Effects of MAD and MiFID on earnings forecast optimism in the German stock market.

Transcription

1 Effects of MAD and MiFID on earnings forecast optimism in the German stock market. Jörg Prokop * and Benno Kammann # January 15, 2016 Abstract European regulators recently adopted the Market Abuse Directive (MAD) and the Markets in Financial Instruments Directive (MiFID) to improve investor protection by reducing biases in analysts equity research. We analyse the effect of these directives on optimism in financial analysts earnings forecasts in the German stock market. We find that post-mad and, in particular, post-mifid affiliated analysts (over)optimism is mitigated. However, while in case of a negative outlook affiliated analysts tended to be overly pessimistic before the implementation of MAD and MiFID, they now seem more reluctant to issuing negative earnings forecasts. As a consequence, the directives mitigating effect on financial analysts optimism comes at the cost of reduced informational efficiency. JEL: G12, G14, G24, G28 Keywords: conflicts of interest, equity, financial intermediary, regulation, security analyst * # Corresponding author; University of Oldenburg, Area Finance and Banking, D Oldenburg, joerg.prokop@uni-oldenburg.de University of Oldenburg, Area Finance and Banking, D Oldenburg, benno.kammann@uni-oldenburg.de

2 Effects of MAD and MiFID on earnings forecast optimism in the German stock market 1 Introduction This study examines the impact of the Market Abuse Directive (MAD), and of the Markets in Financial Instruments Directive (MiFID), two European regulatory initiatives that aim at improving investor protection by forcing financial service firms to install independent remuneration schemes, and internal controls in order to avoid potential conflicts of interest related to financial analysts activities. In particular, we focus on the directives effects on potential biases in financial analysts equity research in the German market due to conflicts of interest. In this context, potential sources of conflicts of interest may be, for instance, current or prospective investment banking ties of the analyst s employer with the company analysed, or an analyst s relationship to the target company s management. This study focuses on conflicts of interest arising from investment banking ties of the analyst s employer and the mitigating effects of recent regulatory changes. Any analyst who issues an earnings forecast on a firm for which his employer has acted as an underwriter (SEO, IPO, or debt issuance) or as an M&A advisor over the last 12 months is considered as affiliated in this study. Empirical evidence reveals that affiliated analysts are prone to being influenced by conflicts of interest, and tend to issue more optimistic signals than non-affiliated analysts (e.g. Lin and McNichols (1998), Michaely and Womack 1999). In this context, European regulators recently adopted two directives intended to increase analyst signals quality, and to foster investor protection in the EU. Following a recent trend in US regulation (NASD Rule 2711, NYSE Rule 472, regulation fair disclosure (RegFD) and Global Analyst Research Settlement ) European regulators enacted the Market Abuse Directive (MAD) in In Germany, MAD became effective October 28 th, In addition, European regulators passed MiFID, which became effective in Germany November 1 st, Our main aim is to study the impact of MiFID on earnings forecast optimism for affiliated and non-affiliated equity analysts. Optimism is measured as the deviation of an individual analyst s earnings forecast for a firm from the respective firm s consensus forecast. Using a multiple regression model and controlling for the effect of the banking crisis, we show that the implementation of MiFID was followed by a decrease in optimism for affiliated analysts. However, the latter still tend to be overoptimistic compared to their nonaffiliated peers in case of smaller firms, and longer forecast horizons. In addition, we find that after the implementation of MAD and MiFID affiliated financial analysts seem more reluctant 1

3 to issuing negative earnings forecasts, which may indicate a deterioration in their earnings forecasts information content. The remainder of the paper is organised as follows. Section two outlines the regulatory context and develops the research hypotheses. Section three describes the data as well as the methodology used. Section four discusses the study s results, and section five concludes. 2 Regulatory context and research hypotheses 2.1 US regulation Empirical evidence on biases in equity research combined with uncovered malfunctioning of Chinese Walls in investment banks recently led US regulators to adopt several regulatory changes in order to reduce financial analysts (over)optimism. In 2000, Regulation Fair Disclosure (RegFD) was adopted, banning firms from selectively disclosing information to market participants. In 2002, NYSE and NASD passed rules concerning the separation of research and investment banking, banning banks from offering favourable analyst reports to attract investment banking business and from linking analyst compensation to specific investment banking deals. In addition analysts were required to publish information on the distribution of their recommendations, on accuracy of investment recommendations for companies covered, as well as on private stock trades (NASD Rule 2711, NYSE Rule 472). Later in 2002, the Global Analyst Research Settlement was reached between the largest US investment banks and the U.S. Securities and Exchange Commission, NASD, NYSE and other domestic regulators. The settlement led to enforcement actions aiming at further mitigating conflicts of interest in investment research. After the passage of these regulatory changes, several studies examined the effects on analysts recommendations and earnings forecasts. According to the empirical evidence gathered, the respective measures successfully mitigated biases in financial analysts output (e.g., Cornett, Tehranian and Yalcin (2006), Kadan et al. (2009), Ertimur et al. (2006), Chen and Chen (2009), and Guan, Lu and Wong (2012)). 2

4 2.2 MAD The Market Abuse Directive (MAD) was adopted in European countries between 2004 and 2006 (2004 in case of Germany). It is intended to guarantee the integrity of European financial markets and increase investor confidence. The objective is to create a level playing field for all economic operators in the Member States as part of the effort to combat market abuse 1. MAD addresses the following issues related to equity analysts (Commission directive 2003/125/EC): First, analyst recommendations are defined as research or other information recommending or suggesting an investment strategy, explicitly or implicitly. This means that all types of analyst outputs (i.e., recommendations, earnings forecasts, and target prices) are addressed. Second, the general standard for fair disclosure is defined as follows: Facts are clearly distinguished from interpretations, estimates, opinions and other types of non-factual information, all sources are reliable or, where there is any doubt as to whether a source is reliable, this is clearly indicated and all projections, forecasts and price targets are clearly labeled as such and [..] the material assumptions made in producing or using them are indicated. Third, the general standard for disclosure of interests and conflicts of interests determines that relevant persons disclose all relationships and circumstances that may reasonably be expected to impair the objectivity of the recommendation, in particular where relevant persons have a significant financial interest in one or more of the financial instruments which are the subject of the recommendation, or a significant conflict of interest with respect to an issuer to which the recommendation relates. In summary, MAD forces analysts to issue investment advice based on reliable information, to disclose material assumptions underlying estimates and forecasts, to disclose potential conflicts of interest and to clearly distinguish personal opinion from facts. For a detailed analysis of the differences between MAD and respective US regulation see Dubois, Fresard and Dumontier (2014). 2.3 MiFID In 2007 the Markets in Financial Instruments Directive (MiFID) was adopted in European markets. MiFID aims at improving the competitiveness of EU financial markets by creating a 1 n_securities/l24035_en.htm 3

5 single market for investment services and activities, and ensuring a high degree of harmonized protection for investors in financial instruments. 2 The following issues of MiFID are relevant for the mitigation of possible conflicts of interest faced by equity analysts (Commission directive 2006/73/EC): First, internal controls have to ensure the independence of analysts, as the measures and arrangements adopted by an investment firm to manage the conflicts of interests that might arise from the production and dissemination of material that is presented as investment research should be appropriate to protect the objectivity and independence of financial analysts and of the investment research they produce. Second, analysts have to separate themselves from activities that interfere with their objectivity, as financial analysts should not become involved in activities other than the preparation of investment research where such involvement is inconsistent with the maintenance of that person's objectivity. Third, the remuneration of analysts has to be in line with their independence, as financial analysts, and other relevant persons involved in the production of the investment research must not accept inducements from those with a material interest in the subject-matter of the investment research. 2.4 Hypothesis development There is a large body of literature analysing optimism in analysts earnings forecasts. First of all there is empirical evidence that analysts earnings forecasts are, on average, optimistic and above actual earnings (e.g. Stickel (1990), Abarbanell (1991), Dreman and Berry (1995), and Chopra (1998)). One source of overoptimism are conflicts of interest arising from underwriting relationships of the analyst s employer to the firm, the analyst is covering. If the analyst issues optimistic earnings forecasts, he may be able to attract additional investment banking business. On the other hand, the analyst s reputation and career perspectives are mainly driven by the accuracy of his forecasts. These conflicts of interest faced by analysts are subject of various studies on the US market, showing that earnings forecasts from analysts who are with brokerage firms that have further investment banking ties to the firms covered tend to be overoptimistic (e.g. Dugar and Nathan (1995), Hong and Kubik (2003), and Chan, Karceski and Lakonishok (2003)). They conclude that brokerage firms reward optimistic analysts who generate additional investment banking business. So for analysts there seems to be a trade-off between reputational gains 2 4

6 and direct rewards from their employer. For a US literature overview on conflicts of interest and their effects on the quality of analysts output, see Ramnath, Rock and Shane (2008). As Capstaff, Paudyal and Rees (1998), Lehar and Randl (2006), and Bessler and Stanzel (2009) show, findings for the US market regarding overoptimistic earnings forecasts of affiliated analysts also seem to hold for the German equity market. However, evidence taking into account potential effects of recent European financial market regulation is limited to Dumontier and Dubois (2009), Dubois, Fresard and Dumontier (2010 and 2014) and Höfer and Oehler (2014). Dubois, Fresard and Dumontier (2014) examine how MAD affects analyst behaviour regarding stock recommendations in the EU between 1997 and They find that affiliated analysts were overoptimistic before the adoption of MAD and systematically published recommendations above the consensus. However, MAD significantly reduced this optimism. Höfer and Oehler (2014) examine the effects of MAD and MiFID on the median recommendation for the largest IPOs in Europe and differentiate between affiliated and non-affiliated analysts. They conclude that the behaviour of analysts did not change after the passage of the two directives because the median recommendation level of affiliated and non-affiliated analysts remained constant. However, as they only analyse this particular aspect, their results do not allow for a conclusion on whether recent regulation reached its aforementioned aims. To the best of our knowledge, this study is the first to focus on the impact of both MAD and MiFID on earnings forecast optimism. We contribute to the respective body of literature by testing four distinct hypotheses at the individual analyst s level. We thereby extend the research design of Dubois, Fresard and Dumontier (2014) to the effects of MiFID, and expect that conflicts of interest in equity research are mitigated for the post-mifid period in particular, as intended by the directive. Our first hypothesis is motivated by Lehar and Randl (2006), and Bessler and Stanzel (2009) who show that pre-mad conflicts of interest led to biased earnings forecast by affiliated analysts in the German stock market. Following Lehar and Randl (2006), we allow for a potential asymmetry in earnings forecast optimism with respect to forecasts above and below the consensus level: Hypothesis 1a(b): In the pre-mad period, affiliated analysts earnings forecasts are more optimistic than those of non-affiliated analysts in the German stock market for forecasts above (below) the consensus forecast. 5

7 The second hypothesis focuses on recent regulation and addresses the role of MiFID in reducing earnings forecast optimism. According to Dubois, Fresard and Dumontier (2014) the effect of MAD on optimism in recommendations issued by non-affiliated analysts is not significant. However, as MiFID puts stronger emphasis on internal controls and on preventing conflicts of interest than MAD we expect the implementation of MiFID to have a more pronounced mitigating effect on analyst optimism. In addition, in line with Dubois, Fresard and Dumontier (2014) who show that MAD reduced overoptimism of affiliated analysts with regard to recommendations, we expect a stronger post-mifid reduction in affiliated analysts earnings forecast optimism. Taking into account prior evidence by Kadan et al. (2009) that recent US regulation successfully eliminated over-optimism in stock recommendations, but did not succeed in preventing under-pessimism of affiliated analysts, our second hypothesis also allows for such a potential asymmetry in analyst behaviour with respect to forecasts above and below the respective consensus forecast. Hypothesis 2a(b): Earnings forecast optimism is reduced and over-optimism of affiliated analysts is eliminated in the post-mifid period for forecasts above (below) the consensus forecast. There is mixed evidence regarding the impact of firm characteristics on earnings forecast optimism. Dugar and Nathan (1995) show that firm size, measured as the market value of equity, does not have an effect on the findings of relative optimism in earnings forecasts. However, Lim (2001) finds evidence that forecast bias decreases with firm size. Thus, following Lim (2001) we expect to observe a lower impact of affiliation on large firms, and the reduction in optimism after MAD and MiFID is expected to be more pronounced for larger firms earnings forecasts as market and supervisory attention is higher: Hypothesis 3a(b): Earnings forecast optimism is inversely related to firm size for affiliated analysts in particular, and post-mifid earnings forecast optimism is reduced more strongly for larger firms for forecasts above (below) the consensus forecast. Empirical evidence by Lin and McNichols (1998), Cowen, Groysberg and Healy (2006), and Agrawal and Chen (2012) suggests that analysts tend to be more optimistic in the long run, but reduce optimism in earnings forecasts for shorter forecast horizons. This might be an indication of analysts managing their forecasts to reach two competing goal, i.e., attract further in- 6

8 vestment banking business and maintain accuracy (and hence reputation) at the same time. Following this line of reasoning, we expect to observe a stronger impact of affiliation on optimism, as well as a less pronounced reduction in optimism post-mifid for longer forecast horizons. Hypothesis 4(a/b): In case of forecasts above (below) the consensus level, optimism is more pronounced for longer forecast horizons, and MiFID has a stronger impact on shorter forecast horizons especially for affiliated analysts. 3 Data and methodology 3.1 Data and sample construction The sample period reaches from 2001 to 2013, with the pre-mad period ranging from 01/01/2001 to 10/27/2004, the post-mad period from 10/28/2004 to 10/31/2007, and the post- MiFID period from 11/01/2007 to 12/31/2013. Analysts earnings forecasts are taken from I/B/E/S International Detail History database which contains individual forecasts as well as information about the issuing analyst and the respective broker. We focus on annual earnings per share (EPS) forecasts for the actual/first year. Information about analysts affiliation is provided by Security Data Company s (SDC) database. Following Dubois, Fresard and Dumontier (2014), affiliation is defined as follows: Analysts are considered affiliated if they issue an earnings forecast on a firm for which their employer (brokerage firm) has acted as bookrunner or underwriter in a SEO, IPO, or debt issuance, or as M&A advisor over the last 12 months. SDC accordingly provides information on European IPOs, SEOs, debt issuances, and M&A deals. The full I/B/E/S sample comprises 185,503 earnings forecasts from 2001 to 2013 for the German market. In a first step, 15,563 forecasts are deleted because of missing analyst information, or duplicate or ambiguous forecasts. In a second step, 29,237 forecasts with insufficient consensus information are excluded from the sample since earnings forecast optimism is defined as the difference between an analyst s earnings forecast and the consensus forecast, divided by the standard deviation of forecasts for the respective company. The minimum number of available forecasts in order to calculate a meaningful consensus is set to five observations within the last 180 days preceding the respective individual forecast. In a third step, forecasts are excluded when matching them to SDC database is not possible, which may be due to two reasons: On the firm level, about 9,080 forecasts are excluded be- 7

9 cause of a missing link from SDC s CUSIP to the corresponding I/B/E/S Ticker. On a broker level, SDC affiliation information has to be matched manually to earnings forecasts from I/B/E/S, as only full broker names are provided in the SDC database. In order to conduct the matching by full broker names, the latter have to be added to earnings forecasts in I/B/E/S first. This was done manually based on a linking table provided by Thomson Reuters, resulting in a loss of 21,675 earnings forecast observations. The final sample comprises 105,305 earnings forecasts in the period January 2001 to December These forecasts were issued on 268 firms in the German stock market by 2,804 analysts employed by 173 brokers. Table A summarizes the sample selection process. [Please insert table A here] As table B shows, the number of earnings forecasts in our sample increases over the years with a peak in Average analyst coverage lies between10.8 and 18.5 analysts per firm while broker coverage is slightly lower, which indicates that typically only one analyst per broker covers one firm. The number of firms covered per year lies between 117 and 171 firms. [Please insert table B here] Over the entire sample period 4,391 earnings forecasts are issued by affiliated analysts, amounting to four percent of the entire sample of 105,305 forecasts. The fraction of affiliated earnings forecasts increases from two percent in 2001 to six percent in 2013, which is in line with the numbers reported by Dubois, Fresard and Dumontier (2010) for affiliated analysts recommendations. Between 19 and 38 percent of the brokerage firms, and between 6 and 16 percent of the analysts contained in our sample are considered as affiliated. Variations in the fraction of affiliated earnings forecasts, analysts and brokers can to some extend be explained by the total amount of deals from the SDC database. For instance, we observe a drop in the percentages of affiliated brokers and analysts in 2008, and at the same time the number of SDC deals decreases from 325 to 254. [Please insert table C here] 8

10 3.2 Empirical strategy In order to test hypotheses 1 and 2, we first compare earnings forecast optimism for affiliated and non-affiliated analysts in the pre-mad, post-mad and post-mifid periods. Following Clement (1999), Cowen, Groysberg and Healy (2006), and Guan, Lu and Wong (2012), earnings forecast optimism relative to the consensus earnings forecast EF _ OPTitj is determined as the individual earnings forecast EF itj AVE _ EFit for firm i, divided by the standard deviation of earnings forecasts STDEV EF ). Earnings forecast optimism observations are winso- ( it rized at the one percent level in order to avoid potential biases due to extreme outliers. EF _ OPT itj EFitj AVE _ EFit = STDEV ( EF ) it (1) We use EF _ OPTitj as a measure for above consensus forecasts if OPTitj EF _ > 0, and for below consensus forecasts if EF _ OPTitj < 0. In the next step, we compare the difference in mean earnings forecast optimism between affiliated and non-affiliated analysts for forecasts above and below consensus level. The mean comparison analysis tests for differences between affiliated and non-affiliated analysts as well as for differences between the pre-mad, post- MAD, and post-mifid periods, respectively. In addition, we conduct a multivariate OLS regression analysis to measure the impact of affiliation, regulatory changes, and other factors on forecast optimism. Tying in with Kadan et al. (2009), and Dubois, Fresard and Dumontier (2014), the regression model is specified as follows: EF _ OPTitj = α + βpost _ MAD + γpost _ MiFID + δaffiliationitj pre _ MAD + υaffiliationitj post _ MAD + η Affiliatio n post _ MiFID + φcontrol _ variables + ε (2) itj itj Depending on whether the analyst is considered affiliated or not, or zero. Likewise Affiliatio n is either unity, post _ MAD and post _ MiFID are dummy variables which are set to one if the forecast is issued during the post-mad or post-mifid period, respectively, and zero otherwise. In addition, in order to control for the specific influence of recent regulation on affiliat- 9

11 ed analysts, interaction dummies Affiliation itj pre _ MAD, Affiliation itj post _ MAD and Affiliation itj post _ MiFID are included in the regression model. In line with prior studies, we control for other potential determinants of analyst forecast optimism. First, Prior firm return and Prior market return are used to control for a momentum bias that potentially leads analysts to be optimistic (pessimistic) about firms after a positive (negative) prior stock market development. Prior market volatility captures the impact of market uncertainty on earnings forecast optimism. In order to capture the incremental information content of each forecast for a given firm, the control variable Information environment, measured as the number of analysts covering a firm, is included in the regression. Broker size controls for the case that larger institutions might be less prone to deviating from the consensus. Herding controls for optimism driven by an analyst following other analysts signals. Financial crisis and Dotcom crisis control for a general increase in uncertainty during crisis periods. Earnings announcement controls for deviations from the consensus due to new information around an earnings announcement. Table D provides a detailed description of all variables employed. 3 [Please insert table D here] In order to test hypothesis 3, the above two-part analysis incorporating a mean comparison test and a subsequent multiple regression analysis is conducted for the subsamples of small and large firms to control for the influence of the covered firms size on earnings forecast optimism. Finally, the impact of the forecast horizon is measured by splitting the sample into two subsamples with forecast horizons of 0 to 180 days and more than 180 days, respectively. Again, we conduct a mean comparison analysis as well as a multivariate regression analysis in order to test whether hypothesis 4 holds. 3 Control variables chosen are by and large consistent with Dubois, Fresard and Dumontier (2014). However, unlike the latter we do not include a dummy variable accounting for whether it is the first time that a specific analyst issues an opinion on a specific firm ( Initiation ) due to data restrictions. Moreover, our crisis and market volatility variables are not used in their regression model. 10

12 4 Results 4.1 Analyst optimism pre-mad, post-mad, and post-mifid Univariate analysis In case of above consensus forecasts (table E, panel A, columns 2-4) affiliated analysts are significantly more optimistic about firms future earnings than non-affiliated analysts before MAD (1.06 vs. 0.95), which supports hypothesis 1a. Moreover, in the post-mad period affiliated analysts are still significantly overoptimistic (1.02 vs. 0.93). In the post-mifid period however optimism is significantly lower, and there is no significant difference between affiliated and non-affiliated analysts (0.86 vs. 0.86) anymore, which is in line with hypothesis 2a. [Please insert table E here] The results for below consensus forecasts are shown in panel A, columns 5-7, of table E. In the pre-mad period affiliated analysts are significantly less optimistic than their nonaffiliated peers (-0.92 vs ) so hypothesis 1b is rejected. Further, considering affiliated analysts only, the results illustrate that after the implementation of MAD and MiFID pessimism is significantly lower, implying an increase in optimism (-0.92 vs post MAD / post- MiFID). At the same time, non-affiliated analysts pessimism increases significantly from to post-mad, and -0.9 post-mifid. As a result, after MAD/MiFID affiliated analysts are significantly less pessimistic, or more (over-)optimistic than their non-affiliated peers(-0.74 vs post MAD / vs post-mifid). Thus, hypothesis 2b has to be rejected. One reason for the increase in optimism post-mad/mifid might be that analysts are not only less influenced by conflicts of interest, but also suddenly lack access to value relevant (negative) information. If this was the case, it would imply a decrease in affiliated analysts signals information content due to tightened regulation. Focusing on the situation post-mifid, our findings are in line with Kadan et al. (2009) who show that affiliated analysts are reluctant to issue pessimistic recommendations after respective regulatory changes in the US. Regarding the full sample of earnings forecasts (columns 8-10), are affiliated analysts are significantly more optimistic than non-affiliated analysts, but only for the post-mifid period (table E, panel A, columns (2)-(4), 0.06 vs ). In addition, overall analyst optimism seems to have increased after the implementation of MAD in particular, both in terms of deviations of 11

13 individual forecasts from the consensus level, and in terms of the percentage of above consensus forecasts relative to all forecasts (as shown in Table E, panel B) Multiple regression analysis To corroborate the above findings, we conduct a multiple regression analysis, and we again distinguish between earnings forecasts above and below consensus level. In particular, we study the effects of MAD, MiFID and affiliation on earnings forecast optimism. Results are reported in columns 1 and 2 of table F. Concerning the impact of affiliation on above consensus forecasts, the coefficient for affiliated analysts is positive and significant (0.139, significant at the 5% level) in the pre-mad period, supporting hypothesis 1a. In the post-mad period the impact of affiliation is still positive and significant, but considerably lower (0.112, significant at the 5% level). The test of differences in the coefficients Affiliation x pre_mad and Affiliation x post_mad reveals that this difference is significant at the 5% level. Thus, MAD has a mitigating effect on affiliated analysts optimism. In the post-mifid period, there is no significant impact of affiliation, and again the test of differences in the coefficients Affiliation x post_mad and Affiliation x post_mifid is significant. As a consequence, the affiliation bias disappears pos-mifid, which supports our hypothesis 2a. As to the impact of MAD and MiFID, only MiFID seems to reduce optimism of nonaffiliated analysts. In the post-mad period ranging from 10/28/2004 to 11/01/2007, optimism is not reduced. This confirms earlier results reported by Dubois, Fresard and Dumontier (2014) who study analyst recommendations and conclude that non-affiliated analysts optimism is not affected by MAD. In the post-mifid period ranging from 11/01/2007 to 12/31/2013, we find that optimism is reduced by as compared to the pre-mad period (significant at the 1% level). Again, this supports our hypothesis 2a concerning the mitigating effect of MiFID in terms of a reduction in optimism. [Please insert table F here] Results for the regression analysis on below consensus earnings forecasts confirm our findings from the univariate analysis. In the second regression model in table F (column 2), affiliation has a mitigating effect on optimism pre-mad (-0.117, significant at the 1% level). This implies that affiliation leads to overly pessimistic forecasts, which runs contrary to hypothesis 12

14 1b. There is, however, a significant positive impact of affiliation on optimism in the post-mad period (0.083, significant at 10%) and an even higher impact in the post-mifid period (0.098, significant at 1%). The tests of differences in coefficients are significant at the 1% level. This implies that compared to non-affiliated analysts, affiliated analysts seem to have become overoptimistic, which is at odds with hypothesis 2b. After the implementation of MAD and MiFID, affiliated analysts show less negative deviation from the consensus than non-affiliated analysts. This finding could be explained by the information set available to affiliated analysts getting smaller due to higher Chinese Walls.Both MAD and MiFID decrease optimism in general. In the post-mad period optimism is decreased by (significant at the 1% level). In the post- MiFID period optimism, compared to the pre-mad period, decreases by (significant at the 1% level). Thus hypothesis 2b is only supported concerning the impact of MAD and MiFID on optimism of non-affiliated analysts. In this regard, the directives seem to have enhanced the informational content of (non-affiliated) analyst forecasts. The results for the entire sample in column 3 of table F also by and large support the results from the univariate analysis. There seems to be no significant effect of affiliation, which as a comparison of columns 1 and 2 indicates is probably due to offsetting effects from above and below consensus forecasts by affiliated analysts. The impact of the control variables is as expected in all regressions. Positive prior firm return has a negative but small impact on optimism. Prior market return and volatility have a significantly positive effect on optimism which may be driven by a momentum effect and higher overall market uncertainty. A better information environment significantly reduces deviations from the consensus ( for above consensus forecasts and for below consensus forecasts, significant at the 1% level). One explanation could be that the incremental information content of each forecast is reduced on average because more analysts are covering the stock. As expected, broker size is associated with lower deviations from consensus level. Herding has a significantly positive influence on optimism, and during the financial crisis optimism is mitigated significantly, probably because of increased forecasting uncertainty. In contrast, the dotcom crisis is even associated with higher optimism in earnings forecasts. Simultaneous earnings announcements significantly amplify optimism in earnings forecasts. In summary, the coefficients of the control variables are in line with Dubois, Fresard and Dumontier (2014). 13

15 4.2 The effect of firm size on optimism Univariate analysis We split the sample into earnings forecasts for large firms and small firms in order to test hypothesis 3. The mean comparison analysis is conducted for above consensus (table G, panel A) and below consensus forecasts (table G, panel B) separately. For large firms, affiliation only leads to higher optimism in the pre-mad period concerning above consensus forecasts (1.09 vs. 0.95, significant at the 5% level). Both affiliated and non-affiliated analysts exhibit significantly lower optimism in the post-mad and, in particular, post-mifid period. In contrast, for small firms higher optimism of affiliated analysts exists in the post-mad period (1.12 vs. 0.95, significant at the 5% level), but not in the other periods. In general, optimism is reduced for small firms after MAD/MiFID but not as strongly as observed for large firms. However, after MiFID it is still higher for small firms than for large firms (0.91 vs. 0.81, significant at the 1% level). In line with Lim (2001), we find that over the entire period, earnings forecast optimism is higher for smaller firms, supporting our hypothesis 3a. With regard to below consensus forecasts (table G, panel B), we find that for both large firms and small firms, affiliated analysts were less optimistic before MAD/MiFID, but became significantly over-optimistic after implementation of the directives. Surprisingly, optimism is more pronounced for large firms for both affiliated and non-affiliated analysts, which is at odds with our hypothesis 3b. This apparent asymmetry in analyst behaviour may be partly explained by the fact that large firms attract more investor attention and therefore show smaller average deviations from the consensus level of earnings forecasts, as the incremental information content of every individual forecast is smaller. To this effect, panel C shows that affiliated analysts tend to mainly cover large firms, presumably due to the latter s higher involvement in capital market transactions involving investment banks. [Please insert table G here] 14

16 4.2.2 Multiple regression analysis In the next step we extend the baseline regression model from chapter 4.1 and split the sample into large and small firms by adding corresponding interaction dummies. Results for above and below consensus forecasts are reported in table H, columns 2 and 3. The multiple regression analysis for above consensus forecasts confirms the above findings. First, the mitigating effect of MiFID is significantly stronger for large firms compared to small firms (-0.092, significant at the 1% level vs , significant at the 10% level) according to the test of differences in coefficients ( diff (i) (ii): 0.00). Second, affiliated analysts are more optimistic only in the pre-mad period for large firms (0.171, significant at 5% level), but still more optimistic in the post-mad period for small firms (0.184, significant at 5% level). Again, the test of differences in coefficients turns out significant. In the post-mifid period there is no more impact of affiliation on optimism. Even after MiFID analysts seem to be more optimistic regarding small firms compared to large firms, but there is no difference between affiliated and non-affiliated analysts anymore. Thus, post-mifid the affiliation bias is mitigated for small firms in particular, and over-optimism in affiliated analysts forecasts for large firms has already been mitigated with the implementation of MAD. In summary, we conclude that hypothesis 3a is supported. [Please insert table H here] With regard to below consensus forecasts, the regression analysis shows that optimism of non-affiliated analysts is lower for both small and large firms after MiFID ( / , significant at the 1% level), and the effect is even stronger for small firms post-mad (-0.109, significant at the 1% level). Affiliated analysts are significantly less optimistic before MAD especially for small firms. Post- MiFID, affiliated analysts turn over-optimistic for large firms, but do not differ from their non-affiliated peers for small firms. Thus, hypothesis 3b is not fully supported by our results. 4.3 The role of the forecast horizon Univariate analysis To test whether analyst optimism changes with the forecast horizon, the sample is split into forecasts with a horizon of more than 180 days until the announcement of the actual earnings 15

17 (long forecast horizon) and forecast with an horizon of less than 180 days (short forecast horizon). The results of the mean comparison test for above consensus forecasts are reported in table I, panel A. For long forecast horizons affiliated analysts turn out significantly more optimistic before MiFID (1.05 vs. 0.95, significant at 5% level). After MiFID, especially affiliated analysts show lower optimism so that the above difference no longer exists. For short forecast horizons there is no difference in optimism between affiliated and non-affiliated analysts, but optimism for both groups is significantly lower in the post-mifid period. The comparison of the means for long and short horizons reveals that due to a stronger reduction in optimism after MiFID for short horizons, optimism is significantly higher for long horizons post-mifid. In summary, hypothesis 4a is supported. [Please insert table I here] The results for the below consensus forecasts (table I, panel B) first show that for all periods, affiliated and non-affiliated analysts are significantly more optimistic for longer forecast horizons. The pessimism observed for affiliated analysts pre-mad only applies to short forecast horizons. Post- MAD and post-mifid, affiliated analysts are more optimistic than their non-affiliated peers for both long and short forecast horizons. The difference in optimism seems larger for long forecast horizons, which would imply a weaker impact of MAD/MiFID on forecasts over longer horizons. Overall, hypothesis 4b is supported Multivariate analysis In order to corroborate hypothesis 4, we also conduct a multiple regression analysis to study the effects of affiliation and regulatory changes on earnings forecasts for long and short forecast horizons. Table J summarizes the respective results. Regarding above consensus forecasts, in column 2 we find evidence that the impact of affiliation on earnings forecast optimism is significant only for long forecast horizons in the post MAD period (0.123, significant at the 5% level), which lends support to hypothesis 4a. However, after the implementation of MiFID, we do not find any impact of affiliation on optimism for long or short forecast horizons. Regarding the mitigating impact of MAD and MiFID changes on analysts optimism, our results reveal that the strongest effect occurs for short forecast horizons. Moreover, the effects of MiFID are larger than those of MAD for both forecast horizons, which is again in line with hypothesis 4a. 16

18 The results for below consensus forecasts are reported in column 2 of table J. The results reveal that on the one hand the overall reduction in optimism is more pronounced for the short forecast horizon ( vs , significant at the 1% level). On the other hand, post- MAD/MiFID the affiliation bias is more pronounced for the longer forecast horizons (0.0117, significant at the 1% level vs significant at the 5% level). Thus, our results support hypothesis 4b. 5 Conclusion Our findings provide several insights into the effectiveness of recent EU regulation with respect to the reduction of conflicts of interest in equity research. According to our findings MAD and MiFID only partly reached their aim to improve investor protection by reducing optimism in earnings forecasts. Our findings may be summarized as follows: First, we provide evidence that the impact of MiFID on analysts biases is asymmetric, as MiFID lowers over-optimism of affiliated analysts for above consensus forecasts but leads to over-optimism for below consensus forecasts. In our view, this finding has two implications. On the one hand, affiliated analysts were not necessarily influenced by conflicts of interest, but may rather have taken into account information available through their employer`s investment banking ties in their research. On the other hand, MiFID seems to mitigate the dissemination of negative information, and hence reduces the financial market s informational efficiency. In fact, the observed reduction in affiliated analysts pessimism could be attributed to the fact that the latter have lost access to valuable information from their broker s investment banking unit due to higher Chinese Walls. Second, our findings reveal that MiFID mitigates affiliated analysts over-optimism with respect to small firms in particular. Concerning large firms, affiliated analysts overoptimism is already significantly lowered post--mad. One explanation for this might be that regulatory attention of MAD in terms of enforcement focuses on large firms. As a consequence, analysts may be reluctant to change their behaviour with regard to smaller firms, since negative effects due to regulatory actions seem less likely. In this regard, MiFID seems to have indeed enhanced investor protection. Third, our results show that affiliated analysts over-optimism is present mainly for long forecast horizons. This finding suggests that affiliated analysts may be well aware of the trade-off between short-term accuracy and related reputational concerns, and potential rewards to optimism in terms of additional investment banking business created. 17

19 Fourth, forecast optimism is higher for small firms and long forecast horizons for all analysts before and after the implementation of the two directives. MiFID seems to lead to a significant reduction in optimism for all analysts, whereas the highest decrease is observable for large firms, and short forecast horizons. Finally, our findings raise several questions for further research. For instance, it would be interesting to focus on the reduction in non-affiliated analysts optimism post-mifid, and to further analyse the determinants of this development. In addition, another promising area of research may be the analysis of potential changes in the information content of earnings forecasts from an investor s perspective, again with a focus on the post-mifid period. 18

20 Tables Table A Sample selection process earnings forecasts from of German firms This table reports the sample selection process of the earnings forecasts from I/B/E/S database. For every step in the sample selection process, the number of forecasts, analysts, brokers and covered firms is provided. The full sample of German earnings forecasts from 2001 to 2013 is first reduced by forecasts w ith missing analyst information and duplicate/ ambiguous forecasts. Second, forecast for firms w ith insufficient analyst coverage are eliminated. Third, forecasts w ith missing firm or broker link information for merge to SDC database are excluded. Steps in selection process Earnings Forecasts Broker Analysts Firms covered # # # # (1) (2) (3) (4) (5) Full sample of German firm earnings forecasts from I/B/E/S 185, ,560 1,110 Less: Forecasts w ith missing analyst information -3, Duplicate/ambiguous forecasts by analysts -11, Forecasts for firms covered by less than five analysts -29, Forecasts w ith missing firm information for SDC merge -9, Forecasts w ith missing broker information for SDC merge -21, ,038-4 Forecasts for firms w ith data for final analysis -4, Final sample for analysis on forecast optimism 105, , Table B Earnings forecasts final sample by year from of German firms This table reports the distribution of the final sample over the sample period from 2001 to First the total number of earnings forecasts, broker, analysts and firms is reported by year as w ell as the entire period. Second, the average number of earnings forecasts, broker and analysts per firm and year is reported. Year Earnings Forecasts Broker Analysts # Ø per firm # Ø per firm # Ø per firm # (1) (2) (3) (4) (5) (6) (7) (8) , , , , , , , , , , , , , , , , , , , Firms 19

21 Table C Proportion of earnings forecasts from affiliated analysts by year from 2001 to 2013 of German firms This table reports the proportion of earnings forecasts from affiliated analysts over the sample period from 2001 to First the total number of earnings forecasts, broker and, analysts is reported by year. Second, the total number and percentage of affiliated earnings forecasts, broker and analysts is reported. Regarding broker and analysts, the Affil/ %Affil figures indicate w hether the broker or the analyst w as "affiliated" during the year/ period but not necessarily the entire period. In column (11) the number of deals from the SDC database is reported. Year Earnings Forecasts Broker Analysts SDC-Deals # Affil % Affil # Affil % Affil # Affil % Affil # (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) , % % % , % % % , % % % , % % % , % % % , % % % , % % % , % % 1, % , % % 1, % , % % 1, % , % % 1, % , % % % , % % % ,305 4,391 4% % 2, % 3,590 20

22 Table D Definition and sources of variables Variable Defintion Source EF_OPT itj Affiliation pre_mad post_mad post_mifid Earnings forecast optimism, measured as the difference betw een one specific analyst j 's earnings forecast for company i over period t, and the respective consensus forecast, deflated by the standard deviation of all other analysts' respective earnings forecasts for this firm. Affiliation dummy variable that equals 1 if the employer (broker) of the analyst had business ties w ith the covered firm in the year preceding the forecast Dummy variable that equals 1 if the forecast is issued in the period before the implementation of MAD in Germany (01/01/ /27/2004) Dummy variable that equals 1 if the forecast is issued in the period after the implementation of MAD in Germany and before the implementation of MiFID (10/28/ /31/2007) Dummy variable that equals 1 if the forecast is issued in the period after the implementation of MiFID in Germany (11/01/ /31/2013) I/B/E/S SDC Prior firm return Return of the corresponding firm one year preceding the forecast Compustat Prior market return Prior market volatility Information environment Broker size Herding Financial crisis Dotcom crisis Earnings announcement Large firms/ Small firms/ Firm size Long forecast horizon/ Short forecast horizon/ Forecast horizon Value-w eighted return of all listed firms in Germany one year prior to the forecast Standard deviation of the market return one year prior to the forecast Logarithm of the number of analysts covering the firm in the six months prior to the forecast Number of firms covered by the broker in the six months prior to the forecast Dummy variable that equals 1 if in the 10-day-period preceding the forecast at least 2/3 of the forecasts have the same sign of forecast optimisms Dummy variable that equals 1 if the forecast is issued during the financial crisis (08/25/ /25/2009) Dummy variable that equals 1 if the forecast is issued during the Dotcom crisis (04/11/ /09/2002) Dummy variable that equals 1 if the forecast is issued during the 3-day-period (-1,+1) around an earnings announcement The full sample of earnings forecasts is split into small firm and large firm forecasts depending on the market capitalization of the corresponding firm The full sample of earnings forecasts is split into forecasts w ith long (>180 days) and short horizons (<180 days) depending on the number of days until the announcement of the actual earnings. Compustat Compustat I/B/E/S I/B/E/S I/B/E/S - - I/B/E/S Compustat I/B/E/S 21

23 Table E Analysis of differences in earnings forecast optimism of affiliated and non-affilated analysts This table reports the results of the analysis of the difference in optimism for affiliated (affil.) and non-affiliated (non-affil.) analysts. The mean comparison test is reported in panel A. From column (2) to column (3) the mean optimism of affiliated and non-affiliated analysts is show n, respectively. In column (4) the s of the mean comparison test is reported. The analysis is conducted for the pre-mad period, the post-mad period and the post-mifid period. The s regarding the difference in optimism betw een the periods are reported in the row s below. Panel B show s the percentage of above consensus forecasts for affiliated and nonaffiliated analysts. Panel A: Mean comparison test Above consensus forecasts Below consensus forecasts All earnings forecasts Mean affil. Mean nonaffil. Mean affil. Mean nonaffil. Mean affil. Mean nonaffil. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Pre-MAD (i) * ** Post-MAD (ii) ** ** Post-MiFID (iii) *** *** (i) (ii) *** 0.06* 0.00*** 0.00*** (ii) (iii) 0.00*** 0.00*** *** 0.05** 0.00*** *** ** Panel B: Proportion of above consensus forecasts Period Affiliated analysts Non-affiliated analysts All earnings forecast Above consensus forecasts Share of above consensus forecasts All earnings forecast Above consensus forecasts Share of above consensus forecasts (#) (#) (%) (#) (#) (%) (1) (2) (3) (3)/(2) (4) (5) (5)/(4) Pre-MAD % 21,314 9,631 45% Post-MAD % 19,806 10,954 55% Post-MiFID 2,909 1,500 52% 59,794 30,278 51% ,391 2,206 50% 100,914 50,863 50% 22