Thesis BSc Economics and Business Specialization Economics and Finance
University of Amsterdam Faculty of Economics and Business
Supervisor: Lukáš Tóth
The Effect of the Financial Crisis 2008 on the Level of
Forecast Accuracy and Bias of Financial Analysts in the
United States of America
Niklas Kammer 10100253 August 19th, 2013
Amsterdam The Netherlands
Abstract
This study examines the effect of a macroeconomic shock on the forecast behaviour of financial analysts. More specifically, this research emphasises the possible impact of the Financial Crisis 2008 on the accuracy and bias level of financial analysts in the U.S. equity market. Based on a sample of 1420 enlisted companies on the NYSE Stock Exchange, I find that the accuracy of financial analysts does not alter depending on the market conditions after controlling for firm-specific risk. Furthermore, I find significant results that financial analysts are less optimistic after a market crash controlling for firm-specific risk.
Table of Contents
1. Introduction ... 4
2. Literature Review ... 5
a. Definition of Dependent Variables ... 5
b. Forecast Accuracy ... 6
c. Forecast Bias ... 7
3. Development of Hypotheses ... 8
4. Sample Choice and Measures ... 9
a. Sample ... 9
b. Measures ... 10
Forecast Accuracy (ACCURACY) ... 10
Analysts’ Bias (BIAS) ... 10
Financial Crisis Dummy (FC) ... 11
Market Value (SIZE) ... 11
Analysts Following (FOLLOW) ... 12
Dispersion (DISP) ... 12
Volatility (SD) ... 12
Loss Dummy (LOSS) ... 13
5. Empirical Results ... 13
6. Conclusion ... 17
1. Introduction
Financial market crashes are often fuelled by irrational behaviour and expectations of market participants. Financial analysts build the foundation of the investment community and play an important role in the process of price discovery in these financial markets. Likewise, analysts’ forecasts have been accepted in empirical research as a proxy for investors’ earnings expectations, due to their ability to predict earnings more accurately than mechanical models. Moreover, research shows that forecast accuracy matters to investors in their investment decisions (Mikhail et al. 1999). Therefore, the question arises whether financial analysts are influenced by, or are even a reason for optimistic expectations of financial markets during a stock market rally, which ultimately ends in a disaster for most of the people involved.
Finally, the fact that “rewards [for financial analysts] were less sensitive to accuracy and more sensitive to optimism during the stock market boom of the late 1990s” (Hong and Kubik, 2003, p.346) and that financial analysts show herding behaviour (Trueman, 1994) could have major implications for the role of research analysts in financial markets.
On the one hand, studies focusing on the behaviour of financial analysts suggest that different incentives can deteriorate the forecasts’ effectiveness to the disadvantage of investors. On the other hand, the literature lacks evidence on the role financial analysts play in a market crash. I try to fill this void by examining the level of accuracy and bias of analysts during the Financial Crisis and the Credit Crunch of 2008. In 2005, Hope and Kang conducted a research comprising a sample of 21 different countries focusing on the effect of macroeconomic uncertainty on forecast accuracy resulting in a lower accuracy considering a higher uncertainty. Although this research gives a clear indication of the effect of market uncertainty on the forecast accuracy, it fails to adjust for the overall market perception of the future. Therefore, I collect quarterly data for a time spanning from 2005-2010 including only actively traded firms on the NYSE Stock Exchange to evaluate whether the financial crisis affected the accuracy level of financial analysts.
To all my knowledge and after conducting extensive research, I conclude that there is no existing literature regarding the effects of macroeconomic uncertainty on the bias of analysts. However, the nature of my research allows me to build upon
literature concerning the bias level in different time periods to examine a possible effect of the Financial Crisis 2008 on forecasted earnings bias.
This research is organized as follows. First, I will present the literature review including a short summary of the dependent variables ACCURACY and BIAS in the recent literature. Next, the hypotheses are postulated and the methodology of this analysis is discussed. After a description of the sample and the variables included in this research, I present the empirical results. Finally, this paper closes with a discussion of the findings, limitations and the conclusion.
2. Literature Review
a. Definition of Dependent Variables
The literature distinguishes between accuracy and bias. Accuracy is the absolute forecasting error scaled by the share price (see section 4b). Furthermore, there are two types of analyses: analyst and firm level. In analyst-level analysis, the absolute forecasting error is calculated as the absolute value of the difference between actual earnings and the predicted earnings for every single analyst within the sample. In firm-level analysis, a consensus forecast is used, which includes the earnings forecasts of all analysts following the company. In this study, I conduct a firm-level analysis and follow therefore the literature by using the mean consensus as the predicted earnings of the analyst community.
Bias is declared as the difference between the predicted and actual earnings scaled by the share price (see section 4b). Only a few literatures deviate from this approach. Hilary and Hsu (2013) use the median of the consensus forecasts to calculate the bias of analysts, while they use the mean value to calculate their accuracy variable. Hong and Kubik (2003) use a dummy variable, which equals one if the analyst’s forecast is above the average consensus forecast. Therefore, I choose to follow the vast majority of the literature and declare the difference between the predicted and the actual earnings as my bias variable. Moreover, I use the mean of reported forecasts as consensus value to be in line with the scaling of my accuracy variable.
b. Forecast Accuracy
The existing literature varies on the size of forecasting errors by financial analysts. Dreman and Berry (1995) found the forecasting error increasing over time, while Brown (1997) shows that the forecasting error decreases. Brown (1997) additionally presents relationships between the market value of the firm, the absolute value of earnings forecast, analyst following and the accuracy level of financial analysts. Brown et al. (1987) support these results and show a positive and significant influence of firm size on accuracy exists. Although the dependent variable ACCURACY (absolute forecast error) is scaled differently, my research confirms the positive effect of firm size on accuracy.
On the other hand, Kross et al (1990), who investigate the effect of firm specific determinants on stated forecast accuracy, discover a negative effect of firm size on the forecast error. However, Lang and Lundholm (1996) note that the inches of print in the Wall Street Journal (news coverage), as employed by Kross et al. (1990) in their analysis, deteriorates the effect of the variable firm size. The amount of media coverage is an exogenous variable, which depends on the newspaper’s policy and other events. Therefore, I do not expect a negative effect of the market value on ACCURACY.
Alford and Berger (1999) show a significant influence of number of analyst following, volatility and special items on the forecast accuracy. These findings support earlier research by Lang and Lundholm (1996), which indicate that firm disclosures result in reduced volatility of forecast revisions and a more accurate earnings forecasts. My results show that there is no significant influence of number of analysts following on the forecast accuracy, even though I follow Alford and Berger’s approach in scaling the dependent variable ACCURACY and the independent variable FOLLOW (Analysts following).
Further literature focuses on country specific elements to explain variances in accuracy levels across countries. Hope and Kang (2005) find strong evidence that the accuracy of forecasted earnings decreases with higher macroeconomic uncertainty. This result adds to the findings of Hope (2003), which show that a higher level of disclosure as well as a stronger enforcement of accounting rules have positive effects on accuracy. These conclusions are vital to suggest possible results of this research, as I study the effect of a macroeconomic crisis on the forecast accuracy and bias of financial analysts. Though, in contrary to Hope and Kang (2005), I include the
standard deviation of the daily returns of a firm over the quarter instead of a measure of macroeconomic risk to account for uncertainty. This approach differs from Hope and Kang (2005), because I conduct a sample of firm specific measures as opposed to a sample of country specific measures.
c. Forecast Bias
Regarding the bias of financial analysts, prior literature on the analyst-level concludes that financial analysts use optimistic forecasts to gain better access to a firm’s management private information (Das et al., 1998 and Ke and Yu, 2006). However, these findings are in contrast to a more recent study by Hillary and Hsu (2013), which finds that financial analysts deliver pessimistic forecasts to increase their consistency. The differing results can be explained by the fact that the studies employ different assumptions about the core incentives a financial analyst faces – access to superior information or career prospects.
I include the independent variables DISP (Dispersion) and FOLLOW in the model for the bias level, to account for the willingness of financial analysts to issue optimistic forecast reports to favour the management. It can be argued that a higher number of analyst following leads to a stronger competition between analysts, affecting the bias level (Das et al., 1998). The results of the study by Das et al. (1998) show that analyst following has no effect on the bias level. My findings support these results.
Hong and Kubik (2003) find that a more optimistic earnings forecast increases the chance for a financial analyst to be promoted. Moreover, job separation of financial analysts showed to be more sensitive to optimism than accuracy in the financial bubble of 2000, indicating that brokerage houses reward analysts who promote stocks. Therefore it can be reasoned that a higher dispersion amongst analysts might have a positive effect on the bias level, i.e. a higher uncertainty amongst analysts about the firm’s future prospects increases the stated optimism about its earnings. DISP is therefore expected to be positive.
Finally, Hong and Kubik (2003) define BIAS as the optimism or pessimism relative to the consensus of forecasts as opposed to the difference of average forecasted and actual earnings. Nevertheless, this thesis finds supportive evidence that financial analysts promote stocks with optimistic forecasts during a market rally.
3. Development of Hypotheses
Following the literature review, I use this chapter to develop and postulate my two hypotheses. The first hypothesis builds upon the research conducted by Hope and Kang (2005), who find that macroeconomic uncertainty affects the accuracy level negatively.
H1 : The Financial Crisis 2008 affects the accuracy level of financial analysts
significantly.
Controlling for several firm-specific factors, I investigate whether the Financial Crisis 2008 had a significant effect on the accuracy level of financial analysts. First, I construct a model to empirical test whether there was an overall change in the accuracy level, due to the Financial Crisis 2008. The model is as follows:
Model 1: Forecast Accuracy = α0 + α1 (FC) + α2 (DISP) + α3 (LOSS) + α4 (SIZE)
+ α5 (FOLLOW)
Further, I conduct a second model to examine whether the change in the level of accuracy is due to firm-specific uncertainty. In the event of a market crash, market disruptions and an increase in the market uncertainty is expected. During the Financial Crisis 2008, the credit channel and bank-lending channel dried up, pushing insolvent and solvent companies alike into liquidity traps. To ensure that the change of accuracy during the Financial Crisis is not due to the higher uncertainty of the survival of the companies, but rather resulting from a reversed behaviour by financial analysts, I control for the firm-specific risk (volatility of daily stock returns over the quarter). The model is as follows:
Model 2: Forecast Accuracy = α0 + α1 (FC) + α2 (DISP) + α3 (LOSS) + α4 (SIZE)
+ α5 (FOLLOW) + α6 (SD)
My second hypothesis builds upon the research by Hong and Kubik (2003) stating that financial analysts followed incentives to report optimistically biased forecasts
during the financial crisis in the year 2000. I relate this finding to the recent Financial Crisis 2008 and postulate my second hypothesis as follows:
H2 : The Financial Crisis 2008 affects the bias level of financial analysts significantly.
Controlling for several firm-specific factors, I investigate whether the Financial Crisis 2008 had a significant effect on the bias level of financial analysts. I construct a model to empirical test whether there was an overall change in the bias level, due to the Financial Crisis 2008. The model is as follows:
Model 3: Forecast Bias = β0 + β1 (FC) + β2 (DISP) + β3 (LOSS) + β4 (SIZE)
+ β5 (FOLLOW) + β6 (SD)
4. Sample Choice and Measures
a. Sample
The sample consists of 1430 listed companies on the NYSE stock exchange in the United States of America. The included companies are all actively traded during the time span from 2005 to 2010. Choosing the companies enlisted on the NYSE stock exchange ensures that the companies are all subject to the same strict regulations by the Security and Exchange Commission (SEC). Moreover, the data is available without any significant values or time spans missing.
Finally, choosing a U.S. stock exchange controls for other shocks that could influence the results. The United States of America exhibit strong and secure market conditions including the labour and money markets. This fact increases the chance of capturing the pure effect of the financial market shock in the collected sample.
The data is composed quarterly, dividing the sample into 24 periods. Observations containing at least one missing value are dropped from the sample, as they are not included in the pooled OLS-regression. Furthermore, I discard all observations for companies with a share price below three, due to the small denominator effect, i.e. only companies with a share price of three or more enter the regression analysis.
The time span of 2005 to 2010 is chosen to capture and compare pre-financial and post-financial crisis levels. Extending the sample time span could influence the results, due to the aftermath of other recent financial crises, e.g. terrorist attacks on the World-Trade-Center in 2001. Therefore, the sample choice is used to control for externalities of the regression analysis.
The sample comprises 25,572 observations, 11,592 observations from the beginning of 2005 to the end of 2007 and 13,980 observations from the beginning of 2008 to the end of 2010.
Using the I/B/E/S database, the latest reported mean forecast within the forecasting period is used. All other consensus forecasts reported in the quarter are discarded from the sample. The maximum forecast horizon is therefore one quarter of a year.
b. Measures
Forecast Accuracy (ACCURACY)
The first dependent variable, ACCURACY, is measured as the negative of the absolute value of the difference between the consensus (mean) of the forecasted earnings and the actual earnings per share deflated by the share price. The forecasted earnings, the actual value of reported earnings and the share price are collected for each firm from the I/B/E/S database.
𝑨𝑪𝑪𝑼𝑹𝑨𝑹𝑪𝒀 = −𝟏 ∗ 𝒂𝒃𝒔(𝒎𝒆𝒂𝒏 𝒇𝒐𝒓𝒆𝒄𝒂𝒔𝒕𝒆𝒅 𝒆𝒂𝒓𝒏𝒊𝒏𝒈𝒔 − 𝒂𝒄𝒕𝒖𝒂𝒍 𝒆𝒂𝒓𝒏𝒊𝒏𝒈𝒔
𝒑𝒓𝒊𝒄𝒆 )
Analysts’ Bias (BIAS)
The second dependent variable, BIAS, is calculated by subtracting the actual earnings per share from the consensus (mean) of the forecasted earnings per share deflated by the share price. Therefore, a positive value resembles a positive (optimistic) bias by the analyst community, while a negative value suggests a negative (pessimistic) bias.
𝑩𝑰𝑨𝑺 =𝒎𝒆𝒂𝒏 𝒇𝒐𝒓𝒆𝒄𝒂𝒔𝒕𝒆𝒅 𝒆𝒂𝒓𝒏𝒊𝒏𝒈𝒔 − 𝒂𝒄𝒕𝒖𝒂𝒍 𝒆𝒂𝒓𝒏𝒊𝒏𝒈𝒔 𝒑𝒓𝒊𝒄𝒆
Financial Crisis Dummy (FC)
To examine whether the level of accuracy and bias of forecasted earnings change due to the financial crisis, I employ a dummy variable FC. The dummy variable FC splits the sample into two equal parts, each containing 12 quarters (3 years) of observations in the pooled OLS regression. The split-point is the first of January 2008.
Market Value (SIZE)
I include the firm size as a control variable in both models. As Das et al. (1998) note, “firm size has often been used as a proxy for the amount of information that is publicly available” (p.284). This suggestion is in line with Bhushan (1989), who postulated that the benefits of information collection might increase with firm size. Furthermore, Lang and Lundholm (1996) find that the accuracy of forecasted earnings increases with firm size, which is also supported by Brown et al. (1987). Including the firm size in all three models guarantees that the results are not driven by firm size variation, which is important given the market value fluctuations within the sample. During the financial crisis 2008, including the real estate bubble in the United States of America, stock prices in the U.S. equity market fell roughly 50% from October 2007 to March 2009.
Another reason for controlling for firm size is the strong correlation found in earlier studies between firm size and analysts following. Bhushan (1989) and Das et al. (1998) find a strong relation between the firm’s market value and the number of financial analysts issuing forecast reports related to the company.
I construct the variable SIZE as the logarithm of the average quarterly market value for each quarter per firm included in the sample, i.e. the average of the market value at the beginning and the end of the quarter. The market value is calculated by multiplying common shares outstanding at the end of the quarter by the closing share price at same point in time. The data is collected from the Compustat database.
Analysts Following (FOLLOW)
The variable FOLLOW is used to control for the number of analysts following the firm. Behavioural studies suggest a positive effect on forecast accuracy, while the effect on the bias level is vague. An analyst reporting forecasts for a company with fewer analysts following could issue more optimistic predictions to obtain non-public information. This would increase the bias and reduce the accuracy of earnings forecasts. Furthermore, as Das et al. (1998) point out that the following of several analysts on one company could create a higher level of competition between analysts, suggesting a reduced bias and increased accuracy. The number of analysts following a firm is obtained from the I/B/E/S database.
Dispersion (DISP)
Based on evidence by Alford and Berger (1999), I include the dispersion between analysts’ forecasts as a control variable. Alford and Berger (1999) and Hope and Kang (2005) show a negative relationship between forecast accuracy and dispersion between analysts. The effect of dispersion between analysts’ reported forecasts on the bias level is expected to be positive. The variable DISP is constructed by obtaining data from the I/B/E/S database, which provides the standard deviation between the earnings forecasts. DISP is used in per-cent values.
Volatility (SD)
Hope and Kang (2005) find a negative relationship between macroeconomic uncertainty and forecast accuracy. Due to the nature of the research question, firm-specific uncertainty (SD) is included as a control variable for the variable of interest FC. During a financial crisis, an increase in the firm-specific uncertainty is expected, given the definition of a crisis. Therefore, I control for uncertainty to overcome the effect that the difference of accuracy before and after the Financial Crisis 2008 is driven by market uncertainty. This research employs the equity volatility SD of each firm included in the sample for every quarter to account for firm-specific uncertainty. The firm-specific volatility is the standard deviation of daily stock returns, calculated by obtaining daily stock prices from the CRISP database. SD is used in per-cent values.
Loss Dummy (LOSS)
According to Hwang et al. (1996), analysts’ accuracy decreases in case the company reports negative earnings. Hope and Kang (2005) support these results and find a negative relationship between forecast accuracy and reported losses. Therefore, I include a dummy variable LOSS controlling for negative earnings.
5. Empirical Results
Table 1 presents the descriptive statistics. The table shows that over the time period 2005 to 2010, the forecasted earnings are on average slightly positive or optimistic, namely 0.17302% of the stock price. As a comparison, the sample of Das et al. (1998) shows a stronger and negative bias of 1.5% to 3.5%. The forecast accuracy level of -0.68% of the stock price is relatively small, compared to results in other studies. Lang and Lundholm’s (1996) sample showed a mean forecast accuracy level of 4.2%, while Hope and Kang’s (2005) forecast error was -2.7% on average. Furthermore, the table shows that the observations are equally spread over both time periods of interest, before and after the financial crisis. The mean of the dummy variable FC is 54.67%.
TABLE 1
Descriptive Statistics (period 2005q1-2010q4) Sample size = 1430 firms (1 obs per firm per quarter)
Obs. per firm mean= 17.8825, min=1, max=24
Variable Obs Mean Std. Dev. Min Max
BIAS 25572 .0017302 .0375842 -1.127877 1.716418 ACCURACY 25572 -.0068921 .0369873 -1.716418 0 FOLLOW 25572 9.303965 6.105253 2 39 SD 25572 10.7964 9.948247 1.0406 204.2178 SIZE 25572 7.928571 1.474054 3.575908 13.13227 DISP 25572 5.758838 12.60369 0 485 LOSS 25572 .0915845 .2884442 0 1 FC 25572 .5466917 .4978248 0 1
Table 2 shows the Pearson correlations. The correlation matrix exhibits that every explanatory variable is significantly correlated to both dependent variables (ACCURACY and BIAS) in the predicted direction. FC is negatively and significantly correlated with ACCURACY. This is consistent with the first hypothesis that the occurrence of the Financial Crisis had an impact on the forecast accuracy level of financial analysts. Furthermore, FC is positively and significantly correlated with the dependent variables BIAS. This is in line with the second hypothesis.
FOLLOW and SIZE are strongly correlated (r=0.5857) as predicted by prior research. Nevertheless, all 3 models do not suffer from multicollinearity.
Table 2 Correlation Matrix
ACCURACY BIAS FOLLOW DISP LOSS FC SD
ACCURACY 1.0000 BIAS -0.7226 1.0000 FOLLOW 0.0579 -0.0221 1.0000 DISP -0.2628 0.1745 -0.0131 1.0000 LOSS -0.2716 0.2694 -0.0953 0.1931 1.0000 FC -0.0736 0.0163 0.0183 0.0709 0.1031 1.0000 SD -0.2172 0.1295 -0.0728 0.1339 0.2163 0.2354 1.0000 SIZE 0.1092 -0.0509 0.5857 0.0321 -0.1937 -0.0938 -0.2268
The number of observations is 25,572 for all correlations. The reported correlation coefficients are Pearson correlations. The boldfaced correlation coefficient is significant at the 5% level. All other correlation coefficients are significant at the 1% level.
Table 3 shows the pooled-OLS regression of the first hypothesis including Model 1 and Model 2. The number of analysts following (FOLLOW) is not significant and has no influence on the level of accuracy, neither in Model 1 nor in Model 2. This might suggest that the level of available information does not increase with the number of analysts following. Hence, regulations by the financial authorities like the SEC to make public information available to every analyst alike and the restrictions of insider tips might be effective. Furthermore, a higher level of competition, i.e. more analysts following, does not increase accuracy as suggested by Das et al. (1998).
The standard deviation between analysts’ forecasts (DISP) shows the predicted and anticipated direction, as well as the dummy variable LOSS. Even
though the influences are small compared to Hope and Kang (2005), DISP and LOSS are both highly significant in explaining the variation of the level of accuracy. The results confirm Hope and Kang’s (2005) conclusion that financial analysts find it harder to predict the earnings of a company that is reporting losses. Dispersion among analysts’ forecasts decreases accuracy, as predicted by earlier studies (Alford and Berger, 1999 and Hope and Kang, 2005).
The market value of the firm is positively related to the accuracy level as predicted. This finding supports the hypothesis that bigger firms are easier to forecast, due to more stable operations or a higher degree of available information in the public market.
Model 1 shows that the level of forecast accuracy decreased after the outbreak of the financial crisis. The result is highly significant, but controlling for firm-specific risk in Model 2 resolves this effect. After including the volatility of daily stock returns per firm in the regression, the variable of interest FC looses its explanatory power. This result suggests that financial analysts’ accuracy is not influenced by optimistic or pessimistic market perceptions. Moreover, financial analysts’ performances appear to be only affected by the level of disclosures and information available within the financial markets. Finally, analysts are showing a constant accuracy level regardless of the financial markets experiencing a boom or bust, after taking firm-specific risk measures into account.
My second hypothesis questions whether the occurrence of the Financial Crisis 2008 had an effect on the earnings forecast bias of financial analysts. Table 4 shows the pooled-OLS regression of Model 3.
FOLLOW and SIZE are both insignificant. DISP is positively related to forecast bias. This suggests that in case of a stronger uncertainty in the community of financial analysts about a firm’s earnings, the collective of financial analysts reports more optimistic earnings forecasts. The same is true for SD (Volatility), which is consistent with prior studies. Higher uncertainty increases optimism about company’s earnings. Both findings are highly significant. These results can be explained by behavioural characteristics of financial analysts like herding behaviour or simple career concerns. Sell-side analysts might be more optimistic about volatile stocks to keep them attractive for investors.
Table 3
Pooled-OLS Regressions of H1
Regression: ACCURACY = α0 + α1 (FC) + α2 (DISP) + α3 (LOSS) + α4 (SIZE) + α5 (FOLLOW)
+ α6 (SD)
Independent Variables Prediction Model 1 Model 2
FC (α1) ? -0.0021079 -0.000173 (-5.04***) (-0.36) DISP (α2) - -0.0006534 -0.0006129 (-7.36***) (-7.14***) LOSS (α3) - -0.0271132 -0.02467 (-14.20***) (-13.97***) SIZE (α4) + 0.0019853 0.001211 (7.65***) (4.70***) FOLLOW (α5) + -0.0000667 -0.0000080 (-1.45) (-0.18) SD (α6) - -0.0005064 (-8.08***) F-statistic 68.76 58.62 Adj. R2 0.13 0.14
***,** and * denote significance at the 1%, 5% and 10% levels, respectively.
The number of observations is 25,572 for the regression model (See Table 1). Intercept is included but not reported.
The variable LOSS is positive and highly significant, signalling a more optimistic forecasting for companies reporting losses.
Finally, the variable of interest, FC, is highly significant and negative. This shows that financial analysts issue more pessimistic forecasts for companies listed on the NYSE Stock Exchange after the Financial Crisis unravelled.
This result can have several implications. First, financial analysts might be involved in creating an overly optimistic set of expectations in the investment community before a financial crisis, by overstating earnings estimates. Second, the bias of analysts merely reflects the overly positive anticipations of the investment crowd. And finally, financial analysts might trade their bias for accuracy or consistency as suggested by Hilary and Hsu (2013). Given the turmoil and uncertainty in the financial markets after an event like the Financial Crisis 2008, decreasing their bias of positive earnings surprises might be a rational trait. The overall pessimistic perception in the financial markets and the slowing economy after a major financial
shock might justify the more negative bias. However, this study does not examine whether this change in the level of biased forecasts is rational.
Table 4
Pooled-OLS Regressions of H2
Regression: BIAS = β0 + β1 (FC) + β2 (DISP) + β3 (LOSS) + β4 (SIZE)
+ β5 (FOLLOW) + β6 (SD)
Independent Variables Prediction Model 3
FC (β1) ? -0.002523 (-5.13***) DISP (β2) + 0.0003648 (4.39***) LOSS (β3) + 0.0305554 (-16.88***) SIZE (β4) - -0.0000359 (-0.13) FOLLOW (β5) ? 0.0000516 (1.09) SD (β6) + 0.0002666 (4.05***) F-statistic 68.76 Adj. R2 0.13
***,** and * denote significance at the 1%, 5% and 10% levels, respectively.
The number of observations is 25,572 for both regression models (See Table 1). Intercept is included but not reported.
6. Conclusion
This study investigates the effect of a financial crisis, like the real estate bubble in 2007, on the forecast accuracy level and bias of predicted earnings. The sample consists 1430 companies listed on the NYSE Stock Exchange in New York, USA. The main reasons for choosing this sample are the market stability, including labour markets, financial markets, money markets, and the level of regulations of the financial and corporate markets. Using data from the NYSE Stock Exchange reduces the chance that other shocks, apart from the Financial Crisis 2008, influence the outcomes.
The empirical results show no effect of the Financial Crisis on the accuracy level of forecasted earnings by financial analysts. There is no indication within this study that financial analysts are less or more accurate before or after a market shock given the information environment and firm-specific uncertainty.
Furthermore, I show that the Financial Crisis had a significant and negative effect on the bias level of financial analysts. Reports issued by analysts appear to be more pessimistic after the financial crisis than before. However, this study does not examine whether this change in the level of biased forecasts is rational. The overall pessimistic perception in the financial markets and the slowing economy after a major financial shock might justify the more pessimistic bias.
Financial analysts fulfil an important role in the act of price discovery and value generation in financial markets. Moreover, researchers tend to use earnings forecasts as estimations for market anticipations and expectancies. Therefore, it is essential for investors and the academic community to be aware that financial analysts might be more negatively biased after a financial market shock.
One of the main limitations of this study is the use of firm-level data, which is averaged per company, hence excluding possible valuable information on the analyst-level. Further, the low explanatory power of the models of the variation in the level of accuracy and bias given the vast sample collected might pose a limitation to this study. Thirdly, the high regulations on financial markets set by the SEC in the United States of America control for possible changes in accuracy level and bias. An application to less regulated financial markets might yield different results. Thus, the high regulations by the SEC jeopardize the external validity of this study. Another threat to the external validity of this study are the characteristics of the Financial Crisis 2008. The results of this study might not be of significance for other financial market malfunctions, as every financial crisis tends to be of different nature.
Even though this study gives some valuable insights in the effect of a financial market shock on the accuracy and bias of financial analysts, more research is needed to confirm the results, improve the models and test the hypothesis on different samples.
7. Bibliography
Alford, Andrew W. and Philip G. Berger. 1999. "A Simultaneous Equations Analysis of Forecast Accuracy, Analyst Following, and Trading Volume." Journal of Accounting, Auditing & Finance 14 (3): 219-240.
Brown, Lawrence D. 1997. "Analyst Forecasting Errors: Additional Evidence." Financial Analysts Journal 53 (6): 81-88.
Brown, Lawrence D., Gordon D. Richardson, and Steven J. Schwager. 1987. "An Information Interpretation of Financial Analyst Superiority in Forecasting Earnings." Journal of Accounting Research 25 (1): 49-67.
Bhushan, Ravi. 1989. “Firm characteristics and analyst following.” Journal of Accounting and Economics 11 (2–3): 255-274.
Das, Somnath, Carolyn B. Levine, and K. Sivaramakrishnan. 1998. "Earnings Predictability and Bias in Analysts' Earnings Forecasts." The Accounting Review 73 (2): 277-294.
Dreman, David N. and Michael A. Berry. 1995. "Analyst Forecasting Errors and their Implications for Security Analysis." Financial Analysts Journal 51 (3): 30-41.
Hilary, Gilles and Charles Hsu. 2013. "Analyst Forecast Consistency." The Journal of Finance 68 (1): 271-297.
Hong, Harrison and Jeffrey D. Kubik. 2003. "Analyzing the Analysts: Career Concerns and Biased Earnings Forecasts." The Journal of Finance 58 (1): 313-351.
Hope, Ole-Kristian. 2003. "Disclosure Practices, Enforcement of Accounting Standards, and Analysts' Forecast Accuracy: An International Study." Journal of Accounting Research 41 (2): 235-272.
Hope, Ole-Kristian and Tony Kang (2005) The Association between Macroeconomic Uncertainty and Analysts' Forecast Accuracy. Journal of International Accounting Research: Spring, Vol. 4, No. 1, pp. 23-38.
Hwang, L., C. Jan, and S. Basu. 1996. “Loss firms and analysts’ earnings forecast errors.” Journal of Financial Statement Analysis 1 (Winter): 18-31.
Ke, Bin and Yong Yu. 2006. "The Effect of Issuing Biased Earnings Forecasts on Analysts' Access to Management and Survival." Journal of Accounting Research 44 (5): 965-999.
Kross, William, Byung Ro, and Douglas Schroeder. 1990. "Earnings Expectations: The Analysts' Information Advantage." The Accounting Review 65 (2): 461 - 476.
Lang, Mark H. and Russell J. Lundholm. 1996. "Corporate Disclosure Policy and Analyst Behavior." The Accounting Review 71 (4): 467-492.
Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis. “Does Forecast Accuracy Matter to Security Analysts?”. The Accounting Review: April 1999, Vol. 74, No. 2, pp. 185-200.
Trueman, B. 1994. "Analyst Forecasts and Herding Behavior." Review of Financial Studies 7 (1): 97-124.
Disclosure
Wharton Research Data Services (WRDS) was used in preparing this research paper. This service and the data available thereon constitute valuable intellectual property and trade secrets of WRDS and/or its third-party suppliers.
