Prospective Clustering Tehran Stocks Exchange for Portfolio Management

Clustering methods commonly use the past data, but this paper has tried to use prospective data and compare to former. After using time series methods to create prediction model for stocks, they clustered by k-means, fuzzy c-means and self-organized maps (SOM). In addition, all stocks were clustered by those methods. After clustering, the stocks could be selected from these groups for building a portfolio. Portfolios optimized by Markowitz model to impose the lowest risk to investor for a certain return, and the best portfolio were selected by Sharp Ratio. The following indicators, return, standard deviation, P/E, Beta, number of buyers, number of deals and value of transaction have been used at different times from the Tehran Stock Exchange for April 2010 to April 2014. Result depicts that retrospective clustering present the better portfolio compared to prospective clustering, and k-means creates the most compact cluster compared to others.

کلیدواژه ها:

Portfolio Management ، Time Series Regression ، K-means Clustering ، Self-organized maps (SOM) ، Fuzzy C-means