilities just after the earthquake hit, but when production starts again, efficiency has increased due to better facilities, thus increasing returns. This can explain the negative effects of both methods for the first 6 months, and the positive effects of the following years.

6_{This table represent the CAR method results. *, ** and ***, indicate statistical significance at the 10%, 5% and}

(27)

5.5 Limitations

As already briefly discussed, there are limitations to the model and the reliability of abnormal returns. In this chapter these limitations are further discussed.

5.5.1 Fama and French significance

As discussed in chapter 5.1.1, the significance of the value and growth factors for calculating the normal return are not optimal. This poses a serious limitations to the reliability of the calculated abnormal returns and the drawn conclusions. Further research should be done to better the normal returns. A possible method for gaining better regression outputs is the portfolio approach (Kolari et al.). This model is commonly used when event dates cluster in time, which is the case for the regional portfolio since the earthquake happened at the same time for every company. The portfolio approach does not regress every company i, but regresses a portfolio with either equally or value weighted stocks with the market model. However, this model is proven to work better for short term than long term studies.

5.5.2 Difference in results BHAR and CAR

The BHAR and CAR give contradictory results when it comes to the significance of the 1, 3 and 5 year periods. This peaks interest since the signs are the same. The question arises which model predicts the periods better. Fama (1988) argues that the most appropriate way to test for abnormal performance is with the CAR method. However others disagree. To obtain more conclusive answers, more types of models should be used to get consensus about the accurate findings

5.5.3 Incomplete portfolio data

Another limitation of this thesis is the limited amount of data the Japanese companies make public and is understandable for non-Japanese people. Since all but one of the companies of the regional portfolio are small cap listed companies, there is not a lot of English documentation written by or about them. Therefore it is difficult to attain information about the places the companies conduct business. This is an important variable to have for each company, because it tells us whether a company truly lost a lot of assets in the FIM prefectures as a result of the earthquake. For example, it would be wrong to put a company in the regional portfolio if it only has its main seat in the FIM prefectures, but has all of its facilities and sales in Tokyo. This thesis failed to gain this data and therefore imposes another limitation on the correctness of the found values.

Another limitation is the bias that arises by only adding listed companies to the database. Of course this is a logical decision, since private companies do not have daily stock prices available. However, listed companies are often large and act on a national or global level. Therefore, they will be more resistant to a natural disaster than smaller private companies who will lose most of their assets and sales. Therefore, it would be beneficial for gaining the true impact the Tōhoku earthquake had on the regional economy of the FIM prefectures, to add private company results as well.

(28)

21

CHAPTER 6 Conclusion

This paper focuses on the effect of the 2011 Tōhoku earthquake on the Fukushima, Iwate and Miyagi prefectures. This was done by employing both the BHAR and the CAR method to do an event study on the listed companies of the FIM prefectures with the Tōhoku earthquake as t = 0. All the listed companies from the FIM prefectures were put into the regional portfolio. The sample size for the regional portfolio consisted of 36 listed companies with all but one being in the small cap segment. For every company the normal return was calculated with the Fama and French 3 Factor Model, using a 1 year estimation period. With the normal returns the BHAR and CAR methods were used to find the abnormal returns for a 6 months, 1 year, 3 year and 5 year post-event window. This resulted in the abnormal returns stated in table 5 and 6. The results show several significant abnormal returns. However, the BHAR and the CAR method were each other’s opposites with regard to significance levels. The BHAR only shows significance for the 6 month period, while the CAR method shows significance for the other 3 periods. With these results both hypotheses can be both accepted or rejected with the different methods.

(29)

REFERENCES

Bloom, N. (2013). Does uncertainty reduce growth? Using disasters as natural experiments.

The National Bureau of Economic Research Working Paper.

Barber, B. M. and Lyon, J. D., (1997). Detecting long-run abnormal stock returns: The

empirical power and specification of test statistics. Journal of Financial Economics,

43(3), 341-372.

Boehmer, E., Musumeci, J. and Poulsen, A. B., (1991). Event-study methodology under

conditions of event-induced variance. Journal of Financial Economics, 30(2), 253

272. Bourdeau-Brien, M. and Kryzanowski, L., (2017). The impact of natural disasters on the

stock returns and volatilities of local firms. Quarterly Review of Economics and

Finance, 63, 259-270

Downton, M. W. and Pielke, R. A., Jr. (2005). How accurate are disaster loss data? The

case of U.S. flood damage. Natural Hazards, 35(2), 211–228.

Fama, E. F. and French, K. R., (2015). Common risk factors in the returns on stocks and

bonds. Journal of Financial Economics, 33(1), 3-56.

Fama, E. F. and French, K. R., (2015). A Five Factor Asset Pricing Model. Journal of

Financial Economics, 116(1), 1-22.

Francis, J. A., Vavrus, S. J., (2012) Evidence linking artic amplification to extreme weather in

mid-latitudes. Geophysical Research Letters, 39(6), 1-6.

Kazama, M. and Noda, T. (2012) Damage statistics (Summary of the 2011 off the Pacific

Coast of Tōhoku Earthquake damage). Soils and Foundations, 52(5), 780-792.

Kolari, J. W., Fraser, D. R., Anari, A. (1988) The effects of securitization on mortgage market

yields; a cointegration analysis. Real Estate Economics, 26(4), 677-693.

Luo, N. (2012). The impact of disasters on global stock market: The case of the Japanese

2011 earthquake (Master Research Project). Saint Mary’s University.

Morse, D. (1984). An econometric analysis of the choice of daily versus monthly returns in

tests of information content. Journal of Accounting Research, 22(2), 605-623

Nakamura, E., Steinsson, J., Barro, R. and Ursúa, J. (2013). Crises and recoveries in an

empirical model of consumption disasters. American Economic Journal:

Macroeconomics, 5(3), 35–74.

Schumacher, I. and Strobl, E (2011). Economic Development and Losses due to Natural

Disasters. The Role of Hazard Exposure. Ecological Economics, 72, 97-105

(30)

23

Skidmore, M. and Toya, H. (2002), Do Natural Disasters Promote Long-Run Growth?

Economic Inquiry, 40(4), 664-687.

Skidmore, M. and Toya, H (2007). Economic Development and the Impact of Natural

Disasters. Economic Letters, 94(1), 20-25.

Strobl, E. (2011). The economic growth impact of hurricanes: Evidence from U.S.

coastal counties. Review of Economics and Statistics, 93(2), 575–589.

Valizadeh, P., Karali, B. & Ferreira, S. (2017). Ripple effect of the 2011 Japan earthquake on

international stock market. Research in International Business and Finance, 41,

556-576.

Von Ungern-Sternberg, T. (2010). Risk Selection in Natural-Disaster Insurance. Journal of

Institutional and Theoretical Economics, 166(2), 344-364.

Wang, L. and Kutan, A. M. (2013). The impact of natural disasters on stock markets:

Evidence from japan and the US. Comparative Economic Studies, 55, 672–686.

Worthington, A. C. and Valadkhani, A. (2004). Measuring the impact of natural disasters

on capital markets: An empirical application using intervention analysis. Applied

Economics, 36(19), 2177–2186.

Yamori, N. and Kobayashi, T., (2002). Do Japanese insurers benefit from a catastrophic

event? Market reactions to the 1995 Hanshin–Awaji earthquake. Journal of the

(31)

APPENDIX A Company abnormal returns

Table 7

Company abnormal returns per post-event window

Company name 6 Months 1 Year 3 Years 5 Years

YAKUODO -0.27917 -0.29183 -0.29675 -0.29584 NCXX -0.13099 -0.17579 -0.16194 -0.29048 BANK OF IWATE -0.18529 -0.08985 -0.06246 -0.05244 TOHOKU BANK -0.1113 -0.08091 -0.06766 -0.01703 KITA-NIPPON BANK -0.14141 -0.1067 -0.11593 -0.07019 JOBAN KAIHATSU 0.175512 0.211596 0.522869 0.576096 JOBAN KOSAN -0.33668 -0.29519 -0.32036 -0.31068 HONEYS HOLDINGS 0.023892 -0.04745 -0.06362 -0.11224 ASAKA RIKEN -0.19042 -0.13212 -0.13642 -0.12689 KOURAKUEN -0.06569 -0.06677 -0.0633 -0.07855 XEBIO HOLDINGS -0.07033 -0.16078 -0.15251 -0.14442 TOHO BANK -0.14187 -0.17014 -0.11676 -0.09006 FUKUSHIMA INDUSTRIES -0.00877 -0.03306 -0.03562 -0.0061 FUKUSHIMA BANK -0.15964 -0.11823 -0.10053 -0.12939 FUKUSHIMA PRINTING -0.11586 -0.09869 -0.08839 -0.05894 DAITO BANK -0.24389 -0.17613 -0.13828 -0.08933 YURTEC 0.315967 0.367174 0.462179 0.472917 KARULA -0.37051 -0.29402 -0.21181 -0.23278 TOHO ACETYLENE -0.18836 -0.13892 -0.11797 -0.10762 TOSNET -0.064 -0.14212 -0.15436 -0.07413 KURAMOTO -0.11057 -0.0189 -0.08199 -0.11264 TOYO KNIFE -0.19082 -0.1467 -0.1141 -0.1209 YAMADAI -0.43714 -0.45077 -0.39665 -0.00162 KOHSOKU -0.23614 -0.14963 -0.06443 -0.06738

FUJI CORPORATION LIMITED 0.076832 0.067433 0.023545 0.017171

KAMEI -0.2918 -0.09033 -0.09947 -0.15943 77 BANK -0.1374 -0.16779 -0.12809 -0.11569 FIDEA HOLDINGS -0.11348 -0.12203 -0.0301 -0.03439 SENKON LOGISTICS -0.10149 -0.14006 -0.14753 -0.14725 TOHOKU ELECTRIC PWR. -0.2512 -0.18981 -0.12504 -0.1779 TOHOKU CHEMICAL -0.24252 -0.12486 -0.14843 -0.14538 TOHOKU STEEL -0.22229 -0.13902 -0.15379 -0.16407 MAXVALU TOHOKU -0.07283 -0.06075 -0.04159 -0.02148 UEMATSU SHOKAI -0.21439 -0.12297 -0.13092 -0.08431 YAMAYA -0.26161 -0.22454 -0.16397 -0.14878 SATOH -0.08747 -0.006 -0.03045 -0.02547

(32)

25

APPENDIX B Fama and French 3 factor regression output

Table 8

YAKUODO Coef. St.Err t-value p-value Sig.

RmktRf 0.493 0.109 4.51 0.000 ***

SMB 0.344 0.238 1.45 0.149

HML -0.597 0.298 -2.00 0.046 **

_cons 0.001 0.001 0.88 0.381

Mean dependent var 0.001 SD dependent var 0.015

R-squared 0.076 Number of obs 260.000

F-test 6.986 Prob > F 0.000

Akaike crit. (AIC) -1442.575 Bayesian crit. (BIC) -1428.333 *** p<0.01, ** p<0.05, * p<0.1

Table 9

NCXX Coef. St.Err t-value p-value Sig.

RmktRf 1.327 0.406 3.27 0.001 ***

SMB 2.111 0.884 2.39 0.018 **

HML 0.289 1.108 0.26 0.795

_cons 0.004 0.003 1.11 0.269

F-test 4.927 Prob > F 0.002

Table 10

BANKOFIWATE Coef. St.Err t-value p-value Sig.

RmktRf 0.841 0.084 10.06 0.000 ***

SMB -0.537 0.182 -2.95 0.003 ***

HML -0.225 0.228 -0.98 0.326

_cons -0.001 0.001 -1.47 0.144

Mean dependent var -0.001 SD dependent var 0.014

F-test 52.648 Prob > F 0.000

Table 11

TOHOKUBANK Coef. St.Err t-value p-value Sig.

RmktRf 0.728 0.067 10.81 0.000 ***

SMB -0.469 0.147 -3.20 0.002 ***

HML -0.014 0.184 -0.07 0.940

_cons 0.000 0.001 -0.27 0.788

F-test 65.926 Prob > F 0.000

(33)

Table 12

KITANIPPONBANK Coef. St.Err t-value p-value Sig.

RmktRf 0.712 0.089 8.00 0.000 ***

SMB -0.176 0.194 -0.91 0.366

HML 0.194 0.243 0.80 0.425

_cons 0.000 0.001 -0.25 0.806

F-test 34.066 Prob > F 0.000

Table 13

JOBANKAIHATSU Coef. St.Err t-value p-value Sig.

RmktRf -0.028 0.169 -0.17 0.867

SMB 0.169 0.368 0.46 0.647

HML -0.267 0.461 -0.58 0.563

_cons 0.000 0.001 0.21 0.830

F-test 0.405 Prob > F 0.749

Table 14

JOBANKOSAN Coef. St.Err t-value p-value Sig.

RmktRf 0.653 0.091 7.14 0.000 ***

SMB 0.165 0.199 0.83 0.409

HML 0.058 0.250 0.23 0.815

_cons 0.000 0.001 -0.50 0.621

F-test 21.788 Prob > F 0.000

Table 15

HONEYSHOLDINGS Coef. St.Err t-value p-value Sig.

RmktRf 1.019 0.219 4.65 0.000 ***

SMB 0.982 0.477 2.06 0.041 **

HML 0.170 0.598 0.28 0.776

_cons 0.002 0.002 0.86 0.388

F-test 8.862 Prob > F 0.000

(34)

27 Table 16

ASAKARIKEN Coef. St.Err t-value p-value Sig.

RmktRf 0.794 0.161 4.93 0.000 ***

SMB -0.007 0.351 -0.02 0.985

HML 0.334 0.439 0.76 0.448

_cons 0.001 0.001 0.56 0.575

F-test 12.660 Prob > F 0.000

Table 17

KOURAKUEN Coef. St.Err t-value p-value Sig.

RmktRf 0.401 0.044 9.03 0.000 ***

SMB 0.008 0.097 0.08 0.935

HML -0.034 0.121 -0.28 0.776

_cons 0.000 0.000 -0.19 0.847

F-test 34.900 Prob > F 0.000

Table 18

XEBIOHOLDINGS Coef. St.Err t-value p-value Sig.

RmktRf 0.872 0.100 8.71 0.000 ***

SMB 0.013 0.218 0.06 0.952

HML 0.185 0.273 0.68 0.500

_cons 0.000 0.001 0.25 0.801

F-test 36.120 Prob > F 0.000

Table 19

TOHOBANK Coef. St.Err t-value p-value Sig.

RmktRf 0.897 0.076 11.84 0.000 ***

SMB -0.129 0.165 -0.78 0.434

HML -0.271 0.207 -1.31 0.191

_cons 0.000 0.001 -0.74 0.458

F-test 59.184 Prob > F 0.000

(35)

Table 20

FUKUSHIMAINDUS TRIES

Coef. St.Err t-value p-value Sig.

RmktRf 0.668 0.132 5.07 0.000 ***

SMB 0.152 0.287 0.53 0.597

HML 0.499 0.360 1.39 0.167

_cons 0.001 0.001 0.88 0.379

F-test 14.263 Prob > F 0.000

Table 21

FUKUSHIMABANK Coef. St.Err t-value p-value Sig.

RmktRf 0.857 0.131 6.56 0.000 ***

SMB -0.200 0.285 -0.70 0.482

HML 0.653 0.357 1.83 0.068 *

_cons 0.001 0.001 0.82 0.410

F-test 27.664 Prob > F 0.000

Table 22

FUKUSHIMAPRINTI NG

Coef. St.Err t-value p-value Sig.

RmktRf 0.588 0.145 4.05 0.000 ***

SMB 0.289 0.316 0.91 0.362

HML -0.919 0.397 -2.32 0.021 **

_cons 0.000 0.001 0.04 0.970

F-test 5.769 Prob > F 0.001

(36)

29 Table 23

DAITOBANK Coef. St.Err t-value p-value Sig.

RmktRf 0.939 0.119 7.88 0.000 ***

SMB -0.646 0.260 -2.49 0.014 **

HML -0.204 0.326 -0.62 0.532

_cons 0.000 0.001 0.45 0.654

F-test 33.496 Prob > F 0.000

Table 24

YURTEC Coef. St.Err t-value p-value Sig.

RmktRf 0.715 0.094 7.62 0.000 ***

SMB 0.088 0.204 0.43 0.666

HML 0.492 0.256 1.92 0.056 *

_cons 0.000 0.001 -0.59 0.553

F-test 32.363 Prob > F 0.000

Table 25

KARULA Coef. St.Err t-value p-value Sig.

RmktRf 0.189 0.075 2.53 0.012 **

SMB 0.235 0.163 1.44 0.150

HML 0.137 0.204 0.67 0.502

_cons 0.000 0.001 -0.26 0.794

F-test 3.181 Prob > F 0.025

Table 26

TOHOACETYLENE Coef. St.Err t-value p-value Sig.

RmktRf 0.702 0.120 5.84 0.000 ***

SMB 0.471 0.262 1.80 0.073 *

HML 0.933 0.328 2.84 0.005 ***

_cons 0.000 0.001 0.29 0.771

F-test 22.211 Prob > F 0.000

(37)

Table 27

TOSNET Coef. St.Err t-value p-value Sig.

RmktRf -0.046 0.176 -0.26 0.794

SMB 0.686 0.384 1.79 0.075 *

HML 1.022 0.481 2.13 0.035 **

_cons 0.001 0.002 0.57 0.572

F-test 2.038 Prob > F 0.109

Table 28

KURAMOTO Coef. St.Err t-value p-value Sig.

RmktRf 2.225 0.371 6.00 0.000 ***

SMB 1.735 0.808 2.15 0.033 **

HML 0.317 1.013 0.31 0.754

_cons 0.005 0.003 1.66 0.098 *

F-test 14.607 Prob > F 0.000

Table 29

TOYOKNIFE Coef. St.Err t-value p-value Sig.

RmktRf -0.059 0.235 -0.25 0.802

SMB 0.383 0.513 0.75 0.456

HML -0.018 0.642 -0.03 0.977

_cons -0.001 0.002 -0.43 0.670

F-test 0.324 Prob > F 0.808

Table 30

YAMADAI Coef. St.Err t-value p-value Sig.

RmktRf 0.125 0.207 0.61 0.545

SMB 0.358 0.451 0.79 0.428

HML -0.010 0.566 -0.02 0.986

_cons 0.001 0.002 0.43 0.669

F-test 0.278 Prob > F 0.841

Measuring the impact of the Japanese Tōhoku earthquake on regional stock prices : using the BHAR and CAR methods to check for abnormal performance