Does cross listing in the U.S. really enhance the value of emerging market firms?

ECONOMICS DEPARTMENT WORKING PAPERS SERIES N184/12/07 Does cross listing in the U.S. really enhance the value of emerging market firms? by Thomas G. O’Connor Department Of Economics National University of Ireland, Maynooth NATIONAL UNIVERSITY OF IRELAND, MAYNOOTH MAYNOOTH CO. KILDARE A. IRELAND http://www.nuim.ie/academic/economics/ http://ideas.repec.org/s/nuim/mayecw.html http://eprints.nuim.ie/ 1 Does cross listing in the U.S. really enhance the value of emerging market firms? Thomas G. O’Connor♣ June 25th, 2007. Abstract In this paper, I study the valuation effects of cross listing in the U.S. for a panel of emerging market firms over the period from 1990 to 2003. In line with Kristian-Hope et al. (2007), I find that only those firms from high disclosure regimes gain from Level 2/3 listing in the U.S. The gains are not immediate, but materialize once the firm has listed in the U.S. for at least five years. I also document long-term, but not immediate valuation gains for Level 1 over-thecounter issues. In contrast to Level 2/3 issues, the gains are concentrated amongst firms from low-disclosure regimes. I find no positive valuation effects for Rule 144a private placements. The results suggest that the decision on the part of the majority of firms from low-disclosure regimes not to list as exchange traded depositary receipts is warranted. JEL Classification: G15, G34, G35. Key Words: Cross listing, corporate valuation, emerging markets. Acknowledgements: I would like to acknowledge the receipt of financial support from the Maynooth Finance Research Group (MFRG) and Institute for International Integration Studies at Trinity College Dublin. Helpful and invaluable comments were gratefully received from Tom Flavin, Donal O'Neill, Denis Conniffe, Gerald Dwyer Jnr., Adriana and Piotr Korczak, Vicentiu Covrig and from participants at the IIIS Miniconference on International Financial Integration, Maynooth November 2004, the Maynooth Postgraduate Colloquium, Maynooth March 2005, the Emerging Markets Finance Conference, CASS Business School, May 2005, and the Global Finance Conference, Dublin, June 2005. All remaining errors are entirely my own. 1. Introduction ♣ Department of Economics, NUI Maynooth, Maynooth, Co. Kildare, Ireland. Contact information: Tel: +353 1 7083452 E-mail Address: [email protected] This paper was presented at the Money Macro Finance (MMF) Annual Conference (2005) under the heading “Cross listing in the U.S: correlated and causal effects”. This paper is a revised version of Chapter 5 of my PhD thesis at NUI Maynooth. 2 Up to recently, the prevailing wisdom decreed that the greatest gains to exchange-cross-listing in the U.S. should accrue to those firms domiciled in countries where investors are poorly protected. Grounded in what is commonly referred to as the ‘legal finance’ literature, an exchange-cross-listing in the U.S. provides a remedy for firms to overcome their financing constraints at home, and thus facilitate their hitherto stagnated growth (See La Porta et al., 1998 and Demerguc-Kunt and Maksimovic, 1998 for the legal finance view, and Coffee, 1999, 2002 and Lins et al., 2005 for arguments specific to cross-listing). Although this option is not cheap, and is greatest for emerging market firms (i.e. cost of U.S. G.A.A.P. compliance), the belief was that the potential benefits from listing, would be more than sufficient to meet these additional on-going costs. In fact, in their theoretical model, Doidge et al. (2004), using valuation metrics theorize that the greatest valuation gains to listing, what they refer to as a ‘cross listing premium’, should accrue to firms domiciled in emerging markets. Bris and Brisley (2006) reach similar theoretical conclusions. In support, using standard event study analysis, Miller (1999) and Serra (1999) find empirical findings in favour of this view. However, this view has been recently challenged. First, in contrast to Doidge et al. (2004), Kristian-Hope et al. (2007), also using valuation metrics conclude that exchange-traded firms from low-disclosure regimes receive a lower valuation than their counterpart highdisclosure domiciled firms. They theorize that the benefits from listing may not be sufficient to meet the sizable costs (or large enough to ensure that low disclosure regime firms gain most). As a result, exchange-traded firms from high disclosure regimes gain most from listing1. Second, and consistent with this view, the number of firms that have de-listed from depositary receipt programs has intensified in the last few years (See Witmer (2006) and Marosi and Massoud (2006)). In many instances, firms have cited the costs associated with SEC compliance as the primary reason for delisting, which have intensified since the imposition of the Sarbanes-Oxley Act in 20022. Taken together, both suggest that the benefits from listing may not be sufficient to meet the costs of listing for emerging market firms. This suggests that the valuation gains, documented by Miller (1999) and Serra (1999) may be shortlived. In this paper, I examine the valuation effects of listing in the U.S. for a sample of 583 cross-listed emerging market firms. Like Doidge et al. (2004) and Kristian-Hope et al. (2007), I use valuation metrics, namely Tobin’s q. However, unlike them, I abstract from their cross-sectional approach and examine the valuation effects of listing within a panel setting. This approach allows me to examine the valuation effects in calendar (as Doidge et al., 2004 and Kristian-Hope et al., 2007 do), but also in event time. The benefit of the later is that it allows me to examine the effects of listing on firm value in the short and long run3. The primary drawback with the cross-sectional 1 Firms from high-disclosure regimes have also cited the costs of SEC complinace as their primary reason for delisting. For example, Skyepharma, a U.K. firm delisted from the Nasdaq in 2007 due to the “expense and burden associated with maintaining compliance with SEC and Nasdaq rules”. Healthcare Finance, Tax & Law Weekly. 2 The greater costs of listing post-Sarbanes-Oxley are not the only reasons for voluntary firm delistings. Chaplinsky and Ramchand (2007) show that a sizable proportion of those that voluntarily cross-delist are low quality firms i.e. firms with low average profitability, assets, and market capitalization, poor stock price performance (50% decline since listing), and no analyst coverage (60%). In connection, Hostak et al. (2006) show that those firms that voluntarily cross-delist have weaker corporate governance systems. 3 I could, of course have examined the long-term valuation effects of listing using the event-study approach of Miller (1999) and Serra (1999). However, this approach has two major drawbacks. First, Kothari and Warner (2005) highlight the limitations of long-horizon event study methods. Second, event studies may not adequately control for self-selection bias. These concerns are voiced by Heidle (2003) and Mittoo (2003). In his synopsis of Mittoo (2003), Heidle (2003, pg. 1664) concludes, “As with all event studies, the analysis in this paper suffers from a potential self-selection bias”. In fact Mittoo (2003, pg. 1659) explicitly acknowledges this shortcoming in her conclusion, “…long-term performance is generally difficult to measure and our results should be interpreted with some caution because of several limitations of our 3 approach of Doidge et al. (2004) and Kristian-Hope et al. (2007) is that the effect of listing is estimated using a cross-sectional of firms with varying degrees of exposure to listing. Thus, in effect, the cross-sectional estimate assumes that gains from listing are homogenous in each post-listing event year. However, there is ample evidence to suggest that this is not the case. Using a series of pooled ordinary least squares and treatment effect regressions, I find support in favour of the predictions of Kristian-Hope et al. (2007). For emerging market Level 2/3 issues, the valuation gains accrue only to firms from countries where investors are well protected. However, the gains are not immediate. I find that the gains to listing materialize, once the firm has listed in the U.S. for at least five years. I also document long-term, but not immediate valuation gains for Level 1 over-the-counter issues. In contrast to Level 2/3 issues, the gains are concentrated amongst firms from low-disclosure regimes. I find no positive valuation effects for Rule 144a private placements. The paper proceeds as follows. In the next section, I outline the sample of emerging market firms. In Section 3, I present univariate statistics and proceed to outline and discuss the regression analysis employed (pooled ordinary least squares and treatment effects). Section 4 concludes. 2. Data I begin by sourcing a full list of emerging market countries with firms cross-listed in the United States. From each, I identify those firms with a cross listing in the U.S. All information on cross-listed firms is sourced from the Bank of New York, and cross-referenced with information from Deutsche Bank, JP Morgan, the New York Stock Exchange, and Nasdaq. From my cross-listed sample of firms, I classify firms according to their first cross listing, and classify simultaneous Level 1/Portal ‘listings’ as Level 1 issues. My final sample (Table 1) is comprised of 4,563 non-cross-listed, non-financial firms and 583 cross-listed firms. The cross-listed sample is comprised of 260 Level 1 firms, 142 exchange-listed Level 2/3 issues and 181 firms that trade under Rule144a. I supplement my original sample of 4,563 non-cross-listed non-financial firms, with an additional 1,031 financial firms to ensure appropriate matches for our financial cross-listed firms. I do not include these financial firms in my fixedeffects, pooled ordinary least squares, and treatment-effects models since the valuation ratios for financial firms are not comparable to those for non-financial firms. Thus, the elimination of financial firms facilitates a greater comparison of firms across countries. Finally, I only include firms with average total assets greater than 100 million U.S. dollars. This latter approach facilitates a greater comparison between cross-listed and non cross-listed firms. In Table 1 I outline by country the number of non-cross-listed firms, and the number of cross-listed firms listed in the United States by depositary receipt level. I provide the percentage that each country (i.e. number of firms) contributes to the total number of firms in each depositary receipt level and adopt an identical approach for my non-cross-listed sample. Taken together South Korean and Malaysian firms comprise almost 28% of the noncross-listed sample: Colombian firms contribute just over half of 1%. Hong Kong firms provide the greatest number of Level 1 firms (37.31%), while Argentina provides no firm. Brazil and Mexico equally provide the greatest share of exchange Level 2/3 issues, while India and Taiwan supply the majority of firms that trade in the methodology. First, benchmarking performance with market indexes as done in our study could lead to serious biases and measurement problems”. I seek to control for self-selection bias using firm-fixed effects and treatment effects regressions. They are outlined later in the text. 4 U.S. under Rule 144a on the Portal. An interesting feature evident from Table 1 is that across and within countries there exists significantly differing preferences for the different types of depositary receipt listings. For example, the majority of firms from Hong Kong trade over-the-counter as Level 1 issues. This contrasts notably with the preference of Indian and Taiwanese firms for a Rule 144a issue. Israeli firms that are predominantly high-tech firms reveal a strong preference for exchange-listed (Nasdaq) depositary receipts. Although now seen as developed nations, I include Hong King and Singapore, as they were deemed ‘emerging’ during a sizable portion of the sample period. I follow Doidge et al. (2004, 2006, 2007), and Kristian-Hope et al. (2007) and employ Tobin’s q to measure firm value, where Tobin’s q is defined book value of debt + market capitalization book value of assets ⎛⎞ ⎜⎟ ⎝⎠ where book value of debt is calculated as book value total assets less the book value of equity. All variables are expressed in local currency, sourced from Worldscope and are collected on the 31st of December in each year from 1990 to 2003. I employ the following firm-level control variables in my empirical specifications: I use the average sales growth over the last two years (geometric average) and Global Industry q to account for firm and industry growth, respectively. Based upon primary standard industry classifications, the (yearly) mean Global Industry q is calculated as the average q of all global firms within each classification. I employ over 15,000 international firms from the Worldscope database to calculate the mean Global Industry q for each year. I use the log of total assets ($) to control for firm size. To remove the influence of outliers, I remove the top 1% of observations for Tobin’s q, and two-year average sales growth, and total assets. I include La Porta et al. (1998) country-level governance variables in order to examine the valuation effects of listing across different governance regimes. I employ legal origin (English Common, French, Scandinavian and German Civil Law), anti-director rights index, an equally weighted index of 6 different shareholder rights, which ranges from a low of 0 to a high of 5, and judicial efficiency, which ranges from 0 to 10. A higher rating implies greater judicial efficiency. The country-level governance variables are outlined by country in Appendix 2. In Appendix 3, I present correlation coefficient estimates for all firm and country variables employed in the analysis. By and large, the correlation coefficients are of the correct sign, and statistically different from zero. Tobin’s q is increasing in Global q, and anti-director rights. Larger firms are worth less. In the remaining column, I present variance inflation factors. Multicollinearity is of no concern in this study. 3. Empirical results In this section, I outline the main results on cross-listing and firm value. I begin with univariate comparisons, whereby I compare the value of cross-listed to non-cross-listed firms in calendar and event time. I then proceed to panel regression estimates. 3.1 Year-by-year valuation comparisons 5 In Table 2 (Panel A) I compare the value of cross-listed firms to non-cross-listed firms in each year from 1990 to 2003. For each subset of cross-listed firms, I outline the value of the mean and median firm in each year. In the remaining columns, I calculate the mean valuation difference [Mean Diff] between the average cross-listed firm, and non-cross-listed firm. Numbers in bold signify that the mean difference is statistically significant at conventional levels. To complement these numbers, I present in Figures 1-3, the mean and median value of crosslisted firms, and the mean value of non-cross-listed firms. The mean difference is calculated using only large firms i.e. firms with average total assets greater than one hundred million U.S. dollars. This facilitates a greater comparison between cross-listed and non-cross-listed firms. The median valuation differentials are available from the author upon request. In the final column, I outline the Average Effect of the Treatment on the Treated [ATT] using propensity score matching. The ATT, calculated as CL NCL E(q q |D 1, X) − =, is the difference in value between each crosslisted firm, and a matched sample of non-cross-listed firms, with an almost identical probability of listing i.e. propensity score. Cross-listed firms are matched to non-cross-listed firms based upon size (total assets), growth (two-year sales growth), legal origin, and industry group using propensity score matching. Li and Zhao (2006) adopt an identical approach in their study of seasoned equity offerings. In Appendix 1, I outline in greater detail the calculation of the estimated propensity scores and the ATT estimates. The summary measures presented in Table 2 (Panel A) are somewhat consistent with the findings of Doidge et al. (2006, 2007). Specifically, the matching estimates suggest that exchange traded firms experience the largest cross-listing premia. Next I find that the cross-listing premium tends to vary over time. For example, emerging market Level 2/3 firms are worth more, but not statistically so in every period. The cross-listing premium is greatest for these firms in 1994. In contrast, for Level 1 and Rule 144a firms, the valuation difference tends to vary from discount to premium over time. 3.2 Event time valuation comparisons I compare in Table 2 (Panel B), the value of cross-listed firms to non-cross-listed firms in event time. I denote the list year as ‘0’, and compare cross-listed to non-cross-listed firms for the five years before to five years after listing. I outline the mean value of cross-listed firms, and calculate the mean difference [Mean Diff] between both sets of firms. The (unreported) median differentials are similar, and are available from the author upon request. In the final column, I calculate the ATT in event time ( CL NCL E(q q |D 1, X) − =). Specifically, in the year preceding the list year, I match cross-listed to non-cross-listed firms based upon size (total assets), growth (two-year sales growth), legal origin, and industry group. I also include time dummies to ensure that the matches are generated prior to listing. For each group of firms, I outline the ATT for each year (including the list year) up to five years post-listing. In Appendix 1a, I outline the corresponding first-stage probit estimates4. The column labeled ‘Matches’ refers to the number of cross-listed to non-cross-listed matches. In Appendix 1b, I present alternative probit specifications. The ATT 4 In their study, Li and Zhao (2006, pg. 358) estimate separate propensity score models for each year. I carry out a similar exercise in Table 2 (Panel A). They refrain from estimating a pooled propensity score model over the entire period because of the year-by-year analysis provides a “flexible specification for business cycle”. Although I am aware of the limitations of the pooled specification to adequately account for business cycle effects, I am primarily motivated in this paper to examine the valuation effects in event time, and not in calendar time. 6 estimates are similar, irrespective of the probit specification employed. Finally, I also present in Figures 1-3, the time series behaviour of Tobin’s for the average and median cross-listed firm, from ten years prior to listing to ten years after listing. Panel B suggests the following. First, Level 1 firms list after a period of poor performance (i.e. falling value). The absolute and relative value of Level 1 firms, falls in the pre-listing period, and continues to fall postlisting. The greatest fall-off in value occurs in the pre-listing period. For example, in the five-year period immediately prior to listing, the average Level 1 firm loses just less than 25% of its value. After listing in the U.S., value continues to fall, but at a much reduced pace. Specifically, after five years of listing, average value falls by 14.53%. The evidence from Figure 1 suggests that after five years of listing, mean and median value tends to level off. The fall in absolute value over this period results in Level 1 firms losing their valuation premium over noncross-listed firms. The average Level 1 firm is worth more than domestic firms pre-listing, but on a par with these firms, post-listing. In contrast to Level 1 firms, firms that trade under Rule 144a appear to ‘time’ their decision to list in the U.S; for the average (and unreported median) firm, value increases dramatically in the years prior to listing, followed by a corresponding fall off thereafter. This trend is depicted graphically in Figure 3. In essence, these firms take advantage of favourable market conditions, and thus raise equity, via private placements during boom markets. Finally, the univariate summary measures for Level 2/3 firms highlight that Level 2/3 firms are not worth more than non-cross-listed firms in every year up to five years post-listing. This is in contrast to the calendar year cross listing premia from Panel A. Thus, the question remains; are the results reconcilable with one another? I believe so. From Panel A, I find that the cross listing premia manifest in the later half of the sample period, with the exception of 1994. In the later years of the sample, the majority of the sample of Level 2/3 firms has been listed in the U.S. for five years or more. The calendar year cross listing premia may therefore be a result of long-term valuation gains from listing, which have yet to materialize after five years of listing. I return to this in the next section. 3.3 Regression analysis In this section I examine the effect of cross listing on firm value. I begin with the following specification, whereby I model firm value as a function of firm characteristics: (1) it it 1 it 2 it 3 it it q X Level 1 Level 2/3 Rule 144a u=α+ β+δ +δ +δ + Where is a set of exogenous observable characteristics of the firm, it X i t Level 1 , , it Level 2/3 i t Rule 144a are standard dummy variables that take the value of 1 if the firm trades in the United States as a Level 1, Level 2/3, or under Rule 144a on Portal, respectively. is a standard idiosyncratic disturbance term, and is a vector of parameters to be estimated. it u123 {,, , , }αβδ δ δ I explicitly acknowledge the non-randomness of the cross-listed sample, and model their decision to cross list as follows: 7 * i titit * it it * it it CL Z CL 1 if CL 0 CL 0 if CL 0 = γ+η = > = < (2) Where ( * it CL it it it it Level 1 ,Level 2/3 ,Rule 144a CL ∈ ) is an unobserved latent variable, is a set of observable firm-level characteristics that determine the decision to cross-list in the United States, and is a disturbance term. Selection bias arises because of the correlation between it Z it η i titit Level 1 ,Level 2/3 ,Rule 144a it and u. This correlation can arise in two instances i.e. (1) selection on observables5 which arises through correlation between i t Z and i t u, or (2) through selection on unobservables i.e. correlation between i t η and i t u . Both instances render ordinary least squares estimates of the effect of cross listing on value, biased. In my analysis, I estimate the effect of listing on firm value using two approaches. First, I estimate firmfixed effect regressions. In this specification, I assume that the unobservables are time-invariant. Thus, the inclusion of firm-fixed effects is sufficient to adequately model, and thus control for unobservables. Unlike matching estimates, I do not assume away unobservables; I just assume that I can adequately control for them. Second, I must assume that the unobservables, in addition to being time-invariant, have no causal effect in precipitating cross listing (See Li and Prabhala (2005) for a discussion). Next, I explicitly model for unobservables by proxying for them. To do so I estimate a treatment effects model, whereby I augment the second stage equation with a selection correction term namely the inverse mills ratio, from a first-stage probit model. The inverse mills ratios are generated on a year-by-year basis (using yearly probit models), thus resulting in a series of time-variant unobservables in the second stage equation. Next, I outline both methods in greater detail. Firm-fixed effects. I begin with a standard fixed-effects specification. I augment Eq. (1) with time-fixed effects and estimate the following two-way fixed effects model6: i t i it 1 it 2 it 3 it t it q X Level 1 Level 2/3 Rule 144a=α +β +δ +δ +δ +α +υ (3) t α are time-fixed effects that account for contemporaneous correlation, and i α are firm specific fixed effects, which reflect differences across firms that are constant, but unobserved over time. Next, I estimate a pooled version given my concerns over violations of strict exogeneity7. I specify the individual specific effects as Mundlak (1978) corrections: T i ii i s1 1 Xa, where X X T= α= ζ+ = ∑it . Substituting into Eq. (3) yields the following: 5 If I assume selection on observables, I must assume that unobservables (private information) do not influence the decision to list and/or influence post-listing value. The significance of the ‘inverse-mills ratio’ in later tables suggests that this is not the case. However, in Table 2 I outline the average effect of the treatment on the treated (ATT) for cross-listed firms, in calendar and event time. The ATT is the difference in value between the cross-listed and matched non-cross-listed firm. These propensity score matching estimates assume that the decision to crosslist is a function of observable factors only. 6 The results from both the standard Hausman (1978) test, and Mundlak (1978) auxiliary regression specification confirm that in this instance a random effects specification is not appropriate. 8 it it 1 it 2 it 3 it i it q X Level 1 Level 2/3 Rule 144a X = α+ β+δ +δ +δ + ζ+μ (4) Treatment Effects In this section I outline a standard treatment effects model, whereby I correct for the probability of listing based upon unobservable factors. This approach is similar, but not identical to the standard Heckman (1979) twostage estimation procedure8. I begin by referring to Eq. (2). Now I assume that the decision to cross-list in the United States is a function of unobservable characteristics. Campa and Kedia (2002), Colak and Whited (2005), and Villalonga and Amit (2006) estimate similar ‘pooled Heckman’ models. Thus, the impact on firm value conditional on being cross-listed in the United States as: (5) it it it 1 1 it it it E(q |CL 1) X CL E( |CL 1)==α+ β+δ +υ = Given Eq. (2) and assuming that the errors terms from both Eq. (1) and (2) are bivariate normal, the unobservable component from Eq. (2), the generalized residual from the probit model is defined as: i tit 1 it E(q |C 1) ( Z ) υ = =ρσ λ β (6) Where: i t 1it i t (Z) (Z) (Z) ϕ β λβ = φβ (7) The latter is commonly referred to as the Inverse Mills Ratio, and is a series of time-specific ‘inverse mills ratios’. In the second-stage, I add this selection-correction term, yielding the following: (8) it it 1 1 it 1 2 i it qXC c=α+ β +δ +λβ + +υ In addition, I specify the unobserved heterogeneity as in Mundlak (1978) i.e. T i ii i s1 1 cX a, where X X T= =ζ+ = it ∑ , and estimate the following: it it 1 1 it 1 2 i it qXC X=α+ β +δ +λβ + δ+υ (9) In their pooled ‘Heckman’ specification, Dewenter et al. (2005) control for unobserved heterogeneity by estimating least squared dummy variable model, whereby, as the name suggests they include a dummy-variable for each firm9. Given the disadvantage of using this approach in large samples, I specify the unobserved heterogeneity by including Mundlak (1978) correction terms as an additional set of regressors in Eq. (9). I estimate treatment effects models for each set of cross-listed firms separately. The coefficient estimates for Eqs. (3, 4, 9) are presented in Table 3 and suggest the following. First, and in line with Doidge et al. (2004, 2006, 2007) I find a statistically significant cross listing premium for Level 2/3 issues. 7 I formally test for this possibility, following Wooldridge (2002), by inserting the one-year forwarded cross-listing variables as independent variables and testing whether their coefficients are jointly equal to zero. 8 Technically, the Heckman (1979) two-stage procedure is not a treatment effects model. In addition to the standard Heckman (1979) model, a treatment effects model includes, unlike the Heckman (1979) model, the selection indicator from the first stage probit as a regressor in the secondstage regression. 9 I would like to thank Kathryn Dewenter and Walter Novaes for clarifying to me their estimation procedure. 9 References 1. Blundell, R., and Costa Dias, M., 2000. Evaluation methods for non-experimental data. Fiscal Studies 21, 427-468. 2. Bris, A., and Brisley, N., 2006. A theory of optimal expropriation, mergers and industry competition. Yale working paper. 3. Campa, J.M., and Kedia, S., 2002. Explaining the diversification discount. Journal of Finance, 4, 1731-1762. 4. Chaplinsky, S., and Ramchand, L., 2007. From listing to delisting: foreign firms entry and exit from the U.S. University of Virginia working paper. 5. Coffee, J., 1999. The future as history: The prospects for global convergence in corporate governance and its implications. Northwestern University Law Review 93, 641-708. 6. Coffee, J., 2002. Racing towards the top: The impact of cross-listings and stock market competition on international corporate governance. Columbia Law Review, 102, 7, 1757-1832. 7. Colak, G., and Whited, T.M., 2005. Spin-offs, divestitures, and conglomerate investment. Review of Financial Studies (Forthcoming). 8. Dehejia, R., 2005. Practical propensity score matching: a reply to Smith and Todd. Journal of Econometrics, 125, 355-364. 9. Demirguc-Kunt, A., and Maksimovic, V., 1998. Law, finance, and firm growth. Journal of Finance, 53, 2107-2137. 10. Dewenter, K., Kim, C., Lim, U., and Novaes, W., 2005. Committing to protect investors in emerging markets: can local exchanges provide value-enhancing bonding mechanisms? University of Washington working paper. 11. Doidge, C., Karolyi, G.A., Stulz, R.M., 2004. Why are foreign firms listed in the U.S. worth more? Journal of Financial Economics, 71, 205-238. 12. Doidge, C., Karolyi, G.A., and Stulz, R.M., 2006. The valuation premium for non-U.S. stocks listed in U.S. markets. NYSE Research Paper. 13. Doidge, C., Karolyi, G.A., and Stulz, R.M., 2007. Has New York become less competitive in global markets? Evaluating foreign listing choices over time. Fisher College of Business, Ohio State University working paper. 14. Hausman, J.A., 1978. Specification tests in econometrics. Econometrica, 46, 1251-1271. 15. Heckman, J.J., 1979. Sample selection as a specification error. Econometrics 47, 153-161. 16. Heidle, H.G., 2003. Discussion of ‘Globalization and the value of a U.S. listing: Revisiting Canadian evidence’. Journal of Banking and Finance 27, 1663-1665. 17. Hostak, P., Lys, T., and Yang, Y., 2006. Is the Sarbanes-Oxley Act scaring away lemons or oranges? An examination of the impact of the Sarbanes-Oxley Act on the attractiveness of U.S. capital markets to foreign markets. Northwestern University working paper. 18. Kothari, S.P., Warner, S.B., 2005. Econometrics of event studies. Handbook of Corporate Finance: Empirical Corporate Finance, B. Espen Eckbo, ed., Elsevier/North-Holland. 16 19. Kristian Hope, O., T. Kang, and Zang, Y., 2007. Bonding to the improved disclosure environment in the United States: Firms’ listing choices and their capital market consequences. University of Toronto working paper. 20. La Porta, R., F. Lopez-de-Silanes, Shleifer, A., and Vishny, R., 1998. Law and finance. Journal of Political Economy, 106, 1113-1155. 21. La Porta, R., Lopez-de-Silanes, F, Shleifler, A, and Vishny, R., 2002. Investor protection and corporate valuation. Journal of Finance 57, 1147-1170. 22. Li, K., and Prabhala, N.R., 2005. Self-selection models in corporate finance. B. Espen Eckbo (ed.) Handbook of Corporate Finance: Empirical Corporate Finance, Chapter 1 2005. 23. Li, X., and Zhao, X., 2006. Propensity score matching and abnormal performance after seasoned equity offerings. Journal of Empirical Finance, 13, 351-370. 24. Lins, K., D, Strickland, and Zenner, M., 2005. Do non-U.S. firms issue equity on U.S. stock exchanges to relax capital constraints? Journal of Financial and Quantitative Analysis, 40, 109-133. 25. Marosi, A., and Massoud, N., 2006. You can enter but you cannot leave – U.S. securities markets and foreign firms. University of Alberta working paper. 26. Miller, D., 1999. The market reaction to international cross listing: evidence from depositary receipts. Journal of Financial Economics, 51, 103-123. 27. Mittoo, U.R., 2003. Globalization and the value of U.S. listing: revisiting Canadian evidence. Journal of Banking and Finance, 27, 1629-1661. 28. Mundlak, Y., 1978. On the pooling of time series and cross section data. Econometrica, 1, 69-85. 29. Reese, W., and Weisbach, M., 2002. Protection of minority shareholder interests, cross-listings in the United States, and subsequent equity offerings. Journal of Financial Economics 66, 65-104. 30. Sarkissian, S., and Schill, M.J., 2007. Are there permanent valuation gains to overseas listings? Evidence from market sequencing and selection. Review of Financial Studies (Forthcoming). 31. Serra, A.P., 1999. Dual-listings on international exchanges: the case of emerging markets’ stocks. European Financial Management 5, 165-202. 32. Villalonga, B., and Amit, R., 2006. How do family ownership, control and management affect firm value? Journal of Financial Economics, 80, 2, 33385-417. 33. Witmer, J., 2006. Why do firms cross-delist? An examination of the determinants and effects of cross delisting. Bank of Canada working paper. 17 Table 1: Sample Description Country NCL SIC 6 % Level 1 % Level 2/3 % Rule 144a % Total CL Sample Argentina 60 7 1.31 0 0.00 11 7.75 5 2.76 16 76 Brazil 246 29 5.39 26 10.00 25 17.61 3 1.66 54 300 Chile 113 35 2.48 2 0.77 17 11.97 2 1.10 21 134 China 89 4 1.95 8 3.08 12 8.45 4 2.21 24 113 Colombia 27 6 0.59 1 0.38 1 0.70 4 2.21 6 33 Hong Kong 540 167 11.83 97 37.31 7 4.93 1 0.55 105 645 Hungary 23 4 0.50 2 0.77 1 0.70 9 4.97 12 35 India 278 23 6.09 5 1.92 9 6.34 50 27.62 64 342 Israel 83 16 1.82 1 0.38 8 5.63 0 0.00 9 92 Korea 636 74 13.94 4 1.54 7 4.93 20 11.05 31 667 Malaysia 638 153 13.98 12 4.62 0 0.00 0 0.00 12 650 Mexico 71 14 1.56 18 6.92 25 17.61 11 6.08 54 125 Peru 45 8 0.99 3 1.15 1 0.70 1 0.55 5 50 Philippines 110 70 2.41 5 1.92 1 0.70 6 3.31 12 122 Poland 56 15 1.23 1 0.38 1 0.70 11 6.08 13 69 Singapore 407 67 8.92 19 7.31 1 0.70 1 0.55 21 428 South Africa 313 151 6.86 37 14.23 8 5.63 3 1.66 48 361 Taiwan 404 60 8.85 0 0.00 6 4.23 42 23.20 48 452 Thailand 296 98 6.49 14 5.38 0 0.00 1 0.55 15 311 Turkey 128 30 2.81 5 1.92 1 0.70 7 3.87 13 141 Total 4,563 1,031 100% 260 100% 142 100% 181 100% 583 5,146 This table outlines the final sample. To enable matching for financial cross-listed firms, I include a set of non-cross-listed financial firms (outlined in column 3). These firms are not included in the valuation regressions. All firms are obtained from the Worldscope Country Lists. All information on firms cross-listed in the U.S. are obtained from the Bank of New York, and cross-referenced with data provided by Deutsche-Bank, JP Morgan and Citibank. Rule 144a ADRs trade on Portal; Level 1 ADRs trade over-the-counter as pink sheet issues, and Level 2/3 trade on the NYSE or NASDAQ. 18 Table 2: Valuation comparison of cross-listed and non-cross-listed firms in calendar and event time. Level 1 Level 2/3 Rule 144a CL Med CL Mean Mean Diff PS ATT CL Med CL Mean Mean Diff PS ATT CL Med CL Mean Mean Diff PS ATT Panel A Calendar Time 1990 1.40 1.49 (0.22) (0.20) 1.42 1.42 (0.29) - - - - - 1991 1.30 1.40 (0.45) (0.57) 1.57 1.66 (0.19) - 1.17 1.17 (0.68) - 1992 1.40 1.78 (0.05) (0.06) 1.86 1.81 (0.02) - 1.76 1.82 (0.01) - 1993 1.67 1.83 (0.06) (0.31) 2.20 2.11 0.22 - 1.71 1.77 (0.12) 0.64 1994 1.70 1.81 (0.28) (0.26) 1.99 2.05 (0.04) 0.70 2.40 2.61 0.52 0.67 1995 1.56 1.75 (0.11) 0.20 1.64 1.68 (0.18) 0.03 1.84 1.96 0.10 0.19 1996 1.43 1.73 (0.10) 0.32 1.63 1.79 (0.04) 0.19 1.51 1.70 (0.13) (0.18) 1997 1.54 1.79 0.00 0.51 1.70 1.96 0.17 0.32 1.53 1.76 (0.03) 0.16 1998 1.20 1.42 (0.03) 0.33 1.28 1.44 (0.01) 0.26 1.26 1.50 0.05 0.18 1999 1.27 1.53 0.00 (0.07) 1.36 1.61 0.08 0.22 1.35 1.60 0.07 0.00 2000 1.29 1.52 0.01 0.01 1.39 1.70 0.19 0.41 1.31 1.64 0.13 0.10 2001 1.21 1.38 0.01 (0.05) 1.25 1.42 0.05 0.16 1.21 1.37 0.00 0.19 2002 1.23 1.41 (0.01) 0.03 1.20 1.38 (0.04) 0.37 1.24 1.43 0.01 0.09 2003 1.36 1.51 (0.01) (0.02) 1.29 1.51 (0.01) 0.32 1.38 1.54 0.02 0.21 Panel B Event Time Level 1 Level 2/3 Rule 144a Mean CL Mean Diff PS ATT Mean CL Mean Diff PS ATT Mean CL Mean Diff PS ATT -5 2.31 0.65*** - 1.81 0.15 - 1.57 (0.09) - -4 1.89 0.23*** - 1.61 (0.05) - 1.66 0.00 - -3 1.78 0.12** - 1.61 (0.05) - 1.91 0.25*** - -2 1.72 0.06* - 1.73 0.07 - 2.24 0.58*** - -1 1.77 0.11** - 1.73 0.07 - 2.10 0.44*** - 0 1.72 0.06 (0.12) 1.70 0.04 0.32 2.18 0.52*** 0.42** 1 1.63 (0.03) (0.02) 1.60 (0.06) 0.15 1.96 0.30*** 0.22 2 1.55 (0.11) 0.07 1.57 (0.09) 0.12 1.72 0.06 0.21 3 1.61 (0.05) 0.14 1.62 (0.04) 0.18 1.64 (0.02) 0.29 4 1.60 (0.06) (0.04) 1.56 (0.10) 0.06 1.50 (0.16) (0.03) 5 1.47 (0.19) 0.14 1.48 (0.18) 0.15 1.44 (0.22) (0.07) Event Time (Before-After Value) Mean Median Mean Median Mean Median Before 1.87 1.59 1.66 1.45 1.89 1.43 After 1.56 1.33 1.59 1.36 1.62 1.34 Difference (0.31)*** (0.26)*** (0.07) (0.09) (0.27)*** (0.09) This table compares the mean performance of cross-listed (Level 1, Level 2/3, and Rule 144a) to non-cross-listed firms in each year from 1990 to 2003, and in event time (5 years pre-listing to 5 years post-listing). Firm value is proxied using Tobin’s q, where. For each subset of cross-listed firms, I also outline the average effect of the treatment on the treated [] using propensity score matching [PS ATT], in calendar and event time. In the Appendix I outline exactly how I calculate the ATT. For cross-listed firms, I also outline their median value. ***, **, * Represents significance at the 1%, 5%, and 10% level, respectively. CL NCL E(q q |D 1, X)−= 19 Table 3: Regression estimates of the impact of cross-listing on firm value. Level 1 Level 2/3 Rule 144a POLS FE TE POLS FE TE POLS FE TE Level 1 0.04 [0.55] -0.07 [1.55] 0.07 [0.85] Level 2/3 0.24 [2.89]*** 0.16 [2.07]** 0.15 [1.57] Rule 144a 0.03 [0.43] 0.08 [1.23] 0.04 [0.57] Global q 1.13 [8.05]*** 0.60 [7.56]*** 1.10 [7.75]*** 1.12 [8.11]*** 0.60 [7.58]*** 1.05 [7.55]*** 1.13 [8.07]*** 0.60 [7.62]*** 1.12 [7.96]*** Sales Growth 0.54 [4.83]*** 0.45 [7.21]*** 0.44 [3.90]*** 0.53 [4.74]*** 0.45 [7.29]*** 0.50 [4.43]*** 0.53 [4.80]*** 0.45 [7.26]*** 0.42 [3.70]*** Total Assets -0.10 [5.95]*** -0.26 [14.40]*** -0.10 [6.25]*** -0.26 [14.64]*** -0.10 [5.75]*** -0.26 [14.54]*** Anti-Director 0.12 [8.81]*** 0.12 [9.21]*** 0.12 [9.03]*** Lambda () λ - - 0.01 [1.17] - - 0.04 [5.49]*** - - 0.07 [5.62]*** Time Dummies No Yes Yes No Yes Yes No Yes Yes # Obs 7,001 7,426 7,426 7,001 7,426 7,426 7,001 7,426 7,426 2 R 0.14 0.07 0.09 0.14 0.08 0.10 0.14 0.08 0.10 Pr F> 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Pr F(Mundlak)> 0.000 - - 0.000 - - 0.000 - - This table presents coefficient estimates from panel regressions. I present three sets of panel data estimates; pooled ordinary least squares with Mundlak (1978) corrections [POLS], firm-fixed effects [FE], and Treatment Effects [TE]. The Treatment Effects regressions are estimated as three separate regressions based upon the different ADR sub-sample of firms. For each ADR level, we estimate a first-stage probit model where the decision to list is determined in terms of size (Log (Total Assets)), and Legal Origin (French, German). To satisfy the exclusion restrictions, these variables are excluded in the second-stage regressions. The first stage probit estimates are available from the author upon request. Firm value is proxied using Tobin’s q. The independent variables are defined in the text. A full set of year specific dummy variables are reported (except in the case of the pooled ordinary least squares estimates) but not reported. I report t-statistics in parentheses. The pooled ordinary least squares t-statistics are calculated using standard errors clustered at the firm level. # Obs is the number of observations and is the coefficient of determination (I report the overall for the firm-fixed effect estimates). I report two F-Stats: (joint significance of all RHS variables) and which tests the joint significance of the included (unreported) Mundlak (1978) time-averaged correction terms. ***, **, * Represents significance at the 1%, 5%, and 10% level, respectively. 2 R 2 RPr F> Pr F(Mundlak)> 20 Table 4: Regression estimates based on country legal characteristics. Anti-Director Rights Index Judicial Efficiency English Common Law All Above Median Below Median Above Median Below Median Common Law Civil Law Level 1 -0.02 [0.20] 0.25 [1.78]* -0.03 [0.24] 0.26 [3.01]*** 0.04 [0.39] 0.08 [0.61] 0.28 [2.62]*** Level 2/3 0.20 [1.67]* 0.16 [1.39] 0.39 [2.15]** 0.12 [1.25] 0.51 [3.04]*** 0.15 [1.74]* Rule 144a -0.08 [0.70] 0.24 [3.09]*** -0.03 [0.25] 0.18 [2.29]** -0.10 [0.84] 0.25 [3.42]*** -0.03 [0.40] Global q 0.79 [4.10]*** 1.38 [7.49]*** 0.91 [4.66]*** 1.26 [6.82]*** 0.84 [4.77]*** 1.36 [6.99]*** 1.13 [8.05]*** Sales Growth 0.11 [0.75] 0.87 [5.87]*** 0.11 [0.66] 0.76 [5.40]*** 0.31 [1.97]** 0.73 [5.03]*** 0.44 [3.74]*** Total Assets -0.10 [3.62]*** -0.12 [6.57]*** -0.10 [3.64]*** -0.11 [5.93]*** -0.10 [3.54]*** -0.12 [6.38]*** -0.11 [6.35]*** Level 1 * Sales Growth * Anti -0.21 [2.83]*** Rule 144a * Sales Growth * Anti 0.16 [1.30] Time Dummies No No No No No No No # Obs 3,863 3,563 3,259 4,167 3,835 3,591 7,027 2 R 0.12 0.17 0.13 0.14 0.12 0.16 0.11 Pr F> 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Pr F(Mundlak)> 0.000 0.000 0.000 0.000 0.000 0.000 0.000 This table presents coefficient estimates from pooled ordinary least squares regressions (with standard errors clustered at the level of the firm) based on legal characteristics. These characteristics are Anti-Director Rights Index, Judicial Efficiency, and Legal Origin (English common law or not). All three variables are taken from and defined in La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1997). Firm value is proxied using Tobin’s q. The independent variables are defined in the text. . # Obs is the number of observations and is the coefficient of determination (I report the overall for the firm-fixed effect estimates). I report two F-Stats: (joint significance of all RHS variables) and which tests the joint significance of the included (unreported) Mundlak (1978) time-averaged correction terms. ***, **, * Represents significance at the 1%, 5%, and 10% level, respectively. 2 R2 R Pr F>Pr F(Mundlak)> 21 Table 5: Cross-listing and the evolution of Tobin’s q. Level 1 Level 2/3 Rule 144a All Com Civil All Com Civil All Com Civil List year 0.03 [0.37] -0.02 [0.18] -0.20 [2.58]*** 0.38 [1.83]* 0.66 [1.73]* 0.33 [1.43] 0.34 [2.33]** 0.60 [1.46] -0.20 [2.58]*** 1 year after list 0.08 [0.99] 0.06 [0.55] -0.18 [2.31]** 0.21 [1.42] 0.49 [1.46] 0.20 [1.40] 0.23 [1.83]* -0.07 [0.32] -0.18 [2.31]** 2 years after list 0.07 [0.73] 0.01 [0.07] -0.02 [0.21] 0.02 [0.14] 0.35 [1.37] 0.06 [0.42] 0.12 [1.31] -0.16 [0.93] -0.02 [0.21] 3 years after list 0.03 [0.30] -0.06 [0.47] 0.05 [0.38] 0.10 [0.86] 0.41 [1.82]* 0.10 [0.85] 0.08 [0.93] -0.25 [1.94]* 0.05 [0.38] 4 years after list 0.17 [1.50] 0.16 [1.07] 0.04 [0.22] 0.06 [0.56] 0.56 [1.61] -0.02 [0.23] 0.06 [0.80] -0.03 [0.17] 0.04 [0.22] 5 years after list 0.13 [1.16] 0.08 [0.57] 0.09 [0.52] 0.02 [0.23] 0.52 [4.98]*** -0.02 [0.23] 0.01 [0.05] -0.24 [2.52]** 0.09 [0.52] > 5 years after list 0.22 [1.69]* 0.11 [0.67] 0.32 [1.60] 0.21 [1.27] 1.09 [4.62]*** 0.04 [0.41] -0.06 [0.87] -0.17 [1.39] 0.32 [1.60] Global q 1.16 [8.24]*** 0.86 [4.79]*** 1.39 [6.93]*** 1.16 [8.28]*** 0.82 [4.75]*** 1.38 [6.95]*** 1.15 [8.25]*** 0.85 [4.72]*** 1.39 [6.93]*** Sales growth 0.41 [3.75]*** 0.33 [2.15]** 0.73 [5.02]*** 0.39 [3.52]*** 0.31 [1.97]** 0.71 [4.83]*** 0.39 [3.52]*** 0.30 [1.94]* 0.73 [5.02]*** Total Assets -0.10 [6.53]*** -0.09 [3.36]*** -0.10 [5.33]*** -0.10 [6.50]*** -0.10 [3.62]*** -0.10 [5.46]*** -0.10 [6.22]*** -0.08 [3.12]*** -0.09 [5.33]*** Time Dummies No No No No No No No No No # Obs 7,452 3,853 3,599 7,452 3,853 3,599 7,452 3,853 3,599 2 R 0.11 0.11 0.15 0.11 0.12 0.15 0.11 0.11 0.15 Pr F> 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Pr F(Mundlak)> 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 This table presents coefficient estimates from pooled ordinary least squares regressions (with standard errors clustered at the level of the firm) for the full sample and by legal origin. ‘Com’ is English common law. Civil is civil law. Firm value is proxied using Tobin’s q. The independent variables are defined in the text. The single year cross-listing dummies equal one in the referred year, and zero otherwise. The ‘> 5 years after listing’ dummy equals one after the fifth year of listing and zero before. . # Obs is the number of observations and is the coefficient of determination (I report the overall for the firm-fixed effect estimates). I report two F-Stats: (joint significance of all RHS variables) and which tests the joint significance of the included (unreported) Mundlak (1978) time-averaged correction terms. ***, **, * Represents significance at the 1%, 5%, and 10% level, respectively. 2 R2 R Pr F>Pr F(Mundlak)> 22 Figure 1 Value of Level 1 firms in calendar and event time 23 Figure 2 Value of Level 2/3 firms in calendar and event time 24 Figure 3 Value of Rule 144a firms in calendar and event time 25