Efficiency in the Brazilian stock market
Is the São Paolo stock exchange semi-‐strong form efficient?
Economics and Finance
University of Amsterdam
Bachelor Thesis
Harriëtte ten Brinke
6132030
19
thof June 2014
Abstract
This thesis focuses on semi-‐strong form efficiency in Brazil, namely the São Paolo stock exchange. In a semi-‐strong form efficient market prices of securities reflect all publicly available information, such as past prices and prospects. An event study, with the event being stock splits between 2000 and 2011, is conducted to test for efficiency. CAPM is used as the expected return model. Using a sample of 18 companies’ returns in various sectors, cumulative abnormal returns for the market are not significantly different from zero. This indicates semi-‐strong form efficiency.
Table of contents
Abstract 1 1. Introduction 3 2. Literature review2.1 Brazil as an emerging economy 4
2.2 Equity market in Brazil 5
2.3 Efficient Market Hypothesis 6
2.4 Forms of efficiency and its early empirical results 8 2.5 Criticism regarding the Efficient Market Hypothesis 10 2.6 The implications of efficiency in emerging economies 13 2.7 Evidence on efficient capital markets in emerging economies 14 2.8 Efficient capital markets in BRIC countries 15
3. Data 17 4. Methodology 20 5. Empirical results 23
6. Limitations and recommendations 26
7. Conclusion 27
8. References 28
1. Introduction
Over the last decades, Brazil has experienced tremendous economic growth. The country’s government is very keen on attracting investors and maintaining a high growth rate. Current president Roussef stated that, addressing captains of industry and bankers, emerging economies such as Brazil ‘’have the biggest investment
opportunities’’ (Follath & Hesse, 2014).
One of the contributors to a high growth rate can be an efficient capital market. Efficiency in the market supports efficient resource allocation, which in turn leads to higher economic growth (Huang, Khurana & Pereira, 2009). As for India, another emerging economy and fellow BRIC country, the World Bank announced in its
development report of 2013 that if India were to allocate its resources more efficiently, economic growth could rise 60 percent (Jha, 2013).
It is therefore interesting to study efficiency in the stock market in relation to a country that has experienced high growth, like Brazil. The semi-‐strong form of efficiency in the São Paolo stock exchange will be studied in this thesis, by conducting an event study. This event study involves the announcement of stock splits for a multitude of companies listed on the exchange.
First, the existing literature will be reviewed. The focus is on Brazil itself, the Efficient Market Hypothesis (EMH), its critics, and the implications of the EMH for emerging countries plus relevant evidence. The next two sections will elaborate on the data used for the event study and the methodology in conducting the study. The Capital Asset Pricing Model is used as an expected returns model in order to obtain abnormal returns. The empirical results show that the average and cumulative abnormal returns for
individual companies and the market exhibit no value significantly different from zero at appropriate significance levels. Therefore, it cannot be rejected that the Brazilian market is semi-‐strong form efficient. However, declaring semi-‐strong form efficiency might be too extreme. Reasons behind this will be discussed in ‘Limitations and
recommendations’.
2. Literature review
2.1 Brazil as an emerging economy
Brazil experienced one of the highest economic growth rates in the mid-‐twentieth century, but the debt crises of 1980 and 1990 did stagnate this growth. After the 2003, growth rate picked up again, to about five percent in 2004 (Thomas, 2006).
Brazil was attained the status of an emerging economy under the power of president Lula, with his presidential term starting in 2003. Foreign investors were attracted and the central bank was given higher autonomy to separate political decision making from decision making out of an economic perspective (Carrasco & Williams, 2012).
Emphasis on infrastructure is one of the policies supporting the emergence of Brazil. With the ‘Growth Acceleration Plan’, commencing in 2007, infrastructure is improved to facilitate economic growth. Decent infrastructure leads to social inclusion so that
everyone can participate in the economy (Carrasco & Williams, 2012).
The attraction of foreign investors was facilitated in several ways. Import tariffs have been lowered to make it attractive for foreign companies to sell their products in Brazil. High interest rates and high returns associated with rapid growth have stimulated foreign direct investment. The level of foreign investment is not set to a maximum or minimum, unlike many other countries. The stock exchange of São Paolo was made up to standard with the markets around the world, with information on disclosure
requirements for listing on the exchange. The availability of information creates confidence of foreign investors (Carrasco & Williams, 2012).
The US financial crisis of 2008 did have some effect on Brazil’s GDP, but was short-‐lived. Brazil’s exposure to mortgage-‐backed securities and dependence on trade with
developed countries was low (Carrasco & Williams, 2012).
The ongoing efforts by the government to create a more favorable debt level and
reforms of social security, labor markets and taxation could improve growth prospects. However, a crucial factor is the improvement in income distribution; the poor need access to financial services in order to achieve productivity growth. When more people are added to the growth process, the strength and sustainability of growth will increase (Thomas, 2006).
When compared to the world’s economic growth, measured by annual percentage change in GDP, Brazil on average either has a higher growth rate or just below the
worlds rate. One exception is 2012, where Brazil’s GDP growth rate is 1,5 percent lower relative to the world’s growth rate. The following graph represents this1.
Annual GDP growth of the world and Brazil (in %)
2.2 Equity market in Brazil
The major stock exchange in Brazil is the Bolsa de Valores, Mercadorias e Futuros de São Paolo (BM&F Bovespa). The BM&F Bovespa can be considered a small stock exchange; it has an average of 500 listed companies and is largely dependent on international
investors. However, in 2010 it did rank 11th of the world stock exchanges with a market capitalization of US$1545 million (Sandoval, 2012).
Stocks of mining, oil, biofuel and gas companies make up a large part of the exchange. The stock exchange is more volatile than stock exchanges in developed countries, exhibiting more risk but also more opportunities to gain (Sandoval, 2012).
The economic activities by mining, oil and metallurgy, financial and construction firms are most important for the Brazilian stock market, indicated by an asset graph
established by Sandoval (2012) in which connections between companies are made at different distances. Distance measures how uncorrelated two returns are from one another. Sandoval finds that for the smallest value of distance, firms in the above mentioned sectors connect. Adding more distance, more sectors are present with
1 Source: www.worldbank.com
accompanying firms connecting, but the sectors mining, oil and metallurgy, financial and construction have the most connections.
2.3 Efficient Market Hypothesis
The Efficient Market Hypothesis (EMH) was first introduced by Eugene F. Fama (1970). An efficient market is a market in which prices always fully reflect all available
information and stock prices are unpredictable. The definition of ‘all available
information’ is connected to the form of efficiency, which will be discussed in the next section.
First, one must define what ‘fully reflect’ actually means. Fama distinguishes three approaches:
With every ‘Expected Return Model’, the expected price for a security at a given time is equal to the price of the same security in a preceding time period, adjusted for the equilibrium expected return, given all available information. The fact that one can
express the conditions of market equilibrium in terms of expected equilibrium returns is an assumption, but this assumption is needed to empirically test the EMH. This
assumption also leads to a ‘fair game’; when expected prices include all available information, the actual price will be equal and thus rules out any abnormal returns. The Submartingale Model describes that an expected price for a security in a future period, conditional on the information known in the current period is greater than or equal to the price in the current period. Expected returns in the future period, given all available information in the current period, would therefore always be non-‐negative. This implies that expected profits of buying and holding in the future period and trading based only on the information are equal.
The Random Walk Model states that a full reflection of available information suggests that returns are independent and identically distributed (i.i.d.). This model is an addition to the Expected Return model, as it focuses more on stochastic properties of returns (Fama, 1970).
Fama (1970) recognizes that a full reflection of all available information might be a bit extreme:
‘’Though we shall argue that the model stands up rather well to the data, it is obviously an extreme null hypothesis. And, like any other extreme null hypothesis, we do not expect it to be literally true’’. (p. 388)
Fama (1991) reviews the work on EMH twenty years after initial introduction. He finds that the notion of ‘market efficiency’ faces two problems: there are information and transaction costs and there is a joint-‐hypothesis problem. For prices to reflect all available information, information costs must be zero. In reality these costs are always positive, hence the extreme version of market efficiency must be false. However, the EMH excluding these costs can serve as a benchmark for the adjustment of prices to different kinds of information. The joint-‐hypothesis problem arises from the fact that the market efficiency is jointly tested with an asset-‐pricing model. However, Fama argues that this does not mean that the joint-‐hypothesis problem makes empirical work
uninteresting. He states: ‘’The empirical literature on efficiency and asset-‐pricing models passes the acid test of scientific usefulness. It has changed our view about the behavior of returns, across securities and through time.’’ (p. 1576)
Fama (1991) now focuses on tests for return predictability, event studies and tests of private information, instead of returning to the forms of efficiency introduced in his earlier work.
There has been a change of focus in tests for return predictability, from a short horizon to a long horizon. The evidence on the existence of predictability in returns is growing; there seems to be variation of expected returns through time. This leads to a joint-‐ hypothesis problem: is this variation rational or do stock prices behave irrationally and deviate from fundamental value? Moreover, the supposed predictability might be spurious, due to data mining or just simply based on chance (Fama, 1991).
If the event date and price movement can be determined accurately, Fama (1991) sees the issues with an equilibrium-‐pricing model as a ‘’second-‐order consideration’’ (p. 1607) for event studies. Therefore, evidence obtained from event studies is less impacted by the joint-‐hypothesis problem. Recent evidence is largely supportive of efficiency, as the price adjustment to firm-‐specific information occurs quickly (Fama, 1991).
New evidence on private information is in line with earlier evidence that insiders possess private information that is not reflected in a stock price. The test whether
investors actually possess private information is associated with the assessment of abnormal returns for a long period. This leads again to a joint-‐hypothesis problem, because abnormal returns can be a sign of either inefficiency or a failing pricing model (Fama, 1991).
2.4 Forms of efficiency and its early empirical results
The Efficient Market Hypothesis recognizes three forms of efficiency, the weak-‐, semi-‐ strong and strong form efficiency (Bodie, Kane & Marcus, 2011).
In a weak-‐form efficient market stock prices reflect all information found in past trading, such as prices and volume. This past trading data has little to no cost to obtain and thus is available to everyone (Bodie, Kane & Marcus, 2011). Weak-‐form efficiency is consistent with the Random Walk Model. Bachelier (1900) first tested this model in his paper about speculation and found that the prices of commodities do follow a random walk.
The work of Kendall (1953) involved studying whether time-‐series of 19 US industrial share prices in the period 1928-‐1938 show any serial correlation. He concluded that the stocks show very small serial correlation. But, Kendall claims: ‘’Such serial correlation as is present in these series is so weak as to dispose at once of an possibility of being able to use them prediction.’’ (p. 9) This is in line with the weak form of EMH.
The semi-‐strong form efficiency hypothesis entails that, in addition to past trading data, the prospects of a company are also embedded in the share price; all publicly available information is reflected in the price. It focuses on the speed of price adjustment, given the occurrence of an event. These events are, for instance, announcements of stock splits or annual earnings announcements. A multitude of events can occur and for each event a test must be carried out separately. Each separate test will then add to the validity of the Martingale Model, that is, expected price for a security in a future period, conditional on the information known in the current period is equal to the price in the current period (Fama, 1970).
The pioneers in testing for semi-‐strong form efficiency are Eugene F. Fama, Lawerence Fisher, Michael C. Jensen and Richard Roll (1969). Their paper focuses on the
from 1927 up to the end of 1959, they investigated returns for 940 splits for stocks traded on the New York Stock Exchange (NYSE).
Results were as follows: the cumulative abnormal return increases substantially in the time period prior to the split, but after the split there is no further movement. The fact that the cumulative abnormal returns increase before the stock split, might be due to a changing view on future earnings potential of the company, or leakage of information but has not so much to do with the stock split itself. When the date of the announcement of the split was used in the research, rather than the split date, results did not differ. They conclude: ‘’the evidence indicates that on average the market’s judgment concerning the information implications of a split are fully reflected in the price of a share at least by the end of the split month but most probably almost immediately after the announcement date.’’ (p. 20)
When using annual income number announcement dates in relation to the movement of stock prices, Ball and Brown (1968) found that for 261 companies listed on the NYSE in the period 1946 to 1966, abnormal returns considerably move during the period before the announcement. The movement is upwards for ‘good news’ announcements and downwards for ‘bad news’. Their explanation for this is timeliness. Annual income numbers are given in annual reports, but the content within these reports are by then already addressed by more prompt sources of media. This already known information accounts for 85 to 90 percent of the annual report.
Strong-‐form efficiency, as the name already suggests, is the strictest view on efficiency. It builds on the two former forms of efficiency and adds that in addition to public information, information only available to insiders, i.e. private information, is reflected in stock prices (Bodie, Kane & Marcus, 2011).
Beck-‐Dudley and Stephens (1989) found that if an individual investor is aware of information before it is publicly announced, the investor could make excess returns. However, there is a limited timeframe for this strategy. This study was based on Wall Street Journal columnist R. Foster Winans, who wrote the ‘Heard’ column, basically a gossip column concerning stocks traded on Wall Street. He would then, between October 1983 and February 1984, give the contents of his column to stockbroker Brent before it was published, so Brent could profit. With the use of an event window around the publication date and calculating the expected return, real returns and the difference in
those returns for the window (the abnormal return), Beck-‐Dudley and Stephens found that one could only achieve the excess returns if information is available to the
individual four to five days prior to announcement. When one would buy a stock the day before announcement and sell in the one to two days following the purchase, one would not profit as much. This is because the information has now already reached the other investors. Beck-‐Dudley and Stephens state that the possession of specific private information can result in profits for an insider or specialist.
2.5 Criticism regarding the Efficient Market Hypothesis
In the early 2000’s the Efficient Market Hypothesis started becoming less widely accepted. The belief that stock prices are partially predictable was gaining popularity among statisticians and economists (Malkiel, 2003). Some of the anomalies found in EMH will be discussed in this section.
Lo and MacKinley (1999) used 1216 weekly observations retrieved from the CRSP database for a period over twenty years and found significant positive serial correlation for holding period return. This study used returns instead of excess returns, but argues that the results would not differ had they used excess returns. However, the fact that stock prices do not follow a random walk does not mean they reject the Efficient Market Hypothesis. There is momentum in the stock prices, meaning that good or bad recent performance of a stock continues over time (Bodie, Kane & Marcus, 2011). However, Lo and MacKinlay (1999) discuss that more explicit models of price-‐generating
mechanisms are needed in order to conclude that if prices do not follow a random walk, the market is inefficient.
Behaviorists attribute momentum to underreaction. When investors underreact to an announcement, it leads to positive serial correlation. The total impact of an
announcement takes time to be reflected in the stock price, due to investors who underreact initially (Malkiel, 2003).
Malkiel (2003) does not diminish the existence of momentum in stock prices, but does claim that any pattern of this kind cannot create an investment strategy yielding excess returns. Buying stock with positive serial correlation most certainly can generate relative positive returns, but can also generate relative negative returns, which was the case for the period during the late 1990s to 2000.
Another anomaly (and behavioral critique) is found in the long-‐run return reversals of stock prices. When investors are overconfident in their ability to predict stock prices with the use of valuation parameters, a relative large weight is given to a low probability of a return (Kahneman & Tversky, 1979). De Bondt and Thaler (1990) found that when analyzing the forecasts of analysts on the NYSE between 1976 and 1984 and comparing to stock returns, predicted changes were more volatile than actual changes. They found this to be consistent with overreaction; this overreaction will then reverse itself as prices reach their fundamental values again.
With this pattern, a strategy could be implemented that would involve buying stocks with unfavorable forecasts, and sell when the reversal has happened, also known as a contrarian strategy (Malkiel, 2003).
However, this does not necessarily mean markets are inefficient. Instead of attributing the reversal to a behavioral bias, Malkiel (2003) argues that the reversal might come from the volatility and mean-‐reverting characteristics of interest rates. Increasing interest rates decrease bond prices, which then in turn increases stock prices. The same principle holds for decreasing interest rates.
The implied profit one can make when following the strategy mentioned above is also not a given. Fluck, Malkiel and Quandt (1997) could not reject that excess returns can be made from following a contrarian strategy. They took a sample of 1000 firms and
ranked their Price/Earnings ratio (P/E ratio) and Price/Book ratio from low to high. The firms with low ratios, and thus out of favor, did on average yield higher excess returns. The time period assessed was 10 years, from 1979 until 1988. They also carried out their analysis in an out-‐of-‐sample timeframe, adding six more years. In this timeframe, the high returns still prevailed. Therefore, it is not clear whether the positive abnormal result of a contrarian strategy is attributable to picking the unfavorable stocks or to an inappropriate risk measure.
Dividend yields also seemingly have some predictive power with respect to future returns. When comparing dividend yields of the S&P 500 index to the subsequent ten-‐ year return, for a period from 1926 to 2001, low dividend yield on the index resulted in a low return on the index, whereas a high dividend yield resulted in a high return (Malkiel, 2003).
Yet, Malkiel (2003) again refutes this seemingly predictable pattern and thus
inefficiency by relating the movement of stock prices to interest rates. With high interest rates, stock prices tend to be high and dividend yield is lower. When the interest then decreases, the stock prices inevitably also decrease.
Malkiel (2003) states: ‘’ Consequently, the ability of initial yields to predict returns may simply reflect the adjustment of the stock market to general economic conditions’’ (p. 7)
The implication that firms with high P/E ratios yield low returns and low P/E ratios yield high returns, based on the same sample as above, would also imply some type of predictability regarding valuation parameters. However, when taking for example June 1987 where the P/E ratio was relatively high and the dividend yield relatively low, the total return on the S&P 500 was a lot higher than compared to the corresponding values obtained by the study of the index. Given this difference in results, Malkiel suggests that one must be very careful in using these ratios as a prediction for future returns (Malkiel, 2003).
Fama and French (1993) find evidence that small firms have higher average returns than large firms and that the higher the book-‐to-‐market equity ratio a firm has, the higher average returns will be. They ranked each firm in the NYSE into quintiles and used 25 stock portfolios based on size and book-‐to-‐market equity ratio. The premium for size increases excess returns 0,46% per month. The premium for the book-‐to-‐market ratio yields 0,40% per month.
When using the factors excess market return, Small minus Big (SMB) relating to firm size and High minus Low (HML) relating to the book-‐to-‐market equity in a regression on excess returns, they found both the beta on SMB and the beta on HML explain the
variation in excess return. This model is also known as the three-‐factor model. Both SMB and HML serve as risk premiums, as they are not dependent on the market beta. The authors argue that SMB and HML are risk factors that are not captured by the Capital Asset Pricing Model, instead of it being signs of inefficiency (Fama & French, 1993).
Behavioral financial economists also argue that the existence of market crashes is a sign of inefficiency. With fundamental elements of valuation not changing in the time of the crash of 1987, they believe this crash can solely be explained by psychological factors,
implying there is no rational explanation for the sharp decrease in stock values (Malkiel, 2003).
However, Malkiel (2003) suggests a few rational factors that could have led to those decreases. Long-‐term Treasury bond yield increased by 1,5% prior to the crisis and risk perceptions also rose. Rate of return on a stock consists of the yield on the Treasury bond plus a risk premium, and with the two latter increasing, rate of return on a stock must also increase. If the expected growth rate of the stock and cash dividends remains the same, this higher rate of return can only be achieved by a declining price.
The occurrence of ‘bubbles’, prices rising above intrinsic value and continuing to rise (Bodie, Kane & Marcus, 2011), is seen by behaviorists as another sign of prices not being rational. In hindsight, it is clear when a bubble occurred, but in the period itself not so much. Investors may have believed they have acted rationally. In the case of the Dotcom bubble, projections on growth of the Internet, and companies affiliated with this, were unsustainable. But at the time being, these forecasts were not seen as extreme (Malkiel, 2003).
There was no clear arbitrage strategy during this bubble; even if one were to disagree with the forecasts, one could not be entirely sure, since all signs did point to growth. (Malkiel, 2003). And even if one were to act upon ones contrarian beliefs, there would still be costs of short selling, difficulty obtaining stocks to sell short and the possibility that you are right, but the adjustment to correct prices happens outside your time window. Markets can stay irrational longer than you can stay solvent (Bodie, Kane & Marcus, 2011).
2.6 The implications of efficiency in emerging economies
An efficient market contributes to efficient resource allocation. Investments in assets such as know-‐how, plant and equipment move with prices of financial assets. If stocks were mispriced, and thus inefficient, resources would be systematically misallocated. On the one hand, overinvestment in companies with overpriced securities can create bubbles, as the price deviates from its fundamental value. On the other hand, companies with underpriced securities cannot invest as much as wanted, because cost of raising capital is too high (Bodie, Kane & Marcus, 2011).
When the cost of obtaining information about companies is too high, capital may not flow to its highest value use (Levine, 1997). Hence, resources are not allocated efficient and could slow down growth (Huang, Khurana & Pereira, 2009).
Urrutia (1995) notes on the emerging markets of Argentina, Brazil, Chile and Mexico: ‘‘these emerging markets are potentially important contributors to the growth and development of the economies of their countries’’ (p. 300).
2.7 Evidence on efficient capital markets in emerging economies
Magnusson and Wysick (2002) researched weak-‐form efficiency in Africa’s capital markets. Monthly data for eight African emerging markets was used and compared to stock markets in developed and other emerging economies: the US, Asia and Latin America. Data series prior to 1998 were used, with the observations depending on the availability of the data, so timeframes vary per market.
Three forms of a random walk were distinguished. RW3 states that prices are
uncorrelated. RW2 adds to RW3 by stating that price changes are independent and non-‐ identically distributed; past volatility does not predict future volatility. RW1 is again an addition to RW2, where prices are independent and identically distributed; past prices predict neither future prices, nor future volatility (Magnusson & Wysick, 2002).
Cote d’Ivoire, Botswana, Kenya, Mauritius, Nigeria and South Africa do conform to RW3, but not RW2 and RW1. Zimbabwe and Ghana conform to none of the three forms of a random walk. When compared to the market in the US, African markets do not match the level of efficiency in the US, whose market conform the RW1. Africa compares favorably to Latin America and Asia in terms of efficiency. Several of these African and Latin American countries, however, only are efficient in the RW3 form when returns are measured in US Dollar, but are not weak-‐form efficient when returns are measured in home currency, implying that it are the international investors creating efficiency (Magnusson & Wysick, 2002).
Urrutia (1995) studies weak-‐form efficiency in Argentina, Brazil, Chile and Mexico, as these emerging equity markets can have an important supporting role to economic growth and development of said countries. For these four countries, monthly stock index prices are used in a sample from December 1975 to March 1991. In order to test for a random walk in prices, Urrutia uses a variance-‐ratio method of Lo and MacKinlay.
If a time series is a pure random walk, the variance of returns grows proportionally with the holding period. He rejects the random walk for all four countries, as the variance does not grow proportionally and variances larger than one are a sign of positive return autocorrelation. When performing a runs test, a test for independence of returns, which does not rely on the assumption returns are normally distributed, independence is not rejected for all four countries. Hence, Argentina, Brazil, Chile and Mexico are weak-‐form efficient. The existence of autocorrelation can just be an indicator of economic growth, instead of being a sign of inefficiency in these equity markets.
In addition to the findings above, Claessens, Dasgupta and Glen (1995) studied return predictability in twenty emerging markets, with a few of these markets also included in the studies mentioned above. The countries studied are: Argentina, Brazil, Chile,
Colombia, Greece, India, Indonesia, Jordan, Republic of Korea, Malaysia, Mexico, Nigeria, Pakistan, Philippines, Portugal, Taiwan, Thailand, Turkey, Venezuela and Zimbabwe. They also used the variance-‐ratio method for monthly data of returns of twenty
countries in the Emerging Market Data Base, with a sample period varying per market, but all time series end in 1992. They find high returns and standard deviations,
generally higher than in industrial economies. Significant positive autocorrelation is found for nine economies (Chile, Colombia, Greece, Mexico, Pakistan, the Philippines, Portugal, Turkey and Venezuela) and compared to industrial countries; seven emerging markets have higher autocorrelation. There is a significant high degree of predictability of returns, but the authors are not able to attribute this to those markets being
inefficient, as it can also be the result of varying risk premiums, large structural changes in the markets or regime switching effects, where the ex post returns are higher than ex ante returns.
2.8 Efficient capital markets in BRIC countries
Brazil, Russia, India and China, a group known as BRIC countries, are the four largest emerging economies. The share of these countries in world output was 46%, lower than in previous periods, where it has exceeded 50%. However, the BRIC countries did rank in the world’s top ten economies of 2008 (Gay, 2008).
Gay (2008) studied whether macroeconomic factors like the foreign exchange rate and oil price explain stock market returns between 1999 and 2006 for the four BRIC
countries. He did not find a significant relationship between those two variables and stock market returns. Moreover, when performing a Durbin-‐Watson test, the
autocorrelation was not present for Brazil, Russia and China at the five percent
significance level. When the case of India was further tested for autocorrelation using a Q-‐test, autocorrelation turned out not be significant for India either. With no
relationship between present and past returns, Gay suggests the BRIC countries are showing signs of weak-‐form efficiency.
Interestingly, different results were found by Majumder (2012) when comparing BRIC markets with the US market. Majumder utilizes the Hurst exponent to test for
dependence of a time series. Any value different from 0,5 indicates dependence in returns. Three time series for the five countries were tested for independence. The first time series covers the entire sample period 2001 to 2011, the other two are sub sets of the first; pre crisis time series ending in 2007 and during-‐and-‐post crisis time series starting in 2007. He found that for the total sample all five markets exhibited
inefficiency. India and China were relatively efficient pre crisis, but not in the second period. Russia’s market was inefficient in all three time series. Brazil’s market exhibited inefficiency pre crisis, but became relatively efficient during and post crisis. This shows that the label ‘efficient’ very much depends on the time period assessed.
3. Data
An event study will be performed in order to analyze the semi-‐strong form efficiency of the Brazilian capital market. To create uniformity among results, stock splits are chosen as the ‘event’. For the companies included, these splits are mostly 2-‐for-‐1. In the total sample of companies split factor ranges from 5-‐for-‐4 to 39-‐for-‐1.
Selection of stocks
To capture the broad range of the market, this study will focus on companies by industry classification. From the Brazil 50 index (IBrX 50), companies from each industry are chosen, if they meet further criteria. In total there are 25 industries for the São Paolo stock exchange, also known as BM&F Bovespa.
Criteria for inclusion in the event study: 1. A stock split has occurred.
2. Only one stock split is present in the estimation window and event window combined. These windows will be further discussed in the section Methodology. The occurrence of a single split in these windows simplifies the analysis and limits the bias in the results, as it is now the only event.
3. No earnings announcements have been made during both the estimation and event window. Again, this ensures limited bias, as announcements can be seen as other types of events.
4. Availability of data. Prices and dates need to be available.
For the 25 industries and 25 related companies, 18 companies meet the criteria
mentioned above. This leaves a sample of stock prices for those 18 companies, which all have one single stock split in the sample period.
For companies that had more than stock split that met the criteria, the most recent one is chosen to perform an event study with. This keeps the time frame relatively small and more focused on efficiency in recent history. Overall, the data ranges from 2000 to 2011. The event dates were found on the website of the stock exchange, as the date on which the split was agreed upon2. Earnings announcements and announcement dates for stock splits were crosschecked using Reuters3. Stock prices were retrieved from Datastream.
2 Source: www.bmfbovespa.com
An overview of the included companies (sorted on Ticker): 3 Source: www.reuters.com
Industry Company Ticker Event Date
Consumer Non-‐Cyclical / Food Processors
BRF BRFS3 31/03/2010
Basic Materials / Chemicals Braskem BRKM5 20/10/2003
Financial / Real Estate BR Malls BRML3 23/09/2010
Utilities / Electric Utilities Cemig CMIG4 26/04/2007
Construction and
Transportation / Engineering
Cyrela Brazil Realty CYRE3 07/12/2006
Basic Materials / Steel and Metallurgy
Gerdau GGBR4 30/05/2008
Consumer Cyclical / Textiles, Apparel and Footwear
Cia Hering HGTX3 29/10/2010 Consumer Non-‐Cyclical / Diversified Hypermarcas HYPE3 30/12/2009 Financial / Financial Intermediaries
Itau Unibanco ITUB4 27/08/2007
Consumer Cyclical / Retail Lojas Renner LREN3 03/10/2006
Consumer Non-‐Cyclical / Cleaning Products
Natura Cosmeticos NATU3 29/03/2006
Telecommunications / Fixed Line Communications Oi OIBR4 12/09/2000 Consumer Non-‐Cyclical / Retail Distribution Companhia Brasileira de Distribuicao PCAR4 30/07/2007
Oil ,Gas and Biofuels Petrobras PETR4 24/03/2008
Diversified Localiza Rent a Car RENT3 24/04/2007
Utilities / Water Utilities SABESP SBSP3 30/04/2007
Financial / Holdings -‐ Diversified
Ultrapar UGPA3 10/02/2011
Selection of market index
For the market index, the Bovespa Index is used. It is a value-‐weighted index, currently consisting of 72 companies. It is a total return index, existing for 42 years. This ensures sufficient data for the time period assessed is available. Index values were retrieved from Datastream.
Selection of the risk free rate
The SELIC rate is used as the risk free rate. This is the overnight interbank exchange rate. Datastream recommends SELIC, as the Brazilian Government Bonds are not risk free4. The rate was retrieved from Datastream.
4 Source: extranet.datastream.com
4. Methodology
An event study will be performed to test for semi-‐strong form efficiency. This event study focuses on abnormal returns. The methodology follows the work of Kothari and Warner (2007). Abnormal returns are obtained by subtracting predicted excess returns from actual excess returns. Predicted excess returns are obtained by the use of the Capital Asset Pricing Model (CAPM). Estimated returns for the event window are based upon an estimation window.
First, the estimation window and event window are established. Both window sizes are chosen a bit arbitrarily, since there is no particular standard for window size in the literature on event studies. For the estimation window, 120 trading days are chosen. This number of trading days is suggested by MacKinlay (1997).
The event window is set at 41 days, 20 days prior to the event and 20 days after. The following figure shows the time span, with day 0 being the announcement of the stock split.
Estimation window Event window
-‐140 -‐20 0 +20
Then, a pricing model is established. For this study, the CAPM is used. Its
implementation is very straightforward and simple, but does have some flaws (which will be discussed in the Limitations section). In addition to its simplicity, CAPM is chosen as a benchmark because it is a widely accepted pricing model around the world. It is the best method available to decompose risk into systematic and firm-‐specific risk. In addition to that, there is evidence that the central theory revolving around CAPM, that the market portfolio is efficient, is not far from being valid (Bodie, Kane & Marcus, 2011).
Returns are generated with the following formula:
𝑟!" = 𝑃!,!− 𝑃!,!!! 𝑃!,!!!
Where:
ri,t = return on stock at time t, for company i
Pi,t = adjusted price of stock at time t, for company i
Pi, t-‐1 = price of stock at time t-‐1, for company i
Prices are not adjusted for dividends and/or other announcements. The assumption is made that dividends are already included in the adjusted price. Other announcements during the time period assessed are checked for and not present.
CAPM is stated by the following formula:
𝑅!,! = 𝛼 + 𝛽 𝑟!,!− 𝑟!,! + 𝜖!,! 𝑅!,! = 𝑟!,! − 𝑟!,!
Where:
ri,t = return on stock at time t, for company i
rf,t = risk free rate (SELIC rate) at time t rm,t = return on market index at time t εi,t= error term at time t, for company i
The alpha and beta are found by performing an Ordinary Least Squares regression on the excess return of company i with observations that fall within the estimation window, between t=-‐21 and t=-‐140.
The estimated alpha and beta are then used to predict values for excess return on stock of company i for the estimation window, between t=-‐20 and t=20.
These predicted excess returns are now subtracted from actual excess returns and this gives the abnormal returns, during the event window.
𝐴𝑅!,! = 𝑅!,!− 𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑𝑅!,!
Per company, the cumulative abnormal return (CAR) on day +20 is calculated and a T-‐ test is performed, to test whether the average abnormal return is significantly different from zero.
𝐶𝐴𝑅(𝑡!, 𝑡!)! = AR!,! !! !!!! Where: t1=-‐20 t2=20
The T-‐statistic is obtained using the following formula: 𝐶𝐴𝑅 𝑡!, 𝑡!
𝐿 ∗ 𝜎 𝐴𝑅 Where:
L = t2 – t1 + 1 L=41 for event window
For the standard deviation of abnormal returns, the observations in the event window are used.
In addition to this, the average abnormal return is also tested for significance for two other time periods. The event window of 41 days is a rather large window; therefore smaller windows of 11 days (5 prior to event, 5 after) and 5 days (2 prior to event, 2 after) are tested for significant abnormal return as well. The same formulas as above apply.
For each company individually, the absolute value of the T-‐statistic will be compared to critical values at different significance levels. Under the H0 hypothesis, that markets are efficient, the average abnormal excess returns should not be different from zero and neither should CAR. Otherwise, significant abnormal excess returns during the event window can be made, which indicates inefficiency.
More formally stated:
𝐻!: 𝐶𝐴𝑅 = 0, 𝑚𝑎𝑟𝑘𝑒𝑡 𝑖𝑠 𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝐻!: 𝐶𝐴𝑅 ≠ 0, 𝑚𝑎𝑟𝑘𝑒𝑡 𝑖𝑠 𝑖𝑛𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡
For the aggregate of companies the CAR is also tested, by performing a regression on only a constant and checking for a value that is significantly different from zero for this constant.