Research on how to use large amount of data in time series forecasting to be published in JASA.
"A team involving the S3RI member Professor Zudi Lu is publishing work on using large amount of data in time series forecasting in the prestigious Journal of the American Statistical Association.
In this paper, two semiparametric model averaging schemes are suggested to obtain more accurate estimates and forecasts of time series by using a large number of conditioning variables. These schemes are based on the useful ideas either of screening out the unimportant regressors marginally by a nonlinear KSIS technique developed or of estimating the common factors for a large number of exogenous forecasting variables. With the survived variables, a semiparametric penalized method of Model Averaging MArginal Regression (MAMAR) is further suggested to select the most helpful forecasting variables by an optimal combination of the significant marginal regression and auto-regression functions. Theoretical justifications are developed together with both simulation and an empirical application demonstrated."