The kernel trick for nonlinear factor modeling

Factor modeling is a powerful statistical technique that permits common dynamics to be captured in large panel of data with few latent variables, or factors, thus alleviating the curse dimensionality. Despite its popularity and widespread use for various applications ranging from genomics finance, this methodology has predominantly remained linear. This study estimates factors nonlinearly through kernel method, which allows flexible nonlinearities while still avoiding We focus on factor-augmented forecasting single time series high-dimensional setting, known as diffusion index macroeconomics literature. Our main contribution twofold. First, we show proposed estimator consistent it nests linear principal component analysis well some nonlinear estimators introduced literature specific examples. Second, our empirical application classical macroeconomic dataset demonstrates approach can offer substantial advantages over mainstream methods.