Forecasting for Monetary and Financial Stability

1. What problem were Dynamic Factor Models (DFMs) originally designed to solve?

• Estimating exchange rates

• Predicting stock market crashes

• Calculating interest rates

• Extracting common signals from many variables when data is limited

Why can a large VAR with many variables lead to overfitting?

• Because it estimates too many parameters relative to observations

• Because it ignores time dynamics

• Because it uses too many lags

• Because it uses seasonal data

What does “latent” mean in the term “latent factors”?

• Directly observed in the data

• Seasonal pattern in the data

• Hidden or unobserved but inferred from the data

• Random noise in the data

In the 2-variable DFM example (GDP and interest rates), the measurement equation links:

• Factors to lag structures

• Factors to seasonal patterns

• Observed variables to latent factors

• Observed variables to other observed variables

In a DFM, the persistence parameter ϕ close to 1 indicates:

• The factor changes rapidly

• The factor is slow-moving and persistent

• The factor is random noise

• The factor has no memory

DFMs can handle datasets with monthly and quarterly data because they:

• Assume all data is quarterly

• Use only the higher frequency data

• Use state-space models and the Kalman filter

• Ignore the missing observations

In Stock–Watson (2016), how many disaggregated series were used in the real activity dataset?

• 58

• 11

• 3

• 120

What method did Stock–Watson (2016) use to determine the number of factors?

• Akaike Information Criterion (AIC)

• Cross-validation

• Bai–Ng Information Criteria

• Rolling window estimation

How many factors did Stock–Watson (2016) select using the Bai–Ng criteria?

• 11

• 3

• 5

• 1

Which of the following is an advantage of DFMs?

• They guarantee causality

• They can handle mixed data frequencies

• They only work with a single variable

• They remove the need for model re-estimation