A Hybrid Machine Learning and Dynamic Nonlinear Framework for Determination of Optimum Portfolio Structure

Abstract

Capital market investment is a growing stream of the economic literature. It has been a prime concern of a large number of investors belonging to different clusters or income groups for two reasons mainly. First, the construction of a portfolio, which deals with the selection of the stocks. Second, the formulation of an appropriate investment strategy, which calls for minimizing the risk while maximization of the return, i.e., optimization of the constructed portfolio. Following the broad framework as suggested in the seminal work of Markowitz [1], this research attempts to address the issue of portfolio optimization based on risk and return parameters while dynamically allocating the weights to the constituent stocks. In the first part of this study, k-means clustering is applied to a heterogeneous sample of 53 number of stocks enlisted with the NSE during the year 2012–2017. The purpose is to classify the stocks in three categories (such as low stock price, medium stock price, and high stock price) based on their monthly closing return. In the second phase, this study focuses on finding out the distribution of weights among the stocks belonging to the portfolio by using the generalized reduced gradient (GRG) method under the dynamic environment. Finally, this study attempts to validate the results by applying perception mapping. We have found eight stocks in the cluster of low stock price which is the sample studied in this research. We have observed that dynamic allocation of weights led to minimization of risk and the finding is validated through a perceptual map.