Microsoft Word – FNCE 435 Fall 2021 Assignment 6
Page 1
Case Western Reserve University
Weatherhead School of Management
FNCE 435 – Empirical Finance
Fall 2021
Assignment 6
_______________________________________________________________________
Part I – Concepts
1. Suppose you are examining the relationship between bonus paid to the CEO and its
firm’s future performance. More specifically, you want to test the hypothesis that paying
more to the CEO of a firm in a year leads to superior stock market performance for that
firm in the near future (say, over the next 3 years).
Explain how you can test the hypothesis. Assume you have yearly data on salary paid to
the CEOs of the Fortune 500 companies in US from 1989 to 1995. Be clear about what
data you would need and what statistical test you would perform. Finally, recall that we
already discussed the determinants of firm returns—that is, perhaps it is not only salary
paid to CEO that determines the returns of a firm’s stock. In particular, you know that a
firm’s beta is a determinant of its stock market performance.
(You only need to explain which data would you need and which test would you perform.
No need to write an algorithm or a code. In particular, you do not need to worry about
how to collect data; just assume any data you need will be collected with no effort!)
2. A real estate company developed a linear regression model to explain sales price (SALE,
in $) in its market. The model uses three independent variables: X1=appraised land value
of the property (in $); X2=appraised value of the improvements (home value) (in $); and
X3=area of living space on the property (in square feet). A regression model
SALE = β0+β1*X1+ β2*X2+ β3*X3+ε
was run for the recent 1,187 real estate transactions, and the result appears below. For each
coefficient, the table shows the coefficient’s estimate, its t-stat (between brackets) and its
standard error (between parentheses).
a) Can you give an interpretation for the coefficient β0? Does it make sense in this
setup?
b) Compute the predicted sales price for a house that has its land appraised at $39,000,
its construction (home value) appraised at $149,000, and with a living area of 1,950
square feet.
FNCE 435 Fall 2021 Assignment 6
Page 2
c) Your friend has a construction company and she usually argues with her clients that
home improvements pay off, in the sense that each dollar invested in the home
value (what appears as value of improvements) translates into a dollar more in sale
price. Is her claim true given the model’s results?
d) If you are thinking of expanding your property with an additional unit that would
increase the living area by 100 square feet, what is the predicted increase in sale
price of your house? Assume no other characteristic of the house will change with
the expansion project.
e) Discuss the explanatory power of the model.
3) You want to study the effects of disposable income, anti-smoking advertising and
cigarette prices on the per capita consumption of cigarettes in US. You collect data on these
variables across the 51 states over 20 years. You thus have 51×20=1,020 data points. For
each state and year, you measure: Y=per capita consumption of cigarette (the fraction of
the state population that smoked in that year); X1=per capital income in the state (in US$);
X2=total expenditure in anti-smoking advertising in that state (in US$); and X3=average
cigarette price in the state (in US$). You can then run a model:
Y = β0+β1*X1+ β2*X2+ β3*X3+ε
Explain how you would test that the model is useful in predicting Y: How you would test
the alternative hypothesis that you have at least one of the coefficients β1, β2 or β3 is
different from zero. (That is, the null is that all three coefficients are zero.) Use a
significance level α=0.05.
(Again, there is no need to “run” the data. Just explain what you would do. Assume you
have all data ready to go, and you have access to any statistical package.)
Part II – Empirical Examination
This part will empirically explore the event of expiration of “quiet periods” of Initial Public
Offerings (IPO’s). Until 2001, quiet period referred to the 25 days after the IPO’s offering
date, a period in which sell-side analysts could not publish information about the new firm.
This examination is based on the article “The Quiet Period Goes out with a Bang”, by
Bradley, Jordan and Ritter (2003). A copy of this paper can be found on Canvas (under the
“Modules”/”Supplementary Materials”/“Papers and Articles” folder).
An IPO refers to the moment a company goes public. This moment is marked by issuance
of public equity, so the IPO is also the first moment the company raises equity in a stock
market. To avoid undue trading volatility in the first few days the stock becomes public,
SEC prohibits sell-side analysts from writing reports or issuing forecasts for the stock (such
as the recommendations we examined in assignment 4). This is known as the quiet period,
and (until 2001) it lasted for the first 25 days after going public.
As the title of Bradley, Jordan and Ritter’s (2003) paper suggests, the expiration of quiet
period usually comes with very high abnormal returns. This might be surprising at first,
given that such dates are completely predictable. In a sense, if such abnormal returns are
significantly positive, it is a rejection of market efficiency, since it is public information
when the quiet period ends. Plus, one could potentially profit from such pattern—by buying
FNCE 435 Fall 2021 Assignment 6
Page 3
the stock right before the end of the quiet period. The goal of this assignment is to explain
why such pattern exists and that in fact it is not that easy to make money out of it.
We have a sample of 902 IPOs between 1998 and 1999. The sample appears in the dataset
“a6_ipo.sas7bdat”, available on Canvas. Each observation contains the firm’s PERMNO
and the date of the end of the quiet period (QUIET_PERIOD_END).1
Your first task is to run an event study of abnormal returns around the expiration of quiet
periods. Use an event window of between –5 and +5 relative days around the end of the
quiet period. For abnormal returns, use the market adjusted returns method, that is,
abnormal return is the IPO return (variable RET in the DSF dataset) minus market return
(variable VWRETD in the DSIX dataset). The methodology will closely mimic the analysis
in assignment 4. In fact, create and show in your write-up two outputs for your event study:
a summary statistic such as in Figure 2 of assignment 4 (one of the 5 panels appearing in
that figure), and the pattern of average cumulative abnormal returns such as in Figure 3 of
assignment 4. Briefly analyze the patterns shown in these two outputs.
Now that (hopefully!) your analysis has shown a pattern of increasing cumulative abnormal
returns around the expiration of quiet periods, let’s go to the second task: that of analyzing
one potential cause for the pattern. It happens that when the quiet period ends, lots of
analysts, which before that could not issue any report on the newly public form, start
issuing recommendations. I collect data on those recommendations. The dataset
“a6_quiet_ext.sas7bat” contains the data: for each IPO (defined by PERMNO), the dataset
has the following variables:
N_SB contains the number of strong buy recommendations received by the IPO at
the end of the quiet period;
N_B contains the number of buy recommendations received by the IPO at the end
of the quiet period;
N_H contains the number of hold recommendations received by the IPO at the end
of the quiet period.
(Note: there was not a single sell recommendation issued for these IPOs.)
The dataset has one more variable, SIZE, containing a proxy for the IPO’s size. It is a
number between 0 and 1, with 0 being the smallest and 1 being the largest (by market
value) IPO in the sample.
Besides these measures, you should compute one more: the variable N_REC will be simply
the sum of all the recommendation issued to an IPO, or
N_REC=N_SB+N_B+N_H
Your task will be to relate in a regression framework the market reaction during the end of
the quiet period and the variables in this new dataset. For market reaction, we follow the
directions in Bradley, Jordan and Ritter (2003) and define a variable CAR for each IPO as
the summation of abnormal returns for that IPO in the window [–2,+2], or
1 Our sample is different from the one used in Bradley, Jordan and Ritter (2003). While our analysis may
yield results that are qualitatively similar to the ones reported in the paper, the results of this assignment
should not be directly compared to those of the paper.
FNCE 435 Fall 2021 Assignment 6
Page 4
CARi=ARi,t=-2+ ARi,t=-1+ ARi,t=0+ ARi,t=+1+ ARi,t=+2
You will need to compute that variable as a final step in your event study.
(You can work on the regression analyses even without concluding the event study. The
dataset “a6_car.sas7bdat”, available on Canvas, contains a measure of the CAR variable
generated by the instructor. Each row of this dataset has the variables PERMNO and CAR,
so that you can use this dataset to get your CAR.
The CAR measure in “a6_car.sas7bdat” is not exactly the measure you may obtain from
your event study, and thus should not be used to evaluate whether you event study is
correct. However, the CAR measure in this dataset is close enough to the true measure and
can be employed in the regression analyses.)
Generate univariate statistics (number of observations, mean value, standard deviation, min
and max) on the variables used in the study. You should generate and show the following
output:
Figure 1. Summary statistics
Use the data from Figure 1 to analyze whether average reactions at the end of quiet periods
is different from zero. That is, formally test whether the CAR measure is significantly
different from zero. (No need of a regression here.)
Now we finally run our regression analyses . You should run different regressions models
to fill in the table in Figure 2. For each coefficient in a model, report the estimated value
and its t-stat (between brackets).
Figure 2. Regressions results
Variable N Mean Std Dev Min Max
Car 0 0.0000 0.0000 0.0000 0.0000
N_rec 0 0.0000 0.0000 0 0
N_sb 0 0.0000 0.0000 0 0
N_b 0 0.0000 0.0000 0 0
N_h 0 0.0000 0.0000 0 0
Size 0 0.0000 0.0000 0 0
Model Int. N_rec N_sb N_b N_h Size # obs Adj-R
2
I 0.0000 000 0.0%
[0.00]
II 0.0000 0.0000 000 0.0%
[0.00] [0.00]
III 0.0000 0.0000 0.0000 0.0000 000 0.0%
[0.00] [0.00] [0.00] [0.00]
IV 0.0000 0.0000 0.0000 0.0000 0.0000 000 0.0%
[0.00] [0.00] [0.00] [0.00] [0.00]
FNCE 435 Fall 2021 Assignment 6
Page 5
The first row of the table (Model I) runs a regression with only an intercept.2 Then for
Model II shows the results of the following regression model
CAR = β0+β1* N_REC+ε
and so on.
You have a bunch of regressions results; now it is time to interpret your findings. Formally
write down each hypothesis being examined and discuss the inferences based on the results
in Figure 2. These are some questions that you should explore:
What is the relationship between having recommendations and the cumulative
abnormal return during the end of the quiet period? Take as the null that no such
relationship exists, and the alternative hypothesis that there is some relationship.
Then explore whether it is significant, and, if so, what is the magnitude of the effect
of having one more recommendation.
Using the same framework, now narrow your analysis to each type of
recommendation. What is the effect, if any, of having an extra strong buy? An extra
buy? An extra hold?
Do small firms have more pronounced market reactions during the end of the quiet
period?
Do you see improvements in the explanatory power of the model as you add more
explanatory variables?
Finally, interpret the intercepts in model II, III and IV. What does it say with respect
to the event study’s results that “end of quiet period goes out with a bang”? Then
interpret whether is it easy to make money out of the pattern of significantly
positive abnormal reactions around the expiration of quiet periods.
2 To run a regression of Y with only the intercept in SAS, use the syntax
MODEL Y=;
in the PROC REG statement.