Thesis
Has there been a more pronounced structural change in the betas of EU firms cross-listed in the U.S. than of single cross-listed EU firms as a result of the global financial
crisis?
Evidence from ADRs for three industries
Name: Justus Poldermans Student number: 10201874 Programme: Economie & Bedrijfskunde
Abstract
This paper explores whether the recent global financial crisis has impacted cross-listed firm returns more severely than single listed firm returns. In particular, whether the cross listing in the form of American Depositary Receipts (ADRs) causes a more pronounced structural change in the betas of the U.S. cross listed EU firm compared to single listed EU firms. The returns of 27 cross-listed and 130 single listed EU are regressed using an extension of the traditional CAPM model, which also controls for the home country of a firm. The event window spans from 15/09/2003-15/12/2013, where the event under investigation is the fall of Lehman Brothers on 15/09/2008. The CAPM Beta coefficients for (global) market risk premium estimated are found to be significant even after controlling for home country. However,
subsequent analysis regarding a structural change in betas for single listings and ADRs yield insignificant results. There is no statistical evidence that the betas changed structurally following the fall of Lehman Brothers. Therefore, there is no statistically sound basis to compare changes in betas of a particular listing type across industries.
Introduction
According to economic theory, firms competing in a frictionless world market are indifferent to the market in which their securities are traded. Yet, opposing this notion of market irrelevance, one finds that firms overwhelmingly choose to list initially their equity for trading on the respective home country stock exchange
(Miller, 1999). This result flows from cross-border barriers in listing requirements and information flow largely causing a bias towards home country stock markets.
Nevertheless, many firms after listing in their home country choose to subsequently list their equity in one or more foreign markets. This observation is supported by vast amounts of evidence regarding the positive attitude of firms towards cross listing. In particular, firms increasingly see foreign listings as an important component of their overall strategy (Pagano et al., 2002). The interest in choosing to list in a foreign market can be explained by gains to overcoming capital and informational barriers. In addition, firms list abroad due to preferential market characteristics such as liquidity,
tax treatment and disclosure (Sarkissian et al., 2009). When having decided to cross-list a firm may choose between several different types of cross cross-listing including: American Depositary receipts (ADRs), Global Depositary receipts (GDRs) ordinary shares, global shares and New York Registry Shares (NYSE, 2013). This paper will focus specifically on ADRs. ADRs are certificates of ownership of shares of a firm registered in any foreign country apart from the United States of America, but cross-listed in the U.S. These certificates are cross-listed and tradable on U.S. stock markets and denominated in US dollars (Jayaraman et al., 1993). ADRs and the underlying share can be regarded as virtually perfect substitutes after correcting for the transaction costs associated with the ADRs. This is because the holder of an ADR has the right to exchange the ADR for the underlying share (Miller, 1999). The selection of ADRs in particular is supported by evidence of ADRs being the primary form of cross listing in the U.S. (Pagano et al., 2002).
A great deal of research has been conducted in the field of cross listing. At large, the scope of this prior empirical research has been valuation gains as well as the various costs, benefits and motives of cross listing. Moreover, a great deal of
empirical research has also been conducted on betas in general and in particular the change in the Betas of firms in times of financial crisis. Yet no research has been conducted on whether this change in beta is more substantial for cross-listed firms compared to single listed firms in a global financial crisis.
In contrast, this particular research will depart from earlier papers by examining the following question: Has there been a more pronounced structural change in the betas of EU firms cross-listed in the U.S. than of single listed EU firms as a result of the global financial crisis? If these effects are present, then financial crises may pose an additional cost to cross listing. Overall, this paper attempts to answer two questions: (1) Is there a more pronounced structural change in the beta’s of cross listed firms compared to single listed firms? (2) Does this structural change (if present) differ across industries?
It is important to answer the aforementioned research question, as it would pose an additional cost to cross listing. This is because economies are susceptible to financial crises and firms may experience a contagion effect of a crisis in another country, other than their home country. Verifying the validity of the research question
would in that case mean that firms defer from cross listing in times of crisis or when the host country in which a firm will cross-list frequently faces severe financial crises.
The rest of this paper is organized as follows. The section Literature review will expand on the most relevant literature in the field of cross listings. This section will cover the benefits and costs and the impact of cross listing on risk and return in both the short run and the long run. The section Methodology provides a step-by-step account of how the data collection and manipulation proceeded. In addition, the Methodology section will state the hypotheses to be tested and the variables included in the various regressions. Moreover, the section Data and results provides a tabular overview of the descriptive statistics of the sample(s) considered. The section will also provide insight into the results of the statistical testing procedures. This will be carried out in order to determine the significance of estimated coefficients. Finally, the Conclusion will summarize and conclude the aforementioned sections.
Literature Review
To date, a great deal of research has already been conducted in the field of cross listing. The main results of the most influential papers refer to the benefits and costs associated with cross listing, the valuation gains of cross listing (Lau et al., 1994 and Sarkissian & Schill, 2009), and the motivation for a particular host country
(Roosenboom & van Dijk, 2009). In particular, the key reasons for European firms to cross-list in the U.S. are driven by the need to fund growth and foreign sales
expansion. Overall the key reasons are: Funding abroad may be easier or more easily available as a firm becomes more visible in foreign stock exchanges. Second, a cross listing may strengthen the competitive position of the company in its industry. This is achieved by enhancing its reputation with suppliers, employees, and customers through increased visibility as well as broadening the shareholder base (Baker et al., 2002 and Merton, 1987). Moreover, a firm may also benefit from greater liquidity and exploit relative mispricing of securities. The reason for this is: a security listed in one country may be overpriced compared to another country. Consequently, cross listing will provide a firm with more capital than if it were to issue equity in only one country (Baker et al., 2002). In addition, when a firm becomes cross-listed it faces an increase in both absolute returns and volume reactions. This is especially the case if
the firm’s home country is a developed country. Reasons for this are disclosure requirements, changes in market liquidity, ownership or trading venue (Pagano et al., 2002). Finally, one of the most significant advantages of cross listing is the resulting lower cost of capital as a company. The lower cost of capital is the consequence of a firm making its shares more easily available to foreign investors. These foreign investors would otherwise find it less beneficial to hold a firm’s shares due to the existence of international trade barriers.
The main costs seem to be heavily outweighed by the benefits of cross listing. The costs of cross listing mainly pertain to direct costs such as listing charges and fees for professional advice. In addition, a cross-listing firm faces costs associated with the compliance with local accounting standards. For example a cross listing firm must comply with Generally Accepted Accounting Standards (GAAP) when listing in the U.S. (Doidge et al., 2009). These costs mainly contain a fixed element, which implies that they bear more heavily on small companies. These costs are considered fixed as they are invariant of firm size. In other words, larger companies are more likely to cross list (Pagano et al. 2002). Doidge et al. (2004) even find that a considerable fraction of the costs associated with cross listing are borne by the depositary banks issuing depositary receipts (ADRs and GDRs).
Various other papers in the field of cross listing concentrate on whether there are short run or long run valuation gains from cross listing. For instance, Sarkissian et al. (2009) conduct empirical research using a global sample of 1,676 listings placed in 25 countries. In this paper Sarkissian et al. (2009) suggest there are no valuation gains from cross listing in the long run. More specifically, there is convincing statistical evidence that there are transitory valuation effects rather than permanent gains. Cross listings follow a pattern of a significant rise in pre-listing returns and a substantial post-listing decline in company returns during the five years before and after the listing. One event study by Miller (1999) focusing on ADRs in particular, discusses the impact of such a cross listing on the risks and returns of underlying stocks. The results suggest using this type of cross listing leads to a positive 1.15% average abnormal return of the underlying stock for 183 ADR-initiating announcement dates between 1985 and 1995. In addition, various papers find that cross listing results in a permanent increase in the volatilities of the returns on these stocks (Jayaraman et al.,
1993 and Miller 1999). In addition, Jayaraman et al. (1993) briefly discuss the impact of the 1987 October crash. According to the findings of Jayaraman et al. (1993) the October crash arguably induces an upward bias in the variances of the returns on ADRs reported. The reason for this is the drastic increase in the return volatilities around this time period. After several statistical tests it is evidenced that the October crash did not contribute to the experienced increase in return volatility. In other words the paper evaluated whether idiosyncratic risk of stocks (volatility) increased as a result of the October crash. The volatility did increase but was not a consequence of the crash. However, the paper does not discuss whether the stocks betas (sensitivity to systematic risk) changed and whether this change differed between cross-listed and single listed stocks, which is the scope of this investigation.
Overall, the papers reviewed above discuss the more general results following a cross listing (the event to be studied is generally the announcement date) with regard to risks and (long run) returns. However, none of these papers look at a structural change in betas of cross-listed and single firms and in the case of a change if it was more substantial for cross-listed firms. This presents an opportunity to investigate whether there is a more pronounced structural change in betas of cross-listed firms compared to single listed firms following the fall of Lehman Brothers on 15 September 2008. Moreover, it presents an opportunity to investigate whether, if present, this structural change is more pronounced in particular industries than in other industries.
Methodology
For this particular event study the characteristics of interest are market segmentation (serving a larger investor base) and visibility. This thesis will look at how these aspects have impacted the sensitivity to macro-economic shocks as measured by betas of NYSE, Nasdaq and NYSE MKT cross-listed EU firms
compared to the betas of single listed EU firms. As mentioned above one of the core motivations for firms to engage in cross listing is the increased visibility a firm experiences as its stock is now more explicitly traded on foreign stock exchanges. This means that a firm can now serve a larger investor base (Baker et al, 2002 and Merton, 1987). Consequently, a cross-listed firm faces greater demand and greater
changes in demand, especially in a global financial crisis, compared to single listed firms. On average, this implies there must be a more than proportionate decline in the return of cross-listed firms compared to single listed firms. In addition, the beta of the cross-listed firms should change more than the betas of single listed firms as a result of the crisis, which influences the returns of the firms based on the CAPM
assumptions. Hence the scope of this thesis is whether one observes statistically significant changes in the betas of the different types of firms as a result of the global financial crisis instigated by the fall of the Lehman Brothers investment bank in 2008.
To narrow the scope of the investigation, three industries will be considered in particular to highlight low beta (β<1), uniform beta (β =1) and high beta (β>1)
industries and to evaluate whether, if present, changes were more pronounced in one particular industry compared to another. The three particular industries to be
investigated are based upon prior research regarding the empirical evidence
surrounding the CAPM model. The choice for the different industries is as follows: a low beta industry (β<1) is the oil & gas industry. This is because when there are macro-economic shocks, the stock price of such firms is stable (i.e. no substantial changes in supply and demand). More specifically, the returns of these firms respond less than proportionately to changes in market risk premium (β<1). This observation is explained as follows: the demand for the products of these firms is insensitive to macro-economic shocks. This is because Oil products are deemed a necessity and therefore generally face a relatively inelastic demand (Bernardo et al., 2007 and Fama & French, 1997). For a uniform beta industry (β =1) the Telecommunications industry was selected. A more or less uniform beta implies that returns on the stock of this type of firm respond approximately proportionate to market wide shocks (Bernardo et al., 2007 and Fama & French, 1997). Finally, there are high beta industries (β>1). The industry to be investigated for this type of beta is the High tech industry. A high beta means that the stock prices of these firms experience an amplified or a more than proportionate effect than the market portfolio as a result of shocks in the economy as a whole (Bernardo et al., 2007 and Fama & French, 1997). Now that the different industries of interest have been defined the hypothesis of interest can be stated more clearly for each industry. In particular the format of the hypothesis will be as follows:
H0 : There is no more pronounced change in the post-crisis betas of cross-listed firms compared to single listed firms.
H1 : There is a more pronounced change in the post-crisis betas of cross-listed firms compared to single listed firms.
These hypotheses will be tested using several statistical hypotheses including t-tests to verify whether various estimated betas are statistically significant. More importantly, Chow’s F-test for structural change will be employed to investigate whether there has been a structural change in betas from the post-crisis period compared to pre-crisis betas. Finally, a comparison will be made between ADRs and single listings to assess whether ADRs in a particular industry experienced a more pronounced change in beta’s than single listed firms in the same industry.
To further clarify some of the terms and events mentioned in the hypothesis, the global financial crisis mentioned above refers to the fall of Lehman Brothers in the United States on 15 September 2008 instigating a wave of bank failures around the world subsequently causing the global financial crisis (Diamond and Rajan, 2009).
The sample of U.S. cross-listed EU firms included in this particular
investigation consists of firms listed on various specific U.S. stock exchanges rather than all U.S. stock exchanges. Namely, cross-listed EU firms listed in the U.S. on the NYSE, NASDAQ or the NYSE MKT (previously NYSE American Exchange). The reason for this choice is that the trade in other cross-listed firms’ equity in the US generally occurs in OTC markets and the data regarding such transactions are less widely available (Gong et al. 2007). In addition, the type of equities to be analyzed is ADRs: the most common form of trading foreign equities on the NYSE and NYSE MKT. The reason for the popularity of this type of security is the relative simplicity of issuing depositary receipts. In addition, ADRs do not require an additional issuance of equity in the host country, which is the case when a firm issues ordinary shares when cross listing (Jayaraman et al., 1993).
The sample data of cross-listed firms was identified and retrieved from
Compustat based on incorporation code, FINC. The event window to be considered is the period between 15-09-2003 and 15-12-2013, where the event under investigation is the fall of Lehman Brothers on the 15th of September 2008. As the investigation
type of interest is an event study a benchmark model is required. The benchmark model to be used is based on the following estimation window:
15/09/2003-15/07/2008. This benchmark model represents the pre-crisis period and is employed to determine whether there are structural changes in the post crisis betas, compared to the pre-crisis betas of both single listed and cross-listed firms. The estimation window cut off point is at 15/07/2008 rather than 15/09/2008 when Lehman Brothers actually collapsed. This is because from the first week of August 2008 onwards investors started to lose confidence in the stock markets and banks in particular prior to the actual fall of Lehman Brothers, due to write-downs on subprime mortgages by various banks (Ivashina and Sharfstein, 2010 and Mishkin, 2011). As the sample consists of monthly returns the cut off point becomes 15/07/2008 rather than 15/08/2008, to include the whole of August.
Furthermore, the data for companies that became cross-listed within the event window were omitted from the sample, as these would bias results. In addition, the 2003 file of Non U.S. issuers of equity (NYSE, 2003) on the NYSE, NASDAQ and NYSE MKT was used in the selection of cross-listed firms for the sample. This particular document was used because one faces a survivorship bias problem when only companies still cross-listed today (using the NYSE 2013 document) are included in the sample. Similarly, only companies in countries that were EU members prior to the start of the event window (15-09-2003) were included. This is because becoming a EU member potentially influences the stock prices, returns and betas of the companies located in these new member countries. This means that the single-listed firms to be considered must also be situated in the EU-member countries prior to the start of the event window and single-listed firms in other EU countries are ignored.
For this particular thesis there are two pools of data: 1. EU firms cross-listed in the U.S. and 2. Single listed EU firms. Both sets of data contain three subsets of data to subdivide cross-listed and single listed firms into their respective industries based on the value of their beta (either low beta: low beta (β<1), uniform beta (β =1) or high beta (β>1)).
Once the data has been collected and the potential difficulties associated with the data have been dealt with, one can run the OLS regressions using the datasets described above. The OLS-regressions will be run in several stages to highlight
different aspects of the sample. Every regression is based on the CAPM model for estimating the cost of capital. The original CAPM model as put forward by Sharpe (1964) and Lintner (1965) is as follows: Ri=rf+β(Rmkt-rf). The CAPM model provides an intuitive approach to the required rate of return on a security. The model is made up of two components the risk free rate: rf and the market risk premium demanded by investors β (Rmkt-rf). The market risk premium consists of the market excess return: (Rmkt-rf) and a measure of a security’s sensitivity to market wide shocks: the beta (β). As the sample compiled for this event study includes firms from across the globe, a new interpretation for the market return term Rmkt and the risk free rate term rf of the original CAPM model are required. More specifically, the Benchmark index Rmkt used in the CAPM model above refers to the S&P Global 100 index rather than the S&P 500 commonly used as a proxy for the market portfolio. Although this deviation of the traditional CAPM model may inhibit the validity of the CAPM, Dumas and Adler (1983) and Stulz (1981b) provide evidence for the plausibility of the CAPM model using a global market portfolio as a benchmark index. The S&P Global 100 index represents the 100 largest firms globally i.e. it represents the global market portfolio. This particular index is used to facilitate the comparison of returns in both samples considered. As both samples represent different countries/regions: the U.S. and the EU. Using the market portfolio of each respective country/region would inhibit cross-sample comparison. Furthermore, in the traditional CAPM model 1-month U.S. T-bills are used as a proxy for the risk free rate of return. Once again, this event study relies on a global sample of firms. Hence, a different risk free rate of return is required based on a global bond index for the risk free rate return. This global bond index together with the global market index will yield a global risk premium component for CAPM. Using a global risk premium consisting of global equity and bond indices is deemed appropriate based on empirical literature by Ferson and Harvey (1993) and Chandar et al. (2009). The bond index to be used is the J.P. Morgan Global GBI (Government Bond Index), a proxy for the global risk free rate of return. This index tracks government bond issuances from high-income countries spanning North America, Europe, and Asia (J.P. Morgan, 2014). The two units combined (Rmkt-rf) make up the global market risk premium. The CAPM model to be investigated is based on excess returns and looks as follows:
Regression 1 Ri- rf= β1(Rmkt-rf)
Note: there are new interpretations for the terms Rmkt and rf
Three regressions will be run based on the above model. Firstly, regressions will be run for the sample as a whole (including cross listing and single listing) for the full sample, the pre-crisis and the post-crisis periods to determine if the estimated coefficients are significant using t-tests. Subsequently, a regression will be run for cross listings and single listings individually for the same three periods, similarly to determine if the coefficients are statistically significant. Finally, several regressions will be run for every industry separately, once again for the three periods to assess significance. These regressions are carried out to assess whether there is a more pronounced effect (if present) in certain sectors compared to other sectors. The different industries are low beta (oil), uniform beta (telecommunications) and high beta industries (high tech firms).
Note that for all the non-US single listed firms included in the sample home country returns are converted to US dollars using the corresponding monthly exchange rates from DataStream.
Following these regressions the original CAPM model will be extended to include several additional (dummy and control) variables. Namely, Home country: as every country responds differently to global market wide shocks (Ferson and Harvey, 1993). In particular, different countries respond differently to the current global financial crisis i.e. for France the dummy is FRi and equals 1 if the home country is France etc. (the EU countries are: the UK, Spain, Portugal, the Netherlands, Italy, Ireland, Germany, France, and Finland). These additional variables yield the following extension of the CAPM model (using excess returns):
Regression 2:
Ri- rf= β1(Rmkt-rf)+ β2Fini+ β3Geri+β4ITi+ β5Irei+ β6NLi+β7Pori +β8 Spi + β9 UKi
Note: observe that for the 9 countries considered in the sample, only 8 Dummies for home country are included. The dummy ‘Fra” is omitted to deal with the dummy variable trap which would result in perfect multicollinearity (Stock and Watson, p.243, 2012). Moreover, the variable ‘Home’ refers to any statistically significant dummy for home country.
Using this extended CAPM model, several regressions will be run. In
the full sample, secondly for the pre-crisis period and finally, for the post-crisis period. The estimated coefficients from these regressions will be used to evaluate whether a firm’s home country was significant if impacted its return, and if it altered the estimated β1.
Finally, several statistical hypotheses will be tested to determine whether the coefficients obtained where statistically significant using a t-test. In addition, several Chow’s F-tests will be conducted to determine if there has been a structural change in post-crisis betas compared to post-crisis betas. Finally, a several hypothesis tests will be conducted in which two populations are compared (single listings and ADRs) to evaluate whether the change in betas following the crisis was more pronounced for ADRs as compared to single listings.
Data & Results
Based on the methodology described in the previous section the event study surrounding the fall of Lehman Brothers in 2008 was carried out. The sample(s) obtained consists of 28 ADRs and 130 single listed firms. In table 1 on the next page the home countries and the particular industries of the firms included in the sample are summarized.
Home country
Petrol (β<1) Telecom (β=1) High Tech (β>1) Petrol (β<1) Telecom (β=1) High Tech (β>1) Finland 1 1 3 France 1 1 1 4 3 20 Germany 1 2 1 4 21 Italy 1 2 2 3 10 Ireland 2 6 The 2 3 1 1 3 Type of Listing
Netherlands Portugal 1 2 Spain 2 1 2 The UK 1 2 5 20 6 16 Total per Industry 5 9 14 35 22 73 Total 28 130
Table 1: Number of firms per listing type organized per industry and per home country.
Table 2 exhibits the returns and standard deviation of returns for the subdivision of the sample (ADRs and single listings). As can be observed the mean returns of ADRs are far lower than those of single listings for the full sample period as well as during the pre- and post-crisis periods. This observation is consistent with the observations of Kadlec and McConnell (1994), Foerster and Karolyi (1999) Baker et al. (2002) and Sarkissian and Schiller (2009). The explanation is an increase in the number of shareholders after cross listing and is associated with lower returns required by investors compared to the required return prior to the cross listing. In addition, it is clear that the standard deviations are substantially smaller for the ADRs than those for single listings across the three periods considered.
Period Mean Standard Deviation
15/08/2003-15/12/2013 (Full sample) ADR 0,67 9,12 Single listing 1,68 16,45 15/08/2003-15/06/2008 (Pre-crisis) ADR 0,84 7,26 Single listing 1,87 18,28 15/07/2008- 15/12/2013 (Post-crisis) ADR 0,52 10,53 Single listing 1,51 26,66
Table 2: The mean returns and standard deviation of returns per listing type organized per period (in percentages).
In Table 3 below one can observe the mean returns and standard deviations for the particular industries under investigation for both single and cross listings. The returns and standard deviations estimated support the results of literature regarding risk and returns for different beta industries. In effect, low beta industries have low returns and volatilities; high beta industries have higher returns and volatilities etc. (refer to Bernardo et al., 2007 and Fama & French, 1997).
Period Industry Mean Standard Deviation
15/08/2003-15/12/2013 (Full sample) ADR Oil 0,58 6,92 Telecom 0,35 9,37 High Tech 1,04 9,90 Single listing Oil 1,75 18,52 Telecom 0,95 11,96 High Tech 1,90 26,16 15/08/2003-15/06/2008 (Pre-crisis) ADR Oil 1,12 5,87 Telecom 0,56 6,48 High Tech 0,96 8,58 Single listing Oil 3,13 16,28 Telecom 1,14 12,74 High Tech 1,54 17,56 15/07/2008- 15/12/2013 (Post-crisis) ADR Oil 0,15 11,38 Telecom 0,087 8,53 High Tech 1,11 10,97 Single listing Oil 0,43 20,16 Telecom 0,77 11,44 High Tech 2,23 32,03
Table 3: The mean returns and standard deviation of returns per period and per industry for both types of listings (in percentages).
In addition, in line with the findings of Hamilton and Lin (1996) and Schwert (1990) it is observed that the volatilities (standard deviation) increase post-crisis both for
the standard deviation of petrol from 5,87% to 11,38%, for telecom an increase from 6,48% to 8,53% and for high tech there was an increase from 8,58% to 10,97%. Similar results were obtained for single listings: an increase in the standard deviation of petrol from 16,28% to 20,16% and a substantial increase in the standard deviation of high tech from 17,56 % to a 32,03%. While the standard deviation of telecom actually declined from 12,74% to 11,44%.
Subsequently, several regressions were run based on the data under
investigation. In table 4 below one can observe the beta estimates for different subsets of the sample across different time periods (pre- and post-crisis). It can be observed that all betas are significant based on the following hypotheses: H0: β1= 0 versus H1: β1 ≠ 0 at a 5% significance level. Hence there is a linear relationship between excess security returns (Ri-rf) and excess market returns (Rmkt-rf).
Period Sample Coefficient
b (standard error)
t-values t(b) 15/08/2003-15/12/2013
(Full sample period)
Full Sample 1.110 *** (0.066) 16.75 ADRs 0.946 *** (0.110) 8.63 Single listings 1.273 *** (0.070) 18.26 15/08/2003-15/06/2008 (Pre-crisis) Full Sample 0.935 *** (0.101) 9.24 ADRs 0.764 *** (0.156) 4.88 Single listings 1.106 *** (0.115) 9.57 15/07/2008- 15/12/2013 (Post-crisis) Full sample 1.184 *** (0.089) 13.34 ADRs 1.024 *** (0.153 ) 6.70 Single listings 1.343 *** (0.090) 14.99
Table 4: The estimated Betas and t-values for the whole sample, ADRs and single listings in various periods. Tcr=1.96 at α=0.05. Tcr=2.58 at α=0.01. *** indicates statistically significant at α=0.01, **, is statistically significant at α=0.05, and * at α=0.10.
In addition, if one compares pre-crisis and post crisis betas for the different subsets of data at first sight, it seems as if in every case the beta has increased as a result of the financial crisis. In other words, one can observe that on average all firms experienced
an increased responsiveness to market wide shocks as measured by betas following the crisis.
To further investigate whether these changes in the betas are statistically significant one must turn to Chow’s F-test for structural change. The aforementioned significance of the estimated betas allows this further investigation. In particular Chow’s F-test for structural change can be employed to determine whether there was a structural break in betas following the global financial crisis. In short, Chow’s F-test compares the Residual Sum of Squares of the linear regression estimated for the pre-crisis and post-pre-crisis period, to assess the presence of a structural change. The F-test for structural change has the following format:
The null and alternative hypothesis for an F-test have the following format: H0: β1,t1 = β1,t2 and H1 : β1,t1 ≠ β1,t2. The H0 implies that there is no structural change as the pre and post-crisis are approximately the same. The F-statistics observed for the Full sample, the ADRs and Single listings can be observed in Table 5 below.
Table 5: The F-statistics for the full sample, ADRs and single listings. This F-test focuses on k=1 parameters and has n1+n2-2k=120 degrees of freedom The Fcr1,120=3.92 at α=0.05.
All the F-statistics can be rejected at the 5% significance level. As a result it can be concluded that there is insufficient evidence to infer that there was a structural change in the beta for the various datasets. Consequently, a further analysis examining whether the structural change was more pronounced for ADRs compared to single listed firms cannot be carried out.
Although the F-statistics were insignificant for the sample as a whole as well as the listing types individually, one can narrow down the scope of the investigation to determine if there was a structural change in the betas in certain industries. Hence,
Sample F-statistic
Full sample 3.041
ADRs 1.172
the next step in the regressions considers the various industries separately for both ADRs and single listings. The relevance of these regressions is that in a later stage these estimates will be used to determine if there was a structural change in betas, and in the presence of a structural change, whether this change was more pronounced for the various industries separately. This is to assess whether certain industries were affected more severely by the financial crisis than other industries.
Period Type of listing Industry Coefficient
b (standard error)
t-values t(b)
15/08/2003-15/12/2013 (Full sample period)
ADRs Oil 0.748 *** (0.100) 7.45 Telecom 0.811*** (0.097) 8.34 High Tech 0.826 *** (.100) 8.26 Single listings Oil 1.220 *** (0.109) 11.23 Telecom 1.072 *** (0.0631) 16.98 High Tech 1.356 *** (0.089) 15.20 15/08/2003-15/06/2008 (Pre-crisis) ADRs Oil 0.673 *** (0.162) 4.14 Telecom 0.694 *** (0.115) 6.05 High Tech 0.758 *** (0.159) 4.78 Single listings Oil 0.906 *** (0.193) 4.71 Telecom 1.134 *** (0.113) 10.05 High Tech 1.175 *** (.148) 7.94 15/07/2008- 15/12/2013 (Post-crisis ADRs Oil 0.769 *** (0.132) 5.84 Telecom 0.860 *** (0.144) 5.96 High Tech 0.860 *** (0.133) 6.46 Single listings Oil 1.325 *** (0.125) 10.58 Telecom 1.045 *** (0.078) 13.38 High Tech 1.443 *** (0.114) 12.63
Table 6: The estimated Betas and t-values for the various industries as organized per listing type in various periods. Tcr=1.96 at α=0.05. Tcr=2.58 at α=0.01. *** Indicates statistically significant at α=0.01, **, is statistically significant at α=0.05, and * at α=0.10.
The beta coefficients estimated in Table 6 are significant and can therefore be used to gain further insights into whether there was a structural change in betas for the various listing types across industries. In addition, one can observe that the beta coefficients
estimated for the different industries across periods is largely in line with previous literature (refer to Fama & French, 1997). For instance, one can observe across the sample periods that the betas for single listed High tech firms are substantially more than one in most cases, hence as earlier empirical literature suggests this means a high beta industry. However the betas of cross listed (ADR) high tech firms across periods depart from earlier empirical results as in most cases the estimated betas are below one and therefore imply a low beta industry. The same results are found for the Oil and Telecom industries.
The next step is to include the Dummy variables for home country (8 in total) into the regression equation to assess whether home country specific factors affected a firm’s return. The F-test statistic column in Table 71 displays the estimated F-statistic for the joint hypothesis test on multiple coefficients of Regression 2 (Ri- rf= β1(Rmkt -rf)+ β2Fini+ β3Geri+β4ITi+ β5Irei+ β6NLi+β7Pori +β8 Spi + β9 UKi). The hypothesis is as follows: H0:β1= 0, β2= 0, …, β9= 0, versus H1: βj≠ 0, at least one j, j=1,…9. The critical F-values differ across the sample periods based on the number of observations in each subset of data. The F-test is computed, as a t-test for every individual
coefficient would yield misleading results. This is because the rejection rate under the null hypothesis does not equal the desired significance level of 5 percent if one were to use individual t-tests (Stock & Watson, 2011). If one inspects the results obtained in Table 7, it is evidenced that for every subset of data the H0 for the F-test can be rejected at a 5 percent significance level. Consequently, the addition of control variables for home country produces statistically significant results and must be included in the extended CAPM model. Hence, one can infer that the home country of a firm does indeed impact its returns, where returns are modelled according to the CAPM assumptions.
Table 7: Summary table containing the estimated beta coefficients for regression 2 and the estimated joint hypothesis F-test statistics. In brackets are the estimated t-statistics for the beta coefficient. Tcr=1.96 at α=0.05, Tcr=2.58 at α=0.01. The Fcr10,114=1.91 at α=0.05, F
cr
10,114=2.48 at α=0.01 *** Indicates statistically significant
at α=0.01, **, is statistically significant at α=0.05, and * at α=0.10.
Following these statistically significant insights one can continue the investigation for structural changes by applying the F-test for structural change. This F-test is
computed for the statistically significant betas (in particular, the estimates for β1 in each subset of data) estimated in table 7.
Listing Type Industry F-statistic
ADRs Oil 0.141
Telecom 0.105
High Tech 0.063
Single listings Oil 0.260
Telecom 0.100
High Tech 0.024
Table 8: The F-statistics for the full sample, ADRs and single listings. This F-test focuses on k=1 parameters and has n1+n2-2k=114 degrees of freedom The Fcr10,114=1.91 at α 5 percent significance level. . *** Indicates
statistically significant at α=0.01, **, is statistically significant at α=0.05, and * at α=0.10 in terms of t-tests for individual coefficients
Based on the values retrieved from table 8 one can observe in all cases that the null hypothesis cannot be rejected at a significance level of 5 percent. Hence, one cannot infer there was a structural change in the betas across industries. Therefore,
ADR Industry Period (Rm-rf) Market risk premium (t-statistic) F-test statistic Single Listing Industry Period (Rm-rf) Market risk premium (t-statistic) F-test statistic Oil Full sample 0.751 *** (13.26) 43.96*** Oil Full sample 1.126 *** (19.14) 77.22*** Pre-crisis 0.653 *** (7.53) 14.30*** Pre-crisis 0.974 *** (9.29) 17.54*** Post-crisis 0.781 *** (10.25) 26.31*** Post-crisis 1.294 *** (19.90) 66.28*** Telecom Full sample 0.860 *** (13.84) 32.24*** Telecom Full sample 1.040 *** (22.63) 64.43*** Pre-crisis 0.674 *** (7.70) 10.68*** Pre-crisis 1.072 *** (12.24) 18.91*** Post-crisis 0.933 *** (10.76) 19.52*** Post-crisis 1.0199*** (19.25) 47.45 *** High Tech Full sample 1.256 *** (9.11) 17.16*** High Tech Full sample 1.343 *** (25.54) 106.89 Pre-crisis 1.026 *** (4.83) 4.73*** Pre-crisis 1.189 *** (14.05) 33.10*** Post-crisis 1.370 *** (7.37) 11.43*** Post-crisis 1.416 *** (20.18) 68.85***
one cannot take the next step in the analysis of evaluating whether the structural change was more pronounced for ADRs as compared to single listings. In addition, one cannot conduct such analysis across the various industries under investigation.
Conclusions
This thesis aimed at answering two questions 1. Is there a more pronounced structural change (if present) in the betas of cross-listed firms compared to single listed firms? 2. Does this structural change (if present) differ across industries? Although the estimated betas across listing type and industry were significant and the home country was controlled for, subsequent analysis with regard to structural change has proven to be fruitless. In particular, the empirical results obtained imply in every case there were no structural changes in the betas at all. In turn, this means one cannot infer that the global financial crisis may pose an additional cost to cross listing.
The reason for the statistical insignificance of results may be imbedded in exogenous factors omitted from the model. For instance, the fall of Lehman brothers was
preceded by a subprime mortgage crisis as well as followed by subsequent crises across the globe. As a result it may prove difficult to isolate the effects of the Lehman Brothers fall on single- and cross listings. In addition, a great deal of empirical
literature has revealed that as firms mature, their betas tend to converge to 1 (Bernardo et al., 2007). This implies, in the long run, there will be no or perhaps statistically insignificant (structural) changes in the betas of firms even in the event of a financial crisis. This observation may be one of the key reasons why no structural change was discovered; in general cross-listed firms are mature stable firms (Pagano et al., 2002). However, other factors omitted from the model may also contribute to there being no manifestation of a structural change in betas. The results obtained in this paper call for further future empirical analysis with regard to other factors affecting the betas of cross-listed firms.
Reference list
Baker, H.K, Nofsinger, J.R. and Weaver, D.G., (2002). International Cross-Listing and Visibility.The Journal of Financial and Quantitative Analysis, Vol. 37 (3). 95-521.
Bailey, W.G., Karolyi, A. and Salva, C. (2006). The economic consequences of increased disclosure: Evidence from international cross-listings. Journal of
Financial Economics, Vol. 81 (1), 175–213.
Bernardo, A.E., Chowdhry, B. and Goyal, A., (2007). Growth Options, Beta, and the Cost of Capital. Financial Management, Vol. 36 (1), 5-17.
Chan, K.C and Chen, N.F. (1991). Structural and Return Characteristics of Small and Large Firms. Journal Of Finance, Vol. 46(4), 1467-1484.
Chan, L.K.C., Karceski, J. and Lakonishok, J. (1998). The Risk and Return from Factors. The Journal of Financial and Quantitative Analysis, Vol. 33(2), 159-188.
Chan, J.S.P., Hong, D. and Subrahmanyam, M.J. (2008) A tale of two prices:
Liquidity and asset prices in multiple markets. Journal of Banking & Finance,
Vol. 32(6), 947–960.
Chandar, N. Patro, D. K. and Yezegel, A. (2009). Crises, Contagion and Cross-listings. Journal of Banking and Finance, Vol. 33(9), 1709-1729.
Diamond, D. W., and Rajan, R. (2009). The credit crisis: Conjectures about causes and remedies. American Economic Review, Vol. 99(2), 606-610.
Doidge, C., Karolyi, A. and Stulz, R.M. (2004). Why are foreign firms listed in the U.S. worth more? Journal of Financial Economics, Vol. 71, 205–238. Doidge, C.G., Karolyi, A. and Stulz, R.M. (2009). Has New York become less
competitive than London in global markets? Evaluating foreign listing choices over time Journal of Financial Economics, Vol. 91 (3) 253–277.
Fama, E.F. and French, K.R. (1997). Industry costs of equity. Journal of Financial
Economics, Vol. 43 (2),153–193.
Ferson, W.E. and Harvey, C.R. (1993).The Risk and Predictability of International Equity Returns. The Review of Financial Studies, Vol. 6 (3), 527-566.
Foerster,S . R., and G. A. Karolyi. (1993). International Listings of Stocks: The Case of Canada and the U.S. Journal of International Business Studies, Vol. 24, 763-784.
Hamilton, J.D. and Lin, G. (1996) Stock market volatility and the business cycle.
Journal Of Applied Econometrics, Vol.11(5), 573-593.
Ivashina, V. and Scharfstein, D. (2010). Bank lending during the financial crisis of 2008. Journal of Financial Economics, Vol. 97, 319–338.
Kadlec, G. B., and McConnell. J. J. (1994). The Effect of Market Segmentation and Liquidity on Asset Prices: Evidence from Exchange Listings. Journal of
Finance, Vol. 49, 611-636.
Lau, S.T., Diltz, J.D. and Apilado, V. P. (1994). Valuation effects of international stock exchange listings. Journal of Banking and Finance, Vol. 18, 743-755. Lintner, J. (1965). The valuation of risk assets and the selection of risky investments
in stock portfolios and capital budgets. Review of Economics and Statistics,
Vol. 47, 13-37.
Merton, R. C. (1987). Presidential address: A simple model of capital market equilibrium with incomplete information, Journal of Finance. Vol. 42, 483-510.
Miller, D.P. (1999). The market reaction to international cross-listings: evidence from Depositary Receipts. Journal of Financial Economics, Vol. 51 (1) 103–123. Mishkin, F. S. (2011). Over the Cliff: From the Subprime to the Global Financial
Crisis. Journal of Economic Perspectives, Vol. 25 (1), 49-70.
Narayanan, J., Kuldeep, S. and Kishore, T. (1993). The impact of international cross listings on risk and return: The evidence from American depository receipts.
Journal of Banking & Finance, Vol. 17 (1), 91–103.
Pagano, M. , Röell, A.A. and Zechner, J. (2002). The Geography of Equity Listing: Why Do Companies List Abroad? The Journal of Finance, Vol. 57 (6), 2651-2694.
Roosenboom, P. and van Dijk, M.A. (2009). The market reaction to cross-listings: Does the destination market matter? Journal of Banking & Finance, Vol. 33, 1898-1908.
listing? Review Of Financial Studies, Vol. 22(1), 371-412.
Scherer, F.M. (1965). Firm Size, Market Structure, Opportunity, and the Output of Patented Inventions. The American Economic Review, Vol. 55(5), 1097-1125. Schwert, G.W. (1990). Stock volatility and the crash of ’87. Review Of Financial
Studies, Vol.3(1), 77-106.
Shalit S. S. and Sankar, U. (1977). The Measurement of Firm Size. The Review of
Economics and Statistics, Vol. 59 (3), 290-298.
Sharpe, W. F., (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of Finance, Vol. 19, 425-42.
Stock, J.H. and Watson, M.W. (2012). Introduction to Econometrics. Pearson, Essex, p. 243.
Stulz, R. M. (1981). On the Effects of Barriers to International Investment. Journal of Finance, Vol. 36, 923-934.
Listing
Type Industry Period
Regression 1 Regression 2 (Rm-rf) Market risk premium (Rm-rf) Market risk premium
Fin Ger It Ire NL Por Sp UK F test
statistic ADR Oil Full sample 0.748 *** (7.45) 0.751 *** (13.26) 0.012 (0.02) 0.030 (0.040) -0.253 (0.651) 43.96*** Pre-crisis 0.673 *** (4.14) 0.653 *** (7.53) 0.141 (0.15) -0.310 (-0.32) -0.476 (-0.49) 14.30*** Post-crisis 0.769 *** (5.84) 0.781 *** (10.25) -0.101 (-0.08) 0.328 (0.27) 0.148 (0.12) 26.31*** Telecom Full sample 0.811 *** (8.34) 0.860 *** (13.84) 0.427 (0.4) -0.834 (-0.79) -0.561 (-0.53) 0.511 (0.47) 0.457 (0.43) 0.057 (0.05) 32.24*** Pre-crisis 0.694 *** (6.05) 0.674 *** (7.70) -0.390 (-0.33) -1.57 (-1.31) -0.471 (-0.39) 0.478 (0.39) 0.977 (0.82) -0.425 (-0.36) 10.68*** Post-crisis 0.860 *** (5.96) 0.933 *** (10.76) 1.145 (0.68) -0.188 (-0.11) -0.640 (-0.38) 0.574 (0.28) -0.007 (-0.00) 0.481 (0.28) 19.52*** HighTech Full sample 0.825 *** (8.26) 1.256 *** (9.11) 0.472 (0.20) 0.839 (0.39) 0.395 (0.18) 3.290 (1.53) 1.480 (0.69) 17.16*** Pre-crisis 0.758 *** (4.78) 1.026 *** (4.83) 0.393 (0.40) 0.058 (0.02) -0.213 (-0.08) 1.094 (0.41) 0.348 (0.13) 4.73*** Post-crisis 0.860 *** (6.46) 1.370 *** (7.37) 0.538 (0.37) 1.526 (0.46) 0.930 (0.28) 5.221* (1.57) 11.43*** Single Listing Oil Full sample 1.220 *** (11.23) 1.126 *** (19.14) -0.078 (-0.09) 2.135 ** (2.32) -0.543 (-0.59) -0.278 (-0.30) 0.864 (0.94) 77.22*** Pre-crisis 0.906*** (4.71) 0.974 *** (9.29) -0.223 (-0.17) 3.424 ** (2.59) -1.501 (-1.15) -0.790 (-0.61) 1.011 (0.77) 17.54*** Post-crisis 1.324 *** (10.58) 1.294 *** (19.90) 0.049 (0.04) 1.360 (1.07) 0.300 (0.24) 0.172 (0.14) 0.733 (0.58) 66.28*** Telecom Full sample 1.072 *** (16.98) 1.040 *** (22.63) -0.102 (-0.11) -0.095 (-0.10) -0.801 (-0.88) -1.153 (-1.27) -0.691 (-0.76) 0.026 (0.03) -0.077 (-0.09) 64.43*** Pre-crisis 1.134 *** (10.05) 1.072 *** (12.24) -0.265 (-0.19) -1.199 (-0.87) -0.803 (-0.58) -0.417 (-0.30) -0.814 (-0.59) -0.728 (-0.53) -1.361 (-0.99) 18.91*** Post-crisis 1.045 *** (13.38) 1.0199 *** (19.25) 0.041 (0.03) 0.877 (0.73) -0.799 (-0.67) -1.800* (-1.51) -0.583 (-0.49) 0.689 (0.58) 1.051 (0.88) 47.45 *** HighTech Full sample 1.356 *** (15.20) 1.343 *** (25.54) -0.674 (-0.74) 0.860 (0.95) -0.657 (-0.72) -0.0560 (-0.06) 0.439 (0.48) 106.89 Pre-crisis 1.175 *** (7.94) 1.189 *** (14.05) -0.175 (-0.15) -0.176 (-0.15) -0.554 (-0.48) -0.729 (-0.63) -0.359 (0.31) 33.10*** Post-crisis 1.443 *** (12.63) 1.416 *** (20.18) -1.11 (-0.81) 1.771 (1.29) -0.748 (-0.55) 0.535 (0.39) 0.508 (0.37) 68.85***
Table 9: The regression coefficients for the original CAPM model and the extended CAPM model across various listing types, industries and periods. In brackets is each coefficients estimated t-test statistic. The far right column displays the joint hypothesis F-test for multiple regression coefficients. Fcr =1.91 (Full sample), Fcr =2.00 (Pre-crisis), Fcr =2.03 (Post-crisis). *** Indicates statistically significant at α=0.01, **, is statistically
