Application of Artificial Neural Networks to the Tactical Asset Allocation Model
Date of Award
Doctor of Philosophy (PhD)
Graduate School of Computer and Information Sciences
Michael J. Laszlo
Tactical asset allocation (T AA) is an investment strategy that switches an investment among different types of asset classes such as stocks, bonds or cash. The strategy consists of identifying the assets with the best performance potential within a defined short period of time. Artificial neural networks (ANN) have been successfully used to model nonlinear systems such as stock and bond price series. ANN have been used to forecast stock and bonds prices. The goal of most investment managers is to beat the market. This means to outperform an index representative of a specific overall market. The objective of T AA is to switch an investment to the asset class that will yield the best performance during the upcoming period. The success of TAA depends on the accuracy of the prediction of which asset class will yield the best performance.
The purpose of this dissertation project was to investigate the effectiveness of artificial neural networks in forecasting the probability that one asset class would outperform two others by the end of a 30-day period. The asset classes considered were stocks, bonds and money market in the United States. An ANN was trained with fundamental and technical historical data. ANNs with different topologies were trained before arriving at an optimal configuration. A three-layer feedforward neural network offered the lowest generalization error. The selected ANN was trained to forecast the probabilities of each of the three investment asset classes outperforming the other two. With the forecast probabilities two TAA portfolios were created. The first portfolio was 100% invested on the asset class with the highest probability of outperforming the other two. The second portfolio followed a risk-neutral tactical asset allocation strategy based on the forecast probabilities.
The accumulated returns at the end of the test period were compared to four benchmarks that represented buy-and-hold strategies. The first T AA portfolio outperformed all benchmarks. The risk-neutral portfolio outperformed all but the accumulated returns of the S&P500 index. The tests proved that an ANN is effective in forecasting the probabilities of one asset class outperforming others. Also, the results were used to create portfolios that outperformed the benchmarks.
C. Augusto Casas. 2002. Application of Artificial Neural Networks to the Tactical Asset Allocation Model. Doctoral dissertation. Nova Southeastern University. Retrieved from NSUWorks, Graduate School of Computer and Information Sciences. (445)