DETERMINANTS OF ADR RETURNS BEFORE AND AFTER DOMESTIC STOCK SEASONED EQUITY OFFERINGS: EVIDENCE FROM ASIAN AND LATIN AMERICAN EMERGING MARKETS

This paper examines the critical determinants of American depository receipt (ADR) returns before and after domestic stock seasoned equity offerings (SEOs) for Asian and Latin American emerging economies during 1990–2007, which has never been probed in related issues. We employ the Time Series Cross Section Regressions and General Method of Moments methods to document that domestic stock returns play a vital role in explaining Latin American ADR returns, while US investor sentiment is crucial in explaining Asian ADR returns. Local investor sentiment is found to be considerably important than domestic stock returns in Asian ADR returns, while Latin American local investor sentiment (US investor sentiment) is more important before (after) domestic stock SEOs. The results do not support the view that ADR-reconciled earnings per share (EPS) and stock EPS provide signifi cant information to explain ADR returns in Latin American and Asian emerging markets both before and after SEOs. Furthermore, international market differences in a specifi c geography should be considered when diversifying investments and effi ciency accounting communication with accounting convergence does not need to be emphasized.


Introduction
The determinants of American depository receipt returns (hereafter ADRNs) have constituted one of the most hotly debated issues in the past two decades, yet never before has a consensus been reached as shown in Table 1. We fi nd that even if the same country and method are employed, the results vary. Regarding the comparison of domestic stock earnings per share (DEPS) and ADR-reconciled earnings per share (AEPS), the results of Chan and Seow (1996) and Luchs (2004), both of whom use regression models to test UK data, are confl icting. For instance, one shows that DEPS dominates the AEPS in infl uencing ADRNs, while the other indicates that DEPS and AEPS are equally important for UK ADRNs. The fi ndings of previous studies that explore whether local investor sentiment (LD) has dominated US investor sentiment (USI) or vice versa are quite distinct. For example, the results of Jiang (1998) and Ely and Salehizadeh (2001), both of whom use error-correction models to examine the same countries (e.g., Germany, UK, and Japan), are also confl icting, as one shows that LD dominates USI in infl uencing ADRNs, while the other shows that the opposite situation holds. The differences in these fi ndings might be due to the limitations of the specifi c information transmitted or the incompleteness of the variables considered.
The purpose of this paper is to expansively investigate and compare the changes in the infl uence of ADRN determinants for Latin American and Asian emerging markets by controlling domestic stock seasoned equity offering (SEO) events. The role of ADRs in the development of emerging markets brings advantages of liquidity, transparency, and ease of trade that characterize US markets. Firms issuing ADRs are required to fi le their domestic GAAP fi nancial statements and reconcile their US GAAP accounting procedures with the SEC, thus providing investors with two sets of accounting information. However, the informativeness of the two sets of accounting information is subject to controversy.
This study comprehensively includes two accounting variables (DEPS and AEPS), two investor sentiment variables (proxy by market indices LD and USI), and domestic stock returns (SRs), as well as compares the variable impact changes on ADRNs before and after domestic stock SEOs. If the ADRNs do not fl uctuate as do the SR, then there is a question regarding the causes of ADR deviation in response to SRs, LD, and /or USI, DEPS, and /or AEPS. Time Series Cross Section Regression (TSCSREG) is applied in comparing the signifi cance among variables. Evidence of the importance among variables may have a signifi cant bearing upon accounting harmonization and diversify investment. If DEPS contains more information than AEPS, then it signifi es that earnings based on foreign GAAP may convey information that may be lost in reconciliation with US GAAP fi nancial statements (Chan, Seow 1996). However, if DEPS and AEPS both contain sparse information regarding ADRNs, then the arguments for accounting harmonization may focus on fi rms' cost reduction and not on the asymmetry reduction arguments (Kirch 2007). An understanding of an ADR's role in diversifi cation, its interrelationships with the market of origin, and its pricing factors certainly benefi t many players in the ADR market (Jiang 1998 Pope and Rees (1992) 1987 Cross-sectional regression UK Local. GAAP earnings changes have incremental information content. US GAAP earnings adjustments appear to add to explanatory power marginally. Webb et al. (1995Webb et al. ( ) 1985Webb et al. ( -1989 Cross-sectional Alaganar and Bhar (2001) 1994 Bivariate GARCH model Australia The 'law of one price' holds. ADRs have an economically signifi cant higher reward than underlying stocks. ADRs have a low correlation with the US market under high states of global and regional shocks. Ely and Salehizadeh (2001 Cointegration and error-correction models UK, Japan, and Germany ADRs are cointegrated with ordinary shares. Local markets are a more important source of information.
Authors Period Empirical methods
ADR may not exactly be foreign shares. Luchs (2004Luchs ( ) 1993Luchs ( -1998  Earning release Bae et al. (2008) 1998 Time-series model Australia, France, Japan, and UK Local market plays a bigger role in determining ADR returns. ADR returns are signifi cantly positively related to exchange rate.
End of Table 1 Two streams of earlier research are related to this paper. One has to do with a comparative study of ADR-related accounting variables, and the other with a comparative study of SRs, LD, and USI. Using data from the UK for 1987, Pope and Rees (1992 found DEPS dominates AEPS in explaining ADRNs, indicating that the market exhibits a limited response to changes in accounting information. Empirical studies were later extended to cover other countries (Kirch 2007;Luchs 2004). However, the fi ndings have been empirically inconclusive. Luchs (2004), for example, held the view that DEPS and AEPS provide the same level of explanatory power in regard to ADRNs. Furthermore, Kirch (2007) claimed that there were no surprises related to ADRNs during releases of both DEPS and AEPS.
As for one of the leading studies on a comparison of the infl uence of SRs, LD, and USI, Webb et al. (1995) found a signifi cant relationship between ADRs and US market returns, thus signifying the leading role played by USI in relation to ADRNs. Jiang (1998) extended the analysis to apply the VAR and GARCH models that took the exchange rate into consideration and came up with opposite fi ndings. Later results of related studies were found to be fruitful (Alaganar, Bhar 2001;Choi, Kim 2000;Ely, Salehizadeh 2001), although the empirical fi ndings vis-à-vis such relationships were found to be mixed if not downright contradictory. Most of the fi ndings examined the related ADR issue, either without specifi c events (Choi, Kim 2000;Kim et al. 2000;Patro 2000), or with selected events, such as the US market's reaction to ADR initial public offerings (IPOs) (Foerster, Karolyi 2000;Miller 1999;Sundaram, Logue 1996), the domestic stock reaction to ADR IPOs (Alexander et al. 1988;Foerster, Karolyi 1999), fi nancial crises (Wang 2003), and profi t warnings regarding the underlying stock (Jackson, Madura 2003). However, studies related to fi nancial crises and profi t warnings are rare. Furthermore, for ordinary specifi c event settings based on a discussion of domestic stocks, none of the studies include both accounting and market variables at the same time.
This study is related to, but distinct from, the extensive and growing literature on crosslisting. We probe into more generalized events to understand the intrinsic nature of ADRNs. To date and to the best of our knowledge, no previous study has explained the infl uence of changes in accounting information and investor sentiment on ADRNs arising from the issuance of SEOs in the home country. Choi and Kim (2000) examined the determinants of ADRNs and mentioned that developed and emerging markets have divergent fi ndings. While Latin American and Asian emerging markets have extensive experience at launching their own ADR programs during the 1990s in terms of pace, breadth, and trading activity (Karolyi 2004), the differences between emerging markets are still open to question.
Level II and III ADR-listed fi rms must provide fi nancial statements with limited reconciliations moving from local GAAP to US GAAP. This study examines whether the reconciled fi nancial information fully communicates the intrinsic value of the ADRs, and if not what factors are mostly signifi cant in explaining ADRN changes. Barberis et al. (2005, hereafter BSW) classifi ed two distinctive views regarding the co-movement of equity prices: one adopts a traditional or fundamental-based approach based on friction-less economies with fully rational investors, while the other utilizes a behavioral (trading or investor sentiment-based) approach, namely, the market friction approach. BSW (2005) proposed that the behavioral approach is superior to the traditional approach. According to the fi ndings of BSW (2005), investor sentiment rather than fundamental information is a factor in explaining ADRNs. If country-specifi c market sentiment is crucial, then ADR price movements can be expected to be affected primarily by local and/or US market sentiment, rather than by fi rm stock price and accounting information.
This investigation aims to contribute to the literature in four ways. First, the mixed fi ndings of prior studies may be due to the variety of information transmitted from the local market, the US market, or domestic fi rms. However, few studies have set event limitations. We analyze the related variables' infl uence changes on ADRNs both before and after home stock SEOs, which are a common event, especially compared to infrequent fi nancial crises and profi t warnings. ADRN determinants can thus be further analyzed during a non-crisis period. Second, this work provides empirical evidence on the signifi cant informativeness of SR, fundamental-based (DEPS and AEPS) and sentiment-based information in explaining ADRNs. Third, prior studies point to the distinct market factors that infl uence ADRNs between developed and emerging markets. To our knowledge, none of the studies compare the differences among emerging markets in different regions. Finally, most previous studies use time series or cross-sectional regressions. Without controlling the time effects and fi rm effects, a divergence of fi ndings occurs. This study employs panel data to control for unobservable heterogeneity and eliminates the risk of obtaining biased results due to this heterogeneity (Moulton 1986). Time Series Cross Section Regression (TSCSREG) is employed to compare the importance of the variables, while Panel system General Method of Moments (GMM) models are applied to compensate for the greater accuracy 1 .
The remainder of this paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the data. Section 4 explains the model and methodology. Section 5 discusses the empirical results, and Section 6 concludes the paper.

Latin American and Asian ADRs in US Markets
In early 1997, as emerging markets boomed, most Asian and Latin American economies looked strong and investors had a strong appetite for Argentine, Brazilian, and Indonesian securities. Asian and Latin American emerging markets seemed also not to experience any major meltdown in 2005 (MacDonald 2005

The Effect from Seasoned Equity Offerings
SEOs occur when a listed fi rm issues additional shares. Smith (1977) was the fi rst to document signifi cant under-pricing of SEOs. Montier (2002) later stated that insiders realize the stock is overpriced and thus send a signal via a SEO, forcing the market to correct its misperceptions and creating negative announcement returns. Most relevant studies depict the SEO announcement effect as approximately minus 3% in US markets (Mola, Loughran 2004;Foerster, Karolyi 2000).
On the other hand, SEO studies yield diverse fi ndings regarding SEO underpricing. Spiess and Affl eck-Graves (1995), along with Denis and Sarin (2001), documented abnormal stock price reactions from post-SEO earning announcements as reliably negative only for the smallest quartile of equity issuers. Lee (1997) and Gombola et al. (1997Gombola et al. ( , 1999 demonstrated that mature fi rms do not share the same negative experience. Nevertheless, it remains unclear whether listed fi rms from emerging markets that have issued ADRs also exhibit SEO under-pricing. Jayakumar (2002) stated that cross-listing provides a credible commitment to greater fi rm information disclosure, less information asymmetry, increased transparency, and enhanced value via pure cash fl ow effects by cutting agency costs. With more analysts following ADR-listed fi rms, considerable fi rm information is publicly available (Ejara, Ghosh 2004). Based on the above fi ndings, we estimate that ADRNs react negatively after domestic stock SEOs.

Co-movement between ADRNs and SRs
BSW (2005) classifi ed co-movement theories into two groups. One contains the traditional or fundamentals-based approach of frictionless economies involving fully rational investors, while the other comprises behavioral (investor sentiment-based) or market friction approaches. Their fi ndings support the latter approach. Moreover, Coakley and Kougoulis (2004) explained that the fundamentals-based approach means that comovement in fundamental value is instantly mirrored by a return co-movement -that is, price equals fundamental value. Recent studies have indicated that the co-movement of securities prices signifi cantly exceeds their common fundamentals, casting doubt on the orthodox views of the rational pricing model.
Several related studies mention that SRs affect ADRNs. Choi and Kim (2000) showed that ADRs and underlying stocks have closely correlated returns. Alaganar and Bhar (2001), Kato et al. (1991), Maldonado and Saunders (1983), Park and Tavakkol (1994), and Wahab et al. (1992) confi rmed that ADR markets are priced effi ciently, and that a 'law of one price' exists, with dominant information fl ow from underlying stocks to ADRs. Foerster and Karolyi (1999) found that ADRs follow on the heels of abnormally positive stock price performance in the year leading up to the listing. Similar evidence exists for Canadian listings and ADRs, as reported by Alexander et al. (1988). Few investigations have explored ADRNs with SR co-movement in common specifi c events other than ADR IPOs, fi nancial crises, and profi t warnings, i.e., SEOs. This work, distinct from the previous literature, explores home stock SEO information transmission to ADRs. Large fi rms, as ADR-listed fi rms, enjoy an information-rich environment and fi rm-specifi c information is preempted in the market. Thus, we hypothesize that SRs affect ADRNs equally, regardless of whether they are measured before and after SEOs.

Investor Sentiment Infl uences ADRNs
Behavioral fi nance studies routinely challenge the conventional argument that market participants behave rationally. Studies on investor sentiment are therefore crucial: they educate us regarding biases in investor forecasts on the stock market and also in order to gain additional returns. Then again, sentiment indicators are widely recognized as a reliable contrarian indicator of market movement (Siegel 1998). The behavioral fi nance theory of DeLong et al. (1990, hereafter DSSW) predicts noise trader sentiment that persists in fi nancial markets and changes in investor sentiment are obviously diffi cult to forecast, otherwise they would be arbitraged away. Black (1986) and DSSW (1990) documented how noise traders acting together in response to non-fundamental signals cause asset prices to deviate from intrinsic values. A prime example of twin security puzzles involving Royal Dutch and Shell is mentioned by Froot and Dabora (1999). They showed Royal Dutch and Shell as being claims on the same cash fl ow. In an effi cient market, Royal Dutch and Shell should trade at a constant ratio, but Royal Dutch is more sensitive to moves in the US market, while Shell is more sensitive to the UK market. Specifi cally, twin securities are correlated with their present market, despite this behavior being unrelated to changes in fundamentals.
If the USI exerts less of an impact on ADRNs than LD, then ADRs can be employed to signal global risk diversifi cation; if USI exerts greater infl uences than LD, then movements in USI will dominate LD. Alaganar and Bhar (2001) and Kim et al. (2000) viewed ADRNs as being more strongly affected by SRs than by US market movements. Jiang (1998) confi rmed a relationship between ADR and home market returns. Conversely, if investor-based specifi c market sentiment is critical to equity pricing, then ADRNs can be expected to be infl uenced by USI, with ADRs being traded by US traders (Suh 2003). Based on the announcements of SEOs regarding listed fi rms, ADRNs can be infl uenced by LD, yet studies demonstrate that listed fi rms in emerging markets are strongly infl uenced by global leading US capital markets (Ehrmann, Fratzscher 2005). The correlation between ADRNs and US (local) market indices indicates that US (local) market sentiment infl uences ADRN movements. Offi cer and Hoffmeister (1987) found little covariance between ADR prices and the prices of the underlying securities.
Moreover, Webb et al. (1995) used information on ADRs and USI to confi rm US dominance in lead / lag relationships among equity markets. Since the US is a global leader among capital markets, USI may exert greater infl uences on ADRs than LD. Therefore, we examine and compare the change in the infl uence of LD and USI on ADRNs before and after domestic fi rm SEOs.

ADR Accounting Communication and Accounting Convergence
ADRs issued by US agencies for non-US fi rms cross-listed on US markets are negotiable securities issued by a US commercial bank backed by the equity shares of non-US parent fi rms. Four levels of ADR programs exist: Levels I, II, III, and 144a 3 . Levels II and III comply with all SEC registration and reporting requirements, including ADR program registration on SEC Form F-6, as well as annual reporting on Form F-20, with either partial or full reconciliation of fi nancial statements to increase investor confi dence by providing fi nancial statement information. Registration allows issuers to list ADRs on one US stock exchange (NYSE, AMEX, and NASDAQ) provided the listing requirements are met (Callaghan et al. 1999).
Previous research on international fi nancial reports identifi ed notable qualitative differences among nations in their fi nancial reporting standards (Meek, Saudagaran 1990). Baumol and Malkiel (1993) argued that further disclosure of ADR issuing requirements in accordance with US GAAP provides scant benefi t to investors, contending that information on foreign GAAP numbers may be untranslatable, because of differences in tax laws, corporate governance, inter-corporate ownership of securities, and other institutional features. Level II and III reconciliations may mislead US investors into making unwarranted inferences. Amir et al. (1993) as well as Meek (1983) also presented evidence that 20-F announcement dates contain no incremental information. Bradshaw et al. (2004) identifi ed several reasons why 20-F may not provide an effective substitute for accounting choice in primary fi nancial statements. Chan and Seow (1996) demonstrated a closer association with 12-month returns for domestic GAAP income than for returns reconciled to US GAAP income. This study expects that if the reconciled items from moving local GAAP to US GAAP in ADR fi nancial statements can effectively eliminate information asymmetries between local and US investors, then reconciliation for ADR listing fi rms can provide an intermediate by means of attracting US investment and can effectively transfer local information to its ADR price. Conversely, if reconciliation items cannot effectively eliminate such asymmetry between investors in the domestic country and the US, ADR price movements are caused by other factors. The International Accounting Standards Committee strives to achieve a global set of accounting principles, also known as the International Financial Reporting Standards (IFRS). Intuitive logic regarding convergence is that a thriving global capital market requires greater investor understanding and confi dence. Embracing common high-quality accounting standards can reduce costs not only for issuers, but also reduce the costs to fi rm personnel involved in complying with the requirements of multiple jurisdictions emanating from cross-listing. Moreover, such convergence reduces the costs of preparers and auditors in juggling the application of several sets of national standards within the same consolidated group. Convergence helps optimize the resources dedicated to setting standards and enables more resources to be devoted to establishing a unifi ed accounting model for a particular topic rather than those resources being spread out in the pursuit of separate national accounting standards for the same topic. IFRS can hopefully expedite fi rms in listing across borders, integrate national capital markets, and increase competition in those markets 4 . If changes in ADRNs cannot be attributed to earnings reconciled from local stock and parent fi rm earnings, then based on comparability and consistency, global accounting convergence is required.

Data
Firms must satisfy the following criteria for inclusion in our study samples: First, ADRs from 13 emerging markets in Latin America (Brazil, Chile, Colombia, Mexico, Peru) and Asia (China, India, Indonesia, South Korea, Malaysia, Philippines, Taiwan, and Thailand) included in the MSCI Emerging Markets Free Index (EMF) as of 2010 and with effective listings following 1990/12/31 are used for investigation 5 . ADRs used in this investigation must belong to Level II and III programs 6 -namely, larger, highprofi le ADRs listed on major US exchanges, rather than smaller ADRs listed and traded in OTC and private-placement issues. Data for Level II and III ADRs are listed on the website of the Bank of New York.
Second, a sample fi rm should conduct an SEO at least one year following the effective issue dates of the underlying ADRs. SEO samples were gathered from the Global New Issues database of Securities Data Company (SDC) during 1990 to 2007. According to D'Mello et al. (2003), SDC fi ling dates serve as the announcement dates. Since this work examines the performance of equity issuers over one year intervals pre-and post-SEO, it is a requirement that fi rms have not conducted an equity offering over one preceding year. In other words, upon a fi rm's completion of a SEO, it cannot re-enter the sample for at least one year following the SEO date. Third, this investigation gathers ADRNs and domestic listed fi rm earnings data from Compustat. SR, USI, and LD, as well as exchange rate (EXR) data, are obtained from Datastream. After eliminating fi rms for which relative data are missing, the fi nal sample contains 31 SEO events from (Indonesia, South Korea, Taiwan) The market index is one of the most popular indicators of daily investor sentiment (Minana 2003;Wang 2003). Minana (2003) and Wang (2003) employed the local index and S&P 500 as proxies for local and US market sentiment. Hence, this study takes the S&P 500 and local composite price index as a proxy for USI and LD. Table 2 indicates that the largest contingent comprises seven Chilean fi rms, followed by six Brazilian fi rms.
Other countries are ordinary listings. Among the sample fi rms, six are in Asia while 17 are in Latin America 7 . Table 3 lists yearly distributions for SEOs in this sample. Table 4 provides summary statistics of the sample organized according to listing year and listing exchange. Most listings occur on the NYSE, with only three fi rms listing on the NASDAQ. Manufacturing and telecommunications dominate the industry distribution listed in Table 5.

2000-2004 8
End of Table 2 Journal of Business Economics and Management, 2011, 12(2): 248-277    End of Table 6 Table 7 reveals that, fi rst, Latin American and Asian ADRNs are strongly positively correlated with LD for the entire period, meaning that LD is important in explaining Latin American and Asian ADRNs. Second, for the Latin American sample the only difference in the results for the two sub-periods is that AEPS is strongly and negatively correlated with ADRNs pre-SEO, while USI signifi cantly impacts ADRNs, meaning the change in AEPS decreases with increasing ADRNs. Third, USI is positively correlated with Asian ADRNs for the entire period.

Random Effects Model
Follow the methodology of Areal et al. (2007), this study uses panel data regression method, which takes the fi rm-specifi c and time effects to analyze the relation between ADRNs and its correlated variables. Panel data follow the same fi rms (countries) over time, helping to facilitate the analysis of dynamic responses and the control of unobserved heterogeneity (Arellano 2003). By combining time series of cross-section observations, panel data provide more informative data and are better suited to study the dynamic of change that simply cannot be observed in pure cross-section or pure time series data (Gujarati 2003). With the indication that the impacts on ADRN are different among SR, LD, USI, DEPS, and AEPS, additional analysis is needed to determine if indeed the differing infl uences are diverse between domestic stock before and after SEO. The basic structure for analyzing the panel data is given by the following equation (1): where N denotes the number of cross sections, T represents the length of the time series for each cross section, K is the number of exogenous or independent variables, and β is the estimated coeffi cient of vectors across cross-sectional observations, where i denotes the company and t denotes time. Therefore, y it is the dependent variable pooling N cross-sectional observations and T time-series observations, and X itK are the independent variables pooling N cross-sectional observations and T time-series observations. Here, u it is a random error term, and u it = u it + v it , where v it denotes the remainder of the disturbance.
The two-way (time-invariant / fi rm-specifi c) effect is randomly distributed and is uncorrelated with the vector of exogenous variables. Using the Hausman (1978) test, the sample data employ random effects, rather than fi x effects. The panel data have four model types: basic model, individual-effect model, fi xed effects model, and randomeffects model. Both fi xed-effects and random-effects models can be further divided into one-way and two-way types of models. The difference between ordinary least squares (OLS), fi xed-effects model, and random-effects model is that OLS calculation can only be analyzed either through cross-sectional or time-series data at a time. Therefore, when a combination of data appears, using OLS may overlook the differences embedded in the cross-sectional data and thus generate unreliable estimate results. While the fi xedeffects model and random-effects model can deal with the two data types simultaneously, given special consideration to the differences within cross-sectional data, we can eliminate discrepancies among samples. The estimation result gained will also be more effi cient and consistent (Cheng et al. 2010).

Time Series Cross Section Regression (TSCSREG) Analysis
The TSCSREG procedure is used to compare the signifi cance of variables. One of the main strengths of the TSCSREG design is that it allows for controlling heterogeneity bias, or the confounding effect of time-invariant variables omitted from the regression model (Nielsen 1999). The TSCSREG procedure involves panel datasets that comprise time series observations for each of several cross-sectional units. The performance of any estimation procedure for the model regression parameters depends on the statistical characteristics of the model's error components. The TSCSREG procedure requires that the time series for each cross section have the same number of observations and cover the same time range. The test performs F-tests of linear hypotheses regarding regression parameters in the preceding models. Each equation specifi es a linear hypothesis to be tested. All hypotheses in one test are tested jointly. Variable names in equations must correspond to regressors in the preceding model, and each name represents the coeffi cient of the corresponding regressor (Santiago-Castro, Brown 2007). Chan and Seow (1996) incorporated annual accounting earnings under US GAAP and foreign GAAP as the independent variables in order to probe the association between SR and foreign GAAP earnings versus earnings adjusted to US GAAP. We extend their models by including investor sentiment and accounting information variables. The hypotheses are tested using the following panel regression model (2): Here, the βs are the estimated coeffi cients and ε is the random disturbance term. Employing the balanced panel dataset, equation (2) is investigated to examine the impact change of the main variables on ADRNs.

Empirical Results
The Hausman (1978) test results provide the choice of the random effects model for all conventional signifi cance levels in four sample groups (Latin American and Asian data pre-and post-SEO). In this study the fi xed effect model is thus infeasible, because annual EPS variables exhibit minimal time variation. Consequently, this study employs a random effects model to examine the variables determining the ADRNs. The TSCSREG procedure is mainly employed to compare the signifi cant associations of related variables with ADRNs. Tables 8 and 9 provide Latin American and Asian results of the two-way random effects models and TSCSREG, where the dependent variable is daily ADRN. While the models (Latin American and Asian pre-and post-SEO results) do not have high F values and R 2 , they exhibit a key variable impact on ADRNs, and the probabilities are signifi cant at the 1% level. Furthermore, models excluding insignifi cant variables in Tables 8 and 9 are also examined to check the stability of results, as shown in Tables  10 and 11. Finally, this study uses the panel system GMM to perform the robustness check. Table 12 summarizes the fi ndings. Table 8 presents estimation results as well as some traditional tests for Latin American markets. The Durbin-Watson statistics are near 2, implying that there is little evidence of serial correlation in our sample since the null hypothesis of no autocorrelation in the residuals would not be rejected if the DW statistic is near 2 (Brooks 2002).  On the other hand, the data are plagued by heteroskedasticity, and thus we adopt robust estimates to correct for cross-sectional heteroskedasticity. The estimation results show that SR variables exert a signifi cantly positive effect on ADRNs both before and after domestic stock SEOs. The null hypotheses of SRs affecting ADRNs equally for domestic stock both pre-and post-SEO cannot be rejected. In other words, SR equally and considerably affects ADRN full period returns in the Latin American sample. However, LD markedly affects Latin American ADRN pre-but not post-SEOs and USI post-but not pre-SEOs. This study rejects the hypotheses that LD and USI affect ADRNs equally pre-and post-SEO. Variables which are shown to insignifi cantly impact ADRN include DEPS and AEPS for the full period, implying that investors do not refer to DEPS and AEPS when investing in Latin American ADRs, either pre-or post-SEO. Table 8 shows the TSCSREG tests to examine the relative importance of different factors on ADRNs and whether the difference between these factors is signifi cant. The results show that SR in Latin American markets has a uniform effect on full period ADRN and that the effect is signifi cantly lower than LD pre-SEO, but higher post-SEO. The intuition behind this observation is as follows. Firm capital will expand after SEOs, and investors will pay more attention to the issue of whether SR can remain after the expansion of fi rm capital since SEO may dilute the EPS or reduce the rate of return on equity. As with the pecking order theory of capital structure, the issuance of stock should be the fi nal choice when the fi rm needs to raise funds, because of asymmetric information between fi rm managers and investors. Thus, it could be inferred that investors of ADRs might pay more attention to the change of SRs after SEOs.

Latin American Results
The results on the TSCSREG tests of other pairs present a simpler pattern. The SR has a dominant effect over USI, DEPS, and AEPS on ADRNs after SEOs, but the difference is not so signifi cant pre-SEO except for the comparison with AEPS. This somewhat corresponds to the observation above that the effect of SRs is important post-SEO. The effect of investor sentiment dominates that of EPS, with LD being signifi cant pre-SEO and USI post-SEO. The infl uence of USI is higher than that of LD, but only in the post-SEO period. Overall, our fi ndings indicate that SR constitutes the most important determinant of Latin American ADRNs in the two sub-periods. However, EPS has no effect on ADRNs, whether it is represented by local accounting standards or reconciled to US GAAP. Moreover, the relative importance between DEPS and AEPS cannot be distinguished as well. Finally, the effect of investor sentiment is mixed in the two subperiods. Table 9 lists the Asian regression results both pre-and post-SEO. As shown, the problem of serial correlation is absent because the DW statistic is near 2. The robust standard error estimates are still employed because of heteroskedasticity problem. The SRs, LD, and USI exert roughly equal positive and signifi cant infl uences over ADRNs during the full period, except for LD as it strongly and negatively impacts ADRNs post-SEO. The null hypotheses that the SRs, LD, and USI affect ADRNs equally during domestic stock pre-and post-SEOs cannot be rejected. DEPS and AEPS exert an insignifi cant effect on ADRNs for the full period. Thus, the Asian sample exhibits limited informativeness of AEPS and DEPS in amplifying ADRNs, which is consistent with the fi ndings for Latin America.  (2007), Asian markets are more responsive to the US markets than to other regional markets. The TSCSREG results cannot distinguish the relative importance of DEPS versus AEPS both for the Asia and Latin American samples, which is consistent with Kirch's (2007) fi ndings that indicate that DEPS and AEPS contain little information about ADRNs. Choi and Kim (2000) examined the determinants of ADRNs and mentioned that developed and emerging markets have divergent fi ndings in terms of the determinants of ADRNs. Our fi ndings indicate that even emerging markets have diverse fi ndings in terms of the determinants of ADRNs. Ferguson (2000) stated that investor sentiment shifted more rapidly in Asia than it did in Latin America, because investors were already more familiar with the regional structural ineffi ciencies and diffi culties in Latin America than in Asia. This article to some extent refl ects Ferguson's point since our fi ndings demonstrate that investor sentiment, either in the local or US markets, plays a more important role on ADR returns for the Asia sample than for the Latin American sample.

Robustness Checking
To examine the stability of previous results, we exclude insignifi cant variables in the initial model and perform the same analysis for the four sample groups. The results are presented in Tables 10 and 11. As can be seen from the two tables, our fi ndings are qualitatively the same as the results above. The domestic stock return still plays a critical role both pre-and post-SEO for Latin American samples, USI dominates ADRNs over other factors for the Asia sample, and the signifi cance of other variables is similar to the initial model. Hence, our observations are robust.
Since SR might be an endogenous variable, an instrumental variable approach should be used (Pangan, Ullah 1988). The instrumental estimation technique used in this study is the panel system GMM estimator. This technique estimates a system by combining two sets of equations. An alternative would be the fi rst-differenced GMM procedure. However, the procedure suffers from weak instrument problems and can produce biased results -for instance, owing to heteroskedasticity problems and autoregressive parameters with values around unity (Blundell, Bond 1998). In all estimations, this study controls for time effects and fi rm effects by adopting random effects models.   Sample characteristics might be the reason for this difference. As shown in Table 6, the pre-SEO Latin American sample reveals some difference in skewness and kurtosis in SR and ADRNs compared to other groups. Even so, the dominance of SR remains in the system GMM specifi cations. Largely, the above robustness checks produce similar empirical results.

Conclusions
What drives the returns of ADR? This paper initially studies the issue of impact changes regarding accounting/investor sentiment and local/US variables on ADRNs by controlling the information transmission of domestic stock SEOs for ADRs in Latin American and Asian emerging markets from 1990 to 2007. We empirically contrast the longrun determinants of ADRNs both before and after domestic stock SEOs and contrast Latin America with Asia emerging economies. By following the model of Chan and Seow (1996) and Grossmann et al. (2007), we apply the two-way random effects panel data models and TSCSREG to investigate the changing infl uence of relative factors on ADRNs. Finally, panel system GMM models are also employed in the robustness check.
This study incrementally contributes to the behavioral fi nance and accounting literature by providing evidence of the informativness of investor sentiment and ADR-reconciled accounting information.
This paper demonstrates the following empirical results. First, the estimated coeffi cient on the SR is positively and statistically signifi cant in Latin American pre-and post-SEO models. The estimated coeffi cient on the SR in Table 8, for example, is 0.6. This means that a one unit increase in SR after domestic stock SEO increases ADRNs by 0.6% over a 1-year period. The relationship between SRs and Latin American ADRNs is consistent with 'law of one price'. However, USI positively dominates in the Asian sample, indicating one unit increase in USI increases Asian ADRN by 0.2 to 0.3% over a 1-year period before and after domestic stock SEO, respectively. Remarkably, LD even dominates SR in explaining Asian ADRNs. The results provide similar evidence with Ferguson (2000) in that investor sentiment shifted more rapidly in Asia than it did in Latin America, because investors were already more familiar with the regional structural ineffi ciencies and diffi culties in Latin America than in Asia. Consistent with Grossmann et al. (2007) and Kirch (2007), this study ascertains that the USI has a leading relationship among equity markets, and AEPS and DEPS convey little information regarding ADRNs. However, our study expands the research sample into the Latin American / Asian emerging markets and makes before and after domestic stock SEO comparisons. The TSCSREG results also permit us to conclude that SR is more important than USI for Latin American ADRNs, with the reverse holding true in Asia.
Second, regarding the difference between pre-and post-SEOs, TSCSREG and twoway random effects models specify that USI (LD) considerably affects Latin American ADRNs post (pre)-but not pre (post)-SEO. The sole difference between the pre-and post-SEOs based on the Asian market is that SRs dominate LD in explaining ADRNs post-SEO, but this state is not signifi cant pre-SEO. Moreover, SRs, LD, and USI exert signifi cant infl uences on ADRNs during the full period for the Asian sample. Both of Latin American and Asian markets' results cannot pinpoint the comparative significance of DEPS versus AEPS. However, Latin American and Asian fi ndings identify USI presents more salient infl uence than LD post-SEO. Although classic fi nance theory does not acknowledge behavioral factors as market sentiment, the empirical fi ndings imply that asset prices are affected by the place they are traded (Suh 2003), especially confi rmed in Asian ADRs of the research. We offer that ADRs from Latin America and Asia emerging markets have distinct and similar features that need to be noticed when planning investment strategies. Also, our fi ndings suggest that ADRs provide diversification benefi ts through two different sources: a market of origin diversifi cation, and a specifi c feature of domestic stock diversifi cation.
Finally, the estimated coeffi cients on the DEPS and AEPS variables for both Latin American and Asian ADRNs are statistically insignifi cant at conventional levels, indicating that changes in DEPS and AEPS have little impact on ADRNs from Latin American and Asian emerging markets, whether pre-or post-SEO. This fi nding may result from the distinctly important infl uence of annual accounting information versus daily investor sentiment data. Our fi nding is different from both Pope and Rees (1992) and Chan and Seow (1996), showing that DEPS dominates AEPS, or from Luchs (2004), indicating that DEPS has the same importance as AEPS. Their results may be mainly caused by no information transmission considerations and /or no comparison with other factors in the previous literature. Thus, we support the cost reduction view of accounting convergence and not that of a reduction in asymmetrical information.