Practice Midterm

Introduction to Time Series, Stat 153

October 13, 2023

Name:
Each question is either in true/false format, or multiple choice. For multiple choice, just choose the single
best option. In each case, make sure to fill in the box according to the answer you choose (true or false, or
the multiple choice option) completely. All questions are worth 1 point.
1. An i.i.d. sequence with zero mean and finite variance is always a white noise sequence.
⇤ True
⇤ False
2. In general, uncorrelatedness does not imply independence.
⇤ True
⇤ False
3. In general, the sample cross-covariance estimator assumes that the given processes are:
⇤ a. jointly strongly stationary
⇤ b. jointly weakly stationary
⇤ c. jointly periodic
⇤ d. jointly symmetric
4. For a moving average of white noise, the auto-covariance matrix is:
⇤ a. diagonal
⇤ b. banded
⇤ c. dense, but decaying off the diagonal
⇤ d. dense, and with no general structure
5. Applying a moving average filter to white noise __ the variance, compared to that of the original white
noise sequence itself.
⇤ a. “increases”
⇤ b. “decreases”
⇤ c. “does not change”
⇤ d. there is no universal rule, so none of the above apply
6. Even with no assumptions on their form, it is possible to accurately estimate the mean and variance
functions of a time series from a single observed sequence.
⇤ True
⇤ False
7. White noise has a __ mean function and __ variance function.
⇤ a. “constant”, “constant”
⇤ b. “constant”, “zero”
⇤ c. “zero”, “constant”
⇤ d. “zero”, “unit”
8. The mean and variance functions reflect the spread and trend, respectively, of a random sequence over
time.

1
 ⇤ True
⇤ False
9. A forecast that was made using only information that is available at the time the forecast was issued is
called:
⇤ a. ex-ante
⇤ b. ex-post
⇤ c. proper
⇤ d. calibrated
10. In order to fit a linear model and use traditional tools to conduct valid inference on the coefficients
(such as hypothesis tests or confidence intervals), we must assume that the true relationship between
the response and covariates is linear.
⇤ True
⇤ False
11. A forecast that was made using information would have only been available later, after the forecast was
issued, is called:
⇤ a. ex-ante
⇤ b. ex-post
⇤ c. proper
⇤ d. calibrated
12. The Gauss-Markov theorem asserts that the least squares estimator has the lowest __ among all linear
unbiased estimators, in a setting where we assume a linear model for the response with __ errors.
⇤ a. “variance”“,”uncorrelated”
⇤ b. “mean squared error”, “white noise”
⇤ c. “mean squared error”, “independent”
⇤ d. “variance”, “normal”
13. Using lagged features in a regression model allows us to:
⇤ a. make ex-ante forecasts when it may otherwise not have been possible
⇤ b. capture a richer dependence of the response on past feature values
⇤ c. achieve stationarity by removing dependence from recent time points
⇤ d. both a and b
14. Compared to MAPE, a notable difference of MASE is that it is:
⇤ a. not sensitive to small response values
⇤ b. not sensitive to correlations
⇤ c. scale-dependent
⇤ d. both b and c
15. The Gauss-Markov theorem is often paraphrased as “least squares is the BLUE”. The L here refers to
linearity in what sense?
⇤ a. The mean of the response y is assumed to be linear in the features X in the underlying model.
⇤ b. The least squares estimator is a linear function of y, and is being compared to other such
estimators.
⇤ c. Both a and b, at the same time.
⇤ d. None of the above.
16. Compared to MAE, a notable difference of MSE is that it is:
⇤ a. less sensitive to large error values
⇤ b. more sensitive to large error values
⇤ c. scale-dependent
⇤ d. robust to correlations
17. In any regression problem, for a feature matrix X with more columns p than rows n, there will be
infinitely many least squares solutions.

2
 ⇤ True
⇤ False
18. A key difference between ridge regression and the lasso is that ridge coefficient estimates are:
⇤ a. available in closed-form
⇤ b. dense
⇤ c. sparse
⇤ d. both a and b
19. The lasso and ridge regression cannot be applied to fit a model in which we have more lagged features
than observations.
⇤ True
⇤ False
20. A kernel smoother is a linear filter where the weights aj take the form, for a kernel function K and
bandwidth b:
⇤ a. aj = K(j/b) q
⇤ b. aj = K(j/b)/r i=1 K(i/b)
n

⇤ c. aj = K(j/b)/ i=1 K(i/b)

⇤ d. either a or b
21. White noise has a __ spectral density.
⇤ a. constant
⇤ b. decaying
⇤ c. diverging
⇤ d. oscillating
22. Any process that is of the form xt = U1 cos(2fiÊt) + U2 sin(2fiÊt) is stationary provided that:
⇤ a. U1 , U2 are uncorrelated
⇤ b. U1 , U2 are uncorrelated, each with mean zero
⇤ c. U1 , U2 are uncorrelated, each with mean zero and equal variance
⇤ d. none of the above
23. Large values of the periodogram indicate __ in the given time series.
⇤ a. “which points correspond to outliers”
⇤ b. “which lags are most auto-correlated”
⇤ c. “which frequencies we can ignore”
⇤ d. “which frequencies are predominant”
24. Any process that is of the form xt = A cos(2fiÊt + „) for an amplitude A and phase „ can also be
written as xt = U1 cos(2fiÊt) + U2 sin(2fiÊt) for two amplitudes U1 , U2 .
⇤ True
⇤ False
25. The relationship between the periodogram and the DFT is:
⇤ a. the DFT is the squared modulus of the periodogram
⇤ b. the periodogram is the squared modulus of the DFT
⇤ c. the DFT is the Fourier transform of the periodogram
⇤ d. there is no general relationship