The arrival of Generative Artificial Intelligence (Gen AI) has taken the world by storm. The advent of Artificial intelligence (AI) and Machine Learning (ML) algorithms has had a significant impact on the financial services industry as they have the ability to revolutionize how the industry functions. Gen AI is proving immensely useful in all parts of Financial services with use cases such as AI-based Chatbots, Robotic Process Automation, Natural and Large Language Models, and Gen AI-based text inferencing. The consistent Regulatory push via Guidelines is trying to regulate this new age area to minimize the risks.
Even though regulators across the world are trying to develop more clarity on the use and governance of AI/ ML algorithms, the Model Risk Management (MRM) domain has been left more or less wanting. With still-emerging regulatory clarity, the onus lies with the Banks and Financial Institutions (FIs) to ensure proper risk management practices that can account for additional risks associated with AI and ML models. Another major disadvantage of AI/ML algo is the black-box nature of these models. Unlike traditional modeling techniques, the complex nature of AI and ML models makes it difficult to interpret results.
This article summarizes Evalueserve's capabilities with AI and ML models and how it can help Banks and FIs incorporate these techniques in their risk models. In addition, we give a domain-wise summary of AI/ML-related activities that Evalueserve has performed for its clients as part of its Risk Management practice.
AI/ML Capabilities Development - Areas of Collaboration
Credit Risk and Stress Testing
Credit Risk and Stress Testing Models Enhancements
Evalueserve has led multiple model upliftment engagements with FIs to enhance the model's performance. From IRB capital calculation to CCAR/IFRS9 models, Evalueserve has used complex AI/ML techniques like Decision Trees, Naïve Bayes, and XGBoost to improve the performance of models. To tackle the black-box nature of these models, we have leveraged advanced methods like Aletheia to interpret the model outcomes.
Although these advanced techniques improve model performance, because the regulatory landscape is not clearly defined for AI/ML models, Banks and FIs prefer using traditional modeling techniques to develop regulatory models. In such cases, Evalueserve has used AI/ML techniques at other stages of the Model Life Cycle like Feature Selection and Hyper Parameter Tuning.
Credit Analytics Framework and Model Enhancement
We have helped enhance Credit Analytics – Decision Science Models using Ensemble methods to combine several base models to produce one optimal predictive model. In addition, we have successfully leveraged NLP models/methods and Bloomberg-based ChatGPT model to develop sentiment analysis models to help optimize campaign/marketing efforts and budgets.
Market Risk
Value at Risk (VaR) Exception Classification Model
Evalueserve has also used its expertise with AI/ML techniques for multiple Market Risk related engagements. For example, we have used SVM classifiers to classify VaR exceptions into "market move" and "VaR model Issues." Parameter selection through cross-validation and our expertise in working with Market Data enabled us to achieve excellent prediction results in the model.
Proxy Selections for Bonds Spreads using ML Models
The shortage of bond liquidity impacts the calculation of pricing and risk measures such as Credit Value Adjustment (CVA) and Value at Risk (VaR). The problem has plagued Market Risk models for a long time. Using a proxy, in which the missing data points are replaced with artificial data to mimic the behavior of the original asset, can solve this limitation. Evalueserve's proposed methodology outperforms the standard proxy methods, benefiting from modeling the non-linear behavior using ML algorithms and considering the time structure included in the analysis via historical information.
Model Validation and Model Governance Framework
Model Validation Principles for AI ML Models
As mentioned earlier, the black-box nature of AI/ML models poses new challenges to the MRM function. Additionally, the lack of regulatory clarity makes it difficult to leverage these models for regulatory purposes. Though the current validation framework of FIs covers the requirements for traditional models, it may fall short in dealing with the risks associated with AI/ML models. With its domain expertise, Evalueserve has developed a validation toolkit for AI/ML models while consolidating guidelines from prominent regulators. This can help FIs enhance their current validation frameworks to address the risks associated with AI/ML models. In addition, we can assist organizations in refining their model inventory and governance structure, policies and procedures, and model validation practices to cover the complex nature of AI/ML models.
Model Governance – EUDA and AI ML Model Regulatory Guidelines Implementation
Regulators like the Prudential Regulation Authority (PRA), European Banking Authority (EBA), and the Federal Reserve (Fed) expect banks to have precise classification between Models (esp. AI/ML models) and non–models. In this regard,
- Banks are expected to identify all 'End User Defined Applications –EUDAs' and distinguish the ones with modeling components.
- Enhance the current testing framework for AI/ML models.
- Revise the Governance policies, Attestation processes, and Inventory systems to ensure compliance with these guidelines.
With the help of in-house developed smart bots, Evalueserve can help identify EUDAs. These bots can read through code snippets to enable model classification. Further, the Evalueserve inventory tool – MRMOne -has been embedded with regulatory requirements and can help institutions create an inventory and conduct attestation of the AI/ML models.
AML and KYC Analytics Support
Conducting online checks for counterparties is an integral part of an institution's AML/KYC operations. Web articles and links are searched for counterparty-specific information through manual effort. Through accelerators like AIRA, Evalueserve has helped its clients leverage GenAI and LLM techniques to automate this time and effort-intensive task.
Banking and Risk Operations
Optical Character Recognition (OCR), Chatbots and RPAs
OCR applications can extract data from scanned documents, photos, and image-only PDFs. Evalueserve's InsightFirst can convert images and scanned documents into readable text format.
Chatbots and RPAs have found utility in banking systems as well. Chatbots are being widely used to replace manual efforts in supporting customer calls and have been used to develop 24*7 customer query response systems. In addition, RPA is helping banking and risk operations handle large data and files. Advanced RPA capabilities help write access systems, write documents, and further make inferences for tests conducted for model validation exercises. Evalueserve has partnered with Automation Anywhere and developed demonstratable capabilities in MRM areas, especially for Model Monitoring.
Conclusion
In this article, we have summarised the domains where Evalueserve can collaborate with FIs and create efficiencies using AI and ML techniques. We have a proven track record of helping Banks and FIs improve their model predictive powers by replacing traditional modeling techniques with AI/ML approaches. We can help enhance the current risk frameworks to prepare organizations for risks associated with these models and get them under the umbrella of regulatory risk models. With Evalueserve accelerators like AIRA, MRMOne, and InsightsFirst, we have helped our clients in KYC/AML, Model Risk Governance, and Risk Operations. We have also partnered in the space to provide best-in-class automation solutions to our clients.
