CEA Economic Forecasting Techniques 3 — Questions and Answers
Question 1: In a Vector Autoregression (VAR) model, lag length selection is critical. Which information criterion penalizes additional parameters most heavily, often selecting the most parsimonious model?
- Akaike Information Criterion (AIC)
- Hannan-Quinn Criterion (HQC)
- Bayesian Information Criterion (BIC) (Correct answer)
- Final Prediction Error (FPE)
Correct answer: Bayesian Information Criterion (BIC)
The BIC (Schwarz criterion) applies the strongest penalty for additional parameters, consistently selecting shorter lag lengths than AIC in large samples.
Question 2: Which scenario best illustrates the 'Lucas critique' in economic forecasting?
- A model fails because of multicollinearity among regressors
- Policy changes alter agent behavior, invalidating forecasts based on historical relationships (Correct answer)
- Seasonal adjustment introduces spurious cycles
- Measurement error in GDP causes forecast bias
Correct answer: Policy changes alter agent behavior, invalidating forecasts based on historical relationships
Robert Lucas argued that structural parameters estimated from historical data change when policy regimes shift, rendering reduced-form forecasts unreliable.
Question 3: A fan chart in economic forecasting is used to:
- Display point forecasts across multiple models simultaneously
- Communicate forecast uncertainty through probability bands around a central forecast (Correct answer)
- Compare in-sample fit versus out-of-sample performance
- Show historical revisions to past data releases
Correct answer: Communicate forecast uncertainty through probability bands around a central forecast
Fan charts show widening probability intervals (e.g., 50%, 75%, 90% confidence bands) to convey that uncertainty grows with the forecast horizon.
Question 4: The Diebold-Mariano test is used to:
- Test whether a time series is stationary
- Compare the predictive accuracy of two competing forecasting models (Correct answer)
- Detect cointegration between two non-stationary series
- Identify the optimal smoothing parameter in exponential smoothing
Correct answer: Compare the predictive accuracy of two competing forecasting models
The Diebold-Mariano test assesses whether differences in forecast accuracy between two models are statistically significant using a loss differential statistic.
Question 5: Regime-switching models, such as Hamilton's Markov-switching model, are particularly useful for economic forecasting because they:
- Eliminate unit roots from macroeconomic time series
- Allow model parameters to change discretely across unobserved states (e.g., recession vs. expansion) (Correct answer)
- Provide exact probability forecasts without estimation error
- Replace structural models with atheoretic statistical methods
Correct answer: Allow model parameters to change discretely across unobserved states (e.g., recession vs. expansion)
Markov-switching models capture nonlinearities by allowing parameters to shift probabilistically across latent states, improving recession and expansion forecasting.
Question 6: Which of the following is an example of a leading economic indicator used in composite index forecasting?
- Unemployment rate (lagging)
- Manufacturing new orders (leading) (Correct answer)
- Prime interest rate (lagging)
- Consumer price index (coincident)
Correct answer: Manufacturing new orders (leading)
Manufacturing new orders tend to rise before actual production increases, making them a leading indicator included in the Conference Board's LEI.
Question 7: When a forecaster reports a 95% prediction interval, it means:
- The model is 95% accurate on in-sample data
- There is a 95% probability the actual value falls within the interval under the assumed model (Correct answer)
- The forecast error is less than 5% of the actual value
- 95% of past forecasts were within this range
Correct answer: There is a 95% probability the actual value falls within the interval under the assumed model
A 95% prediction interval is constructed so that, if the model is correctly specified, 95% of future realizations will fall within the stated bounds.
CEA Economic Forecasting Techniques 3
In a Vector Autoregression (VAR) model, lag length selection is critical.
Which information criterion penalizes additional parameters most heavily, often selecting the most parsimonious model?