Abstract

An algorithm is presented that locally approximates the nonlinearity of stochastic unit root (STUR) models by n linear models. The previous integer n is chosen so that the Hadamard matrix of order n can be defined. The strategy STUR(n), then consists in creating n linear models from this Hadamard matrix and taking their average forecast. A purchase (sell) signal is made if the obtained average forecast is positive (negative). Subsequently, a comparison is made with respect to competing models (Moving average strategies) to assess their ability to forecast the variation of five international indexes. It is found, after taking account transaction costs, that STUR(n) generates generally the highest profitability in the out-of-sample data.

This work is licensed under a Creative Commons Attribution 4.0 License.