Decision Trees and Financial Variables

Literature Review

Decision trees in the management sciences are used to model the likely consequences of various decisions for a particular problem domain (Corner & Kirkwood, 1991). Within artificial intelligence, a decision tree algorithm is a classification method where the induction process “constructs a flowchart-like structure where each internal (non-leaf) node denotes a test on an attribute, each branch corresponds to an outcome of the test, and each external (leaf) node denotes a class prediction” (Han & Kamber, 2006). The decision tree models will often “help users to investigate how each feature weighs in the decision-making process and uncover the logical relations between the features and classified results” (Lin, Wang, & Chung, 2010). For instance, in an effort to improve the efficiency of call centers, researchers have created a decision tree to help identify the factors that were most crucial in influencing the performance of the call center (Kyper, Douglas, & Blake, 2012).

The value of automated decision trees for financial forecasting has been established (Tsang, Yung, & Li, 2004). Roko and Gilli (2008) feed fundamental and technical ratios to a decision tree algorithm to forecast high earning stocks. Given the Adaptive Market Model, one might want decision trees to adapt over time, and research has shown how to evolve decision trees (Fu, Golden, & Lele, 2003). Of course, a robust approach to multi-attribute decision making in finance might use more than decision trees, and other methods, such as fuzzy correlation rule mining (Robinson & Amirtharaj, 2014) or interpolative Boolean algebra (Mandic, Delibasic, & Radojevic, 2015), could be applied.

For decades researchers have been identifying patterns in accounting data and predicting stock returns by exploiting those patterns (Richardson, Tuna, & Wysocki, 2010). Some research shows that investors make false assumptions about companies with extreme book-to-market (BM) values and with large accruals. To earn above average returns, Piotroski (2000) first identifies high BM stocks and then ones with strong financial fundamentals. Sloan (1996) showed that stock prices do not adequately reflect extraordinary items in the accruals and said, “stock prices are found to act as if investors 'fixate' on earnings, failing to reflect fully information contained in the accrual and cash flow components of current earnings until that information impacts future earnings.” Galdi and Hermesmeyer (2010) apply Piotroski's method to stocks in Brazil but find it prudent to put different weights on the independent financial variables. High BM stocks are considered 'value' stocks because they are deemed to be undervalued relative to their assets, while low BM stocks are considered 'growth' stocks. To successfully forecast which stocks will be high-earners, Mohanram (2005) first screens for 'growth' stocks and then identifies which independent financial variables can be used to forecast high earners among the growth stocks.