خوشه‌بندی داده‌های بازار سهام هند برای مدیریت پرتفوی

In this paper a data mining approach for classification of stocks into clusters is presented. After classification, the stocks could be selected from these groups for building a portfolio. It meets the criterion of minimizing the risk by diversification of a portfolio. The clustering approach categorizes stocks on certain investment criteria. We have used stock returns at different times along with their valuation ratios from the stocks of Bombay Stock Exchange for the fiscal year 2007–2008. Results of our analysis show that K-means cluster analysis builds the most compact clusters as compared to SOM and Fuzzy C-means for stock classification data. We then select stocks from the clusters to build a portfolio, minimizing portfolio risk and compare the returns with that of the benchmark index, i.e. Sensex.

One of the decision problems in the financial domain is portfolio management and asset selection. Under the extremely competitive business environment, in order to face the complex market competitions, financial institutions try their best to make an ultimate policy for portfolio selection to optimize the investor returns. A formal model for creating an efficient portfolio was developed by Markowitz (1952). In his model the return of an asset is its mean return and the risk of an asset is the standard deviation of the asset returns. Risk was quantified such that investors could analyze risk-return choices. Moreover, risk quantification enabled investors to measure risk reduction generated by diversification of investment. So diversification of investment is essential to create an efficient portfolio. The problem of selecting well diversified stocks can be tackled by clustering of stock data. Clustering as defined by Mirkin (1996) is “a mathematical technique designed for revealing classification structures in the data collected in the real world phenomena”. Clustering methods organize a data set into clusters such that data points belonging to one cluster are similar and data points belonging to different clusters are dissimilar. In this paper we demonstrate the implementation of stock data clustering using well known clustering techniques namely K-means, self organizing maps (SOM) and Fuzzy C-means. The stock market data is clustered by each of the above methods. The optimal number of clusters for the stock market data using each clustering technique is carried out. The stock data contains attributes as a series of its timely returns as well as the valuation ratios to present a clear position of their market value. These are the direct investment criteria that are being considered for stock selection. Thus the resulting clusters are a classification of high dimensional stock data into different groups in view of the difference between return series along with current market valuation of stocks. After clustering stock samples are selected from these clusters to create efficient portfolio. The process is simulated for certain iterations and average risk and return is found out. It is easy to get the portfolios with lowest risk for a given level of return, using certain optimization model which is demonstrated at the end of the paper. In order to create efficient portfolios with Markowitz model, we use the clustering method to select stocks in the paper, called clustering-based selection in our paper. The remainder of the paper is organized as follows. The remainder of this paper is organized as follows. Section 2 describes relevant literature review. Section 3 presents the clustering-based stock selection method. Section 4 shows problem description. Section 5 depicts experimental results. In Section 6, the conclusion is presented.

This paper suggests how to integrate clustering techniques like K-means, SOM and Fuzzy C-means into portfolio management and build a hybrid system of getting efficient portfolios. It can reduce a lot of time in selection of stocks as stocks of similar categories can be easily grouped into a cluster and thus best performing stocks from those groups can be selected. Our work can find a lot of applications in software development for areas like asset management, algorithmic trading and Investor’s technical information in financial markets. In our research we chose timely stock returns and valuation ratios, however from investment perspective other dimensions or factors that influence the performance of a stock can also be considered and thus that would help to refine the classification. Cases where certain temporary macroeconomic factors affect market performances for a short period of time could also be considered in our approach. We considered data for the fiscal year 2007 and since, it was the beginning world recession with markets worldwide acting bearish, and most of the stocks considered in the data have negative returns. But from clustering perspective we have compared the portfolio performances with the benchmark index, i.e. BSE Sensex. We summarize relevance of our work as: 1. In this study, various stocks from BSE were analyzed for their timely returns and market valuations. The data collected was actual rather than simulated data and the results can be considered more practical. 2. After analysis of the clustering methods, for our stock data K-means formed well compact clusters as compared to Fuzzy C-means and SOM neural network. 3. A clustering approach to portfolio management and selection of stocks to reach the efficient frontier was demonstrated.