Predictions, nonlinearities and portfolio choice

Meer over het boek

Finance researchers and asset management practitioners invest considerable effort into optimal asset allocation, leading to extensive research on portfolio decision-making, quantitative modeling, and prediction models. This study integrates three typically isolated research areas: the predictability of asset returns and their covariance matrix, optimal portfolio decision-making, and nonlinear modeling through artificial neural networks, focusing on their implications for predictions and portfolio construction. The work emphasizes incorporating predictability into asset allocation, particularly addressing nonlinearities. It contributes to portfolio choice literature in two significant ways. First, it analyzes the impact of nonlinear predictions on portfolio performance, motivated by evidence of linear predictability, using empirical predictions for an investor in equities (DAX index), bonds (REXP index), and a risk-free rate. Second, it presents a solution to the dynamic programming problem for intertemporal portfolio choice, employing functional approximations of the investor’s value function with artificial neural networks. This method adeptly handles multiple state variables, allowing for an analysis of the effects of predictive parameters, estimation biases, and a Bayesian investor's perspective on intertemporal portfolio choice. A key empirical finding reveals that residual correlation among state variables significantly i