Stock market analysts’ recommendations: Difference in the Dutch stock market abnormal return between the small cap, mid cap and large cap index.
Faculty of Economics and Business Administration Finance Department Master thesis in Finance Student Name:
R.H. Knippenburg
Supervisor:
Drs. J. Grazell
Administration number:
698699
Defense Date:
27 September 2012
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!!
Stock market analysts’ recommendations: Difference in the Dutch stock market abnormal return between the small cap, mid cap en large cap index.
Abstract: This master thesis investigates the recommendation revisions of analysts in the Dutch stock market. Empirical research shows that companies listed in the small-capitalized index (AScX) have a greater price reaction and volume increase after a revision than companies in the mid-cap index (AMX) and large-cap index (AEX). Similar, companies listed in the AMX yield a greater cumulative abnormal return and volume increase than companies in the AEX. In the three day period around the event date, upgraded recommendations for AEX, AMX and AScX yield on average a cumulative abnormal return of respectively 0.96%, 1.29% and 1.35% while it yields a cumulative abnormal return for downgraded recommendations of -1.11%, -1.66% and -2.22%. The difference in returns can be attributed to the information-asymmetry and the differences in risk between small and large companies. Keywords: Analyst, recommendations, revisions, upgrades, downgrades, AEX, AMX, AScX, small firms, large firms, size effect, abnormal returns, information asymmetry, excess returns, stock markets
!
#!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table of contents 1. Introduction
3.
2. Literature review
5.
2.1 The Efficient Market Hypothesis
5.
2.2 Methodologies to issue recommendations
8.
2.3 Independence of analysts
10.
2.4 Herding behavior of analysts
14.
2.5 Analysts and stock-market bubbles
16.
2.6 The evolution of the post-recommendation drift literature
17.
2.7 Difference between small and large companies
23.
3. Empirical research
26.
3.1 Data
26.
3.2 Methodology
28.
3.3 Hypotheses
32.
4. Research results
37.
4.1 Descriptive statistics
37.
4.2 Testing the hypotheses
38.
4.3 Robustness test
44.
4.4 Additional test
45.
5. Summary and main conclusions
47.
6. References
50.
Appendix: Figures and tables
!
$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 1. Introduction The existing literature about the impact of analyst recommendations on stock markets is extensive. The research of Cowles (1933) was one of the first in this area, and it described that investors are not able to add value to the market when they follow the recommendations issued by analysts. These conclusions of Cowles, followed later by Colker (1963), Diefenbach (1972), Logue and Tuttle (1973), Bidwell (1977) and Groth, Lease, Lewellen and Schlarbaum (1979) confirmed the existence of the semistrong and strong form of the efficient market hypothesis of Fama (1970). This line of literature clashes with the fact that research departments of brokerage firms and investment banks spend large amounts of money on security analysis, presumably because these firms and their clients believe that it generates positive returns. This appearance is confirmed by a vast amount of empirical researches, started with the findings of Davies and Canes (1978) among others and which is still justified nowadays. Investigating the recommendations of analysts offers an opportunity to study analyst judgement and preferences across large samples of stocks. The majority of the empirical research finds that on average markets react favourable on upgrades of recommendations and react negatively to downgrades of recommendations. These empirical researches are presumably based on US data, while the data aimed on Europe is relatively scarce. Jegadeesh and Kim (2006) researched the Group of Seven (G7) countries, including some European countries and Azzi, Bird, Griringhelli and Rossi (2004) investigated analyst recommendations in fifteen European Countries, including the Netherlands. To append the empirical researches in Europe, it has been determined to examine the Dutch stock market and in particular the differences between the AEX, AMX and the AScX indices which contains respectively the small, middle and large capitalized firms. Examinations of Stickel (1995), Womack (1996) and Jegadeesh and Kim (2006) showed that the market reaction of small-capitalized firms is significantly larger than large-capitalized firms.
!
%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! To see whether this result also holds for the Dutch stock market, the following research question will be investigated: “Are analysts’ recommendation revisions able to 1) add more value to the Dutch small cap index than the mid-cap or large-cap indices and to 2) add more value to the mid-cap index compared to the large-cap index?” The results of this empirical research of this master thesis showed that in the three day period around the event date, buy recommendations for AEX, AMX and AScX yield on average a cumulative abnormal return of respectively 0.96%, 1.29% and 1.35% while it yields a cumulative abnormal return for sell recommendations of -1.11%, -1.66% and-2.22%. The difference in returns can be partially attributed to the information-asymmetry between small and large companies. On average, there is less information available for small companies than large companies. Therefore, one can expect that a recommendation revision for a small stock yields a larger cumulative abnormal return than a revision for large companies, where already lots of information are available. Furthermore, the differences can be attributed to the differences in risk between small and large companies. Small companies are generally young, volatile, unprofitable, non-dividend paying, growing and financially distressed firms. Therefore, an investment in these particular firms should give a larger return than large companies since there is a higher risk involved in these investments. The abnormal return of companies was being corrected for this risk, as was being measured by the beta of the companies. The remainder of this master thesis is organized as follows. Chapter two describes the current state of the existing literature about analyst recommendations. Literature that has common themes with the literature of analysts’ recommendations, such as the earnings estimation literature, will be discussed, though not in depth. Chapter three presents the data, the research methodology and the hypotheses. The fourth chapter evaluates the results of the research and chapter five summarizes the findings and discusses some of their implications.
!
&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 2. Literature review Financial analysts are concerned with providing comprehensive understandings of investment opportunities. Financial analysts gather information by studying public records and fillings of companies, as well as by participating in public conference calls where they can directly ask their questions to the management of the target company. Most analysts focus on a specific industry and therefore they typically cover between the five and twenty-five stocks. Financial analysts are often employed by mutual and pension funds, hedge funds, securities firms, banks, investment banks and insurance companies. They help their companies or their clients to make investment decisions. Financial analysts values a company by studying the entire industries in which the company is involved, but also its current trends in business practices, products, industry competition and regulation or policies that may affect the profitability of the industry. There are three main types of analysts: buy-side analysts, sell-side analysts and unaffiliated analysts. Buy-side analysts work for institutional money firms like mutual funds, pension funds, trusts and hedge funds. These types of analysts are concerned with buying assets or securities for their own accounts. Sell-side analysts works for brokerage firms that manage individual accounts. They evaluate companies for future growth and other investment criteria and they make recommendations on the securities they cover. Sell-side analysts often place recommendations on stocks and other securities to the clients of the firm, typically phrased as buy, sell, hold, outperform, underperform, etc. These recommendations help clients to make decisions to buy or sell stocks and securities, which is beneficially for the brokerage firm because they receive a commission on each transaction the clients makes. Unaffiliated analysts are not associated with institutional money firms or brokerage firms. Independent analysts often sell their recommendation reports to clients on a subscription or other basis. 2.1 The Efficient Market Hypothesis Analysts exist to promote market efficiency and to help investors to value companies more accurately. As Grossman (1975, 1977) showed in past research, there is an incentive to create a market whenever there are differences in beliefs that are not !
'!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! completely arbitraged. Since the market of analysts’ recommendations clearly exists of differences in beliefs, there is an incentive to create a market for the recommendations of analysts. This way of thinking is not in line with the alternative theory of expectations developed by Muth (1961). According to this view, analysts have rational expectations, which means that their expectations will be identical to optimal forecast using all available information. Even though a rational expectation equals the optimal forecast using all available information, a prediction based on it may not always be perfectly accurate. There are two reasons why expectations of analysts may fail to perfectly accurate: 1) Analysts might be aware of all available information, but find it takes too much effort to make their expectation the best guess possible; 2) Analysts might be unaware of some available relevant information, so their best guess of the future will not be accurate. This rationale is also applicable for the financial market. This theory is called the Efficient Market Hypothesis (EMH) and was developed by Fama at the beginning of the sixties. The theory states that the current prices in a financial market will be set in such a way that the optimal forecast of a security’s return using all available information, equals the security’s equilibrium return. Or stated in a more simple way: in an efficient market, a security’s price fully reflects all available information. This implies that there are no arbitrage and unexploited profit opportunities, because these are all eliminated by the market. The EMH suggests that if information is already publicly available, a positive or negative announcement about a company or an analyst’s recommendation will not, on average, affect the price of its stock, since the information is already reflected in the stock price. However, the work of analysts does promote market efficiency since it helps investors to value companies’ assets more accurately. Because empirical researches have found consistently problems with the EMH, Fama (1970) developed three forms of financially market efficiency. In the weak form efficiency future changes cannot be predicted by analyzing prices from the past. Therefore, stock prices follow a random walk, which means that future changes in stock prices are unpredictable. This offers the opportunity to generate excess !
(!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! returns by doing fundamental analysis. The semi-strong form efficiency implies that share prices adjust to publicly available new information rapidly. This means that no excess returns can be earned by trading on that information. The strong form efficiency implies that every individual investor have monopolistic access to any information relevant for price formation. Share prices reflect all public and private information and therefore no one can earn excess returns. The semi-strong or strong form of the efficient market hypothesis is confirmed in the research of Cowles (1933), whose research was one of the first empirical researches that investigated the post-recommendation drift of analysts’ recommendations. Cowles concluded that the recommendations given by analysts do not create value for the investors. These conclusions are supported by the findings of Diefenbach (1972) and Logue and Tuttle (1973). These researches supported the view that the market is perfectly efficient and that analysts are not able to add value with their recommendations since any information they have would already be reflected in the market price. On the other hand, there is a vast amount of literature that rejects the semi-strong and strong form of the efficient market hypothesis. This line of though suggests that the recommendations of analysts do influence the stock market return and gives analysts reasons for existence. Grossman and Stiglitz (1980) showed in a mathematically way that the price system does not reveal the information about the true value of the risky asset. They argued that information is costly and therefore prices cannot perfectly reflect all available information. If prices would reflect all information available, analyst who spent resources (time and money) to obtain information would receive no compensation. The only way for analysts to receive compensation for their activity of gathering information is to use information to take positions in the market that are better than the positions of uninformed traders. According to the supportive rationale of Michaely and Womack (2005), the marginal costs of gathering and obtaining information should be equal to the marginal costs, which are the rents analysts receive. These arguments offers reasons to believe that analysts’ efforts to security analysis indeed leads to positive returns on the stocks they analyze and is supported by a vast amount of literature. The research of Davies and Canes (1978) was one of the first researches that proves that analyst recommendation do add value. Their research !
)!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! investigated buy and sell recommendations, based on the Wall Street Journal column ‘Heard on the Street’, in 1970 and 1971. They show that there is an abnormal price movement on the day after the day of publication of buy and sell recommendations of analysts. Black (1973), Groth, Lewellen, Schlarbaum and Ronald (1979), Givoly and Lakonishok (1979), Copeland and Mayers (1981), Dimson and Marsh (1984), Liu, Smith and Syed (1990) among many more researchers support the findings of Davies and Canes. The findings of these researches will be expounded later on. 2.2 Methodologies to issue recommendations In the 1980s and earlier, recommendations were private or quasi-private information and not easily available for private investors who were not customer of a brokerage firm. However, in the information age nowadays, it has become easy for investors to access large amounts of information and recommendations about firms. Analysts use valuation models to transform this readily available information into earnings forecasts. These earnings forecasts are compared to the stock prices, which is followed by a typical recommendation, for instance, sell, hold or buy. There is a wide range of valuation models available for the analysts, varying from simple valuation models, such as the Price-Earnings Growth (PEG) models, to more sophisticating models like the abnormal earnings growth model and the residual income model. One would expect that an analyst would use more sophisticated valuation models, since simple valuation models are faced by major drawbacks. Discounted cash flow models are impractical since many firms pay zero dividends and the main disadvantage of the discounted free cash flow valuation method is that it does not always measure value added over a period correctly, because if a firm invests more cash in operation than it takes out from operations, the model perceive this as value destroying. Investment is treated as a ‘bad’ rather than a ‘good’. In contrast to these shortcomings, empirical research shows that analysts base their valuations and recommendations on stocks on simple valuation methods. Bradshaw (2004) shows that the consensus recommendations of sell-side analysts are not consistent with the firm value that is derived when sophisticated valuation models are used, such as the residual income model and the discounted cash flow model. In fact, the recommendations are more consistent with heuristic valuation models, such as the !
*!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! PEG models and long-term growth models. Analysts often use disclosed target prices in their recommendations, to empower their recommendation report. For example, broker A issued a buy recommendation for company X on the second of July 2012, and its target price is ! 33. A research of Bradshaw (2002) showed that analysts are less likely to disclose target prices when they are more uncertain about the forecasted earnings for the company. The research stated that PEG valuations are significantly related to the analysts disclosed target prices and stock recommendations. However, when the PEG valuation method is used as a proxy for analysts’ undisclosed prices, Bradshaw did not find similar support for the recommendations in these reports. These findings suggest that the private information used to issue a recommendation by analysts, differs from the true recommendations of a residual income model. Analysts may fail to do so, for several reasons. First, valuing companies is a complicated task and some analysts do not have the ability and experience to efficiently transform earnings forecasts into recommendations. It is shown in a research from Ke and Yu (2007) that analysts who have less firm-experience, the transformational efficiency is lower. Secondly, as Francis and Philback (1993) suggest, an analyst issue more optimistic recommendations, which is different than the true valuation in order to ‘curry favor with management’. According to Francis and Philback, unfavorable stock recommendations result in greater deterioration of the analyst’s relation with management. They postulate that analysts’ earnings estimations are optimistic, on average, and are more optimistic for sell and hold recommendations than buy recommendations. This asymmetric use of target prices is an incentive for analysts to justify their recommendations. The third reason is in line with Lin and McNichols (1998), who depict that a different valuation can be issued to enhance investment-banking relations. Finally, according to Graham (1999) analysts can also herd with other analysts. The last two reasons will be appended more in the following paragraphs.
!
+!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 2.3 Independence of analysts Large groups of investors follow the recommendations issued by analyst. Principally, investors do this based on two main reasons. In the first plays, analysts base their recommendations on important sources of information, which often are not accessible by private investors. Analysts have direct links with the companies they follow and the recommendations based on this information offer valuable insights on companies and industries. By issuing recommendations, analysts transfer the information to investors which improve the efficiency of the financial market since individual investor can actively take part in the financially system. Behavioral finance offers an additional view about why investors would follow the recommendations of analysts. As Shefrin (2002) concluded, investors have a great aversion to regret. Shefrin described regret as an emotional feeling, experienced after the awareness that a different decision in the past would have been better than the decision chosen. With this knowledge, investors are willing to follow the recommendations of analysts, because if the stock price of a company follows the predicted target price, the decision to not follow the recommendation of the analyst would cause regret. However, analysts, the analysts’ firms, their clients and their clients companies face conflicts of interests and competing pressures. Companies do not want to fall short of the analysts’ forecasts. Doing so, will cause an earnings disappointment that often is interpreted by the market as bad news and goes along with a plummeting stock price. This will not only lower the attention of the investors, but it will also lower the managers’ compensation since it is linked to the evolution of the stock price. The company will try to manipulate the earnings, which can be done in several ways. First of all, they can manipulate earnings through accruals as is showed by Chan, Chan, Jegadeesh and Lakonishok (2006). Other ways to avoid disappointments is to manipulate forecasts through overstating the long-term growth rate in earnings or price targets or to adjust forecast earnings to ensure that the actual earnings will not come up short on the forecasted earnings. Chan, Karceski and Lakonishok (2007) found evidence that firms are more likely to be associated with non-negative surprises. Companies that avoid negative surprises are rewarded with a greater stock price increase compared to companies that do have negative earnings surprises. A last way !
#,!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! to avoid disappointments is to lobby with analysts, in order to receive buy recommendations of analysts. Moreover, analysts of sell-side firms are benefited when their clients are successful over time. Giving advice to clients is the broker’s main service and if they succeed to fulfill their core activities it improves the reputation and success of the brokerage firm. There are several factors that create pressure on an analyst’s independence and objectivity. The commissions of the brokerage firms are a first factor that creates pressure on the analyst’s independence and objectivity. Analysts of brokerage firms usually do not charge for their recommendation reports, however, by issuing favorable recommendations on stocks, investors can be tempted to trade more which generates more brokerage commissions to the brokerage firm. Another cause for a conflict of interest comes from the compensation and bonuses of analysts. The compensation and bonuses are often linked to the profitability of the company or on the number of investment banking trades. This can result in more favorable reports from the analyst, in order to generate more brokerage commission that results in a higher profit or to attract more investment banking deals. Ownership in a particular firm by an analyst, or the company of the analyst, can result in a biased recommendation report. Also indirect ownership, like employee stockpurchase pools can bias the view of the analyst. A growing trend in the investmentbanking environment is called ‘venture investing’, in where the firm of the analyst, or the analyst himself, acquires a stake in the start-up of the company. Through these practices, analysts are more reluctant to issue less favorable recommendations. At last, there are conflicts of interest within the investment banking relationships. When a company issues new securities, they hire investment bankers to ask advice, to construct the deal and help with the actual offering. These activities are very profitable for an investment bank and sometime even more profitable than the brokerage operations or research reports. Analysts are an important part of the investment banking team by researching the valuation of the company. By issuing a positive recommendation of the company, just after the initial published offering took !
##!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! place, it could lead to a higher demand of the companies stocks. This in turn leads to a higher possibility that the company issues more or new securities, which would lead to a higher profit of the investment bank. Another reason why analysts issue favorable recommendation to boost the profit of the investment part of the bank comes forth from that client companies prefer positive research reports. As Pratt (1993) described, issuing a sell recommendation comes along with substantial costs in the investment community because sell recommendations can harm a brokerage firm’s present and potential investment banking relationships. An unfavorable research report can result in a lower stock price of the customers company, which is disadvantageous since it lowers the amount of money it can collect from future seasoned equity offerings, it lowers the investors attractiveness and it increases the change to be taken over. Hence analyst are discouraged by investment bankers to issue sell recommendations. The preference of positive recommendation is confirmed in several researches. The research of Hong and Kubik (2003) showed that analysts who issue a large fraction of positive recommendations are more likely to go up in the brokerage house hierarchy. Pratt (1993) further described that top management and investment contacts may limit or cut off the flow of information if an analyst issues unfavorable ratings, since sell recommendations
are
more
visible
and
less
frequent
relatively
to
buy
recommendations. Therefore, an incorrect judgment on a sell recommendation is more likely to be costly for the analyst’s and firm’s reputation than an incorrect buy recommendation. Moreover, analysts are more reluctant to issue sell recommendation arise from the fact that issuing positive recommendations results in a higher profit for the investment bank because favorable recommendation reports attract new clients. Even analysts who are not employed by involved underwriters, issue more favorable recommendations to attract investment-banking deals in the future as is shown by Bradly, Jordan and Ritter (2008). Clients are unlikely to hire an underwriter who has a negative view on the clients company and therefore will hire investment banks that have a favorable view on the company.
!
#$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Empirical research proves that analyst recommendations are influenced by investment banking relationships as is showed by Lin and McNichols (1998) and Michaely and Womack (1999). Most sell-side analysts work for brokerage firms whose primary businesses are investing banking and sales and trading. Growth stocks, and firms with higher trading activity are attractive investment opportunities for investment banking clients. These firms also tend to be widely held by the investment banking clients. As is shown in various researches (e.g. Womack, 1996) brokerages firms are more likely to issue a buy than sell recommendation on stocks, to make the institutional clients places trades with the brokerage firms. Therefore, sell-side analysts have significant economic incentive to issue buy recommendations. Lin and McNichols (1998) examine the effect of investment banks’ relationships on analysts’ earnings forecasts and recommendations. They make a distinction between affiliated and unaffiliated analysts. According to them, affiliated analysts are employed by the lead bank, which underwrites seasoned equity offers (lead underwrite analysts) and analysts who are employed by the co-underwriter bank (counderwriter analysts). Analysts that do not serve as a lead or co-underwriter for a firm are called unaffiliated analysts. Their findings indicate that the recommendations of affiliated analysts are more favorable than unaffiliated analysts, although their earnings forecasts are not. There is, however, no difference found in the postannouncement returns between the recommendations of affiliated and unaffiliated analysts. These findings reflect the greater incentive of affiliated analysts to issue overly favorable recommendations to maintain the client relations. Nevertheless, an investor who follows the recommendations of affiliated analysts, does not experience weaker investment performance than if it would follow the recommendations of unaffiliated analysts. Not only private investors follow the recommendations of analysts, also bank-affiliated investors like asset-management divisions, follow the recommendations issued by their own analysts. According to Jordan, Liu and Wu (2012) bank-affiliated investors follow the recommendations of sell-side analysts in general. They increase their relative holdings following positive recommendations and they decrease their holdings follow negative recommendations. The change in holdings respond more strongly to recommendations issued by their own sell-side analysts, than to those issued by analysts of other banks. !
#%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! There are actions taken to restrict the links between the investment banking and the stock research departments. A settlement in 2004 of the New York Attorney General and the SEC major Wall Street firms agreed to restrict links with the investment banking and the stock research departments, to promote independent research. This settlement tries to anticipate that if conflicts of interest are removed, analyst recommendations will be able to more accurately discriminate undervalued stocks from overvalued stocks. Still, buy recommendations are more commonly issued compared to sell recommendations. The research of Jegadeesh and Kim (2006) showed that in the Group of Seven countries (G7) there are 15.3% sell or strong sell recommendations, compared to 46.9% buy or strong buy recommendations. Even in these countries the percentage of (strong) buy and (strong) sell recommendations differs indicating by the higher percentage of (strong) buy and lower percentage of (strong) sell recommendations in the US compared to the other countries. 2.4 Herding behavior of analysts Herding can be defined as behavior patters that are correlated across individuals. Imitation and mimicry is a basic instinct in human nature, since humans have a strong desire to be part of the group and if something goes wrong it is more confortable to be part of a group rather than to be alone. Keynes nicely paraphrases this as “Worldly wisdom teaches that it is better to fail conventionally than to succeed unconventionally”. Herding occurs in non-economic activities (e.g. fashion), but has also been linked to multiple economic activities, such as price behavior of IPO’s, mergers and acquisitions, earnings forecasts, portfolio managers and bank runs. In the field of analysts’ recommendation, herd behavior occurs when analysts issue the same earnings estimations to mimic the actions of others. Scharfstein and Stein (1990) described herding as one would stick to the way of behaving that has become usual and therefore follows the consensus. Michaely, Thaler and Womack (1995) founds that herding is common for earnings estimate revisions, but is relatively rare in the case of recommendations changes. !
#&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Devenow and Welch (1996) divided herding into four categories. The first type of herding is called informational cascades. Informational cascades occur when the existing aggregate information becomes so overwhelming, that an individual’s single piece of private information is not strong enough to reverse the decision of other analysts. Therefore the analyst chooses to discard its private information and mimic the recommendation of others. If this rationale holds for one individual, then it is likely to hold for the analysts acting after the first person. This domino effect is often referred as a cascade. The second category of herding is called reputational herding. Just like informational cascades, the analyst chooses to ignore its private information and mimic the action of other analysts who has build up a good reputation from the past. The third form of herding is called investigative herding and occurs when an analyst chooses to investigate a piece of information, who believes that other analyst also will investigate. The analyst who discovers the information first, can benefit only from an investment, when other investors follow and push the price of the asset in the direction the analyst predicted by the first analyst. If the investors do not follow, the analyst may saddle with an asset that cannot be sold profitable. The last form of herding that Devenow and Welch (1996) described is empirical herding. This means that analysts follow empirical researches that other analysts use, such as buying past winners, or repeating the predominant buy or sell pattern from previous period. Graham (1999) used these types of herding as variables to build a model. In the model there are four main incentives for analysts to discard private information, and to mimic the actions of others increases with the initial reputation of the analyst. First of all, an analyst with a high reputation and corresponding salary is likely to herd, in order to protect his initial level of reputation and payment. This is in line with the research of Prendergast and Stole (1996), where young analysts exaggerate private information to look knowledgeable, while older analysts make more conservative recommendations. In the same was as Prendergast and Stole (1996), Jegadeesh and !
#'!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Kim (2010) showed that analysts from more reputed brokerage firms are more likely to herd than analyst who works for less reputed brokerages. Secondly, the incentive to herd decreases with the ability of the analyst. Thirdly, the incentive to herd increases in the strength of prior public information that is consistent with the leaders action and finally it increases with the level of correlation across informative signals. This means that analysts have the same information and act all in the same way, which is acknowledged by Welch (2000). He shows that analysts’ recommendations are influenced by the recommendations of previous analysts. Given that a stock is on average followed by 20 to 30 analysts, it is not surprising that analysts’ opinions are positively correlated. Jegadeesh and Kim (2010) found stronger herding effects for downgrades recommendations than for upgrades. They interpreted this, as that analysts are more reluctant to stand out for the crowd, when analysts’ recommendations contains negative information. 2.5 Analysts and stock-market bubbles As the current state of literature showed, analysts are prone to conflicts of interest and herding behavior. Analysts therefore issues recommendations that are different from their private expectations, which can be accompanied with stock market bubbles since great numbers of investors follow the recommendation of analysts. As Abreu and Brunnermeier described (2003), bubbles can survive, as the coordination element do not burst. Some investors are aware that the price of an asset surpassed the fundamental value, but they want to ride the bubble as it continues to growth. They would like to exit the market just before the crash, as Keynes nicely described as ‘beat the gun’. When a sufficient mass of arbitrageurs is sold out, the lack of decreasing buying pressure will start to burst the bubble. Balakrishnan, Schrand, Vashishtha (2011) showed that analyst recommendation concentration has a positive association with stock-market bubbles. During the bubble-period, the firms involved in the bubble faces a higher percentage of buyrecommendations, more upgrades and fewer downgrades relative to firms that are not involved in the bubble. When the bubble burst, the involved companies experienced a significant increase in downgrades.
!
#(!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 2.6 The evolution of the post-recommendation drift literature The literature about the effect of recommendations on stock market returns is bilateral. On the one hand, there is a school of researchers who believes that the recommendations of analysts do not have an impact on the stock market returns. As is described earlier, Cowles (1933) was one of the first researchers who investigated the post-recommendation drift. He found that the 7500 recommendations on the twenty fire insurance companies and sixteen financial services compiled an average recorded that was worse than that of the average common stock by 1.43 per cent annually. After Cowles, the literature about the impact of recommendations on the stock market was not being amplified since the 1960s and 1970s. During these decades researches from Colker (1963), Diefenbach (1972), Logue and Tuttle (1973), Bidwell (1977) and Groth, Lease, Lewellen and Schlarbaum (1979) confirmed the findings of Cowles. Colker (1963) examined the recommendations issued in the Wall Street Journal from the years 1960 and 1961 and uses the SP425 as benchmark. He found that the recommendations slightly outperformed the market, but he classifies the small outperformance as an inability to translate their financial acumen into an impressive degree of prophecy. Logue and Tuttle (1973) investigated the recommendations of the six largest brokerage houses, which were available in the ‘The Wall Street Transcript’. They found that the brokerage houses did not obtained a superior investment performance that could be expected, since the costs of obtaining these recommendations are quite large. The inability to obtain superior investment performance is also showed in the research of Bidwell (1977), when he used examined a beta-adjusted benchmark of eleven brokerage firms. Groth, Lease, Lewellen and Schlarbaum (1979) analyzed the recommendations issued in the years 1964-1970 and found that there is an excess return prior to the announcement instead of after the announcement. On the other hand, there is a vast amount of literature that proves that recommendations issued by analysts do add value. A few researchers in the 1970s, 1980s and 1990s use prominent Journals as input for their researches, such as the monthly issued Wall Street Journal. Davies and Canes (1978) investigated price reactions with buy and sell recommendations printed in the Wall Street Journal column ‘Heard on the Street’, based on the years 1970 and 1971. In contrary of the !
#)!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! findings of Colker (1963) they do found abnormal price movements the day of the publication and the day after. They also detect an asymmetric price reaction to recommendations, whereas buy sell recommendations have a greater postrecommendation drift compared to buy recommendations. The findings of Davies and Canes are supported by researches of Liu, Smith and Syed (1990) and Beneish (1991), whose researches are also based on the recommendations issued by the ‘Heard on the Street’ column. Liu, Smith and Syed (1990) showed that if information is leaked about forthcoming recommendations in the column, it still causes a significant market reaction on its publication date. Beneish (1991) founds a price reaction of about 2.0 percent for buys and about -3.0 percent for sells. Furthermore, he states that analysts have an incentive to leak their recommendations to the media before they inform their clients, to establish their reputation. Another column published by the Wall Street Journal, is called the ‘Dartboard’ column. In this column, four investment analysts give a recommendation to one stock each, and their picks are compared to a portfolio of four randomly chosen stocks selected by the staff of the Journal. Barber and Loeffler (1993) found an abnormal return of four percent for the two days following the publication of the analyst recommendations. The pattern is reversed, however, within 25 trading days. According to Barber and Loeffler the results can be explained by two driving factors. On the one hand, the naïve buying pressure after the recommendation of the analysts, temporary creates buying pressure, which causes the abnormal return. On the other hand, the issued recommendation reveals relevant information and therefore it presents a fundamental revaluation of the security. Dimson and Marsh (1984) performed a different analysis and this study was a measurement of the correlation between forecast and actual returns. They examined 4,187 one-year forecasts of 206 UK stocks made by 35 brokers in the years 1980 and 1981. Their findings suggest that analysts were able to distinguish losers from winners, although high forecasts tend to be overestimated and low forecast to be underestimated. Instead of looking at the impact of a recommendation on the stock price, many researches look at the impact of a revision of the recommendation on the stock price. A benefit is that this approach allows determining the extent of mispricing, analysts !
#*!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! are able to detect, at the time of their recommendations change. Tests also showed that the predictive power of recommendation revisions is more robust than the level of analyst recommendations. Another reason why this method is useful, is that up- and downgrades of recommendations are among the most prominent news items and are most likely to convey to important institutional customers of brokerage firms. This method is used more intensively from 1990, in researches of Stickel (1985,1995), Elton, Gruber and Grossman (1986), Womack (1996) and Jegadeesh and Kim (2006). Stickel (1985) finds abnormal event-window returns of 2.4 percent for firms added to the highest rank, a buy recommendation and only a -0.3 percent for firms added to the lowest rank, a sell recommendation. Elton, Gruber and Grossman (1986) employed a larger and more carefully constructed dataset than those used in previous researches. They choose to focus on large capitalized firms and they eliminated stocks that were followed by less than three analysts. The recommendations of 33 brokerage firms during the years 1981-1983 were investigated and they find evidence that buying the brokerage firm list leads to excess return, but acting on recommendation revisions leads to a greater excess return. They find an excess return of 1.9 percent for addedto-buys and -0.4 percent for added-to-sells. The usage of more comprehensive datasets and more careful empirical analysis increased the precision of the results. In the researches of Dimson and Marsh (1984) and Elton, Gruber and Grossman (1986) monthly calendar-return data is used, while the studies from 1990 on, use daily returns. Moreover, these studies uses event study methodology, which makes it possible that earnings release dates and cross-sectional characteristics of firms and analysts making the recommendations, can be used. Stickel (1995) for instance, examines 17,000 recommendations issued in the years 1988-1991 and finds a positive stock return of 1.16 percent for buy recommendations and a negative stock return of -1.28 percent over the 11-business days event window. He also showed that downgrades to sell or strong sell have greater negative impact than downgrades to hold. The opposite holds, to a lesser extent. Upgrades to strong buys have a greater positive impact than upgrades to buy recommendations. Stickel argued that because of these results, analysts are able to detect the extent to which a stock is over- or undervalued. He also investigated the price reaction of recommendations that skip a rank (e.g. from hold to buy; from strong buy to sell, etc.). !
#+!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! He finds that revisions that skip a rank have a larger price effect than changes in recommendations that do not skip a rank. The precision of the results is also established by using newer techniques of benchmarking that uses Fama-French factors (excess returns, small minus low and high minus low factors) and industry-adjusted returns. This method is first used in Womack (1996) and it adjusts more appropriately for risk. Womack analyzed new buy and sell recommendations of stocks by security analysts at the U.S. market. Womack documented significant price and volume reactions. On average stocks increased with 3.0 percent for buy recommendations and drop 4.7 percent for sell recommendations in a three-day event period window, indicating that the market reaction is asymmetric. The post-recommendations returns are not mean reverting, are significant and in the direction forecast by the analysts. He also found that stock prices rise after the firms are added to the buy list and that stock prices have been rising prior to the removals from the buy list. The opposite pattern holds for sell recommendations. Barber, Lehavy, McNichols and Trueman (2005) analyzed the distributions of stock ratings and examines whether these distributions can be used to predict the profitability of analysts’ recommendations. They found that upwards revisions to buy recommendations issued by brokers with the smallest percentage of buy recommendations, significantly outperformed the upward revisions of brokers with the greatest amount of buys. Also the opposite holds, downgrades to hold or sell from brokers with the highest percentage of sell recommendations significantly outperformed those of brokers with the lowest amount of sell recommendations. Jegadeesh and Kim (2006) examine analyst recommendations in the Group of Seven (G7) industrialized countries and evaluate the value of these recommendations over the 1993 to 2002 period. They found that analysts do more often issue buy and strong buy recommendations than sell and strong sell recommendations. This shows that the analysts’ recommendations contain an element of bias towards being favorable. Furthermore, they observed a general pattern of changes in buy and sell. The buy recommendations increased from 1993 to 2000, followed by a decrease in 2001. The observed pattern was observed for sell recommendations. This pattern indicates that !
$,!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! analysts tend to issue more buy recommendations during an expanding economy (bull markets) and more sell recommendations during a contracting economy (bear markets). A different perspective than researching the recommendations or researching the revisions of recommendations is to measure the profitability of a strategy based on the recommendations of analysts. This is a perspective Barber, Lehavy, McNichols and Trueman (2001) used. Their primary goal is to measure the average price reaction to changes in analysts’ recommendations in an event-study setting, while the study of Barber et al examines whether investors can profit from the publicly available recommendations of analysts. Therefore, the transaction costs are taken into account to calculate the net return to investors. The period between 1985-1996 was investigated, which includes over 360,000 recommendations. They used the recommendation consensus of analysts to daily rebalance the portfolio of the investor at the end of the trading day, such that the portfolio contains the most highly recommended stocks. After controlling for market size, book-to-market and price momentum-effects, the portfolio generated an abnormal gross return of 4.13 percent, whereas the portfolio, which includes the least recommended stocks, generated an abnormal gross return of -4.91 percent. When controlling for transactions costs, however, the strategy does not generate an abnormal net return that is reliable greater than zero percent. A less frequent portfolio rebalancing, for instances weekly, monthly or yearly instead of daily, will only lower the return of the portfolio. Therefore, private investors cannot successfully exploit the market inefficiencies by following the recommendation consensus of analysts. Another study of Barber, Lehavy, McNichols and Trueman (2003) analyzed the returns on the buy and sell recommendations of analysts during the 1996-2001 period. They found negative returns for the least favored stocks for the years 1996-1999, but different results for the years 2000 and 2001. During those two years, the least favored stocks of analysts outperformed the market with 13.44 percent, while the most highly recommended stocks underperform the market by 7.06 percent. This pattern was observed for both technology and non-technology firms. This reverse pattern can be explained by the tendency of analysts to continue to recommend small !
$#!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! capitalization firms (growth stocks), despite that these firms were not the favored stocks anymore and it was better to step aside from small capitalized firms. Barber et al questioned the usefulness of analyst stock recommendations and alerted researchers when excluding turbulent times like the 2000-2001 period in sample periods of an empirical research, have a significant impact on the conclusions they draw. The observation that recommendations of analysts change across different economic conditions, is also identified by Jegadeesh, Kim, Krische and Lee (2004) and Kaplinski and Levy (2010). Jegadeesh, Kim, Krische and Lee (2004) showed that analysts from sell-side firms generally prefer ‘glamour’ stocks to ‘value’ stocks. They defined glamour stocks as stocks with a positive momentum, high growth, high volume and relatively expensive. They show that stocks with favorable recommendations
on
average
outperform
the
stocks
with
less
favorable
recommendations. Azzi, Bird, Griringhelli and Rossi (2004) provided an extension on the study of Jegadeesh et al. (2004). They evaluated the recommendations of analysts in fifteen European countries over the period 1994-2004. They investigated the extent of diversity within Europe by splitting the data in four countries (France, Germany, Italy and the UK) and the two regions Other Europe (Austria, Belgium, Greece, Ireland, the Netherlands, Portugal and Spain) and Scandinavia (Denmark, Finland, Norway and Sweden). The results showed that the recommendations of European analysts are biased towards large, high momentum growth stocks. Their results showed that German and Italian analysts were able to add value to the stocks, while the analysts of the UK and France did not. Kaplinski and Levy (2010) showed in a time-series analysis of analyst recommendations that when investor sentiment is positive, analysts issue more optimistic recommendations, while they issue more pessimistic recommendations when the investor sentiment is negative. Recommendations corresponding to small, young, volatile, unprofitable, non-dividend paying, growing and financially distressed firms are more affected by the investor sentiment. Moreover, recommendations corresponding to stocks with a large number of recommendations are also more affected by the sentiment. Kaplinski and Levy define this as sentiment takes a more dominant role in the forming of recommendations when herding takes places. !
$$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! But how exactly can the post-recommendation drift be explained? Barber and Loeffler (1993) offer two theories in what drives the abnormal performance associated with recommendations. The price pressure theory states that the recommendations of analysts create temporary buying pressure by naïve investors in the recommended securities and this buying pressure causes the observed abnormal returns. The information theory maintains that the analysts’ recommendations reveal relevant information and therefore the abnormal performance of the announcement of recommendations represents a fundamental revaluation of the security. Herding is another explanation of the post-recommendation drift. As is explained before, herding can be defined as behavior batters that are correlated across individuals, which means in this case that investors buy the same stocks as others do. Lakonishok, Shleifer and Vishny (1991) observed this kind of behavior in a research under pension fund managers in small stocks. They believed that herding is most likely to occur on the announcement of well-publicized analysts’ recommendations. 2.7 Difference between small and large companies Small-capitalized firms have larger stock price reactions to revisions, compared to large-capitalized firms. This is shown in several studies, for instance Stickel (1995) and Womack (1996). Glezakos, Mylonakis and Papadopoulos (2011) observed that small companies could provide returns up to four times as large compared to large firms. This apparition is closely related to the theory of the small-firm effect. This effect premises that stocks of smaller companies tend to outperform stocks of larger companies. The reasons of the small-firm effect can be applied on the apparition that small firms have larger stock price reactions to revisions, compared to large firms. The first reason is called the neglected-firm effect and it describes that recommendations of small capitalized firms reacts more strongly than for large capitalized firms, because small-capitalized firms are less favored by (institutional) investors, analysts and the media, compared to large firms. Michaely and Womack (2005) reports that 57% of the recommendations by the top 14 brokerage houses involves the 20% largest firms, while only 1% of the recommendations involves the 20% smallest firms. Therefore there is less information available about smaller firms. !
$%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Large firms are globally focused and there are large amounts of information available for these firms while small firms are often locally active with less information available. Existing empirical research shows that analyst following is negatively correlated with information asymmetry. This can be explained as either because analysts reduce the information asymmetry or because they extent coverage to more transparent companies. Hence, this information availability hypothesis suggest that a recommendation of an analyst would provide more additional information to a small firm compared to a large firm and therefore the post-event price drift of small companies is larger because small firms are less covered by analysts. Another explanation why small firms react more strongly to recommendation revisions is that an investment in small firms is often riskier and therefore should be rewarded with higher returns. Small firms have many characteristics of risky firms since they are generally young, volatile, unprofitable, non-dividend paying, growing and financially distressed firms. Moreover, they are often less diversified relatively to large firms, which makes the profits of small firms vary more over the business cycle compared to large firms. Another reason why small firms are more risky compared to large firms is offered by Caves (1970). He mentioned that the ratio of equity to total asset rises with an increasing firm size and therefore large firms have a less risky capital structure. Therefore, an investment in small-capitalized firms should offer a greater return relatively to large firms since these investments exhibit more risk. Small caps are listed on the broad exchange, which makes a stock initially easier to sell. This often does not hold however, since the liquidity is bad as investors less favor small-capitalized firms. Therefore the trading volume and turnover, as is measured by multiplying the trading volume by the stock price), on average will be lower. Moreover, these firms are also more susceptible for a higher volatility on the stock market compared to large caps. Due to the low amount of outstanding shares, combined with the low trading volume and turnover, a higher volatility on smallcapitalized stocks appears. The current state of literature about liquidity of stocks shows that less liquid stocks outperforms more liquid stocks, as is for example shown by Brennan and Subrahmanyam (1996). Pastor and Stambaugh (2003) attribute these high returns by !
$&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! low-volume stocks to a liquidity risk premium, which is a premium that investors receive for holding a low-liquid asset, because the turnover is low and therefore exhibit more risk. Glezakos, Mylonakis and Papadopoulos (2011) proposed two additional reasons that might explain the relatively large returns of small-capitalized firms. Small companies have a higher probability to be taken over due to an acquisition, since these companies have a smaller market capitalization. The probability of a takeover might results in higher stock price movements after buy-recommendations and therefore returns are bigger. Finally, it could be that growth rates of small companies are more sustainable compared to the growth rates of large companies. Since small companies are active in new (and often risky) markets, there is more opportunity to achieve a sustainable growth rate. When firms are growing bigger, competition increases and the market is getting saturated, the growth rate will decrease.
!
$'!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 3. Empirical research The first section of this chapter describes how the data used in this empirical research is abstracted from different databases. Next the methodology used to conduct the analysis will be defined and the last section describes the hypotheses of this empirical research. 3.1 Data The Euronext Amsterdam includes the exchanges Amsterdam Exchange Index (AEX), Amsterdam Mid Cap Index (AMX), the Amsterdam Small Cap Index (AScX) and local funds index. The AEX exhibit 25 shares that have the largest market capitalization and are the most traded stocks of the Amsterdam Stock Exchange. The AMX contains shares 26-50 and the AScX contains shares 51-75 measured by its market capitalization and most trades shares. These shares are included in this empirical research and the index in where a share is recorded will be used as a measurement of its size. The shares that are listed on the local funds index will not be examined, since these shares are not traded very often and therefore are not very marketable. The composition of the indices will be looked up on the Euronext website. Because the AScX started to exist from 2005, the analyst recommendations from January 2005 to December 2011 will be used in the research. The stock prices data will be looked up on DataStream. The official closing price is chosen as the stock price for the equities. The ‘current’ prices taken at the close of market are stored each day. These stored prices are adjusted for subsequent capital actions, like dividend pay-out and mergers and acquisitions. The corresponding stock price return is defined as the difference in the natural logarithms of the closing prices over the period of one day:
!! ! !"
!! !!!!
The main advantage of using the natural logarithm returns instead of arithmetic returns is because of the asymmetry of the returns.
!
$(!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! The analyst recommendations can be obtained using the Institutional Brokers Estimate System (I/B/E/S) database, which is accessible via Wharton Research Data Services (WRDS). The retrieved analysts’ recommendations from the I/B/E/S database, exhibit information about the ticker/cusip, company name, the announcement date and time of the recommendation, the name of the analyst and the analyst recommendation. Note that the data sample also exhibit firms that do no longer exists in order to avoid the survivorship bias. The most commonly analyst recommendations for stocks are ‘strong buy’, ‘buy’, ‘hold’, ‘sell’ and ‘strong sell’. Most brokerage firms do use an expanded system, which include other labels to recommend stocks, for instance ‘market underperform’, ‘market outperform’ or ‘underweight’ and ‘overweight’. I/B/E/S however, standardizes the recommendations and converts them to numerical scores, where 1 is ‘strong buy’, 2 is ‘buy’, 3 is ‘hold’, 4 is ‘sell’ and 5 is ‘strong sell’. To allow for a more intuitive interpretation of the results, the recommendations are coded in such a way that the more favorable recommendations earn a higher score, so that 1 is ‘strong sell’, 2 is ‘sell’, ‘3’ is hold’, 4 is ‘buy’ and 5 is ‘strong buy’. The recommendation revisions of analysts will be chosen to analyze the recommendations. As is presented in the literature, there are two main reasons based on this choice. In the first place, recommendation revisions has as benefit that its approach allows determining the extent of mispricing analysts are able to detect, at the time of their recommendations change. Secondly, test showed that the predictive power of recommendation revisions is more robust than the level of analyst recommendations. As Stickel (1995) described, there are some criteria that the data must satisfy. These criteria are imposed since the primary focus is on the performance of stocks after recommendation revision. First of all, there should be at least one analyst who issues a recommendation for the stock and revises the recommendation during the sample period. Therefore, recommendations are not included in the sample if an analyst makes only one recommendation for the stock. Also, all recommendations that are reiterations of previous recommendations are excluded from the sample, since recommendations are categorized as either upgrades or downgrades compared to the !
$)!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! previous recommendations from the stock by the same analyst. The second criterion is that the analyst code should de available on I/B/E/S. It is impossible to identify the revision of analysts without an analyst code. At the same time, the stock return data on the recommendation revision date should be available. If no stock data is available around the estimation or event window, it is not possible to identify the post-event recommendation drift. Finally, the stock price should be at least ! 1 on the day before the event date. If the stock price is lower on the day before the event revision date, the results could be biased. This is due to the possibility that the stock price shows a relatively large change, since the stock price is low. Table one presents the matrix of changes in recommendations that met the criteria above. Throughout the master thesis, buy recommendations are defined as all upward revisions to a strong buy from buy, hold, sell or strong sell recommendations and all buy recommendation upgrades coming from hold, sell, or strong sell recommendation. Sell recommendations are defined as all downward revisions to a strong sell coming from a sell, hold, buy or strong buy recommendation, all sell recommendation downgrades coming from hold, buy or strong buy recommendation and all downward revisions to a hold coming from buy or strong buy. 3.2 Methodology The event study is a commonly accepted method to research the recommendations of analysts. Inspired by the five-step methodology to conduct an event study of Bowman (1983), De Jong and De Goeij (2011) developed a three-step approach, which will be used in this research. First of all, the timing of the event has to be identified. The timing of the event is equal to the announcement day of the recommendation and is provided in the date of the I/B/E/S database. Next, the benchmark for normal stock return behavior has to be identified. Since stocks of the AEX, AMX and AScX indices of the Amsterdam exchange are used, there is chosen for the Amsterdam All Share index. The Amsterdam All Share index is a weighted index based on all shares of the Euronext Amsterdam. The index is the broadest benchmark-index of Euronext Amsterdam. The reason to choose the Amsterdam All Share index as benchmark is two folded. First of all, data of the AMX and AScX index were not available in DataStream and therefore no academic source was available to abstract the data. !
$*!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Secondly, stocks can upgrade/downgrade across the three indices during the estimation or event period and therefore it is more insightful to work with one general benchmark instead of three different benchmarks. The last step in the event study methodology is to calculate and analyze abnormal returns around the event date. Before the abnormal returns can be calculated, first the normal returns has to be calculated wherefore the benchmark can be used. The market model adjusts normal returns for market wide stock price movements, since many announcements of analyst recommendations occur at the same point in time. Otherwise, the abnormal returns are based on general movements on the Amsterdam exchange instead due to analyst recommendations and therefore the returns might be biased if the whole market goes up or down at the time the analyst issues its recommendations. The abnormal returns return has to be adjusted for differences in beta, since the beta of each stock is not equal to one. Therefore, abnormal returns can be expressed as residuals of the market model: !!" ! !! ! !! !!" ! !!" The market and beta adjusted normal returns can be defined as the residuals or prediction errors of the model above, which can be mathematically expressed as: !"!" ! !! ! !! !!" , where !! and !! are the Ordinary Least Squares (OLS) estimates of the regression coefficients. The determination of normal returns is performed over the estimation period !!! !! !! !, which precedes the event period !!! ! !! !. The time period of the estimation period is set equal to the estimation period of the classical paper about event studies of Brown and Warner (1980), and therefore an estimation period from !! ! -180 to !! !-40 is chosen. The days (weekends and holidays) that there is no trade on the Euronext Amsterdam are excluded from the research, since these days always exhibit a zero percent return and including these days would underestimate the results.
!
$+!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Because of the uncertainty of the announcement date of the analysts’ recommendations, the information content event window is expanded to include days before the announcement date. The I/B/E/S/ database uses the announcement date of the recommendation of the brokerage firm. Nevertheless, it is possible that the proposed change in recommendation is disseminated before the actual event date. Analysts at brokerage firms frequently meet each other to discuss the proposed changes in recommendations. After the meeting, analysts write the report containing information about the proposed change in recommendation, which might be discussed with customers of the brokerage firms before the actual announcement date. This could result in a pre-event drift before the announcement of the change of the recommendation and therefore is useful to expand the event window. As Stickel (1995) showed, the pre-event drift may occur twenty days before the event and the post event drift may last up for six months after the event as is showed by Womack (1996), therefore an event period from t=-20 to t=120 is chosen. Next, the abnormal returns can be calculated. Abnormal returns are defined as actual returns minus normal returns, which can be mathematically considered as: !"!" ! ! !!" ! !!" , where !!" is the return to security i on day t and !!" is the return of the market index. Next the abnormal returns are averaged to improve the analysis. The cross-sectional average of abnormal returns (AAR) at time t can be expressed as: ! !!"! ! ! !
!
!"!" !!!
At the same time, the AR at time period !! to !! can be cumulated to obtain the cumulative abnormal returns (CAR), to acknowledge: !!
!"#! !
!"!" !!!!
Subsequently, the CARs are divided by the number of observations to obtain CAAR, which is equal to the aggregated AARs:
!
%,!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! ! !""# ! ! !
!!
!
!"#! ! !!!
!!"! !!!!
To test whether the CAAR is statistical significant, support of statistical tests are needed. These tests answer the question whether the calculated abnormal returns are significant different from zero at a certain significance level. But first there are some conditions that the expected error term must satisfy, which are also known as the Gauss-Markov conditions: - ! !! ! !! ! ! !! ! ! !. This means that the expected value of the abnormal returns is zero and that on average, the linear regression line is correct. This is also referred as linearity; - The abnormal returns ! !! ! ! ! ! !! and the independent variable !! ! ! ! ! !! are independent; - There is zero correlation between different error terms, which excludes any form of auto-correlation, this is written as: !!! ! !! !=0, for i, j, = 1,…N and ! ! !; - !!!! ! ! ! ! ! ! ! !! ! ! !. This means that all error terms have the same variance, which is also referred as homoscedasticity. First the outliers are taken into account. To see whether there are outliers, a graphical inspection of the variables is made as can be seen in figure one. The variables that are used will be defined later. As can be seen, there are some consistent outliers. To prevent certain biases in the analysis, all observations that deviates more than the average plus or minus three times the standard deviation, will be excluded from the analysis. This means that approximately, observation with abnormal returns with -15% and more than +15% are excluded from the analysis. Next, the standard normal distribution of the error term needs to be checked. Following the Central Limit Theorem, the abnormal returns should follow a standard normal distribution since the number of cross-sections of events in the sample is quite large. To check whether the residuals of the CAR of the event period from t=-20 to t=180 are normally distributed, a histogram of the abnormal returns is drawn, as can be seen in figure two. As can be seen, the data is fairly normally distributed.
!
%#!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! When looking at the Gauss-Markov conditions, the second condition is not likely to holds, because error times might be correlated with each other as events are clustered since several events occur in the same period. This assumption can be relaxed somehow to !! and !! are independent, which is based on an asymptotic distribution. Therefore, the OLS estimator still is consistent. Next, the standard error of the coefficients needs to be adjusted as several statistical tests and figure three points out, that not all error terms have the same variance and therefore exhibit is heteroskedasticity. Moreover, when a Breusch-Pagan / CookWeisberg test is performed, the null-hypothesis that error term has a constant variance can be rejected with 99% confidence. The heteroskedasticity of the error term can be explained due to the fact that some stocks are more volatile than others. Therefore a weighted average of abnormal returns is used that puts a lower weight on abnormal returns with a high variance. This is easily done in Stata by using the command robust. Moreover, the sample exhibits cross-sectional dependence because events are clustered abnormal returns since several events occur in the same period. This causes a biased estimator, since the variance estimator underestimates the variance of the average abnormal returns. As a result the t-statistics are biased upwards and therefore the null-hypothesis is rejected too often. To adjust the standard error of the estimator, the command cluster will be used, which also includes the properties of the command robust. This will correct for the cross-sectional and time-sectional autocorrelation. After these corrections, the OLS estimates will be consistent. 3.3 Hypotheses As the greater part of the literature showed, stock prices increased after buy recommendations and decreased in stock price after sell recommendation. This empirical observation will be expected in the results but the size of the impact of the revision is unknown, since it varies across the literature. As is showed in different studies, small capitalized shares reacts more to analyst recommendations than to large capitalized shares. Therefore, it can be expected that the market reaction for AScX shares is significantly larger compared to AMX and AEX shares and similar, market reactions for AMX shares are significantly larger than AEX shares. This results in the following hypothesis:
!
%$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Hypothesis 1: Recommendation revisions for companies listed in the AScX index have greater price impact than revisions for companies listed in the AMX or AEX index. Likewise, revisions for companies listed in the AMX index have greater price impact than revisions for companies listed in the AEX index. Performing three different regressions, one for the AEX firms, one for the AMX firms and one for the AScX firms will test this hypothesis. Next, these regressions will be conducted across multiple event periods, in order to see if this hypothesis holds at different points in time around the event date. Small firms have often lower coverage of analysts compared to large firms. This is known as the information asymmetry. One would expect that if an up or downgrade follows for a company with low analyst coverage, the price reaction should be larger since more information becomes publicly available relative to companies with high analyst coverage. Therefore, the following hypothesis is proposed: Hypothesis 2:
Recommendation revisions for stocks with low analyst coverage have a greater positive price impact than stocks with high analyst coverage.
The independent variable COVER is included that equals the number of different analysts that made a recommendation revision during the 2005-2011 period. It can be expected that the coefficient of the variable is negative for both buy and sell recommendations, since higher analyst coverage should come along with a lower abnormal return. Small-capitalized stocks have a lower trading volume and turnover relatively to largecapitalized stocks. The low trading volume decreases the bid-ask spread for small firms, which should result in higher positive returns after buy-recommendations of analysts since the demand to buy shares is larger than the supply to sell shares. Also the reverse should hold; a higher negative return after sell-recommendations since the supply to sell shares is larger than the demand. Therefore, the following hypothesis !
%%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! should hold: Hypothesis 3:
Recommendation revisions for illiquid stocks have a greater price impact than revisions for liquid stocks.
The literature pointed out, that illiquidity can be measured by using the turnover and ask-bid spread of stocks. Therefore the independent variables LOGTURN and ASKBID are included. The variable LOGTURN is equal to the log of the trading volume of a stock multiplying by its closing price and ASKBID is equal to the spread between the ask and bid price. The log of the turnover is taken, in order to linearize the relationship between the turnover and the abnormal return. It can be expected that the coefficient of both variables is negative, since a higher liquidity reduces the cumulative abnormal return. Stickel (1995) argued that strong buy and strong sell are respectively the highest and lowest recommendation level and therefore it is possible that a recommendation revision to or from this category conveys a strong signal about the analyst’s opinion that a revision to or from any of the other recommendation levels. Therefore, prices should react more to strong buys than to buy recommendations. Equally, the negative price reaction should react more to strong sells than sell recommendations. However, one can expect that recommendation revisions to a strong buy or sell, should differ in strength between large and small-capitalized firms. This result in the following hypothesis: Hypothesis 4:
Recommendation revisions to strong buy or sell recommendations have a larger price impact in the AMX index than recommendation revisions to a strong buy or sell recommendation in the AEX index. Likewise, recommendation revisions to strong buy or sell recommendations have a larger price impact in the AScX index than recommendation revisions to a strong buy or sell recommendation in the AMX index.
!
%&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! To test this hypothesis the dummy STRONG is included if recommendations are revisions to strong buy or strong sell recommendations. It can be expected that the coefficient of the variable is positive if the revision is set to a strong buy recommendation and negative if it set to a strong sell. Moreover, it can be expected that the coefficient of STRONG is smaller for the AEX index, compared to the AMX and AScX index and that the coefficient of the AMX index is smaller than the parameter of the AScX index. Stickel (1995) showed that upwards recommendation revisions that skip a rank (e.g. from neutral to strong buy) have an additional positive effect in the short event window (over the event window !! ! !!to !! ! !!). In the same way, downwards recommendations that skip a rank (e.g. from buy to strong sell) have an additional negative effect in the short event window. It can also be expected that the price impact of recommendation revisions that skip a rank for small firms is larger compared to large firms, since less information is available for small firms and therefore conveys a stronger signal. Therefore, the following hypothesis is suggested: Hypothesis 5:
Up- or downgrades that skip a rank in the AEX index have a smaller price impact than recommendation revisions that skip a rank in the AMX. Likewise, revisions that skip a rank in the AMX index have smaller price impact than revisions that skip a rank in the AScX.
The dummy SKIPRANK is included if recommendations skip a rank, for instance from sell to buy, or from strong buy to sell. The dummy equals one, if the recommendations skip a rank and it zero otherwise. It can be expected that the coefficient is positive for buy recommendations and negative for sells. Moreover, the coefficient of SKIPRANK should be smaller when the size of the index is increasing. As is showed in the literature review section, the investor sentiment and the current economic condition have a severe impact on recommendations of analysts. Kaplinski and Levy (2010) showed that when investor sentiment is positive, analysts issue more optimistic recommendations, while they issue more pessimistic recommendations when the investor sentiment is negative. The data sample used for the empirical !
%'!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! research, involves the time period in when the American subprime crises evolved to a global financial crisis and affected worldwide the investor sentiment and economic conditions. This affects the recommendations of analysts, and therefore the following hypothesis is suggested: Hypothesis 6:
There are more upgrades in analyst recommendations in years prior the financial crisis and more downgrades after the start.
To test this hypothesis if this hypothesis is right, year dummies over the period 20052011 are included (YEAR2005 – YEAR2011). Each year exhibit a different dummy and it equals one if the year is equal to the year of the dummy and zero otherwise. It can be expected that there are more positive revisions relative to negative revisions in the years 2005 – 2007 and more negative recommendation revisions in the years 2008-2011.
!
%(!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 4. Research results This chapter describes the results of the empirical research as described in the previous chapter. In the first section of this chapter, the descriptive results of the research are shown. Next, the Ordinary Least Squares (OLS) technique and crosssectional regressions will test the proposed hypotheses. A robustness check to test the validity of the results is covered in the third paragraph and the last section covers an additional test to investigate the abnormal volume during the event period. 4.1 Descriptive statistics Before testing the hypotheses, one first has to take a look at the descriptive statistics. In the sample data, a total of 4,689 recommendation revisions were given for 98 companies in the period January 2005 to December 2011. Of these recommendation revisions, 3,917 revisions were used in the research. Table two represents the mean of the non-dummy variables of the data sample for the AEX, AMX and AScX indices. Not surprisingly, the average market capitalization for firms in the AEX index, is substantial larger compared to firms in the AMX and AScX index. Similar, the market capitalization for firms listed in the AMX index is larger than for firms in the AScX index, on average 2.5 times. AEX firms in the data sample not only have a larger market capitalization, they were on average also being followed by a larger number of analysts. Firms listed on the AEX index were on average being followed by 44 analysts, compared to 20 analysts for medium-capitalized firms and 13 analysts for small-capitalized firms. These results are in line with the finding of Welsh (2000), who observed that on average 20 to 30 analysts follow a stock. An analyst issued on average three revisions per firm during the 2005-2011 period, regardless the size of the company. The descriptive statistics also shows that shares of large-capitalized firms have a higher turnover and a lower ask-bid spread, compared to smaller-capitalized companies. This indicates that larger companies are more liquid and hence easier to trade, compared to small-capitalized firms. This statement, as is formulated in hypothesis two, will be researched more closely in paragraph 4.2. Table three shows the distribution of the independent dummy variables. As can be seen, more than half of all recommendation revisions of analysts was issued for firms in the AEX. The fewest recommendations were issued for small-capitalized firms. !
%)!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! The number of downgrades is higher than the number of upgrades, which is different from most researches in literature. This can be explained by the negative investorsentiment in the time-period in which this research is conducted. Analysts were quite bullish when they upgraded a recommendations, since 53.91 percent of the buy recommendation revisions were upgrades to a strong buy recommendation. This is asymmetric compared to downgraded recommendations, since only 13.01 percent of the sell recommendations were downgrades to a strong sell recommendation. In both up- and downgrades, around 47 percent of the recommendations skipped a rank. When investigating the year dummies, the number of upgraded and downgraded recommendations was constant during the years 2005 to 2007, but it fluctuated more in the years thereafter. At the time the crisis started in 2008, the number of upgraded recommendations decreased, whereas the number of sell recommendations increased. This pattern reversed in the year thereafter, when the indices hit their lowest level in years and showed a strong recovery in the years afterwards. In 2010 and 2011 both buy and sell recommendations decreased, indicating that analysts are unsure about forecasts of companies. 4.2 Testing the hypotheses To test the hypotheses, several regressions were ran and stored in tables to see whether the hypotheses can be accepted or rejected. In the following section, each hypothesis is tested separately. Table four and table five show respectively the average abnormal returns (AAR) and cumulative average abnormal returns (CAAR) around different event dates. Tables six to nine seven clarify the cumulative abnormal returns with a linear regression model, including several explanatory variables. Two regressions were run, since the variables COVER and LOGTURN are highly correlated with each other. This might have caused multicollinearity and comes along with untruthful OLS estimates. Therefore, the variable LOGTURN is excluded from the regression model of table six and seven and the variable COVER is excluded from the regression model of table eight and nine. The correlation between the explanatory variables can be found in table ten. When focusing on the R-squared of the regressions, one can see that the variables in the model do not explain much of the variation of the cumulative abnormal return since the R-squared varies between the 0.01 and 0.15. Obviously, the model does !
%*!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! explain more variation in the CAR for small and medium capitalized firms compared to large firms. The F-test of the regressions is lower than the 0.05 for most cases, which indicates that not all of the variables are equal to zero. This shows that the included variables in the regression model do make sense in most of the cases. Hypothesis 1 The first hypothesis stated that revisions for companies listed in the AScX index have greater price impact than revisions for companies listed in the AMX or AEX index and similar, revisions for companies listed in the AMX index have greater price impact than revisions for companies listed in the AEX index. To check whether this hypothesis is valid, first abnormal returns were calculated for single event days as can be seen in table four. Before t =-2, there are hardly any significant abnormal returns for upgraded recommendations. Looking at the coefficients from t =-2 to t =+1 one can see that most of the abnormal returns are significant different from zero. The statistical significance fades away after t =+1. It is not always the case, however, that the abnormal return for companies listed in the AScX index, is always larger than for companies listed in the AMX and AEX index and, similarly, that abnormal returns for companies listed in the AMX index are larger than abnormal returns for companies listed in the AEX index. The abnormal returns for the companies located at the three indices at t=0, statistically do differ significantly from each other. The abnormal return for recommendation revisions for sells, are larger in magnitude compared to upgraded recommendations. The abnormal returns for downgrades for small-capitalized firms are more negative compared to middle-capitalized firms and in the same, abnormal returns of middle-capitalized firms are more negative relatively to large-capitalized firms. The effects also seems to start a day earlier, at t=-3, compared to upgraded recommendations. The abnormal returns across the indices are not always significantly different from each other. When looking at the Cumulative Average Abnormal Returns (CAAR) of buy recommendations (table five), one can see that when an analysts issues an upgrade, the CAAR is statistically significant for the time periods which includes the event date, except the event period !! ! !!!to !! ! !!"#. CAARs are negatively related to the company’s firm size, as was not the case at the single event period. Hence if firms !
%+!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! are listed on the AEX index, the CAAR is significantly higher compared to AMX companies and ASCX companies. The same holds for downgraded recommendations. For time periods which include the event date, with exception of the event period !! ! !!!to !! ! !!"#, the CAAR is statistically different from zero. The cumulative average abnormal returns are becoming more negative whenever the company is listed in a smaller index. These results are in line with the results of Stickel (1995) Womack (1996) and Glezakos, Mylonakis and Papadopoulos (2011), who also found that the post-recommendation drift of small companies is larger compared to large companies. A graphical representation of the performance over the event period !! ! !!"!to !! ! !!"# can be found in figure four and five for upgraded and downgraded recommendations respectively. Figure four shows that upgrades initially react positively and then shows a reverse pattern to zero percent return. This is different from the pattern of downgraded recommendations, where shares show an initially negative reaction to downgraded recommendations and then shows a strong recovery to an even positive return. The remarkable result, however, is that small-capitalized firms do not react equally on a downgraded recommendation compared to medium and large-capitalized firms, since they show a continuously declining pattern after a sell recommendation revision. This cumulative average abnormal return is statistically significant from zero, in contrary to most of the other returns over the event period !! ! !!"!to !! ! !!"#. Because of these findings, it can be assumed that recommendation revisions for companies in the AScX index have greater price impact than revisions for companies the AMX or AEX index around the event date. Likewise, the results shows that around the event date recommendation revisions for companies listed in the AMX index have greater price impact than revisions for firms listed in the AEX index. Hypothesis 2 The second hypothesis states that recommendation revisions for stocks with low analyst coverage have a greater positive price impact than stocks with high analyst coverage. To test this hypothesis, the following regression was run across different !
&,!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! event periods and different indices: !"#!!! !!! ! ! !! ! !! !"#$% ! !! !"#$%& ! !! !"#$%& ! !! !"#$%&'"! ! !! !"#$!""# ! !! !"#$!""# ! ! ! !!" !"#$!"## The variable COVER is also included in the AEX, AMX and AScX regressions, since the number of analysts that follows a company is dispersed within these groups and therefore the coefficient estimator might be statistical meaningful. The results of these regressions can be found in table six (upgraded recommendations) and table seven (downgraded recommendations). The term COVER is in most of the cases negative for the buy recommendation revisions, with exception of the event period !! ! !!!to !! ! !!"#. The regressor is often statistical significant when the regression is run on all firms or AEX firms and not in the regressions for only AMX and AScX firms. Therefore, it does not clarify the differences within the indices. The variable COVER is, as predicted, positive and often significant for sell recommendation revisions. The coefficient estimator often explains the CAR better for all firms in the data sample than for the indices separately. As was the case for upgraded recommendations, the sign and significance is different in the event period !! ! !!!to !! ! !!"#. The results suggest that the hypothesis can be accepted. Hence, when analysts upgrade a stock, the variable seems to predict a lower positive CAR for companies that are covered by more analysts. The reverse holds for downgraded recommendations; a lower negative CAR for companies that are covered by more analysts. The hypothesis does only hold when all firms are considered, except for the event period !! ! !!!to !! ! !!"#. Although the coefficients of the regressor vary a lot, on average the coefficient is not high, which suggests that the results are not economically meaningful. Hypothesis 3 The third hypothesis is formulated as: “Recommendation revisions for illiquid stocks have a greater price impact than revisions for liquid”. To test this hypothesis, the following regression was run separately for buy and sell recommendations, and for all
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! AEX, AMX and AScX firms: !"#!!! !!! ! ! !! ! !! !"#$%& ! !! !"#$%&' ! !! !"#$%& ! !! !"#$%&'"! ! !! !"#$!""# ! !! !"#$!""# ! ! ! !!" !"#$!"## The results of these regressions can be found in table eight and nine. The signs of the variables ASKBID and LOGTURN for upgraded recommendations (table eight) are in most cases negative. The coefficient estimators, however, are not consistently statistically significant and therefore no clear statement can be made about the regressions results. When looking at the downgraded recommendations, one can see that the signs of the variables ASKBID and LOGTURN are in some cases positive and in other cases negative. These coefficients are not always statistically significant neither. Therefore, there is not enough evidence to accept the hypothesis that recommendations for illiquid stocks have a greater post recommendation drift than recommendation revisions for liquid stocks. Hypothesis 4 The fourth hypothesis suggests that recommendation revisions to strong buy or sell recommendations have a larger price impact on AMX companies than recommendation revisions to a strong buy or sell recommendation on AEX companies. Equally, recommendation revisions to strong buy or sell recommendations have a larger price impact on firms in the AScX index than recommendation revisions to a strong buy or sell recommendation on firms in the AMX index. Tables six to nine can be consulted, to see whether this hypothesis can be accepted or rejected. The dummy variable STRONG should be positive and significant for upgrades and its coefficient estimator should be more positive for small-capitalized firms than for middle and large-capitalized firms. When looking at the results of the regression output for upgraded recommendations, we can see that this is not the case. The coefficient estimator is only positive for the event period !! ! !!to !! ! !!" and often negative in all other cases. Moreover, the size of the parameter varies across the different indices and the t-stat suggests that most of the results are not statistically significant at a minimum of a ten percent confidence interval.
!
&$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! When looking at the variable STRONG for downgrades, the estimator is predicted to be negative and its value should be more negative for AScX and AMX shares than AEX shares. Again, the sign of the STRONG dummy varies a lot across the indices, although it seems to be often negative for AScX companies. Since the parameters are not consistently statistically significant, the hypothesis cannot be accepted neither for sell recommendations. Therefore, the alternative hypothesis that recommendation revisions to strong buy or sell recommendations have a larger price impact in smallcapitalized indices compared to larger-capitalized needs to be rejected for both buy and sell recommendations. Hypothesis 5 Up- or downgrades that skip a rank in the AEX index have smaller price impact than revisions that skip a rank in the AMX. Likewise, up- or downgrades that skip a rank in the AMX index have smaller price impact than recommendation revisions that skip a rank in the AScX. Therefore, it can be expected that the coefficient estimator is positive for buy recommendations and increasing across the size of the indices and that it should be negative and decreasing for sell recommendations. When looking at the dummy variable SKIPRANK in tables six to nine, one can see that the sign for this dummy is positive except for the AMX index in the event period !! ! !!!to !! ! !!. The t-statistic shows that the coefficient estimator is not consistently statistical significant. Although it seems to be the case that buy recommendation revisions that skip a rank have a greater positive return than recommendation revisions that do not skip a rank, there is not enough statistical evidence to accept this and hence it can also not be accepted that buy recommendation revisions for smallcapitalized companies have a greater positive price reaction compared to largecapitalized companies. When focusing on the downgraded recommendations, the sign of the coefficient estimator is in some cases positive and in some cases negative. Moreover, the parameter is not always statistically significant. Hence, there is also not enough evidence for downgraded recommendations to accept that recommendation revisions for small-capitalized firms have a greater price movement than for large-capitalized firms. Therefore, hypothesis five has to be rejected.
!
&%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Hypothesis 6 Table three can be consulted to see whether there are more upgrades in analyst recommendations in years prior to the financial crisis and more downgrades after the start. As is already mentioned in the beginning of this chapter, the financial crisis started in late 2007/beginning 2008. This crisis had an impact on the number of recommendations, as the number of upgrades dropped with 28% from 2007 to 2008 and the number of downgrades increased by more than 10%. This pattern was reversed in the subsequent year, where stock markets showed a strong recovery after hitting it lowest level since the burst of the internet bubble in 2001. In 2009, the number of upgrades increased with 27% and the number of downgrades decreased with 11%, compared to 2008. In 2010 both up- and downgrades dropped significantly, respectively 23% and 24% and stayed constant in the next year. These results are in favor of the hypothesis, though the results seem to be short-lived since the pattern quickly reverses. Moreover, the results are in line with the results of Kaplinski and Levy (2010), who found that when investment sentiment is more positive, more buy recommendations are issued and when the investment sentiment is negative more sell recommendations are issued. When looking at tables six and eight, we can see that the dummy variables YEAR2008 and YEAR2009 for upgrades around event period !! ! !!!to !! ! !! and event period !! ! !!to !! ! !!" are statistically significant different from zero. This finding holds when the regression model includes only dummies of the years after 2008. For downgrades the dummy variables YEAR2008 and YEAR2009 around the three day event period !! ! !!!to !! ! !! is statistically different from zero. In addition, the dummy YEAR2011 is statistically significant in all event periods. This shows that during the years 2008, 2009 and 2011 the cumulative abnormal return is more negative compared to 2005. Again, this observation still holds when the regression model includes only dummies of the years after 2008. 4.3 Robustness test A robustness test can be done, by using the market capitalization of firms that is obtained from DataStream. Market value is measured by multiplying the number of ordinary shares by its corresponding closing price. Then the following regression is run separately for buy and sell recommendation revisions for all firms: !
&&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! !"#!!! !!! ! ! !! ! !! !"# ! !! !"#$%& ! !! !"#$%& ! !! !"#$%&'"! ! !! !"#$!""# ! !! !"#$!""# ! ! ! !!" !"#$!"## The variable CAP is the logarithm of the market capitalization of firms. The variables COVER and LOGTURN are not included, since these variables are strongly correlated with the variable CAP. The results of this regression can be found in table eleven. The coefficient estimator of the variable is negative and significant for upgraded recommendations and positive and significant (except the event period !! ! !!! to !! ! !!"# ) for downgraded recommendations. This result gives additional support that small-capitalized firms have higher cumulative abnormal return compared to large companies. Hence, this means that upgrades for small firms have a larger positive cumulative abnormal return than for large firms and that downgrades for small firms have a larger negative cumulative abnormal return price impact than for large firms. This finding gives additional evidence in favor of hypothesis one. 4.4 Additional test Womack (1996) found that the trading volume on the event day differs significantly compared to the average volume. To test if this is also the case in this empirical study, an additional test was performed to test whether the trading volume of companies of the different indices differs on the event date. The abnormal volume is being measured by using the formula that Womack proposed in his study:
!"!!
!
!!! !
!!" ! !!!! !!
!
!" ! ! !! !
! ! !"#
!
where !"!! is the abnormal volume and !!! is the actual trading volume from stock i at time t. The average trading volume is being measured at 60 days before the event period !! ! !!!to !! ! !! and 60 days after this event period. Then the average of all observations i is taken at time period t, to calculate the average abnormal trading volume. The abnormal volume of upgraded recommendations can be found in figure
!
&'!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! six and seven and the abnormal volume of downgraded recommendations can be found in figure eight and nine. As can be seen visually, the abnormal volume of buy and sell recommendations do not differ significantly. This result is supported by the Hausman difference tests at the event period !! ! !!!to !! ! !! with p-values of respectively 0.0386, 0.1314 and 0.6885, which suggests that there is no significant different in the coefficients. There are however, differences across the indices. On the three day period around the event date, the abnormal volume is statistical larger for the AScX shares than the AMX and AEX shares and the abnormal volume in the AMX index is larger than the AEX shares. The three p-values 0.0002, 0.000 and 0.0348 for respectively the three event dates !! ! !!!to !! ! !! rejects the nullhypothesis that suggests that the abnormal volume for companies listed on the three indices are the same. During the event days, the abnormal volume can increase with 50% compared to the average volume. All these abnormal volumes are strongly statistically significant at a confidence level of 99%. The abnormal volume converges to its average again within approximately six days, although the abnormal volume of AScX shares still can fluctuate.
!
&(!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 5. Summary and main conclusions The existing literature about the impact of recommendation from analysts on stock prices is large and comprehensive. Literature describes all kind of approaches that shed light on the analyst anomaly, from the methodologies to issue recommendations, to the herding behavior of analysts and the risk to create a stock-market bubble. A substantial amount of researches has shown that analysts do add value to the stock market, which is in contrary to the semi-strong and the strong form of the Efficient Market Hypothesis (EMH) that states that no excess return can be earned by acting on publicly available new information since share prices reflect all public and private information. According to literature the post-recommendation drift differs in size across the researches. Mostly the effect varies between one and four percent, depending on whether it is a buy or sell recommendation and on the companies’ firm size. The size of the abnormal return of sell recommendations is predominantly larger relative to buy recommendations. This is primarily explained by the fact that analysts do often issue more sell than buy recommendations. This shows that the analysts’ recommendations contain an element of bias towards being favorable. Therefore, a sell recommendation conveys a strong negative signal, since sell recommendations are more conspicuous than buy recommendations. As literature shows, it can be expected that small-capitalized firms have a greater price reaction to recommendation revisions compared to large companies. That is, small firms tend to react more positively to buy recommendations and more negatively to sell recommendations. The literature explains this small-firm effect due to the information asymmetry, the neglected-firm effect and due to a higher level of risk for small firms. To extend the existing literature about the small-firm effect, this master thesis investigated this effect on the Euronext Amsterdam. The main research question was: Are analysts’ recommendation revisions able to 1) add more value to the Dutch small cap index than the mid-cap or large-cap indices and to 2) add more value to the midcap index compared to the large-cap index?
!
&)!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Instead of looking at the impact of a recommendation on the stock price, this study looked at the impact of a revision of the recommendation on the stock price. A benefit is that this approach allows determining the extent of mispricing that analysts are able to detect, at the time of their recommendations change. Literature showed that the predictive power of recommendation revisions is more robust than the level of analyst recommendations. In the data sample used for this empirical research, there are more downgraded recommendations issued than upgraded recommendations. This is unusual, since most researches find that analysts’ recommendations contain an element of bias towards being favorable. This could suggest that analysts on the Euronext Amsterdam are more independent compared to other countries (primarily the USA) and that they face less conflicts of interest and competing pressures to issue buy recommendations. The sell recommendations issued by analysts have a greater price reaction and a greater trading volume reaction than buy recommendations. In the three day period around the event date, buy recommendations for AEX, AMX and AScX yield on average a cumulative abnormal return of respectively 0.96%, 1.29% and 1.35% while it yields a cumulative abnormal return for sell recommendations of -1.11%, -1.66% and -2.22%. The asymmetry that stocks reacts more to sell recommendations than buy recommendations was expected, as was showed in researches of Stickel (1995), Womack (1996) and Jegadeesh and Kim (2006). Moreover, the results show that revisions for AScX companies have a greater price reaction and trading volume increase than revisions for AMX and AEX companies and that the recommendation revisions for AMX companies have a greater price reaction and trading volume increase than AEX firms. The remarkable result is that firms in the AScX index react extremely negative (-7.00%) to downgrades over the six months period after the event. This evidence does not support the strong and semi-strong form of the efficient market, since the conclusion can be drawn that analysts are able to 1) add more value to the Dutch small cap index than the mid-cap or large-cap indices and to 2) add more value to the mid-cap index compared to the large-cap index. Furthermore, it was tested whether the information-asymmetry could explain the small-firm effect. The information-asymmetry was tested using the number of analysts that follows stocks. Results point out that stocks in the AScX index are on !
&*!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! average being followed by fewer analysts, compared to companies that are included in the AMX and AEX. A relationship is found between the number of analysts that follows a stock and the abnormal return. Small-capitalized firms experience a greater post-recommendation drift compared to large firms. This result shows that the information-asymmetry theorem holds. Another explanation for the small-firm effect is the illiquidity of small-capitalized stocks, as is shown in literature. However, after testing whether illiquid stocks have a large post-recommendation drift after up-and downgrades, no result was being found. Therefore, illiquidity does not explain the differences in cumulative abnormal return across the indices. The results of the empirical research showed that in 2008, the year the crisis started, the number of upgraded recommendations decreased with 28% and the number of downgraded recommendations increased with 10%. This result shows that when investor sentiment is negative, analysts issue more pessimistic recommendations. This pattern was reversed in the year 2009, where stock markets showed a strong recovery after the indices AEX, AMX and AScX hit it lowest level since years. In this year, compared to 2008, the number of upgraded recommendation increased with 27% and the number of downgraded recommendations decreased with 11%. The results showed that the buy and sell recommendations issued in 2008 and 2009 have a greater post-event drift compared to other years in the sample, suggesting that analysts add more value during a pessimistic sentiment. For further research, it would be interesting to find other factors that might explain the differences between small-capitalized and large-capitalized companies. The variables in the regression model of this research do not explain much of the variation of the cumulative abnormal returns, although R-squared tends to increase when the research is focused on small-capitalized firms. Furthermore, it would be interesting to extend this research to all countries in Europe, to see if there are differences within Europe. This should be expected since capital markets in Eastern European countries are less developed, compared to West-Europe. Another nice extension of this research would be to investigate the differences between small and large companies in developing countries where the cumulative abnormal returns between companies could be even larger for less efficient stock markets exchanges. !
&+!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! 6. References Abreu, D., Brunnermeier, M. (2003), Bubbles and Crashes. Econometrica, 71, (1), 173-204. Azzi, S., Bird R., Griringhelli, P., Rossi, E. (2004), Biases and Information in Analysts’ Recommendations: the European Experience. European Financial Management Association, 1-42. Balakrishnan, K., Schrand, C., Vashishtha, R. (2011), Analyst Recommendations and Higher Order Beliefs: Explaining Bubbles and Price Drift. Working paper. Barber, B., Loeffler, D. (1993), The ‘Dartboard’ Column: Second-hand Information and Price Pressure. Journal of Financial and Quantitative Analysis, 28, (2), 273-284. Barber, B., Lehavy, R., McNichols, M., Trueman, B. (2003), Reassessing the Returns to Analysts’ Stock Recommendations. Financial Analysts Journal, 59, (2), 88-96. Barber, B., Lehavy, R. McNichols, M., Trueman, B. (2005), Buy, Holds and Sells: The Distribution of Investment Banks’ Stock Rating and The Implications For The Profitability of Analysts’ Recommendations. Working Paper. Beneish, M. (1991), Stock Prices and the Dissemination of Analysts Recommendations. Journal of Business, 64, (3) 393-416. Bidwell, C. (1977), How Good is Institutional Brokerage Research, Journal of Portfolio Management, 3, 393-416. Black, F. (1973), Yes Virginia, There is Still Hope: Test of the Value Line Ranking System. Financial Analyst Journal, 29, 10-14. Bowman, R. (1983), Understanding and Conducting Event Studies. Journal of Business Finance and Accounting, 10, (4), 561-584.
!
',!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Bradley, D., Jordan, B., Ritter, J. (2003). The Quiet Period Goes Out With a Bang. Journal of Finance, 58, (1), 1–36. Bradshaw, M. (2002), The Use of Target Prices to Justify Sell-side Analysts’ Stock Recommendations. Accounting Horizons, 16, 27-41. Bradshaw, M. (2004), How Do Analysts Use Their Earnings Forecasts in Generating Stock Recommendations? Accounting Review, 79, 25-50. Brennan, M., Subrahmanyam, A. (1996), Market Microstructure and Asset Pricing: On the Composition for Illiquidity in Stock Returns. Journal of Financial Economics, 41, 441-64. Brown, S., Warner, J. (1980), Measuring Security Price Performance. Journal of Financial Economics, 8, 205-258. Caves, R. (1970), Uncertainty, Risk Structure and Performance: Galbraith as Conventional Wisdom. University of Boston. Chan, K., Chan, L, Jegadeesh, N., Lakonishok, J. (2006), Earnings Quality and Stock Returns. Journal of Business, 79 (3), 1041-1082. Chan, L., Karceski, J., Lakonishok, J. (2007), Analysts’ Conflict of Interests and Biases in Earnings Forecasts. Journal of Financial and Quantitative Analysis, 42, (4), 893-913. Colker, S. (1963), An Analysis of Security Recommendations by Brokerage Houses. Quarterly Review of Economics and Business, 3, 19- 28. Copeland, T., Mayers, D. (1982), The Value Line Enigma: A Case Study of Performance Evaluation Issues. Journal of Financial and Quantitative Analysis, 35, 485-498.
!
'#!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Cowles, A. (1933), Can Stock Market Forecasters Forecast? Econometrica, 1, 309324. Davies, P., Canes, M. (1978), Stock Prices and the Publication of Second-Hand Information. Journal of Business, 51 (1), 43-56. Devenow, A., Welch, I., (1996), Rational Herding in Financial Markets. European Economic Review, 40, 603–616. Diefenbach, R. (1972), How Good is Institutional Brokerage Research? Financial Analyst Journal, 28, 54-60. Dimson, E., Marsh, P. (1984), An Analysis of Brokers’ and Analysts’ Unpublished Forecasts of UK Stock Returns. Journal of Finance, 39, (5) 1257-1292. Elton, E., Gruber, M., Grossman, S. (Discrete Expectational Data and Portfolio Performance. The Journal of Finance, 41 (3), 699-713. Fama, E. (1970), Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal of Finance, 25, (2), 383-417. Francis, J., Philbrick, D. (1993). Analysts’ Decisions as Products of a Multi-Task Environment. Journal of Accounting Research, 31, (2), 216-230. Givoly, D., Lakonishok, J. (1979), The Information Content of Financial Analysts’ Forecast of Earnings. Journal of Accounting and Economics, 1, (3), 165-185. Glezakos, M., Mylonakis, J., Papadopoulos, J. (2011). Do Revised Analyst’ Recommendations Influence Investors’ Decisions and Include a Size Effect? Journal of Money, Banking and Investing, 21, 21-31. Graham, J. (1999), Herding Among Investment Newsletters: Theory and Evidence. The Journal of Finance, 44 (1), 237-268. !
'$!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Groth, J., Lease, R., Lewellen, W., Schlarbaum, G. (1979), An Analysis of Brokerage House Securities Recommendations. Financial Analysts Journal, 35, (1), 32-40. Grossman, S., Stiglitz, E. (1980), On the Impossibility Efficient Markets. The American Economic Review, 70, (3), 393-408. Grossman, S. (1975), Essays on Expectations. Unpublished doctoral dissertation. Grossman, S. (1977), The Existence of Futures Markets, Noisy Rational Expectations and Informational Externalities. Revised Economic Studies, 64, 431-449 Hong, H., Kubik, J. (2003), Analyzing the Analysts: Career Concerns and Biased Earnings Forecasts. The Journal of Finance, 58, (1), 313-351. Jegadeesh, N., Kim, W., Krische, S., Lee, S. (2004), When do recommendations add value? Journal of Finance, 59, (3), 1083-1124. Jegadeesh, N., Kim, W. (2006), Value of Analyst Recommendations: International Evidence, Journal of Financial Markets, 9, (3), 274-309. Jegadeesh, N., Kim, W. (2010), Do Analysts Herd? An Analysis of Recommendations and Market Reactions. The Review of Financial Studies, 23 (2), 901-937. Jordan, B., Liu, M., Wu, Q. (2012), Do Investments Banks Listen to Their Own Analysts? Journal of Banking & Finance, 36, 1452-1463. Kaplanski, G., Levy, H. (2010), Sentiment Effect on Analysts’ Recommendations: Time-Series and Cross-Section Analyses. Journal of Finance, 95, (2), 174-201. Ke, B., Yu, Y. (2007), The Effect of Ability, Independence, and Investor Sentiment on Analysts’ Propensity to Use Their Own Earnings Forecasts in Stock Recommendations. Working Paper.
!
'%!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Lin, H., McNichols, F. (1998), Underwriting Relationships, Analysts’ Earnings Forecasts and Investment Recommendations. Journal of Accounting and Economics, 25, 101-127. Liu, P., Smith, S., Syed, A., (1990), Stock Price Reactions to The Wall Street Journal’s securities Recommendations. Journal of Financial and Quantitative Analysis 25, 399-410. Logue, D., Tuttle, D. (1973), Brokerage House Investment Advice. Financial Review 8, (1), 38–54. Michaely, R., Thaler, R., Womack, K. (1995), Price reactions to dividend initiations and ommisions: Overreaction or drift? Journal of Finance, 50, 573-608. Michaely, R., Womack, K. (1999), Conflict of Interest and the Credibility of Underwriter Analyst Recommendations. Review of Financial Studies, 4, (12), 653686. Muth, J. (1961), Rational Expectations and the Theory of Price Movements. Econometrica, 29, (3), 315-335 Pastor, L., Stambaugh, R. (2003), Liquidity Risk and Expected Stock Returns. Journal of Political Economy, 111, 642-685. Pratt, T. (1993), Wall Street’s Four-letter Word, Investment Dealers Digest. 59, (13), 18-22. Prendergast, C., Stole, L. (1996), Impetuous youngsters and jaded old-timers: Acquiring a reputation for learning. Journal of Political Economy, 104 (6), 1105– 1134. Scharfstein, D. Stein, J. (1990), Herd Behavior and Investment. American Economic Review, 80 (3), 465-479. !
'&!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Shefrin, H. (2002), Beyond Greed and Fear: Understanding Behavioral Finance and the Psychology of Investing. Boston, Mass: Harvard Business School Press. Stickel, E. (1985), The Effect of Value Line Investment Survey Rank Changes on Common Stock Prices. Journal of Financial Economics. 14, 121-143. Stickel, E. (1995), The Anatomy of Performance of Buy and Sell Recommendations, Financial Analyst Journal, 51, 25-39. Welch, I. (2000), Herding Across Security Analysts. Journal of Financial Economics. 58, (3), 369–396. Womack, K. (1996), Do Brokerage Analysts’ Recommendations Have Investment Value? The Journal of Finance 51, (1), 137-167. Womack, K. (2005), Brokerage Recommendations: Stylized Characteristics, Market Responses and Biases. Advances in Behaviorial Finance II.
!
''!
Stock market analysts’ recommendations: Difference in the Dutch stock market abnormal return between the small cap, mid cap and large cap index.
Appendix: Figures and tables
Faculty of Economics and Business Administration Finance Department Master thesis in Finance Student Name:
Ruud Knippenburg
Supervisor:
Drs. J. Grazell
Administration number:
698699
Defense Date:
27 September 2012
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!!
!"#$%&'()'*%+,-".+/'"01,&.2"30'34'3$2/"&%1''
!
0
.1
Density .2
.3
.4
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!!
-40
-20
0 Abnormal Returns
20
40
! !"#$%&'5)'6"123#%+7'34'2-&'8&01"29'34'+:03%7+/'%&2$%01';"2-'3$2/"&%1''
!
! !"#$%&'&'+>&%+#&'+:03%7+/'%&2$%0'34':$9'%&.377&08+2"30'%&>"1"301'
)!
!"#"$%&'()*+,-./#%$*0)&"/-*1!+02*
*! +! #)+! #*+!
+!
*+! )+! (+! '+! &+! %+! $+! ,+! -+! *++! **+! *)+!
#*! #)! #(! #'! .//!012340!
#&!
.56!712340!
#%! #$!
.86!712340! .796!712340! 3()-&*4%56*
!"#$%&'A)'@$7$/+2">&'+>&%+#&'+:03%7+/'%&2$%0'34'1&//'%&.377&08+2"30'%&>"1"301'
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! (!
'! (:&!
+,-./#%$*7.$"#)*
):&!
(! ):&!
)!
)!
*:&!
*:&! *!
*!
+:&! +!
+:&!
#)+! #*,! #*%! #*'! #*)! #*+! #,! #%! #'! #)! +! )! '! %! ,! *+! *)! *'! *%! *,! )+!
+!
#+:&! #*!
!"#"$%&'()*+,-./#%$*0)&"/-*1!+02*
'
.;!.//!712340! 3/$7&'34'+//'1-+%&1':$9'%&.377&08+2"30'%&>"1"301'
! ! !
(!
'! (:&!
+,-./#%$*7.$"#)*
):&!
(! ):&!
)!
)!
*:&!
*:&! *!
*!
+:&! +!
+:&!
#)+! #*,! #*%! #*'! #*)! #*+! #,! #%! #'! #)! +! )! '! %! ,! *+! *)! *'! *%! *,! )+!
+!
#+:&! #*!
!8#"$%&'()*+,-./#%$*0)&"/-*1!+02*
!
.;!.56!712340! 3/$7&'34'CEF'1-+%&1':$9'%&.377&08+2"30'%&>"1"301'
!
'!
):&!
(!
)!
)!
*:&!
*!
*!
+!
+:&!
#*!
+!
#)!
!"#"$%&'()*+,-./#%$*0)&"/-*1!+02*
(!
#)+! #*,! #*%! #*'! #*)! #*+! #,! #%! #'! #)! +! )! '! %! ,! *+! *)! *'! *%! *,! )+!
+,-./#%$*7.$"#)*
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!!
.;!.86!712340! chi2
-0.56
0.0729
0.73
0.5471
0.1391
1.42
-0.1095
-1.13
0.1949
0.79
-0.0317
-0.42
0.1553
1.51
0.3738
0.68
-0.0034
-0.03
0.0989
0.83
0.8121
2.12
-0.1164
-1.14
0.0395
0.33
0.1187
-2.73
0.0250**
Mean (%)
t-Stat
Mean (%)
t-Stat
Mean (%)
-10
0.0425
0.79
0.0853
1.21
-0.0695
-9
0.0192
0.42
0.0078
0.13
-8
0.0444
1.02
0.0441
-7
0.0412
0.84
0.0419
-6
0.0464
0.97
!!!"#$!!
-0.0367
-4
0.0023 -!!!"#$
!!
-!!!"#$
!!!
-1
-!!!!"#
!!!
0
-!!!"#"!!! !
-3 -2
-0.69
0.0104
0.16
0.0841
0.66
-0.3300
0.04
0.0512
0.85
0.0397
0.32
-0.1854
-2.37
-0.7883
-1.24
-!!!"#$
!!!
-6.81
-!!!"#!
!!!
-11.09
-!!!"#"!!!
-4.66
!!!
Difference test
Mean (%)
t
-5
AScX
-0.0529
-0.47
***
-1.50
0.2184
!!
-2.52
0.1760
-0.5774
***
-3.65
0.1195
***
-4.34
0.0628*
-!!!"#$
-!!!"#$
!
-4.26
-!!!"#$
!!!
-3.22
-0.7628
-7.92
-!!!"#$!!!
-5.39
-0.9091***
-5.75
0.1149
-4.00
-!!!"#!
!!!
-4.95
-0.5495
***
-4.71
0.0090***
-2.80
-1.94
+1
-!!!"#$
!!!
-7.74
-!!!!!"
+2
-!!!""#!!!
-4.16
-0.0149
-0.25
-!!!"#$!!!
-5.02
-0.3769***
-2.75
0.0001***
+3
-!!!"#!!!
-2.12
-0.0607
-1.05
-0.1204
-1.32
-0.1571
-1.43
0.6891
+4
-0.0396
-0.88
-0.2031
-0.34
-0.9367
-1.06
-0.2123
-0.20
0.7749
-2.86
0.0282**
+5
-0.0641
-1.42
0.0081
0.14
-0.0136
-0.15
+6
-0.0110
-0.26
0.1413
0.25
-0.0760
-0.91
0.0056
0.05
0.6616
+7
0.0245
0.59
0.0484
0.85
0.0058
0.07
-0.0172
-0.19
0.8063
+8
-!!!""#!
-1.84
-0.0515
-0.92
-0.0695
-0.83
-0.1624
-1.64
0.6207
-1.96
0.0120
0.15
0.0613
0.61
0.2266
0.68
-0.1169
-1.30
0.1203
1.13
0.1925
+9
-0.0442
-1.06
-!!!"#"
+10
0.0127
0.30
0.0362
N=2132
!
N=1160
!!
N=556
-0.3332
***
N=416
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 5 Cumulative Average Abnormal Returns T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The differences between the coefficients estimators of the AEX, AMX and AScX are calculated by using the Hausman test.
>?@32A40! All shares t! (-20, -11)
Mean (%)! -0.3083**
t-Stat! -2.18
(-10, -1)
0.7418***
4.61
1.1177
***
(0, +10)
1.0010
***
(+11, +20)
AEX !
Mean (%)! -0.1666
t-Stat! -0.90
0.6065***
3.05
0.9574
***
6.43
0.7484
***
-0.2846**
-2.15
(+21, +30)
-0.0402
(+31, +40)
0.0060
(-1, +1)
**
AMX !
AScX
Difference test
Mean (%)! -0.2692
t-Stat! -0.66
! !
p->chi2 0.3827
1.43
1.3202***
3.00
!
0.2614
7.05
1.3516
***
5.09
!
0.1874
3.30
1.8870
***
4.41
!
0.0540*
Mean (%)! -0.5974**
! -2.38
0.4654
!
1.2906
***
3.77
0.9861
***
-0.6256***
-3.84
-0.0026
-0.01
0.3671
0.99
!
0.0153**
-0.31
-0.0466
-0.29
0.3570
1.36
-0.7338**
-2.13
!
0.0413**
0.04
0.1217
0.74
-0.0085
-0.03
-0.3616
-0.80
!
0.5879
-2.11
-0.2954
-1.64
-0.1878
-0.73
-0.4543
-1.22
!
0.8379
0.0406
0.08
!
0.9497
11.68
7.83
(+41, +50)
-0.2898
(+51, +60)
-0.0628
-0.45
-0.0613
-0.36
-0.1230
-0.50
(+61, +70)
-0.0425
-0.28
0.1536
0.81
-0.4744
-1.59
0.0690
0.18
!
0.2016
(+71, +80)
-0.2128
-1.57
0.1006
0.62
-0.7374***
-2.78
-0.3338
-0.82
!
0.0239**
(+81, +90)
-0.0721
-0.56
0.0644
0.42
-0.4600*
-1.68
0.1621
0.46
!
0.2066
-1.82
-0.4898
*
-1.82
-0.5014
-1.31
!
0.8051
*
-2.79
!
0.0312**
-0.3828
***
(+101, +110)
-0.3463
**
(+111, +120) (-5, +120)
(+91, +100)
-2.87
-0.3018
*
-2.52
-0.0472
-0.29
-0.4962
0.1390
1.03
-0.0447
-0.28
0.0621
0.09
0.3477
0.36
N=1842
***
-1.74
-1.9035
0.4233
1.48
0.2518
0.69
!
0.3213
-0.6646
-0.52
0.3998
0.22
!
0.8033
N=1010
N=535
!
N=297
Downgrades (-20, -11)
0.4623
***
(-10, -1)
-0.7166
(-1, +1)
***
3.34
!!!"#$
!!
2.55
0.4983* *
1.29
!
0.9796
-1.95
-1.9463
***
-4.38
!
0.0027***
1.74
0.4116
-3.90
-0.2468
-1.09
-0.7770
-1.4724***
-13.72
-1.1136***
-9.07
-1.6608***
-7.20
-2.2213***
-7.48
!
0.0008***
(0, +10)
-1.5260***
-9.69
-0.9428***
-5.08
-2.1340***
-6.71
-2.3394***
-5.25
!
0.0004***
(+11, +20)
0.0961
0.71
0.1224
0.68
0.3569
1.36
-0.3259
-0.98
!
0.2697
-1.82
!
0.0424***
(+21, +30)
0.1129
0.79
0.3519
(+31, +40)
-0.0722
-0.55
(+41, +50)
0.1748
(+51, +60) (+61, +70)
*
1.89
0.1494
0.50
-0.6022
-0.0853
-0.50
0.1357
0.53
-0.3137
-0.96
!
0.5492
1.25
0.1639
0.91
0.4699
1.64
-0.1895
-0.56
!
0.3281
-0.1404
-1.00
0.0641
0.36
-0.4083
-1.28
-0.3530
-1.13
!
0.2938
0.1866
1.25
0.2553
1.38
0.2903
0.87
-0.1433
-0.41
!
0.5753
-2.18
!
0.0143**
1.46
!
0.5739
(+71, +80)
-0.0047
-0.03
-0.0329
-0.21
0.6194
(+81, +90)
0.2725*
1.93
0.2809
1.61
(+91, +100)
0.2397
*
1.74
0.2383
(+101, +110)
0.2591*
1.89
0.3342*
(+111, +120)
0.1519
1.10
0.2355
(-5, +120)
-1.1600
*
-1.69
N=2132
!
*
0.4328 N=1160
*
**
1.93
-0.7605
0.0518
0.18
0.5442
1.35
0.3674
1.31
0.0729
0.21
!
0.8030
1.93
0.4022
1.45
-0.1417
-0.40
!
0.4214
1.36
0.2750
0.93
-0.2456
-0.74
!
0.3958
-4.76
!
0.0001***
!
!
0.48
-0.1128 N=556
-0.08
-7.0004
***
N=416
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 6 Determinants of Stock Price Performance of Upgrades The intercept is the constant of the regression. The variable COVER is the number of analysts that covers a company. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock, multiplied the trading volume of that day. STRONG equals one when the recommendation is revision a strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 is omitted to prevent multicollinearity. YEAR2006 to YEAR2011 are year dummies, which equals one when the date of the recommendation revisions equals the corresponding year dummy and it is zero otherwise. T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The differences between the coefficients estimators of the AEX, AMX and AScX are calculated by using the Hausman test.
All shares
AEX
AMX
t-Stat
p->chi2
2.49
-3.1929*
-1.87
0.0105**
-0.0810**
-2.23
0.2153**
2.21
0.0230**
-2.5047
-0.79
-4.3608
-1.52
0.0395**
Mean (%)
t-Stat
Mean (%)
Intercept
1.2700***
2.68
Panel A: !"#!!!"!!!! 1.5399** 2.38 2.8026**
COVER
-0.0164*
-1.92
-0.0145
-1.15
ASKBID
-1.9683*
-1.80
-2.0185*
-1.73
STRONG
-
-
-0.8825**
SKIPRANK
1.2046
YEAR2006
***
-2.35
-
-
-0.3935
Difference test
Mean (%)
Variable
LOGTURN
t-Stat
AScX
Mean (%)
-
-
-0.84
-1.9799*** ***
3.23
0.3203
0.67
2.0841
-0.0890
-0.20
-0.0171
-0.04
YEAR2007
0.6271
1.33
0.4971
YEAR2008
-0.2116
-0.32
YEAR2009
1.0234
YEAR2010 YEAR2011
-2.61
-
-
-0.9821
-
-0.93
0.0207**
2.80
0.0031***
***
2.93
2.9963
-0.9208
-1.04
1.2045
0.94
0.0513*
0.95
0.2950
0.32
1.9834
1.39
0.0569*
-0.5373
-0.69
-0.1823
-0.12
1.2017
0.73
0.0554*
1.51
0.8608
1.03
0.1503
0.12
3.4160
1.58
0.0556*
-0.9626**
-2.18
-1.2816***
-2.89
-0.9253
-0.96
1.3403
0.81
0.0449**
-1.2223***
-2.85
-1.0920**
-2.32
-2.0070**
-2.29
-0.6875
-0.54
0.0337**
Prob > F
0.0000
0.0639
0.0212
0.0507
R-Squared
0.0207
0.0173
0.0390
0.0612
Panel B: !"#!!!!!!! Intercept
1.2507***
COVER
***
ASKBID LOGTURN STRONG
-0.1670 -0.6973
3.99 -3.23 -0.44
-
-
-0.5286
***
-2.63
1.3205*** -0.1733 -0.2359 -0.8508
1.47
0.5728
YEAR2006
0.0255
0.10
0.2872
1.14
1.6051
***
1.6300
***
YEAR2010
0.5571
**
YEAR2011
0.7411
YEAR2008 YEAR2009
3.96 4.32
2.06
0.6840
0.61
0.8270
-1.99
-0.0061
-0.28
-0.0040
-0.08
0.8236
-3.20
2.8280
0.51
0.0515*
-
0.3018 0.3111
1.4856**
-0.31
-
SKIPRANK YEAR2007
**
3.00
***
**
0.5026
*
2.0081
***
1.4099
***
-5.8875
***
-
-
-
-
-0.4217
-0.97
0.1841
0.33
0.4873
2.27
-0.2001
-0.47
0.0705
0.13
0.5120
0.98
-0.4114
-0.74
-0.0358
-0.05
0.7922
1.67
0.5308
3.71 2.96
0.83
-0.7799
-1.08
0.4926
1.2174
**
2.02
0.9049
0.66
0.7919
2.1526
***
3.01
1.14314
1.20
0.6976
**
1.99
1.3474
1.44
0.2832
0.48
0.6845
0.84
0.7763
2.06
0.4327
0.15
1.2153
0.28
-0.1751
-0.61
0.2886
Prob > F
0.0000
0.0000
0.0003
0.4697
R-Squared
0.0329
0.0474
0.0598
0.0329
!
-
-3.48
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 6 – Continued All shares Variable
Mean (%)
AEX
t-Stat
Mean (%)
AMX t-Stat
AScX
Mean (%)
Difference test
Mean (%)
t-Stat
!
p->chi2
Panel A: !"#!!!"!!!! ! Intercept
0.7607
1.62
COVER
-0.0254
ASKBID
-1.2285
LOGTURN
***
-3.06
-0.0240
-0.68
-0.5274
-
-
1.51
-1.1323
-0.99
2.3830
1.27
!
0.1599
-1.80
0.0146
0.44
-0.0360
-0.39
!
0.2599
-0.36
-3.5586
-1.24
-2.5393
-0.38
!
0.1554
-
!
-
1.0040 *
-
-
-
-
-
STRONG
0.2080
0.65
0.0875
0.22
0.3734
0.54
0.0464
0.06
!
0.4395
SKIPRANK
0.4380
1.34
0.3599
0.87
0.7657
1.14
0.1217
0.14
!
0.4462
YEAR2006
0.2968
0.71
0.4540
0.99
0.1769
0.20
0.1331
0.12
!
0.3048
1.82
0.8196
0.82
-1.5646
-1.39
!
0.1182
3.37
0.9442
0.41
!
0.1734
1.80
!
0.2633
YEAR2007
0.4813
1.09
2.1586
***
2.1153
***
YEAR2010
0.8115
*
YEAR2011
0.0842
YEAR2008 YEAR2009
3.24 3.33
0.9461
*
1.3113
*
1.76
1.3430
1.55
4.5116
***
3.4819
***
3.00
2.7013
*
1.83
-0.0632
-0.05
!
0.3866
0.85
0.3398
0.22
!
0.3993
1.95
0.4639
0.99
1.8632
0.20
-0.2275
-0.50
0.7593
*
Prob > F
0.0000
0.2690
0.0001
0.5075
!
R-Squared
0.0231
0.0121
0.0650
0.0315
!
Panel B: !"#!!!!!!! Intercept
0.5362
0.27
-0.0972
-0.03
3.3440
0.73
-2.8943
COVER
0.0021
-0.05
0.0276
0.48
-0.1965
-1.34
0.4728
ASKBID LOGTURN STRONG SKIPRANK YEAR2006
-4.3889
-0.87
-
-
-2.6337 4.2275
*
-1.68
***
-3.4341
**
2.64 -2.17
6.8677
1.28
-
-
-3.9636
*
-1.88
3.0510 -1.8587
1.35 -1.06
-17.554
-1.39
-
-
!
0.6844
-2.13
-1.3490
-0.32
!
0.6950
-2.33
!
0.1135
0.29
!
0.8094
3.68
!
0.0531*
-1.55
!
0.3042
-3.88
!
0.0194**
0.4722
0.17
2.3361
0.68
-3.7818
-0.64
1.8728
YEAR2011
-8.9026
***
0.01
0.7171
0.41
1.9871
-5.14
-5.7116
-0.03
-10.39
-
1.02
YEAR2008
***
!
3.9061 -12.213
0.0141
-
2.36 0.60
YEAR2010
0.0577*
**
2.1926 11.84
!
0.7448
0.07 0.27
-1.99
!
0.1697 1.2311
0.2203
0.26
-6.6889
2.32
!
1.0140
-0.95
7.2379
1.31
-1.03
-1.8250
YEAR2009
0.6977
-
YEAR2007
**
!
**
-2.9873 6.8378
-28.610
**
-0.45
***
***
2.59
23.827
0.45
-7.7562
-2.75
-24.14
**
***
Prob > F
0.0000
0.3735
0.0000
0.0000
!
R-Squared
0.0315
0.0266
0.0781
0.1480
!
N=1842
N=1010
N=535
N=297
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 7 Determinants of Stock Price Performance of Downgrades The intercept is the constant of the regression. The variable COVER is the number of analysts that covers a company. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock, multiplied the trading volume of that day. STRONG equals one when the recommendation is a revision to strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 is omitted to prevent multicollinearity. YEAR2006 to YEAR2011 are year dummies, which equals one when the date of the recommendation revision equals the corresponding year dummy and it is zero otherwise. T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The differences between the coefficients estimators of the AEX, AMX and AScX are calculated by using the Hausman test.
All shares
AEX
AMX
Variable
Mean (%)
t-Stat
Mean (%)
t-Stat
Intercept
-1.6394***
-3.18
Panel A: !"#!!!"!!!! ! -1.1971 -1.23 -2.0423*
COVER
0.0407
***
ASKBID
0.2031
AScX Mean (%)
t-Stat
!
p->chi2
-1.81
-1.2448
-0.66
!
0.8384
2.39
-0.0570
-0.54
!
0.2601
-0.04
1.9216
0.66
!
0.8812
-
Mean (%)
**
3.83
0.0253
1.20
0.1044
0.11
2.666
1.14
-0.2098
Difference test
!
-
STRONG
-0.5717
-0.92
0.6257
0.79
-0.3124
-0.27
-3.5287**
-2.25
!
0.1895
SKIPRANK
-0.0314
-0.08
0.0695
0.15
-0.1758
-0.22
-0.1403
-0.15
!
0.9730
-1.67
-0.8262
-0.90
0.6243
0.47
!
0.6970
0.96
1.3069
0.99
1.3410
1.06
!
0.9561
-2.65
-0.4276
-0.28
!
0.1147
1.74
!
0.5541
LOGTURN
YEAR2006 YEAR2007 YEAR2008
-
-
-0.6555 1.0972
*
-0.9481
YEAR2009
0.9028
YEAR2010
-0.0900
YEAR2011
-2.5484
-
-1.27
-1.2042
1.89
0.7555
-1.48
***
-
0.4614
0.59
1.40
0.2587
-0.20
-0.4518
-4.08
*
-2.4532
***
-
-
-3.8685
0.39
0.8748
-0.98
0.9303
-3.80
-2.2330
***
*
-
*
0.59
3.9772
0.84
-0.0922
-0.07
!
0.8334
-1.73
-3.6691
-1.60
!
0.9623
Prob > F
0.0000
0.0100
0.0045
0.0103
!
R-Squared
0.0228
0.0199
0.0453
0.0554
!
Panel B: !"#!!!!!!! Intercept
-1.9413***
COVER
***
ASKBID LOGTURN
0.0333
-2.1388
-
-0.3224
SKIPRANK
-0.1919
YEAR2006
-0.6037
YEAR2007
0.2743
YEAR2009
5.15 -0.78
-
STRONG
YEAR2008
-5.90
**
-1.0990
***
-0.8675
**
-1.3072*** 0.0199
**
0.7384
-0.83
0.2117
-2.04
-0.8064
0.82
0.0537
***
-1.8909*** 0.0413
*
1.6081
-2.61
-2.3322**
-2.09
!
0.6148
1.67
0.3586
0.67
!
0.8280
2.22
!
0.7185
-
0.69
3.6850
**
!
-
-0.97
0.2254
0.26
-1.0370
-0.89
!
0.7767
0.84
-0.5934
-1.29
-0.7154
-1.21
!
0.2131
-2.76
0.0056
0.01
-1.0069
-1.05
!
0.5621
-
-0.4500
-2.34
2.03 0.39
-
-0.78
-2.67
-2.69
-
-
-
0.13
0.3348
0.49
0.5769
0.67
!
0.8943
-1.3577
***
-2.89
-0.9384
-1.06
-0.5529
-0.50
!
0.8712
-0.8267
**
-2.20
-1.2900
-1.36
-0.5412
-0.45
!
0.8612
***
-2.68
-0.1693
-0.25
0.8131
0.91
!
0.4362
-2.96
-1.3673*
-1.92
-1.8482
-1.10
!
0.8264
YEAR2010
-0.4277
-1.46
-0.8111
YEAR2011
-1.2768***
-5.90
-1.1531***
Prob > F
0.0000
0.0325
0.1906
0.1986
!
R-Squared
0.0231
0.0169
0.0245
0.0326
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 7 – Continued All shares
AEX
AMX t-Stat
AScX
Difference test
Mean (%)
t-Stat
!
p->chi2
-0.62
-3.0370**
-2.22
!
0.2498
-0.0094
-0.29
0.0771
0.87
!
0.3692
-2.5312
-0.76
3.3265
1.21
!
0.3930
-
Variable
Mean (%)
t-Stat
Mean (%)
Mean (%)
Intercept
-1.6019***
-3.52
Panel A: !"#!!!"!!!! ! -0.6139 -0.90 -0.6011
COVER
0.0334
***
3.50
0.0160
1.22
ASKBID
-0.2186
-0.12
-1.0150
-0.46
!
-
STRONG
0.1527
0.25
1.4080*
1.88
0.4741
0.43
-3.1405**
-2.02
!
0.0340**
SKIPRANK
-0.6980**
-2.19
-0.6730*
-1.74
-1.1328*
-1.83
-0.6349
-0.72
!
0.5143
-2.61
-0.9479
-1.15
-0.0655
-0.05
!
0.5039
-1.51
-0.9258
-1.12
-0.4734
-0.51
!
0.5921
-0.41
-1.0880
-0.87
1.3474
0.91
!
0.4100
1.69
2.2081
1.13
!
0.0248**
-0.69
1.6644
1.45
!
0.1917
-1.97
!
0.0519*
LOGTURN
-
-
-
-
-0.9428
**
-2.24
-1.2306
YEAR2007
-0.8412
**
-2.03
-0.8632
YEAR2008
-0.3130
-0.54
-0.2676
YEAR2006
***
**
-2.10
1.7403
-2.48
-0.7386
YEAR2009
-0.1559
-0.28
-1.4273
YEAR2010
-0.6644
-1.55
-1.1836**
YEAR2011
-2.0430
***
-3.55
-0.9566
-
*
-1.94
*
-3.1480
***
-
-
-3.28
-4.9251
**
Prob > F
0.0002
0.1184
0.0537
0.0124
!
R-Squared
0.0157
0.0133
0.0324
0.0541
!
Panel B: !"#!!!!!!! Intercept
1.0575
0.54
6.9832**
2.27
4.0173
0.88
-8.3036
-1.53
!
0.0452**
COVER
-0.0338
-0.87
-0.1276**
-2.08
-0.2512
-1.62
0.2530
0.81
!
0.0442**
-4.03
***
-3.20
**
-2.41
-14.227
-1.11
!
0.1419
!
-
ASKBID LOGTURN
-31.83
***
-
-
-30.95 -
-
-30.71 -
-
-
-
*
1.66
-3.4803
-0.75
!
0.0562*
STRONG
2.6825
1.11
3.5235
1.06
7.8972
SKIPRANK
-2.0517
-1.44
-2.5379
-1.26
-5.5255*
-1.93
1.1313
0.41
!
0.0811*
YEAR2006
-1.4212
-0.81
-0.6408
-0.32
-2.3551
-0.63
0.0213
0.00
!
0.1597
-2.78
!
0.0004***
1.59
!
0.1329
4.09
!
0.0000***
0.23
!
0.1123
-2.46
!
0.0013***
**
YEAR2007
-4.0057
YEAR2008
6.021*** 12.07
YEAR2010
0.3295
YEAR2011
-2.10
-1.4351
-0.54
1.0347
0.28
-11.60
3.01
5.250***
2.23
9.0981**
1.98
7.3947
***
***
YEAR2009
-7.274
***
***
3.79
2.7434
0.71
32.52
0.20
-0.4668
-0.27
2.4970
-3.55
-3.0962
-1.46
-12.136
**
***
5.19
24.58
0.60
1.0267
-2.46
-15.47
**
Prob > F
0.0000
0.0706
0.0000
0.1494
!
R-Squared
0.0470
0.0148
0.1433
0.1494
!
N=2132
N=1160
N=556
N=416
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 8 Determinants of Stock Price Performance of Upgrades The intercept is the constant of the regression. The variable COVER is the number of analysts that covers a company. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock, multiplied the trading volume of that day. STRONG equals one when the recommendation is a revision to strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 is omitted to prevent multicollinearity. YEAR2006 to YEAR2011 are year dummies, which equals one when the date of the recommendation revision equals the corresponding year dummy and it is zero otherwise. T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The differences between the coefficients estimators of the AEX, AMX and AScX are calculated by using the Hausman test.
All shares
AEX
AMX
Variable
Mean (%)
t-Stat
Mean (%)
Intercept
1.2004
1.31
Panel A: !"#!!!"!!!! ! 3.3739 1.47 -6.0205*
COVER
-
-
-
t-Stat
*
AScX
Mean (%)
-
-1.77 -
Difference test
Mean (%)
t-Stat
!
p->chi2
-0.9765
-0.35
!
0.0012***
-
!
-
-
-1.75
0.1909
0.05
-4.5235
-1.59
!
0.0038***
-0.2196
-1.07
0.7902**
2.01
0.1627
0.44
!
0.0041***
-2.28
-0.4119
-0.88
-1.9001**
-2.52
-0.9865
-0.92
!
0.0062***
1.2197***
3.27
0.3184
0.66
2.2758***
3.22
2.7837***
2.61
!
0.0016***
YEAR2006
-0.0518
-0.12
0.0472
0.10
-1.1110
-1.20
0.9712
0.75
!
0.0026***
YEAR2007
0.6589
1.39
0.5827
1.09
-0.1634
-0.16
1.4506
0.99
!
0.0077***
YEAR2008
-0.2306
-0.35
-0.5291
-0.68
-0.2067
-0.14
0.3077
0.19
!
0.0090***
YEAR2009
1.0157
1.50
0.7880
0.94
0.6991
0.53
2.8136
1.32
!
0.0045***
YEAR2010
-1.0069**
-2.28
-1.3556***
-3.09
-0.6990
-0.72
0.4605
0.28
!
0.0013***
YEAR2011
-1.2383***
-2.90
-1.1469**
-2.43
-1.7157*
-1.92
-1.2808
-0.97
!
0.0079***
ASKBID
-1.6202
-1.47
-2.0132
LOGTURN
-0.0509
-0.61
STRONG
-0.8531**
SKIPRANK
Prob > F
0.0001
0.0012
0.0001
0.0591
!
R-Squared
0.0192
0.0175
0.0390
0.0474
!
Panel B: !"#!!!!!!! Intercept COVER
0.9839* -
1.73 -
2.1366 -
1.61
-5.5734**
-
-
-2.53 -
**
-1.7484
!
0.0072***
-
!
-
-0.94
-
-2.20
2.9113
0.53
!
0.0012***
0.7998***
3.19
0.3357
1.46
!
0.0033***
-3.51
-0.4411
-1.04
0.1557
0.28
!
0.0142**
0.5692**
2.25
-0.1184
-0.28
0.0484
0.09
!
0.0089***
0.24
0.3388
1.15
-0.6947
-1.26
0.0483
0.07
!
0.0087***
1.21
0.5555*
1.79
-0.1058
-0.18
-0.8949
-1.17
!
0.0114**
3.87
1.9704***
3.64
0.9912
1.62
0.8575
0.64
!
0.0278**
4.28
***
3.29
1.3420
1.44
!
0.0328**
1.65
!
0.0071***
1.06
!
0.0251**
ASKBID
-0.2635
-0.17
-0.1257
-0.17
-4.1731
LOGTURN
-0.0321
-0.65
-0.1398
-1.18
STRONG
-0.4925**
-2.45
-0.8575***
SKIPRANK
0.3195
1.55
YEAR2006
0.0627
YEAR2007
0.3319
YEAR2008
1.5761***
YEAR2009
1.6225
***
YEAR2010
0.5077
*
YEAR2011
0.0543
1.3545
2.86
2.2921
*** **
1.86
-0.0365
-0.13
1.2471
0.20
-0.2236
-0.78
0.2971
2.00
1.4805
0.50
0.8165
*
Prob > F
0.0000
0.0000
0.0899
0.2653
!
R-Squared
0.0282
0.0449
0.0899
0.0417
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 8 – Continued All shares
AEX
AMX
Variable
Mean (%)
t-Stat
Mean (%)
Intercept
1.6681*
1.71
Panel A: !"#!!!"!!!! ! 2.8949 1.36 -5.6936
COVER
-
-
t-Stat
-
-
AScX
Mean (%) -1.45
-
-
Difference test
Mean (%)
t-Stat
!
p->chi2
1.6834
0.55
!
0.1492
-
!
-
-
ASKBID
-1.0987
-0.61
-0.4341
-0.30
-2.5909
-0.88
-2.5011
-0.38
!
0.2559
LOGTURN
-0.1805**
-2.03
-0.2613
-1.38
0.5698
1.28
0.0195
0.05
!
0.1136
STRONG
0.2307
0.73
0.0698
0.18
0.3344
0.48
0.0433
0.05
!
0.4030
SKIPRANK
0.4490
1.38
0.3557
0.86
0.7958
1.18
0.1537
0.18
!
0.4070
YEAR2006
0.3582
0.86
0.5394
1.17
-0.0455
-0.05
0.1831
0.17
!
0.3816
1.98
0.3228
0.35
-1.4928
-1.31
!
0.0843*
**
YEAR2007
0.5925
1.33
1.0444
YEAR2008
2.1784***
3.24
1.2853*
1.75
4.3000***
3.18
1.0847
0.48
!
0.0539*
YEAR2009
2.1017***
3.30
1.2489
1.46
3.4772***
3.03
2.8278*
1.87
!
0.2122
1.79
0.1010
0.09
!
0.2483
0.79
0.4556
0.30
!
0.3376
YEAR2010
0.7650
YEAR2011
0.0774
*
1.84
0.3487
0.75
1.8364
0.19
-0.3041
-0.66
0.6938
*
Prob > F
0.0011
0.3454
0.0704
0.0311
!
R-Squared
0.0214
0.0111
0.0704
0.0311
!
Panel B: !"#!!!!!!! Intercept
7.9187*
1.91
14.289
1.48
35.208***
2.73
32.057**
2.33
!
0.3411
-
!
-
ASKBID
-7.3220
-1.33
5.5444
1.09
-23.601*
-1.73
-29.976*
-1.96
!
0.0669*
LOGTURN
-0.7471*
-1.95
-1.1789
-1.36
-4.2511***
-2.90
-3.904**
-2.08
!
0.0457**
STRONG
-2.7953*
-1.79
-4.1273*
-1.95
-2.5810
-0.91
1.3621
0.37
!
0.3633
2.38
3.7668
1.03
!
0.5318
-1.57
-2.8650
-0.64
!
0.5246
-2.23
!
0.0288**
COVER
SKIPRANK YEAR2006
-
-
4.1400
-
-
-
-
***
2.60
3.0704
1.36
6.7428
**
-2.14
-1.6534
-0.91
-4.9271
-3.4019
**
-
YEAR2007
-1.3901
-0.71
0.6845
0.28
6.0983
1.57
-11.812
YEAR2008
0.8282
0.30
2.9356
0.87
-1.9702
-0.35
0.7147
YEAR2009 YEAR2010
7.2249
**
0.1624
YEAR2011
-8.7731
Prob > F
0.0000
R-Squared
!
2.33 0.10
*
1.91
0.9583
0.21
0.7437 -5.8304
0.43 ***
-3.10
0.2799
12.359
***
2.4152 -9.5704
2.72 0.54
**
0.0000
-2.48
**
0.11
!
0.6180
***
3.02
!
0.0698*
-11.267
**
-2.30
!
0.0988*
-26.967
***
-4.23
!
0.0155**
20.146
!
0.0000
0.0287
0.0120
0.0919
0.1912
!
N=1842
N=1010
N=535
N=297
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 9 Determinants of Stock Price Performance of Downgrades The intercept is the constant of the regression. The variable COVER, is the number of analysts that covers a company. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock multiplied the trading volume of that day. STRONG equals one when the recommendation is a revision to strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 is omitted to prevent multicollinearity YEAR2006 to YEAR2011 are year dummies, which equals one when the date of the recommendation revision equals the corresponding year dummy and it is zero otherwise. T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The differences between the coefficients estimators of the AEX, AMX and AScX are calculated by using the Hausman test.
All shares
AEX
AMX
Variable
Mean (%)
t-Stat
Mean (%)
Intercept
-3.7891***
-3.69
Panel A: !"#!!!"!!!! ! -7.0019** -2.46 -11.017***
COVER
-
-
-
t-Stat
-
AScX
Mean (%)
-
-3.10 -
Difference test
Mean (%)
t-Stat
!
p->chi2
3.8414*
1.72
!
0.0002***
-
!
-
-
ASKBID
0.3478
0.18
3.2254
1.49
-0.5198
-0.10
0.7703
0.26
!
0.0006***
LOGTURN
0.3525***
3.77
0.6228**
2.44
1.2917***
3.23
-0.8103***
-2.78
!
0.0000***
STRONG
-0.4512
-0.72
0.5946
0.74
0.4125
0.35
-3.9495**
-2.52
!
0.0006***
SKIPRANK
-0.0896
-0.23
0.0871
0.18
-0.5419
-0.68
-0.1636
-0.18
!
0.0015***
YEAR2006
-0.7447
-1.43
-1.4587*
-1.92
-1.1643
-1.25
0.7815
0.58
!
0.0012***
YEAR2007
0.8952
1.57
0.2712
0.35
0.6703
0.56
1.7268
1.39
!
0.0013***
YEAR2008
-0.8503
-1.32
0.4441
0.57
-3.3776**
-2.36
-0.5986
-0.41
!
0.0004***
YEAR2009
1.1037
1.62
0.4681
0.70
0.9422
0.63
3.6143
1.59
!
0.0007***
YEAR2010
-0.0109
-0.02
-0.3955
-0.85
1.2059
1.05
-0.5474
-0.41
!
0.0007***
YEAR2011
-2.3322***
-3.78
-2.2308***
-3.56
-2.0502
-1.63
-4.3172*
-1.90
!
0.0022***
Prob > F
0.0000
0.0053
0.0007
0.0003
!
R-Squared
0.0231
0.0249
0.0519
0.0717
!
Panel B: !"#!!!!!!! Intercept COVER
-2.4380
***
-
-4.07 -
-1.6062
-0.95
-
-
-3.6221 -
-1.42 -
4.0486***
2.69
!
0.0071***
-
-
!
-
ASKBID
1.3557
1.31
0.2547
0.14
1.2926
0.59
2.3916
1.38
!
0.0251**
LOGTURN
0.1618***
2.88
0.1042
0.68
0.2993
1.00
-0.8093***
-3.79
!
0.0011***
STRONG
-0.2553
-0.61
-0.4194
-0.90
0.4289
0.50
-1.3927
-1.20
!
0.0356**
SKIPRANK
-0.2511
-1.14
0.1774
0.71
-0.7226
-1.55
-0.7896
-1.36
!
0.0083***
YEAR2006
-0.6335**
-2.11
-0.8347***
-2.72
-0.0318
-0.06
-0.8891
-0.05
!
0.0200**
YEAR2007
0.1714
0.52
-0.0453
-0.11
0.2354
0.36
0.8234
0.98
!
0.0254**
-2.77
-0.7594
-0.86
-0.9223
-0.87
!
0.0373**
YEAR2008
-1.0155
**
***
-2.46
-1.2952
YEAR2009
-0.7378**
-1.98
-0.7426*
-1.95
-1.3032
-1.36
-1.0178
-0.92
!
0.0348**
YEAR2010
-0.3524
-1.19
-0.7734**
-2.55
-0.1091
-0.16
0.0959
0.12
!
0.0134**
YEAR2011
-1.1033***
-2.77
-1.0354***
-2.71
-1.2859*
-1.81
-2.7678*
-1.65
!
0.0382**
Prob > F
0.0030
0.1091
0.2562
0.0015
!
R-Squared
0.0146
0.0135
0.0224
0.0679
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 9 – Continued All shares
AEX
AMX
Variable
Mean (%)
t-Stat
Mean (%)
Intercept
-2.4608***
-2.65
Panel A: !"#!!!"!!!! ! 2.4220 1.04 -1.4293
COVER
-
-
t-Stat
-
-
AScX
Mean (%) -0.42
-
-
Difference test
Mean (%)
t-Stat
!
p->chi2
0.8404
0.40
!
0.6391
-
!
-
-
ASKBID
-0.8476
-0.46
-2.1159
0.99
-2.3298
-0.70
2.2514
0.78
!
0.5996
LOGTURN
0.1932**
2.23
-0.2119
-1.01
0.0742
0.18
-0.3834
-1.35
!
0.0985*
STRONG
0.2201
0.37
1.4751**
1.99
0.4838
0.43
-3.3427**
-2.13
!
0.0632*
SKIPRANK
-0.7569**
-2.39
-0.7378*
-1.92
-1.1139*
-1.77
-0.6646
-0.75
!
0.8987
-2.33
-1.0042
-1.18
-0.0154
-0.01
!
0.7952
-1.26
-1.0056
-1.14
-0.4070
-0.43
!
0.9060
YEAR2006
0.9825
YEAR2007 YEAR2008
**
-2.31
-1.1214
-0.2343
-0.41
-0.7269
-0.0313
-0.06
-0.1586
**
-0.24
-1.1193
-0.91
1.0162
0.67
!
0.7384
**
-2.11
1.7700
1.72
1.9194
0.99
!
0.0660*
-2.43
-0.7195
-0.67
1.1682
1.01
!
0.3371
-2.14
!
0.0728*
YEAR2009
-0.5927
-1.38
-1.4212
YEAR2010
-0.5927
-1.38
-1.1582**
YEAR2011
-1.8699
***
-3.31
-0.9053
*
-1.86
-3.1729
***
-3.28
-5.5250
**
Prob > F
0.0036
0.1172
0.0541
0.0093
!
R-Squared
0.0126
0.0133
0.0323
0.0562
!
Panel B: !"#!!!!!!! Intercept COVER
5.4491 ***
1.45
34.631***
3.15
43.238***
3.27
-
-
-
-
-
-4.30
-33.425
***
-3.49
-31.827
-3.37
-2.1097*
-1.83
!
0.0004***
1.11
5.4335
1.13
-4.5586
-0.99
!
0.0622*
-2.6089
-1.29
-4.4914*
-1.61
0.9840
0.36
!
0.1265
0.5768
0.28
-0.6088
-0.16
0.3293
0.07
!
0.2124
-2.56
!
0.0147**
1.31
!
0.1982
3.93
!
0.0000***
-0.28
!
0.1628
-3.10
!
0.0036***
-3.05
STRONG
2.4126
1.00
3.6596
SKIPRANK
-2.0091
-1.42
YEAR2006
-1.2489
-0.71
YEAR2009
11.993
YEAR2010
0.2701
YEAR2011
-7.4798
***
-1.92
0.8991
0.33
4.0430
2.96
5.3085**
2.31
7.7666*
3.97
1.7191
0.16
-0.7483
-3.67
-
-5.161***
-2.9933***
***
!
0.2006
-1.58
5.8895***
-
!
-0.5496
YEAR2008
0.0574*
-1.38
LOGTURN
-3.6892
!
-19.126
-34.729
YEAR2007
-
1.26
-2.69
ASKBID
*
***
10.609
-4.1950
**
0.46
31.971
-0.44
1.361
-2.04
***
-12.484
**
1.03
-11.056
1.77
5.9942 ***
5.19
23.175
0.34
-1.2471
-2.51
**
-18.295
***
Prob > F
0.0000
0.0047
0.0000
0.1592
!
R-Squared
0.0481
0.0218
0.1590
0.1592
!
N=2132
N=1160
N=556
N=416
!
!
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 10 Correlation matrix of the explanatory variables The correlation matrix shows the level of correlation between the explanatory variables that are used in the regressions. The variable COVER is the number of analysts that covers a company. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock, multiplied the trading volume of that day. STRONG equals one when the recommendation is a strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 to YEAR2011 are year dummies, which equals one when the
COVER
1.00
ASKBID
-0.26
1.00
LOGTURN
0.79
-0.26
1.00
STRONG
-0.03
0.00
-0.04
1.00
SKIPRANK
-0.07
0.02
-0.07
0.34
1.00
YEAR2005
-0.03
0.00
-0.03
-0.03
-0.03
1.00
YEAR2006
-0.05
0.06
0.00
-0.02
-0.05
-0.18
1.00
YEAR2007
-0.07
0.06
0.03
0.01
0.02
-0.17
-0.18
1.00
YEAR2008
0.00
0.04
0.00
0.01
0.03
-0.17
-0.18
-0.17
1.00
YEAR2009
0.05
-0.04
0.00
0.05
0.04
-0.18
-0.19
-0.18
-0.18
1.00
YEAR2010
0.04
-0.06
0.00
-0.00
0.03
-0.16
-0.16
-0.16
-0.15
-0.16
1.00
YEAR2011
0.07
-0.07
0.01
-0.02
0.03
-0.16
-0.16
-0.16
-0.15
-0.16
-0.14
!
YEAR2011
YEAR2010
YEAR2009
YEAR2008
YEAR2007
YEAR2006
YEAR2005
SKIPRANK
STRONG
LOGTURN
ASKBID
COVER
date of the recommendation equals the corresponding year dummy and it is zero otherwise.
1.00
! ! ! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""!! Table 11 Determinants of Stock Price Performance All Shares The intercept is the constant of the regression. The variable CAP, is the log of the market capitalization of the companies. ASKBID is the spread between the ask and the bid price. LOGTURN is the log of the turnover, which is the closing price of a stock, multiplied the trading volume of that day. STRONG equals one when the recommendation is a strong buy or strong sell and zero otherwise. SKIPRANK equals one when the change in recommendation skipped a rank and zero otherwise. YEAR2005 is omitted to prevent multicollinearity. YEAR2006 to YEAR2011 are year dummies, which equals one when the date of the recommendation equals the corresponding year dummy and it is zero otherwise. T-statistics with an absolute value of 1.65, 1.96 and 2.58 indicate significance at the 0.10, 0.05 and 0.01 levels, respectively. *** indicates statistical significance at less than the 0.10 level, ** indicates statistical significance at less than the 0.05 level and * indicates statistical significance at less than the 0.01 level. The difference test is calculated by using the Hausman test.
!"#!!!"!!!! Variable
Mean (%)
t-Stat
!"#!!!!!!! Mean (%)
t-Stat
!"#!!!!!"!
!"#!!!!!!"#!
Mean (%)
t-Stat
Mean (%)
t-Stat
Buy Recommendations Intercept
2.2114
2.45
2.7716
5.11
3.1741
3.62
7.8977
2.11
CAP
-0.1786
-1.93
-0.2477
-4.57
-0.3888
-4.25
-0.8739
-2.18
ASKBID
-1.7792
-1.63
-0.6392
-0.41
-1.1648
-0.65
-6.098
-1.18
STRONG
-0.8883
-2.36
-0.5517
-2.75
0.1696
0.53
-2.8565
-1.83
SKIPRANK
1.1991
3.22
0.2862
1.40
0.4124
1.27
4.0990
2.57
YEAR2006
-0.5472
-0.12
0.0600
0.23
0.3493
0.84
-3.4347
-2.16
YEAR2007
0.6822
1.44
0.3867
1.40
0.6026
1.35
-1.5658
-0.81
YEAR2008
-0.2438
-0.37
1.5765
3.90
2.1159
3.19
0.5226
0.19
YEAR2009
0.9505
1.41
1.5314
4.06
1.9608
3.09
6.9137
2.22
YEAR2010
-0.9967
-2.26
0.5302
1.98
0.7720
1.86
0.1108
0.06
YEAR2011
-1.2353
-2.88
0.0652
0.25
0.0715
0.17
-8.8471
-5.04
R-Squared
0.0208
0.0378
0.0280
0.0288
Sell Recommendations Intercept
-5.8378
-5.41
-4.8678
-7.07
-3.0780
-3.07
9.1355
2.29
CAP
0.6579
5.67
0.4781
6.43
0.3063
2.73
-1.0904
-2.55
ASKBID
0.2440
0.14
1.9700
1.99
-0.9621
-0.56
-33.688
-4.35
STRONG
-0.4451
-0.72
-0.2252
-0.55
0.2314
0.39
2.5188
1.05
SKIPRANK
0.0511
0.13
-0.1336
-0.61
-0.7043
-2.22
-2.2928
-1.61
YEAR2006
-0.7389
-1.42
-0.6616
-2.25
-0.9715
-2.30
-1.2593
-0.71
YEAR2007
0.9420
1.63
0.1584
0.48
-0.9253
-2.25
-3.7759
-1.97
YEAR2008
-0.6242
-0.99
-0.8256
-2.10
-0.1198
-0.21
5.5828
2.80
YEAR2009
1.2270
1.93
-0.6144
-1.67
0.0624
0.11
11.689
3.90
YEAR2010
-0.0032
-0.01
-0.3543
-1.21
-0.5845
-1.37
0.2779
0.17
YEAR2011
-2.3581
-3.88
-1.1191
-2.87
-1.8798
-3.34
-7.4141
-3.59
R-Squared
!
0.0325
0.0354
0.0146
0.0499