Forecast accuracy of analysts
during the global financial crisis
Loes van der Jagt – 10244611 June, 2015 - University of Amsterdam Bsc in Business Economics, field Finance
Supervisor: Dr. J.E. Ligterink
Abstract
This paper examines the effect of the global financial crisis on the accuracy of analysts’ forecast errors. Therefore 388 companies which were listed on the S&P500 at the end of the sample period are investigated. As expected the accuracy of earnings forecast by analysts decreased during the crisis compared to the pre-crisis and post-crisis. The amount of debt seems to have a negative effect on the forecast error which is opposite of previous studies. But the error of forecasts for firms with more debt is bigger during the crisis period than before and after the crisis. The size of the firm has no large significant effect on the forecast error. A reason for this could be that the sample consists of large firms.
Table of Contents
1. Introduction ... 2
2. Literature Review ... 4
2.1 Biased forecasts ... 4
2.2 Reasons for optimism ... 4
2.2.2 Strategic reporting bias explanation ... 5
2.2.2 Selection bias explanation ... 5
2.2.3 Cognitive bias explanation ... 6
2.3 Influence of debt ... 6
2.4 Influence of firm size ... 7
2.5 Hypotheses ... 8 3. Methodology ... 10 3.1 Sample Selection ... 10 3.2 Method ... 11 4. Results ... 13 4.1 Summary statistics ... 13
4.2 Absolute Forecast Error ... 13
4.3 Signed Forecast Error ... 16
4.4 Forecast Dispersion ... 16
4.5 Remarks on the model ... 17
5. Conclusion and Remarks ... 19
References ... 20
2
1. Introduction:
The accuracy of analysts’ earnings forecasts is widely studied. It’s an important topic because many professionals rely on the forecasts of these analysts. For example investment bankers, financial advisors and stockbrokers. But also individual investors attach great importance to the forecasts of analysts (Ciccone, 2002). Analysts play an important role in the smooth operation of the capital market. It is assumed that they could better predict earnings than others could do because they are experts in collecting and processing information of firms (Ang and Ma, 2001). According to Loh and Mian (2003) good forecasts have the highest value during periods of heightened uncertainty. In the crisis of 2008 there was a lot of uncertainty and therefore the forecasts of analysts have higher values.
This study examines the effect of the global financial crisis on the accuracy of financial analysts. Previous studies showed that the accuracy is decreasing during
macroeconomic uncertainty (Sidhu and Tan, 2011; Hope and Kang, 2005). I will examine if the same holds for firms of the S&P5000 during the crisis of 2008.
The size of the firm and the level of debt are taken into account as control variables. It is proven that if the amount of debt increases, the accuracy will decrease (Bradshaw,
Richardson and Sloan, 2006; Vanstraelen et al., 2003). A higher amount of leverage causes greater volatility of earnings which makes it harder to predict future earnings. This leads to less accurate forecasts. Besides that also firm size has an influence on the accuracy of analysts. The bigger the firm the more accurate the forecasts are (Basi, et al., 1976; Atiase, 1985; Jaggi and Jain, 1998; Lim, 2001; Thomas 2002; Brown, Richardson and Scwager, 1987b). A reason for this could be that larger firms are more transparent and followed by more analysts (Eddy and Seifert 1992; Zhang, 2006; Chen et al., 2010; Hussain, 2000; Lim, 2001). Ang and Ma (2001) conclude that analysts were more accurate for large firms
compared to small firms during the Asian financial crisis. This study examines if this is also the case during the global financial crisis of 2008. Besides that it will also be investigated if the level of debt has influence on the accuracy of analysts during the crisis.
3
Previous research questioned the accuracy of analysts and it is found that on average analysts are too optimistic (De Bondt and Forbes, 1999; O’Brien, 1988, De Bondt and Thaler, 1990; Richardson et al., 2004). One reason for this could be that the optimistic forecasts are made to improve the relations with the management of the firms and therefore will ensure access to private information (Das, Levine and Sivaramakrishnan, 1998; Lim, 2001). The second reason could be to improve the market perceptions of firms. As a result analysts could benefit from covering these firms which subsequently do well (McNichols and O’Brien, 1997). Another reason is that optimism could serve as an incentive to encourage trading (De Bondt and Thaler 1990). Finally, optimistic expectations are often present in our everyday life and may be in our nature (Neil Weinstein, 1980).
This paper will examine if the accuracy during the financial crisis decreases because it is harder to predict earning during periods of macroeconomic uncertainty. Besides that it will be investigated if the accuracy increases when the crisis ended. The idea behind this is that the macroeconomic uncertainty decreased and therefore the earnings become better to predict and the forecast accuracy will increase. The size of the firm and the level of debt are taken into account because earnings for small firms or firms with a high debt ratio are harder to predict and I will examine if forecasting earnings of firms which are hard to predict become even harder during a crisis.
To answer the questions stated above, the paper is structured as follows. The next section will be a review of the relevant literature to provide the theoretical basis and discuss previously published findings on this topic. At the end of this section the hypotheses are formulated. The research methodology will be discussed in Section 3. In Section 4, the results are provided and will be analyzed to answer the research question. In the last section the most important findings will be given. Besides that some remarks and suggestions for future research on this topic will be made.
4
2. Literature Review
This section serves as a review of the relevant literature. The most severe bias of analysts is optimism and this will be discussed at first. Then the reasons for optimism are explained. Next the influence of debt on the accuracy will be discussed an after that I will look at the influence of the size of the firm.
2.1 Biased forecasts
Many researchers investigated the accuracy of analysts’ earnings forecasts. Several studies showed that analysts could forecast earnings better than time-series models (Brown 1987; O’Brien 1988). But there are also some limitations about the accuracy of analysts’ forecasts. Many studies showed that besides the fact that analysts are more accurate than time-series models, the forecasts have some severe biases.
One of the greatest biases of analyst forecasts is that nearly all of them are excessively optimistic (De Bondt and Forbes 1999; O’Brien 1988; De Bondt and Thaler 1990). Optimism means that the earnings forecasts of the analysts are structurally higher than the actual
earnings. Especially at long time horizons there is a systematic optimism, but this optimism will shrink when the announcement date of earnings becomes closer. This means that
forecasts become more accurate and less biased as they approach the earnings announcement date (Richardson et al. 2004; Ramnath et al. 2008). O’Brien (1988) implies that the most current forecast available is more accurate than either the mean or the median of all available forecasts (Chrichfield, Dyckman and Lakonishok, 1978). This seems logical because analyst could incorporate new information into their forecast as the year progresses (O’Brien, 1988; Collins and Hopwood, 1980; Elton, Gruber and Gultekin, 1984). However according to Richardson et al. (2004) there will be analyst optimism at all forecast horizons, not only at long time horizons. But he agrees that the forecasts become increasingly less optimistic as the horizon shrinks towards the announcement date.
2.2. Reasons for optimism
There are several reasons why analysts might be too optimistic. These reasons could be
divided in three groups. The strategic reporting bias assumes that analysts are rational but they issue optimistic forecasts to promote revenue-generating business for their brokerage firms and to facilitate information access to the management. The second explanation for optimism is based on a selection bias. It assumes that analysts are rational and truthful but they only release forecasts for firms for which they have positive expectations. The last category is the
5
cognitive bias explanation, it assumes that analysts are irrational forecasters who systematically make too optimistic forecasts. (Gu and Wu, 2003)
2.2.1 Strategic reporting bias explanation
One reason for optimism could be the strategic reporting bias, which actually consists of two things. The first is that the optimistic forecasts are made to improve the relations with
corporate management and ensure access to private information (Das, Levine and
Sivaramakrishnan,1998). The company’s management is an important source of nonpublic company information. For managers it is favorable if forecasts are high because this supports higher capital market valuations and leads to higher compensation levels. The information which is given to the analysts could be limited if analysts issue unfavorable forecasts, even if these forecasts are justified. Especially for companies with more uncertain information environments the management’s information is very important and therefore uncertain information environments are associated with more optimistic forecasts (Lim, 2001).
Secondly, strong pressure from the company management for which the analyst works could encourage too optimistic forecasts. Many analysts work for brokerage and investment banking firms. These firms make money by encouraging trading. All customers are
potentially interested in a buy recommendation, but only the current stock holders are interested in a sell recommendation. That’s why optimistic forecasts are more profitable, because these could serve as an incentive to encourage trading. (De Bondt and Thaler 1990).
2.2.2 Selection bias explanation
Instead of making forecasts by purpose too optimistic, another method to make more profitable forecasts is to make only forecasts for companies of which their true expectations are favorable. If analysts do not report their expectations when these are sufficiently low, then the lower tail of the distribution of forecasts will not be published. Therefore the average of the published forecasts will be higher than the (unobserved) mean of all forecasts. It is showed that the ratings which analysts assign to stocks they have just added to their lists of followed stocks are heavily weighted towards “Strong Buy” recommendations compared with ratings of stocks with previous recommendations. Analysts stop covering stocks which have lower ratings than those whose coverage continues. Also the realized return on equity is higher for stocks that analysts add to their covering lists than for stocks that analysts covered in the past and for stocks that analysts subsequently drop. This shows that analysts prefer to follow stocks with favorable future expectations. If analysts stop covering a company or avoid
6
updating their forecast when new unfavorable information is available, the sample of
observed forecasts will be on average too high. This is because the analyst’s prior forecast is no longer a good forecast given the unfavorable information received by the analyst. Analysts do not report forecasts when they stop covering a stock, so the last forecast prior to dropping the stock from coverage will fail to reflect the negative information that induced the drop decision. Therefore at least some portion of the phenomenon that analysts’ forecasts of earnings are too optimistic could arise from self-selection by censoring unfavorable forecasts. (McNichols and O’Brien, 1997).
2.2.3 Cognitive bias explanation
Analysts are assumed to be irrational and systematically make to optimistic forecasts. There is a problem by analysts in how they react to information. According to Easterwood and Nutt (1999) analysts both underreact to negative information but on the other side overreact to positive information. Taken this together the forecasts will be on average too optimistic. This is the same when you look at analysts’ revisions of forecasts in response to the prior year’s forecast error. It is found that analysts underreact to abnormally negative forecast errors and overreact to abnormal positive forecast errors. These findings are consistent with analyst being irrational and systematically optimistic.
The unrealistic optimism by the forecasts of analysts is also consistent with a general experiment of Neil Weinstein (1980) and others who find overoptimistic expectations of individuals in everyday life. The study of Weinstein consists of an experiment in which he investigated the tendency of people to be unrealistic optimistic about future life events. He asked students about their expectations about their own chances in several events and found that they see their own chances above average for positive events but their chances for negative events below average. This shows that it looks like it is in our nature to be overoptimistic.
2.3 Influence of debt
In this study I look at the influence of debt and firm size on the accuracy and the optimism by analysts. A study about earnings forecasts by analysts in Belgium, Germany and the
Netherlands shows that a higher amount of leverage is related to a lower accuracy (Vanstraelen et al. 2003). More financial leverage causes greater volatility of earnings.
Greater variability in earnings makes it harder to predict earnings which leads to less-accurate forecast. Another consequence of more variability in earnings is that there would be more
7
dispersion among forecasts, which means that the forecasts vary more between analysts. (Eddy and Seifert, 1992; Thomas, 2002; Hope, 2003).
Also Bradshaw, Richardson and Sloan (2006) investigate the influence of debt on earnings forecast error. They conclude that the degree of overoptimism, on both one and two year forecasts, is increasing when debt increases. Because overoptimism means the forecasts of earnings by analysts are higher than the realized earnings, overoptimism leads to less accurate forecasts. The accuracy increases when analysts downgrade their forecasts.
Bradshaw et al conclude that there is a negative relation between external financing activities and future stock returns. It is suggested that companies attract external capital if they think that their shares are overvalued. A reason for this could be that analysts are overoptimistic, which leads to prices of shares being too high and force companies to attract external funding. Especially when there is on the short run overoptimism by analysts, companies use debt. So companies with more debt have on average short run overoptimistic analysts. Thus according to Bradshaw et al (2006), the amount of debt is more a consequence rather than a cause for optimism. All together it can be concluded that the greater the financial leverage of a firm is, the greater the prediction error will be. (Eddy and Seifert 1992; Chen et al., 2010).
2.4 Influence of firm size
The other firm characteristic that is included in the analysis is the size of the firm. Also this characteristic seems to influence the forecast error. Earlier studies showed that analysts’ earnings forecasts are more accurate for larger firms than they are for smaller firms (Basi, et al., 1976; Atiase, 1985; Jaggi and Jain, 1998; Lim, 2001; Thomas 2002). Brown, Richardson and Schwager (1987b) conclude that the superiority of I/B/E/S Summary forecasts over a random walk down increases with firm size. Larger firms also have less forecast dispersion (Thomas, 2002).
One reason that forecast errors for large firms are smaller, is that there is more
information available about them (Eddy and Seifert 1992). In their study they placed firms in four different groups based on their size. They found that the smallest group has a
significantly higher prediction error than any of the other three groups. They assumed that the reason for this is that there is less information generated about small firms. Large firms are more likely to be transparent, so more information will be available for these firms. The larger firms also tend to attract more public attention and media coverage, which also leads to more available information (Chen et al., 2010). According to Zhang (2006) firm size is often used as a proxy for the amount of information that is publicly available about a firm. He thinks it is
8
plausible that large firms have more information available for the market than small firms. Larger firms have more customers, suppliers, shareholders and these firms may be more able to bear high disclosure preparation costs. So the general information environment is likely to be richer for large companies than for small ones.
Second, because there is less information about smaller firms, analysts have stronger incentives to issue more optimistic forecasts for smaller firms. This is in line with the strategic reporting bias explanation (Gu and Wu, 2003) which is stated above. Analysts make higher forecasts about firms to improve relations with the management to create access to private information. Because there is less public information for small firms, management
communication is in this perspective more important for smaller firms than for larger firms. So the forecasts for smaller firms will be more influenced by improving management relations and therefore the forecasts will be more optimistic and therefore the accuracy will be lower.
2.5 Hypotheses
From previous research it can be concluded that analysts are biased and on average too optimistic. The first research question in this paper is whether the forecast accuracy will be the same during a period of crisis compared to the period before and after the crisis. It seems logical that the accuracy will decrease when the financial crisis started in 2008 because it was an unforeseen big event in the economy, which had an influence on almost all companies and no one had foreseen it.
Earlier research confirms the intuition that the forecast error becomes larger during a financial crisis (Sidhu and Tan, 2011). The same study also shows that the forecast dispersion becomes bigger. The forecast dispersion could be seen as the degree of disagreement among analysts. (Sidhu and Tan, 2011). Another research also shows that the forecast accuracy decreases in the level of macroeconomic uncertainty (Hope and Kang, 2005). Although Hope and Kang took inflation and foreign exchange volatility as measures of macroeconomic uncertainty, it seems logical that their outcomes are also valuable for the global financial crisis which is the subject of this paper. A research about the Asian financial crisis showed that analysts not only failed to anticipate on the uncertain environment before the financial crisis but also refuse to learn after the crash. They did not use the available market
information after the crisis to significantly downgraded their forecasts. (Ang and Ma 2001; Coën and Desfleurs, 2004). During periods of negative earnings growth, which is mostly the case during a crisis, their forecasts will become overly optimistic. (Ding et al., 2004). There is
9
no evidence that analysts panicked during a financial crisis and go from too optimistic to too pessimistic (Ang and Ma, 2001)
The second question is what effect debt and leverage, which is concluded to have an effect on forecast accuracy during normal economic periods, have during a crisis. As stated above, debt causes greater volatility of earnings. The effect of greater volatility is that the earnings become harder to predict, so the forecast will become less accurate. It could be that that this property will become more severe in times of more uncertainty. There could also be an interaction effect between firm size and the crisis. Larger firms mostly have more stable incomes and their earnings are more differentiated. For this reason the forecast of these firms might be less affected by the crisis because these large firms are on average more stable.
The focus of this study is if the financial crisis has an influence on the accuracy of analysts’ earnings forecasts and if the degree of disagreement between analysts become larger. Therefore the hypotheses are:
𝑯𝑯𝑯𝑯𝟎𝟎: the forecast error will become larger during the financial crisis and will decrease after
the crisis has ended
𝑯𝑯𝑯𝑯𝟎𝟎: the forecast dispersion will become larger during the financial crisis and will decrease
10
3. Methodology
This section first discusses the companies that were included in this study and which time span is taken. Also the disadvantages of this sample selection are explained. After that the definition of the variables used is discussed and the method of the of investigation is described.
3.1 Sample selection
Forecasted and actual earnings are extracted from the Institutional Brokers’ Equity System (I/B/E/S). Information about the firm characteristics are gathered via the databases of
COMPUSTAT. In this study we look at all companies which are listed at the S&P 500 at the end of 2014. A selection is made with only companies for which earnings forecast and actual earnings data are available for the fiscal year ends between 31 December 2005 and 31
December 2014. This resulted in a data sample of 386 companies with every month a forecast for the end of the fiscal year over a period of 10 years. Firms on the S&P 500 are chosen because it’s one of the most commonly followed indices and it is considered as one of the best representations of the stock market of the United States. As a consequence these firms also are followed by many analysts so much forecasts are made for these companies which results in more data for this research.
The disadvantage of this selection is that we must take into account that this selection is subject to a sampling bias. Firms which are part of the S&P500 in 2005 but are not in 2014, are not included in this study. The other way around, firms that were not that large in 2005 but grow their way to inclusion in the S&P500 in 2014 are part of the analysis. Since we consider that the size of the firm has an effect on the accuracy this is an important point to mention. As a fact only firms which were small at the beginning of the time span but grew to one of the 500 biggest companies listed on the NYSE or NASDAQ are included. The forecasts errors for these companies could be different compared to companies of the same size which did not grew to this size. On the other hand, firms that were that big that they were included in the S&P500 in 2005 but shrink a lot during the years or were passed in size by other companies, are not included in the sample. The omission of these companies in the analyses may also influence the results. For example forecasts for these firms might be too optimistic because analysts did not expected that these companies would shrink that much. If this is indeed the case is not investigated in this study, but it is good to think about it when reading the results.
An effect of the sample that is chosen, is that the results might not be externally valid because on average we look only at very big companies, as a consequence the results of this
11
study may not apply to smaller companies. Another disadvantage of this sample is that these results also may not hold for big companies which are not listed on the NYSE or NASDAQ, because only big companies that were included in the S&P500 are included in this
investigation. A difference in forecast accuracy may arise by the fact that these companies are subject to the strict regulations of the Security and Exchange Commission (SEC) while other big firms which are not included in this analysis may not meet these strict rules.
Finally another fact that we have to keep in mind that only firm size and the amount of debt are taken into account in this investigation. Other variables might also have an influence, for instance some part of the effect which is now assigned to one variable should actually be assigned to another variable. For example the amount of analysts that cover a company might have influence on the forecast error. Since there is a relation between the amount of analysts which cover a company and the size of the firm, the influence of size might be overstated. Therefore by drawing conclusions at the end of this study we have to take all this in mind.
3.2 Method
Following the research of Sidhu and Tan (2011) and Eddy and Seifert (1992) I use two different methods to derive accuracy of analysts. One is the absolute forecast error which is the absolute value of the difference between the average forecasted earnings per share and the actual earnings per share, divided by the actual earnings (Sidhu and Tan, 2011; Eddy and Seifert, 1992) Forecast errors measures how wrong the analysts are from actual earnings. The absolute value is taken because otherwise positive and negative errors will compensate each other. When the forecasted earnings are much larger or smaller than the actual earnings you could have an average forecast error of zero, while the analysts are far from accurate.
But to derive if the analysts are too optimistic or too pessimistic on average I will take another forecast error, namely the signed forecast error. This forecast error is the difference between the estimated earnings per share and the actual earnings per share divided by the absolute value of the actual earnings. This formula is not used to calculate the accuracy but to calculate if analysts are on average too optimistic. Positive outcomes by using this formula indicate optimistic bias and negative values indicate pessimistic bias (Sidhu and Tan, 2011).
To examine how different the expectations about company results are from each other I look at the forecast dispersion. This is the absolute value of the standard deviation of EPS forecasts divided by the mean forecast. The forecast dispersion is also named as the degree of “disagreement” among analysts (Arping and Sautner, 2013; Sidhu and Cheng Tan, 2010).
12
I took the bankruptcy of Lehman Brothers on September 15th 2008 as start of the crisis. Therefore the pre-crisis period is defined from fiscal year end 2005 till September 2008. However it is hard to state a date that the crisis ended, following a working paper of the National Bureau of Economic Research from Hatzius et al. (2010) I took a point during the course of 2009 as the end of the crisis. The exact date is March 9th because at this time the S&P500 was at his lowest point. This point is also taken as the end of the global financial crisis, for example in a research of Chang et al. of 2011. To control for the fact that it is difficult to indicate one date as the end date of the crisis I did the same analysis for another date. Based on the statement of Hazius et al. I took the last date of the year in which according to them the crisis ended as the other end date. This results in a second end date of December 31st 2009. I will look if the results of a different end date influence the outcomes of this investigation. In summary this leads for the main analysis to a pre-crisis period from fiscal year end 2005 till September 15th 2008, a crisis period from September 16th 2008 till March 9th 2009 and an after crisis period from March 10th 2009 till December 31st 2014
To answer the research questions stated above I used the Ordinary Least Squares method. For the first research question, if the forecast error will become larger during the crisis and become less after the crisis, the absolute forecast error is regressed on the period the forecasts are made. I took 𝑥𝑥1 and 𝑥𝑥2 as dummy variable for time. With 𝑥𝑥1 is one and only one for observations before the crisis and 𝑥𝑥2 is one and only one for the forecast errors after the crisis. When the forecast errors are during the crisis both 𝑥𝑥’s are zero. To investigate if there is a difference in the degree of optimism by analysts in the different time periods, the signed forecast error is regressed on the period the forecasts were made. To test if the forecast dispersion will become larger during the crisis and will decrease afterwards, I regress the forecast dispersion on the period the forecasts were made.
Besides that I expanded previous regressions with the size of the firm and the debt level to see if these firm characterises have an influence on the forecast errors. According to other investigations about forecast accuracy, the natural logarithm of the market value of equity is used as the measure of firm size (Das, Levine and Sivaramakrishnan, 1998; Gu and Wu, 2003). The debt-ratio used is the amount of long term debt plus the current liabilities divided by the total assets (Thomas, 2002; Eddy and Seifert 1992). All together this leads to the following regression.
𝐹𝐹𝐹𝐹𝑗𝑗𝑗𝑗 = 𝛽𝛽0+ 𝛽𝛽1𝑥𝑥1+ 𝛽𝛽2𝑥𝑥2+ 𝛽𝛽3𝐿𝐿𝐿𝐿𝐿𝐿 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽4𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽5𝑥𝑥1log 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑗𝑗𝑗𝑗
13
4. Results
This section analyses the empirical results from the Ordinary Least Squares regression method which takes March 9, 2009 as the end date of the crisis. In the first paragraph the summary statistics of the variables that were used are discussed. After that the regression results regarding the absolute forecast error will be discussed. The next paragraph discusses the results about the signed forecast error and the last part looks at the forecast dispersion. Finally the results for the analysis which takes December 31, 2009 as end date of the crisis are given in the appendices.
4.1 Summary statistics
This paragraph discusses the summary statistics. In Table 1 the number of observations, mean, standard deviation, minimum value, maximum value of the variables are presented. The first three variables are the different dependent variables that are used, the other variables are the independent variables. There are in total 46,314 absolute forecasts errors measured and 46310 for the 388 companies that are included in this study. The number of 388 companies is the result of all the S&P500 firms that were included at December 31, 2014 and for which monthly forecasts and actual earnings were present for all these years in the I/B/E/S database.
As you can see the absolute forecast error is three times higher than the signed forecast error, this makes sense because for the signed forecast error positive errors could be settled by negative forecast errors. A shortcoming of this analyses is that there are some large outliers which should be eliminated in a new investigation.
4.2 Absolute Forecast Error
This paragraph analyzes the results of the regression of the absolute forecast error. Table 2 presents the results of the OLS-regression for the absolute forecast error. It can be concluded that the errors are larger during the crisis than they are before September 2008. The estimated beta’s for the period before the crisis are all negative for all the three different regression equations. These results are significant at the 1% level. The fact that the forecast error is higher during the crisis than before is in line with the earlier research (Sidhu and Tan, 2011; Hope and Kang, 2005). A second result is that the forecast error is also larger during the crisis than it is after the crisis at a 1% significant level. This is consistent with the results of similar research about the Asian financial crisis (Ang and Ma,2001).
The debt ratio on his own has a negative effect on the forecast error which is opposite of previous studies (Vanstraelen et al., 2003; Eddy and Seifert, 1992; Chen et al., 2010). The
14
Table 1 Summary Statistics
This table presents the characteristics of all the variables that were used in the three different regression equations. The first three lines are the different dependent variables of these equations, the absolute forecast error, the signed forecast error and the forecast dispersion. After that the eight independent variables are presented. A dummy variable is used to indicate to what time the forecast error or dispersion can be recognized. The crisis is from September 15, 2008 to March 9, 2009. LNSIZE is the natural logarithm of the market value of equity and DEBT is the sum of the long term debt and the current liabilities divided by the total assets.
Variable Obs Mean Std. Dev. Min Max
(1) (2) (3) (4) (5)
Absolute Forecast Error 46314 0.1468 1.0035 0 51.5602
Signed Forecast Error 46314 0.0496 1.0129 -9.3198 51.5602
Forecast Dispersion 46310 0.0686 0.4890 0 45.0000
(Dummy) before crisis 46314 0.32 0.466 0 1
(Dummy) after crisis 46314 0.59 0.491 0 1
DEBT 46314 0.390 0.1999 0.0000 1.734
LNSIZE 46314 9.531 1.0852 5.5613 13.348
(Dummy) Before crisis * LNSIZE 46314 3.0112 4.4493 0.0000 13.1309
(Dummy) After crisis * LNSIZE 46314 5.7247 4.8034 0.0000 13.3480
(Dummy) Before crisis * DEBT 46314 0.1226 0.2102 0.0000 1.5423
(Dummy) After crisis* DEBT 46314 0.2326 0.2476 0.0000 1.7342
results that the debt ratio has a negative effect are significant at the 1% level with a t-value of -6.070 and -11.556. Therefore the relation of more debt leads to a greater variability of earnings which makes it harder to predict these earnings (Eddy and Seifert, 1992; Thomas, 2002; Hope, 2003) cannot be confirmed with these results. Besides that there is evidence that the amount of debt has a different influence during the crisis than before and after the crisis. The error of forecasts for firms with more debt is bigger during the crisis period than before and after the crisis. This is consistent with our expectations formulated in section two that the forecasts of more leveraged firms are more effected by the crisis.
The size of the firm does not have a significant effect at the one or five percent level. At the 10% significance level it seems that the size of the firm has a negative effect on the error. Which is consistent with the findings of for example Basi et al. (1976), Atiase (1985), Jaggi and Jain, (1998), Lim (2001), Thomas (2002). Reasons for this relation could be that there is more information available about large firms and less incentive to issue more
15
the accuracy at a 10% significance level. A reason for this could be that the sample consists of large firms, namely firms that are listed on the S&P500 at the end of our sample period so for all these firms there might be a lot of information available. To make better analyses of the influence of the size of the firm it would be better for further research to include also smaller firms. Besides that there is no interaction effect between the size of the firm and the crisis on the accuracy of analysts. This means that the forecast errors are not different for smaller firms than for larger firms before and after the crisis.
Table 2 Absolute Forecast Error
This table presents the results of three different analyses. The first analyses only looks at the influence of the crisis, the second analyses takes the size of the firm and the debt-ratio of the firm into account and the third also looks if there is an interaction effect between the crisis and the firm characteristics. The results show that the forecast error is bigger during the crisis than before and afterwards. At a 5% significance level there is no evidence for an size effect. More debt seems to be followed by a lower forecast error, but besides that there is an interaction effect between the amount of debt and the period in which the forecast is made.
𝐴𝐴𝐴𝐴𝑠𝑠𝐿𝐿𝑙𝑙𝐴𝐴𝐴𝐴𝑠𝑠 𝐹𝐹𝐿𝐿𝑙𝑙𝑠𝑠𝐹𝐹𝑙𝑙𝑠𝑠𝐴𝐴 𝐹𝐹𝑙𝑙𝑙𝑙𝐿𝐿𝑙𝑙𝑗𝑗𝑗𝑗 = 𝛽𝛽0+ 𝛽𝛽1𝑥𝑥1+ 𝛽𝛽2𝑥𝑥2+ 𝛽𝛽3𝐿𝐿𝐿𝐿𝐿𝐿 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽4𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽5𝑥𝑥1log 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑗𝑗𝑗𝑗+ 𝛽𝛽6𝑥𝑥2log 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽7𝑥𝑥1𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽8𝑥𝑥2𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝜀𝜀𝑗𝑗𝑗𝑗 (1) (2) (3) Before crisis -0.173 (-9.721)*** -0.165 (-9.278)*** -0.522 (-3.339)*** After crisis -0.117 (-6.974)*** -0.105 (-6.195)*** -0.497 (-3.345)*** LNSIZE -0.020 (-4.547)*** -0.026 (-1.823)* DEBT -0.143 (-6.070)*** -0.896 (-11.556)***
Before the crisis * LNSIZE 0.004
(0.275)
After the crisis * LNSIZE -0.007
(0.463)
Before the crisis * DEBT -0.813
(9.160)***
After the crisis * DEBT -0.837
(10.065)***
Adjusted R2 0.002 0.003 0.005
F 49.020 37.224 31.825
16
There are a few differences if we take December 31, 2009 as the end date of the crisis. The size of the firm has in this analyses a significant effect on the forecast error. The larger the firm the lower the forecast error is which is in line with the literature described in section two. Besides that there is a significant interaction effect between the size of the firm and the period in which the forecast is made. The forecasts made before and after the crisis for large firms seems to have bigger forecast errors. This is not what we expected, but again for further research the effect could better be investigated if smaller firms will be included in the analysis.
4.3 Signed Forecast Error
This paragraph analyzes the results of the regression of the signed forecast error which are presented in Table 3. As expected it can be concluded that the analysts present more
optimistic forecast during the crisis than before and afterwards. The same effect was seen in an investigation of forecast errors during negative earnings growth, which is mostly the case during a crisis (Ding et al., 2004). The crisis was not foreseen so the forecasts of analysts did not include the effect of the crisis when it broke out in September 2008. Second, a higher amount of debt leads to less optimistic forecasts which is the opposite of an investigation of Bradshaw, Richardson and Sloan (2006).
The size of the firm does not have an influence on the degree of optimism when March the 9th is taken as end date of the crisis. But if we take December 31, 2009 as end date the forecasts for large firms are less optimistic. This is consistent with the fact that for smaller firms analysts are more dependent of management information. For managers it is favorable if forecasts are high because this leads to higher compensation levels (Das et al., 1998).
4.4 Forecast Dispersion
This paragraph analyzes the results of the analyses on the forecast dispersion. The results are presented in Table 4. As expected the forecast dispersion is smaller before the crisis than during the crisis. But there is evidence that the forecast dispersion after the crisis ended.
If December 31, 2009 is taken as the end date of the crisis the forecast dispersion is also smaller in the post-crisis period. These results are consistent with previous research of Sidhu and Tan (2011).
17
Table 3 Signed Forecast Error
This table presents the results of three different analyses. The first analyses only looks at the influence of the crisis, the second analyses takes the size of the firm and the debt-ratio of the firm into account and the third also looks if there is an interaction effect between the crisis and the firm characteristics. The results show that the analysts present more optimistic forecast during the crisis than before and afterwards. Second, a higher amount of debt leads to less optimistic forecasts.
𝑆𝑆𝑠𝑠𝐿𝐿𝑆𝑆𝑠𝑠𝑆𝑆 𝐹𝐹𝐿𝐿𝑙𝑙𝑠𝑠𝐹𝐹𝑙𝑙𝑠𝑠𝐴𝐴 𝐹𝐹𝑙𝑙𝑙𝑙𝐿𝐿𝑙𝑙𝑗𝑗𝑗𝑗= 𝛽𝛽0+ 𝛽𝛽1𝑥𝑥1+ 𝛽𝛽2𝑥𝑥2+ 𝛽𝛽3𝐿𝐿𝐿𝐿𝐿𝐿 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽4𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽5𝑥𝑥1log 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑗𝑗𝑗𝑗+ 𝛽𝛽6𝑥𝑥2log 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽7𝑥𝑥1𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝛽𝛽8𝑥𝑥2𝑙𝑙𝑠𝑠𝑙𝑙𝑠𝑠𝑙𝑙𝑙𝑙𝐿𝐿𝑠𝑠𝑗𝑗𝑗𝑗+ 𝜀𝜀𝑗𝑗𝑗𝑗 (1) (2) (3) Before crisis -0.227 (-12.690)*** -0.229 (-12.756)*** -0.621 (-3.924)*** After crisis -0.197 (-11.589)*** -0.200 (-11.614)*** -0.955 (-6.375)*** LNSIZE 0.005 (1.081) -0.020 (-1.407) DEBT -0.064 (-2.696)*** -0.938 (-11.994)***
Before the crisis * LNSIZE 0.006
(0.372)
After the crisis * LNSIZE 0.040
(2.570)**
Before the crisis * DEBT 0.884
(9.875)***
After the crisis * DEBT 0.996
(11.880)***
Adjusted R2 0.004 0.004 0.007
F 82.435 43.586 40.883
N 46314 46314 46314
4.5 Remarks on the model.
It should be noted that only a little portion of the variability in the earnings forecast errors is explained by the above stated models. The percentage of variability explained by the models are somewhere between 0.01% and 0.07%. But all models used do have explanatory power according to the F statistic.
18
Table 4
Regression of the forecast dispersion
𝐹𝐹𝐿𝐿𝑙𝑙𝑠𝑠𝐹𝐹𝑙𝑙𝑠𝑠𝐴𝐴 𝐷𝐷𝑠𝑠𝑠𝑠𝐷𝐷𝑠𝑠𝑙𝑙𝑠𝑠𝑠𝑠𝐿𝐿𝑆𝑆𝑗𝑗𝑗𝑗= 𝛽𝛽0+ 𝛽𝛽1𝑥𝑥1+ 𝛽𝛽2𝑥𝑥2+ 𝛽𝛽3𝐿𝐿𝐿𝐿𝐿𝐿 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑗𝑗𝑗𝑗+ 𝜀𝜀𝑗𝑗𝑗𝑗 Before crisis -0.030 (-3.442)*** After crisis 0.004 (0.544) Adjusted R2 0.001 F 24.024 N 46310
19
5. Conclusions and Remarks
As expected the accuracy of earnings forecast by analysts decreased during the crisis
compared to the pre-crisis and post-crisis period. This result is consistent with earlier research of Sidhu and Tan (2011), Hope and Kang (2005) and Ang an Ma (2001). But the debt seems to have a negative effect on the forecast error which is opposite of previous studies
(Vanstraelen et al., 2003; Eddy and Seifert, 1992; Chen et al., 2010). But as expected the error of forecasts for firms with more debt is bigger during the crisis period than before and after the crisis. The size of the firm has no large significant effect on the forecast error. A reason for this could be that the sample consists of large firms.
As expected it can be concluded that the analysts present more optimistic forecast during the crisis than before and afterwards, which is consistent with the study of Ding et al.(2004). But a higher amount of debt leads to less optimistic forecasts which is the opposite of an investigation of Bradshaw, Richardson and Sloan (2006). Finally the forecast dispersion is smaller before the crisis than during the crisis. But there is no evidence that the forecast dispersion is higher during the crisis compared to the period after the crisis.
There are also some limitations to this study. First I could not find an objective date for the end of the crisis. To control for this fact I took a second date, fortunately there are no big differences between this analyses but for further research this could be an area for improvement. A second limitation is the size of the firms included in this analyses. Since we consider that the size of the firm has an effect on the accuracy this is an important point. Of course there is a difference between the size of the firms, but it is not as big if you take a company which is not on the S&P500. Therefore we cannot conclude that the results stated above are applicable to all firms in the U.S..
Another disadvantage of the sample that is chosen is that the results may not hold for big companies which are not listed on the NYSE or NASDAQ, because only big companies that were included in the S&P500 are included in this investigation. A difference in forecast accuracy may arise by the fact that these companies are subject to the strict regulations of the Security and Exchange Commission (SEC) while other big firms which are not included in this analysis may not meet these strict rules.
Besides that it seems obvious that there are way more firms characteristics and other conditions which have influence on the forecast accuracy. It is concluded that the amount of analysts covering firms is increasing in size of the firm. It therefore might be that firm size does not have an influence but that the amount of estimates by different analysts has an influence.
20
Reference List
Ang J, Ma Y. (2001) “The behavior of financial analysts during the Asian financial crisis in Indonesia, Korea, Malaysia, and Thailand”, Pacific-Basin Finance Journal 9, pp. 233– 263.
Basi, B., K. Carey and R. Twark, (1976), “A comparison of the accuracy of corporate and security analysts’ forecasts of earnings”, The Accounting Review, April, 244-254.
Bradshaw, M., Richardson, S., Sloan, R., (2006), “The relation between corporate financing activities, analysts’ forecasts and stock returns”, Journal of Accounting and Economics
42, 53–85.
Brown, L., R. Hagerman, P. Griffin, and M. Zmijewski, (1987a), "Security Analyst Superiority Relative to Univariate Time-Series Models in Forecasting Quarterly Earnings," Journal of Accounting and Economics, no. 1, pp. 159-193.
Brown, L., G. Richardson and S. Schwager, (1987b), “An information interpretation of financial analyst superiority in forecasting earnings”, Journal of Accounting Research
25, 49-67.
Chen, C. C. P., Ding, Y., & Kim, C. F. (2010), “High-level politically connected firms, corruption, and analyst forecast accuracy around the world” Journal of International
Business Studies, 41(9), pp. 1505–1524.
Chang et al. (2011) “Risk Management of Risk under the Basel Accord: Forecasting Value-at-Risk of VIX Futures”
Ciccone, S.J. (2002), “Improvement in the Forecasting Ability of Analysts”
Coën, A., & Desfleurs, A. (2004). “The evolution of financial analysts’ forecasts on Asian emerging markets”, Journal of Multinational Financial Management, 14, 335–352
21
Das, S., C.B. Levine and K. Sivaramakrishnan (1998), “Earnings Predictability and Bias in Analysts’ Earnings Forecasts”, The Accounting Review, Vol. 73, pp. 277–94.
DeBondt, Werner F. M., and William P. Forbes (1999), “Herding in analyst forecasts:
Evidence from the United Kingdom”, European Financial Management 5, pp. 143-163.
DeBondt, Werner F.M., and Thaler, Richard H. (1990), “Do security analaysts overreact?” ,
American economic review, pp.793-805
Ding, D. K., Charoenwong, C., & Seetoh, R. (2004). “Prospect theory, analyst forecasts, and stock returns”, Journal of Multinational Financial Management, 14, pp. 425–442.
Eddy, A. and B. Seifert, (1992) “An Examination of Hypotheses Concerning Earnings Forecast Errors”, Quarterly Journal of Business and Economics 31, pp. 22–37
Elton, E. J., M. J. Gruber, and M. N. Gultekin (1984), “Professional expectations: Accuracy and diagnosis of errors”, Journal of Financial and Quantitative Analysis 19, 351-363.
Gu, Z. and S. Wu, J. S., (2003), “Earnings skewness and analyst forecast bias”, Journal of
Accounting and Economics, 35, 5-29.
Hatzius, J., Hooper, P., Mishkin, F., Schoenholtz, K. and Watson, M., (2010), “Financial Conditions Indexes: A Fresh Look after the Financial Crisis”, NBER Working Papers,
No. 16150.
Hope, 0. (2003), “Accounting policy disclosures and analysts' forecasts”, Contemporary
Accounting Research 20: pp. 295-322.
Hope, O. K., & Kang, T. (2005). “The association between macroeconomic uncertainty and analysts’ forecast accuracy”, Journal of International Accounting Research, 4(1), pp. 23–38.
22
Hussain, S. (2000), “Simultaneous determination of UK analyst following and institutional ownership”, Accounting and Business Research, 30, 111–124.
Jaggi, B. and R. Jain (1998), “An Evaluation of Financial Analysts’ Earnings Forecasts for Hong Kong Firms”, Journal of International Financial Management and Accounting, 9 (3): pp. 177–200.
Lim, T. (2001), “Rationality and Analysts’ Forecast Bias”, Journal of Finance, Vol. 56, pp. 369–86.
Loh and Mian. “The Quality of Analysts’ Earnings Forecasts During the Asian Crisis: Evidence from Singapore” 2003
O’Brien, Patrica (1988) “Analysts’ forecasts as earnings expectations.”, Journal of
Accounting and Economics, 10, pp. 53–83.
Ramnath, S., Rock, S. and Shane, P. (2008) “Financial Analyst Forecasting Literature: A Taxonomy with Trends and Suggestions for Further Research”, International Journal of
Forecasting, 24: pp. 34–75.
Richardson, S.A., Teoh, S.H. and Wysocki, P.D. (2004) “The Walkdown to Beatable Analsyt Forecasts: The Role of Equity Issuance and Insider Trading Incentives”, Contemporary
Accounting Research, 1985, 21, 4: 885–924.
Sidhu, B. and Tan, H. (2011) , “The Performance of Equity Analysts During the Global Financial Crisis”, Australian Accounting Review, 21, 1: 32–43.
Thomas, S., (2002), “Firm diversification and asymmetric information: Evidence from analysts’ forecasts and earnings announcements”, Journal of Financial Economics 64, pp. 373–396.
Vanstraelen, A., Zarzeski, M. T., & Robb, S. W. G. (2003). “Corporate nonfinancial disclosure practices and financial analyst forecast ability across three European countries”, Journal of International Financial Management and Accounting, 14(3), 249–278.
23
Weinstein, Neil (1980), "Unrealistic Optimism about Future Life Events," Journal of
Personality and Social Psychology, pp. 806-20.
Zhang, F., (2006),“Information uncertainty and analyst forecast behavior”, Contemporary
24
Appendix
The following tables are the same as the tables presented in the paper except one thing. The Crisis in the paper ended on March 9, 2009, in the following tables the end date of the crisis is taken as December 31, 2009 to compare if the results are very different dependent on the end date of the crisis
Table 5 Absolute Forecast Error Crisis from sept 2008 till 31 December 2009
(1) (2) (3)
Before the crisis -0.132
(-9.611)***
-0.127
(-9.221)***
-0.481
(-3.852)***
After the crisis -0.084
(-6.555)*** -0.072 (-5.596)*** -0.549 (-4.629)*** LNSIZE -0.020 (-4.509)*** -0.043 (-4.317)*** DEBT -0.141 (-5.981)*** -0.476 (-8.692)***
Before the crisis * LNSIZE 0.022
(1.761)*
After the crisis * LNSIZE 0.034
(2.811)***
Before the crisis * DEBT 0.393
(5.637)***
After the crisis * DEBT 0.419
(6.575)***
Adjusted R2 0.002 0.003 0.004
F 46.182 35.453 23.954
25
Table 6 Signed Forecast Error Crisis from sept 2008 till 31 December 2009
(1) (2) (3)
Before the crisis -0.105
(-7.574)***
-0.105
(-7.566)***
-0.392
(-3.112)***
After the crisis -0.067
(-5.184)*** -0.067 (-5.094)*** -0.754 (-6.295)*** LNSIZE 0.001 (0.195) -0.030 (-2.915)*** DEBT -0.065 (-2.752)*** -0.438 (-7.910)***
Before crisis * LNSIZE 0.016
(1.230)
After crisis * LNSIZE 0.053
(4.370)***
Before crisis * DEBT 0.384
(5.447)***
After crisis * DEBT 0.492
(7.640)***
Adjusted R2 0.001 0.001 0.003
F 28.684 16.327 17.655
26
Table 7 Forecast Dispersion
Crisis from sept 2008 till 31 December 2009
Before the crisis -0.066
(-9.843)***
After the crisis -0.044
(-7.018)***
Adjusted R2 0.002
F 48.524
