Prediction Based Portfolio Optimization Model Using Neural Networks with an Emphasis on Leading Stocks of NSE

This work discusses most frequently traded stocks of National stock exchange of India. A prediction based portfolio optimization model is considered to present an ideal portfolio out of the considered stocks. Neural network has been used to predict stock returns and a risk measure is derived that has the same foundation as that of mean variance model. The architecture of the network is designed experimentally. The cross-correlation structure between the fluctuations of price for frequently traded stocks in national stock exchange has been studied for a period of last one year. The structure of interactions is studied using the spectral properties of the cross-correlation matrix. Spectral decomposition of the cross correlation matrix is used in finding the most influencing underlying stocks of NIFTY 50. The portfolio formed by the presented model is considered to verify the presence of influencing stocks as determined by the spectral analysis.