Increases accuracy in predicting which firms most/least likely to come under financial distress
Thomson Reuters, the world’s leading source of intelligent information for businesses and professionals, today announced that it has introduced the first commercial credit risk model to measure corporate financial health by quantitatively analyzing text. The StarMine Text Mining Credit Risk Model (TMCR) assesses the credit risk of publicly traded companies by analyzing the text in various documents. StarMine TMCR systematically assesses the language in Reuters news, StreetEvents conference call transcripts, corporate filings and select broker research reports to predict which firms are likely to come under financial distress and which are likely to thrive.
The model is a complement to StarMine’s other two credit risk models, the StarMine Structural Credit Risk Model and the StarMine SmartRatios Credit Risk Model. StarMine TMCR builds on Thomson Reuters aims to help professional investment managers generate more profitable investment ideas and mitigate risks, giving them an edge over their peers.
“Mining text is the new frontier in quantitative analysis,” said Ryan Roser, principal quantitative research analyst at Thomson Reuters and the chief architect of StarMine TMCR. “What we have created is a very powerful and flexible framework for analyzing and making predictions from text. We have successfully applied the model to default prediction and foresee applying it in other applications as well.”
In creating the StarMine TMCR, the StarMine quant group developed a new specialized dictionary of credit and financial-health related language to enable robust predictions of credit quality beyond what is possible when using a standard dictionary in a classic “bag-of-words” text mining framework. The StarMine TMCR provides transparency into the origin of its final scores, ratings and default probabilities by creating overall component scores for each class of document it processes: news, transcripts, filings, and select broker research. The StarMine TMCR also provides insight into the types of language the model picks up on, categorizing the language into four groups – income statement related, balance sheet and debt structure related, legal obligations and terms, and external & market events – and producing scores for each of the four categories. In addition, the StarMine TMCR scores each individual document for a company, allowing analysts to quickly zero in on the documents that matter.
“The Text Mining model is probably the most innovative model our StarMine team has ever created,” said Dr. George Bonne, director of quantitative research at Thomson Reuters. “Not only is StarMine TMCR powerful in predicting default and generating alpha, it will also increase the efficiency of analysts and financial professionals by helping to identify the documents that matter most out of the potentially hundreds or thousands of pages they may be responsible for reading every day.”
The model produces daily updated estimates of the probability of default or bankruptcy within one year for over 20,000 companies globally, including financials. The default probabilities are also mapped to traditional letter ratings and ranked to produce 1-100 percentile scores (with 100 being comparable to an AAA rating). StarMine TMCR is provided as a daily data feed, as well as through a range of Thomson Reuters desktops.