Time Series Analysis Practice test with solutions Q1. Consider the time series xt = −2t + wt + 0.1wt−1 , where xt are the observations and wt ∼ iid N (0, 1). (a) Is this time series stationary? Justify your answer. (2 pts) Answer: E(xt ) = E[−2t] + E[wt ] + 0.1E[wt−1 ] = −2t The mean is not constant and depends on t. No, xt is not stationary. (b) Prove that the first difference series 5xt = xt − xt−1 is stationary. (12 pts) Answer: 5xt = −2t + wt + 0.1wt−1 + 2(t − 1) − wt−1 − 0.1wt−2 = −2 + wt − 0.9wt−1 − 0.1wt−2 First we compute the mean: E[5xt ] = E[−2] + E[wt ] − 0.9E[wt−1 ] − 0.1E[wt−2 ] = −2 Second we compute the autocovariance: γ5x (h) = cov(5xt+h , 5xt ) = cov(−2 + wt+h − 0.9wt+h−1 − 0.1wt+h−2 , −2 + wt − 0.9wt−1 − 0.1wt−2 ) = cov(wt+h , wt ) − 0.9cov(wt+h , wt−1 ) − 0.1cov(wt+h , wt−2 ) −0.9cov(wt+h−1 , wt ) + 0.81cov(wt+h−1 , wt−1 ) + 0.09cov(wt+h−1 , wt−2 ) −0.1cov(wt+h−2 , wt ) + 0.09cov(wt+h−2 , wt−1 ) + 0.01cov(wt+h−2 , wt−2 ) By independence of wt ,we have: γ5x (h) = 1.82 −0.81 −0.1 0 1 h=0 h = ±1 h = ±2 elsewhere. The mean is constant and the auto-covariance only depends on s, t through their difference h, so we conclude that 5xt is stationary. Q2. Let {xt , t = 0, 1, 2, . . .} be a random walk with constant drift µ, defined by x0 = 0 and xt = µ + xt−1 + wt , t = 1, 2, 3, . . . where xt are the observations and wt ∼ iid N (0, σw2 ). (a) What are the mean function and autocovariance function of this process {xt }? Is this time series stationary? Justify your answer. (6 pts) Answer: First we compute the mean: xt = µ + wt + xt−1 = µ + wt + µ + wt−1 + xt−2 xt = wt + wt−1 + 2µ + xt−2 ... xt = t−1 X wt−j + tµ + x0 = j=0 t−1 X wt−j + tµ j=0 " t−1 # X E[xt ] = E wt−j + E[tµ] = tµ j=0 Second we compute the autocovariance: γ(t, t+h) = cov t−1 X j=0 γ(t, t+h) = t−1 X wt−j + tµ, t+h−1 X ! wt+h−i + (t + h)µ i=0 cov (wt−j , wt−j ) = tσw2 j=0 xt is not stationary because the mean and the autocovariance function depends on t. (b) Prove that the first difference series 5xt = xt − xt−1 is stationary. (6 pts) Answer: 5xt = xt − xt−1 = µ + wt 2 First we compute the mean: E[5xt ] = E[µ] + E[wt ] = µ Second we compute the autocovariance: γ5x (h) = cov(5xt+h , 5xt ) = cov(µ + wt+h , µ + wt ) By independence of wt ,we have: γ5x (h) = σw2 h=0 0 |h| > 0 The mean is constant and the auto-covariance only depends on s, t through their difference h, so we conclude that 5xt is stationary. Q3. Consider the two series xt = w t yt = wt − 0.5wt−1 + t , where wt and t are independent white noise series with variances 2 and 3, respectively. (a) Determine the cross-covariance function (CCF), ρxy (h) relating xt and yt . (10 pts) Answer: First we compute the cross-covariance function of xt and yt . γxy (h) = cov(xt+h , yt ) = cov(wt+h , wt − 0.5wt−1 + t ) Therefore our cross-covariance function is: 2 h=0 γxy (h) = −1 h = −1 0 h 6= 0, −1 Second we compute the autocovariances of x and y: γx (h) = cov(xt+h , xt ) = cov(wt+h , wt ) 3 By independence of wt ,we have: γx (h) = 2 h=0 0 |h| ≥ 1 γy (h) = cov(yt+h , yt ) = cov(wt+h − 0.5wt+h−1 + t+h , wt − 0.5wt−1 + t ) Again, by independence of wt and t ,we have: 5.5 h=0 γy (h) = −1 |h| = 1 0 |h| ≥ 2 Now, we compute the cross-correlation function of xt and yt . γxy (h) γxy (h) ρxy (h) = p = √ 11 γx (0)γy (0) √ 2/ 11 √ ρxy (h) = −1/ 11 0 h=0 h = −1 h 6= 0, −1 (b) Show that xt and yt are jointly stationary. (14 pts) Answer: Two time series, say, xt and yt , are said to be jointly stationary if they are each stationary, and the cross-covariance function γxy (h) = cov(xt+h , yt ) is a function only of lag h. First we compute the mean of xt and yt : E[xt ] = E[wt ] = 0 E[yt ] = E[wt ] − 0.5E[wt−1 ] + E[t ] = 0 Second the autocovariance functions of x and y by part(a): 2 h=0 γx (h) = 0 |h| ≥ 1 4 5.5 γy (h) = −1 0 h=0 |h| = 1 |h| ≥ 2 The mean is constant and the auto-covariance only depends on s, t through their difference h, so we conclude that xt and yt are stationary. Also, the cross-covariance function of xt and yt 2 γxy (h) = −1 0 h=0 h = −1 h 6= 0, −1 is a function only of lag h. Therefore, xt , yt are each stationary and the crosscovariance function only depends on the lag h, we can conclude that xt and yt are jointly stationary. 5