Page 1 Case Western Reserve University Weatherhead School of
Management FNCE 435 – Empirical Finance Fall 2021 Assignment
8 ________________________________________________________________________ Part
I – Concepts You are examining the relationship between bonus paid to
the CEO and its firm’s future performance. Your alternative hypothesis
is that the higher the bonus paid, the higher the performance of the
firm. For a bonus paid in year t, future performance (PERF) is measured
as the firm’s average monthly return in the 3 years following t (that
is, t+1 through t+3). To measure bonus, the study defines LBONUS as the
natural logarithm of the bonus paid (US$ thousands) to the CEO in year
t. The other variables of the study are: LSIZE: the logarithm of
market value of equity (US$ millions) for the firm measured at year t.
BEME: the ratio of book value of equity to market value of equity, both
measured at the end of year t CHAIR: dummy equal to 1 if the CEO is
also the Chairman of the Board in year t TENURE: the number of years
since the person started in the CEO position Summary statistics for the
data as of t=2006 is shown in Figure 1. For our sample, the 3- year
average performance of firms is 2.63%, firm’s average size is $8.4
billion, average book-to-market is 0.52, and average CEO bonus is
$738,000. Finally, 18% of the firms in our sample have CEO cumulating
the role of Chairman of the Board, and the average CEO tenure is 3.43
years. Figure 1: Summary statistics The results of a SAS execution of
the regression explaining future performance, PERFi=β0+ β1*LBONUSi +
β2*LSIZEi + β3*BEMEi + β4*CHAIR+ β5*TENURE + εi appear in Figure 2. FNCE
435 Fall 2021 Assignment 8 Page 2 Figure 2: Regression results a)
Examine the effect of bonus payments on company’s future performance.
Please formulate your hypothesis clearly and how it is being tested in
the model above. Do bonus payments matter How so b) Define a 95%
two-tailed confidence interval for the coefficient on tenure
(TENURE). Then conclude whether and how tenure affects the future
performance of the firms. c) Interpret the coefficient β0 in the model
above. Is it meaningful in this regression d) Your colleague argues
that, since you are interested in the relationship between BONUS and
future performance, you should be running a simple regression model,
as in PERFi=β0+ β1*LBONUSi + εi Your colleague believes that this would
give a cleaner measure of association between these measures, and that
the introduction of other right-hand side variable can confound the true
association between bonus and performance. Please argue which
model—your colleague’s model or the extended version used to produce the
results in Figure C.2— is more suited to the analysis of the
relationship between bonus and performance, and why. e) After running
the regression model whose result appears in Figure 2, you were
shown the following pattern for the errors in the regression. FNCE 435
Fall 2021 Assignment 8 Page 3 Figure 3: Errors from regression model
explaining firm performance What does that illustrate as a problem with
respect to the assumptions required to run a regression model How can
you confirm this as a problem, and what measure can you use to remedy
the problem Part II – Empirical Examination of Reactions to Earnings
Announcements We will examine reactions to earnings announcements.
Companies release earnings numbers every quarter. Depending on the
expectations that investors have regarding the soon-to-be-released
numbers, these earnings announcements might convey good, bad, or no news
to investors. We can then look at market reactions to earnings news to
verify the financial implications of earnings numbers. Why this study
matters Not only for academic purposes. There is a huge industry out
there that tries to predict where earnings will be. If market do react
to earnings news (our hypothesis) and you know how to predict the news,
then you have a money machine! See in Figure 5 an article from The New
York Times about this. Our sample involves all quarterly earnings
announcements for U.S. public firms trading at NYSE or NASDAQ for the
first quarter of 2013. The sample is in the
dataset “a8_earnings.sas7bdat” file, available on Canvas. Each data
point involves an earnings announcement. An example of such data point
is show in Figure 4. Figure 4. One data point of the sample of earnings
announcements It shows the earnings announcement for the company with
PERMNO=10225 (named “BEAM INC”) and for the quarter ending in
PENDS=03/31/2013. Earnings per share for that quarter were released in
05/02/2013 (variable ANNDATS). The earnings per share released is stored
in the variable VALUE, while the market expectation is stored in
the variable MEANEST. So, for the first quarter of 2013, J & J’s
earnings number was $0.64, a bit more than the market expectation of
$0.54. -15 -10 -5 0 5 10 15 20 25 30 -6 -4 -2 0 2 4
6 Residual Predicted AR permno gvkey comnam pends anndats value meanest
car_ea n_buys n_holds n_sells 10225 1408 BEAM INC 31-Mar-13 2-May-13
0.64 0.54 0.0194 1 0 0 FNCE 435 Fall 2021 Assignment 8 Page 4 As for
market reaction, we define a variable CAR_EA that measures the
cumulative abnormal return for the window [0,1] around the earnings—that
is, it adds the abnormal return in the day after the announcement day
and the abnormal return at the announcement day. This makes sense since
some earnings announcements happen after the close of the market, and so
their effects materialize only in the next trading day. For the first
quarter of 2013, the abnormal reaction to BEAM INC’s earnings
announcement was 1.94% (notice that returns here are stored in
decimals). The idea is to examine and quantify the relationship between
the market reaction and the surprise in the earnings announcement. We
define SURPRISE as the gap between the actual earnings and the analyst’s
expectation about the earnings right before the announcement date. In
our example above, the surprise was VALUE – MEANEST=0.64– 0.54=$0.10;
or, 10 cents per share. Given that earnings were above the market
expectations, the announcement amounts to good news! We are interested
in a linear relationship between CAR_EA and surprise, as
follows: CAR_EA= β0+β1*SURPRISE+ε Earnings surprise might matter for
market reactions, but there are other explanations for the reactions to
earnings announcements. The same Wall Street analysts that
produce earnings forecasts also produce recommendations on the stocks of
the firms they follow. (We’ve seen them in assignment 6 already.)
Recommendations come with extensive research reports, but in the end a
recommendation amounts to a statement about whether someone should buy,
hold, or sell the fim’s stock. If recommendations do affect the price of
each share, and recommendations are issued around earnings
announcements, we may see market reactions to the earnings announcement
due to recommendations, rather than (or on top of) the surprise in the
announcement. In order to examine this possibility, we also collect, for
each earnings announcement, the recommendations issued for the firm in
the day right before the earnings announcement. The new variables are:
N_BUYS (the number of buy recommendations issued for the firm in the day
preceding the earnings announcement for that quarter); (3) N_HOLDS
(the number of hold recommendations); and (5) N_SELLS (the number of
sell recommendations). Our example in Figure 4 shows that there was one
buy recommendation (but no hold, nor sell recommendation) issued for
PERMNO=10225 one day before the earnings for the first quarter of 2013
was released. We now have to think about which control variables to
adopt to our examination. First, we will control the issuance of
recommendations, and thus use the variables N_BUYS, N_HOLDS, and
N_SELLS. Our question thus becomes: after controlling for the
issuance of recommendations, does surprise in earnings numbers have any
say on reactions to earnings Page 5 Figure 5. Earnings announcements
and trading strategies (NYT, 01/27, 2008, Sunday Business Section, Page
6) The second control variable is firm size. We know that size is
related to returns. Plus we want to examine that relationship based on
rates of change in firm size. We will use two proxies for firm size:
The variable LMVE is defined as the natural logarithm of the firm’s
market value of equity in the year before the earnings. Since our data
refer to 2013 earnings, you FNCE 435 Fall 2021 Assignment 8 Page 6 will
collect market value of equity as of December 2012. Market value of
equity is abs(PRC)*SHROUT, where both PRC (stock price) and SHROUT
(shares outstanding) are variables available in the CRSP dataset
“msf.sas7bdat” (located in “/wrds/crsp/sasdata/a_stock”); The variable
LTA is defined as the natural logarithm of firm’s total
assets, measured in 2012. Total assets is the variable AT in Compustat
dataset FUNDA, located at “/wrds/comp/sasdata/nam”. The baseline model,
using LVME as our proxy for firm size, thus becomes: CAR_EA=
β0+β1*SURPRISE+β2*LMVE+β3*N_BUYS+β4*N_HOLDS +β5*N_SELLS+ε But, before we
run the regression model, generate and show some summary statistics
for the basic variables of your model. The summary statistics you should
have for each variable are: the number of observations, the average,
the standard deviation, the minimum and maximum value. A first look at
reactions to earnings announcements: Let’s first test reactions to
earnings announcements, conditioned on the type of surprise, in some
event studies. You will run 4 event studies. First, test whether
reactions to earnings with positive surprises (that is, the sample
of earnings with SURPRISE>0) are significantly positive. That is,
formally test: Ho: E(CAR_EASurprise>0)=0 Ha:
E(CAR_EASurprise>0)>0 Second, you test whether reactions to
earnings with negative surprises (that is, the sample of earnings with
SURPRISE<0) are significantly negative. Third, you test
whether reactions to earnings with no surprises (that is, the sample of
earnings with SURPRISE=0) are significantly different from
zero. Finally, let’s look at reactions to very small surprises—the ones
with positive but up to 1 cent surprise. You test whether reactions to
earnings with small positive surprises (that is, the sample of earnings
with 0Ho: E(CAR_EA0Ha: E(CAR_EA00 Here is the output you should produce and show: Figure 6. Event study results What can you learn from these event studies Sample # obs Average CAR_EA t Earnings with negative surprise 0,000 0.0000 0.00 Earnings with no surprise 0,000 0.0000 0.00 Earnings with positive surprise 0,000 0.0000 0.00 Earnings with small (1 cent) surprise 0,000 0.0000 0.00 FNCE 435 Fall 2021 Assignment 8 Page 7 A first look at the regression results: Recall that our basic regression model is CAR_EA= β0+β1*SURPRISE+β2*LMVE+β3*N_BUYS+β4*N_HOLDS +β5*N_SELLS+ε but there is some kind of misspecification in this model. We know that returns and firm size are negatively related—that smaller firms tend to have higher returns. You (probably!) learned from the event studies that reactions to earnings announcements are asymmetrical and dependent on the surprise: abnormal returns are positive following good news (positive surprises) and negative following bad news (negative surprises). We may thus see that reactions to positive surprises are more positive the smaller the firm size (a negative correlation between CAR_EA and firm size for the sample of positive surprises), but reactions to negative surprises are more negative the smaller the firm size (a positive correlation between CAR_EA and firm size for the sample of negative surprises). In sum, the coefficient on LMVE may vary dependent on the subsample being analyzed. Let’s thus run the regression model above for three subsamples as described in Figure 7. Figure 7. Regression results, baseline model Interpret the effect on CAR_EA, if any, of each of the right-hand side variables. Describe whether the effect is dependent on which subsample you are employing. Let’s now examine the robustness of our results to the concerns regarding heteroscedasticity: From this point onwards, let us focus on the subsample of earnings with positive surprises—that is, with SURPRISE>0. Let’s now examine the robustness of our results to the concerns regarding heteroscedasticity. Thus, for the model II in Figure 7, generate a plot of residuals vs. predicted CAR_EA. Does heteroscedasticity stand out visually from the graph Then, run the regression with the White test for heteroscedasticity. Does the test reject the null of homoscedasticity If so, rerun the regression asking for t-statistics (and standard errors) that are robust to heteroscedasticity. Create a new output (as the output from model II in Figure 7) for the regression results, and comment on whether the inferences are robust to this change. Intercept Surprise Lmve N_buys N_holds N_Sells # obs Adj-R 2 I All 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] II Surprise>0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] III Surprise≤0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] Model FNCE 435 Fall 2021 Assignment 8 Page 8 We then examine the robustness of our results to the concerns regarding correlation of residuals: (Still, use for this analysis only the subsample of earnings with positive surprises—that is, with SURPRISE>0.) First, make sure the data is sorted by the announcement date, that is, first run the code proc sort data=d; by anndats; run; then run the regression asking for the Durbin-Watson test for serial correlation and comment whether serial correlation seems to be a problem for this regression specification. Now test for the Durbin-Watson again, but, before running the regression, sort the data by the firm identifier, that is, run the statement: proc sort data=d; by permno; run; Comment on the results you get for the Durbin-Watson test. (Hint: this is a test about correlation of consecutive residuals, so it depends on sorting the data…) Now exploring some other potential explanatory variables: (Still, use for this analysis only the subsample of earnings with positive surprises—that is, with SURPRISE>0.) Recall you also collected a different proxy for firm size, LTA. Let’s change the model a bit. Instead of using LMVE as a control variable, use instead LTA, CAR_EA= β0+β1*SURPRISE+β2*LTA+β3*N_BUYS+β4*N_HOLDS +β5*N_SELLS+ε Now your research team argues that you should use both LMVE and LTA in the same model. In sum, you will the three regressions shown in Figure 8 below. (The first model in Figure 8 you already ran.)That is, run regressions separately with LMVE and LTA, then add them all in the same specification. Make sure that for all regressions here you use the standard errors adjusted for heteroscedasticy. Figure 8. Regression results, using different proxies for size Intercept Surprise Lmve Lta N_buys N_buys N_sells # obs Adj-R 2 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0,000 0.0% [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] [0.00] FNCE 435 Fall 2021 Assignment 8 Page 9 Do you see any issues with the inferences, regarding the proxies for firm size (LMVE and LTA) Can you get a hint of what is going on You can use a PROC CORR to help you in your analysis.
