HEC Paris

Professor Ioanid Rosu MIF Securities Markets

Fall 2023

Solutions to Assignment #2
CAPM/APT; Market Efficiency

Problem 1. You are asked to calculate the cost of capital for Apple (ticker: AAPL) and K-
Mart (ticker: KMRT) for the period 2007-2008. Use the Excel spreadsheet “ps2_prob1.xls”
to compare the APT- and CAPM-predicted risk premia on Apple and K-Mart. It contains
monthly returns data for the two stocks (and many others), as well as for the three Fama–
French portfolios (MKT, SMB, and HML). To make it easier for you, I already showed in the
spreadsheet how to perform the same calculations for the period 2004-2008.

Note: By convention, MKT is the risk factor that represents the market. Therefore, the
value of MKT in month t is the return on the market at t minust the risk-free rate at t:

MKT t = rM,t − rf,t .

(a) Find the α, the β, and the R2 of the monthly CAPM regression for the two stocks for
the period 2007-2008. Write down also the standard errors and the t-statistics for α
and β.
Use the Excel command LINEST. When using the LINEST formula, select with the
mouse a 3 × 2 rectangle (3 rows and 2 columns), and type the appropriate formula. For
example, to perform the CAPM regression for Apple for 2007-2008 you have to type
=LINEST(H$5:H$28,$B$5:$B$28,TRUE,TRUE) . Keep in mind that the result is an array
(a 3 × 2 matrix), so in order to get all the coefficients you have to press

Ctrl-Shift-Enter

Which estimates are statistically significant? Would a statistically significant value of

α constradict CAPM?
(Hint: You can simplify the process by copying the relevant blocks for 2004-2008, and
changing it so that the data is taken for the 2007-2008 period.)

Solution: For AAPL the CAPM regression produces:

AAPL estimate standard error t-statistic

α 0.043 0.024 1.81
β 1.768 0.423 4.18
R2 44.30%

Since we use statistical significance with 95% confidence, we want the t-statistic to be
more than 1.96 in absolute value (that is the t-statistic should be either > 1.96 or
< −1.96). So α is not significant, while β is significant.
For KMRT we get:

1
If the CAPM holds perfectly, alpha should be equal to zero. This is because alpha represents the residual return or "excess return" above what
can be explained by beta. In other words, alpha measures the portion of an investment's return that cannot be attributed to market risk but may
be due to the skill of the fund manager or other factors. In an efficient market where all assets are fairly priced based on their risk, there should
be no excess return (alpha) left to capture.

KMRT estimate standard error t-statistic

α -0.021 1.493 -0.84
β 1.493 0.444 3.36
R2 33.91%

Again, α is not significant, while β is significant.

According to CAPM, the true alpha of any stock should be zero. In the case of AAPL,
we see that the estimate of α is statistically significant at a 95% confidence level. That
means that even if the true α of AAPL is zero, it might still happen (with 5% probability,
to be precise) that the estimate of α is statistically significant. This is especially if we
choose (after the fact) a successful stock like AAPL. Therefore, one significant α is not
by itself strong evidence against CAPM.
(b) Find the α, the β, and the R2 of the monthly APT Fama-French 3-factor regression for
the two stocks for the period 2007-2008. Write down also the standard errors and
the t-statistics for α and β.
Use again the Ctrl-Shift-Enter as in (a). Also, make sure you pay attention to the
order of the coefficients, which LINEST for some obscure reason places in reverse
order! )

Solution: or AAPL the FF3 regression produces:

AAPL estimate standard error t-statistic

α 0.034 0.022 1.53
βMKT 1.837 0.395 4.65
βSMB 0.075 1.179 0.06
βHML -2.687 1.115 -2.41
R2 59.58%

For KMRT we get:

KMRT estimate standard error t-statistic

α -0.012 0.017 -0.71
βMKT 1.239 0.017 4.02
βSMB 1.479 0.308 1.61
βHML 3.242 0.869 3.73
R2 73.61%

(c) Find the cost of capital estimates for Apple and K-Mart for the period 2007-2008
using both CAPM and APT (the Fama-French 3-Factor model). Assume that the risk-
free rate is rf = 1%, and that the factor risk premia λMKT , λSMB and λHML are obtained
using all the available data.

Solution: From all the available data, the factor risk premia are λMKT = 7.08%, λSMB =
2.78%, and λHML = 4.91%.
For AAPL (βAAPL = 1.768), CAPM predicts:
CAPM
EAAPL = = 1% + 1.768 × 7.01% = 13.52%

2
 For KMRT (βKMRT = 1.493), CAPM predicts:
CAPM
EKMRT = 1% + 1.493 × 7.08% = 11.57%

For AAPL, FF3 predicts:

FF3
EAAPL = 1% + 1.837 × 7.08% + 0.075 × 2.78% − 2.687 × 4.91% = 1.02%

For KMRT, FF3 predicts:

FF3
EKMRT = 1% + 1.239 × 7.08% + 1.479 × 2.78% + 3.242 × 4.91% = 29.80%

(d) We see that for the period 2004-2008, CAPM produces similar estimates for the cost of
capital for AAPL and KMRT, while FF3 produces very different estimates. Do we get
similar results when we compute the estimates for the period 2007-2008?

Solution: Yes, the results for the period 2004-2008 are similar to the results for the
period 2007-2008. We get similar CAPM estimates, because the CAPM beta is not
very different for the two companies: thus, AAPM and KMRT are almost equally risky
according to CAPM. By contrast, FF3 accounts for the extra risk of small and value
firms (such as KMRT), and discounts for the fact that large and growth firms (such as
AAPL) are less risky.
Note that in practice it is better to use the whole 5-year period 2004-2008 when esti-
mating the CAPM and FF3 betas, not just the 2-year period 2007-2008.

(e) Which model should be used, CAPM or APT, in the estimation of the cost of capital
for AAPL and KMRT? Why do you think CAPM is used in practice much more than
APT? (It is ok to speculate in your answer to the last question.)

Solution: In principle, we should use the correct model of risk. We know that in an
empirical race, CAPM does not work well, especially when it comes to stocks. For
stocks, such as AAPL and KMRT, we know that FF3 produces better results. So it is
better in principle to use FF3 for computing the cost of capital.
In practice, CAPM is still widely used to compute the cost of capital. There are at least
three reasons for this. First, there is the force of habit: the industry has used CAPM for
a long time and understands it well, while APT is relatively newer. Second, CAPM is
a rigorous theoretical model with a clear intuition, while APT lacks a clear theoretical
foundation. Finally, if a firm gets close to bankruptcy (and thus, just like KMRT, is
likely to be a small value firm), it makes sense for its manager to use CAPM even if the
model is wrong. Indeed, in the example above, suppose the manager of KMRT used
APT instead and obtained a cost of capital of 26.91%, which is probably closer to the
correct value than 14.91%, which is the cost of capital according to CAPM. But with
a discount rate of 26.91%, most projects would be rejected because they would have a
negative NPV; while with a discount rate of 14.91% more projects would be accepted.
It is true that these projects would actually lose money on average (since the true NPV
is likely negative), but it would be in fact optimal for the manager of a company close
to bankruptcy to take risky projects. This is because a successful project leads to a
gain for the shareholders, while an unsuccessful project does not lead to a large loss for

3
 the shareholders: due to the company’s value being close to zero, the losses would be
absorbed by the bondholders instead! This is called risk shifting. It might not be nice
to the bondholders of the firm, but it is optimal for the shareholders, and remember
that legally the manager represents the interests of the shareholders of the firm, not the
interests of the bondholders.

Problem 2. Which of the following observations would provide evidence against the weak-
form efficient market hypothesis? Explain in detail your reasoning.

(a) You do a study which shows that past capital expenditures of a company have predictive
power for future abnormal returns.

Solution: No. This is evidence against the semi-strong-form EMH, not against the
weak-form EMH. Evidence against the weak-form EMH would be to show that past
prices have predictive power for future abnormal returns.

(b) You find that weekly returns are negatively correlated.

Solution: Yes. If the weak-form EMH is true, there should be no correlation of price
changes, and therefore no correlation of returns. Indeed, since you found that returns
are negatively correlated, this implies that a positive change in price today predicts (on
average) a negative change in price tomorrow. This would violate the weak-form market
efficiency: you have found a way to use past prices to predict future price changes. Note
that this argument only works in the short run, such as at a weekly horizon. Over the
long run, one needs to worry about expected returns.

(c) Stocks returns in September are significantly lower compared to other months of the
year.

Solution: Yes. According to the weak-form EMH, you should not be able to predict
any future returns, including returns in September.

(d) You find that daily prices are positively correlated.

Solution: No. Since the best predictor of the price tomorrow is the price today (ac-
cording to EMH), prices must be correlated. What are not correlated according to EMH
are price changes.

Problem 3. True, False, or Uncertain? Explain in detail your reasoning.

(a) If markets are not semi-strong-form efficient, it implies that markets cannot be weak-
form efficient, either.

Solution: False. Markets can be weak-form efficient, in the sense that information
contained in past prices cannot predict future returns — while at the same time other it
is possible that other publicly available information, such as various accounting measures
for the firm, can predict future returns.

4
 (b) The Efficient Market Hypothesis always implies the Random Walk Hypothesis.

Solution: False. This is only true in the short run. In general, if there are no dividends
until the next period, we have the equation

Pt+1 = Pt (1 + r) + et+1 ,

where et+1 is unpredictable at t (and has mean zero). Over the short run (if the time
interval less than one month), the expected return r can be approximated to zero. In
that case, Pt+1 ≈ Pt + et+1 , which is by definition a random walk. Over the long run,
the expected return r can no longer be ignored.

(c) Suppose you find that an investment strategy based on publicly available information
that produces positive average returns, even after you adjust for market risk. This
would indicate that the market is inefficient (in the semi-strong form).

Solution: Uncertain/False. You are told that your strategy’s CAPM alpha is positive,
but you would like to also know that is significant. Also, CAPM is unlikely to be the
correct risk model, so you should also compute the FF3 alpha of your strategy, to be
sure that you still have a positive and significant alpha.
Of course, you have to wonder about data mining, and whether your finding is econom-
ically significant. Have you tried to see whether your alpha is positive out-of-sample or
in various sub-samples? Can you check it in other countries? Does alpha stay positive
even after you account for transaction costs and short-sale constraints?

Problem 4. You read the following news in the financial media:

NEW YORK (CNN/Money), Friday, January 28, 2005 – Procter & Gamble announced
the largest acquisition in its history Friday, agreeing to buy Gillette
in a $57 billion deal that combines some of the world’s top brands and
could lead to further mergers involving products consumers know and love.
‘‘It’s a dream deal,’’ Buffett said, adding that he would increase his
holdings so that he would end up with 100 million shares of P&G by the
time the deal closes. (The deal gives Buffett a one-day profit, on paper,
of roughly $645 million and a whopping $4.4 billion profit overall.)
Under the deal announced early Friday, Procter & Gamble will pay 0.975
share of its common stock for each share of Gillette common stock. Based
on Thursday closing prices, that would represent an 18 percent premium
for Gillette shares.
The news sent P&G stock down about 2 percent while Gillette jumped about
12 percent in afternoon trading.

A historical prices search on Yahoo! Finance gives you the following (end-of-day) stock prices
for the two companies:

5
 Date P&G Gillette
26 Jan 2005 54.33 44.56
27 Jan 2005 54.22 45.40
28 Jan 2005 53.07 51.28
31 Jan 2005 52.17 50.40
1 Feb 2005 51.84 49.99
2 Feb 2005 52.34 50.38

Also, what is not reported in the news is that on the merger announcement date (Fri, Jan
28), Gillette also paid a dividend of $0.16 per share.

(a) What is the return on P&G and Gillette on the announcement day? (Do not forget to
include the dividend.)

Solution:
PPG,28Jan05 + dPG,28Jan05 53.07 + 0.00
par rapport au jour d'avant r̃PG,28Jan05 = −1= − 1 = −2.12%
PPG,27Jan05 54.22
P + dG,28Jan05 51.28 + 0.16
r̃G,28Jan05 = G,28Jan05 −1= − 1 = 13.30%
PG,27Jan05 45.40

Notice that the news above only reports a 12% jump for Gillette (instead of at least
13%), but this is probably because the news came out before the end of the trading day.

(b) Suppose markets are efficient. If Gillette had not payed its dividend on 28 Jan 05, what
do you think its closing price would have been at the end of that day?

Solution: The market would not have discounted the cash dividend of $0.16/share, so
the price would have been

PG,28Jan05 = 51.28 + 0.16 = 51.44

(c) According to the news, using Thursday closing prices, P&G offered Gillette 0.975 ×
$54.22 = $52.86 per share, which represents a premium of (52.86 − 45.40)/45.40 =
16.44%. (Notice the slight difference from the reported figure of 18%.) However, instead
of the price of Gillette jumping all the way to the offer price of $52.86, it only jumped
to $51.44 (including the dividend).
Assuming markets are efficient, what likelihood did the market place on the fact that
the merger between P&G and Gillette will be successful? Give a number for that
probability. What are the assumptions that you use to compute this likelihood?

Solution: From Thursday’s closing price of $45.40, Gillette jumped to $51.44 instead of
going all the way up to $52.86. Therefore the market-implied probability of the merger
should be approximately
51.44 − 45.40
Probabilitymerger = = 81%.
52.86 − 45.40
This is a high number, which indicates that the market is quite confident that the merger
will be successful (and as we know, in the end the merger was indeed successful).

6
 We assumed that if the merger is not successful, the price of Gillette would revert to what
it was before the announcement ($45.40). This is not necessarily the case, since even if
the merger fails the announcement might be interpreted as good news (for example as
the beginning of a bidding war), and the price of Gillette would revert to a higher level
than $45.40.
Problem 5. Please answer briefly the following questions, based on the clip from the article
reproduced below. You do not need to read the article in detail, just make sure you get a
general idea before you answer the questions.
Here is what happens when a company announces that it will split its shares. In a typical
split, a company exchanges each existing share for two new shares, each worth half as much
as the previous shares at the time of the exchange. Companies often advocate splits as a
means of broadening their shareholders base.

(a) According to the article, what happens to the share price after a stock split announce-
ment? What are the explanations offered for the findings? Which explanations do you
think are more plausible and why?
Solution: There are in fact two types of rises in price described in the article:
1. Stocks which announce a 2-to-1 split jump on average 3.4%.
2. In the subsequent year after the split, stocks continue to rise, and beat the market
by 8%.
As mentioned in the article, finding 1 can be due to
ˆ The halo effect: underreaction of investors to the stream of good news that accom-
pany firms experiencing stock splits. Once investors hear about the stock split,
they regard it as additional good news, since they realize there is more good news
to come in the near future.

7
 ˆ The signalling hypothesis (suggested by Ikenberry): managers who announce the
stock split are in effect announcing that the stock will continue to grow. As this is
new information, the price rationally goes up.
Notice that the halo effect does not quite explain why investors regard the stock split
as additional good news. Since they keep underreacting to other good news, why not
assume that they underreact to the stock split as well, or even not react at all? The
signalling hypothesis sounds in the end more plausible.
Finding 2 (the subsequent rise after the stock split) is essentially a form of momentum,
and the author offers the typical behavioral explanation for momentum: investor un-
derreaction. Nevertheless, momentum can be also rationally justified (although perhaps
less plausibly), if we agree that the FF4 model—which includes momentum—is rational.
(b) What are the implications for market efficiency of this article?

Solution: Finding 1 (an immediate jump in price after the stock split announcement)
is in line with market efficiency. Finding 2, however, contradicts market efficiency if
we accept that there is a positive subsequent price drift after accounting for the correct
risk factors. However, as we know from the course, it is not clear what the correct risk
factors are. Besides, the article is silent about which risk factors are used in the event
study. For instance, is the momentum factor (UMD) used? Since the article does not
discuss these issues, overall the implications for market efficiency of the article are not
clear.
(c) Why are the smaller firms and those with lower price-book ratios “just where you’d
expect management’s signals to be ignored”? Explain this phrase according to the
rational vs. behavioral debate (Fama vs. Shiller) in market efficiency: which side does
the author of this article prefer?

Solution: The rational (“Fama”) view is that small/value stocks are riskier: they have
a higher exposure to the systematic risk factors SMB/HML, and this is why they have
higher expected returns. The behavioral (“Shiller”) view regarding, e.g., the low prices
(and high expected returns) of value stocks is that analysts do not pay enough attention
to the value segment of the market, so prices are low compared to the “glamorous”
segment of growth firms. The author of the article seems to side with the behavioral
view by saying that magangement’s signals are ignored. This “ignorance” would not
occur in the rational view.
(d) When doing an event study, you should remove the market trends by comparing the
actual stock returns with the returns predicted by a benchmark risk model. According
to the article, what benchmark model do you think it was used: the CAPM or the
Fama–French 3-factor model (FF3)?

Solution: It is not entirely clear what the benchmark model is. The article says: “Over
the first 12 months, his stocks beat the market by an average of 8% annually.” This
indicates that the CAPM was used. (Technically, the abnormal return should subtract
beta times the market return, but beta is probably close to 1, so it’s ok to just subtract
just the market return.)
Could it be that FF3 was used? The article is from 1996, 4 years after the “beta is
dead” Fama–French paper, so it is in principle possible. However, the article mentions

8
 that “additional research by Ikenberry indicates that the underreaction is greatest for
smaller firms and for those with lower price/book ratios.” Since the FF3 model accounts
for the value and small firm effects, it probably was not used.

(e) Describe the steps that you would take to reproduce the results of this event study.

Solution: This is a standard event study:

ˆ Choose a benchmark model, typically CAPM or FF3 (or FF4, etc.).

ˆ For each stock in the sample, denote by t = 0 the day of the announcement of
the split. Estimate the parameters of the benchmark model using data from the
pre-split announcement period (before t = 0).
ˆ Then calculate the returns predicted by the model for the period immediately
surrounding the announcement. The difference between the actual and predicted
values is the abnormal return (or abnormal residual). Denote by ARt the abnormal
return on each day t for a fixed period (for example between t = −15 until t = 30).
ˆ Compute the cumulative abnormal return:

CARt = AR−15 + AR−14 + · · · + ARt−1 + ARt .

ˆ For each day t, take the average CARt for all the stocks in the sample, and plot
this average.