Assignment 2

Sebastian Vogel
[email protected]@epfl.ch
Extranef 127

1. Utility theory (10 points). Consider an individual whose end of period wealth
will be W = Y R where Y := ey is his labor income and R = e is a multiplicative
and insurable risk (e.g., car accident). Assume that Y and R are jointly lognormally
distributed with E[y] = µY , E[] = −µ , V [y] = σy2 , V [] = σ2 and corr[y, ] = ρ.
Assume that the investor is risk-averse with CRRA utility given by

Wα
u(W ) = , α 6= 0. (1)
α

(a) Derive the coefficients of relative risk aversion and absolute risk aversion of this
agent.
(b) Compute the maximum relative insurance premium π the individual would be
willing to pay to insure against the risk R. That is, find π such that

E[u(Y R)] = E[u(Y e−π )].

(c) How does π change with µy , σ , σy , ρ, α. Interpret these findings. Are they rea-
sonable?

2. Normal distribution for stock returns (10 points). Using the Python code
developed in the first lecture, download daily stock returns over the period December
31st , 1999, to December 31st , 2018 from WRDS for the following 5 companies: Apple,
General Electrics, Goldman Sachs, Microsoft, and Procter & Gamble (use the permco’s
given in Problem Set 1).

(a) For each stock compute the mean and variance of daily simple returns.

1
 (b) Plot the empirical density function of stock returns (i.e., an histogram estimate of
the underlying density function) and compare to (i) the normal distribution with
the same mean and variance as the empirical distribution and (ii) the normal
distribution with mean and variance of the ‘winsorized’ empirical distribution.
To compute the ‘winsorized’ empirical distribution, simply keep only the daily
returns with absolute value less than 4%.
(Effectively you should make one separate graph for each stock with the empirical
histogram and two different normal densities, corresponding to the two calibra-
tions of means and variances).
(Hint: If S is a panda DataFrame, then the method S.plot.bin might be useful.)
(c) For each stock, compute the 95% and 99% Value-at-Risk from the empirical dis-
tribution of returns, and compare to the Value-at-Risk that would obtain if the
distributions were normal with corresponding means and variances.
(Hint: If S is a panda DataFrame, then the method S.quantile might be useful.
Also, you might want to import the function scipy.stats.norm which contains
a lot of useful methods linked to the normal distribution.)
(d) Based on your computations, does the normal distribution seem to describe ap-
propriately the distribution of daily stock returns?

3. Risk, return, and diversification in MSCI indices (20 points). The MSCI
Global Equity Indices are widely tracked global equity benchmarks. They cover more
that 75 countries and span both developed and emerging markets (EMs). They also
serve as the basis for over 500 exchanged traded funds throughout the world. The
Excel file MSCI.xlsx contains total return indices of seven MSCI indices (sourced from
Bloomberg) from December 1987 to February 2016:

• Four developed markets: US, Japan, Switzerland, and Germany

The total return indices are in USD and include reinvested dividends.1
1
The time series for Europe & Middle East exclude reinvested dividends as these are only available from
2000 onwards; hence, the average return is downward biased for this index. MSCI also computes total return
indices in local currency.

2
(a) Download the data into Python (Hint: Clean the Excel file “MSCI.xlsx” and then
save it as a “.csv” document. Finally, use the command:
S=pd.read csv(‘filepath\MSCI.csv’,delimiter=‘,’,index col=‘Date’).)
(b) For each country/region, compute monthly returns and then the arithmetic mean,
geometric mean, standard deviation, skewness, kurtosis, and 95% VaR. Plot his-
tograms of returns. Compare the riskiness of emerging market equities with de-
veloped market equities.
(Hint: Recall that, if S is a panda DataFrame, then the method S.describe()
automatically computes basic summary statistics. Note also that the package SciPy
contains lots of useful functions for statistics. For example, to import a function
which computes the skewness, use the code from scipy.stats import skew. For
the 95% VaR, the function np.percentile can be useful.)
(c) Verify the rule of thumb that geometric mean ≈ arithmetic mean - 1/2 variance
(d) Compute the correlation matrix for returns in developed markets. Then, compute
the correlation matrix for returns in Switzerland and the EM regions. For a
Swiss investor, which category (developed or EM) offers the best diversification
opportunities in the sense of being least correlated with Swiss index returns on
average?
(Hint: If S is a panda DataFrame, then the method S.corr() might be useful.
Also, remember that if you want to apply some method (let’s call it method) only
to a subset of columns, you can use the command S.iloc[:,col idx].method,
where col idx is a list of columns.)
(e) Consider a rolling 24-month window. For each window, compute the average
correlation across all markets. Plot the time series of the average correlation.
What could be the reason for the trend that you see in the time series. When does
the time-series peak and what does that tell you about the benefits of diversifying
internationally.
(Hint: If ts corr is a time-series of all correlation pairs, then the command
ts corr.loc[date].values[idx].mean() will compute the average correlation
of the pairs idx at the date date.)

3
Figure 1: Total return index for MSCI Switzerland