A New Data-Driven Stock Selection Model Framework Using Portfolio Theory

Abstract Stock filtering and selection in a portfolio is a critical problem given the large number of stocks across industries, market capitalizations, tenures, product types; the amount of information regarding their financial health and performance, valuation, company management and the different degrees of uncertainty and volatility associated with each of them. Previous studies propose theoretical models considering mean- variance method and several risk constraints. This paper describes a practical methodology of filtering and selecting stocks with solid fundamentals and available at an attractive valuation, out of thousands of available stocks in the index. Along with Piotroski F-score that considers three parameters- profitability, liquidity, and operating efficiency, we develop two new scores based on comprehensive data analysis- Custom and Valuation score, to gauge the company’s fundamental performance and valuation/expensiveness of the stock price. We consider NSE, India for all analysis, and NIFTY as the index for comparing market performance. We crawl data (source: Bloomberg) of 2,000 exchange-traded stocks for FY 2013–2017 and filter it down to 47 best stocks. Further, we narrow down to 10 final stocks keeping in mind diversification across industry segments, market capitalization and attractiveness of the firm. For validation, we check the performance of these 10 final stocks between 1st April, 2017 and 12th February, 2018. We find that our basket of stocks performed significantly better and give 9.5 times higher returns than market returns. Finally, we also validate the performance as of today with metrics like compounded sales growth, compounded profit growth, stock price CAGR, and Return on Equity.