An examination of the merits of inductive reasoning in accounting research utilizing a data mining methodology.

摘要

Although unsanctioned by classical and modern philosophers, the use of inductive research methods in business and academics is increasing due to the presence of very large data sets, accessible problem-solving algorithms and inexpensive computer hardware. Tools and techniques know collectively as data mining approach research from an inductive, hypothesis free, perspective. The inductive perspective has been praised in accounting research as having the potential to reveal non-intuitive relationships in data. Inductive data-mining techniques have the potential to reveal actionable information from accounting data at a much faster rate than could be attained using purely deductive methods of hypothesis formation and testing. This paper examines the merits of inductive methods in accounting research by comparing the results of a fundamental analysis using inductive data-mining tools and methods with results discovered in previous research. Financial statement fundamentals and stock returns are organized into a data mart and analyzed graphically and statistically for the relationships they contain. Models of winner stocks are produced using two algorithms: (1) a neural regression and (2) a decision tree based on chi-square automatic interaction detection (CHAID). It is found that the decision tree modeled by the CHAID algorithm is superior to the more general neural regression in predicting winner stocks. The portfolio returns for the decision tree model exceed those found in previous accounting research by a wide margin. Based on the success of the inductive tools and methods, it is concluded that induction should be given some credit for its merits as a research methodology. Future research is needed to refine the methodology for applying inductive tools and methods to accounting data.