兩篇財務工程論文類神經網路與Copula的應用

Abstract

This thesis encompasses two essays in financial engineering. In the first essay, we apply artificial neural network methodology to evaluate mutual fund performance. Application of financial information systems requires instant and fast response for continually changing market conditions. The purpose of this essay is to construct a mutual fund performance evaluation model utilizing the fast adaptive neural network classifier (FANNC), and to compare its performance in classification and forecasting with those from a backpropagation neural network (BPN) model. FANNC is a newly-developed model which combines features of adaptive resonance theory and field theory. In our experiment, the FANNC approach requires much less time than the BPN approach to evaluate mutual fund performance. Root mean square error is also superior for FANNC. These results hold for both classification problems and for prediction problems, making FANNC ideal for financial applications which require massive volumes of data and routine updates. Consequently, an on-line evaluation system can be established to provide real time mutual fund performance for investors. In the second essay, we apply a copula methodology to study the dynamic dependence structures between Chinese market and other major markets in the world, admitting China’s increasing integration with the global economy. Through the use of time-varying copula models, our data shows conditional copulas outperform unconditional copula and the conventional GARCH models. We consistently find Chinese market experiences a high degree of dependence with markets in Japan and in the Pacific. Meanwhile, the variations of these dependences are also higher. For investors interested in China’s market, our results can provide more timely suggestions on portfolio diversification, risk management, and international asset allocation than those implied by static models.

This thesis encompasses two essays in financial engineering. In the first essay, we apply artificial neural network methodology to evaluate mutual fund performance. Application of financial information systems requires instant and fast response for continually changing market conditions. The purpose of this essay is to construct a mutual fund performance evaluation model utilizing the fast adaptive neural network classifier (FANNC), and to compare its performance in classification and forecasting with those from a backpropagation neural network (BPN) model. FANNC is a newly-developed model which combines features of adaptive resonance theory and field theory. In our experiment, the FANNC approach requires much less time than the BPN approach to evaluate mutual fund performance. Root mean square error is also superior for FANNC. These results hold for both classification problems and for prediction problems, making FANNC ideal for financial applications which require massive volumes of data and routine updates. Consequently, an on-line evaluation system can be established to provide real time mutual fund performance for investors. In the second essay, we apply a copula methodology to study the dynamic dependence structures between Chinese market and other major markets in the world, admitting China’s increasing integration with the global economy. Through the use of time-varying copula models, our data shows conditional copulas outperform unconditional copula and the conventional GARCH models. We consistently find Chinese market experiences a high degree of dependence with markets in Japan and in the Pacific. Meanwhile, the variations of these dependences are also higher. For investors interested in China’s market, our results can provide more timely suggestions on portfolio diversification, risk management, and international asset allocation than those implied by static models.