Amsterdam Business School
MSc. Finance
Master thesis
The wealth effect of the 2016 U.S. election in the U.S., Canada and Latin
American stock markets
Student: Diana Stefanie Tornell Flores Supervisor: Dr. Tanju Yorulmazer
Statement of Originality
This document is written by Student Diana Stefanie Tornell Flores who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgement
I would like to express my profound gratitude to my parents and to my brother for providing me with unfailing support and continuous encouragement to achieve every project in my life. This accomplishment would not have been possible without them. Thanks!
Abstract
This paper analyzes the effect of the 2016 U.S election in seven industries stock prices from Canada and eight Latin American countries. In addition, the study suggests as a robust and more reliable econometric approach, the SUR method for abnormal security-price performance. Following Trump’s victory, the U.S. market anticipated a decrease in the Energy and Banking regulation as well as a higher investment in the Construction industry derived from his campaign proposals, presuming these industries as the winners. In contrast, the Energy and Construction industry from the other countries were presumed as the losers. The Banking industry in the other countries shows a mixed effect.
INDEX
I. Introduction ... 1
A. Rationale ... 1
B. Background ... 1
II. Literature review ... 6
III. Hypothesis and Methodology ... 10
A.Hypothesis ... 10
B. Methodology ... 11
1. Parametric test ... 12
2. Non-Parametric tests ... 13
a) Sum Rank Test ... 13
b) Sign Rank Test ... 14
3. Robust test SUR method ... 14
4. Regresssions ... 15
IV. Data and descriptive statistics ... 17
A. Databases ... 17
B. Descriptive Statistics... 19
V. Results ... 22
VI. Robust Check ... 26
VII. Conclusions ... 34
Limitations ... 35
VIII. Appendix ... 37
A. Definition of variables ... 37
B. Results ... 38
C. Distribution of Abnormal Returns ... 50
D. Investment Banks and Systemic Important Banks ... 52
1
I. Introduction
A. Rationale
Many event studies have analyzed the impact of the monetary policy, earnings, dividends or macroeconomic announcements but there are only few ones related to the effects of the U.S. elections in the stock market. Given the big controversy of the last U.S. election and the evident effect in the U.S. stock market, the aim of this study is to analyze for seven industries the impact in the wealth of their shareholders in U.S. and the spillovers in Canada and eight Latin American countries as well as to identify which variables suggest a relationship to the change of the stock prices.
To empirically investigate the mentioned issues, this research puts in practice the event study methodologies widely used in several studies as well as a non-commonly econometric method, using daily security-return data. An important focus of this analysis is the selection of the empirical method so as to determine consistent and efficient abnormal returns based on the assumptions of uncorrelation and nonnormality in daily returns. Depending on the assumption and the method chosen, the results may convey misleading conclusions.
B. Background
The market’s reaction to the U.S. Election was discerned on the policies proposed by the Democratic nominee Hillary Clinton and the Republican nominee Donald Trump during the presidential race. Among their proposals, the industries targeted were the Automotive, Construction, Health Care, Energy, Pharmaceutical and Banking.
Both candidate’s ideas about global trade were opposite, Trump clearly stated his aim of increasing the manufacturing and goods-based jobs in America by changing the North American Free Trade Agreement (NAFTA) and the Trans-Pacific Partnership (TPP). He believed that there was an abuse from these trades by taking the manufacture out of the country. In the particular case of the Automotive industry, many carmakers established factories abroad to decrease their manufacturing costs which will be affected if the trades were changed.
2 Despite both candidates were interested in improving the infrastructure of the country, Trump denote a higher interest when he mentioned a drastically higher amount of investment than the one proposed by Clinton. In addition, he emphasized that the country needed to rebuild roads, bridges, airports, schools and hospitals. This was translated as a benefit for the suppliers to the Construction industry.
Another remarkable difference between the candidate’s proposals was the abolition of the Obamacare pledged by Trump. This health care act center in providing more Americans with access to good quality and affordable health insurance. Thus, the Health care industry could be affected.
In this context, the Pharmaceutical industry would suffer changes if the drug pricing be relaxed. Unlike Trump whose proposal involved to import cheaper drugs, Hillary warned to curb the unjustified outrageous drugs prices.
With respect to the job creation, Trump seek to maintain the US energy independent by proposing new areas of the country to oil and gas development and by lessening the regulation on energy production. Hence, the Petroleum and Oil & Natural Gas industries would receive a boost.
The Banking industry was also likely to be affected. Clinton’s plan strengthened the financial regulation by tackling risks from shadow banks and loopholes in the Volcker Rule, increasing the regulation for the too-big institutions, imposing fees based on the risk and size of the bank and prosecuting individuals or corporations that break the law. On the other hand, Trump’s plan was to dismantle the Dodd-Frank Wall Street Reform which purpose is to promote the financial stability of the United States by improving accountability and transparency in the financial system, as a response of the 2007-2008 crisis. After the crisis, this Act was disapproved by some financial institutions and it faced many constitutional challenges such that many states jointly lawsuit the Government. Some key components of this Act were the introduction of the Volcker Rule which effectively separates the investment and commercial functions of a bank in addition to limit the banking entities ownership in hedge fund or private equity, the creation of the Financial Stability Oversight Council which prevent systemic risk and the regulation of credit default swaps by the Securities and Exchange Commission (SEC) and the Commodity Futures Trading Commission (CFTC).
3 Hence, the unexpected victory of Trump and his previous proposals called the attention of the company’s shareholders such that some were winners and some others losers in the financial markets. The industries will suffer the aftermath in different magnitudes and given the reliance of many countries on the U.S., the effect will go beyond the local financial market. Then, the scope of this analysis will be limited to the U.S. and nine countries from America: Argentina, Brazil, Canada, Chile, Colombia, Jamaica, Peru, Mexico and Venezuela.
The Figure 1 shows the observed cumulative returns performance for each US industry, around the Election day. This figure reflects a stable performance the first 7 days, from -22 to -15 days, where the cumulative returns were between -5% and 5%. Afterwards, the volatility of the Pharmaceutical and Construction industries increased, reflecting a negative return for the former and a positive return for the latter. Three days previous to the Election, the Pharmaceutical industry denoted an improvement from -18% to -14% in the election day and to -3% afterwards. The Automotive, Oil & Oil Products and Petroleum & Natural Gas industries had a negative performance whereas the Construction, Banking and Health Care had a positive performance. In addition, the chart illustrates the market return as a benchmark. It is remarkable the flat performance of the market unlike that of each industry.
The Figure 2 shows the observed cumulative returns performance for each non-US industry, around the Election day. Unlike the US industries, the performance of the non-US industries
Figure 1. US Industry’s returns around the US election. This figure shows cumulative returns on the portfolios of each US industry from June 13 to October 7, 2016 (days −22 to +5 around the US elections). The straight line represents the US market return as a benchmark.
4 was relatively stable fluctuating between -5% and +5% until 5 days prior to the Election day. The Pharmaceutical and Automotive industries had the worst performance, during the two weeks previous the announcement the former changed its cumulative return from 2% to -11% and the latter from -1% to -12%. In the case of the Petroleum & Natural Gas industry, the effect was observed one week previous to the announcement where the change was from 3% to -10%. Compared to these non-US industries, the Banking, Oil & Oil products and Health care were less volatile.
The Figure 3 illustrates the observed cumulative returns for the Banking industry of each country, around the election day. Unlike the other industries, the performance of the Banking industry has little fluctuations previous to the U.S, Election with the exception of Argentina which has a negative trend reaching up to -15%. The evident result after the election was a significantly higher return for Venezuela as well as for the US. In the particular case of Venezuela, the Bolsa de Valores de Caracas (BVCI) presented big changes around the Election day. Compared to the other countries, Mexico and Argentina denote negative returns.
Figure 2. Non-US Industry’s returns around the US election. This figure shows cumulative returns on the portfolios of each non-US industry from June 13 to October 7, 2016 (days −22 to +5 around the US elections).
5 The proposals of Trump along the visual evidence of the stock prices reflect an effect from the US election. Despite the news announced which companies were considered to profit or loose from the election, there is no statistical evidence that determine which industries were considered to be the losers and the winners.
This research will make some contributions to the literature. First, it will examine the shareholder’s wealth reaction of the recent U.S. election in seven industries not only in the U.S. but also in nine countries from America. Moreover, it will scrutinize the effect by geographical data source, domestic and international, or by location, North and South. Second, this article will provide a deeper focus of the effect in the Banking industry based on the assumption of a potential benefit risen from the announcement of changing the Dodd Frank Act. Third, the study will evaluate if the response of the market prices of the Investment Banks, Global Systematically Important Banks (GSIBs) and Domestic Systematically Important Banks (DSIBs) differ from that of the rest of the banks. Banks with different characteristics are likely to be differently affected. Fourth, the research will provide evidence of the variables that suggest a relation in the returns. Lastly, the research will provide evidence of the advantages of using an efficient empirical approach, SUR method, by comparing its results with the parametric method. The differences in results will also provide evidence of misleading conclusions derived from an underestimation of the standard error risen from the cross-sectional dependencies in
Figure 3. Bank’s returns around the US election. This figure shows cumulative returns on the portfolios of each country from June 13 to October 7, 2016 (days −22 to +5 around the US election).
6 firm residuals. Furthermore, the empirical methods rely not only in commonly used parametric and non-parametric tests but also in their prediction power.
The analysis proceed as follows. In Section II, I provide a background of the event studies related to announcement, political events and changes in regulation by using daily stock prices in addition to wide literature with the methodology in event studies. Based on the findings in the methodology, I explained the empirical approach in Section III. The data is described in Section IV. Section V documents the results. In Section VI, I examine the results from the previous section with a robust check and document the impact of the elections. The conclusions are discussed in section VII.
II. Literature review
Based on the Efficient Market Hypotheses (EMH) which state that an asset's prices fully reflect all available information in particular the semi-strong hypothesis that as soon as the information is publicly available, the prices will adjust immediately, the result of the US election triggered the market to adjust the prices.
There are empirical findings in the financial institutions stock prices based on the type of announcements. O’Hara and Shaw (1990) studied the effect on bank equity values of the Comptroller of the Currency's announcement that some banks were "too big to fail" (TBTF), they found positive wealth effect which magnitude differed on the size and bank solvency. Eysell and Arhsadi (1989) examined the wealth effects of the announcement of the Basel Committee’s intention to impose a pre-determined minimum level of risk-adjusted capital. They found evidence that the equity of publicly traded banks decreased at the time of the announcement. Correa et al (2014) explored the effects of the sovereign credit rating downgrade on the banks stock return by analyzing the data of 37 countries between 1995 and 2011. The results suggest that the downgrade have a large negative effect on bank stock returns for those banks that are expected to receive stronger support from their governments. Krakaw and Zenner (1986) shown that High Leveraged Transactions (HLT)-related announcements had different effects in the banking industry and that these effects are relative to the size of HLT.
7 Studies related to U.S. politic events have exposed empirical evidence of the effects on the stock market. Luksetich and Riley (1980) examined the Presidential elections and the stock market over the 1900-1968 period and Found evidence that the stock market prefers the republicans. Lobo (1999) examined the impact of U.S. elections and partisan politics on the stock market using jump–diffusion models of daily stock returns from 1965 to 1996 and found that that small stocks perform better under Democrats relative to Republicans. Gift and Gift (2011) examined the stock market reactions of U.S. casino-related businesses to the president's statements, they found that President Obama's statements were followed by significant negative abnormal returns in the segment of companies targeted more towards conventions, trade shows, and tourism, and by significant positive abnormal return for companies with more of a local/regional focus. Hardin et al (2004) evaluated the presidential election cycle within the context of the turn-of-the-month effect in stock returns. The results provide evidence of higher month returns in the second half of presidential terms. The higher turn-of-the-month returns account for most of the additional returns found in the second half of presidential terms evidenced in prior research.
Some studies have analyzed the effects of the introduction of the Dodd’s Frank Act (DFA) regulation. Gao et al (2014) provided evidence that larger and more interconnected financial institutions experienced more negative abnormal stock returns and more positive abnormal bond returns as compared to other banks in the sample. In the case of the Big 6 banks both shareholder and bondholders experienced negative returns. Cyree and Balasubramnian (2016) investigated if the market discipline improved after the DFA by showing a lower discount of size of yields spreads. The discount is reduced by 47% and for the Too Big to Fail, 94%. Kostas (2016) measure the effects of key legislative events of the DFA, the results indicate a mixed response by financial institutions during the various stages of the DFA’s legislative process. In particular, national banks, finance companies and insurance companies seem to welcome concessions made to the industry when the House and Senate Bills were reconciled regardless of the size. In contrast, the reaction was negative when the conference report was passed/agreed to in the Senate. Schäfer, Schnabel and Weder (2015) analyzed the impact of four major regulatory reforms in the banking industry with an even study so as to determine the effect in the banks stock returns and the CDS. Among these reforms, the DFA, in specific the Volcker Rule had the strongest influence.
8 Regarding the methodology to address this research, there is a rich literature on how the so-called event studies are used in economics, finance, accounting and other social sciences to investigate the impacts of news announcements on firms. The parametric tests are the most common ones whereas the non-parametric methods are used as robust tests to verify the findings. The former acts under the premise that the returns are normally distributed. The choice of test statistic is based on the data statistical issues, in particular clustering leads to either cross-sectional correlation of abnormal returns or distortions in volatility. As consequence, both issues overstate the t-statistic by introducing a downward bias in the standard deviation. Thus, there is an over-rejection of the null hypothesis.
Brown and Warner (1985) examined the properties of daily stock returns and its implications in the event studies as well as event study methodologies. He employed the parametric test with the Ordinary Least Square (OLS) market model to determine the Abnormal returns (AR). Brown and Warner (1985) also mentioned that the degrees of skewness and kurtosis in the test statistics changes according to the size of the sample. Moreover, they stated that the cross-sectional variance can lead to well-specified tests and that the non-parametric sign and Wilcoxon rank test were also found to be unaffected by the variance. Campbell and Wasley (1996) compared the performance of a nonparametric test statistic with the parametric test statistic used in prior researches by studying the samples of NASDAQ securities as well as samples of NYSE/ASE securities. They found that the non-parametric test is more powerful when used in clustered events in conjunction with the market model abnormal returns. Corrado (1989) stated that the general problem of the parametric test is that the stock prices returns are not normally distributed. The importance of the normal distribution is driven by the Central Limit Theorem which states that the power of the t-test is affected by the mean and variance of the returns rather by its shape. In particular, the Sum Rank Test has more power under the alternative hypothesis of abnormal security-price performance compared to its parametric counterpart’s test. In addition, Corrado (1989) showed that the Rank Test correctly specifies the significance of the event regardless of how skewed is the cross-sectional distribution of excess returns. Moreover, this test is less affected by the variance of the excess returns than the way a parametric test is. As Brown and Warner (1989) referred in their research, Wilcoxon (1945) defined the ranking method as a way for testing the significance of the differences between two samples. Another approach to test non-normal distribution was
9 studied by Ajinkya and Jain (1989) who documented the behavior of daily stock market trading volume in the NYSE by studying the abnormal volumes around an event. To fix the skewness of the prediction errors, they used a natural logarithm transformation in order to simulate the normal distribution.
Zellner (1962) proposed the SUR method and test for aggregation bias, he encountered that the regression coefficient estimators obtained by a set of regression equations are at least asymptotically more efficient than those obtained by an equation-by-equation application of least squares. Salinger (1992) discussed the effect of the intertemporal or contemporaneous correlation of residuals can result in a significant underestimation of the standard errors. He explained why the standard procedure of assuming estimated ARs could affect the test results significance as well as he proposed the SUR method to capture the correlation of the estimated ARs. Gibbons (1982) based his research of testing the CAPM with the multivariate methodology proposed by Zellner. He stated as an advantage that the method has a better predictive power. Lamdin (2001) discussed the way to test for the impact of regulatory changes in event studies by using a standard approach to test for the abnormal returns. This approach was the modified version of the market model where the modification employs a dummy variable for the event period.
The SUR model is used in many researches as part of the system of equations for the SUR method. Schipper and Thompson (1983) analyzed the impact of the merger-related regulations with the joint return-generating process which takes advantage of cross-sectional dependencies that may significantly affect test statistics and conclusions. They observed that the test statistics ignored the cross-sectional dependencies in firm residuals and the time-series dimension of the residual variability as well as the interdependence of the announcement months across regulations. Doidge and Dyck (2015) documented the interactions between tax incentives and corporate policies after the announcement of imposing corporate taxes on a group of Canadian publicly traded firms. They estimated a modified version of the market model for each trust as a system of equations using SUR to account for cross-correlations in stock returns.
10
I.
Hypothesis and Methodology
A. Hypothesis
Based on the previous methodology and the background information, the hypotheses will be addressed in the following way:
1) The Construction, Oil & Oil Products, Petroleum & Natural Gas Producers, Pharmaceutical and Banking American Industries were the winners from the U.S. Election.
Based on the lowering in the regulation for some industries and the high investments announcements along the job creation promises for other industries, the market reacted positively.
2) There was a negative effect in the market prices of the Industries in the rest of the Americas
The dependence of many industries from North and South American in the trades with the U.S. were affected after the announcements of changing the NAFTA and TPP.
3) The response of the market prices from the companies that are classified as Domestic provided that their data source is local, is higher than that from the companies classified as Domestic & International, for the Automotive, Construction, Health Care, Oil & Oil Products, Petroleum & Natural Gas producer and Pharmaceutical industries. Derived from the announcement of “America First”, the gains for the local companies is expected to be higher than the gains from the companies nonmined as local & international.
4) The impact of the companies from North America was higher than that of the companies of South America
The proximity to the U.S. implies a closer business relation in regards to the number of companies. As a result, the companies from North America are more likely to be harmed after Trump’s victory.
11 5) Provided of the tentative changes in the Dodd Frank Act, the Investment banks, GSIBs
and DSIBs had bigger gains than the rest of the banks.
The establishment of the DFA had a negative impact in the wealth of the shareholders. As a consequence, the banks will have the opposite behavior after pulling down this Act.
6) Based on event studies, the drivers of book-to-market ratio and leverage of the U.S companies are related to the observed abnormal returns
Some research identified a positive relation between the returns and the book-to-market ratio and leverage however, it is unclear if the relation is preserved with the abnormal returns.
B. Methodology
An event study examines the “event window” to determine whether these returns were abnormally positive or negative. The event window is defined here as the entire length of time over which one may look for a stock price reaction to what the analyst has identified as “news” received by investors. A common way to assess the expected (benchmark) return is to apply an asset-pricing model. Abnormal returns (AR) exist when significant differences between the expected (estimated) and realized firm returns over specific periods of time exist.
The observation window contains the information from June 13, 2016 to October 7, 2016, this period denotes the performance during the 100 days previous to the U.S. election. In case that the uncertainty close to the election day could bias the estimation of the returns, I defined the observation window as of one month previous to the election. In this way, the event window considers 3 scenarios, during 1 and 2 days surrounding the event [-1,1] and [-2,2], respectively, being the event day November 9, 2017, [0,0]. The definition of the windows is based on the fact that as the day unfolder, the market’s reaction softens but the shareholders immediately react. Assuming semi-strong market efficiency, any delay in the response of new information is possible only for few days. Hence, the magnitude of model misspecification will be small when measuring few-day AR.
I assessed the event study by using the parametric test and non-parametric test mentioned before. As a first step, I applied the parametric test with the Ordinary Least Square (OLS) market
12 model estimation discussed by Brown (1985). I only employed this method to determine the Abnormal returns (AR) by company; it is not possible to use it for analyzing the portfolios not only because the size of the sample could affect the significance levels but also because of the cross-sectional dependencies among the residuals.
Second, to test the effect of the event among the portfolios I used two non-parametric tests, the Sum Rank test and the Sign Rank test. As stated before, the general problem of the parametric tests is that the stock prices are not normally distributed and that there is a cross-sectional distribution of excess returns.
As the robust check, I used the non-parametric test of Seemingly Unrelated Regressions (SUR) because even the abnormal returns are uncorrelated, estimated abnormal returns could be correlated. Autocorrelations could arise when all the traders do not trade within one day on information they use to rebalance their portfolios, in some cases, the investors adjust their holdings later than others because of the asymmetry information or because they trade only in certain moments to minimize transaction costs. Therefore, the SUR was used to determine the AR per company and the Sum Ranked test and Signed Rank test were used to determine the impact among the portfolios.
Lastly, the data were divided in two samples, Banking industry and the rest of the industries. The purpose of this split was to analyze into detail the effect of the US election by type of bank given the conceivable benefits to the shareholders according to the Trump’s proposals.
1. Parametric test
The null hypothesis is that the mean excess return (AR), for each of the estimation windows, is equal to zero. Thus, the event affects the returns to shareholders. In order to determine this, the market model is the most commonly used method as benchmark of few-days of AR:
𝑅𝑖,𝑡= 𝑎𝑖+ 𝛽𝑖∗ 𝑅𝑚𝑖,𝑡+ 𝜖𝑖,𝑡
where 𝑅𝑖,𝑡 is the return on stock i at time t, i=number of companies; 𝑅𝑚𝑖,𝑡 is the return on
market at time t; 𝑎𝑖 and 𝛽𝑖 are the regression coefficients of the (OLS) for the company i.
The 𝑅𝑖,𝑡 is calculated as the logarithm difference between two daily stock prices whereas the
13 𝑅𝑖,𝑡 = ln(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡) − 𝑙𝑛(𝑃𝑟𝑖𝑐𝑒𝑖,𝑡−1)
𝑅𝑀,𝑡 = ln(𝐼𝑛𝑑𝑒𝑥𝑖,𝑡) − 𝑙𝑛(𝐼𝑛𝑑𝑒𝑥𝑖,𝑡−1)
The estimated coefficients 𝑎̂𝑖 and 𝛽̂𝑖 will replace 𝑎𝑖 and 𝛽𝑖, calculating the 𝐸(𝑅𝑖,𝑡). Then, the
abnormal return is calculated as the difference between the return and the expected return 𝐴𝑅𝑖,𝑡= 𝑅𝑖,𝑡− 𝐸(𝑅𝑖,𝑡). The sum of the daily abnormal returns is the Cumulative Abnormal
Return (CAR) and the Cumulative Average Abnormal Return (CAAR) over the event window is the Average of CAR which for purposes of this thesis the CAAR is called CAR in the result tables. The derived formulas are 𝐶𝐴𝑅𝑖 = ∑𝑡2𝑡=𝑡1𝐴𝑅𝑖,𝑡 and 𝐶𝐴𝐴𝑅𝑖,𝑡= ∑ 𝐴𝐴𝑅𝑡 =
1 𝑁∑ 𝐴𝑅𝑖,𝑡 𝑁 𝑖=1 𝑡2 𝑡=𝑡1 .
As mentioned before, the null hypothesis denotes 𝐸(𝐶𝐴𝑅𝑖) equal 0, assuming that the
𝐴𝑅𝑖~𝑖. 𝑖. 𝑑. (0, 𝜎2). This implies that 𝐴𝐴𝑅~𝑁(0, 𝜎2
𝑁). The test examines the difference between
the announcement period average daily returns and the N average daily returns from outside the announcement period. Hence, the test statistics are given by 𝑇𝑆 = √𝑁𝐴𝐴𝑅𝑠 ~𝑁(0,1) where N is the number of days in the estimation period. If N is large enough, the quantiles of the normal distribution can be used as critical values for the t-test. In event studies, N > 30 is typically sufficient for this. A remarkable feature of the test statistic is that the autocorrelations are assumed to be zero.
2. Non-parametric tests
The sign test examines the proportion of positive cumulative abnormal returns (CARs) in the event window by testing whether it is significantly different from the proportion of cumulative unadjusted returns (CURs) or observed returns while the rank test examines the rankings of all CARs by testing whether the ranking of event window CARs is significantly higher or lower than the ranking of CURs.
a) Sum Rank Test
This test is also known as the Mann –Whitney two-sample statistic. The logic of the Sum Rank test is to verify if two independent samples, X1 and X2, are from populations with the same distribution. Let X1 have size n1 and X2 have size n2. In this case, the samples are the cumulative unadjusted returns and cumulative abnormal returns per company; it is used the cumulative
14 value in order to reflect the time series effect in each of the estimation windows. If the data are tied, average ranks are used. The returns are ranked as 𝑇 = ∑𝑛𝑖=1𝑟𝑎𝑛𝑘𝑖. The randomization
consists of the (𝑛𝑛
𝑖) ways to choose the ranks where n= n1+ n2 and the mean and variance are
defined as 𝐸(𝑇) =𝑛1(𝑛+1)2 and 𝑉𝑎𝑟(𝑅) =𝑛1𝑛𝑛2𝑠2 where s2 is the standard deviation of the combined ranks 𝑠2= 1
𝑛−1∑ (𝑟𝑖− 𝑟̅) 2 𝑛
𝑖=1 .
For large samples, T is aproximately normally distributed, then the standardized value is 𝑧 = 𝑇−𝐸(𝑇)
√𝑉𝑎𝑟(𝑇). Additionally, the probability that determine at which level the medians are not
statistically different is calculated with the next formula 𝑝 = 𝑈
𝑛1𝑛2.
The previous formula is also referred as the area under the Receiver Operating Characteristic Curve (ROC). It is important to mention that the Rank Test is more robust than the t-test given that it is less likely to spuriously indicate significance due to the presence of outliers.
b) Sign Rank Test
This test is used to determine whether two samples have the same distribution. Then the null hypothesis that the matched pairs of two distributions are the same. Let djbe the difference for any matched pair of observations (Cumulative Unadjusted Return and Cumulative Abnormal Return) for any company and let the signed rank be 𝑟𝑗 = 𝑠𝑖𝑔𝑛(𝑑𝑗)𝑟𝑎𝑛𝑘(|𝑑𝑗|). The
randomization of the test statistic T can be computed by 𝑇 = ∑𝑛𝑗=1𝑆𝑗𝑟𝑗 where Sj is either +1 or -1. The test statistic is often expressed (equivalently) as the sum of the positive signed-ranks, T+, with 𝐸(𝑇+) = 𝑛(𝑛+1) 4 and with 𝑉𝑎𝑟(𝑇+) = 1 4∑ 𝑟𝑗 2 𝑛
𝑗=1 . Then, a normal approximation is used
to calculate 𝑧 =𝑇+−𝐸(𝑇+)
√𝑉𝑎𝑟(𝑇+).
3. Robust test
The SUR method is a set of regression equations which determines the coefficient estimators at least asymptotically more efficient than the single-equation least-squares estimators. The main procedure is to simultaneously estimate the coefficient’s regressors which considerate the residual covariances derived from an equation-by-equation application of the OLS.
15 First, the regression for each company is defined by the modified version of the market model:
𝑅𝑖,𝑡= 𝛼𝑖+ 𝛽𝑖∗ 𝑅𝑚𝑖,𝑡+ 𝛾𝑖∗ 𝐸𝑣𝑒𝑛𝑡 + 𝜖𝑖𝑡
Where the 𝑅𝑖,𝑡 is the daily return for the company i, 𝑅𝑚𝑖,𝑡 is the return on the benchmark
country market of the company, and Event is a dummy variable that equals one on the days surrounding the event. The benchmark portfolio includes the Stock Exchange index per country. The regression is estimated for each trust as a system of equations using SUR. The CAR is computed as the multiplication of each 𝛾𝑖 by the number of days in the event period.
The equation system is
[ 𝑦1 ⋮ 𝑦𝑛 ] = [ 𝑋1 ⋯ 0 ⋮ 𝑋𝑖 ⋮ 0 ⋯ 𝑋𝑛 ] ∗ ⌈ 𝛽1 ⋮ 𝛽𝑛 ⌉ + ⌈ 𝑢1 ⋮ 𝑢𝑛 ⌉ = 𝑌 = 𝑋𝐵 + 𝑢
where 𝑋𝑖 considers all the variables involve in the return per each of the n companies. Formerly,
the OLS is estimated such that the equation system turns into:
[ 𝑦1 ⋮ 𝑦𝑛 ] = [ 𝑋1 ⋯ 0 ⋮ 𝑋𝑖 ⋮ 0 ⋯ 𝑋𝑛 ] ∗ ⌈ 𝛽̂1 ⋮ 𝛽̂𝑛 ⌉ + ⌈ 𝑢̂1 ⋮ 𝑢̂𝑛 ⌉ = 𝑌 = 𝑋𝐵 + 𝑈̂
Then, the covariance matrix is defined as 𝑈̂ 𝑈̂ and the equations system is solved. The Breusch ′
and Pagan test 𝑋2statistic is a Lagrange multiplier statistic which test for heteroscedasticity by verifying that the correlation is zero. Thus, the rejection of the null hypothesis means that there is heteroscedasticity. The test is defined by 𝜆 = 𝑇 ∑𝑀𝑚=1∑𝑚−1𝑛=1𝑟𝑚𝑛2 ~𝑥2with
𝑛(𝑛−1)
2 degrees
of freedom where 𝑟𝑚𝑛 is the estimated correlation between the residuals of the n equations
and T is the number of observations. Additionally, it is test that the coefficients 𝐴𝑅𝑖 are jointly
zero in the n regressions. If this is valid then there is no aggregation bias involved and there is a reduction of the Type I-error risks of conducting many t-tests. The hypothesis is repeated for the three estimation windows of each of the portfolios.
4. Regressions
To compare the abnormal returns and to identify the variables that may influence the returns by industry and by estimation window, I defined my control variables based on the ideas
16 developed by Fama and French (1993 and 2015) who mentioned the relationship between the stock returns and the value, size, leverage and growth as well as the notions established by Borio (2013) and Yang and Tsatsaronis (2012) about volatility and leverage in the business cycle. In addition, Bhandari (1985) identified that the stock returns are positively related to the leverage.
To investigate the drivers of the abnormal returns relative to the U.S. industry, the regressions will consider the 𝑅𝑂𝐸𝑖, 𝐵𝑇𝑀𝑖, 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖 and 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖 as well as the dummy variables
𝑆𝑚𝑎𝑙𝑙𝑖, 𝐿𝑎𝑟𝑔𝑒𝑖, 𝐷𝑜𝑚𝑒𝑠𝑡𝑖𝑐𝑖 and 𝐼𝐵𝑖. The first two variables are based on the concepts from
Fama and French (1993) and their purpose is to control the profitability (Return on Equity) of the institution as well as the growth (Book-to-market ratio). The 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖 is the standard
desviation of the returns observed in the estimation window. The 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖 is the debt to
equity.
The 𝑆𝑚𝑎𝑙𝑙𝑖 and 𝐿𝑎𝑟𝑔𝑒𝑖 dummy variables were assigned based on the assets of the companies
by industry. The 𝑆𝑚𝑎𝑙𝑙𝑖 is equal to 1 if the company assets are in the first tercile of its industry
and 𝐿𝑎𝑟𝑔𝑒𝑖 is equal to 1 if the company assets are in the third tercile of its industry. 𝐷𝑜𝑚𝑒𝑠𝑡𝑖𝑐𝑖
is equal to 1 if the source of data for the company is local; in the case of the Banking industry the variable is equal to 1 if the bank is considered as either DSIB or GSIB.
As previously mentioned, the data were divided in two samples, Banking industry and the rest of industries. In regards to the regression analysis, there is a slight difference in the definition of the variables as well as the approach to identify the variables that suggest such that the regressions for Banking are different from those for the other industries.
By industry, the model is defined as 𝐶𝐴𝑅𝑖,𝑡= 𝛼𝑖+ 𝛽1𝑅𝑂𝐸𝑖+ 𝛽2𝐵𝑇𝑀𝑖+ 𝜀𝑖 provided that these
two variables are commonly used to measure the abnormal returns. As a side note, the OLS cannot be computed for small samples, thus those industries with few companies do not meet this requirement.
To have a complete view of the performance of the non-Banking industry, I used a Fixed effect model to control for omitted variables that are common to all industries. The model is 𝐶𝐴𝑅𝑖,𝑡 =
𝛼𝑖+ 𝛽1𝑉𝑎𝑟1𝑖+ 𝛽2𝑉𝑎𝑟2𝑖+ 𝑢𝑖+ 𝜀𝑖𝑡 where the 𝐶𝐴𝑅𝑖,𝑡 is the Cumulative Abnormal Return of
17 𝑉𝑎𝑟2𝑖are 𝑅𝑂𝐸𝑖with 𝐵𝑇𝑀𝑖, 𝐵𝑇𝑀𝑖with 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖, 𝐿𝑎𝑟𝑔𝑒𝑖with 𝐷𝑜𝑚𝑒𝑠𝑡𝑖𝑐𝑖 and 𝐵𝑇𝑀𝑖 with
𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖; the 𝑢𝑖+ 𝜀𝑖𝑡 is the error term and the 𝑢𝑖 is the unit-specific error term.
The regressions for the Banking industry are performed in two ways. The first one is by modelling the CARs for each type of bank with 𝐶𝐴𝑅𝑖,𝑡= 𝛼𝑖+ 𝛽1𝑅𝑂𝐸𝑖+ 𝛽2𝑀𝑇𝐵𝑖+ 𝜀𝑖. The
second one is to model the industry by both using all the predictors at the same time and by using the dummy variables for Investment Banks and Systemic Important Banks. Unlike the other industries, the size of the Banking industry allows the use of a bigger number of predictive variables in the model:
𝐶𝐴𝑅𝑖,𝑡= 𝛼𝑖+ 𝛽1𝑅𝑂𝐸𝑖+ 𝛽2𝐵𝑇𝑀𝑖+ 𝛽3𝑆𝑚𝑎𝑙𝑙𝑖+𝛽4𝐿𝑎𝑟𝑔𝑒𝑖+𝛽5𝐿𝑒𝑣𝑖+𝛽6𝑉𝑜𝑙𝑖+𝛽7𝐷𝑜𝑚𝑖+ 𝜀𝑖
𝐶𝐴𝑅𝑖,𝑡 = 𝛼𝑖+ 𝛽1𝑅𝑂𝐸𝑖+ 𝛽2𝐼𝐵𝑖+𝛽3𝐷𝑜𝑚𝑖+ 𝜀𝑖
where the 𝐶𝐴𝑅𝑖,𝑡 is the Cumulative Abnormal Return of the bank i in the event window t which
can be [0,0], [−1,+1] and [−2,+2]. In all the models, the error terms were corrected for heteroscedasticity.
II.
Data and descriptive statistics
A. Databases
The study focuses on 7 industries which are: Automotive, Construction, Health care, Petroleum & Petroleum Products, Crude Petroleum & Natural Gas, Pharmaceutical and Banking. The source of the data depended on the country which could be: United States (US), Argentina, Brazil, Canada, Chile, Colombia, Jamaica, Peru, Mexico and Venezuela. These countries were selected based on the existing information in the databases.
The data for the US companies was extracted from two databases of Wharton Research Data Services (WRDS). The first one is the CRSP/Compustat Merged, it contains the Fundamentals Annual figures from the most recent fiscal year-end 2015 to classify the company according to the assets, liabilities, book-to-market ratio, equity, net income, SIC, S&P Quality Ranking and International/Domestic flag. These variables will be used to measure the relation between the abnormal return and the company’s attributes. The definition of both variables S&P Quality
18 Ranking and International/Domestic flag is in the appendix. The second database is the CRSP Data which contains the daily stock prices.
The US sample consists of publicly trading US institutions grouped by Standard Industrial Classification (SIC). These are Motor Vehicles and Passenger Car with SIC code 3711, Construction Machinery and Equipment 3531, Home Health Care Services 8082, Petroleum and Petroleum Products Wholesalers, except Bulk Stations and Terminals 5172, Crude Petroleum and Natural Gas 1311 and Pharmaceutical 2834. The Banking industry classified commercial banks with code 6020, these operate under Federal or State charter: National Commercial Banks chartered under the National Bank Act with code 6021, State Commercial Banks 6022 and Commercial banks NEC (Not Elsewhere classified) which do not operate under Federal or State charter, and the Security Brokers, Dealers, and Flotation Companies with code 6211. For further references the Security Brokers, Dealers, and Flotation Companies will be denoted as Investment Banks. The time frame is 2016.
The Federal Reserve System defined the list of 34 DSIBs which for matters of this study, it considers 32. MUFG Americas Holdings Corporation was excluded from the sample because the information was unavailable whereas BancWest Corp belongs to BNP Paribas which is considered as GSIB.
The list of the 30 GSIBs is defined by the Financial Stability Board in consultation with Basel Committee on Banking Supervision. This list was reviewed on November 2016 and did not changed. The information of Group BPCE was unavailable, therefore this study considers 29 GSIBs. The banks classified as Investment Banks, DSIBs and GSIBs are mentioned in the Appendix.
The data for Canada and the Latin American countries was extracted from Datastream. The information available includes the stock prices of all the companies included in the 7 industries. The market return data for all the countries was extracted from Datastream as well.
In order to have reliable predictions in the regressions, I excluded the companies that do not fulfill the 80% of the stock price available information.
The table I displays the 879 companies analyzed in this research distributed by country and by industry; 477 companies belong to the US industries, representing the 54% of the sample, from which there are 297 companies of the Banking industry and 180 companies distributed in the
19 rest of industries. The 402 non-US companies represent the 46% of the sample, being Canada the country with the biggest number of companies. The largest industry is Banking with 371 banks, 297 US banks and 74 non-US banks. The Pharmaceutical industry is the second largest with 147 companies which are mainly located in Canada and the US.
B. Descriptive Statistics
The analysis was performed by splitting the information between U.S. and American companies and between U.S. banks and the Non-U.S. banks given that the Banking industry has a bigger sample.
The table II shows the descriptive statistics of the daily returns observed in the estimation window of July 1, 2016 to October 31, 2016 for Automotive, Construction, Petroleum & Natural Gas, Health Care, Oil & Oil Products and Pharmaceutical industries. The classification Domestic and Domestic & International denotes if the company has local or local & international business with the aim of identifying a segregated performance. The average daily return of the Domestic companies is 0.14% whereas is 0.03% for the Domestic & International ones. The returns for the Domestic companies have a wide range of -79.15% to 80% which values are driven by the Pharmaceutical and Oil & Natural Gas industries. The companies that dragged the values up to these limits are Ampio Pharmaceuticals, Biostar Pharmaceuticals, Enerjex Resources, Infinity Pharmaceuticals, Mast Therapeutics, Resolute Energy and Threshold Pharmaceuticals. Because
Industry Ar Br Ca Ch Co Ja Mx Pe Ve US Total
Banking 6 22 14 5 5 3 9 6 4 297 371
Automotive 1 6 7 0 0 0 2 3 0 23 42
Construction 6 12 23 9 5 1 14 5 2 12 89
Oil & Natural Gas 3 6 2 1 2 0 0 1 0 73 88
Health Care 0 5 59 3 0 1 1 0 0 2 71
Oil & Oil Products 2 2 62 0 1 0 0 0 0 4 71
Pharmaceutical 1 3 76 0 0 0 1 0 0 66 147
Total 19 56 243 18 13 5 27 15 6 477 879
The table shows the distribution of companies by industry and by country. The columns denote the countries Argentina (Ar), Brazil (Br), Canada (Ca), Chile (Ch), Colombia (Co), Jamaica (Ja), Mexico (Mx), Peru (Pe), Venezuela (Ve) and United States (US)
Summary statistics - Distribution of companies
20 of these companies, the kurtosis has big values. The last column of the table shows the kurtosis which represents the concentration of daily returns, if the number is less than 3 it means the distribution produces fewer and less extreme outliers than does the normal distribution. The skewness represents the asymmetry of the probability of distribution such that a negative number indicates that the left tail is longer than the right one. The skewness for the Domestic companies is positive while it is negative for the other companies.
The table III shows the summary statistic for the daily returns for the sample of 325 Non-U.S. companies distributed in the 6 industries and grouped by regions. The estimation window which refers to is the period between June 13 to October 7, 2016. The classification of Region was determined based on the proximity of the country to the US such that Mexico and Canada were classified as North America and the rest of the countries as Central & South America. The population of North America is three times bigger than that of Central & South America, this classification will be useful to determine if there is a distinguished effect by region. The average return is 0.13% for the North America countries and 0.18% for the countries. As it could be expected, the ranges of returns are wide because of the economic factors of each country, in specific the Oil & Natural Gas and Pharmaceutical industries reflect the wider range which is the same situation as in the US companies. The main companies that affected the ranges are CVR Medical, Easy Technologies, Eplay Digital and PREMD. In regards to the skewness, both regions have positive values.
The table IV shows the summary statistics for the returns of the US Banks, the type of bank classification DSIB stands for Domestic Systematically Important Banks and it is based on the list published by the Federal Reserve System whereas the No DSIB stands for Non-Domestic Systemically Important Banks. The State Commercial banks have the biggest share and the wider range of returns. The average daily return of the 299 banks is 0.07%. There is a slight difference between Domestic and No Domestic however; among the Industrial Classification there are bigger differences, 0.08%, .04% and .06% for Investment Bank, National Commercial Bank and State Commercial Bank. Regardless of the Industrial Classification, the skewness is negative.
The table V shows the descriptive statistics of the Non-U.S. Banks daily returns observed in the estimation window of June 13, 2016 to October 7, 2016. The sample size is 74 banks, 23 from North America and 51 from South and Central America. The average daily return is .06% which
21 is smaller than the value of the U.S. banks, this reflects a lower gain to the shareholders. The range of values goes from –35.67% to 17.35%; the lowest value was driven by the Mercantil Inv. Bank from Brazil. The value of skewness is negative so the situation is similar to the one of the U.S. banks. Unlike the U.S. Banks, the Non-U.S. banks have a higher kurtosis which represents a bigger number of outliers.
From the summary statistics of the tables II, III, IV and V, the skewness for the U.S. banks and most of the Non-U.S. banks is negative, denoting a longer or fatter tail than the right side. In contrast, the skewness for the rest of the industries is positive.
Furthermore, from these summaries it can be seen that the kurtosis is higher than 3 which denotes that the average values by either country or industry, respectively, do not fulfill the requirement of a normal distribution. Because of this, the parametric test cannot be performed by portfolio.
The table VI exposes the summary statistics for the fundamental values of each industry. The Return on Equity (ROE), Book-to-Market ratio (BTM), Leverage and Volatility were weighted by Assets in order to implicitly reflect the size of the company in the industry. The industry with the highest average of assets is Banking with 143,426 million whereas Health & Care is the one with the lowest average, 747 million. Yet, the former is highly leveraged.
The ROE variates among the industries, in particular the Construction and Oil & Gas Natural industries denote a negative ROE. As a side note, a negative ROE does not necessarily mean that a company has an unhealthy balance sheet or that it is a bad investment; this can mislead its status so it is recommended to take a look to its cash flows before determining its situation. In this analysis, the ROE will be used as a control variable in the regression and will not determine the position of the company.
The Banking and Construction industries have BTM above one unlike the rest of the industries. The aim of the BTM is to identify if the securities are overvalued or undervalued such that a value above one implies that the stock is undervalued. Thus, it could be inferred that these two industries have undervalued stocks. Finally, the last column shows the volatility which is the stock return volatility during the Estimation Window. The volatility will be useful to determine the effect in the abnormal returns.
22
III.
Results
Provided that the abnormal returns were computed with both the Parametric Method and the SUR method, I considered relevant to present only the evidence for the first three hypotheses in the Results Section whereas all the hypotheses are presented in the Robust Check Section. The aim of addressing all the hypotheses in the Robust Check Section is to perform the comparative of results with both methods along with the validation of the abnormal returns in order to expand the analysis by answering the hypothesis relative to the grouped ARs tests. The previous section presented the description of the population and the way it is classified so as to test the hypothesis. As a first step, I used the Parametric Method to calculate the Abnormal Returns for each of the 879 companies for the three estimation windows [0,0], [-1,+1] and [-2,+2]. This method is employed to determine the impact by company rather than by portfolio.
The histograms of the ARs for the seven industries using the primary 3-day event window are shown in the Figures 3 and 4. The first column refers to the event day and the second and third column to the [-1,+1] and [-2,+2], respectively. Among the industries, the distributions of the ARs are significantly different. Subject to the sample size for both the Oil & Oil Products and Health Care industries, the distributions will not show neither a trend nor a shape. The histograms for the event day can hardly show a distribution with the exception of the Petroleum & Natural Gas and Pharmaceutical industries where the shape could resemble a Normal distribution with a big right tail. By observing the histograms for the estimation window [-1,+1] for the Automotive and Construction industries, the ARs considerably shift to the right side of the plots, however the shape for the Automotive industry for the estimation window [2,+2] looks like a Normal distribution in contrast to the Construction industry. Furthermore, in cases such as both the Petroleum & Natural Gas and the Pharmaceutical Industries for the estimation window [2,+2], the distributions are reasonably symmetric around zero, appearing to support the hypothesis of no effect. Along the tree estimation windows, the banking industry denotes a positive skewness which might be affected by its internal industrial classification of Commercial Bank, National Commercial, State Commercial, Commercial Not Everywhere Classified, Investment Bank or DSIB. This gives rise to the question if the effect of the US
23 election was stratified by the type of bank, being more beneficial to some type of banks while unfavorable to others.
The visual evidence of the ARs provide a basic impression of the impact of the US election, it could be expected in some industries and striking in others, however the statistical tests are required to identify the effect as well as the magnitude of the election in each industry.
A. Hypothesis 1: The Construction, Oil & Oil Products, Petroleum & Natural Gas Producers, Pharmaceutical and Banking American Industries were the winners from the U.S. Election
After the US election, the newspapers proclaimed the industries that were considered as the winners based on the President Trump’s proposal. In general, it was concluded that the Pharmaceutical, Health Care, Infrastructure and Oil industries were profited, the Automotive was damaged and the Banking had a mix effect. The Parametric Method determined the average abnormal returns by industry, nonetheless it cannot determine the statistical significance because of the cross-correlation of the returns. Therefore, the Signed Test examines if the proportion of positive Average Cumulative Abnormal Returns (CAR) are significantly different from the proportion of average Cumulative Unadjusted Returns (CUR). Unquestionably, the null hypothesis is that there is no change between the returns.
It is essential to highlight that even though the average cumulative returns look similar, the internal distribution of the company might be different. Thus, if the two average cumulative returns values are similar between them, it does not necessarily mean that the U.S. Election did not impact on their returns.
The table VII shows the results for the Sign Test of the U.S. Industries. The sample size for Petroleum & Natural Gas as well as for Pharmaceutical is larger than the rest of the industries, then I clustered them based on the median of its Assets as of 2015. The cluster could provide at a first sight if the size of the companies could have an implication in the returns. By considering the clusters, the classification of the industries changed from 6 to 8 sets. For each of the estimation windows [0,0], [-1,+1] and [-2,+2], the CAR and CUR is calculated.
In the event day, the Automotive industry indicate a negative return at the 5% significance level; the significance level improves to 1% for the estimation windows [-1,+1] and [-2,+2] but
24 with positive returns of 1.43% and 4.49%, respectively. The Petroleum & Natural Gas with High Assets also presents a positive significant impact at the level of 1% for the three estimation windows. In contrast, the same industry but with Low Assets do not present a significant effect. The Construction industry has 6.10% and 9.35% returns at the significant level of 5% for the estimation windows [-1,+1] and [-2,+2], correspondingly. The industry with the highest return was the Pharmaceutical - Low Assets with 15.84%, yet it was not significant. In regards to the Health Care and Oil & Oil Products industries, there is not enough data to possibly obtain meaningful results. The Banking industry has a positive and significant abnormal return which growths while the observation window increases.
Hence, it can be concluded that even though all the industries have positive average CAR, it is only significant at 10% level for the Oil & Oil Products industry and at 1% level for the Automotive, Construction, Petroleum & Natural Gas and Banking Industries. The last three industries are aligned with our hypothesis but the Automotive it is not. The unexpected outcome could be explained by the participation share of the U.S. companies that manufacture in Mexico. The number of factories in Mexico and the dependency of the US companies on them might not be relatively high compared to the size of the complete industry. Furthermore, the announcement of renegotiating the North American Free Trade Agreement between the US, Canada and Mexico, might not have immediate consequences as the conditions of the renegotiation were uncertain.
B. Hypothesis 2: The Investment Banks, and DSIBs, GSIBs had bigger gains than the rest of the banks
The detail of the Banking industry is provided in the table VIII where the Industrial Classification refers to the type of bank which can be Commercial Bank, National Commercial, State Commercial, Commercial Not Everywhere Classified (NEC), Investment Banks, DSIB and GSIB. In order to provide a bigger detail of the Banking industry, I clustered the first three groups based on the median of its Assets as of 2015. The CARs are significant at 1% in most of the groups for the three estimation windows. The average CAR in the observation window [-2, +2] reached a maximum return of 15.32% and a minimum of 3.69%. Additionally, the CAR of the banks with lower assets is inferior than that from the banks with high assets. Because the
25 abnormal returns apparently react with the size of the Assets, I raise the question if the distributions can be compared by using the non-parametric Kolmogorov-Smirnov test as well as the Gini Index. Further research will be necessary to understand the relationship and understand the proprieties for measuring abnormal returns.
The table also shows the average CUR and CAR for each of the type of banks where the Investment Banks, DSIB and GSIB have a CAR in the estimation window of [-2,2] of 11.92%, 10.05% and 6.16%, respectively. By comparing the CAR for each row, the straightforward conclusion is that the Commercial, National and State banks with high assets have higher abnormal returns than the returns from the Investment Banks, DSIBs and GSIBs. Yet this comparison does not imply that Investment Banks, DSIBs and GSIBs have lower gains than the rest of the banks given that the comparison is made by type of bank and not by these three type of banks versus the rest of the banks. The conclusion for this hypothesis will be further explained in the Robust Check Section.
C. Hypothesis 3: There was a negative effect in the market prices of the Industries in the rest of the American continent
The table IX shows the results for the Sign Test of the Non-U.S. Industries. The sample size for Construction, Health Care, Oil & Oil Products and Pharmaceutical are bigger than the rest of the industries, then I separated each one in two groups based on the median of its stock return volatility observed during the estimation window. The sample for Banking is partitioned by country in the table X. The groups could sustain a stronger sensitivity of the abnormal returns. By considering the clusters, the classification of the industries increased from 6 to 10 sets. For each of the estimation windows [0,0], [-1,+1] and [-2,+2], the CAR and CUR is calculated. In the estimation window [0,0], the average CAR for the Health, Oil & Oil Products and Pharmaceutical industries with High volatility is significant at the 1% level; from these, only the Pharmaceutical and Oil & Oil products with low volatility are significant at the 5% level. For the second estimation window, the Oil with High Volatility industry maintain it significance level. The Construction, Health Care with High Volatility and Petroleum & Natural Gas industries have negative significant abnormal returns of -1.38%, -3.23% and -4.15%, respectively. In the last
26 estimation window, both the Construction and Pharmaceutical industries with Low Volatility as well as the Health Care with High Volatility disclose -4.71%, -3.80% and -2.58% significant returns at the 5% level. Surprisingly, the Banking industry has a small but positive abnormal return of .63%.
The table X shows the same analysis as the one from the previous table but only for the Banking industry segmented by country. In the estimation window [-1,+1] there is a negative effect for Mexico and Argentina at the significance level of 5% as well as for Brazil at the 1%. In addition, in the estimation window [-2,+2] with a 5% level of significance, the abnormal return is -3.24%, -3.01% and -1.19% for Mexico, Argentina and Chile. On the other hand, Canada has a positive CAR.
Unlike the U.S. industries with the exception for both industries Banking and Pharmaceutical – high volatility, those from the Non-U.S. countries, acquainted negative returns. Considering the estimation windows of [-1,+1] and [-2,+2] and the 5% of significance level, the Construction - Low Volatility, Health Care - High volatility, Petroleum & Natural Gas and Pharmaceutical – Low Volatility industries were damaged. The Banking industry was harmed in Mexico, Argentina, Brazil and Chile.
IV.
Robust Check
This section addresses all the hypothesis by using the SUR method to calculate the Abnormal Returns. The first three hypotheses were answered in the Results Section but in order to further answer the other hypotheses, I considered necessary to validate the abnormal returns with the SUR method, by both industry and country. The abnormal returns were calculated under the SUR assumption that there is a high cross-sectional correlation in the stocks return residuals both because a company belongs to the same industry which might have another affected company and because the event occurs in the same day for all the affected companies. The table XI shows the results of the SUR method for the US industry by industrial classification. Panel A shows the results of all industries except Banking and the Panel B the results of the Banking industry by type of bank. The average abnormal return is determined by the system of equation of each industrial classification. The significant values SUR Breusch-Pagan Test,
27 displayed next to these returns, test the null hypothesis that the correlation of the residuals is zero. Thus, all the ARs computed reject the null hypothesis at a significant level of 1%. The Jointly test of AAR test that the coefficients of the average abnormal return are jointly zero in each one of the industrial classifications, implying no change due to the election. The factual power of a test is verified when testing the jointly hypotheses. By rejecting the null hypothesis that the coefficients are zero, the test demonstrates the coefficients predictive power. The significant values for this test are shown next to the p-values. In the panel A, the Petroleum & Natural Gas – High Assets failed to reject the null hypothesis in the estimation window [-2,+2]. As it was mentioned before, the Health Care and Oil & Oil Products industries do not account with enough data to obtain meaningful results.
The table XII shows the results of the SUR method for the Non-U.S. industries. Panel A shows the results of all industries except Banking while Panel B shows the results for this industry by country. The ARs of most of the industries displayed in the Panel A reject the null hypothesis that the correlation of the residuals is zero with the exception of Health Care industry however, the Jointly test of AAR is significant for the High Assets cluster. Therefore, with the SUR Breusch-Pagan Test, in most of the industries it is found that the correlation of the residuals is different from zero. Apart from the Health Care industry, the Oil & Oil products with low volatility, the Petroleum & Natural Gas and the Pharmaceutical low volatility fail to reject the null hypothesis of the Jointly test of AAR in some estimation windows.
In the Panel B, the banking industry from Peru do not reject the null hypothesis of the SUR Breusch-Pagan Test in any estimation window whereas Mexico only rejects it in the [-2,+2] with a significant level of 10%. Thus, the correlation of the residuals is not zero to all the countries. Relative to the Jointly test of AAR, the banking industry from most of the countries failed to reject the null hypothesis. Thus, the predictive power of the ARs from the Non-U.S. countries is not as strong as the power of the ARs from the US.
As a result, the SUR method computed the ARs for either industry or country by exploiting the OLS residual covariance matrix and establishing the null hypothesis that the correlation of residuals is zero. In most of the cases the null hypothesis is rejected, consequently the coefficients, ARs, determined with the system of equations, SUR, will be more efficient and consistent than those calculated with the one-by-one OLS, i.e. Parametric Method.
28 A. Hypothesis 1: The Construction, Oil & Oil Products, Petroleum & Gas Producer, Pharmaceutical and Banking American Industries were the winners from the U.S. Election
The table XIII shows the results with the Sign Test as in table VII but with the SUR method instead of the Parametric method. The average CURs are the same as those in table VII, the difference lies in the average CARs. Despite the CARs from table VII and table XIII are similar between them, the statistical significance changes. As an example, the significance is different for both the Automotive and Construction industries. Moreover, the results for the Oil & Oil Product industry with the Parametric method in the estimations window [-1,+1] and [-2,+2] indicate an effect at the significance level of 10% whereas the results with the SUR method do not identify any effect.
The table XIV present the results for the Banking industry with the SUR method. The ARs are significant at 1% in most of the groups for the three estimation windows which implies a positive impact in the industry. This table is comparable with the table VIII. Even though the average CARs are similar in both tables, there are minor differences in the coefficients. Instead of the maximum return of 15.32% and the minimum of 3.69% observed in the observation window [-2, +2], the maximum is 15.38% and the minimum 1.19%. The relationship between the abnormal return from banks with high and low assets prevails.
Therefore, the conclusions from the Robust check marginally differ from the conclusions stated in the Results Section. The effects in the Automotive, Construction, Petroleum & Natural Gas and Banking Industries are still present, yet the difference remains on the significance level. Though, the Oil & Oil Products industry does not have significant impact anymore.
B. Hypothesis 2: The Investment Banks, and DSIBs, GSIBs had bigger gains than the rest of the banks
Even though the average CARs are similar in tables VIII and XIV, there are slight differences in the significance levels of National banks with high assets, State banks with low assets, Commercial Banks NEC and GSIB. The biggest difference is for Investment banks which were previously determined to have a significant positive effect at the level of 10% in the estimations
29 window [-1,+1] and [-2,+2]; the SUR method identifies the effect at a significant level of 1% for the first window but also no effect for the second window. As a result, the Investment Banks, DSIBs and GSIBs have a positive and significant effect in the window [-1,+1] and only the DSIBs and GSIBs keep the effect for the window [-2,+2].
This hypothesis was not entirely addressed in the Results section because I considered necessary first to validate the CARs with the Robust check. Then, the panel A from the table XV shows the results from the OLS regression where the CARs are explained by the book-to-market ratio as well as the dummy variables for Systemic Important Banks and Investment Banks. With respect to the relation of the BTM on the returns, Fama and French (1992) found that the average stock returns increase with the BTM, it is inferred that if the average stock returns are expected to increase, the average abnormal returns are expected to decrease. This assumption is based on the definition of the average abnormal return as the difference between the stock return and its expected value, the increase of both will imply a lower gap, consequently a lower abnormal return. Thus, an increase of 1 in the BTM will decrease the CARs in 1.85%, 2.14% and 3.04% in the event day and the 1 and 2 days surrounding the event, respectively. In relation to the Systemic Important Banks (SIBs) and the Investment Banks (IBs), there is a positive effect such that the increase for IBs is higher than the increase for SIBs.
The Panel B indicates the comparison of GSIBs which were divided in US banks and non-US banks so as to provide a direct effect of the US Banking industry. Given the size of the sample, the OLS was not suitable to determine the impact in the CARs but in order to identify if there was a different effect between both samples, I used the Rank test. The values for these comparisons were calculated as the difference between each of their variables and their significance values, next to the difference values, were determined with the two-sample Rank Sum Test. The probability values indicate if the medians of the samples are not statistically different at any level smaller than the one specified. Clearly, the GSIBs – USA have higher abnormal returns than the non-USA banks, accounting a spread of 4.30%, 11.54% and 15.08% for the estimation windows of [0,0], [-1,+1] and [-2,+2], respectively. Furthermore, the comparison between them are significant at the 1% level with probabilities of 9.52%, 17.26% and 23.21%.
To conclude, there is a positive relation between the abnormal returns and the IBs and SIBs, such that the effect for the Investment Banks is higher than the effect for the SIBs in both the
