Daily Speculations The Web Site of Victor Niederhoffer and Laurel Kenner


The Chairman
Victor Niederhoffer



About Victor Niederhoffer

2005 All content on site protected by copyright

Write to us at:(address is not clickable)

Clustering Analysis and Financial Performance, by Victor Niederhoffer

In the course of studying the applicability of the use of clustering methods to financial markets, a field with several hundred thousand methodological and application papers, I came across an interesting line of studies applying qualitative clustering methods to determining companies' future financial performance from the content of their past quarterly reports.

The studies by Kloptchenko from Abo Akademi University take 10 years of annual reports and classify the companies into bad, neutral and good return on equity improvements in the next year. He then takes the words used in each of the three categories of annual reports and sees if there are differences between them in the three categories that the words appear in. He was able to come up with 60% accuracy versus 33% random after the data mining. Good words were admissible:  approach, award, career, vary, wealth. Bad words were: burgeon, chance, collapse, complex. uncertain, undergo, unforeseeable.

For example, "discreet, stockholder, intelligent, profit, diverse, extraordinary, innovative and succeed " were positive in the positive class but negative in the negative predictive model. Similarly, "stress, cumulative, losses, unknown, doubt, " have positive weight in the negative class predictive model and negative weights in the positive class predictive model..

By looking only at words that appeared a certain threshold of times, an improvement in accuracy was possible". The author in another paper has applied this work to analyzing future performance from past quarterly reports  the author concludes: "Before a dramatic change occurs in company financial performance, we see a change in the written style of a financial report. If a company's position is worse during next quarter, the report of the current quarter gets more pessimistic, even though the actual financial performance remains the same."

There are so many variables, so many differences in company performance, so many implicit hypotheses, so much correlation and interaction of the content of the financial numbers with the future results, that with the small sample size of company years considered, possibly less than the number of hypotheses implicitly considered, it is a wonder the author didn't come up with 100% accuracy.. The results are completely useless for practical current work. However, the effort, the type of strategy employed, seems well worth improving upon by specinvestors.

More writings by Victor Niederhoffer