Data Mining in Indian Equity Markets: building low Risk, Market beating Portfolios

• Published in: Finansy: teoriâ i praktika (Online) Vol. 27; no. 5; pp. 115 - 127
• Main Authors: Mitragotri, S. R., Patel, N.
• Format: Journal Article
• Language: English
• Published: Government of the Russian Federation, Financial University 23-10-2023
• Subjects: mining association rules; portfolio-risk; return-determinants; stock returns
• ISSN: 2587-5671; 2587-7089
• DOI: 10.26794/2587-5671-2023-27-5-115-127

Over the last five decades, business academics have identified over 300 determinants that potentially influence stock returns. However, we still do not know whether all return determinants are equally important, or whether there is a smaller set of determinants that has a disproportionately larger influence on stock returns. Can mining historical data help us find this smaller set of return determinants that has a disproportionately higher influence on stock returns? Using historical data from the Indian market, we build a large database of investments with more than 74,000 investments spread over a period of 132 months. From this database, using “association rule mining” method, we are able to mine a strong set of “association rules” that point to a smaller set of “return determinants” that are seen more frequently in investments that beat index returns. From a pool of thirty-seven return determinants, using “association rule mining”, we were able to find out a small set of key return determinants that are seen most frequently in investments that beat index returns in India. Portfolios created from these “association rules” have a portfolio risk lower than the market risk and provide index-beating returns. “Out-of-sample” portfolios created using these association rules have portfolio “Beta” less than one and provide returns that beat the market returns by a significant margin for all holding periods in the Indian market. Through this paper, we demonstrate how portfolio managers can mine “association rules” and build portfolios without any limits on the number of factors that can be included in the screening process.