Forecasting Flashcards
7 cards from real CBE practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Forecasting flashcards as text
Which of the following best describes a stationary time series?
Answer: A series with constant mean, variance, and autocovariance over time
Stationarity requires that statistical properties—mean, variance, and autocovariance—do not depend on the point in time at which the series is observed.
A company's sales forecasts show that actual sales are always below the forecast. This suggests the forecasts are:
Answer: Overestimates (positively biased)
If actual sales consistently fall below forecasts, the model is systematically predicting too high—a positive bias in the forecast.
Holt's double exponential smoothing extends simple exponential smoothing by also smoothing:
Answer: The trend component
Holt's method adds a second smoothing equation for the trend, allowing forecasts to project a changing level rather than a flat one.
Theil's U statistic is used to evaluate a forecast by comparing it to:
Answer: A random walk (naïve) forecast
Theil's U compares the root mean squared error of a model forecast to that of a naïve no-change forecast; U < 1 means the model outperforms the naïve benchmark.
In regression forecasting, multicollinearity primarily causes problems by:
Answer: Making individual coefficient estimates unstable and hard to interpret
Multicollinearity inflates standard errors of correlated predictors, making coefficients unreliable even if overall model fit is high.
Which approach is most appropriate for forecasting a non-stationary economic series with a stochastic trend?
Answer: ARIMA with differencing (d ≥ 1)
A stochastic (random-walk) trend requires differencing to achieve stationarity before ARIMA modeling, unlike deterministic trends removable by regression.
Prediction intervals in forecasting are wider than confidence intervals for the mean because they:
Answer: Also account for the variance of future individual observations
Prediction intervals cover the distribution of a single future value, adding irreducible observation error to parameter uncertainty, making them inherently wider.