Professional Documents
Culture Documents
=
=
=
=
mean(S);
std(S);
mean(logRetS);
std(logRetS);
The results of these computations are row vectors that contain the means
or the standard deviations of the corresponding values in each column.
Each element of these row vectors corresponds to one of the assets: IBM,
GOOGLE and SIEMENS, in that order. We can now proceed to determine
whether the statements are true or false.
The average log returns of the three assets are logRetAverages = [0.4667e
3, 0.3358e 3, 0.1145e 3]. They are close to zero. Therefore, the first
statement is true and the second one false.
It is easy to see that the GOOGLE prices (stored in S(:, 2)) are, on average,
about six times larger than the SIEMENS prices (stored in S(:, 3)). The
historical volatilities for IBM, GOOGLE and SIEMENS are logRetStds =
[0.0155, 0.0210, 0.0267], respectively. The volatility of GOOGLE (0.0210)
is smaller than the volatility of SIEMENS (0.0267). Therefore, the third
statement is true and the fourth one is false.
Finally, we can see from the plots that the time series of prices are clearly
non-stationary. For instance, the averages of the asset prices in the first
half of the time series are markedly different from the corresponding averages in the second half. Therefore, the fifth statement is false.
2
2. We continue studying the time series of asset prices from the previous
exercise. What is the approximate value of the correlation between the
log returns of IBM and GOOGLE?
0.0
0.3
0.4
0.5
0.6
0.5026
1.0000
0.5162
0.6301
0.5162
1.0000
There are significant positive correlations between the log returns of the
three assets. Given that the IBM log returns are stored in the first column
of logRetS and the GOOGLE log returns are stored in the second one,
the required correlation value is the (1, 2) entry of the matrix assetCorr,
which is approximately 0.5.
3. We continue with the analysis of the time series from the previous exercises. We will now compute the autocorrelations of the time series of log
returns, assuming that they are stationary. The autocovariance at lag n
of a stationary time series {X1 , X2 , . . . , XN } is
n = E [(Xm E[Xm ]) (Xm+n E[Xm+n ])] .
The lag-n autocorrelation is the autocovariance normalized by the standard deviations
n =
1 [log-ret] = 0.2,
1 [log-ret] = 0.2,
1 [log-ret] = 0.2,
1 [log-ret] = 0.0,
1 [log-ret] = 0.0,
1 [log-ret] = 0.0,
1 [log-ret2 ] = 0.2
1 [log-ret2 ] = 0.0
1 [log-ret2 ] = 0.2
1 [log-ret2 ] = 0.2
1 [log-ret2 ] = 0.0
1 [log-ret2 ] = 0.2
Hint: The lag-n autocorrelation of a time series can be obtained by computing the correlation between the values of the time series and the values
of the same series displaced n time steps. Assuming that the time series is
stored in vector X the Octave/MATLAB command to compute the lag-n
autocorrelation is
autocorr = corr(X((1+n):end), X(1:(end-n)));
Explanation: The lag-1 autocorrelations of the log returns of IBM are
computed using the Octave/MATLAB commands
lrIBM = logRetS(:,1);
n = 1;
autocorr = corr(lrIBM((1+n):end), lrIBM(1:(end-n)))
The resulting estimate (autocorr = -0.0278) is close to zero.
If we compute the lag-1 autocorrelation for the squares of the log returns
lrIBM2 = lrIBM.^2;
n = 1;
autocorr2 = corr(lrIBM2((1+n):end), lrIBM2(1:(end-n)))
we obtain (autocorr2 = 0.1759), which is significantly different from
zero. These are common features of time series of stock prices: Price
changes in different days are generally uncorrelated. This is consistent
with the hypothesis that price movements are unpredictable. By contrast,
as noted by Mandelbrot in 1963, the magnitude of price movements exhibits positive correlations: large changes tend to be followed by price
changes that are also large, although it is not possible to forecast in which
direction the movement will be. This phenomenon is known as volatility
clustering.
4. Consider an asset whose initial price at t0 (today) is S(t0 ) = S0 . The
owner of a European digital call option on this asset with maturity t0 + T
receives an amount of money A if the value of S(t0 + T ) is above K (the
strike of the option) and 0 otherwise. Therefore, the payoff at maturity of
this digital call option is
A if S(t0 + T ) K
payoff (S(t0 + T )) =
0 if S(t0 + T ) < K
4
expression for C - P , the difference between the prices of the call and
the put options. Use this expression to answer the following question:
how does C P behave when , the volatility of the underlying asset,
increases?
C P decreases.
C P increases.
C P increases if S0 KerT and decreases if S0 < KerT .
C P decreases if S0 KerT and increases if S0 < KerT .
* C P is independent of the volatility.
Hint: The price of a derivative product is the discounted expected value
of the corresponding payoff under risk-neutral probability
Price(t0 ) = erT E[Payoff(t0 + T )],
where r is the risk-free interest rate. The payoff at maturity for a call
option is max(S(t0 + T ) K, 0). Derive the expression of the payoff of a
put option.
Consider an investor who purchases a call option and writes a put option.
The payoff of this investors portfolio is the payoff of the call minus the
payoff of the put. Compute the value of this portfolio as the discounted
expected value of the payoff under risk-neutral probability. This value is
C P , the difference between the prices of the call and the put options.
In the derivation you will need to use the following results:
The discounted expectation of the value of the asset at any time in
the future is its price today
er(tt0 ) E [S(t)] = S0 ,
t > t0 .
PayoffCall PayoffPut
S(t0 + T ) K.
the asset price will be S(t0 + T ) and the cash debt will be KerT erT = K
due to the payment of interest at rate r (the risk-free interest rate).
We have seen that both portfolios pay the same amount at maturity.
Therefore, if their value today were different, a market agent would be
able to make profit without incurring risk by buying the cheap one and
selling the expensive one. If we assume that this free lunch is not possible (the so-called no arbitrage assumption), the value today of both
portfolios must be the same
C P = S(t0 ) KerT .
This is the put-call parity.