We have developed a proprietary machine learning algorithm which analyses textual data. Credit Pulse, which is developed using this algorithm, aspires to provide a qualitative edge to how credit analysis is done. Traditionally credit analysis has been a numbers game; leading to a cat and mouse game between ever more complex accounting policies and analyst creating ways to decipher latest jugglery. But as Goodhart’s law says that - “when a measure becomes a target it ceases to be a good measure”.
When we use traditional credit risk analytical tools that relies completely on numbers there are possibilities of being blindsided to signals which might be indicative of adverse news being unaccounted for in the numbers due to their being a target to be achieved by the management. The acknowledged leverage cycle also teaches us that underwriting standards also weaken according to different phases in the business cycle
The role of market intelligence should be to anticipate surprise, the surprise manifests in form of lower credit quality of company or bankruptcy which can only be derived by backward induction of the communications of the company to the investors in forms other than numbers.
As the 2008 financial crisis has demonstrated that credit risk related metrics relying on only numbers can be deceiving, it is imperative we have an additional layer of analytics that give us signals. These signals can alert us to several aspects of credit quality we might be overlooking using our traditional quantitative models and force us to take a closer look which might reveal inconsistencies.
Our algorithm analyses the text of latest SEC filings and classifies companies as either 'High Risk' or 'Low Risk' for the upcoming quarters. This information is immensely useful to all asset managers as we believe that the textual analysis of management discussions filed by companies can give better insights into predicting their immediate future. As the cognitive aspects of language can reveal to us presence of issues and can help us determine lowering credit quality standards. Credit Pulse has studied from data of 10 years of management discussions filed by companies.
This has allowed it to develop a high level of analytical proficiency, with a current accuracy of 90% and is constantly improving. Our test portfolio provides an annualized return of 30% compared to the 4.6% benchmark.
How is this delivered?
- Accurate Output is available within just 1 minute of a SEC filing, providing valuable lead-time
- Other options are – once a day or once a week
- Universe could be index stocks e.g. S&P 500 or any other user defined universe
The subscription is accessed online through a secure website. You will receive a username and password via email to access your subscription. The subscription period is one year.
As per an article published in Financial Times in April 2016, “Corporate borrowers across the world have defaulted on $50bn of debt so far this year as the number of delinquent companies accelerates at its fastest pace since the US emerged from the financial crisis in 2009”. This clearly signals widespread increase in credit risks in the market. The accounting numbers and traditional analysis thereof put out are proving to be insufficient to generate accurate evaluation of credit risks faced by these companies.
Easy money policies have led to excess liquidity in the market thereby limiting the discipline of the markets, which can produce volatility and lead to situation where traditional metrics many not be sufficient to evaluate the risks.
With the prediction that the world economy will continue to be volatile in coming years producing more bankruptcies, investors need an additional layer of protection around their portfolio to shield themselves from volatility due to numbers not representing the underlying value. Surprise events are detected in form of signals which is detected by qualitative analysis of communications to the market by the company.
Credit Pulse is designed using proprietary algorithm on textual analysis developed by us using machine learning technology. The model uses information from each issuer company's statutory filings to gauge the credit risk and thereby points out the credit quality of the company in question using qualitative data. The final analysis categorizes companies into either 'High Risk' or 'Low Risk' for the upcoming two to three quarters, with significant confidence level. So, Credit Pulse can provide you with all the required insights right-on-time to sustain and augment your investment strategy.
The model is tested using 10 years of real market data (20,000 companies, 10 years of 10K/10Q filings and their credit performance). We have carried out extensive back testing using out-of-sample data.
- The Model identifies the credit risk of a company by putting out a risk score between zero and one classifying companies into high risk and low risk categories. The credit risk of industries as a result can be monitored.
- Companies that have a higher credit risk and lower risk. The information is useful to bond traders, especially those who deal in high yield debt. It can also be useful to hedge funds in creating a short equity portfolio or in diversifying an existing portfolio.
- Companies in our high-risk category have seen consistent median price decrease since the release of our score. Having divided our low risk categories into further two categories, companies with credit score ranging from 0-0.1 and 0.1-0.6 we observe that information for companies in the first case tends to be over discounted. While in the latter category we see median price increases which points to the information availability and its ambiguity regarding evaluation of credit risk of these companies by the market.