董監事質押與公司績效關係探討-以縰橫平滑移轉迴歸模型為例

全文

(1). . Institute of Business and Management National University of Kaohsiung, Taiwan .

(2)

(3) . . . . . . . . . . . . . . The Relationship between Directors’ Collateralized Shares and Firm’s Performance - The Panel Smooth Transition Regression Model. . . :

(4). . . . : . . . . . . . . . . . . . . . . . . . .

(5) .

(6) . !. /. ". <. #. =. T <. >. ?. =. u. t. . v. . . ). ¡. ¯. Û. Ü. Î. ©. q. Î. !. . . Ð. ß <. (. ). à. ö Y. ¥. ¦. î. ë ÷. . ù. . . P. r. (. 9. C. . . Q. e. m ;. ). . . : ;. c d. q. ~. 8. O. b p. $. 7. N. o. #. ). ú. q. A. «. ¯. $. $. '. F. . .. S. &. /. . $. R. f. .. N. Í. . r. q. ä. . Gs. x. !. j

(7). A. ü ©. ¹. º . Ø. æ. ð. ½. ¾ Í. Ù. ñ. ý. Y. ¯. . ). {. . F. °. +. G. Ú. è. ò. þ. . jÎ. ç . D. q. ±. Ì. ×. å. . ° < Z. Ë. ®. û. i. . Ö. . ¯. q ). ã. . ®. Õ. . . . Ê. ° . ª. . ¬. m . . ð. . Gk. . â + . ª. F. Ô. Y. q »¼º. É. Ó. . . º. q. Y. . q. ¹. á. . . ¨ ©. . 9. . . §. È. @. t. M. _. . 6. L. A. . 5. K. n. {. · ¸. 8. ?. ^. ¦. Ò. Y. ï. J. Y. . 4. a _. m. . q. . Æ. Þ ø. Ç. . . 7. ¥. Æ. ). `. . 23. I. _. . ¶ . _. . 1. H. }]. . . 0. @. l. |. . Ñ . . õ. °. . (. D. ]. Í . '. . ë . Z . Þ. I. ¯. . 7. Å. /. ^. :. p `. µ. .. G?. {. . ¤.

(8). . . u . é. ó. . ê. . . ë. ¥. ¦. . jÎ . K. $. í ². H.

(9). Ï. Ý. ì j+. ô. . . A. q. °. p. F. . . £. . - $. E. . . . ¢. ´. ¿ À À Á Â Ã Ä : Î. . . \ ]. E. . . ,. [. !. . . ³. Z. z. . . ². +. D. Y. y. ). *. jk. . . . X. i. <. ). C. W. x. . (. B. W. w. . '. V. . . U. D. . &. A. h. \. $. %. @. >. g ) t. $. ) Z. ñ. . . Y . ,. $. f -. . . .. n. G

(10). ! *. . + è. A. . . " ¬. +. k. . . /. r. ). æ . . ¯ ¥. . . ® °. A. . °. ). X 0. .

(11). q. . Y . <.

(12). #. J ¹. { . m. ñ. . . . S . v. . º ©. q. q. . ê . $. ]. Ú. u. ] %. 2&. jÎ ». Z Î. Y 3. 1 4. 2. jÎ. Î è. é. î. ë. Î. 5 6 6 7 8 9Î. .

(13) . TABLE OF CONTENTS LIST OF TABLES ............................................................................................................. iii LIST OF FIGURES ............................................................................................................. v ABSTRACT ........................................................................................................................ vi ....................................................................................................................................vii. Chapter 1 Introduction ....................................................................................................... 1 1.1 Background and motivations 1.2 Research purpose 1.3 Research framework Chapter 2 Literature Review .............................................................................................. 5 2.1 Ownership structure 2.2 Collateralized shares 2.2 Firm size and performance Chapter 3 Data Selection and Variables ........................................................................... 10 3.1 sample set 3.2 Variables 3.2.1 Dependent variable 3.2.2 Independent variables Chapter 4 Methodologies .................................................................................................. 13 4.1 Panel Unit Root Test 4.1.1 Levin, Lin and Chu (L-L-C, 2002) Panel Unit Root Test . 4.1.2 Im, Pearan and Sin (IPS, 2003) Panel Unit Root Test 4.1.3 Maddala and Wu (MW, 1999) Panel Unit Root Test 4.2 Fixed effect model 4.3 Panel Smooth Transition Regression 4.4 Setting of panel smooth regression model

(14) 4.4.1 Illustrate the model: test the homogeneity

(15) 4.4.2 Parameter estimation 4.4.3 Model evaluations Chapter 5 Empirical Results ............................................................................................. 24 .

(16) . 5.1 The relationship between pledge ratio and firms’ return 5.2 The relationship between pledge ratio and firms’ return 5.3 The relationship between size and firms’ return 5.4 Risk-adjusted result 5.5 The relationship between pledge ratio, divergences and excess return compared to market 5.6 Furthermore References .......................................................................................................................... 36. . .

(17) . LIST OF TABLES Table 1: Variable definitions and summary statistic ............................................................. 38 Table 2: Sample description ................................................................................................ 39 Table 3: Panel unit-root test results ..................................................................................... 40 Table 4: Panel data estimates: Fixed effects of pledge ratio ................................................. 41 Table 5: Panel data estimates: Fixed effects of divergence ratio ........................................... 41 Table 6: Estimation result of PSTR model for pledge ratio and return ................................. 42 Table 7: Estimation result of PSTR model for divergences between voting right and cash flow right and return ........................................................................................... 43 Table 8:Estimation result of PSTR model for pledge ratio (size for threshold variable) ....... 44 Table 9: Estimation result of PSTR model for divergences between voting right and cash flow right (size for threshold variable) ................................................................ 45 Table 10: Estimation result of PSTR model for pledge ratio and risk-adjusted return........... 46 Table 11: Estimation result of PSTR model for divergences and risk-adjusted return ........... 47 Table 12: Estimation result of PSTR model for pledge ratio and risk-adjusted return (absolute beta).................................................................................................................... 48 Table 13: Estimation result of PSTR model for divergences and risk-adjusted return (absolute beta).................................................................................................................... 49 Table 14: Estimation result of PSTR model for pledge ratio and excess return .................... 50 Table 15: Estimation result of PSTR model for divergences and excess return .................... 51 Table 16: Estimation result of PSTR model for pledge ratio and risk-adjusted return (size for threshold variable) .............................................................................................. 52 Table 17: Estimation result of PSTR model for divergences and risk-adjusted return (size for threshold variable) .............................................................................................. 53 Table 18: Estimation result of PSTR model for pledge and excess return (size for threshold variable) .............................................................................................................. 54 . .

(18) . Table 19: Estimation result of PSTR model for divergences and excess return (size for threshold variable) .............................................................................................. 55 Table 20: Estimation result of PSTR model for divergences ratio between voting right and cash flow right and return (1-divergence ratio) .................................................... 56 Table 21: Estimation result of PSTR model for divergences and risk-adjusted return (1-divergence ratio) ............................................................................................. 57 Table 22: Estimation result of PSTR model for divergences and excess return (1-divergence ratio) ................................................................................................................... 58 Table 23: Summary Statistic for pledge ratio as threshold variable ...................................... 59 Table 24: Summary Statistic for pledge divergence ratio as threshold variable .................... 59 Table 25: Summary Statistic for pledge ratio (size for threshold variable) ........................... 60 Table 26: Summary Statistic for divergence ratio (size for threshold variable) ..................... 60. . .

(19) . . LIST OF FIGURES. Figure 1: Transition function of the PSTR model for pledge ratio and return........................ 61 Figure 2: Transition function of the PSTR model for divergences and return........................ 61 Figure 3: Transition function of the PSTR model for pledge ratio and return (size for threshold) .......................................................................................................... 62 Figure 4: Transition function of the PSTR model for divergences and return (size for threshold) .......................................................................................................... 62 Figure 5: Transition function of the PSTR model for pledge ratio and risk-adjusted return ... 63 Figure 6: Transition function of the PSTR model for divergences and risk-adjusted return ... 63 Figure 7: Transition function of the PSTR model for pledge ratio and risk-adjusted return (absolute beta) ................................................................................................... 64 Figure 8: Transition function of the PSTR model for divergences and risk-adjusted return (absolute beta) ................................................................................................... 64 Figure 9: Transition function of the PSTR model for pledge ratio and excess return............. 65 Figure 10: Transition function of the PSTR model for divergences and excess return ........... 65 Figure 11: Transition function of the PSTR model for pledge ratio risk-adjusted return (size for threshold) ..................................................................................................... 66 Figure 12: Transition function of the PSTR model for divergence and risk-adjusted return (size for threshold)............................................................................................. 66 Figure 13: Transition function of the PSTR model for pledge ratio and excess return (size for threshold) .......................................................................................................... 67 Figure 14: Transition function of the PSTR model for divergences and excess return (size for threshold) .......................................................................................................... 67 . . .

(20) . The Relationship between Directors’ Collateralized Shares and Firm’s Performance - The Panel Smooth Transition Regression Model. Advisor: Dr. Yi-Kai Chen Department of Finance, National University of Kaohsiung Dr. Tsang-Yao Chang Department of Finance, Feng Chia University Student: Ju-Hong Teng Institute of Business and Management, National University of Kaohsiung. ABSTRACT Shares collateralization is one of popular issues in terms of corporate governance in Taiwan. The relationship between Taiwan corporate performance and directors’ collateralized shares by using an alternative approach is the objective of this study. Existing literatures regard the relationship as linearity and the results are controversial. Panel Smooth Transition Regression model (PSTR) purposed by González, Terasvirta and Dijk (2004) is employed to test the nonlinearity relationship between firm performance and directors’ pledge ratio. Firm’s size and directors’ pledge ratio are regarded as the threshold variables respectively in the tests by using the corporate data from 1997 to 2007. The results indicate that non-linearity relationship between firm performance and directors’ pledge ratio does exist. The more shares the directors collateralized, the worst the firm perform when the pledge ratio is lower than 17.507%. When the pledge ratio exceeds the threshold value, it has no significant effect on firm performance. The results explain the reason that the empirical studies have no concrete conclusion on the relationship because of the nonlinearity relationship.. Key words: Panel Smooth Transition Regression Model (PSTR), directors’ collateralized shares, pledge ratio, corporate governance. . .

(21) .

(22)

(23) . . . . . . . . . . . . . . . . . Æ : ø <. . ; î. ë =. A ø <. î. ë. =. A. @. A @. . D. 3. Þ. . C. ;E. ? ?. . A. . >. . ë B. . A. A. . >. A ?. @. 4. ï. ?. @. . . . . ! . . i. . U. v. D. F G. G. E V. W. & H. & H. I. G. I. J. & H. Y. ê J I. N. O. ) K. J Y. Z. X. J N 0. K O. L Y [. $ M. %. 0 Z. R. ). N R. [. <. O. b. ° d. c. @ P. â. ] \ X. . ^ ^. e. N Q. : 0. _. ) ` f. GS R. a. [ jD E 0. . jk T. ª. ) g. °. U /. h »i h. González, Terasvirta and Dijk (2004) m n o p q r s ] ^ (Panel Smooth Transition Regression model;PSTR)t u G H & I J v w X N O Y Z : x l R 0 fj y : z 1997 { 2007 2p | n } (balanced panel data)~ ° i ] X G H & I J v w GM . V W j e x l R 0 f N G H & I J v w 17.507% G H & I J v w X N O Y Z i h J 0 2î 17.507% p l R 0 fjç ¨ 17.507%G$ M a G : W ¯ ^ _ ` D E ° / e 0 f b c 2 ¯ ) GG H & I J v w Y N O Y Z x l R 0 fjÎ Î Î Î Î Î 0 ;n } p q r s ] ^ PSTR»G H & I J »I J v w »N O P @ Î . j. . 0. N. 7. p. kl. R. 20. fj). q. . . . C.

(24) . Chapter 1 Introduction. 1.1 Background and motivations The stock market is one of the major sources for corporations to raise funds. This market allows publicly traded firms to raise additional capital for further expansions by selling shares of ownership of the company to the public. Unfortunately, some investors buy the so-called “tank stock” such as: Procomp Informatics Ltd, Rebar Group, and Taichung Commercial Bank. Those prices of shares decline sharply and are suddenly affected by the economic collapse or some artificial events. The main aspect of why the Asian enterprise is unable to promote international competitiveness, is that corporate governance can’t reach the standard of developed country, Organization for Economic Co-operation and Development (OECD) mentions in Council Meeting at Ministerial level. OECD proposed five principles of corporate governance in 1998, and revised into six frameworks in 2004, which might be helpful for company developing. Also after East Asian economic collapse in the late 20th century, the World Bank' s president warned those countries that for the reason of being in a sustainable development, corporate governance has to be good. OECD proposed five principals of corporate governance in 1998, six frame-works are setup. They think these principals might help enterprises to accomplish well governance as a standard. After Jensen and Meckling (1976) proposing a principal-agent theory, which causes problems between principal and agent due to the separation of ownership and management. La Porta (1999) indicates ownership structure is the key factor for well corporate governance in Asia country, because family-controlled company is quite common in this area. Therefore, how the . .

(25) . families hold the companies and how they manage the companies are important for investors and stakeholders.. La Porta et al. (1999) is the first investigation of the. importance of ultimate control. They separate the ownership into direct and indirect ways to determine the ultimate owner (with most voting rights). But, Claessens et al. (2000) defined the ownership as cash flow rights. Also, La Porta et al.2002 purposes the more cash flow right ultimate controller hold the more the firm value. Lemmon and Lins2001point out we also have family-controlled shape in Taiwan. In. general form, Economic health of a nation depends substantially on how sound and ethical businesses are. Recently, in Taiwan, the embezzlement scandals Rebar causes the supervisors started to announce to the investors who owns the shares of public offering company the importance of abnormal collateralized shares variation. In our study, we focus on the supervisor and directors’ collateralized shares issue which compares to the relationship between block shareholders and firms’ performance is rare.. 1.2 Research purpose Jensen and Meckling (1976) propose a principal-agent theory, which causes problems between principal and agent due to the separation of ownership and management. Besides, they argue an incentive effect that is investors who are with large ownership stakes have strong incentives to maximize their firms’ value and are able to collect information and oversee managers, so it can help to overcome one of the principal-agent problems in the corporate governance. But Morck, Shleifer, and Vishny (1988) find an inverse U-shaped relationship between managerial equity ownership and firm valuation for a sample of U.S. firms. One interpretation is that firms’ performance improves with highly managerial ownership, but after a point, managers become entrenched and pursue private benefits at the expense of outside . .

(26) . investors, which is so-called entrenchment effect. In addition, La porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002) find that many large shareholders use the pyramid structure and cross holding enhance the control right, which will be more than cash-flow right out of American. They indicate that there is a positive incentive effect relating to the share of cash-flow rights held by large shareholders and the negative entrenchment effect relates to the share of control rights held by large shareholders. Yeh (2002) pointed out the largest shareholders owned 30.33% of voting right (control right) and exhibit far divergence between cash-flow rights and control rights, that means the shares’ structure is not widely held. In our study, we try to analysis the relationship between collateralized shares and firm’s performance. Unfortunately the collateralized shares are unique in Taiwan. Yeh (2002) proposes that the effect of collateralized shares to performance in Taiwan is equivalent to the effect of divergence between cash flow right and voting right to performance out of Taiwan. Therefore we apply both divergences between cash flow right and voting right and collateralized shares to measure performance. Reviewing literature of collateralized shares, there are some point of views which try to explain the relationship between collateralized and performance (Shen, 2001, Kao and Chiou 2002, Lin 2006, and Chen and Hu 2001). Shen(2001) indicated the collateralized shares, serving stocks as collaterals, are one of financial leverage approaches.. Chiou et al. (2002) point out that the term ”collateralized shares” referred to shareholders’ personal behavior, in other words, there is no relationship between them. Kao and Chiou (2002) and Lin (2006) suggest that directors and supervisor would announce unrealistic message to outside. That might be negative relationship. Finally, we apply panel smooth transition regression model (PSTR) to find a threshold of collateralized . .

(27) . shares to explain the relationship between performance and directors’ collateralized shares.. 1.3 Research framework To achieve these purposes, this study is structured as follows. The first section deals with the background and motivations, purposes and major findings. Literature review offers the relationship between ownership structure and performance also including collateralized share which will be shown in the second section. The third and fourth chapter includes sample selection and data description, panel smooth transition regression. Finally, the empirical results and conclusions are showed in the fifth and sixth section.. . .

(28) . Chapter 2 Literature Review After OECD proposed five principals of corporate governance in 1998, six frame-works are setup. They think these principals might help enterprises to accomplish well governance as a standard. The frameworks are below. 1. Ensuring the basis for an effective corporate governance framework. 2. The rights of shareholders and key ownership functions 3. The equitable treatment of shareholders 4. The role of stakeholders in corporate governance 5. Disclosure and transparency 6. The responsibilities of the board Taiwan Economic Journal (TEJ) uses above structure and concludes the view of firm and academic to develop the corporate governance data base below: divergence of holding and control right, the equitable treatment of shareholders, the disclosure and transparency, the stability of organization and human resource, and the responsibility of society. Under each category, there are few variables to stand for. However La Porta (1999) indicates ownership structure is the key factor for well corporate governance in Asia country. From his point of view, we only concern the rights of shareholders and key ownership functions as key function.. 2.1 Ownership structure La Porta et al. (1999) is the first investigation of the importance of ultimate control. They separate the ownership into direct and indirect ways to determine the ultimate owner (with most voting rights). Direct ownership is the shares directly owned by the ultimate owner, and the indirect ownership trace is calculating the . .

(29) . shares held by other legal entities that are obviously owned or controlled by the ultimate owner. In the other side, Claessens et al. (2000) defined the ownership as cash flow rights. According to Claessens et al. (2002) the divergence between cash flow rights and control rights are negatively correlated with firms’ performance In addition, La Porta et al. (2000) found that the further the divergence between cash flow rights and control rights, the easier the controller might find opportunities to expropriate investors. Bebchuk, Kraakman et al. (2000) category the three basic mechanisms that permit a company’s controller to retain only a minority of the cash-flow rights attached to the firm’s equity: differential voting rights structures, pyramid structures, and cross-ownership structure. Jensen and Meckling (1976) propose a principal-agent theory, which causes problems between principal and agent due to the separation of ownership and management. From viewing those scandal public offering company, it should be noted a phenomenon that owners/block shareholders use little money to own shares with control ship of the company. However, block shareholders can make use of assets totally even if they own little proportion of the total assets. In other words, they are not responsible for all debts if the corporate suffer financial crises. Such a curious situation that the deviation between the control ship and the ownership would be dangerous because it could give block shareholders an incentive to misapply assets due to the self-interest mindset and arising other financial distress. Besides, they argue an incentive effect that is investors who are with large ownership stakes have strong incentives to maximize their firms’ value and are able to collect information and oversee managers, so it can help to overcome one of the principal-agent problems in the corporate governance. But Morck, Shleifer, and Vishny (1988) find an inverse U-shaped relationship between managerial equity .

(30) . ownership and firm valuation for a sample of U.S. firms. One interpretation is that firms’ performance improves with highly managerial ownership, but after a point, managers become entrenched and pursue private benefits at the expense of outside investors, which is so-called entrenchment effect. In addition, La porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002) find that many large shareholders use the pyramid structure and cross holding enhance the control right, which will be more than cash-flow right out of American. They indicate that there is a positive incentive effect relating to the share of cash-flow rights held by large shareholders and the negative entrenchment effect relates to the share of control rights held by large shareholders. Yeh (2002) pointed out the largest shareholders owned 30.33% of voting right (control right) and exhibit far divergence between cash-flow rights and control rights, that means the shares’ structure is not widely held due to the divergence of voting rights and cash-flow rights.. 2.2 Collateralized shares Bebchuk, Kraakman et al. (2000) category the three basic mechanisms that permit a company’s controller to retain only a minority of the cash-flow rights attached to the firm’s equity: differential voting rights structures, pyramid structures, and cross-ownership structure. In Taiwan, collateralized shares is another way to gain control right (Kao, 2002). For instance, Wang family (Rebar group) use collateralized share to enhance the control right, but actually they do not have such shares (money). In tough situation, they might harm firm’s value or expropriate investors’ wealth. Chiou et al. (2002) provide a feasible definition of collateralized shares, the big owners or supervisors borrow money from bank by collateralizing the shares of company they own, and point out that the term “collateralized shares” refers to .

(31) . shareholders’ personal behavior and it is irrelevant to the corporation, in Shen et al. (2001) the collateralized shares, serving stocks as collaterals, are one of financial leverage approaches for individual. But the corporation value could be reduced and other shareholders’ right would be deprived if directors and supervisors collateralized their shares, especially during the period of recession, they could invest in riskier investment or resort to illegal conduct due to the pressure for stock disposals rather than invest in efficient investment. Chen and Hu (2001) also show that the directors and supervisors could exercise their power to invest in riskier investment or resort to illegal conduct during the period of recession. However, it could be a factor for block shareholders to abuse assets due to self-interest motivation. Kao(2002) indicates the bigger the difference between cash flow right and voting right caused principal agent problem between controller and block shareholders severely. Kao and Chiou (2002) and Lin (2006) suggest that directors and supervisor would announce unrealistic message to outside shareholders through accounting earning management. They indicate the collateralized share may cause the agency problem, directors and supervisor would harm shareholders in the long run. But, Hsiung (2000) and Chiou (2002) Find that the higher the proportion of collateralized shares, the poorer the operating performance and thus the higher the possibility of financial crises.. 2.2 Firm size and performance The relationship between firm size and performance has been the subject of many studies, some of which have provided conflicting results. A number of studies have indicated a positive relationship between firm size and performance. The scale economy justification for a positive relationship between firm size and profitability is .

(32) . prominent in the works of Hall and Weiss (1967) and Scherer (1973). In the other way, capital market imperfections provide yet another conceptual argument to support size related differences in profitability. Reingnum and Smith (1983) find that lenders charge risk premiums of small firms that exceed what is justified by increased risk of default. Also Meyer (1967) indicates these differences in borrowing patterns between large and small borrowers as a sourced of increased borrowing cost for small firms. However, Michael Porter (1985 and 1998) offers arguments to suggest that the relationship between firm size and profitability may be non-linear. Porter’s argument focuses on his “stuck in the middle” hypothesis, which suggests that profitable niches are available to both very small and very large firms, but mid sized firms may find it difficult to develop an effective and profitable strategy. According to Porter’s hypothesis, there are profitable opportunities available to small firms serving localized niche markets and profitable opportunities available to large firms following a market wide strategy. Medium size firms, on the other hand, are too large to pursue niche markets but too small to compete against national or international companies whose focus is on serving the entire market. The stuck in the middle hypothesis suggests that the relationship between firm size and profitability is a non-linear cubic function. In empirical study, Amato (2004) finds that a cubic model with a positive, negative, positive sign pattern best describes the relationship between profitability and firm size in the U.S. retailing sector.. .

(33).

(34) . Chapter 3 Data Selection and Variables 3.1 sample set . To explore the relationship between firms’ performance and directors’ pledge. ratio, we employ the panel smooth transition regression model to test if there are thresholds, in addition, firm’s size and directors’ pledge ratio are regarded as the threshold variables respectively in the tests. We conduct our investigation using balanced panel data for a sample of 291 1 selected Taiwan Stock Exchange (TWSE)-listed companies in Taiwan from 1997 to 20072. We obtain all our data from the Taiwan Economic Journal (TEJ) database of Taiwan. However we have all publicly traded companies but neither financial nor insurance companies listed in TEJ. It is because financial and insurance companies are special cases; we can not use the same measurement to discuss within it. In other words, the nature of performance and corporate governance mechanism in these industries is not comparable to those of non-financial firms. The final sample is 291 public trading companies, distributed across the eighteen industry sectors as following: Electronics (66), Textiles (38), Steel and Iron (22), Construction (22), Plastics (17), Food (17), Electric & Machinery (15), Transportation (13), Wholesale and retail (11), Ele. Appliance and Cable (10), Rubber (8), Cement (7), Paper & Pulp (7), Glass & Ceramics (6), Tourism (5), Automobile (4) and Chemicals (2). The residual 21 companies are from the others.. 1. For a balance panel data, we must delete the missing data or abnormal data, therefore we have 291 publicly traded companies during 1997 to 2007. 2 TEJ corporate governance database deal the data from 1996, but there are too much missing data. However we collect the data from 1997. . .

(35) . 3.2 Variables 3.2.1 Dependent variable Various risk-adjusted performance measures have been constructed to single out the selection (stock-picking) ability. Fund performance is defined as the intercept in the time series regression of the excess fund return1 on the excess returns of passive benchmark portfolios.. Currently, most studies use multi-factor models to estimate. Jensen’s alpha. One of the most frequently used specifications is a three-factor model of Fama and French (1993).. The corresponding factor returns, RMRF, SMB (firm. size: stock price times the number of shares), HML (book-to-market equity (the ratio of the book value of the firm’s common stock to its market value, are calculated as the difference between the returns on small and big-stock portfolios and the returns on portfolios with high and low book-to-market equity, respectively. A model that describes the relationship between risk and expected return and that is used in the pricing of risky securities. The general idea behind Capital Asset Pricing Model (CAPM) is that investors need to be compensated in two ways: time value of money and risk. The time value of money is represented by the risk-free (rf) rate in the formula and compensates the investors for placing money in any investment over a period of time. The other half of the formula represents risk and calculates the amount of compensation the investor needs for taking on additional risk. This is calculated by taking a risk measure (beta) that compares the returns of the asset to the market over a period of time and to the market premium (Rm-rf). But in reality, many empirical studies do not support the view point of CAMP, they are questionable that CAPM based on the idea that an asset' s returns can be predicted using the relationship between that same asset and many common risk factors. However Arbitrage Pricing Theory (APT) Created in 1976 by Stephen Ross, this theory predicts a relationship . .

(36) . between the returns of a portfolio and the returns of a single asset through a linear combination of many independent macro-economic variables therefore risk-adjusted performance measures have been constructed in our study by beta as Rit / β . 3.2.2 Independent variables There are two categories of explanatory variables in our panel data. The threshold variable, i.e., pledge ratio is measured by the collateralized shares held by all managers (directors, supervisors, and top executives), is the key variables that we use to investigate whether there are threshold effect of pledge ratio and the convergence between voting right and cash flow right which are conducted in empirical literature on firm performance (Chiou et al., 2002; Kao and Chiou, 2002; Hsiung, 2000). We also introduce three control variables commonly used in the analysis of financial performance (e.g. Mak and Kusnadi, 2005), the natural log of the book value of total assets (Size) to capture intangibles related to the firm size; the ratio of market value to book value (growth); the raw return among industries (industry return). However, those variables, while important , there is considerable disagreement among researchers about the directions and we do not anticipate them either positively or negatively related to dependent variables. In addition, Chiu et al. (2002) and Kao and Chiou (2002) indicate that the size effect is another important factor to affect the performances; we may conduct it to our threshold variable. The clearly definition of variable is listed on Table 1.. . .

(37) . Chapter 4 Methodologies It is because our study is a panel data research, first of all, we have to confirm whether the data is stationary or not by panel unit root test. After the test, the use of panel data technique allows us to determine the temporal evolution of groups of individual rather than analyzing the temporal behavior of each. Moreover we use the panel smooth transition regression developed by González, Teräsvirta and Dijk (2004, 2005) to analyze the smooth transition effect of balanced panel data. Furthermore we apply empirical data to discuss the relationship between pledge ratio and firm’s performance in Taiwan, and is there a threshold value for enhancing firm’s performance.. 4.1 Panel Unit Root Test González,’s (2004) panel smooth transition regression model requires that the variables in the model be stationary in order to avoid spurious regressions and go further estimations of the panel threshold regression. First of all, we must show the unit root test to ensure the stationary status for all variables. We adopt the Maddala and Wu (1999), Levin-Lin-Chu (LLC) (2002), and Im-Pesaran-Shin (IPS) (2003), for panel data. Based on the results of the unit root test of each panel, the variables have stationary characteristics since the nulls of the unit root are mostly rejected in Table 2. 4.1.1 Levin, Lin and Chu (L-L-C, 2002) Panel Unit Root Test . Levin, Lin and Chu (2002) found that the panel approach substantially increases. power in finite samples when compared with the single-equation ADF test, proposed a ^. panel-based version that restricts β i by keeping it identical across cross-industries as follows: . .

(38) . ∆X it = α i + βX i ,t −1 + γ i t + ∑ θ ij ∆X i ,t − j + ε it k. (1). j =1. Where i =1,2,…N indexes across cross-industries. Levin-Lin-Chu tested the null hypothesis of β1 = β 2 = ... = β = 0 against the alternative of β1 = β 2 = ... = β < 0 , ^. ^. with the test based on the test statistic t β = β / se( β ) 4.1.2 Im, Pearan and Sin (IPS, 2003) Panel Unit Root Test . Im et al., (2003) relaxed the assumption of the identical first-order atoregressive. coefficients of the Levin-Lin-Chu test and developed a panel-based unit root test that allows β to vary across regions under the alternative hypothesis. In addition, Im-Pesaran-Shin tested the null hypothesis of. β1 = β 2 = ... = 0 against the. alternative of β i < 0 , for some i. The Im-Pesaran-Shin test is based on the mean group approach. They use the average of the t β statistics to perform the following t-bar statistic: Z =. N [t − E (t )] / Var (t ). (2). Where t = (1 / N )∑ t β , E( t ) and Var( t ) are respectively the mean and variance of N. i =1. each t β statistic, and they are generated by simulations3. 4.1.3 Maddala and Wu (MW, 1999) Panel Unit Root Test An alternative approach to panel unit root tests derives tests that combine the p-values from individual unit root tests. This idea was proposed by Maddala and Wu (1999), If we define π i as the p-value from any individual unit root test for cross-section, then under the null of unit root for all cross-sections, it demonstrates that:. 3. . For further details, see Im et al., 2003 .

(39) . ∑Φ N. 1. Z=. N. −1. (π i ) → N (0,1). (3). t =1. Where Φ −1 is the inverse of the standard normal cumulative distribution function. It reports both asymptotic χ 2 and standard normal statistics using ADF and Phillips-Perron individual unit root tests. The null and alternative hypotheses are the same as for the as IPS.. 4.2 Fixed effect model In an attempt to determine the determinants of directors’ collateralized shares and divergence ratio of cash flow right to voting right, in this study, the panel data techniques have been employed. The use of panel data technique allows us to determine the temporal evolution of groups of individual rather than analyzing the temporal behavior of each. This technique takes into account the individual heterogeneity, allows a larger number of data points, and improves the efficiency of the estimates. Panel data may have group effects, time effects, or both. These effects are either fixed effect or random effect. A fixed effect model assumes differences in intercepts across groups or time periods, whereas a random effect model explores differences in error variances. The Hausman specification test compares the fixed versus random effects under the null hypothesis that the individual effects are uncorrelated with the other regressors in the model (Hausman 1978). If correlated (H0 is rejected), a random effect model produces biased estimators, violating one of the Gauss-Markov assumptions; so a fixed effect model is preferred. Formally the model is: yit = xitβ + αi + uit. (4). where yit is the dependent variable observed for individual i at time t ,. β is the vector of coefficients, . .

(40) . xit is a vector of regressors,. αi is the individual effect uit is the error term. The model (4) is also called the analysis-of-covariance model. Without attempting to make the boundaries between regression analysis, analysis of variance, and analysis of covariance precise, we can say the regression model assumes that the expected value of y is a function of exogenous factor x, while the conventional analysis-of-variance model stipulates that the expected value of yit depends only on the class i to which the observation considered belongs and that the value of the measured quantity, y, satisfies the relation yit = αi + uit, where the other characteristics, uit, are random and are in no way dependent on the class this individual belongs. . regression model enables us to assess the effects of quantitative factors, the analysis-of-variance model those of qualitative factors; the analysis-of-covariance model covers both quantitative and qualitative factors.. 4.3 Panel Smooth Transition Regression Apply this model for completed catching the heterogeneity in the data by individual effect and time effect. In the general panel data model, they assume the parameter is fixed, but in empirical, the fixed of constant may not describe the relationship between variables appropriately, hence cause the wrong result. Therefore there are some panel models allowed constant varied by time including random parameter model and the model which parameter is the function of exogenous variable. The representative of latter one is the panel threshold regression developed by Hansen (1999). This model could separate the observation values of panel data into some different homogenous group or homogenous regimes, and there are different parameters in different regime. The characteristic of Hansen’s model is that using a . .

(41) . time varying threshold variable to separate the panel data into some different regimes, having a jumping effect when the observation is close to threshold value, but the jumping effect is not appropriate in some empirical test. Therefore our study apply the panel smooth transition regression which adding a transition speed into model to modify the jumping process, and use parameter of transition speed to describe the phenomena of smooth transition around transition threshold. Panel smooth transition regression model which compose with a fixed effect of exogenous regression coefficient could be explained by treating the model as linear heterogeneity or non-linear homogeneous panel data. Before using panel smooth transition regression we have to test whether our data is heterogeneity or not, if it exists, we could treat the model as non-linear panel model. Each entity will vary by time due to the time varying transition variable, also could treat PSTR as a kind of non-linear homogeneous panel model, the model will be separated into N+1regimes depending on the setup of transition variable, and the panel model of each regime is homogenous model. Moreover the basic setting of Smooth Transition Autoregreesive (STAR) which is developed by Teräsvirta (1994, 1998) is below: y it = u i + β 0'x it + β 1'x it g ( q it ; γ , c ) + ε it. (5). where, i =1,…,N is the individual entity, t= 1,…,T is time, yit is a scalar, xit is a k dimensional vector stand for the time varying exogenous variable, ui is a fixed effect of entity, . .

(42) . ε it is error term, g ( qit ; γ , c ) is transitional function and a continuous function. Among, qit is a transitional variable between zero and one, γ is a transition speed, and c is a transitional threshold value. Granger and Teräsvirta (1993), Teräsvirta (1994), and Jansen and Teräsvirta (1996) define the transitional function as follow: g ( qit ; γ , c) = (1 + exp( −γ ∏ ( qit − c j ))) −1 m. j =1. indeed γ > 0 , and c1 ≤ c 2 ≤ ... ≤ c m and the c = (c1 ,..., c m )' is a location parameter of m dimension vectors. No matter m=1,2…, γ always affects the slope of g(), the greater of γ the steeper of graph of g() function. When γ → ∞ the model will be the same with Hansen’s model as following: y it = µ i + β 0'x it + β1'xit φ (qit ; c ) + ε it. (7). 1 if qit ≥ c φ (qit ; c ) =  0 if qit ≤ c. In the other hand, when γ → 0 , the g() function will approximately be linear function; the convention of structure is not obvious. Because the g() function is between zero and one, the critical value of coefficient in the regression is β 0' and β 0'+ β1'. But the generalized-PSTR allowed several different regimes, the model as following.. y it = µ i + β 0'x it + ∑ β j'xit g (qit ; γ j , c j ) + ε it r. ( 8 ). j =1. The plot of these transition function, g( ), is determined by (6), j=1,…,r . .

(43) . represent there might be r smooth transition function, and 2 r different regimes. When m=1 and γ → 0 , is a u-shaped special case. Moreover when q lies between c1 and c 2 , g() will be zero, in the other case, when q locates at c1 or c 2 , the g() will be 0 or 1 such a jumping, treat as multi threshold regimes model.. 4.4 Setting of panel smooth regression model 4.4.1 Illustrate the model: test the homogeneity Before setting the model we have to run the testing homogeneity to know whether the testing model is belonged to PSTR. If the data belong to homogeneity, PSTR is not appropriate for analysis but the normal linear panel model. In economic side, relying on the test, we could understand the sensitivity factor in the model at same degree of sensitivity for all panel data, without the structure transition. For testing H 0 : γ = 0 or H '0 : β1 = 0 in PSTR model could reduce as the. homogenous model. There’s no standard for doing so, because it will include disturbed parameter especial location parameter (c) under each null hypothesis. Davies (1977,1987) is the first one to study this problem, later Luukkonen, Saikkonen and Teräsvirta (1988), Andrew and Ploberger (1994) ,and Hansen (1996) purposed the solutions applying by time series. Our study take Luukkone, Saikkonen and Teräsvirta (1988) as solution. They take null hypothesis as γ = 0 when proceeding testing homogeneity. Hence they used the first term of the Taylor expansion to solve problem, and replace ( 5 ) to y it = u i + β 0'* xit + β1'*xit qit + ... + β m'* xit qitm + ε it*. (9). where, β1'*,..., β m'* are the multiplier of γ , ε it* = ε it + R m β1'x it , and Rm is remainder. It is the same in statistic meaning for whether γ equals to zero in term ( 5 ) and . .

(44) . { }. β 0* = ... = β m* = 0 in term (8) and ε it* = {ε it }. Therefore it did not affect asymmetric distribution theory when we use the expansion to approximate to the term. Under null hypothesis, it is convenient to test LM, first of all remove fixed effect from term (8), then calculus LM of transitional model, it could separate Chi statistic and F statistic, calculation as following: 1.. y it = βxit + ε ' ;. calculus. for. residual. sum. of. squares. ( SSR0 ). y it = y it − ∑t y it / T ; x = y it − ∑t x it / T 2. ~ y it = β~ x it + ( xit'qit − ∑t xit'q it / T ,...., ∑t xit'q itm / T ) + ε * ; calculus for SSR1. ~ y it = yit − ∑t y it / T ; ~ x it = y it − ∑t xit / T 3.. 2 LM = TN (SSRO − SSR1 ) / SSR0 ~ χ mk. LM F = {( SSR0 − SSR1 ) / mk }/{RSS1 /(TN − N − mk )} ~ F [mk , TN − N − mk ]. We could examine whether the model exist non-linear relationship by LM of different distribution.. 4.4.2 Parameter estimation For estimation of parameter in PSTR, first we have to remove individual-specific means achieving erase individual effect ( u i ), then run the nonlinear least square, NLS to estimate the parameter. Therefore revised term ( 5 ) into: y it = u i + β 'xit (r , c) + ε it. ( 10 ). Among, xit (r , c ) = ( xit', x it'g (qit ; r , c)) ', β = (β 0', β1') ', then remove the individual mean from term (10): ~ y it = β '~ xit (r , c) + ε~it. ( 11 ). Among,. . .

(45) . ~ y it = y it − y i , ~ xit (r , c ) = ( xit' − x i', xit'g (qit ; r , c ) − wi'(r , c )) ', ε~ = ε it − ε i , y i , xi , wi , and ε i are individual mean,. wi ≡ T −1 ∑t =1 xit g (q it ; r , c) . T. The transition vector ~ xit (r , r ) decided by r and c, but r and c are decided by levels and individual mean. Therefore the most appropriate ~ xit (r , c) in NLS should re-calculate again and again. Then for estimating the parameters of term (11), determinate the most appropriate parameter by using NLS, the parameter could minimize the sum of squared error, SSE, and the formula of SSE is:. Q c (r , c ) = ∑∑ ( ~ y it − βˆ (r , c) ~ xit (r , c)) 2 N. T. ( 12 ). i =1 t =1. In the term (12), βˆ (r , c) is calculated by OLS from term (11). However if the ε it is normal distribution, the estimation process will exactly be the same with maximum likelihood.. 4.4.3 Model evaluations The method of model evaluation divided into testing parameter constancy and test whether the model still exist heterogeneity.. 1. Testing parameter constancy Few studies discuss about parameter constancy in panel model, because the data of cross section is much longer but the time series is not in normal panel model. But Lundbergh, Teräsvirta and Dick (2003) purposed time varying panel smooth transition regression (TV-PSTR) to test parameter constancy, the model setting as: ' y it = ui + ( β 10' xit + β11' xit g (qit ; γ 1 , c1 )) + f (t ; γ 2 , c 2 )(β 20 xit + β 21' xit g (qit ; γ 1 , c1 )) + ε it. . . ( 13 ).

(46) . f() is a transitional function which treats time as transitional variable, Therefore we. could revise it as: yit = ui + ( β10' + β 20' f (t ; γ 2 , c 2 )) xit + ( β11' + β 21' f (t ; γ 2 , c 2 )) xit g ( qit ; γ 1 , c1 ) + ε it. ( 14 ). And we judge whether the parameter is consistency or not depending on f(). in. TV-PSTR, the f() setting as following: f (t ; γ 2 , c 2 ) = (1 + exp( −γ 2 Χ (t − c 2 j ))) −1 h. ( 15 ). j =1. The definition of parameter is the same as term (6), the transitional variable q transfer into t. To replace f( ) by its first-order Taylor expansion around γ 2 = 0 , after rearranging terms this yields the following auxiliary regression: y it = µ i + β10*'x it + β1*'xit t + β 2*'xit t 2 + ... + β h*'xit t h + ( β 20*'xit + β h*'+1 xit t + ... + β 2*'h xit t h ) g (qit ; γ 1 , c1 ) + ε it*. ( 16 ). Where ε it* = ε it + R (t , γ 2 , c 2 ) and R(t , γ 2 , c2 ) is the remainder term. The parameter vectors β *j for j= 1, 2,… , h, h+1,… ,2h are multiples of γ 2 , such that the null hypothesis H 0 : γ 2 = 0 in term (13) can be reformulated as H 0 : β *j = 0. { }. for j= 1, 2,… , h, h+1,… ,2h in the auxiliary regression. Under H 0* ε it* = {ε it } , so the Taylor series approximation does not affect the asymptotic distribution theory. The testing procedures are similar to latter one. Under the null hypothesis, LM χ is asymptotically distributed as χ 2 (2hk ) and. LM F = LM χ / 2hk is approximately. distributed as F(2hk, TN-N-2k(h+1)-(m+1)). When the null model is a homogeneous fixed effects model β11 ≡ β 21 ≡ 0 in (13), (16) collapses into a parameter constancy test in this model. . .

(47) . 2. Testing the hypothesis of no remaining heterogeneity The assumption that a two-regime PSTR mode (5) with (6) adequately captures the heterogeneity in a panel data set can be tested in various ways. In the PSTR framework it is a natural idea to consider an additive PSTR model (5) with r=2, or three regimes, as an alternative, Thus, y it = u i + β 0'xit + β1'xit g1 ( qit(1) ; γ 1 , c1 ) + β 2'xit g 2 ( qit( 2 ) ; γ 2 , c 2 ) + ε it. ( 17 ). Where the transition variables qit(1) and qit( 2 ) can but need not be the same. The null hypothesis of no remaining heterogeneity in an estimated two-regime PSTR model can be formulated as H 0 : γ 2 = 0 in (17). This testing problem is again complicate by the presence of unidentified nuisance parameters under the null hypothesis. As before, the identification problem is circumvented by replacing g 2 ( qit( 2 ) ; γ 2 , c2 ) by a Taylor expansion around γ 2 = 0 . Choosing a first-order Taylor approximation leads to the auxiliary regression y it = u i + β 0*'xit + β1*'xit g1 ( qit(1) ; γˆ1 , cˆ1 ) *' *' β 21 xit g it( 2 ) + β 22 xit ( g it( 2 ) ) 2 + ... + β 2*'m xit ( g it( 2 ) ) m + ε it*. ( 18 ). Where γˆ1 and cˆ1 are estimates under the null hypothesis. The hypothesis of no remaining heterogeneity can be restated as H 0* : β 21* = ... = β 2*m = 0 . When H 0* holds, the. LM χ statistic has an asymptotic. χ 2 ( mk ) distribution, whereas LM F is. approximately distributed as F (mk , TN − N − 2 − k (m + 2)) .. . .

(48) . Chapter 5 Empirical Results 5.1 The relationship between pledge ratio and firms’ return In an attempt to determine the determinants of returns, in this study, the panel data techniques have been employed. The use of panel data technique allows us to determine the temporal evolution of groups of individual rather than analyzing the temporal behavior of each of them. This technique takes into account the individual heterogeneity, allows a larger number of data points, and improves the efficiency of the estimates. When we perform the Hausman test specification, the test recommends the use of fixed effects model. Table 4 and Table 5 report the relevant estimates following the analysis. In Table 4, almost the explanatory variables are seen to be significant in explaining return and are in line with recent empirical evidences. For instance directors’ collateralized shares measured as pledge ratio is seen to be negative and significant with return and abnormal return confirming the results of Chen and Hu (2001) Kao and Chiou (2002) and Shen et al. (2001) for earning management also supporting Hsiung (2000) and Chiou (2002)

(49) poorer the operating performance and thus the higher the possibility of financial crises. Moreover, size market to book value and industrial return are positive and significant in explaining return, risk-adjusted return and abnormal return consistent with Reingnum and Smith (1983), and Mak and Kusnadi (2005). However in Table 5, we find size, market to book value and industrial return are positive and significant in explaining return, risk-adjusted return and abnormal return consistent with previous literature. But there’ s no supporting on the relationship . .

(50) . between divergence ratio and return (La porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002) indicated the negative relationship between them).. 5.2 The relationship between pledge ratio and firms’ return The first step consists of testing the pledge ratio or divergences between voting right and control right against a specification with threshold effects. The dependent variable Yit represents firms’ return, and the independent variable Pit represents pledge ratio which is a threshold variable. We can treat the other variables as control variables or explain variables. In addition, µ i , the fixed effects, represents the heterogeneity of companies under different operating condition. Errors ε it are assumed to be independent and identically distributed with mean zero and finite variance σ 2 (ε it ~ i.i.d .(0, σ 2 )) , where i and t represent individual firm and time period, respectively. The Table 6 shows that the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=44.0321, p-value=0.0000). This first result confirms the nonlinearity of pledge ratio and yield return, but more originally show the presence of strong threshold effects determined by pledge ratio. It will be therefore, necessary in a second step, to determine the number of transition functions required to capture all the nonlinearity of the relationship. In our second test of no remaining nonlinearity, the null hypothesis is not rejected. Thus, our model needs only one transition function. In second step, we analyze the parameter estimates of the final PSTR models. The estimated model from our empirical results is represented as follows: Yit = µ i + 10.6246S − 0.6277 P + 15.6170MB + 0.7182 RM +. g (q it ;4103.4080,17.5071%)(10.0407S − 0.1732 P + 22.9873MB + 0.8695IR) + ε it ( 19 ). In Panel B from Table 6, we could see the regimes are distinguished by the different . .

(51) . regression slope, -0.6277 and -0.1732. In the first regime, where the pledge ratio is less than 17.5071%, the estimate of coefficient is -0.6277 which is significant at the 10% level and which indicates that the firms’ return decrease 0.6277 with and increase of 1% in pledge ratio. In second regime, where the pledge ratio is greater than 17.5071%, the estimate of coefficient is -0.1732 but not significant. These indicate that more the collateralized shares the less firm return, but after amounts, the collateralized shares are not related to firms’ return. In addition, the potential firms need money to invest more not only in first regime but second regime with positive significant coefficients 15.6167 and 22.9873 given the pledge ratio level. In the other word, the firms’ return increase 15.6167% with increase of 1% in ratio of market to book value before pledge ratio reach 17.5071%. But stepper slope (22.9873) after pledge ratio over 17.5071%. That indicates the potential company have the optimal pledge ratio approximate 17.5%. The Table 7 shows that the linearity hypothesis which between divergences between voting right and cash flow right and firms’ return is strongly rejected with Likelihood ratio test of linearity (F=22.3465, p-value=0.0000). This first result confirms the nonlinearity of divergences and yield return, but more originally show the presence of strong threshold effects determined by divergences. It will be therefore, necessary in a second step, to determine the number of transition functions required to capture all the nonlinearity of the relationship. In our second test of no remaining nonlinearity, the null hypothesis is not rejected. Thus, our model need only one transition function. In second step, we analyze the parameter estimates of the final PSTR models. The estimated model from our empirical results is represented as follows: Yit = µ i + 8.7247 S + 0.2001D + 27.0233MB + 0.6318RM + . g (qit ;350.2526,54.5743%)(10.2958S − 0.1608 D + 16.3831MB + 0.8093IR) + ε it . ( 20 ).

(52) . In Panel B from Table 7, we could see the regimes are distinguished by the different regression slope, 0.20005 and -0.1608 From the literature views, we may see the divergences between control right and cash flow right is not a good thing to a company, even some studies treat divergences between cash flow right and control right is equivalent to collateralized shares. But in our empirical results, we have the same indication but not significant.. 5.3 The relationship between size and firms’ return The relationship between firm size and performance has been the subject of many studies, some of which have provided conflicting results. A number of studies have indicated a positive relationship between firm size and performance (Hall and Weiss (1967), Scherer (1973) and Reingnum and Smith (1983)). However, Michael Porter (1985, 1998) offers arguments to suggest that the relationship between firm size and profitability may be non-linear. There are different attitudes about size to performance, therefore we set size as our threshold variable to run PSTR model as previous. Table 8 shows that the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=34.4517, p-value=0.0000), and there is only a threshold at 21.3133. We analyze the parameter estimates of the final PSTR models. The estimated model from our empirical results is represented as follows: Yit = µ i − 0.0328P + 10.8150S + 11.1286 MB + 0.2708IR + g (qit ;2.1849,21.3133)(−0.1983P + 10.0483S + 18.5885MB + 0.8522 IR) + ε it. ( 21 ). In Panel B from Table 8, we could see the regimes are distinguished by the different regression slope, 10.8150 and 10.0483. In the first regime, where the size is less than 21.3133, the estimate of coefficient is 10.8150 which is significant at the 5% . .

(53) . level and which indicates that the firms’ return increase 10.8150 with and increase of 1 in size. In second regime, where the size is greater than 21.3133, the estimate of coefficient is 10.0483 but not significant. These indicate that before the threshold, the greater the size the better firms’ performance due to scale economic (Hall and Weiss, 1967 and Scherer, 1973), but after amounts, the size is not a big deal. Furthermore, given two regime of size, the pledge ratio is negatively related to firms’ performance (-0.0328 and -0.1983 respectively) but not significant. In Table 9, we examine the size effect toward divergence between voting right and cash flow right and firms’ performance. We also find the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=21.4293, p-value=0.0000), and there is only a threshold at 21.4293. The first regime slope is 8.6040 significant at 10% level, indicates the more the size the more profit company earn, but in second regime slope still positive(9.5392) but not significant support by Michael Porter (1985 and 1998) offers arguments to suggest that the relationship between firm size and profitability may be non-linear. The estimated model from our empirical results is represented as follows: Yit = µ i + 0.3561D + 8.6040S + 12.1697 MB + 0.3291IR + g ( qit ;2.6395,21.4293) ( −0.1101D + 9.5392S + 18.5628MB + 0.8588 IR) + ε it. ( 22 ). Moreover, we find small firm’ s divergences are positively related firm’ s performance (0.3561), but larger firms are negatively related to firm’ s performance ( ), but not significant.. 5.4 Risk-adjusted result Theory (APT) Created in 1976 by Stephen Ross, this theory predicts a relationship between the returns of a portfolio and the returns of a single asset through a linear combination of many independent macro-economic variables therefore . .

(54) . risk-adjusted performance measures have been constructed in our study by beta as Rit / β . After taking Rit / β as our dependent variables, we re-run our model4. The results are showed by Table 10 and Table 11. First in pledge ratio (Table 10), we find the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=20.7997, p-value=0.0000), and there is only a threshold at 52.3469%. Unfortunately, there’ s no significant relationship between pledge ratio and risk-adjusted return in both regimes. The estimated model from our empirical result is represented as following: Yit = µi + 42 .5936 S − 0 .5680 P + 19 .4861 MB + 1.2708 RM + g (q it ;13 .5445 ,52 .3469 %) . (45.1247S − 0.6711P + 81.9048MB + 0.7188IR) + ε it. ( 23 ). Furthermore, we find market to book value and industry return yield positively significant effect with risk-adjusted return. Second in divergences (Table 11), we find the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=17.7668, p-value=0.0000), and there is a threshold at 128.4139%. Unfortunately, there’ s no significant relationship between divergences and risk-adjusted return in both regimes. The estimated model from our empirical result is represented as following: Yit = µ i + 23.9751S − 0.6259D + 31.4297MB + 0.5669RM + g (qit ;0.8076,128.4139%) (895864619049.5520S - 209656550259.639D − 135796827302.8990MB. + 15320283485.4180IR) + ε it. ( 24 ). With the same result we discuss above, the potential for development and industry return yield positively significant effect with risk-adjusted return in first regime.. 4. See Table 12 and Table 13, we also exam the absolute value of beta, we have the same result with original. . .

(55) . 5.5 The relationship between pledge ratio, divergences and excess return compared to market Currently, most studies use multi-factor models to estimate Jensen’ s alpha. One of the most frequently used specifications is a three-factor model of Fama and French (1993). The corresponding factor returns, RMRF, SMB (firm size: stock price times the number of shares), HML (book-to-market equity (the ratio of the book value of the firm’ s common stock to its market value, are calculated as the difference between the returns on small and big-stock portfolios and the returns on portfolios with high and low book-to-market equity, respectively. However, we replace return to excess return compared to market as Rm-Rit to capture the relationship between abnormal return and director’s collateralized shares. 5 The results are described by Table 14 and Table 15. First of all, we exam the relationship between abnormal return and pledge, find the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=33.1680, p-value=0.0000), also there is a threshold at 17.93048. But we do not find any significant relationship between pledge ratio and abnormal return in both regimes. The estimated model from our empirical result is represented as following: Yit = µ i + 0.7961S − 0.3362 P + 13.4422 MB + 0.3835RM + g (q it ;2482.8730,17.9305%) (0.3443S − 0.1072 P + 18.2956MB + 0.5483IR) + ε it. ( 25 ). With the similar result with risk-adjusted return the potential for development and industry return yield positively significant effect in regime 1. Also in regime 2, the industry return is still positively significant related to abnormal return with flatter. 5. See Table 16, Table 17, Table 18 and Table 19we also exam the size effect, but there’ s no big deal with collateralized shares and divergence between voting right and cash flow right. . .

(56) . slope (). Turns to divergence between voting right and control right, we find the linearity hypothesis is strongly rejected with Likelihood ratio test of linearity (F=14.3661, p-value=0.0000), also only one threshold at 53.5026. But we do not find any significant relationship between divergence and excess return compared to market. The estimated model from our empirical result is represented as following: Yit = µ i − 0.9873S + 0.2969 D + 21.6626MB + 0.2601RM + g (qit ;1029.8296,53.5026%) (0.3299 S − 0.0765 D + 13.7620 MB + 0.4831IR) + ε it. ( 26 ). With the similar result with risk-adjusted return the potential for development and industry return yield positively significant effect in regime 1. Also in regime 2, the industry return is still positively significant related to abnormal return with flatter slope (2.6374). But the market to book value is inversed with the greater divergence between voting right and cash flow right.. 5.6 Furthermore At the beginning of discussion of the relationship between divergence ratio and return, we set divergence ratio as cash flow right divided by voting right which indicates that high divergence ratio with low difference between cash flow right and voting right also low divergence ratio with high difference between cash flow right and voting right. Practically, it sounds tricky, therefore we set a new variable as 1-divergenc ratio which represents the bigger the value the higher divergences and smaller the value the lower divergences. We rerun the model to see the result is consistent with previous one. Thus Table 20, Table 21 and Table 22 illustrate the estimated result from our model which indicated the same result as we estimate before. But, in Table 20 we find high divergence with poor performance which is consistent . .

(57) . with La porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002). After estimating the threshold for pledge ratio and divergence ratio, the threshold values divide our data into N+1 regimes discuss the relationship within. Therefore we category pledge ratio, divergence ratio and size as high pledge ratio, low pledge ratio, high divergence ratio, low divergence ratio, larger firms, and small firms to see whether the outcomes come from summary statistic are consistent with our empirical result or not. In Table 23, we use 17.507% as criteria to divide data into two part; high and low areas. Panel A and B list the summary statistic for low pledge and high pledge ratio respectively. First of all, we find the companies with low pledge ratio perform better than high pledge ratio companies in return and industrial return. In other words, pledge ratio is toxics for firms’ performance. Also market to book value in low pledge ratio companies is larger than higher one which indicates that the low pledged ratio companies get more opportunities for developing than the other one. In Table 24, we use 54.5743% as criteria to divide data into two part; high divergence ratio and low divergence ratio. Here, the high divergence ratio indicate the small the difference between cash flow right and voting right, the low divergence ratio indicate the bigger the difference between cash flow right and voting right, Panel A and B list the summary statistic for low and high divergence ratio respectively. We find the low divergence ratio companies perform better than high divergence ratio companies which is not consistent with the past literature (La porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002) indicated the negative relationship between them). Also in Table 26, we find that the directors who work in larger firms averagely . .

(58) . collateralized their shares more than who work in small firms which indicate the bigger firm use collateralized shares to control company than small firms. However in Table 25 and Table 26 indicate that larger firms perform better than small firms is consistent with Reingnum and Smith (1983).. . .

(59) . Chapter 6 Conclusions The fundamental question addressed in this study is whether the pledge ratio or divergences between voting right and cash flow right will affect the company performance. The population of our research is domestic listed companies. The examination period is from 1997 to 2007. To control for individual heterogeneity, the panel smooth transition regression, PSTR purposed by González, Terasvirta and Dijk (2004) was conducted to test the nonlinearity relationship between firm performance and directors’ pledge ratio. The empirical results support the claim that the collateralized shares may have contributed to a lower return also the pledge is harmful to company performance in first regime of PSTR model. Thos results are also in line with previous researches (Chiou et al., 2002) which concern the linearity. In our model, we find out thresholds to separate regimes, which allow the non-linearity. The results indicate that non-linearity relationship between firm performance and directors’ pledge ratio does exist. The more shares the directors collateralized, the worst the firm perform when the pledge ratio is lower than 17.507%. When the pledge ratio exceeds the threshold value, it has no significant effect on firm performance. The results explain the reason that the empirical studies have no concrete conclusion on the relationship between directors’ collateralized shares and firms’ performance because of the nonlinearity relationship. From the size view, we find there’ s size effect (scale economic) in both divergences between voting right and cash flow right and pledge ratio in regime one (8.6040 and10.8150). These indicate that before the threshold, the greater the size the better firms’ performance due to scale economic (Hall and Weiss, 1967 and Scherer, 1973), but after amounts, the size is not a big deal offers arguments to suggest that the relationship between firm size and profitability may be non-linear supported by . .