Formal Assignment 1
Formal Assignment 1
In class you saw the most famous asset pricing model, the CAPM. It predicts
that investors will demand higher expected excess returns on stocks which are
more sensitive to market variation. We will now look at firm-specific characteris-
tics of stocks. (Please round your results to 2 decimal places unless more digits
are necessary to be informative.)
Use the 25 Fama-French portfolios formed on size and B/M as test assets.
Consider value-weighted portfolios only. Conduct your analysis for the sample
period from 1963:07 to 2020:06 (this will be the sample period for the entire
assignment). Data is available from Professor Kenneth French’s website.
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/25_Portfolios_5x5_CSV.zip
a. Report arithmetic average raw (i.e. not log) monthly returns and market
betas of these portfolios in two tables with the size-quintiles in rows and the B/M-
quintiles in columns. What, if any, are the broad patterns in these average returns
across assets? Is this what you would expect given the CAPM prediction?
[3 points, 2-3 lines]
b. Report time-series standard deviations of portfolio returns using the same
table format as in point (a). Then, report Sharpe ratios of portfolio returns using
the same table format. What, if any, are the broad patterns in these Sharpe ratios
across assets?
Hint: You can find data on the risk free rate on Kenneth French’s website.
[3 points, 2-3 lines]
c. In the same format as before report t-statistics for tests that each of the 25
portfolio returns is different from zero, and in an additional row (column) report
t-statistics for tests that quintiles 5 minus 1 (i.e. the average return on a portfolio
which is long quintile 5 stocks and short quintile 1 stocks) have average returns
different from zero. Comment on your results.
[5 points, 2-3 lines]
d. Consider your estimates in (a), (b), and (c). Do they pose a challenge to
the CAPM theory? Can you think of an economic explanation for any of these
cross-sectional return patterns?
[4 points, 5-6 lines]
1
Question 2. Factor Models.
To better understand your findings in Question (1), you will now look at factor
models. The main idea is that stocks might be affected by different types of risks.
If the asset-pricing model is well-specified, then both time- and cross-sectional
variation of portfolio excess returns should be fully explained by its exposure to
the chosen set of risk factors.(Please round your results to 2 decimal places unless
more digits are necessary to be informative.)
Consider the following data on factors:
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/F-F_Momentum_Factor_CSV.zip
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/F-F_ST_Reversal_Factor_CSV.zip
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/F-F_Research_Data_Factors_CSV.zip
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/F-F_Research_Data_5_Factors_2x3_CSV.zip
Using the above, you will estimate the following factor models:
i. The Fama-French Five factor model.
ii. The Fama-French Three factor model plus the Momentum factor.
iii. The Fama-French Three factor model plus the ST Reversal factor.
a. Estimate time-series regressions using all 25 portfolios for each of the mod-
els above. For each model, compute and report the multivariate Gibbons, Ross,
and Shanken (GRS) test statistic (see Chapter 12, pp. 230 in the Cochrane text-
book; alternatively, visit support class for discussion of this) for the entire set of
portfolios. What is the null hypothesis that you are testing here? Is it rejected?
Report a total of 3 GRS statistics and their corresponding statistical significance.
[15 points]
b. How do the above three models compare to each other based on the GRS
statistics?
[5 points, 3-4 lines]
c. You will now check how well these three models explain the cross-sectional
variation of portfolio returns. Use the Fama-MacBeth (1973) procedure discussed
in class. Compute the factor risk premia and the pricing error for each model.
Report your estimates together with the corresponding t-statistics. Which null
hypothesis are you testing here? Are all factors priced? What can you say about
the models’ pricing errors?
Hint: You should be running a total of 3T cross-sectional regressions, where T is
the total number of months in the sample.
[15 points]
d. Explain the difference between a “factor” and a “priced factor”.
[5 points, 3-4 lines]
e. Is there statistical support for the inclusion of every one of the factors in the
models above? Suppose some of the factors turn out to be statistically insignificant
in the time-series regressions. Would there still be a reason to include them?
[10 points, 5-6 lines]
2
Question 3. Data Mining.
This question will urge you to think carefully about the problem of data-mining
in empirical asset pricing. With only one history of stock returns it will always
be possible to identify a couple of new factors (corresponding to the major princi-
pal components of the realised returns’ distribution) able to capture the variation
in the panel of returns arbitrarily well in-sample. Since the 1960s over 300 such
factors have been discovered (see Harvey et al.).(Please round your results to 2
decimal places unless more digits are necessary to be informative.)
a. With the above in mind, how would you justify that a newly discovered
factor is important? Which key characteristics should it have? See Goyal (2012)
for a fine discussion.
[10 points, 3-4 lines]
You will now check whether including a new factor has additional explanatory
power over a classic three-factor Fama-French model. Consider the investment
growth factor proposed in Xing (2008). The factor is defined as the return on a
zero-cost trading strategy that consists of a long position in stocks with low in-
vestment growth rates and a short position in stocks with high investment growth
rates.
b. First, check whether there is any abnormal return variation across portfo-
lios of stocks with different investment growth rates. For that you will need to
download monthly returns on 10 U.S. stock portfolios formed on past investment.
Data is available from Kenneth French’s website:
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/Portfolios_Formed_on_INV_CSV.zip
Estimate the CAPM for each of the 10 portfolios and plot the CAPM αs against
investment deciles. Based on the observed pattern, what sign would you expect
the investment factor beta to have and why?
[10 points, 1-2 lines]
You will now construct the investment growth factor. First, identify the high
and low investment growth (IG) groups: the low IG group consists of the 30%
of firms with the lowest past investment (the bottom three deciles of the down-
loaded data), and the high investment group consists of the 30% of firms with the
highest IG (the top three deciles). Second, take the value- weighted return differ-
ence between the low investment growth groups and the high investment growth
groups.1 Such constructed IG factor captures the return difference between firms
with different levels of past investment.
1
Kenneth French’s file provides you with the average size and number of firms in each
portfolio.
3
[5 points, 2-3 lines]
d. Consider the classic Fama-French model and its extension that includes the
investment growth factor:
e e
rp,t = αp + βp rm,t + βp,HM L HM Lt + βp,SM B SM Bt + εp,t , (1)
e e
rp,t = αp + βp rm,t + βp,HM L HM Lt + βp,SM B SM Bt + βp,IN V IN Vt + εp,t , (2)
Estimate the two models given by eq. (1) and (2) on the 25 portfolios used in
Question (1). For each model, report in a table the adjusted R2 , α and its corre-
sponding t-statistics. Then, again for each model, report the GRS test statistics
and the corresponding p-value. Compare your results for the two models: does
adding a new factor improve the Fama-French model? Provide a possible expla-
nation for your findings.
[10 points, 3-4 lines]
References
• Fama, Eugene F.; MacBeth, James D., 1973, “Risk, Return, and Equilib-
rium: Empirical Tests”, The Journal of Political Economy, Vol. 81, No. 3.
(May - Jun., 1973), pp. 607-636.
http://ecsocman.hse.ru/data/965/126/1231/fama_macbeth_tests_1973.pdf
• Harvey, Campbell R.; Liu, Yan; Zhu, Heqing, 2016, “. . . and the Cross-
Section of Expected Returns”, The Review of Financial Studies, Volume 29,
Issue 1, 1 January 2016, Pages 5-68.
https://faculty.fuqua.duke.edu/~charvey/Research/Published_Papers/P118_and_the_cross.PDF
• Xing, Yuhang, 2008, “Interpreting the Value Effect Through the Q-Theory:
An Empirical Investigation”, The Review of Financial Studies, Volume 21,
Issue 4, 1 July 2008, Pages 1767-1795.
https://doi.org/10.1093/rfs/hhm051