COMBINE AND CONQUER: SEGMENTATION, REPUTATION AND THE DEMAND FOR UNDERWRITER SERVICES IN THE INTERNATIONAL DEBT MARKET. +

Transcription

1 COMBINE AND CONQUER: SEGMENTATION, REPUTATION AND THE DEMAND FOR UNDERWRITER SERVICES IN THE INTERNATIONAL DEBT MARKET. + Michael G. Kollo * London School of Economics October, 2006 (Job Market Paper) ABSTRACT I examine the role of investor relationships and reputation in the demand for debt underwriting services. Relationship to the target investors is measured by the underwriter s access to the domestic market of the bond s currency. The introduction of the Euro, and the resulting change in investor segmentation, is used as a natural experiment to measure the change in the value of investor relationships to underwriters. The structural model estimates for the value of investor relationships in the segmented pre-euro European markets are USD 64 million per year, and constituted about 62% of the total demand for European banks services. Reputation is valued by issuers for the more liquid Euro and U.S. Dollar bonds. Key words: Underwriter competition, Underwriter fees, international finance, industrial organisation Journal of Economic Literature Classification Codes: G15 and G24. * Michael G. Kollo is at the London School of Economics and Financial Markets Group. I would like to thank David Webb, Peter Davis, Ayako Yasuda, Timothy Burch, Enrique Schroth, Tom Chemmanur, Otto van Hemert, Andrew Patton, Antoine Faure- Grimaud, Margaret Bray, Bob Nobay, and participants at the London Business School PhD Inter-Continental Seminars. The paper has also benefited from discussions with Nikolaj Schmidt, Asher Curtis, Matt O Connor, Michela Verardo and the PhD Seminar workshops at LSE. All errors are my own. 1

2 "Execution and investor knowledge can be more if not as significant as pricing." Matt O Connor, Managing Director, Fixed Income Finance, General Electric, July The growing sophistication and increasingly central role of the financial services industry has produced a large literature on the pricing and demand for banking and financial intermediation by firms. The common models of underwriter literature examine the equilibrium demand or pricing as a function of the characteristics and relationship between the underwriter and the issuing (borrowing) firm. This approach neglects an important third party in the transaction; the potential investors of the new security. The special relationship between the underwriter and the market was identified as early as Baron (1982) and continues to be acknowledged, albeit not directly examined, by recent literature (Altınkılıç and Hansen (2000), Ljungqvist, Marston and Wilhelm (2004) and Fang (2005)). As investor identity is not public information, few studies 1 have been able to identify investor identity and therefore consider the relationship between the underwriter and investor as determining the equilibrium outcome between issuer and underwriter. I study the importance of the underwriter-investor relationship in the issuer s demand for underwriter services. Using a unique dataset of new debt issues in the global capital markets, I am able to explicitly identify segmentation in the investor pool that favours different groups of underwriters. Using a structural model and an exogenous shift in market structure, I find that investor segmentation does significantly impact underwriter competition. Unlike other models, the source of this friction arises from geographic segmentations and investor home-bias as opposed to information asymmetries or banking relationships (James (1992), Burch, Nanda and Warther (2003), Yasuda (2005)). I identify and test three factors that differentiate competing underwriters as: (i) target investor relationships, (ii) underwriter reputation and (iii) underwriter size. The study is the first to study the role of investor relationships in the market for underwriting services. Geographic segmentations arise in the investor base that leads different distribution abilities for underwriters in different currency bonds. Being local to the 1 Notable exceptions have been Cornelli and Goldreich (2003) who use a proprietary database to examine book-building allocation by underwriters. 2

3 target investors of bonds gives the incumbent underwriter market power and leads to differentiation. The study focuses on the franchise of European banks in their home currencies as a study of geographic segmentation. I am the first to use the natural experiment of the Euro to estimate and identify the significance of this effect in differentiating underwriters. Reputation is often identified as a significant certification device that differentiates underwriters in the debt and equity markets (see Chemmanur and Fulghieri (1994) and Puri (1999) for theoretical arguments and Carter and Manaster (1990), Krigman, Shaw and Womack (2002), Burch, Nanda and Warther (2003), Ljungqvist, Marston and Wilhelm (2004) and Fang (2005) among others for empirical studies). Whilst it is not surprising that reputation acts as a valuable certification 2 method in the IPO and SEO equity markets, it is not clear why it continues to have economic significance in the relatively more information transparent investment grade debt market. The study is the first to provide economic estimates of the value of reputation in attracting demand for underwriting services in the largely investment grade international debt markets. I use the well known nested logit demand function (McFadden (1973)) in estimating the firm s demand for underwriting services. The methodology is closest to the recent study by Schroth (2006) who examines the value of innovation of new financial securities by underwriters in the US domestic market. Schroth uses the nested logit demand model (Berry (1994)) to estimate the difference across the choice of innovator and imitator underwriters. The study differs radically in the aim of the research agenda; the aim is to establish general sources of differentiation between underwriters across a cross-section of different markets. The data consists of the issuer identities, contract characteristics, and fees for 2535 debt issues that were underwritten by 90 international banks from 1994 to I identify the ten largest currency segments by issue volume and these contain the major European (legacy) currencies that were later replaced by the Euro. The legacy currency segments were characterised by fractured investor base and the dominance of local European underwriters. In spite of higher rating than the U.S. Dollar segment, 2 Underwriter reputation fulfils a credible certification device that signals the value of new securities to the market. The value of this service is proportional to the information asymmetry between the inside information of the firm and the market (Chemmanur and Fulghieri (1994)). 3

4 fees were still significant higher in the legacy currency segments (185 versus 134 basis points). By estimating a structural model of issuer demand, I find that investor relationships in these markets provided European underwriters with rents equivalent to 129 basis points or $US 2.50 million for the average issue size. Alternately, they derived 62% of the total demand from their positions. Other non-local underwriters were also present in these markets, but relied on their reputation capital and size to generate demand for their services. European underwriters effectively enjoyed the demand of highly reputable U.S. underwriters before the Euro. The introduction of the Euro saw the amalgamation of the European investor base under a single currency and a shift in market shares from the incumbent European to the reputable U.S. underwriters. The European franchise was severely damaged, as increased competition in fees saw the average underwriter fee fall by over 50% and new issue volume rise two-fold. While increased liquidity was likely a factor in the fall in fees, the removal of investor segmentation and its effects on underwriter competition was very significant. The results complement and extend the existing literature in a number of ways. I extend the literature by testing the two new factors in underwriter competition of investor relationships and underwriter size. I find that the value of investor relationships is larger but comparable to that of banking relationships identified by Yasuda (2005). She finds that lending relationships in the U.S. investment grade corporate bond market are equivalent to 58 basis points (bps), or 39% of the average 150 bps fees reported by Burch, Nanda and Warther (2003). The comparable magnitudes suggest that investor relationships are economically comparable in importance to lending relationships. I extend the literature by confirming that reputation capital is of value to competing underwriters (Krigman, Shaw and Womack (2003) and Fang (2005)). Fang (2005) stresses the significance of reputation for sub-investment grade issues, but I find that the bulge bracket (top five) underwriters derive a significant 24% more demand through their reputation in the U.S. Dollar segment of the market. This result is significant after controlling for the direct costs to the issuer (fee and the yield). 4

5 Finally, I find that the yield on new bonds is not an economically significant source of differentiation among underwriters in this market, consistent with transparent pricing practises. This appears to be in contrast to studies by Gande et al (1997), Livingston and Miller (2000) and Fang (2005) who find that high reputation underwriters obtain lower yields. Though these studies are silent on the economic significance of their findings in investment grade markets, their results are likely to be more pronounced in sub-investment or illiquid markets. The remainder of the paper is structured as follows. In Section I, I briefly review some of the literature on market power and introduce the data and summary statistics in Section II. In Section III I outline the potential determinants of the choice of underwriter followed by the formal model in Section IV. Section V details the estimation, results and robustness and Section VI concludes. 1 Previous Literature Recent work in finance has sought to identify sources of market power in banking underwriting that may impact pricing and access to capital markets. Frameworks with imperfectly competitive agents require increasingly sophisticated methods to model and empirically estimate equilibrium demand and supply relationships in the financial services industry. Imperfect competition in the financial services industry may arise due to information frictions related to relationship specific effects and geographic barriers to investment. Models of information asymmetry leading to market power between the intermediary and firm appear in the audit (DeAngelo (1982)), banking (Rajan (1992)) and underwriting literatures (James (1992)). In the initial period, the provider of the financial service invests some sunk costs into acquiring soft information that is necessary to provide their certification and/or monitoring services. In the case of banks, this may be a screening technology. In consequent rounds of competition, any competing service providers must pay the sunk costs and may also bear the firm s own switching costs. The sunk costs provide natural barriers of entry for new suppliers, thereby leading to some unilateral market power for the incumbent in 5

6 pricing their services and potentially limiting access to capital of firms. An alternate mechanism is outlined by Broecker (1990), who argues that market power of lending banks arises due to the issuer s aversion to send negative signal by leaving their bank. The market power of the bank is inversely proportional to the level of adverse selection in the market for credit arising from the information asymmetry of other borrowers. While these models identify localised market power between borrowers and intermediaries, they do not address wider elements of market power in the market for intermediary services. The second branch of literature investigates the role of geographical segmentation in investment decisions. Recent work by Coval and Moskowitz (2001) highlights the importance of geographic proximity of investors and the investment location. The authors show how mutual funds in the U.S. domestic market favour investments with close geographical proximity and earn higher abnormal returns on these investments. As geographic segmentation forms a component in the performance, it may also play a determining role in the ability of funds to attract new capital inflows. As an example, a mutual fund located in close geographic proximity to a specialised industry may enjoy competitive advantages in investment, generating returns and therefore attracting new funds vis-à-vis other mutual funds. Assuming positive costs of relocation, competition in the mutual fund industry is likely to exhibit geographic patterns, where funds enjoy some market power specific to their geographic locale. Similar patterns of geographic segmentation are identified by Becker (2006) that examines segmentation in the supply of capital in the form of higher deposits of (retired) seniors. Becker shows that the heterogeneity in the supply of capital is a significant factor in regional economic outcomes, and effectively acts to create segmentation in the market for deposits. The geographic segmentation is shown to be reduced by deregulation of interstate branching (Kroszner and Strahan (1999)) resulting in the geographic integration of banking markets. My paper examines a similar phenomenon in the international underwriting markets where the segmentation in the investor base significantly impacts the competition for underwriting services. 6

7 2 Data and Summary Statistics 2.1 Sample Selection and Variables I obtain pricing, yield, issuer identity and lead manager identity information for 4770 non-equity linked straight/fixed-rate international debt securities issued by industrial firms between January 1994 and December 2003 from Thomsons Deals. 3 Financial firms are excluded due to their industry knowledge and internal underwriting capacity that may bias their demand for external underwriting services (see Gande et al (1997), Jewell and Livingston (1998) and Livingston and Miller (2000) among others). The aim is to construct a relatively homogeneous sample of fixed-rate issues to estimate the model. I therefore exclude all deals with multiple tranches 4, equity warrants 5 and implausible values of fees and yields 6. International debt instruments can be denominated in any currency, though most issues favour the larger and more liquid currencies. I therefore include issues denominated in the most liquid currencies of the U.S. Dollar, the Yen, the Euro 7, the pre-euro legacy currencies ( (the Deutschemark, the Italian Lira and the French Franc) and the commonwealth currencies (the British Pound, the Canadian Dollar and the Australian Dollar). The underwriting fees are commonly termed gross spread and expressed a percentage of the total amount. I standardise the yield to maturity of the bond by the relevant government bond rate to obtain the yield premium. The government bond data and methodology are described in the Appendix A1. 3 Thomsons Deals is akin to the commonly used research database SDC Platinum used by Gande et al (1997), Gande et al (1999), Livingston and Miller (2000), Burch, Nanda and Warther (2004), Ljunqvist and Wilhelm (2004) and Kollo and Sharpe (2004) among others. 4 Multiple trenches are groups of smaller bond offerings with different characteristics that are underwritten as part of a larger deal. The reported fees and yield to maturity of individual bonds within a larger deal may reflect aggregation of underwriter services across the deal, and therefore is not comparable to stand-alone offerings. 5 Equity warrants attached to bonds alter the final yield the maturity reported by Thomsons Deals. 6 Livingston and Zhou (2001) note among others that SDC Platinum has some errors in variables. I exclude four observations with the implausibly high underwriter spreads of 22.42%, 42.09%, 93.92% and 165% 7 The Euro was introduced and replaced the European legacy currencies in 1999 in trading, though the physical currency did not enter circulation till a few years later. The Euro effectively removed the barriers of segmentation between the investors of legacy currencies, with the direct effect of reducing exchange rate transaction costs and increasing liquidity. For a detailed analysis of the introduction of the Euro see Detken and Hartmann (2000). 7

8 I identify the lead manager (book-runner) 8 of a transaction as the lead bank and obtain consolidated group accounting information for the parent entity from Compustat Global and Mergent Online. In case of discrepancy, I use the Compustat Global figure. Due to different national banking laws regulating bank capital and reporting of accounts, I adopt a conservative approach to measure bank size using the book value of assets. The approach seeks to minimise cross-country differences in banking regulation that may lead to disparities in asset size management and reporting. I treat mergers carefully by identifying newly merged underwriters as separate entities, except for reputation. A full description of the merger treatment is described in the Appendix A2. I hand-collect information on the underwriter s reputation by using published survey polls from the practitioner periodical Euromoney Institutional. Euromoney publishes annual rankings of underwriters in the international debt markets based on responses from institutional investors and the investment banking community on the perceived quality of service in underwriting activities. The top ranked banks receive special editorials and receive accolades for their performance. The overall ranking is a combination of scores for the underwriting, trading, advisory and transaction processing of the bank. I use the underwriting score to rank the top twenty banks. The final sample consists of 90 underwriters that managed at least two deals from 1994 to 2003 in the top ten currency segments of the international debt market. The evolution of the sample from deal-by-deal to underwriter-market-year is also summarised in the Appendix. 2.2 Market volume and issuers The new issue volume and underwriter spreads for the same currency segments are reported in Table I. The market for Euro denominated fixed-rate international debt issues increased enormously from legacy levels from US $62 billion (94-98) to $216 billion (99-03). 9 The average (median) size of international debt issues rose from U.S. 8 In the case of multiple co-leads, I credit each bank with an equal share of total transaction amount. 9 The boom in new issuance is also documented by Galati and Tsatsaronis (2003). 8

9 $194 ($192) million for legacy currencies to U.S. $362 ($264) million for Euro denominated issues, consistent with greater liquidity of the new Euro currency. In support, the average underwriter fee declined by over 50% from 186 basis points for legacy currency issues to only 84 basis points for Euro denominated issues. 10 The fees for U.S. Dollar issues declined only marginally from 134 to 121 basis points over the same period. The decline in average underwriter spreads was more than offset by the rise in volume of new issues. Underwriting revenue rose from a combined US $230.3 million a year from bonds in the Deutschemark, French Franc and Italian Lira legacy currencies to US $364 million a year for Euro issues. Next, I examine the homogeneity of issuers in this market. Table II shows the characteristics of issuers in the U.S. dollar, Japanese Yen, Euro and legacy currency segments. A majority 94.3% of the number of pre-euro legacy issues and 80.5% of the Euro issues are made by infrequently issuing firms. Those that are repeat issuers are usually large multinational manufacturing firms with high credit rating, i.e. the U.S. multinational firm General Electric. Issues are largely high quality, with only 6.6% of issues in U.S. Dollars and 0.2% in the Yen rated sub-investment grade rated by Moody s. It is interesting to note that the proportion of sub-investment grade debt rose from 2.3% in the pre-euro legacy currencies to 7.5% in the Euro. The greater liquidity and market participation of the new currency appears to have led to greater participation by lower ranked issuers. 2.3 Underwriter characteristics The underwriters in the international debt markets are large predominantly European, Japanese and U.S. commercial and investment banks. Table III shows the underwriter characteristics in the sample. A total of 22 European underwriters operated in the legacy currency segment, 19 of whom acted as lead managers in their home markets. 11 Reputation clearly favoured the U.S. underwriters with a mean value of for the U.S. Dollar segment, while the European underwriters lagged behind with only This is consistent with the economies of scale in direct issue costs found by Altınkılıç and Hansen (2000). 11 This suggests that three European underwriters acted only in currency segments other than their own. These are ABN Amro and two smaller European underwriters whose local currency segments were excluded due to insufficient observations. 9

10 and 3.19 in the legacy and Euro currencies respectively. The difference in mean reputation values between European and U.S. underwriters is further highlighted by Figure 5. Figure 5 shows that U.S. underwriters maintained a clear dominance in reputation in the legacy and later the Euro currency segments over their European counterparts. The U.S. banks were large, well established multinational underwriters with presence in multiple markets that earned them on average, higher rankings. Many of the European underwriters were smaller local banks with limited market shares that did not feature in published rankings. 3 Potential Determinants of the Choice of Underwriter I present three potential sources of differentiation for underwriters in the international debt market as: (i) investor relationships, (ii) global reputation and (iii) placement risk. These sources of differentiation are in addition to the transaction costs to issuers including the fee and the yield to maturity of the new debt instrument. 3.1 Investor relationships I propose that investor relationships are a significant source of differentiation among competing underwriters. The hypothesis contains two parts: (i) potential investors for new issues are segmented and (ii) underwriters have different access to pools of investors. First, investors are segmented due to home to home-bias and/or restrictive domestic regulation that limits the exposure of large mutual funds to foreign currency denominated assets. 12 This suggests that the bond s currency of denomination largely determines the investor segment that will purchase the issue, that is, the domestic market of the bond s currency of denomination. Underwriter s investor relationships arise due to underwriting and/or lending activities in different domestic markets. The strength and exclusivity of these relationships defines the underwriter s placement capacity of bonds in a given currency. Issuers choose the underwriters to place their bonds with a given investor group, and therefore consider the underwriter s relationship to the target investor group. Smaller, less liquid domestic markets are 12 Roldos (2004) reports pension funds investments in foreign assets accounted for eleven percent of portfolio holdings in the U.S., and only five and seven percent in France and Germany respectively in

11 likely to have fewer active underwriters, and therefore investor relationships are likely to be more valuable for issuers wanting to place with investors in those markets. The empirical evidence supports the positive correlation between the bond s currency of denomination and the chosen underwriter s nationality (McCauley and White (1997) and McCauley (1999)). The mean market shares of volume and fee revenue of European banks for bonds denominated in their home currencies is significant before the introduction of the Euro as shown in Figure 1. What is perhaps surprising is the significant decline in market shares for the incumbent European banks at the introduction of the Euro from 70% to 50%. The anecdotal evidence suggests that the importance or exclusivity of investor relationships declined with the widening of the investor base resulting from the introduction of the Euro. I identify an underwriter s relationship to the target investor group using (i) the nationality of the underwriter and (ii) the underwriter s share of the domestic security issuance 13 in the bond s currency s domestic market. Relationships with the target investor group may arise due to domestic commercial lending, advisory or other related banking services provided in the domestic market of the investor. The first proxy captures a nationality effect, in that it is a broad indicator of an underwriter s domestic market presence versus other foreign banks. This largely ignores the possibility that banks have significant (leading) presence in other markets. 14 The second indicator captures this effect directly, but can not capture lending and nonunderwriting financial services that may also lead to greater investor relationships. I term these specialist underwriters, due to their close relationship with the bond s target investor group. In the European case, this represents the value of investor relationships to banks in smaller, more illiquid capital markets before the Euro. The introduction of the Euro provides a natural experiment to test the effect of a change in investor pool on the significance of investor relationships in the demand for underwriting services. In line 13 This measure is based on the market shares of underwriters for all debt, equity and hybrid instruments within the sample currencies domestic markets. 14 While globalisation has led to increased entry and competition by foreign banks in domestic markets, it is still generally the case that the domestic banks command a majority share of banking assets and therefore relationships with the investor base in most developed markets. See Denaus (1996) for the share of banking assets and Roldos (2004) for Pension fund holdings of domestic assets. 11

12 with the raw correlations, I expect to find that the value of investor relationships will decline with, if not be eliminated by, the introduction of the Euro. The demand model will control for a number of factors that may also change during the period and I turn to these now. 3.2 Global reputation Underwriters have different levels of reputation capital that reflect the width and breadth of its investor relationships and its expertise in promoting high quality offerings. The accepted view of reputation is that it is valued by issuers (see Carter and Manaster (1990), Chemmanur and Fulghieri (1994), Krigman, Shaw and Womack (2002), Fang (2005) and Ljungqvist, Marston and Wilhelm (2006) among others). Fernando, Gatchev and Spindt (2005) argue that higher reputation underwriters generate greater surplus for the issuer in their ability to certify, promote, place and support an offering. What is not clear is what the source of the value to issuers is. If reputation reflects the quality of the underwriter, then higher quality underwriters may charge a higher fee but obtain a lower yield for issuers (Fang (2005)). If so, then controlling for yield should produce no reputation effect in the empirical results. Otherwise, the value of using a high quality underwriter may lie in an indirect signal of the overall quality of the firm. Associating with higher reputation underwriters can translate to a number of indirect benefits including greater future access to capital markets, lower costs of capital and/or increased marketing and exposure of the firms name in capital markets. The empirical tests will seek to distinguish the way in which reputation is important, and if so, the economic value of reputation. Empirical identification of the concept of reputation is challenging. The general approach is to obtain rankings from information on completed transactions by using published tombstone 15 and league table rankings 16 (Carter and Manaster (1990) and Megginson and Weiss (1991), Yasuda (2005) and Fang (2005)). A common criticism of the league table rankings is that they may be correlated with underwriter size (capacity) rather than quality of service. I employ a measure that I believe is largely 15 Tombstone announcements are publicly available advertisements of upcoming new issues that list the underwriters involved in order of importance. 16 League table are published by the major data vendors and rank underwriters by the volume of transactions in a given security type. 12

13 free of this size bias by using investor survey information from a publicly available periodical (Euromoney Institutional). 17 I rank underwriters by assigning a value of 20 for the top underwriting bank, 19 for the second and so on. I expect a positive coefficient on the proxy for underwriter reputation to reflect the higher value associated with higher reputation underwriters (Chemmanur and Fulghieri (1994), Puri (1999) and Fang (2005)). I provide a further discussion of the relationship between size and reputation in the robustness section. 3.3 Placement Risk and Insurance Underwriters provide insurance for unsold bonds by purchasing and holding them on their balance sheet. The underwriter s ability to pledge the economic capital on its balance sheet to support the issue placement is directly related to the value of the insurance it provides. Larger issues may demand underwriters with larger balance sheets to provide credible insurance against unsold bonds. 18 Insurance for unsold bonds is just one reason why book value may differentiate underwriters. Banks with large balance sheets may be more capable of making (future) loans to firms. 19 The ability of some banks to bundle their lending and underwriting services by having larger, more robust balance sheets may also differentiate them. In both these cases, there is some argument for the positive relationship between underwriter size and the demand for underwriting services. As lending relationships are unobservable in this market, I focus my discussion on the value of insurance against unsold securities. Anecdotal evidence of the relationship between underwriter size and market share is shown in Figure 3. Larger banks increased their market shares significantly in the pre- Euro legacy currencies and the Euro from 1994 to Some of this trend is likely to be due to growth in banking assets and general consolidation in the banking market. 17 Some of the bias toward favouring larger underwriters with higher reputation may remain even in survey data. This is because investors may not be familiar with smaller underwriters, and therefore can not provide favourable opinions. I further address this bias in two ways. Firstly, by including variables for underwriter (book) size, I control for large differences in capacity. Secondly, the sample only includes relatively large, multinational banks that are likely to be widely known in the investor community. 18 Another mechanism to provide insurance for unsold bonds may be the use of a co-management and sales syndicate team (Corwin and Schultz (2005)). While syndication is increasingly common, a lion s share of the issue volume will be committed by the lead underwriter. 19 These loans are largely unobservable in public databases. The exception is of syndicated loans examined by Drucker and Puri (2005). 13

14 To address this, I sort underwriters by their balance sheet size into four equal declines in each year. I then calculate the market shares of underwriters in each decline in the legacy and Euro and report these in Figure 4. The figure shows a positive shift in market shares to the largest (top 25%) underwriters after the introduction of the Euro. This appears to come at the cost of the smallest (bottom 25%) underwriters that capture less than 10% of the underwriting market in the same period. As underwriter size and market share appears to be correlated, I include it in the determinants of the demand for underwriting services. I use the underwriter s book value of assets to represent their ability to commit economic capital to support the new issue. 20 If underwriters differentiate themselves in size, the proxy will be significant in the demand for underwriting services. 4 The Model of the Demand for Underwriting Services There are three agents in the international debt markets: firms seeking finance, underwriters providing intermediary services and investors that purchase the bonds. Firms (i=1 I) seek financing for a positive net present value project in each period in the public bond markets. Investors exhibit strong preferences for assets denominated in their home currencies. For example, German investors held Deutschemark and then later Euro denominated assets. The introduction of the Euro therefore increased the breath of the investor base by amalgamating the European investors. Issuers employ the services of an intermediary to certify and distribute the new bond to the purchasing investors. 21 Each firm i approaches the market for underwriters with a debt instrument in currency m. Each firm makes a discrete choice of a single underwriter (Gande et al (1999), Puri (1999) and Yasuda (2005)). A finite set of underwriters (j=1 J) intermediate between the issuers and investors (Baron (1982)). 20 The choice of proxy omits the possibility that of heterogeneous commitment levels across banks. As national regulatory restrictions are likely to be a factor in this, I include underwriter nationality indicator variables and find that the results are qualitatively unchanged. 21 As noted before, the usual problems of information asymmetry and certification don t appear to bind in these markets, however almost 100% of issues are underwritten using a bank. This further raises the question of why underwriters are necessary in these markets. 14

15 Underwriters offer to sell the bond at a set yield premium 22 y jmb (price) and post a fee for their service, p jmb. 23 Both the yield and fee are publicly observable characteristics of the underwriter s service and form part of the direct costs of borrowing for the issuing firm. Underwriters offer heterogeneous services based on their relationships with the target investor group, their reputation and their size. As underwriters may exhibit geographic (currency) segment specific market power, I model competition as an oligopoly framework with horizontal differentiation that may vary by market segment and by time. The indirect conditional utility 24 of each firm i purchasing the services of underwriter j to place a bond into currency segment m is given as: V = V ( w, y, p, ξ ) + µ (1) i jt i The mean value of underwriting services across issuers is denoted by V, while the µ i is the additive random component of the issuer s discrete choice function. w and ξ jt denotes the observable and unobservable (by the econometrician) characteristics of the underwriter. The observable characteristics of the underwriter are its geographic proximity to the target investor group, the underwriter s reputation and the underwriter s size. The direct costs of the service are the yield to maturity of the bond y and the fees p charged by the bank. The term µ i specifies the parametric distribution of issuer preferences across different groups of underwriters, markets and over time. A full estimation of correlations is computationally burdensome and requires more detailed dataset that is available to us (Berry, Livingston and Pakes (1995)). Therefore, I follow the literature by assuming a particular parametric distribution for heterogeneity in the term {µ ijmt,, µ ijmt } that is 22 The yield premium is over the relevant government bond (risk-free) benchmark. I use the term yield to refer to yield premium henceforth. 23 This is characteristic of a bought deal where the underwriter guarantees the issuer a price (yield) for their security by agreeing to purchase any unsold securities. Intermediaries may the form syndicates with other banks to assist in the sale of the bonds. The fixed-price offer obliges the members of the syndicate to sell the bond at the price set by the lead intermediary and prevents price undercutting my syndicate members. Both bought deals and fixed-price offers are the dominant form of underwriting bonds in domestic and international markets. 24 It is indirect because the budget constraint is included through the p and conditional because it is conditional on a choice of an underwriter. 15

16 assumed correlated among underwriters but independent across markets The correlation in choice between members of a group of underwriters is estimated from the model with the parameter σ. I group underwriters by their geographic proximity to the target investor group. 25 For non-zero values of this correlation, the model is akin to the well known nested logit structure 26 shown in Figure 2 (McFadden (1973)). Finally, issuers choose underwriters that maximise their valuation of the service. However, if no option is optimal, I allow the firm the option of an outside-choice, the choice not to use any underwriter at all, and abandon the transaction. The utility of this outside choice is set to zero. The empirical definition of an outside good is nontrivial and requires an estimate of the maximum possible market size 27 of all financing demanded by firms to undertake projects. I therefore define the market size as the volume of new debt and equity issues in both the domestic and international markets of the bond s currency of denomination. 28 Given the outside good, Berry (1994) provides a closed-form solution expression which facilitates linear estimation of the multinomial model and the determinants of the mean value V as: V s = ln(ˆ ) ln( ) = ' β α1 y α 2 p + σ s + ξ (2) somt The dependent variable is the function of the market share of underwriter j in market m at time t s and the market share of the outside good s omt. The correlation structure within groups of underwriters is proxied by the within-group market share of each underwriter j, ŝ. The correlation in issuer preferences for underwriters within groups, σ, is an estimated coefficient of the model. 25 I also tried to group underwriters into high, medium and low reputation. Though found that correlation in preferences was positive among reputation groups and out coefficients were comparable in significance to the present specification, the estimates were not as stable across alternate specifications. 26 The nested logit model is a tractable representation where the econometrician specifies ex-ante the correlation pattern between the choices. The structure of choices is therefore represented as a treestructure where the first level determines the choice of local underwriter, while the second level is the choice of a specific underwriter within the local underwriter category. 27 As an example, Nevo (2001) takes the difference between the volume of cereal that would be consumed if the entire population ate cereal each morning and the actual cereal sales. 28 In the case of the Euro, I take the sum of the European member states domestic market new issues for that year. 16

17 4.1 Features and Limitations The base multinomial model assumes that issuers differ only according to the additively separate idiosyncratic component µ i. This is clearly a limiting assumption in favour of tractability that future work following Berry, Livingston and Pakes (1995) may aim to relax. In this paper, I assume that issuers don t differ too much in their preferences because they are usually large, listed, well-known and almost always investment grade rated. The high credit rating is required due to the absence of a single domestic regulation framework to protect debt-holders rights. 29 I add proxies for issuer characteristics in segments as additional controls for systematic differences that may arise in issuer characteristics and therefore preferences across currency segments,. Consistent with previous literature, this model captures elements of static interaction between underwriters and issuers, thereby neglecting any dynamic elements in firmunderwriter interaction that may arise. An example of dynamic model may be of dynamic learning, where issuers change their valuation and therefore choice of underwriter services through repeat transactions or by observing the underwriter over time. These models require the formulation and estimation of a dynamic demand model that will require further simplifications of the present framework. While of interest for future research, the available form of such a dynamic demand model is intractable. 30 Dynamic interactions are also less likely to exist in the international debt markets due to infrequent repeat issuance by firms (94% in the legacy and 81% in the Euro). Firms issue infrequently in the international debt markets, and due to their size and transparency, usually employ a number of different intermediaries for different markets. Kollo and Sharpe (2004) report high instances of switching (70%) in their sample of floating rate notes issued in the international debt market. The high instances of switching are likely to suggest lower scope for dynamic interaction due to switching costs or learning. 29 As an example, the proportion of fixed-rate sub-investment grade issues is less than 5% (Esho, Kollo and Sharpe (2006)) vis-à-vis over 20% in the U.S. domestic market (Jewell and Livingston (1998)). 30 See Rysman and Gowaileran (2006) for a recent possible formulation. 17

18 5 Estimation and Results Estimating the demand model outlined in (2) requires aggregation over individual bond characteristics to obtain indicators for the fee and yield. Both the underwriter fee and the yield vary with bond characteristics like tenure and rating, so these characteristics should be used as controls across currency segments and to derive underwriter specific components of pricing and yield. The empirical method used to obtain the fee (p ) and the yield (y ) as inputs into equation (2) is outlined in Appendix A3. The error term in the demand equation ξ is the unobserved underwriter characteristic 31, and may be correlated with the fees p and within-group market shares ŝ (Berry (1994)). Such a correlation would mean ordinary least squares estimates of the coefficients of demand will be biased. I therefore estimate the coefficients of the model using standard instrumental variable regression technique to correctly identify the demand parameters. The instruments used must be correlated with supply factors which affect the underwriter s cost of supplying the service. I select the instruments carefully with this aim in mind as the two-year and the ten-year government bond interest rate daily volatilities in market m and year t. Higher interest rate volatility is likely to increase the underwriter s risk that some of the bonds will remain unsold due to adverse interest rate movements. Similarly, I include the specialist indicator and the global reputation variables that appear in the main demand model as also possibly correlated with the direct costs of placement. 32 Finally, I include indicator variables for pure investment banks as their cost structures may differ to universal banks and indicator variables for European and U.S. banks to represent different compensation structures for banking personnel across nationalities. The within-group market shares are also instrumented using the same orthogonal instruments. The first stage estimation results are not reported for reasons of brevity 31 Examples of unobservable underwriter characteristics are marketing campaigns, specific client relationships, business networks and any other soft signals of underwriter characteristics that issuers assign a positive value to in their demand function. 32 Excluding the specialisation and global reputation variables has a no qualitative effect and only minor quantitative effects on the results. 18

19 due to the larger number of parameter estimates. 33 The fee index regression has an adjusted R 2 is with a statistically significant F-stat of The within-group market shares regression has an equally high adjusted R 2 of with a statistically significant F-statistic of Estimated demand parameters The regression estimated is of the form: V = λ ) + 0 GS jm α1 y α 2 p + σ ln(ˆ s ξ (3) The difference in coefficient estimates across the pre-euro legacy and the Euro are tested by adding interactive terms for the legacy and the Euro segments. For example, the interactive terms β 1 (Leg m GS jm ) and β 2 (Euro m GS jm ) capture the change in the value of specialisation from the legacy to the Euro segments. If β 1 > β 2 then the value of specialisation decreased, while if they are equal then there was no change with the introduction of the Euro. Table IV presents the results for the base regression and the interactive terms. The adjusted R 2 shows a good fit of the model at a relatively high value of 62.9% and 62.5% respectively. Tables V, VI and VII provide a more detailed set of interactions for each determinant with each currency segment. As these tables provide valuable information about the difference across currency segments, I focus my discussion on the general results of Table IV, and show the summary of the economic values of my estimates in Table IX Market Specialisation Table IV shows that both indicators of investor relationships, both derived from sharing nationality with the target investors and by having domestic underwriting market share, are valued by issuing firms. The interactive terms show that this is especially pronounced in the legacy currency markets, with a marginal positive value of above the other currency segments. The result in consistent with the argument that the significance of investor relationships declined with the introduction 33 These results are available on request. 19

20 of the Euro. The third regression reported finds that investor relationships derived from domestic market underwriting activities were especially important in the legacy currency segments, after controlling for the underwriter s specialisation derived from its nationality. This supports non-local underwriters establishing investor relationships through domestic underwriting subsidiaries. The economic interpretation follows by comparing the estimated coefficients relative to the coefficient of the underwriter fee index (e.g. Yasuda (2005)). The interpretation is as follows. An increase of 100 basis points (bps) on the underwriter spread decreases the demand for underwriter services by units based on the coefficient value on the fee index variable. The marginal value of being a specialist underwriter with investor relationships in the legacy currencies increases demand by units. The trade-off implies a 48 basis point (bps) value of being a specialist relative to other currency segments in the sample. The Euro interaction term is not significantly different from zero, indicating that this benefit was not carried over to the Euro. Ceteris Paribas, this equates to a hypothetical loss of US$1.27 million per deal of possible fee revenue after the Euro owing to the removal of geographic segmentation that maintained the European specialist s demand. This is also equivalent to a relative 34 loss of US$ million 35 per year in fee revenue for European underwriters as a whole. It is also interesting to compare the magnitude of the geographic segmentation with the observed drop in fees after the Euro. Fees dropped by 102 bps from 186 to just 84 bps (Table I), which may suggest that greater competition due to the removal of the effect of geographic segmentation (48 basis points) may have played a significant part. Next I ask the question How much of the total demand was driven by market specialisation in the smaller currency segments?. I obtain the difference in implied demand for a specialist and a non-specialist underwriter in the legacy and Euro 34 This is a relative loss as underwriters received greater overall revenue due to increased volume of transactions and in spite of falling fees. 35 Calculated using the volume of new Euro denominated issues from 1999 to 2003 from Table 3 and the estimated value of 50 basis points per transaction. This figure is likely to underestimate the total economic effect as it is estimated using the sample of fixed-rate international bonds issued by nonfinancial firms. Including that larger universe of bonds (for example floating-rate, convertible equitylinked and asset-backed) and widening the issuer types to government and banks is likely to provide an even higher estimate of the value of specialisation to European banks. I continue to use within-sample economic interpretations as a conservative measure of the value of specialisation. 20

21 currency segments by multiplying the statistically significant coefficients of the model with the mean values for the respective independent variables. Taking the exponential to give d s = s smt /s 0mt for the specialist underwriter and d s = s -smt /s 0mt for the nonspecialist underwriter. Calculating d s d d s s shows that specialist underwriters in the legacy (Euro) currency segment obtained 62% (41.5%) more demand than nonspecialist underwriters. This suggests that specialist European underwriters lost 18.7% of the demand for their services on the introduction of the Euro, Ceteris Paribas. The value of specialisation is likely to have supported the presence of smaller, less reputable European banks that relied heavily on their specialist status to obtain demand for their services. The more detailed analysis of the role of the specialist in Table V suggests an even more severe loss of demand, as the benefits of specialisation appear to accrue only to those banks local to the Yen, Commonwealth (U.K. Pound, Australian Dollar and Canadian Dollar) and legacy currencies, but not those in the more liquid (and more competed for) U.S. Dollar and Euro segments Global Reputation The value of reputation is positive and increased significantly after the introduction of the Euro (Table IV). The coefficient values show that value of reputation in the Euro segment is 0.09 units per rank (to a maximum of 20). 36 The corresponding economic value of a give rank increase is 54 bps per transaction, or USD 1.2 million for a median U.S. Dollar denominated debt transaction. This is consistent with the notion that the valuation of underwriting services shifted from specialisation to global reputation due to the decrease of investor segmentation. As a final test of the economic significance of reputation, I calculate the average value implied by the model for a top (rank=20) underwriter and a mid-rank reputation (rank=10) and compare the marginal value of reputation for the higher ranked underwriter in the Euro segment. The top five to ten underwriters represent the most reputable banks (Fang (2005)). Using the methodology outlined in the previous section, I find that bulge bracket (top five) underwriters obtain 37% more demand for 36 This is also confirmed by the results of Table VII that show that the only two markets where reputation matters is the U.S. Dollar and the Euro. 21