In 2009, it was not so good times for Satyam Computer Services after being brought to its knees by its founders due to financial fraud. The founders admitted to the fraud amounting to more than INR 7,000 crores by falsely inflating the company’s revenues and profitability year after year, thereby artificially boosting share prices. The Satyam scam not only shook the Indian market to its core but also harmed the country’s reputation in the global market.
Credit risk is not a new term. There have been numerous cases of financial wrongdoings in the past. Over the years, financial institutions have recognised the importance of managing credit risk effectively and underscored the potential consequences of failing to do so.
Traditional models of assessing the credit risk of a company involve quantitative analysis and are primarily based on financial numbers. While financial statements provide a meaningful picture of past performance of a company, their utility in providing reliable insights into the future of the company is still far from satisfactory. Today, we live in a world full of technological disruptions which can be instantaneously catastrophic to a particular company as well as an entire industry, even if the latest financial statements paint a healthy picture of the company or sector. Further, traditional credit risk analytical tools might not be able to generate signals related to credit risk for a company as good numbers are always a target for the management.
Number Dressing: Manipulation of Accounts to Paint a Rosy Picture of the Company
Manipulation of accounts, or number dressing, to hide fraudulent activities and portray a company in good stead in order to inflate the share price is a practice that continues to plague capital markets the world over. High-profile cases of fraud and poor financial reporting, such as the aforementioned Satyam and Enron, one of the biggest accounting frauds in history, immediately come to mind. In both cases, crores and crores of investor wealth was wiped off within days as their management routinely falsified accounts for personal gains and to hide bad debts. The reasons for these disasters were manifold, including regulatory oversight and audit failure. However, one can argue that the lack of available tools to predict these frauds at a nascent stage was among the critical reasons for them snowballing into a full-blown crisis.
Credit analysis is traditionally a number crunching game, which companies can expertly manoeuvre to fool even the best of experts. This calls for a comprehensive overview of current credit management practices and a methodology to free them from the overdependence on financial numbers. Thankfully, authorities now have a new weapon in their arsenal to combat accounting fraud: Artificial Intelligence. Financial datasets and models that are appropriate for the situation can help identify and rationally explain fraudulent account behaviour. Systems that use AI to detect fraud can cut down on costs while reducing fraud by 95%. For governmental organisations and regulatory bodies, these findings are highly valuable.
Textual Analytics: Zero Reliance on Numbers
The breakdown of both Enron and Satyam raises some serious questions. Why the sudden meltdown? Did the financial numbers published by the companies preceding the crisis not raise any red flags? Would the reading of management reports and filings generate enough warning signals? All said and done, Enron’s debt was rated as investment grade right before the fraud came to light!
Qualitative signals are more relevant now than before due to increasing complexity of accounting policies, excessive liquidity in markets driven by easy money policies, and a turbulent global economy. An analysis of textual information present in the management discussions filed by companies provides much better insights and is free from manipulation. Management discussions offer valuable insights into a company’s operations, the challenges they face and also their plans for the future. Analysing these discussions through the lens of Machine Learning algorithms can help in identifying signs of potential wrongdoings which carefully doctored financial numbers may hide.
Financial numbers cannot be trusted entirely as they can be easily manipulated by the management to boost stock prices and lure investors. There is an urgent need for a tool that completely ignores financial data and focuses solely on text information. This tool needs to be independent of any assumptions regarding capital or market structure, be impervious to manipulation of financial numbers, with the only focus on analysis of text, thereby resulting in objective output.
One example of such a tool is CreditPulse, the foundation of the tool is the idea that textual analysis provides more insightful information than number crunching. The tool analyses textual information from regulatory filings to detect the bankruptcy risk of a company using Machine Learning techniques.
Tools which have the capability to read textual information use natural language processing (NLP) techniques to analyse data such as financial reports, news articles, social media posts, and other unstructured data sources, to identify potential risks and opportunities. These tools have the potential to be a market disruptor in the credit ratings space and complement current credit scoring methodologies which are based on financial data.
