One of my interests in finance is applying machine learning and computational techniques to areas of finance that are typically more about "feel" and "instinct" like fundamental analysis of companies for "stock picking".
My latest little experiment is on transportation stocks (won't list them here because this isn't a "recommendation"). I started off with 102 stocks and used a clustering algorithm used a lot in "big data" (k-means for anyone who cares) to cluster the stocks based on common sized financial statements.
I took each cluster of stocks in 3 years (2012, 2013, and 2014) and built equally weighted portfolios of the stocks in each cluster and looked at each portfolio's return over the next year. Tacked those returns onto the end of my table and then combined the years into one table.
I looked at that table a couple of ways - a correlation matrix and a graphical 'map' of the relationships between all the data. To skip to the punch line: You might want to look at a company's SG&A and interest expense line items on the income statement - the strongest correlations for returns were between return and SG&A and between return and interest expense.
I'm not sure I would call this 'actionable' but it's definitely interesting and something I'm going to look at with more data... like 500 to 1,000 stocks, I think.
The picture: (it's big)
My latest little experiment is on transportation stocks (won't list them here because this isn't a "recommendation"). I started off with 102 stocks and used a clustering algorithm used a lot in "big data" (k-means for anyone who cares) to cluster the stocks based on common sized financial statements.
I took each cluster of stocks in 3 years (2012, 2013, and 2014) and built equally weighted portfolios of the stocks in each cluster and looked at each portfolio's return over the next year. Tacked those returns onto the end of my table and then combined the years into one table.
I looked at that table a couple of ways - a correlation matrix and a graphical 'map' of the relationships between all the data. To skip to the punch line: You might want to look at a company's SG&A and interest expense line items on the income statement - the strongest correlations for returns were between return and SG&A and between return and interest expense.
I'm not sure I would call this 'actionable' but it's definitely interesting and something I'm going to look at with more data... like 500 to 1,000 stocks, I think.
The picture: (it's big)
Last edited:
