As per recent research, one of the major reasons for increasing credit default is the inability of the banks to predict the emerging stress in accounts that are likely to default. In several countries, regulations have progressed from rule-based surveillance to an early warning detection method over the past two decades (from income recognition and asset classification (IRAC) requirements in the 1990s to surveillance of complex signals) to addressing the problem of unmanageable NPA (non-performing assets). This has led stakeholders to doubt how banks are tracking loan accounts recovery. Over the last few years, the NPA rate has shot-up in many countries, especially the developing and underdeveloped regions. There is an urgent need to reconsider existing practices and rework them according to internationally agreed and innovative approaches according to market participants. Globally, banks need to switch from an increasingly compliance-driven post-facto process to a proactive digital framework focused on controls to track credit risk.
One of the trending approach major tech-oriented lenders doing is tracking publicly accessible news on borrower- and industry-specific results on a regular basis through an AI-based early warning system. Analyzing this information using tools like – Natural Language Processing (NLP), Predictive Analytics, and Clustering helps the lenders in tracking the loan portfolio. Based on the insights generated, loan accounts may be divided into three or five groups of the watch list. Borrowers in the amber area should be closely monitored on the basis of a three-color scale (green, amber, and red) (here we believe customers in the red group have already turned bad, and giving them a digital early warning does not generate any meaningful value). Additionally, it should be noted that because credit quality-related information takes some time to affect the applicant, the system may be configured to operate through batch processes, perhaps daily or weekly, rather than doing it on an absolute real-time basis. The initial investments can be very large for making things operate on an absolute real-time basis.
Potential benefits of tracking and credit risk management using AI-based EWS system
- Default rate reduction at portfolio level: Loan default rate at portfolio level can be decreased by restrictive exposure to unique customer segments where warning signs have been observed. This will assist in managing default probability.
- Covenants tightening: Lenders need to intervene early in order to optimize recovery by tightening the covenants and increasing collateral levels. Upon entry into the watch list of a loan account, daily monitoring can help optimize collateral-related criteria and minimize losses in the event of actual default. This minimizes the loss given default.
- Securitization of assets: Lenders must attempt to proactively securitize or sell assets that are correlated with warning signals and where there are low chances of recovery. Any asset-reconstruction companies (ARCs) may be interested in taking over such assets. This would help businesses get bad loans off their books and get them a better price.
- Identification of business patterns and opportunities: Emerging developments in various markets can be detected, and banks can change their portfolio mix. It would have helped banks, for example, to recognize the telecoms sector as a whole, turning adverse and steadily reducing their exposure to it.
- Decisions on portfolio composition: Portfolio-level monitoring can help reduce the exposure of banks to threatened sectors and even reduce exposure at individual levels by decreasing committed credit lines, i.e. the credit conversion factor.
- Modification of banks’ credit policies: Banks can take high-level credit policy-related decisions based on system information on sector regulations, results, and news.
Our approach for developing an EWS
Developing EWS involved the integration of a wide range of variables that may affect borrowers’ creditworthiness. All the data points may not be accessible at a single source and may vary in their structure as well. The first development process is, therefore, to incorporate all possible data-points for modelling purposes. Post integration, calibration of the algorithms is performed to measure the risk scoring. Risk scoring is a dynamic procedure, and its performance depends on how weight variables adjust for different portfolios of loans. The final solution will be to fill outcomes to help end-users take remediation steps.
Four steps to building a robust EWS model
To ensure the developed EWS is robust to give relevant output, data from both traditional and non-traditional sources must be incorporated. AI-based predictive and cluster-based models are used to process the interconnected data points in order to understand what is and what is not important. The model is used to train the past default database. Training is an important factor, as the portfolio of loans varies across geography and sectors. It is also not one size that fits all strategy. Once the model is trained, parameters for the risk scoring are specified, and plans for risk remediation are formulated.
Data Integration is the backbone of a successful EWS model
Multiple data sources are combined as the first step in developing the EWS platform. A regulatory EWS recognizes borrowers who are at risk of hardship or default. Such frameworks should not be limited to a comprehensive database and accurate statistics; it must also ensure that it incorporates “qualitative” variables. A summary of how various data points provide valuable insight.
- Borrowers’ financial information: Unaudited interim figures, creditor feedback and bank statements may provide financial information. For example, tie-ups can be made to track bank statements, and input from creditors can be gathered via e-mails and automation based on VBA.
- Market information: Data sources are required for the movement of equity prices, yield curves, CP spreads, ratings and other publications from credit bureaux. So it can be beneficial to bring AI or ML algorithms together. Market data sources such as Bloomberg and Thomson Reuters may be used to set up automated tests. Grade publications may be gathered annually for data from credit bureau reports and updated throughout the framework.
- Business news: Vernacular news articles are desirable, along with negative press conferences or reports and news from some other external source. Banks need to build technologies that can be achieved with the aid of Natural Language Processing (NLP) based on ‘Text Mining’ and ‘Sentiment Analysis.’
- External macro scenario: systems should be able to accommodate information on adverse regulatory guidelines or notices as the decisions of the government can adversely affect the company.
Chart II: Detailed view of the critical data points to be included to make the EWS robust
Some illustrative of what should be tracked are :
- What credit rating agencies think about banks, some sectors or the economy
- Loan portfolios of banks across the world, across products and across sectors of industry/customers
- Forums and social media talk about a borrower’s promoters
- Borrowers’ indiscipline of what is published in semi-public sources
- Review of payment transactions by borrowers, including number and type of receivers, regular account opening practices, and so on
- Operations of Business Managers inside and outside the Bank
Below table (Chart III), is a very clear illustration of how various data points may produce useful insights into the borrower in order to take precautionary steps. The idea is to become pro-active with EWS, rather than respond to potential credit defaults. As we can see in the table below, even when designing an emerging EWS system, qualitative factors such as consumer sentiment on the borrower, behavioral and macro-level indicators are important.
Chart III:Illustrative decision output against inputs from various data sources
Source: Cognizant 20-20 Insights
Key factors for a successful EWS development
We focus on the broader picture and not limit our scope only on the development by making sure the final product has integrated work-flows, information system and risk visualization dashboard to enable escalation and review at different levels. We tailor outputs/ insights based on internal processes to help clients get the most out of the developed platform. To ensure internal nuances are identified and managed appropriately, we loop internal team members while finalizing technology architecture and models to be used and schedule frequent meetings/demos with internal IT teams. If needed, we also onboard industry experts when identifying business logic for risk categorization and triggering.
Given that government and regulatory bodies have already forced banks to bring down NPA levels, their lifesaver can be an all-inclusive NPA management solution that has Early Warning Signals. Not only does it help to deal with current NPAs, but it also prevents potential loans from being delinquent by implementing various precautionary approaches.
Model Architecture of EWS
Source: Cognizant 20-20 Insights
The proposed implementation of automated early warning signal monitoring has the potential to improve the consistency, timeliness and bad account detection significantly. To exploit this opportunity, banks need to move quickly to develop the infrastructure, but in a cautious manner. Considering the business volume and complexities, implementation of EWS requires great rigour and time.
Drop a comment or write us an email with any feedback about this article, queries firstname.lastname@example.org. View our portfolio in building Robo advisory, Large Scale Trading Systems, Algo Trading, Stock Sentiments, Price Trends forecasting, Backtesting frameworks, Credit Model, Open Banking, etc. and services offered on our website. to see our Fintech expertise can help you build cutting-edge solutions powered by AI/ML.