Abnormal returns in the Dutch stock market following analyst recommendations
MSc Thesis Tim Jokhorst (s1686402)*
Double Degree MSc International Financial Management 2012/2013
Faculty of Economics & Business, Rijksuniversiteit Groningen
Department of Business Studies and Economics, Uppsala University
Supervisor: N. Brunia
---
Abstract
Analyst revisions (upgrades and downgrades) are followed by significant abnormal returns in the Dutch stock market. These (cumulative) abnormal returns are significantly influenced by stock size, an
analysts’ affiliation with investment banks, and the size of the brokerage house issuing the recommendation. In the Dutch stock market, the abnormal returns following downgrades are significantly bigger than upgrades. This study finds that since the enactment of the Global Settlement
regulations in 2003, the abnormal returns following analyst recommendations have significantly increased. Also, evidence shows that leading and following recommendations lead to significantly
higher abnormal returns induced by analyst recommendations.
---
Keywords: analyst recommendations, Dutch stock market, Global Settlement, event study, abnormal returns
*I would like to express my gratitude to my MSc thesis supervisor N. Brunia for his support and
1 TABLE OF CONTENTS
I . Introduction 2
II. Literature Review 4
Analytical Framework 4
Conflicts of Interest: Analyst Bias and the regulations of 2003 6
Empirical evidence of abnormal returns in the U.S. 7
Empirical evidence of abnormal returns in the rest of the world 9
Stock Characteristics 12
Analyst Characteristics 13
III. Data and Sample Description 16
Composition of the recommendations 18
Continuous variables 20
Dichotomous variables 20
IV. Methodology 20
Event study methodology 21
Test statistics 22
Regressing (cumulative) abnormal returns on stock and analyst characteristics 24
V. Results 27
Event study results 27
Regression results 35
VI. Conclusion 42
VII. References 43
VIII. Appendix 46
I. List of the brokerage houses included in this study 46
II. Discussion of t-tests 47
III. Descriptive Statistics Dependent Variables CAAR and AR for Downgrades
and Upgrades 48
IV. Variance Inflation Factors of the Independent Variables 48
V. Correlation Matrix of the Independent Variables 48
VI. Regression results for Upgrades, including Nationality 49
VII. Regression results for Downgrades, including Nationality 49
2
I. Introduction
The buy, hold, and sell recommendations of various international brokerage houses are of interest to all participants in the capital markets. However, the financial world is at odds on this very topic of the effect of analyst recommendations on stock prices. Financial theorists claim that, following the efficient market theory (Fama 1970), in a semi-strong efficient market all stock-relevant and publicly available information is already reflected in the stock price and prices instantly change to reflect new information. Therefore, investors trading on this new information cannot realize an abnormal return. This would render the activities of the many stock analysts useless. Cowles (1933) supports the claim that analyst recommendations have no value with his study of the forecasting abilities of analysts and their market effect. He found that the stock recommendations of the 16 financial services in his sample had an average effective annual rate of -1.43%. Nevertheless, a lot of money is spent on equity research and much attention is paid by many participants in the international capital markets to the opinions of these equity analysts and their recommendations. In contradiction to the findings of Cowles(1933),evidence of abnormal returns following analyst recommendations is found by Womack (1996), Jegadeesh & Kim (2006), and Stickel (1995). Grossman and Stiglitz (1980) argue that market prices cannot perfectly reflect all available public information due to the cost of information. Information is costly to acquire and if there would be no compensation for the analysts’ activities then why would so many brokerage houses invest large amounts of money in their research departments? If market prices reflect the public available information, there is no incentive to collect (costly) information. Only if prices do not fully reflect the information, there is an incentive to acquire that information. Thus, ‘because information is costly, prices cannot perfectly reflect the available information, if it did, those who spent resources to obtain it would receive no compensation’(Grossman and Stiglitz 1980, p 405) . Grossman and Stiglitz argue that in a competitive and rational market information gatherers (analysts) must be compensated for their activities. Also, investors would only be willing to pay for the analysts’ advice if it would be to their advantage, the expected marginal benefit being at least equal to its marginal cost. Womack (1996) argues that this expected benefit (or value) for an investor would be the abnormal returns following analyst recommendations. Being able to trade on the information of an analyst recommendation before it becomes publicly available allows an investor to obtain abnormal returns once the market prices to the newly available public information,
3 I focus on revisions of analyst recommendations. A revision is a recommendation that has been changed by the analyst in regard to the previous recommendation. Recommendations that are confirmed by the analyst and thus have not changed are referred to as reiterations and are not included in this study. A revision that has positively changed (ascended in rank) is called an upgrade and a negative revision is called a downgrade. For example, a recommendation that moves from a ‘buy’ to a ‘hold’ is labelled as a downgrade. The most common used recommendations are (in descending order of quality) strong buy, buy, hold, underperform, and sell. This approach is different from earlier studies on the abnormal returns following analyst recommendations such as Stickel(1995) who studies the five recommendation ranks separately without regard to the previous recommendation rank. I believe that studying the market effect of analyst recommendations based on the five ranks does not fully capture the market reaction to analyst recommendations. I furthermore believe that revisions contain more information compared to reiterations. Also, a part of the market reaction to analyst recommendations depends on the previous recommendation, e.g. an upgrade from a sell recommendation to a buy recommendation is more interesting and is likely to significantly affect the stock price. This approach is similar to Womack (1996), who studies changes to changes-from and changes-to the extremes ( added-to-buy, added-to-sell, removed-from-buy, and removed-from-sell) because these recommendation revisions would be among the most prominent news items and are most likely to be immediately conveyed to investors.
This study distinguishes itself from prior work due to its relatively large sample size, a long time frame including recent years and years of financial crisis, a focus on the Dutch stock market, and by the discovery of new significant variables in the cross-sectional analysis of the determinants of the abnormal returns following analyst recommendations. Also, the effect of the Global Settlement and its European equivalent Market Abuse Directive on the abnormal returns following analyst recommendations is studied. These regulations, imposed in 2003, sought to solve the conflict of interest among security analysts employed by investment banking firms. As academic literature on analyst recommendations is mainly focused on the U.S. market, research on the occurrence of abnormal returns following analyst recommendations in Europe is scarce. To my knowledge no published research has been done (to this extent) on the reaction to analyst recommendations in the Netherlands. Furthermore, this study covers a period of seventeen years, an unprecedented time span in this particular field of study. This time span allows me to investigate the effects of the regulations of 2003 (the Global Settlement and the Market Abuse Directive) on the composition of analyst recommendations and their subsequent abnormal returns.
4 upgrades and downgrades. For downgrades I find an average abnormal return of -0.38% on the event date whereas an average abnormal return of 0.35% is found for upgrades. Interestingly, the abnormal returns following recommendations have significantly increased since the implementation of the Global Settlement regulation (May 28 2003). This study also intends to provide an in-depth examination of the abnormal returns and focuses on the stock and analyst characteristics in order to determine their effect on the abnormal returns. I find that recommendations on stocks with a smaller market value are followed by significantly bigger abnormal returns. Furtermore, I find that the average abnormal returns following analyst recommendations have significantly after the fall of Lehmann Brothers in 2008. Leading recommendations that are followed by other analyst recommendations also lead to significantly bigger abnormal returns.The amount of analyst coverage and the size of the brokerage house of the analyst are significant contributors to the abnormal returns following recommendations.
II. Literature Review
In this section I will review the empirical evidence on the occurrence of abnormal returns following analyst recommendations. Thereafter, an overview of relevant stock- and analyst characteristics influencing the abnormal returns after stock recommendations is provided. First, I will present the analytical framework I use in this study. After that, I will shortly elaborate on the conflicts that analysts face and the regulations of 2003. Subsequently, I analyze the empirical evidence on analyst recommendations and the abnormal returns following these recommendations. Lastly, I discuss the stock and analyst characteristics that may influence the magnitude of abnormal returns following analyst recommendations
Analytical Framework
In this section, I describe the workings of analysts and their role in the capital markets. Sell-side analysts work for brokerage houses or firms that manage individual accounts. These sell-side analysts make stock recommendations in order to help their clients make profitable buy and sell trades. This activity is beneficial for the brokerage houses because it earns a commission on every trade that their clients make. Buy-side analysts, on the other hand, typically work for pension funds or mutual funds. After analyzing stocks, they make recommendations as well. These recommendations, however, are meant for the employer only and are not available for anyone except the fund that pays them. This study solely focuses on sell-side analysts. The analysts’ tasks can be described as 1) gathering new information on the industry or a specific stock; 2)analyzing these data and forming earnings estimates and recommendations; 3) presenting these recommendations, estimates and models to investors in presentations and written reports. (Michaely & Womack 1999). In effect, an analysts’
5 used recommendations are (in descending order of quality) strong buy, buy, hold, underperform, and sell. Credit Lyonnais, one of the brokerage houses included in this study, labels a stock with an expected return above 20% in the next 12 months as a ‘strong buy’, a stock with an expected return between 10% and 20% in the coming year is labelled a ‘buy’, and a ‘hold’ recommendation has an expected return between 0% and 10% for the next 12 months. A ‘underperform’ recommendation indicates a negative expected performance up to a maximum of 20%. Stocks expected to perform more poorly than minus 20% are labelled ‘sell’.
6 provide a direct test of market efficiency; abnormal returns which persist after an event like the
issuance of an analyst recommendation are inconsistent with the hypothesis that stock prices adjust quickly to fully reflect new information (Brown and Warner 1980).
Cowles (1933), employing a rather out-dated methodology, finds no evidence of stock-picking or forecasting abilities for analysts and concluded that analyst recommendations have no investment value. These findings support the efficient market hypothesis. Womack (1996) finds substantial and significant abnormal returns following buy and sell recommendations for both a short-term event window (immediate reaction) as well as a longer event window (several months after the
recommendation).These results are consistent with the expanded view of market efficiency by Grossman and Stiglitz(1980); there must be returns to information search costs. These information search costs are often assumed to be zero when considering the efficient market hypothesis. Stickel (1995) and Jegadeesh & Kim (2006) also find significant abnormal returns following analyst
recommendations. These findings imply the capital market to be semi-strong efficient. The empiricial evidence regarding abnormal returns following analyst recommendations will be discussed more thoroughly in the next section Firstly, the relation between analysts and investment banks is discussed in the following section.
Conflicts of Interests, Analyst bias and the regulations of 2003
7 In order to investigate the dubious practices of (some) analysts, Congress held hearings in 2001 which were titled ‘Analysing the analysts’. Six weeks before the bankruptcy of Enron, fifteen of the seventeen analysts covering Enron were still recommending Enron as a buy or even strong buy. The controversy went on in 2002, when internal mails from Merrill Lynch completely trashed certain stocks that were promoted by the firm to the public. Most illustrative is the internal memo from Morgan Stanley reading ‘’Our objective… is to adopt a policy, fully understood by the entire firm,
including the Research Department, that we do not make negative or controversial comments about our clients as a matter of sound business practise’’. On April 28 of 2003, the Global Settlement was
enforced to address the conflict of interest that analysts faced. Besides fining the ten largest investment banks in the U.S., the regulation also sought to limit the relation between investment banking departments and research departments in a firm by means of a Chinese wall (a clear and sometimes physical separation between investment banking and research departments). Also, more strict disclosure requirements were established in order to make the research output more meaningful and credible. The Market Abuse Directive, being the European equivalent of the Global Settlement, came into force in April 2003 as well. Kadan, Madureira, Wang, & Zach (2009) analyse the stock recommendations and price reactions before and after the regulations. They find that after the regulations the recommendations have become more balanced as more hold, underperform and sell recommendations are issued. It must be noted that this balancing coincided with a fair amount of banks reshuffling their rating system and moving from a five-rank system to a three rank system. Also, they find that the price reaction to positive recommendations has become stronger after the enforcement of the regulations. However, the results indicate that recommendations have become less informative (proxied by abnormal returns).
Empirical evidence of abnormal returns in the U.S.
8 Diefenbach (1972) also finds that analyst recommendations have no investment value. Covering a period from 1967-1969, he finds that, on average, the returns on the stocks recommended by analysts are not significantly higher than the market’s return.While concluding that analyst recommendations do not out- or underperform (in the case of a sell advice) the market, Diefenbach explains his results by pointing towards the random walk theory, which states that past movements of a stock cannot be used to predict its future movement, and the many linkages and affiliations between analysts and investments bankers.
However, more recent research does find investment value in analyst recommendations. Stickel (1995) finds that brokerage house buy and sell recommendation influence stock prices. After analysing 8.790 buy and 8.167 sell recommendations in the 1988-1991 period, Stickel (1995) finds that buy recommendations are associated with an average price increase of 1.16% over the next 11 market days following the record date of the recommendation. Sell recommendations lead to an average decrease in stock prices of 1.28%. Interestingly, on average the stock subject of a buy (sell) recommendation significantly outperforms (underperforms) the market during the ten-day run-up to the event date with 0.65% (1.06%). This is significant evidence of leakage or the anticipation of the abnormal return after recommendations. Please note that, in contrast to this study, Cowles(1933), Diefenbach (1972) and Stickel(1995) all study the performance of ‘Buy’ and ‘Sell’ analyst recommendations. This study focuses on revisions in the form of upgrades and downgrades as previously explained. For this reason, the evidence found in these studies is not fully comparable with the results found in this study. Nevertheless, the evidence found in these studies is interesting and relevant as it shows the development in research regarding analyst recommendations and provides contrasting evidence on the investment value of analyst recommendations.
9 Bjerring, Lakonishok and Vermaelen (1983) find that recommendations of a Canadian brokerage house lead to significant abnormal returns on the event date for both Canadian as U.S. stocks. A listing on the Recommend List (an upgrade) leads to a first week significant abnormal return of 1.8% for Canadian stocks. Interestingly, the price increase for a Recommended listing is only significant on the week following the listing and thus no evidence is found for a post-event drift nor a leakage effect. Nevertheless Bjerring et al (1983) conclude that there is investment value in analyst recommendations and that following the recommendations an investors would have achieved significantly positive abnormal returns, even after deducting transaction costs. Yet, Barber et al (2001) find that purchasing (selling short) stock with the most (least) favourable consensus recommendations together with daily portfolio rebalancing and a timely response to recommendations changes yields an annual gross return greater than 4%. However, this strategy requires high trading frequencies and when deducting the substantial transaction costs the abnormal returns are not significantly different from zero.
Empirical evidence in the rest of the world
11
Table 1:Evidence found on abnormal returns following analyst recommendations, as discussed in the Literature Review
Author Country Period Buy/Sell Upgrade/Down grade Pre-event Event Date Post-Event Note
Cowles(1933) U.S. 1928-1932 Buy N.A. N.A. -1.42 Annualized returns
Diefenbach (1972)
U.S. Buy N.A. N.A. -0.4% Annualized returns
Sell N.A. N.A. -12,6% Annualized returns
Bjerring et al(1983)
U.S 1977-1981 Buy 0.32% 1.49% 4.09% Pre-event(-7,-1),
Event date(0,+7) Post-event(0,+266)
Stickel(1995) U.S. 1988 - 1991 Buy 0.65% 0.90% 1.81% Pre-event (-10,-)1, event date( 0,+10)
Sell
-1.06%
-0.80% -1.34% Post-Event (0,+60)
Womack(1996) U.S. 1989-1991 Upgrades N.A. 2.98% 0.09% Event date (-1,+2)
Downgrades N.A. -4.69% -9.15% Post Event (0,+180)
Barber et al(2001)
U.S 1985-1996 Buy N.A. 0.76% 4.13% Annualized returns,
Event date(-1,+1)
Sell N.A 1.28% 4.91%
Jegadeesh & Kim (2006)
U.S 1993 - 2002 Upgrade N.A 1.76% 2.1% Post-Event (0,+1)
Downgrade N.A. -3.19% 3.38
Britain Upgrade N.A. 0.18% 0.28%
Downgrade N.A. -0.18% -0.23%
Canada Upgrade N.A. 0.4% 0.54%
Downgrade N.A. -0.45% -0.64%
France Upgrade N.A. 0.32% 0.38%
Downgrade N.A. -0.35% -0.48%
Germany Upgrade N.A. 0.16% 0.21%
Downgrade N.A. -0.22% -0.33
Italy Upgrade N.A. 0.04% 0.05%
Downgrade N.A. -0.09% -0.14%
Japan Upgrade N.A 0.46% 0.74%
12 Using an event study methodology, I test whether analyst recommendations in the Dutch market induce abnormal returns:
H1 Abnormal return Hypothesis Abnormal returns following analyst recommendations occur in
the Dutch market in the period 1995 to 2012
With the reasonable quantity of empirical research done on the phenomenon of recommendation-induced abnormal returns, it is surprising that research on the determinants of recommendation performance is scarce. I will add to this small amount of existing literature by adding variables that prove to be significant in order to identify the relevant determinants of recommendation-induced abnormal stock performance. There are two sets of characteristics which influence the magnitude of the abnormal return following analyst recommendations; Stock characteristics and Analyst characteristics. This study includes both sets. I first discuss the relevant stock characteristics and thereafter the relevant analyst characteristics will be discussed.
Stock characteristics
Market Value
Stickel (1995) finds that stocks of smaller firms , as measured by market value, react more strongly than those of larger firms to broker recommendations. This result is consistent with differences in firms’ information environments; information about smaller firms is reported less frequently, increasing the information content of any single report. However, Moshirian et al(2009) state that investors prefer large stocks over smaller stocks on the basis of their superior information environment. In sum, I expect the relation between the market value of a stock and its abnormal return to be negative.
H2 Market Value hypothesis Smaller stocks have larger abnormal returns following analyst
recommendations
Analyst coverage
13 H3 Analyst Coverage hypothesis A negative relation exists between analyst coverage and the
magnitude of the abnormal returns following analyst recommendations
Analyst characteristics
Nationality
The geographical proximity of analysts in relation to the accuracy of their recommendations is a controversial topic in academic research. Intuitively, one would expect the local analyst to have an advantage over foreign analysts owing to a better understanding of language, culture and business values. Also, the cost of information and the actual proximity of the analyst to the covered firms can be perceived as advantageous. Bae,Stulz & Tan (2008) find evidence for what they call the local analyst advantage. They show that earnings forecasts of local analysts are more precise. However, they note that the local analyst advantage is stronger in countries were disclosures are weaker, where institutional investors are less important, where corporate ownership is more concentrated, and where accounting information is less informative.
Yet, Chang (2009) finds that foreign analysts outperform domestic analysts. He also notes that expatriates also outperform local analysts. Although they share the physical location, both parties are presumably differently endowed. Chang (2009) concluded that global resources and knowledge are very important when it comes to the performance of an analyst’s recommendations. Bacmann & Bolliger (2001) agree with Chang(2009) and show that foreign analysts outperform local analysts in the Latin American market. These authors point to the advantage in information, sophistication and resources that foreign parties have as well. However, evidence from research done in Western markets shows that domestic analysts are more accurate compared to non-domestic analysts. Actually, Orpurt (2004) finds that analysts in Germany and the Netherlands are the best accuracy outperforms.
Therefore I hypothesize as follows:
H5 Nationality hypothesis Recommendations from Dutch brokerage houses lead to higher abnormal
returns
Investment banking
14 recommendations made on non-clients is more positive compared to investment bank clients. Barber et al (2007) find that the recommendations of independent brokerage houses outperform those by investment bank-affiliated analysts by 8% annualized. Nevertheless, Allen & Faulhaber(1989) assume that underwriting firms, because of their client-relation, have more information on that certain firm and thus have an informational advantage. This, in turn, might enable affiliated analysts to make more accurate recommendations. Also, Hirst Koonce & Simko (1995) report that experimental investors react stronger to negative reports from analysts who lack independence. Hence, I hypothesize as follows:
H6 Investment Banking hypothesis Recommendations from investment bank affiliated analysts lead
to lower abnormal returns
Brokerage house size
The seventh hypothesis comes from the suspicion that recommendations made by large brokerage houses will lead to bigger abnormal returns. Elton,Gruber & Grossman(1986) find evidence confirming this suspicion. Logue(1986) discussed these findings and questions whether the abnormal returns following analyst recommendations are the results of ‘true forecasting ability or just the strength of the brokerage house marketing effort. Loque(1986) questions whether analysts are really identifying altered company situations. He argues that it may be the case that the salesmen of the brokerage houses simply convincing enough investors. If this were to be true then recommendations from larger brokerage houses, employing more staff and generating more price pressure, have a greater price effect.
H7 Brokerage House Size A positive relation exists between the size of the brokerage house and the
magnitude of the abnormal returns
Recommendation Strength
According to Stickel(1995), a buy recommendations signals that the analyst believes the company is undervalued by the market. A strong buy signals an even more greatly undervalued company. Thus, prices should react more to strong buys than to buys. This effect also works for sell recommendations; a sell recommendation should lead to a stronger market reaction than an underperform recommendation. Following this line of reason, I anticipate a larger abnormal return following an upgrade to a strong buy recommendation and, likewise, a downgrade to a sell recommendation.
H8 Recommendation Strenght hypothesis Upgrades( downgrades) to strong buy (sell) have greater
15 Likewise, recommendation revisions or changes that skip a rank are expected to lead to larger market reactions. Stickel(1995) hypothesizes that these revisions that skip a rank will have a larger price impact because of the greater change in expectations and thus the bigger the surprise.
H9 Revision Magnitude hypothesis The abnormal returns following a revision that skips at least one
rank are greater compared to those revisions that do not Post-2003 Regulation
Regulators in the U.S. and Europe enacted the Global Settlement and the Market Abuse Directive regulations in order to enhance credibility of analyst reports and limit the relation between research- and investment banking departments. The implementation of the ‘Chinese Wall’ would (partly) solve the conflict of interest that affiliated analysts faced, thereby paving the way for more objective and non-biased recommendations. In this study, I test whether the regulations of 2003 have affected the (magnitude of) abnormal returns following analyst recommendations in the Netherlands. As a preliminary test I conducted a Chow breakpoint test on the date of enactment of the Global Settlement act ( the Market Abuse Directive was enacted shortly after). The Chow breakpoint analysis tests whether a substantial change in variance has taken place in the sample of residuals on the given date. The Chow test and the complementary Recursive Estimation test show that a rather significant breakpoint is located close to April 28, being the enactment date of the regulations, thus suggesting that the regulations have had a significant effect on the price reaction to analyst recommendations. Unfortunately, not much research has been done on this topic. Therefore, this research goes into unchartered territory by investigating the (recommendation) market response to these regulations . Since the regulations are aiming to enhance the objectivity and credibility of analyst recommendations, I hypothesize that the recommendation-induced abnormal returns will be higher after the Global Settlement regulations
H10 Post-2003 Regulations The abnormal returns following recommendations will be higher after
the enactment of the regulations of 2003.
Also, when performing the Chow breakpoint analysis, I found indications that a significant breakpoint occurs in the dataset mid-september 2008, coinciding with the fall of Lehman Brothers, a well-covered event that drew attention from all over the world. This event is often regarded as the start of the financial crisis of 2008. I suspect that the fall of Lehman Brothers, causing consumers confidence to fall, has lead investors to lose trust in financial institutions. Therefore, I expect the abnormal returns following recommendations after the fall of Lehman Brothers to be significantly lower than before. H11 Post-2008 The abnormal returns following recommendations will be lower after the enactment of
16 Leading and Following Recommendations
Welch(1999) finds that the recommendations of financial analysts have a positive influence on the next two recommendations. The phenomenon of analysts following other analysts is called ‘herding’ or ‘mutual imitation’. Elaborating on this idea, it might be interesting to test to what extent herding among analysts takes place. I do this by looking at the clustering of issuance dates of analyst recommendations. A distinction will be made between those analysts that are ‘leading’ the herd; they issue recommendations on a certain stock that are followed shortly thereafter by other analysts ( the ‘followers’). A logical rationale would be that the abnormal return following the recommendation of a ‘leader’ would be of greater magnitude compared to the abnormal returns following the recommendations of ‘followers’. I expect the effect of following recommendations shortly after the issuance of another recommendation to lead to a smaller abnormal return (if any). Therefore, I hypothesize as follows:
H12 Leading Recommendations hypothesis: The abnormal returns following a leading
recommendation will be significantly higher.
Alternatively, it might be interesting to test whether the distinction between leading, following, and independent recommendations proves to be significant.
III. Data and Sample Description
The primary data used in this paper comes from the International Brokers’ Estimate System (IBES). IBES is a database founded in 1976, collecting analysts ‘estimates and recommendations concerning over 60.000 firms with more than 1.200 firms contributing to the IBES database. This data source is frequently used for studies regarding analyst estimates and recommendations ( Jegadeesh and Kim (2006), Moshiran,Ng and Wu (2009),Chan and Hameed (2006) and Chang,Dasgupta & Hilary (2009). A data set from 1 January 1995 to 22 October 2012 is used. This time span is wide compared to similar studies. The data include 1) the IBES ticker symbol, 2) the name of the reviewed stock, 3) the recommendations’ exact date and time, 4) the analysts’ firm and analysts’ name, 5) the
17 For a recommendation to be included in this study; 1) The relevant stock has to be listed on either the AEX or the AMX at the time that the recommendation is published, 2) the recommendation has to be made by one of the 28 brokerage firms selected for this study. This selection is based on a minimum number of 250 recommendations per brokerage firm. 3) daily stock return data for the 250 days prior to be event date has to be available for the relevant stock. 4) The analyst recommendation is a revision of a previous recommendation, reiterations (confirmations of earlier recommendations) are therefore not included in the sample. Finally, I excluded 8 events from the sample based on an examination of the abnormal returns found using the event study. After manually checking these events, I decided to exclude them from the sample. Table 2 shows the selection criteria and their effect on the sample size. A list of the 28 brokerage houses included in this study is provided in appendix I.
Table 2: Selection Criteria and their effect on the sample size
Selection Criteria Sample size (N)
1 Stocks listed on the AEX/AMX in the period 1995-2012 24,527
2 Recommendations from the 28 brokerage houses 16,940
3 250 daily stock returns prior to the event date available 13,756
4 Only revisions 9,715
5 Deletion of clear outliers 9,708
Final Sample 9,708
18
Composition of the recommendations
Womack (1996) reported a buy (strong buys and buys) to sell (underperform and sell) ratio of 7 to 1 in his 1989-1991 sample of analyst recommendations in the U.S. Likewise, Stickel(1995) reports a ratio close to 5 to 1 (55% vs. 12%). More recently, Barber et al(2001) and Jegadeesh & Kim(2006) report even higher ratios,8.3 (54% vs. 6.5% ) and 18.8 (62.2% vs. 3.3%) respectively, found in the U.S. market. With the ratio of less than 3, the Dutch sample employed in this study is substantially less positive compared to the U.S. recommendations. The buy/sell ratio found in this sample is more in line with those found in European neighbours France and Germany (see table 3 and 4). Whereas Michaely & Womack(2005) point to the high buy/sell ratio found in the U.S. market as a sign of optimism bias and a conflict of interest, the substantially lower ratio’s found in Europe and in the Netherlands in specific may indicate that the analysts operating there are less concerned with the analysts’ conflict of interest.
Table 3: Distribution of recommendations (as found in the literature review)
Country Buy%
(Buy and Strong Buy) Sell% (Sell and Underperform) Stickel (1995) U.S. 55% 12% Barber et al (2001) U.S. 54.1% 6.5%
Jegadeesh & Kim
(2006) U.S. 62.2% 3.3% Canada 58% 12.1% Japan 46% 18.3% Great Britain 46.6% 11.8% France 53% 15.9% Germany 38.6% 19.9% Italy 39.2% 13.6%
19
Table 4:Matrix of changes in Recommendations & Composition
New Recommendation rank/scale
Old Rank 1Strong
buy
2.Buy 3.Hold
4.Under-perform
5.Sell Total Percentage
1.Strong buy 348 574 1066 58 82 2128 15.45% 2.Buy 593 779 1678 250 39 3339 24.24% 3.Hold 1038 1562 1122 802 331 4855 35.24% 4.Under-perform 46 236 745 273 86 1386 10.06% 5.Sell 77 48 336 79 40 580 4.21% Initiations 313 410 579 119 67 1488 10.8% Total 2415 3609 5526 1581 645 13756 Percentage 17.53% 26.2% 40.11 % 11.48% 4.68% Buy Sell 43.73% 16.16%
Table 5: Descriptive Statistics of the independent variables (IVs)
Variable Mean Median Maximum Minimum Std. Dev
Average Recommendations 24.86 24.33 39 3.4 8.55
Market Value (Log) 3.598 3.66 4.96 2.05 0.69
20
Continuous variables
The average number of analysts covering a certain stock construct the Analyst Coverage variable. The market value of each stock is measured by the Market Value variable. This variable accounts for the size of all individual stocks included in this study. In order to normalize the distribution I
loglinearized the market values. Stickel(1995) states that smaller stocks lead to bigger price reactions to recommendations because of their informational environment. The Brokerage House size variable indicates the size and , according to Stickel(1995) and Logue(1986), the marketing ability of a
brokerage house. I use two proxies for the size of the brokerage houses included in this study. Firstly, I take the number of analysts per brokerage house in this sample. Thus, the number of analysts working for a certain brokerage or research firm indicate the size of that firm, as recommended by
Stickel(1995). Secondly, I use the number of stocks covered by the firm as a proxy for brokerage or research house size (portfolio complexity).
Dichotomous variables
Those recommendations coming from an analyst working for a Dutch brokerage house are assigned a ‘1’ whereas those who do not are assigned a ‘0’.The same approach is used when assigning ‘1’s to recommendations coming from brokerage houses affiliated to investment banking departments. Revisions skipping a rank, e.g. upgrading from ‘hold’ to ‘strong buy’ are labeled ‘1’.’Strong Buy’ and ‘Sell’ recommendations are also labeled ‘1’. The dummy variables Post 2003 and Post 2008 are assigned a ‘1’ when the recommendation was issued after the enforcement of the Global Settlement regulations in 2003 or after the fall of Lehmann Brothers in September 2008. Lastly, those
recommendations that are not preceded by another recommendation in the preceding three days but are followed by another recommendation are labeled ‘Leader’ and are assigned a ‘1’ . Those
recommendations that are preceded by another recommendation in the three days before its issuance are labeled ‘Follower’ and are assigned a ‘1’. ‘Independent’ recommendations are not preceded nor followed by other recommendations in the surrounding days and are assigned a ‘0’
IV. Methodology
21 the abnormal returns following analyst recommendations.For both the event study and regression, two separate analyses will be performed; 1) upgrades and 2) downgrades.
Event Study Methodology
An event study measures the effect of a specific event on the market value of a security. The event study methodology used in this study closely resembles those of Brown and Warner (1980,1985) and MacKinlay(1997). The initial task of conducting an event study is to define the event of interest and identify the period over which security prices of the firms involved in this event will be examined, the so-called event window. It is customary to define the event window to be larger than the specific period of interest as this permits examination of periods surrounding the event( MacKinlay 1997). In the event study, I test whether abnormal returns occur during the event window. Besides testing whether abnormal return are found on the event date, I may also study a broader event window, depending on the significance of the abnormal returns surrounding the event date. In order to find abnormal returns (ARit), one needs to compare the actual returns (Rit) to the expected returns (E(Rit)).
The actual returns are adjusted closing prices. This means that they are adjusted for stock splits and dividend payments. The abnormal return is the actual return of the stock minus the normal or expected return. Following Brown and Warner (1985), the abnormal return for each stock for every day in the event window is determined using the following formula;
Abnormal return:
AR
it
R
it
E
(
R
it)
All returns are continuously compounded. Pit is the stock price on day t and Pit-1 is the stock price on
day t-1.
Continuously compounded actual stock return:
1
itP
P
Ln
R
it itThe expected returns are estimated using the market model. This model calculates the expected return of stock i based on the stock’s alpha (αi) and beta (βi) which resemble the stock performance relative
to that of the appropriate market benchmark (Rmt). The market model assumes a linear relation
between the market return and the security return. The benchmark market in this study is either the AEX or the AMX, depending on which index the stock is listed. The market model can be expressed as follows:
Market Model:
22 This regression model below estimates the alpha (αi) and the beta (βi) of a certain stock over an
estimation period of 250 days prior to the event date. These estimated parameters are used to calculate the expected or normal return of stock i using the Market Model formula.
Regression equation for estimating i
and i
it mt i i it R R E
) ( withvar(
it)
2Before moving on to a further development of the framework, some notations have to be defined first. As already discussed, t notates the event time. Let t = 0 define the event date, t = T1 to t = T2 defines
the event window and T0 + 1 to t = T1 constitutes the estimation window. L1 = T1 – T0, which equals the
length of the estimation period. The length of the event window is defined by L2 = T2 – T1. The event
study timeline is depicted below.
Figure 1 The event study timeline. From MacKinlay (1997)
The estimation window is used to estimate the expected return of stock i and the event window is used to test whether abnormal returns occur as a result of the event. The post-event window is used to test whether post-event abnormal returns occur. Womack (1996) for example, found significant post-event abnormal returns. To prevent the returns around the event date from influencing the expected returns using the market model, the estimation window and event window do not overlap. In this study the estimation window consists of 250 trading days.
Test statistics
Using the model as explained in the event study section, I test the significance of my abnormal returns hypothesis with various t-statistics. The average abnormal return (AARt) is calculated by dividing the
sum of all individual abnormal returns by the sample size (N):
Average Abnormal Return
23 Dividing this by the appropriate standard error results in the t-statistic. This t-statistic tests whether the average abnormal returns found differ significantly from zero. If so, this indicates that analyst recommendations lead to significant abnormal returns. I employ a multitude of t-tests, of which the results are reported in the tables 7 and 8. There is some discrepancy in significance found among these various t-tests. For reasons of clarity, I will briefly discuss the traditional t-test and the non-parametric Wilcoxon ranked sign test. The other tests are discussed in Appendix II. The t-test statistic equals the average of the event-period abnormal returns divided by the square root of the sum of all securities estimation-period residual variances:
Traditional t-test (as used in this study)
i AR it it AAR AAR t
where
N i i AR 1 2
The next step is accumulating the AARs to calculate the Cumulative Average Abnormal Return (CAAR) in order to measure the cumulative abnormal return over a multi-day event-window. The CAAR is calculated by adding the AARs of the days included in the event-window (
1
to
2
);Cumulative Abnormal Return
2 1 ) 2 , 1 (
t it AR CARCumulative Average Abnormal Return
n i CAR CAAR 1 ) 2 , 1 (
The significance of these cumulative average abnormal returns is also tested using the following t-test;
Traditional t-test for Cumulative Average Abnormal Returns
i CAAR it it CAAR CAAR t ) 2 , 1 (
where(
1
,
2
)
(
)
1 2
N i it i CAAR
AAR
24
Wilcoxon Signed Rank test
i N i i i X R Y SIGN W ( )* 1
Yi denotes the actual return on the event day, whereas Xi denotes the median of the returns on that day.
I have ranked the differences between Yi and Xi according to the absolute value (Ri). The null
hypothesis when using a Wilcoxon Signed Ranked test is that there is no tendency in one direction (positive or negative) and thus the numbers of positive and negative signs would be approximately equal. In that case, the expected value of W would be zero. The next step is summing the positive differences into RANK+ and the negative differences into RANK-;
and
N i i i i i R for Y X R 1 0 ) (The test statistic is the smaller of the two summations described above.
Regressing (cumulative) abnormal returns on stock- and analyst characteristics
In order to analyze the relation between the magnitude of abnormal returns and stock- and analyst characteristics I employ Ordinary Least Squares (OLS) regressions. As mentioned earlier, I
distinguish between upgrades and the downgrades and run separate regressions, one with upgrades and one with downgrades These regressions will include 14 variables including the dependent variable. The following regression equation is estimated using White’s heteroskedasticity consistent standard errors and covariance: (C)ARit = βi + ANALYST_COVERAGE*β1 + MARKET VALUE*β2 +
PORTFOLIO COMPLEXITY*β3 + BROKERAGE_SIZE*β4 + INVESTMENT_BANKING*β5 +
SKIPRANK*β6 + STRONG*β7 + POST_2003*β8 + POST2008*β9 + LEADER*β10 + FOLLOWER
*β11 + εit
The descriptive statistics of the (cumulative) average abnormal returns are reported in Appendix III. Also, the correlation matrix and the variance inflation index are provided in Appendix IV and V. The descriptive statistics show that the Jarque-Bera null hypotheses (the (C)AARs are normally distributed) cannot the rejected. Although Brown and Warner (1985) note that the daily stock returns for an individual stock show departures from normality, they point to evidence provided by Blattberg & Gonedes (1974) and Hagerman (1978) that the distribution of the cross-sectional daily mean return converges to a normal distribution. Also, the Central Limit Theorem, as explained by Billingsley (1979), mathematically proves that the distribution of the sample mean abnormal return converges to normality as the number of securities. With the sample size (N=4751 and N=4957) employed in study, I safely presume a normal distribution of the errors and continue with the regression. MacKinlay (1997) notes that there is no reason to expect the residuals from the regression to be homoscedastic
25 and thus he deems it advisable to use heteroskedasticity-consistent standard errors. Following the approach of White (1980), I adjust my regression model for heteroskedasticity. As mentioned before, I excluded obvious outliers from the sample. Nevertheless, the graphical examination of the distribution of the AARs and the CAARs shows some visual outliers. I have tested whether the exclusion of outliers, defined as returns larger than five times the standard deviation, would significantly affect the average abnormal return. I find that, for the downgrades as well as the upgrades, the exclusion of these (potential) outliers has no significant effect on the average abnormal returns (downgrades: for CAR t = -0.07 p= 0.94 and for AR t = 0.6, p=0.54; upgrades: for CAR t= 0.29 p=0.76 and for AR t=0.09 p=0.92).
The correlation matrix and the Variance Inflation Index are provided in Appendix IV and V. In order to address a potential multi-collinearity problem, I examined both the correlations between the variables employed in the regressions. When examining the Variance Inflation Factors, the threshold of 10, provided by Kutner (2004), was never exceeded. However, the more conservative threshold of 4 (moderate multicollinearity) is exceeded by the Variance Inflation Factor of the ‘Nationality’ variable. The square root of the VIF shows how much larger the standard deviation is with the current level of collinearity compared to the situation in which no collinearity exists among variables.The correlation between the Nationality-variable and the Portfolio Complexity-, Brokerage house size-, and the Investment Banking variable is rather high. In order to handle this potential problem, I will run several regressions; both including and excluding the Nationality-variable.
Dependent Variable
26
Table 6 Summary of Stock and Analyst Characteristics
Hypothesis Literature Variable Measured by Expected sign Down grades Expected sign Up grades H2: Market Value Stickel (1995), Moshirian et al (2009)
Market_Value Log of Market
Value of individual stocks positive negative H3: Analyst Coverage Doukas (2005),Lang (2008), Stickel (1995)
Average_Rec The average
number of analysts following a certain stock
positive negative
H4: Nationality Bae, Stulz & Tan (2008), Chang (2009),Bacmann & Bolliger (2001), Orpurt(2004) Nationality Dutch brokerage house = 1, Non-Dutch = 0 negative positive H5: Investment Banking Michaely & Womack (1999), Barber et al (2007),Allen & Faulhaber (1989), Hirst, Koonce & Simko (1995) Investment Bank Research department affiliated with investment bank = 1, independent = 0 positive negative H6: Brokerage House Size Elton,Gruber & Grossman (1986) Number_of_ Analysts Portfolio Complexity The number of analysts per brokerage houses included in this sample. The number of stocks covered by a brokerage house negative positive H7:Recommenda tion Strength
Stickel(1995) Strong_Rec Strong buy and
Strong sell = 1
negative positive
H8: Revision Magnitude
Stickel (1995) Skip_Rank Revision skips a
rank = 1
negative positive
H9: Post 2003 None Available Post_2003 Post 2003 = 1 , Pre 2003 = 0
negative positive
H10:Post 2008 None Available Post_2008 Post 2003 = 1 , Pre 2003 = 0
positive negative
H11:Leading Recommendation
Welch(1999) Leader Leading = 1
Not-Leading =0
27
V. Results
Event study results
28
Table 7:Daily Average Abnormal Returns and their significance for Upgrades
(all tests are two-tailed, with *** indicating significance at 1% alpha-level with a t-statistic ≥2.58, ** is significant at 5% level with a t-statistic ≥1.97, and * indicates significance at 10% level, t-statistic ≥1.65)
Graph 1: Cumulated Average Abnormal Returns following upgraded revisions
29 Table 8: Daily Abnormal Returns and their significance for Downgrades
(all tests are two-tailed, with *** indicating significance at 1% alpha-level with a t-statistic ≥2.58, ** is significant at 5% level with a t-statistic ≥1.97, and * indicates significance at 10% level, t-statistic ≥1.65)
Graph 2: Cumulated Average Abnormal Returns following downgraded revisions
-0.025 -0.02 -0.015 -0.01 -0.005 0 AR( -10) AR( -9) AR( -8) AR( -7) AR( -6) AR( -5) AR( -4) AR( -3) AR( -2) AR( -1) AR( 0) AR( 1) AR( 2) AR( 3) AR( 4) AR( 5) AR(6) AR( 7) AR( 8) AR( 9) AR( 10) Per e ce n tage
Cumulated Average Abnormal Returns
30
Table 9:(Cumulative) Average Abnormal Returns and their significance
(The reported t-statistic is calculated using the formula from the methodology section, all tests are two-tailed, with *** indicating significance at 1% alpha-level with a statistic ≥2.58, ** is significant at 5% level with a
t-statistic ≥1.97, and * indicates significance at 10% level, t-t-statistic ≥1.65)
Average Abnormal Returns.
Interestingly, significant abnormal returns are found ten days prior to the event date for downgrades. Jegadeesh and Kim (2006) and Stickel (1995) also find evidence for pre-event date abnormal returns. These pre-event abnormal returns are presumably caused by leakage effects; the markets or some investors have already picked up signals of the upcoming downgrades. A brokerage house informing large clients before the announcement date might also be a plausible explanation. Nevertheless, it is surprising that ten days prior to the downgrade significant abnormal returns are already found. At least for six of the ten days prior to the downgrade I find significant abnormal returns. The leakage effect for upgrades seems to be smaller and closer to the event date compared to those found prior to downgrades. Actually, significant pre-event abnormal returns prior to upgrades are only found in the three days prior to the event. It seems that the market is better at anticipating downgrades then upgrades.
Unlike Womack (1996), I find no evidence for post event price drifts. For both upgrades and downgrades the post-event abnormal returns seem to reverse to zero shortly after the recommendation issuance date. For the day after the event, I find significant abnormal returns for all revisions, which seem to indicate that the market needs a day to fully digest the new information conveyed by the analyst recommendation. The quick reversion to zero abnormal performance can be explained in various ways. It seems plausible that once all interested investors have analyzed and interpreted the Days relative to
event date
Downgrades (N=4957) Upgrades (N=4751)
(C)AAR T-statistic (C)AAR T-test
31 (revised) recommendation no excess trading on the relevant stock is done. This quick reversion of abnormal returns to zero is in line with the semi-strong form of market efficiency; the stock prices quickly adapt to the new available information. Based on these results, it can be concluded that these results provide evidence for Grossman and Stiglitz’ (1980) expanded view on efficient markets. In the case of fully efficient markets, no abnormal returns following analyst recommendations should have been found.
Cumulative Average Abnormal Returns.
The predowngrade leakage effect is confirmed by the statistical significance of both the 8day (day -10 to -2 ) cumulative average abnormal return (CAAR) and the 4 day (day -5 to -2) CAAR; for the 8-day CAAR I find a significant cumulative average abnormal return of -1.2% for downgrades whereas the significant 4-day pre-event CAAR is -0.74%. These significant pre-event CAAR is significantly bigger than the abnormal return found on the event date. This might indicate that the majority of the ‘new information’ that is disseminated by the issuance of the downgrades is already anticipated and reflected in the stock prices in the days prior to the event date. However, the fact that the event-date abnormal returns are significantly higher, for both upgrades and downgrades, than those found on the days before and after the event suggest that the analyst recommendation does contain new information that has investment value. For upgrades the leakage or anticipation period is much shorter; the pre-upgrade 8-day CAAR is insignificant. However, the shorter pre-event 4-day CAAR is significant and shows a cumulative abnormal return of 0.30% for upgrades which is significantly lower then the downgrades pre-event 4-day CAAR.
32
Graph 3. Average Abnormal Returns following downgrades
Graph 4. Abnormal returns following upgrades
33 Prior to the discussion of the regression results, I discuss the results displayed in table 10; It shows the (cumulative) average abnormal returns classified by the binary categories according to the variables that are included in the regressions I use to analyze the influence of the stock- and analyst
characteristics as discussed before and reported in table 5 and 6.
Interestingly, I find that recommendations on stocks that are relatively more covered by analysts lead to significantly lower event-day abnormal returns compared to those stocks that are followed by fewer analysts. The intuition that smaller stocks are associated with bigger information asymmetry seems to be confirmed; recommendations on smaller stocks are followed by significantly bigger abnormal returns. I also find that recommendations on stocks listed on the AMX, the smaller index in the Netherlands, are followed by significantly higher abnormal returns when compared to those
recommendations that cover AEX-listed stocks. (downgrades: t-statistic =-2.44,for upgrades: t-statistic =2.40). The rationale that recommendations from an analyst employed by a big brokerage house is also proven right; the (cumulative) abnormal returns associated with big brokerage house
recommendations are significantly bigger compared to those abnormal returns following small brokerage house recommendations. However, portfolio complexity measuring the number of stocks that a brokerage house covers and used as a measure of brokerage house size does not lead to significant differences in abnormal returns. Surprisingly, recommendations from analysts that are employed by brokerage houses that are affiliated with an investment banking department lead to significantly higher (cumulative) abnormal returns. This finding contradicts the expected relationship. The regression output might provide more insight in this matter. Also the expectation that
recommendations from analysts employed by Dutch brokerage houses will lead to higher abnormal returns is proven wrong; recommendations from non-Dutch brokerage houses induce significantly (cumulative) abnormal returns. Although the results are not significant, revisions that skip a rank or ‘Strong Buy’ and ‘Sell’ recommendations seem to lead to lower abnormal returns.
Furthermore, I find evidence that the (cumulative) abnormal returns following recommendations have significantly increased after the Global Settlement regulations of 2003. More surprisingly, the
34
Table 10: (Cumulative) Average Abnormal Returns classified by categories
(For Average Recommendations ‘0’ represents the the group with an recommendations average between 0 and 20; ‘1’ represents the group between 20 and 40; Market Value: ‘0’ is the group with scores between 2 and 4 whereas ‘1’ categorized
those stock whose LN Market Value is between 4 and 6; Portfolio Complexity: ‘0’= between 0 and 50 stocks covered and ‘1’ is between 50 and 100; Brokerage Size: ‘0’ = less than 50 analysts ‘1’ categorizes more than 50 analysts The reported t-test indicates the significance of the difference in (cumulative) average abnormal returns, all tests are
two-tailed, with *** indicating significance at 1% alpha-level with a statistic ≥2.58, ** is significant at 5% level with a t-statistic ≥1.97, and * indicates significance at 10% level, t-t-statistic ≥1.65)
Downgrades N=4957
AR(event date) CAAR (-1,1)
Mean 0 (=No) 1 (=Yes) t-stat Mean 0 (=No) 1 (=Yes) t-stat Average Recommendations -0.38% -0.51% -0.32% -2.21 ** -0.88% -1.09% -0.77% -2.08** Market Value -0.45% -0.22% -2.53** -1.01% -0.54% -2.84** Portfolio Complexity -0.55% -0.34% -1.92* -1.10% -0.81% -1.48 Brokerage Size -0.22% -0.48% 3.93*** -0.82% -2.02% 3.16*** Investment Banking -0.22% -0.58% 4.66*** -0.48% -1.34% 5.92*** Nationality -0.56% -0.28% -3.34*** -1.22% -0.67% -3.64*** Skiprank -0.39% -0.36% -0.39 -0.85% -0.88% 0.22 Strong -0.39% -0.26% -1.04 -0.91% -0.39% -2.24** Post 2003 -0.11% -0.66% 7.02*** -0.33% -1.41% 7.55*** Post 2008 -0.27% -0.93% 6.25*** -0.68% -1.77% 5.67*** Leader -0.29% -0.88% 5.26*** -0.75% -1.56% 3.93*** Follower -0.36% -0.46% 0.95 -0.73% -1.45% 3.89*** Independent -0.64% -0.25% 4.7*** -1.50% -0.55% -6.17*** Upgrades N=4751
AR(event date) CAAR (-1,1)
35 Regression results
In this section I present and discuss the results of the regressions I use to analyze the relevant stock- and analyst characteristics and their relationship with the magnitude of abnormal returns following analyst recommendations. Because the event-date abnormal returns (AARs) and the three-day CAARs are both very significant, I decide to employ both variables as dependent variable in the regression. Thus, I run four regressions; one regression with the event day ARs as the dependent variable and one with the three-day CARs as the dependent variable. This process is done twice; once for the upgrades and once for the downgrades.
In order to mitigate the potential multi-collinearity problem posed by the inclusion of the Nationality-variable, I compare regression results from regressions including and excluding that variable. The Nationality-variable is not significant in either of the four regressions. The exclusion of this variable slightly increases the adjusted R2 and the F-values of the regression models. The regression results for the downgrades are presented in table 11 and the regression output for the upgrades are reported in table 12. The regression results including the Nationality-variable are included in Appendix VI. The high correlation (0.669) between the brokerage house size- and the investment banking-variable is high but not unexpected. It seems logical that large financial firms also have investment banking departments whereas independent research houses tend to be smaller in size. Also, the high correlation (0.456) between the Skiprank- and the strength of the recommendation-variable is easily explained; it seems quite probable that those revisions that skip a rank in regard to the previous recommendation might be revised to either a ‘strong buy’ or a ‘sell’. These skiprank-revisions might be triggered by a sudden detoriation or improvement of circumstances for a certain stock.
36
Table 11 : Regression results for Downgrades
(all tests are two-tailed, with *** indicating significance at 1% alpha-level with a t-statistic ≥2.58, ** is significant at 5% level with a t-statistic ≥1.97, and * indicates significance at 10% level, t-statistic ≥1.65)
Table 12 : Regression results for Upgrades
(all tests are two-tailed, with *** indicating significance at 1% alpha-level with a t-statistic ≥2.58, ** is significant at 5% level with a t-statistic ≥1.97, and * indicates significance at 10% level, t-statistic ≥1.65)
Downgrades Event-date AR (0) 3-day CAAR
Variable Expected Sign Coefficient t-statistic Coefficient t-statistic
Constant -0.00720 -2.31** -0.01225 -2.20** Average rec. + -0.00003 -0.55 -0.00006 -0.68 Market Value + 0.00232 2.98*** 0.00579 4.16** Portfolio Complexity - 0.00003 0.95 -0.00003 -0.50 Brokerage size - -0.00004 -1.63 -0.00009 -2.12** Investment Banking + -0.00183 -1.61 -0.00524 -2.48** Skiprank - 0.00006 0.07 -0.00137 -0.89 Strong - 0.00098 0.59 0.00481 1.58 Post 2003 - -0.00396 -4.32*** -0.00841 -5.02*** Post 2008 + -0.00313 -2.41*** -0.00295 -1.34 Leader - -0.00603 -4.14*** -0.00932 -3.75*** Follower -0.00205 -1.80* -0.00952 -3.99*** R2 2.43% 3.30% Adj R2 2.20% 3.07% F-statistic 10.60 14.54 Prob (F-statistics) 0.0000 0.0000
Upgrades Event-date AR (0) 3-day CAAR
Variable Expected Sign Coefficient t-statistic Coefficient t-statistic
