Credit Rating Prediction Using Financial And Sentiment Data
The credit rating of financial instruments is one of the factors that play a significant role while making investment decisions. Credit rating companies provide information about the credit stability/ transitions (deterioration/ up gradation) of the debt instruments using financial parameters, industry parameters, and external environment at quarterly intervals. In the present scenario, investment managers have to wait for quite some time to know about the credit rating stability/ transition. We propose that incorporating the market news available from different sources in almost real time along with the usual financial parameters should be able to predict the rating stability/ transition well in advance. The present work focus on incorporating sentiment scores extracted from various news sources like new aggregators like Bloomberg, Reuters, Company Web sites, Blogs etc. Depending upon the source of news a weight will be given taking into consideration trust worthiness of the source, Location of the Company information in the body of the news, Frequency of appearance. Using deep learning weights are assigned to each source. A Composite Sentiment score is computed as weighed average of the various individual news sources. This composite score is additional Input along with other financial parameters about eleven to survival analysis which in turn will forecast credit events. Our results show that Return on Capital Employed ROCE and Interest cover ratio along with sentiment scores are the best predictor variables to predict the rating stability/ transition. We tested our methodology on about three hundred companies. Most of them are from mining & textile segment as majority of downgrades and rating transitions have happened in those sectors. Back testing results show that proposed model very well predicts the credit rating stability/ transition. Without using the sentiment score and our weighing scheme the prediction accuracy of a transition was about 45% on hold out Samples (About 30 of 300) and with Financial and sentiment score it went up to about 70%. This work will be very helpful from the regulatory point of view also as one has to compute the Future value of Loan portfolio ie Credit Risk VAR. This method computes the Credit Risk VAR much accurately compared to existing methods of Credit Risk Computation.
Outline/Structure of the Case Study
PPT and Demo of KNIME Work Flow
Learn how sentiment and text mining can be used along with conventional models to predict the rating transition of companies and take corrective action well in advance.
At present there is a long delay between rating upgrade/down grade as many technique use only financial data.
Will help to manage the Credit Risk efficiently.
KNIME is used for this case
Any one interested to know the application of Text Mining for a Financial application Pirticulary for Portfolio Managers in Banks Mutual Funds Rating Analysts Investment Bankers
Prerequisites for Attendees
Some amount of brief about Text Mining though not essential