Nonparametric prediction of stock returns guided by prior knowledge
Michael Scholz (),
Jens Perch Nielsen () and
Stefan Sperlich ()
Additional contact information Michael Scholz: Karl-Franzens University Graz
Jens Perch Nielsen: Cass Business School
Stefan Sperlich: Universite de Geneve
One of the most studied questions in economics and finance is whether equity returns or premiums can be predicted by empirical models. While many authors favor the historical mean or other simple parametric methods, this article focuses on nonlinear relationships. A straightforward bootstrap-test confirms that non- and semiparametric techniques help to obtain better forecasts. It is demonstrated how economic theory directly guides a model in an innovative way. The inclusion of prior knowledge enables for American data a further notable improvement in the prediction of excess stock returns of 35% compared to the fully nonparametric model, as measured by the more complex validated R2 as well as using classical out-of-sample validation. Statistically, a bias and dimension reduction method is proposed to import more structure in the estimation process as an adequate way to circumvent the curse of dimensionality.