Factor Modeling for High Dimensional Time Series

We briefly compare an econometric factor model and a statistical factor model, the latter being to capture the linear dynamic structure of the data y(t) only. We propose a method for decomposing {y(t)} into a common component and a white noise in the sense of a statistical factor model, together with an eye-ball test for finding the number of factors. Rates of convergence for various estimators are spelt out explicitly.