Relationship vs Transaction Based Matching in the Japanese Corporate Bond Market

Transcription

1 Relationship vs Transaction Based Matching in the Japanese Corporate Bond Market C. R. McKenzie a and Sumiko Takaoka b a Faculty of Economics, Keio University, Mita, Minato-ku, Tokyo , Japan b Faculty of Economics, Seikei University, 3-3-1, Kichijoji Kitamachi, Musashino, Tokyo , Japan December 21, 2012 Abstract This paper provides a comprehensive analysis of relationship-based and transactionbased matching in the Japanese corporate bond market over the period by examining commissions, launch spreads, issue sizes, and matching behavior. Both relationshipbased and transaction-based matching are found to have power in explaining observed outcomes. Issuers re-match with the same underwriter when the difference between their reputations is small. If the underwriter is related to issuer s main bank, issuers have a higher probability of re-matching with the same underwriter. Although the impact of the main bank relationship weakens over time, the effects of underwriter and issuer reputations do not. JEL classification numbers: G20, G24, L14 Keywords: Corporate bonds, issuer, Japan, main bank, matching, reputation, underwriter The authors would like to express their thanks to Bill Greene who provided extensive assistance and advice with respect to the panel options in LIMDEP/NLOGIT. They would also like to thank Kian Teng Kwek, Yasuhiko Tanigawa and participants at the 2012 Japanese Economic Association Autumn Meeting and the Thirteenth International Convention of the East Asia Economic Association for their helpful comments, and Junko Tomino for her excellent research assistance. The first author also wishes to thank Sciences Po and the University of Bocconi for their hospitality while part of this paper was being written. Both authors gratefully acknowledge the financial assistance provided by the Japan Society for the Promotion of Science (JSPS) Grant in Aid for Scientific Research (B) No for a project on A Dynamic Analysis of Price Formation in Securities Markets.

2 In the issue market for corporate bonds, some issuers come to the market quite often and others appear only once. The same is true for underwriters, some underwriters may rarely underwrite corporate bonds, while other underwriters are regularly underwriting. For example, in the Japanese straight corporate bond market between 1994 and 2009, there were 553 issuers that issued an average of 9.6 bonds over the period, with the most frequent issuer, Tokyo Electric making 138 issues. Of the 553 issuers, 107 issuers only come to the market once over the sixteen year period. Over the same period of time, there were 53 active underwriters with the most active underwriter, Nomura, handling 1,149 issues and the least active underwriter handling only one. Issuers coming to market more than once allow us to look at how the underwriter they match with changes over time and the factors that influence this matching, for example, the reputations of underwriters and issuers, and the relationships between underwriters and issuers via the issuer s commercial bank. In an extremely innovative paper,? emphasize the importance of the current reputations of issuers and underwriters in determining how issuers and underwriters match in a securities market. Their theoretical analysis suggests the key factors determining which firm matches with which underwriter are the current reputations of the issuing firm and the underwriter. As a result, if an issuing firm (underwriter) can increase its current reputation compared to its reputation at the time of its previous issue, the issuing firm (underwriter) is more likely to match with an underwriter (issuer) that has a higher reputation than the one used for the previous issue.? provide strong evidence from an analysis of IPO and SEO issues of stock in the United States that higher quality issuing firms associate with higher ability underwriters, and lower quality issuing firms associate with lower ability underwriters. Since neither past issuer and underwriter behavior nor relationships between them are determinants of this matching, we term this transaction based matching. In contrast, relationship-based matching in this paper refers to matching of underwriters and issuers where the history of previous matching, in particular, matching with the underwriter associated with the firm s principal commercial bank, is important. This paper provides a comprehensive analysis of relationship-based and transaction-based matching in the Japanese corporate bond market over the period by examining commissions, launch spreads, issue sizes, and matching behavior. Theoretical analyses of mutual matching suggest that the joint surplus is larger when positive assortative matching occurs. The comprehensive analysis of commissions, spreads and issue sizes is designed to verify the validity of the hypotheses underlying the matching analysis. It is shown that the reputations of both issuer and underwriter are important especially in the launch spread and issue size. When the difference of the issuer s reputation and the reputation of the underwriter is small (that is, the matching is positive assortative), the commission is smaller and issue size is bigger. While the commission is a transfer from issuer to underwriter, and does not have an impact on the joint surplus, the matching affects significantly the issue size which is an important component of joint surplus. In this analysis, we investigate the roles of issuer and underwriter reputations and bankissuer relationships in explaining the observed matching of issuers and underwriters in the 1

3 straight corporate bond market in Japan. By examining the validity of the predictions made by? in this market, this paper is able to demonstrate? s (2005) usefulness in explaining issuerunderwriter outcomes for a different security, corporate bonds, and in a different country, Japan. In addition, the examination of commissions, launch spreads, and issue sizes indicates that these outcomes are affected to varying degrees by underwriter reputation and the degree of matching between the underwriter and the issuer. However, there is also reason to believe that relationship underwriting may influence the matching of issuers and underwriters. The banking literature has discussed both transaction lending as well as relationship lending, where relationship lending evolves because banks invest in relationship-specific information collection and evaluate the profitability of these investments through multiple investments with the same customer (see, for example,?). In the underwriting of securities, if investing in the collection of private information on issuers is valuable and this private information has value for a period significantly longer than the issuing period, then we could expect relationship underwriting to occur. When the securities underwriter also is a commercial bank or related to a commercial bank, the opportunities for multiple interactions between the issuer and the underwriter are even greater, and the likelihood of relationship underwriting is even stronger. Using data on lending and public securities offering,? find that in the United States concurrent issuers receive a lower underwriter fee for the equity offering and a discounted yield spread on the concurrent loan, and combining lending with underwriting is an important factor in determining issuer-underwriter pairings. The recent incisive study by? that uses the sudden collapse of Lehman Brothers as a natural experiment reveals that even in the United States the relationship between the underwriter and issuing firm is of significant value. Many firms in Japan have a long-term and relatively stable relationship with a particular bank that is called a main bank relationship (see? for a recent discussion). There are many aspects to this relationship including lending, shareholdings, and settlement accounts, but this firm-bank relationship in Japan has been much stronger than the corresponding relationships in the United States (see?). After banks were allowed to enter the Japanese corporate bond underwriting service through subsidiary securities companies in 1994, corporate bond underwriting services are provided by two types of securities companies: those which are bank subsidiaries, and those securities companies that are independent of banks (see? for a theoretical model suggesting that banks and investment houses can coexist). This mixture of securities company types together with the existence of strong and stable bank-firm relationships suggests the matching of issuing firms and underwriters in Japan may be quite different from what is reported by? for the United States, namely, transaction based matching where firm-securities company transactions are one-off events, and the matching of firms and securities companies is determined principally by their reputations. In contrast to? s (2005) focus on the matching of underwriters and issuing firms, much of the pre-existing research focuses on the issuer s choice of underwriter. For example, in the United States this choice has been analyzed mainly for issues in the equity market (for example,??, and?). Some existing research on an issuing firm s choice of underwriter for 2

4 straight corporate bond issues in Japan focuses on the choice of the underwriter type. For example,? analyze whether or not the issuing firms have a propensity to choose the underwriter which is a bank subsidiary, and whether or not an underwriter is the issuing firm s main bank securities subsidiary (?).? investigates which particular underwriter is chosen as the issuing firm s underwriter. While these papers constitute static analyzes of underwriter choice at one point of time, namely, the time of an issue,? examine the switching of underwriters between the initial public issue of a corporate bond and the second public issue of a corporate bond. This paper examines how three factors, the reputation of the issuing firms, the reputation of the underwriters, and the relationship between an issuing firm and its main bank, affect the matching of firms and underwriters over time in the Japanese market for straight corporate bonds. All issues of straight corporate bonds over the period from 1994 to 2009 are analyzed, so that our analysis contains many repeated issues by the same issuers. In addition to using reputation measures used in?, we also add some original variables: the change in Tobin s q and the change in an issuer s rating. Firms with high values of Tobin q are expected to be better investment opportunities and higher growth potential, so Tobin s q is used here as a measure of the issuing firm s reputation for future profitability. Reputational effects are found to be important in explaining issuer-underwriter matching. To comprehensively investigate the role of reputation of both issuer and underwriter and their relationship in the process of the bond issue, we analyze matching, underwriting commissions, launch spreads, and issue size. Throughout the process of the bond issue, the reputations of issuing firm and underwriter are significant factors. The results for launch spreads suggest that reputations are not only important between the issuing firm and underwriter, but also for market participants. An issuing firm matching with an underwriter that has a good reputation can issue a bond with a lower spread in the primary market. During the sample period of this paper, , the so-called main bank system has been viewed by some as having been an impediment to the growth of the Japanese economy (see? and?). These close firm-bank relationships in the heyday of the main bank system have gradually changed, as the financial market evolves from bank-centered economy to marketcentered economy. For example,? provide empirical evidence that increased capital market competition in Japan is associated with reduced relationship lending, especially for long-term loans. Here, we investigate how the relationship between the issuing firm and underwriter which is a subsidiary of issuing firm s main bank has changed over time by dividing the sample into the two halves. The results from these sub-samples provide some evidence that the impact of the main bank relationship has weakened over time, but that it still has a significant influence on the matching of issuers and underwriters in both sub-samples. Since the firm-bank relationship is significant explaining the matching of issuers and underwriters even after taking account the reputation effects of the issuer and the underwriter, not all the matching in our dataset can be explained by? s (2005) transaction based matching. However, while relationship based matching weakens in the second half period of the sample, reputation effects are significant, and do not weaken over time. As a robustness check of our results, we model the underwriter-issuer matching together 3

5 with the quality of the chosen underwriter using a bivariate probit model. It is natural to think that the issuing firm is interested in the quality of underwriter at the time of the issue of security, so that this decision and the matching decision closely are connected. The estimation results strongly favor modeling both decisions together. The remainder of the paper proceeds as follows. Section I discusses the hypotheses to be tested. Section II describes the model to be estimated. Details of the data used in this analysis are explained in section III. Descriptive statistics and the results of estimating models for commissions, launch spreads, issue sizes and the switching of underwriters are discussed in section IV. Section V presents a conclusion. I Hypotheses In a static theoretical framework of matching behavior,? demonstrate that issuers and underwriters will match in a way that ensures that firm quality/reputation and underwriter ability/reputation are positively related. Even though the observed outcomes at each point in time are unconnected, it follows that changes in the reputation of underwriters and the issuing firm between successive issues will lead to the issuing firm matching with different underwriters.? argue that if the reputation of an issuer improves sufficiently after the most recent past security issue, then the issuer will match with a higher quality underwriter. Similarly, if the quality of the underwriter chosen for the most recent past security issue improves sufficiently, then the issuer will be forced to match with a lower quality underwriter.? highlight that the observed matching outcome does not result solely from the issuer s choice of underwriter, but as a result of the mutual choices made by issuers and underwriters. Several other hypotheses have been suggested in the literature to explain why an issuing firm might be observed to use a different underwriter between successive issues of securities 1. In addition to? s (2005) transaction based matching which emphasizes only the current reputations of the underwriter and issuer, the hypothesis emphasizing the existing relationship between an issuing firm and its main bank is investigated here. There are several possible ways that the existence of a main bank might influence an issuing firm s choice of underwriter. In the course of its continuing relationship with the issuing firm, a main bank is likely to accumulate 1 There are four other hypotheses that have been identified in the literature as explaining underwriter switching but they are not the focus of this paper. The first hypothesis is? s (2001) graduation effect hypothesis which predicts that issuing firms will tend to switch underwriters between IPO and SEO issues when they can obtain the services of an underwriter with a better reputation for the follow-on offering. A second hypothesis suggests that as the time between successive issues increases, the probability of switching also increases. This hypothesis assumes that an underwriter obtains private information on the issuing firm in the course of underwriting a bond issue that is valuable after the bond issue underwriting is complete, and that the value of this private information depreciates over time (see?). A third hypothesis suggests that the underwriter s past performance (for example, the size of mispricing) for the most recent past issue will affect the likelihood of that underwriter being chosen again (see? for some survey results). Finally, a fourth hypothesis is that issuing firms, particularly firms with high ratings, may seek to avoid the lock-in effects that result from repeatedly using the same underwriter. 4

6 private information on the issuing firm through loans or other transactions which might be useful when the issuing firm wishes to issue a new security. It should be noted that in Japan banks are not permitted to directly underwrite securities issues of other firms, but must do so through a securities company subsidiary. Although firewall regulations may restrict the flow of this information between the parent and the subsidiary, the extent to which private information collected by the bank itself may be used by the securities company subsidiary in the underwriting process will depend on how strictly the firewall regulations are enforced. Alternatively, a main bank may be able to apply pressure on the issuing firm to choose the main bank s securities subsidiary as the lead underwriter (for anecdotal evidence in Japan see?). When the Japanese financial market was bank-centered, there are strong reasons to believe that the matching of underwriters and issues based solely on the reputations of issuers and underwriters will not explain all observed outcomes. However, the recent shift to a more marketbased financial market in Japan is obvious, so it is likely that issuing firms and underwriters now put even greater emphasis on their market reputations. This market evolution can be expected to lead to more transaction-based matching over time. Even though the Japanese financial market is evolving from being bank-centered to being market-centered, bank-firm relationships, especially main bank-firm relationships, are still important and prominent in the Japanese economy (see?). We expect a significant effect of the existence of a main bank-firm relationship on underwriting outcomes. That is, once an issuing firm chooses an underwriter which is the securities subsidiary owned by the issuer s main bank, the issuer and underwriter have a higher probability of staying matched than when the previous issues was handled by another underwriter (see?). A great deal of financial deregulation, liberalization and re-organization occurred during the sample period , so the effect of this bank-firm relationship may not be constant over time. As a result, it is important to check for the stability of the relationship based underwriting outcomes and transaction based underwriting outcomes. While the extent of the bank-firm relationship may change depending on the economic situation or each firm s financial condition, this does not necessarily mean that the transaction-based matching is affected by the firm-bank relationship. The matching model also has implications for launch spreads, commissions, the size of issues and the observed reputations of the underwriters used for successive issues. One key assumption of Propositions II-IV in? is that the joint surplus created by the matching of underwriters and issuers is increasing in the ability/reputation of the underwriter. The joint surplus here refers to the value created by the issuing process net of all the costs that incurred by both the issuer and the underwriter. We cannot observe the joint surplus for any matching between the underwriter and the issuer, but two key factors influencing the joint surplus will be the launch spread on the issued bond and the size of the bond issue itself, as these two multiplied together give us an estimate of the direct cost of the bond issue to the issuer. One way that an increase in an underwriter s reputation can increase the joint surplus is through the launch spread, if an underwriter with a higher reputation is able to arrange the issuing of a firm s bonds at lower spreads either because it has a more extensive distribution network or because it can make use of a certification effect. The size of the issue itself is 5

7 also likely to influence the joint surplus. Again due to more extensive distribution networks or certification effects, underwriters with higher reputations may be able to arrange issues of larger size. As? note, the total underwriting commission paid by the firm to the underwriter does not have an impact on the size of the joint surplus, but the monetary allocation of the surplus to the underwriter likely to be non-decreasing in the reputation of the underwriter, and the percentage allocation of the surplus is likely to be a declining function of the underwriter s reputation 2. Since the monetary value of the commission payment is the percentage commission multiplied by the issue size,? s (2005) argument requires that the issue size be non-decreasing in reputation of the underwriter. The degree of matching between the underwriter and the issuing firm will affect the surplus, and the potential channels for this are by changing the size of the launch spread and the issue size. When the matching of the underwriter and the issuer improves, we expect launch spreads to decline and the size of issues to increase. The matching theory developed by? implies that issuing firms with higher reputations are matched with underwriters of higher reputations. If we observe that the issuing firm s reputation has increased between successive issues, then we expect the reputation of the actual underwriter being used by the issuing firm has also increased. When the matching of the underwriter and the issuer improves, we expect the underwriter commission in percentage terms to decline. II Models A. Static Models In order to provide a more comprehensive analysis of? s (2005) matching model, we also estimate models for the commission, the launch spread, and the size of the issue. In the literature on the factors explaining commissions and launch spreads associated with straight corporate bonds, it is common to include the ratings of the issuing firm where higher ratings reduce both the commission and launch spread. Underwriter reputations are not usually included as explanatory variables, but? s (2005) theoretical model of issuer and underwriter matching assumes that increases in firm quality/reputation and underwriter ability/reputation both increase the joint surplus generated by a matching, and the sharing of the surplus between them will be affected by both variables. We assume that the joint surplus is affected by the launch spread and the size of the issue, and the sharing of the surplus is related to the commission paid to the underwriter. For commissions, we estimate following equation: Commission i,j = a Z1 i,j + ϵ1 i,j, (1) 2 Part of the surplus may be allocated to the underwriter through underpricing (overpricing) which would also affect the size of the joint surplus. For Japan, there is evidence that there is significant overpricing of bond issues: see?. 6

8 where Commission is the underwriting commission paid by firm i for issue j, Z1 is a vector of explanatory variables, a is a vector of parameters, and ϵ1 i,j is an error term. Following the literature on underwriter commissions, variables in Z1 include the size of the issue itself (ln(amount)), dummy variables to account for the maturity of the issue, dummy variables to account for the rating of the issuer, a dummy variable to account for whether the underwriter is the subsidiary of bank, and industry dummies (see, for example,? and?. To take account of the effects suggested by?, we also include a measure of the underwriter s reputation and a measure of the degree of matching, Absolute difference of the issuer s reputation and the reputation of the current underwriter. It is expected that improvements in an underwriter s reputation will lead to an increase in the commissions they receive, and that better matching will reduce the commission. For the launch spread, we estimate following equation: LaunchSpread i,j = b Z2 i,j + ϵ2 i,j, (2) where Launch Spread is the launch spread for issue j made by firm i, Z2 is a vector of explanatory variables, b is a vector of parameters, and ϵ1 i,j is an error term. Following the literature on launch spreads, the variables in Z2 include those used in Z1 and the age of the issuer, the Tobin-q of the issuer, the inflation rate and the change in the index of industrial production (see, for example,? and?). In this case, improvements in the underwriter s reputation are expected to reduce the launch spread, and as the matching between the underwriter and the issuer improves, we expect that the launch spread will fall. For the size of the issue, we estimate the following equation: ln(amount) i,j = c Z3 i,j + ϵ3 i,j, (3) where ln(amount) is the logarithm of the size of firm i s bond issue j, Z3 is a vector of explanatory variables, and c is a vector of parameters, and ϵ3 i,j is an error term. The variables in Z3 include the rating of the issuing firm, the reputation of the underwriter, and the degree of matching. As the rating of the issuing firm or the reputation of the underwriter improves, we expect the firm to make larger issues. When the matching of the issuer and underwriter improves, we also expect the firm to be able to raise more funds. B. Dynamic Models In order to test the hypotheses explained in section I, we follow the existing literature by modeling whether or not the issuer matches with a different underwriter for successive issues (or switches from one underwriter to another underwriter). To start with, the following model is assumed to explain the observed switching of the lead underwriter employed by an issuing firm between the most recent prior bond issue and the current bond issue: Switch i,j = d Z4 i,j + ϵ4 i,j. (4) 7

9 Here, Switch i,j is a latent variable. In the literature that emphasizes the matching of issuers and underwriters (for example,?), Switch i,j could be interpreted as being the difference in the net benefits of a bond issued by issuer i when a different lead underwriter is used for the jth issue compared to using the same lead underwriter as was used for the j-1th issue. In equation (??), Z4 i,j is a vector of explanatory variables related to the jth issue by the issuer i, and whose first element is assumed to be unity, d is a vector of unknown parameters, and ϵ4 i,j is a random variable which is assumed to be distributed according to a standard normal distribution. It is assumed that Switch i,j = { 1 Switch i,j > 0, 0 Switch i,j 0, (5) where Switch i,j is an observable variable taking the value unity if different lead underwriters are used by the issuer i for the jth issue and j-1th issue, and zero otherwise 3. In most of the existing literature, when Switch i,j = 1 (= 0), the issuing firm is referred to as a switcher (stayer). The combination of (??) and (??) means that a probit model can be used to explain variations in Switch 4 i,j. The variables used in Z4 are defined as follows. Current reputation of the previous underwriter is the reputation of the lead underwriter used by firm i for j-1th issue measured for the year in which the jth issue was made 5. Change in reputation of the previous underwriter is the change in the reputation of the lead underwriter used by firm i for j-1th issue between the years in which the j-1th issue and the jth issue were made. Difference of the issuer s reputation and the current reputation of the previous underwriter is defined as the difference between firm i s reputation for the year in which the jth issue was made and the reputation of the underwriter used by issuer i for j-1th issue measured for the year in which the jth issue was made. Main Bank UW dummy is a 0-1 dummy variable taking the value one if the lead underwriter for the j-1th issue by issuer i is a securities subsidiary of the firm s main bank, and zero otherwise. 3 Switch i,j is defined only when it is physically possible for the issuer i to choose the lead underwriter chosen for the j-1th issue as the lead underwriter of the jth issue. 4 As noted in?, there are numerous cases of an issuer making multiple issues on the same day, that is, issuing several bonds of different maturities on the same day. For the purpose of defining Switch i,j, these multiple issues were dealt with as follows. Suppose the current issue is a single issue, and the most recent prior issue is a multiple issue. In this case, a switch is deemed to have occurred if the lead underwriter for the current issue differs from all the lead underwriters associated with the multiple issue. In all other cases, the issuing firm is deemed to have stayed with the same underwriter. Suppose the current issue is a multiple issue, and the most recent prior issue is a single issue. Each of the maturities in the multiple issue is treated separately. If the lead underwriter for one of the maturities is the same as the lead underwriter for the single issue, a switch is deemed not to have occurred for the bond with that maturity. Suppose the current issue is a multiple issue, and the most recent prior issue is a multiple issue. Each of the maturities in the current multiple issue is treated separately. If the lead underwriter for one of the maturities of the current multiple issue is the same as the lead underwriter for any one of the issues in the previous multiple issue, a switch is deemed not to have occurred for that current issue. 5 As noted in footnote??, there are many cases of an issuer making multiple bond issues on the same day. If the previous issue is a multiple issue, the reputation of the previous underwriter is computed as the average of the reputations of lead underwriters for each bond in the multiple issue. 8

10 Rating is the difference between the numerical rating of the j-1th issue and the jth issue. Tobin-q is the change in the value of Tobin-q of firm i between the j-1th issue and the jth issue. Industry dummies are also included in equation (??). Following?, it is expected that an improvement in the Current reputation of the previous underwriter will reduce the probability of a switch, and a Change in the reputation of the previous underwriter will increase the probability of a switch. Following the arguments of?, it is expected that an increase in the Difference of the issuer s reputation and the current reputation of the previous underwriter will increase the probability of a switch. The arguments in section I suggest that the coefficient on the Main Bank UW dummy will be negative, so that using the main bank s securities subsidiary as the lead underwriter for the most recent past issue will increase the probability that the same underwriter is chosen the next time round. Rating and Tobin-q are included as variables measuring different dimensions of the change in the firm s reputation. Improvements in the firm s reputation over time are expected to lead to an increased probability of switching occurring. Finally, we examine how the reputation of the underwriter that a firm matches with changes between successive issues by assuming that changes in the reputation of the lead underwriter that is used by firm i for jth issue and j-1th issue can be modeled using the following model Improve i,j = e Z5 i,j + ϵ5 i,j, (6) where Improve i,j is a latent variable, Z5 i,j is a vector of explanatory variables related to the jth issue by the issuer i and whose first element is assumed to be unity, e is a vector of unknown parameters, and ϵ5 i,j is a random variable which is assumed to be distributed according to a standard normal distribution. It is assumed that Improve i,j = { 1 Improve i,j > 0, 0 Improve i,j 0, (7) where Improve i,j is an observable variable taking the value unity, if the reputation of the underwriter for the jth issue by issuer i is higher or the same as the reputation of underwriter for the j-1th issue, and zero otherwise 6. According to?, the key explanatory variable in Z5 is the change in the reputation of the issuing firm. If the issuing firm s reputation improves, then the probability that the issuing firm matches with an underwriter with a better reputation increases. To examine the robustness of the results, we also add the reputation of the underwriter for the j- 1th issue at the time of that issue. It is expected that the higher this initial reputation is, the lower probability of an improvement. We also add the actual change of the reputation of the previous underwriter as an explanatory variable. This variable is potentially endogenous, because if Switch = 0, then negative (positive) values of this variable correspond to Improve = 0 (= 1). In the course of issuer and underwriter matching, the issuer cares about, and chooses the quality of underwriting service determined by the underwriter s reputation. A natural extension 6 It is worth noting that even if the same underwriter matches with the issuer, Improve i,j can take the value one provided that the reputation of the underwriter has not worsened. 9

11 of the issuer and underwriter matching model is the two-equation model containing a switching equation and quality of underwriter choice equation 7. The essential motivation for the bivariate model is that the two decisions might be connected, and there are efficiency gains to be obtained using a bivariate model if the error terms for the two models are correlated. The model for a joint determination of two variables (Switch i,j and Improve i,j) is assumed to be Switch i,j = d Z4 i,j + ϵ4 i,j, ϵ5 i,j Switch i,j =1 if Switch i,j > 0, 0 otherwise, Improve i,j = e Z5 i,j + ϵ5 i,j, Improve i,j =1 if Improve i,j > 0, 0 otherwise, (8) ( ) [( ) ( )] ϵ4i,j 0 1 ρ Z4i,j, Z5 i,j N,, 0 ρ 1 where (ϵ4 i,j, ϵ5 i,j ) have a bivariate normal distribution with means (0,0), variances (1,1) and correlation ρ. In a similar way to the single equation probit model, in order to identify the parameters of both equations, the variances of both errors are standardized at unity (V (ϵ4 i,j ) = V (ϵ5 i,j ) = 1). The set of explanatory variables in each equation is same as in equations (??) and (??), respectively. Define Ψ 2 (A, B, ρ) as being the cumulative distribution of two variables x and y that are distributed according to a bivariate standard normal distribution, ρ is the correlation coefficient between x and y, and Ψ 2 (A, B, ρ) = P r(x < A and y < B). If P ijst is defined as the joint probability of the event Switch i,j = s and Improve i,j = t (s = 0, 1, t = 0, 1), then the probabilities for the four possible outcomes for the model specified in equation (??) are given as follows: P ij11 = Ψ 2 (d Z4 i,j, e Z5 i,j, ρ), (9) P ij10 = Ψ(d Z4 i,j ) P ij11, (10) P ij01 = Ψ(e Z5 i,j ) P ij11, (11) P ij00 = 1 P ij11 P ij10 P ij01, (12) where Ψ is the cumulative distribution function of the standard normal distribution. As can be seen from equations (??)-(??), each of the four probabilities depends on the explanatory variables in both equations. There are cases where the issuing firm re-match with the same underwriter whose reputation changes from at the time of previous time. Samples of non-switchers are also classified into Improve and not Improve. The bivariate probit model reflects the issuing firm s decision that it re-matches with the same underwriter since the reputation of previous underwriter increases 8. 7 The existing literature contains suggestions that the reputation and quality of underwriter are correlated (see, for example,?). 8 Another possible approach here is to use a nested logit model where at the first stage the choice is whether to switch or not, and the second stage choice is whether the reputation of the previous underwriter improves or not. In theory, the two stages could be reversed. 10

12 The combinations of equations (??) and (??), and (??) and (??) mean that for continuous explanatory variables, the marginal effect of an explanatory variable can be computed as: δp r(switch i,j = 1)/δZ4 i,j = ϕ(d Z4 i,j )d, (13) δp r(improve i,j = 1)/δZ5 i,j = ϕ(e Z5 i,j )e, (14) where ϕ(.) is the probability density function of a standard normal distribution. For the bivariate model, we have to first specify the outcome variable of interest. Following?, we focus attention on the following conditional expectation, E{Switch i,j Improve i,j = 1} =Pr{Switch i,j Improve i,j = 1}. For the model in (??), this expectation can be written as: E{Switch i,j Improve i,j = 1} = Ψ 2 (d Z4 i,j, e Z5 i,j, ρ)/ψ(d Z4 i,j ). (15) From equation (??), it is easy to see that unlike the marginal effects defined in (??) and (??) which only depend on Z4 i,j and Z5 i,j, respectively, the explanatory variables of both equations in (??), Z4 i,j and Z5 i,j, appear in this conditional expectation. If the marginal effects are defined as being how a one unit change in an explanatory variable affects the conditional expectation of Switch i,j defined in (??), then the marginal effect of Z4 i,j and Z5 i,j for continuous explanatory variables are given, respectively, by δe{switch i,j Improve i,j = 1}/δZ4 i,j, (16) δe{switch i,j Improve i,j = 1}/δZ5 i,j. (17) Given the difficulty in interpreting the estimated coefficients of the bivariate probit model, we report estimates of the marginal effects based on equations (??) and (??). III Data The sample period used in this paper starts on February 25th, 1994, and ends on December 31st, The starting date is when the first bank subsidiary underwriting of a straight bond issue took place (see?). The dataset used in this paper is constructed from various data sources, and data collected manually. Data on straight corporate bonds issued publicly by individual firms are taken from the Thomson One Investment Banking database. Although our dataset includes bonds issued by unlisted companies and financial institutions, these issues are excluded from our analysis because Tobin s q cannot be computed for these issuing firms. Electric power companies are extremely heavy users of the corporate bond market, and may be argued to behave differently from other firms. The analysis reported in this paper includes electric power companies, but there is no substantive change in the estimation results reported if they are excluded. 11

13 This dataset includes details of the issue itself, the lead underwriter 9 and the underwriting syndicate for each bond issue in the sample period. Information about the relationship between the issuing firm and financial institutions comes from the Nikkei Corporate Quarterly (Nikkei Kaisha Joho). Over the course of the sample period, several new entries, mergers, acquisitions and bankruptcies occurred in both the securities and banking industries (see Figure 1 in?). Information about these mergers and acquisitions, new market entry and bankruptcies is collected manually from the Nikkei newspaper data base, Nikkei Telecom 21. We treat issues underwritten by merged or acquired securities firms as follows. For example, Tokyo Mitsubishi Securities acquired Mitsubishi Trust Securities on July 1st, Tokyo Mitsubishi Securities is treated as the successor of Mitsubishi Trust Securities so that if a firm used Tokyo Mitsubishi Securities as the lead underwriter for its j-1th bond issue, and Tokyo Mitsubishi Securities as the lead underwriter for its jth bond issue, we do not consider this to be a switch of the lead underwriter. If instead, the firm chose Nomura Securities as the lead underwriter for the jth bond issue, this choice would be counted as a change of the lead underwriter. A key variable in this paper is an underwriter s reputation. The reputation of the lead underwriter used by firm i for issue j (issued in year t), Underwriter reputation, is computed as the underwriter s percentile rank of the total amount of straight corporate bond issues it was the lead underwriter for in the three year period (t-2,t-1,t). Current reputation of the previous underwriter is the reputation of the lead underwriter of the previous issue at the time of the current issue. Change in reputation of the previous underwriter is the difference of the reputation of the previous lead underwriter for the previous issuer at the time of the previous issue and at the time of the issue. Reputation of the previous underwriter at the time of the previous issue is the reputation of the lead underwriter at the time of that issue. The issuer s reputation in the issue year is computed as the issuer s percentile rank of the total amount of straight corporate bonds issued in that year. Change in reputation of issuer is the difference between the issuer s reputation at the time of the previous and current issues. The difference of the issuer s reputation at the time of the current issue and the previous lead underwriter s reputation at the time of the current issue is measured as Difference of the issuer s reputation and the current reputation of the previous underwriter. To take account of the matching effects suggested by?, we also include a measure of the underwriter s reputation and a measure of the degree of matching, Absolute difference of the issuer s reputation and the reputation of the current underwriter is the absolute value of the difference of firm i s reputation and the reputation of the lead underwriter for issue j, both measured at the time of issue j. In addition to the percentile rank, more information about the issuing firm such as the firm s rating is used as issuer s reputation measures. AAA rating is a 0-1 dummy variable 9 There are of course issues when there is more than one lead underwriter. In these cases, we choose the lead underwriter that is listed in the Thomson database as being the bookrunner among the listed underwriters for the particular issue. The bookrunner is the underwriter that receives the issuing mandate from the issuer, and is consulted by the issuer regarding the timing, size and pricing of the issue, engages in ex ante marketing with investors, and sets the levels for demand forecasts. 12

14 taking the value unity if the issuing firm s rating is AAA, and zero otherwise; AA rating is a 0-1 dummy variable taking the value unity if the issuing firm s rating is AA+, AA or AA-, and zero otherwise; A rating is a 0-1 dummy variable taking the value unity if the issuing firm s rating is A+, A or A-, and zero otherwise; and BBB rating is a 0-1 dummy variable taking the value unity if the issuing firm s rating is BBB+, BBB or BBB-, and zero otherwise. In the regression analysis, AAA rated bonds are used as the base group. Each of the ratings is assigned a numerical value between one and ten, with AAA (BBB-) being assigned the lowest (highest) value one (ten). Rating is the difference between the numerical rating of the issuing firm s previous and current issues. A positive value of Rating means that the issuing firm s rating has worsened between the j-1th issue and the jth issue. Tobin-q is the value of Tobin-q for the issuing firm at the time of the issue, and Tobin-q is the change in the value of Tobin-q between the previous and current issues. Age is the time in years since the issuing firm was established. In order to measure the relationship between an issuing firm and its main bank, we created Main Bank Underwriter dummy which takes the value unity if the previous bond issue is underwritten by the underwriter that is a subsidiary of its main bank. A firm s main bank is defined as the private financial institution (excluding insurance companies) with the largest shareholding in the issuing firm. A firm is deemed not to have a main bank if there are no private financial institutions (excluding insurance companies) in the list of top shareholders. Shares held by bank is the change in the shares (%) of the issuing firm held by the bank with the top shareholding between the previous and current issues. Information about characteristics of the particular corporate bond issue is also in the estimated models. The variable ln(amount) is the logarithm of the size of each bond issue. Commission is the underwriting commission paid for each issue. Launch Spread is the launch spread. Short-term maturity dummy is a 0-1 dummy variable taking the value unity if the bond s maturity is less than five years. Long-term maturity dummy is a 0-1 dummy variable taking the value unity if the bond s maturity is more than 15 years. ln(size) which is the logarithm of the total value of issues in the straight corporate bond market in the month of the bond issue is included to capture the market trends. The existing literature for Japan points out that underwriting commissions on corporate bonds are lower when the lead underwriter is a bank owned subsidiary?. To account for this, a dummy variable Underwriter type taking the value unity when the lead underwriter is a bank owned subsidiary, and zero otherwise is created. Data on the consumer price index and the index of industrial production were obtained from the Nikkei NEEDS Macroeconomic Data Base. For the purpose of creating the industry dummies, firms were assigned to thirty two industries using the industry classification in the Toyo Keizai s Corporate Quarterly Handbook (Kaisha Shikiho). 13

15 IV Estimation Results A. Descriptive Statistics All the results in this paper were estimated using NLOGIT Version 5 (May 1st, 2012, see?). Table?? presents details of the descriptive statistics for all the variables used in the regression analysis for the full sample. First, for the full sample 66% of issuing firms switch their underwriters between issues in the fifteen year period being examined, while 58% of issuing firms are able to match with an underwriter whose reputation is higher than or equal to the reputation of the previous underwriter employed by the issuer. An underwriter that is a subsidiary of issuer s main bank is employed in 19% of all the bond issues. Table 1 also reports the mean value for each variable for the sub-samples of when issuers switch their underwriters (Switch i,j = 1) and stay with the previous underwriter they used (Switch i,j = 0), and and a p-value for a t-test of the null hypothesis that the mean of the switchers and stayers groups are the same. No significant difference is observed in the commission and the launch spread between switchers and stayers. There is a tendency for issuing firms with a high rating to switch underwriters more frequently. Another measure of the change in the issuer s reputation, Tobin-q, indicates that issuing firms are more likely to stay with the same underwriter when Tobin-q changes for the worse. The current reputation of the previous underwriter is significantly higher for stayers. When issuers employed the underwriter that is a subsidiary of its main bank, they tend to rematch with the same underwriter. These simple statistics suggest that behaviors of switchers and stayers may be different in their response to reputational changes and their main bank relationship. The mean value and the p-value are also reported for another two sub-samples: when the reputation of the current underwriter is higher than or equal to the reputation of the previous underwriter (Improve i,j = 1) and when the reputation of the current underwriter is lower than the reputation of the previous underwriter (Improve i,j = 0). The issue size appears bigger when the reputation of the current underwriter is higher than or equal to the reputation of the previous underwriter. The proportion of issuers with an AAA rating or AA rating groups tend to be significantly higher for when the reputation of the current underwriter is higher. However, the tendency is not as clear as in the category of switchers and stayers. When Tobin-q changes for the worse, issuing firms are more likely to match with an underwriter whose reputation is lower than the reputation of the underwriter used for the previous issue. The reputation of issuer measured using the firm s percentile rank of bond issues is significantly higher in the Improvement group than in Decline group. The descriptive statistics suggest that there is no significant difference in the commission and the launch spread even if the issuers switch underwriters or choose the underwriter with better quality. However, the issuer s reputation when measured using the firm s rating, its percentile rank of bond issues and its Tobin-q shows significant difference between Switchers and Stayers, and the Improvement and Decline cases. A cross tabulation of the issuing firm s switching behavior and the change in the reputation 14

16 of the lead underwriter handling the issue is provided in Panel A of Table??. In the case of Switch (2,634 issues), 1,312 issues of the switchers employ an underwriter whose reputation is higher than or equal to the reputation of the underwriter that handled the previous issues. Nearly the same number of switchers (1,322 cases) employ an underwriter whose reputation is lower than the reputation of the underwriter. That is, the proportions of Improve and Decline are nearly same. On the other hand, in the case of Stay (1,358 cases), the number of Improve cases is much higher than the number of Decline cases(965 cases and 393 cases, respectively). Panel B shows a cross tabulation of the change in the issuer s reputation and the change in the reputation of the lead underwriter handling the issue. When the issuer s reputation is higher than or equal to the issuer s reputation for the previous issue (2,983 issues), Improve (1,758 issues) cases dominate. In contrast, when the issuer s reputation is lower than the issuer s reputation for the previous issue (1,009 issues), there is no substantial difference between the Improve (519 issues) and Decline (490 issues) cases. The results of simple descriptive statistics suggest that there is a relationship between the reputations of issuer and underwriter and matching behavior. Also a relationship between the reputations of issuer and underwriter and the quality of the underwriter chosen is suggested. B. Commissions, Spreads and Issue Size Before estimating any models of switching or matching behavior, we first investigate whether there is any evidence to support the assumptions made by? about the joint returns from matching and the how these returns are split between the underwriter and issuer. We do this by estimating standard equations for commissions and launch spreads (see?) and adding our measure of underwriter reputation, and also by estimating a simple model of issue size that includes both the reputation of the issuer and underwriter. We add the underwriter reputation and absolute difference of the issuer s reputation and the reputation of the current underwriter in the standard commission equation. Table?? shows the results of estimating a standard commission equation with underwriter reputation included. The effects of the rating variables are consistent with that of existing literature, namely, that the commission paid by the issuing firm to the underwriter significantly increases when the issuing firm s rating worsens. When the underwriter is a bank owned subsidiary, its commission is lower (a finding consistent with?). Thus, Table?? replicate the standard results in the literature regarding the issuer s reputation and underwriter type. While the underwriter s reputation does not significantly affect commissions, the key result here is that there is strong evidence that the absolute difference of the issuer s reputation and the reputation of the current underwriter affects the commission. More specifically, the smaller the absolute difference (the better the matching), the lower the commission. Table?? reports the results of estimating a standard launch spread equation with an underwriter reputation variable and a matching variable added. Both the reputations of the issuer and the underwriter impact on the launch spread. The launch spread is lower when the reputation of the underwriter handling the issue is higher suggesting that underwriters are certifying the 15