| Summary: |
This investigation seeks to construct financial conditions indices (FCIs) by the partial least squares (PLS) method with the aims (i) that the FCIs should outperform interest rate, which is conventionally used in small VAR (Vector Auto-Regression) models to present the predictive macro-impacts of the financial markets, and (ii) that the FCIs are adequately invariant during regular updates to resemble non-model based aggregate indices. Both aims are shown to be attainable as long as the FCIs are tailor-made with carefully selected components and suitably targeted macro variables of forecasting interest. The positive outcome sheds light on why the widely used principal component analysis (PCA) approach is ill-suited to the tasks here whereas why the PLS route promises a fruitful way forward.
|
