​​Economics Seminar | Forecasting with Partial Least Squares Using a Large Number of Predictors 

We consider a model in which a large number of predictors are available for a target viable to be forecasted and a subset or a whole set of the common factors in the predictors are determinants of the target variable. For the model, we consider Partial Least Squares (PLS) estimation of the factors relevant for forecasting. Asymptotic and finite-sample properties of the PLS factors are examined. We find that the number of relevant PLS factors crucially depends on the covariance structure of the of the common factors in the predictors The number of the relevant PLS factors is not necessarily equal to the number of the common factors that are correlated with the target variable. In addition, it is shown that the forecasting power of the PLS factors deteriorates when more than the relevant PLS factors are used.