Dash, Gordon, H

Advisor Department

Finance and Insurance




Artificial Neural Network, WinORS, Hedge Fund Returns, Volatility, Forecasting, Econometric

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 License.


Alternate Models for Forecasting Hedge Fund Returns

Michael Holden

Faculty Sponsor: Gordon Dash, Finance and Decision Sciences

Investors have always wanted to improve the efficiency of modeling realized volatility to maximize directional trading returns and substantially improve profitability. As proposed, this honors project will provide evidence from hedge fund returns that a Radial-Basis Function (RBF) artificial neural network (ANN), specifically the Kajiji-4 RBF-ANN dominates other forecast methods in producing one-period ahead change-of-direction when forecasting the expected returns of various hedge fund indexes.

I began this project by collecting historical economic data in monthly increments to serve as the dependent variables. The primary independent variable used in this study are two types of Treasury securities (short-term and long-term) to represent interest rates as well as the Volatility Index (VIX). The VIX index serves as a proxy for options implied volatility in the equity markets. These independent variables are used to predict the returns of multiple hedge fund indexes which serve as the dependent variables.

The data was plugged into the RBF-ANN in order to solve the economic models. The ANN first took time to train using 33% of the data, and then it validated the remaining 67% of the data to measure the fitness. The study proceeded to calculate the residual by taking the difference of the actual data and the RBF-ANN predicted data. The RBF-ANN showed that the data was very fit as the mean square error (MSE) was relatively small.

Overall, I have found that the RBF-ANN has done quite well in predicting the returns of various hedge fund indexes. The scope of the project will examine three well-known hedge fund styles.

Holden_PowerPoint.pptx (1022 kB)