# time series problem set page 40

1. (a) Consider the quarterly earnings of Johnson & Johnson from 1960 to 1980 in the file q-earn-jnj.txt .

Perform a log transformation of the data, detrend and deseasonalize the data, and subtract the mean, in order to obtain a sequence of observations that appears to be stationary with zero mean. Plot the sample autocovariance or autocorrelation function of the obtained time series. Perform the Box-Ljung test for m=5 and m=10 and draw conclusions. Use some forecasting method built in in the software you are using to forecast 24 values and plot the original series together with the 24 predicted values. [Hint: This is fairly straightforward if you use the software ITSM – This will be demonstrated in class].

(b) Consider the accidental deaths between 1973 and 1978 in the file Deaths.txt . Repeat the tasks stated in (a) without the log transformation and forecast 36 values for the deaths time series.

3. #1.8(a) see attachement p.40

and express the autocovariance function of the detrended and deseasonalized time series in terms of the stationary process {Y_t}.

4. Suppose that {X_t} is a stationary time series with mean mu and ACF rho(.). Show that the best mean square predictor of X_{n+h} of the form

a X_n + b is obtained by choosing a = rho(h) and b = mu (1-rho(h)), where the best mean square predictor minimizes the mean square error (MSE) E(X_{n+h} – predictor)^2.

a X_n + b is obtained by choosing a = rho(h) and b = mu (1-rho(h)), where the best mean square predictor minimizes the mean square error (MSE) E(X_{n+h} – predictor)^2.

5. Find the ACVF of the time series

Y_t = Z_t – 1.2 Z_{t-1} – 1.6 Z_{t-2}, where {Z_t} ~ WN(0, 0.25).

Y_t = Z_t – 1.2 Z_{t-1} – 1.6 Z_{t-2}, where {Z_t} ~ WN(0, 0.25).