CEA Economic Forecasting Techniques 5 — Questions and Answers
Question 1: Bayesian Model Averaging (BMA) addresses model uncertainty in forecasting by:
- Selecting the single model with the highest posterior probability
- Weighting forecasts from all candidate models by their posterior model probabilities (Correct answer)
- Using cross-validation to select the best regularization parameter
- Averaging only models that pass a significance threshold
Correct answer: Weighting forecasts from all candidate models by their posterior model probabilities
BMA integrates over model uncertainty by combining forecasts weighted by each model's posterior probability, accounting for the fact that the true model is unknown.
Question 2: In long-horizon economic forecasting, which method is generally preferred over ARIMA models?
- Exponential smoothing with fixed parameters
- Structural economic models anchored by long-run theory (Correct answer)
- Simple random walk with no drift
- Autoregressive models with very long lag lengths
Correct answer: Structural economic models anchored by long-run theory
At long horizons, theoretical anchors (e.g., purchasing power parity, potential output) matter more than statistical time-series properties, favoring structural models.
Question 3: The concept of 'cointegration' is important in economic forecasting because cointegrated variables:
- Must always be differenced before modeling
- Share a long-run equilibrium relationship that constrains their joint dynamics (Correct answer)
- Are always stationary in levels
- Cannot be used in VAR models
Correct answer: Share a long-run equilibrium relationship that constrains their joint dynamics
Cointegrated series are individually non-stationary but have a stationary linear combination, which is exploited in Error Correction Models (ECMs) to improve forecasts.
Question 4: Which of the following best describes 'judgmental adjustment' in economic forecasting?
- Automated outlier correction using statistical rules
- Expert modification of model-based forecasts using information not captured by the model (Correct answer)
- Backcasting to fill historical data gaps
- Adjusting for data revisions using real-time databases
Correct answer: Expert modification of model-based forecasts using information not captured by the model
Judgmental adjustment incorporates soft information (e.g., anticipated policy changes, geopolitical events) that statistical models cannot capture, improving real-world forecast accuracy.
Question 5: In the context of real-time forecasting, 'data vintage' refers to:
- The age of the statistical model being used
- The specific release of a data series available at a given point in time (Correct answer)
- The number of periods used in model estimation
- Annual revisions to national accounts methodology
Correct answer: The specific release of a data series available at a given point in time
Different vintages of GDP, for example, reflect successive revisions; using the correct vintage is essential for realistic out-of-sample forecast evaluation.
Question 6: A forecaster observes that their model systematically underpredicts inflation during supply shocks. This systematic error is best described as:
- Random forecast noise
- Forecast bias (Correct answer)
- Overfitting
- Parameter instability
Correct answer: Forecast bias
Systematic directional errors indicate forecast bias; an unbiased forecast should have errors that are zero on average with no predictable pattern.
Question 7: Which approach is most effective for forecasting rare economic events such as financial crises?
- Standard linear ARIMA models estimated on full sample
- Tail-risk models, extreme value theory, or early warning systems using nonlinear indicators (Correct answer)
- Simple random walk benchmarks
- Fixed-effects panel models with annual data only
Correct answer: Tail-risk models, extreme value theory, or early warning systems using nonlinear indicators
Rare events lie in the tails of distributions; extreme value theory and early warning system models trained on pre-crisis indicators are better suited than standard linear models.
CEA Economic Forecasting Techniques 5
Bayesian Model Averaging (BMA) addresses model uncertainty in forecasting by: