Working Paper: CEPR ID: DP8867
Authors: Rocio Alvarez; Maximo Camacho; Gabriel Perez-Quiros
Abstract: We examine the finite-sample performance of small versus large scale dynamic factor models. Our Monte Carlo analysis reveals that small scale factor models out-perform large scale models in factor estimation and forecasting for high levels of cross-correlation across the idiosyncratic errors of series belonging to the same category, for oversampled categories and, especially, for high persistence in either the common factor series or the idiosyncratic errors. Using a panel of 147 US economic indicators, which are classified into 13 economic categories, we show that a small scale dynamic factor model that uses one representative indicator of each category yields satisfactory or even better forecasting results than a large scale dynamic factor model that uses all the economic indicator
Keywords: business cycles; output growth; time series
JEL Codes: C22; E27; E32
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| small scale dynamic factor models (SSDFM) (C22) | large scale dynamic factor models (LSDFM) (C22) |
| small scale dynamic factor models (SSDFM) (C22) | forecasting accuracy (C53) |
| large scale dynamic factor models (LSDFM) (C22) | forecasting accuracy (C53) |
| highly correlated indicators (C10) | LSDFM performance (C52) |
| redundant data (Y10) | forecasting performance (C53) |
| one representative indicator (C43) | SSDFM performance (C69) |
| all 147 economic indicators (C43) | LSDFM performance (C52) |