Artificial intelligence (AI) methods, such as machine learning (ML), are rapidly transforming the field of risk management, specifically financial risk management, with AI techniques increasingly being used for the management of operational risk, credit risk, market risk, liquidity risk, and compliance risk. This article discusses the growing role and application of AI in financial risk management and recent developments in this field.
Image credit: Generated using DALL.E.3
Importance of AI in Risk Management
In financial risk management, AI solutions can significantly improve productivity and efficiency while decreasing costs as these solutions can rapidly analyze and handle substantial volumes of unstructured data with lower degrees of human intervention. AI-driven solutions can decide the amount a bank can lend to a customer, detect insider and customer fraud, provide warning signals about position risk to financial market traders, and improve compliance.
Thus, this technology can enable financial institutions, such as banks, to reduce their compliance, regulatory, and operational costs while providing them with precise credit decision-making capabilities simultaneously. AI solutions can also generate a large amount of accurate and timely data, which allows financial institutions to develop customer intelligence expertise, enabling effective implementation of strategies and reducing potential losses. Additionally, AI-powered risk management solutions can be utilized for model risk management, including model validation and back-testing, and stress testing based on the requirements of global prudential regulators.
Support vector machines (SVM), Bayesian networks, artificial neural networks (ANN), classification trees, classification and regression tree (CART), k-nearest neighbors (KNN), Lasso logistic regression, random forest (RF), logistic regression (LR), decision tree (DT), fuzzy rule-based system, deep learning (DL), deep belief network, extreme ML, fuzzy SVM, support vector regression (SVR), naïve Bayes (NB), multivariate adaptive regression splines (MARS), Lasso regression, gradient boosting, generalized extreme learning machine (GELM), and cluster analysis are the AI techniques used for financial risk management.
Benefits and Challenges of AI
Higher Forecasting Accuracy: Conventional regression models cannot effectively capture the non-linear relationships between the company financials and the macroeconomy, specifically during stressed scenarios. ML models offer better forecasting accuracy as they can identify the nonlinear effects between risk factors and scenario variables.
Improved Variable Selection Process: Variable/feature extraction processes are extremely time-consuming for risk models utilized for internal decision-making. ML algorithms augmented using big data analytics platforms can process large volumes of data and efficiently extract several variables. A rich variable/feature set with broad coverage of risk factors can result in data-driven, robust models for stress testing.
Data Segmentation with More Granularity: Proper segmentation and granularity are crucial to effectively deal with the changing composition of the portfolio. ML algorithms can be used for superior segmentation and to consider several attributes of segment data. Specifically, both density and distance-based approaches can be combined for clustering using unsupervised ML algorithms to realize greater modeling accuracy and explanatory power.
Cost and privacy are the major challenges of using AI for financial risk management. The processing of substantial amounts of data can be expensive even when cloud-native services are used for the task. Additionally, the use of specialized AI services can also lead to significant expenses.
Privacy is another key challenge due to growing concerns about data privacy with the increasing proliferation of AI solutions. Financial institutions must implement data protection controls, such as obfuscation, tokenization, transport security, and encryption, to alleviate data privacy concerns.
Major AI Applications
Credit Risk Management: Credit risk assessment primarily involves applying a classification technique on past data of customers, including on delinquent customers, to evaluate and analyze the relation between a customer’s characteristics and their potential failure.
Credit risk evaluation occupies an important place within risk management. Techniques such as logistic regression and discriminant analysis are traditionally used in credit scoring to determine the probability of default, while SVM can be utilized to successfully classify credit card customers who can default and to identify features that are crucial for determining the default risk.
Neural networks can be employed in the credit risk decision process and company distress predictions. Studies displayed that a multistage deep belief network-based extreme ML can be utilized effectively for credit risk assessment. The multistage ensemble learning paradigm framework working at three stages can outperform multistage ensemble learning paradigms and single classification techniques with high prediction accuracy.
Similarly, a bilateral weighted fuzzy SVM and a hybrid SVM can be used to analyze and evaluate an applicant’s credit score from input features, respectively. A combined method of partial least squares (PLS)-based feature selection with SVM for information fusion can be employed to predict bankruptcy.
Market Risk Management: In financial markets, volatility forecasting is crucial for asset pricing and risk management. Neural network models can be used to improve the performance of the volatility estimation method.
A model based on the generalized autoregressive conditional heteroskedastic (GARCH) model and extreme ML algorithm can be utilized for volatility estimation. The model can predict the target time series volatility using the GELM-radial basis function (RBF) and extrapolating the predicted volatilities, which enables the calculation of value at risk (VAR) with improved efficiency and accuracy.
ML clustering methods designed to address the stochastic differential equation (SDE) can be applied for developing anticipatable VAR models as a leading risk measure of market regime change to partially address some complexity introduced by the challenging regulatory environment.
Liquidity Risk Management: Several liquidity risk problems can be addressed using ML techniques. For instance, ANN can determine the most influential factors and approximate the general risk trends. Bayesian networks can be employed to estimate the probability of liquidity risk, while BN and ANN implementations can distinguish the most critical liquidity risk factors by measuring the risk using a distributional estimation and a functional approximation, respectively.
Operational Risk Management: ML can also be applied to the operational areas to mitigate, detect, and prevent risks. ML algorithms can be used for suspicious transaction detection and fraud detection. For instance, a model using a logistical regression algorithm can generate a report that enables suspicious transaction detection. Similarly, clustering algorithms can identify customers with identical behavioral patterns and groups of people planning to commit money laundering.
Bayesian algorithms bagging ensemble classifiers based on DT, SVM, and KNN have been used in fraud detection systems. Among them, the bagging ensemble classifier based on DT algorithms is more suitable as it is unaffected by attribute values and can handle class imbalance effectively.
Recent Studies
Black box AI is unsuitable for regulated financial services, which is a major challenge in implementing such methods for financial risk management. Explainable AI models can be used to overcome this problem as these models provide reasons or details to make the functioning of AI easily understandable.
In a study recently published in the journal Computational Economics, researchers proposed an explainable AI model that can be employed in credit risk management, specifically for measuring the risks that emerge when credit is borrowed using peer-to-peer lending platforms.
The model applied correlation networks to Shapley values to ensure that the AI predictions were grouped based on the similarity in the underlying explanations. Thus, the proposed model can explain any prediction in terms of the Shapley value contribution of every explanatory variable.
Researchers used TreeSHAP, an accurate and consistent method available in open-source packages, to implement their model. TreeSHAP is a fast algorithm that can compute Shapley's additive explanation for trees in polynomial time in place of classical exponential runtime.
The proposed model was evaluated using data from the European External Credit Assessment Institution (ECAI), which specializes in credit scoring for peer-to-peer platforms focused on small and medium enterprise (SME) commercial lending. The empirical analysis of 15,045 SMEs asking for credit revealed that both non-risky and risky borrowers could be effectively grouped based on a set of similar financial characteristics, which can be used to explain their credit scores and predict their future behavior.
Thus, the research findings demonstrated that network-based explainable AI models can successfully advance the understanding of the determinants of financial risks, specifically credit risks.
References and Further Readings
Bussmann, N., Giudici, P., Marinelli, D., Papenbrock, J. (2021). Explainable Machine Learning in Credit Risk Management. Computational Economics, 57, 203–216. https://doi.org/10.1007/s10614-020-10042-0
Leo, M., Sharma, S., Maddulety, K. (2019). Machine Learning in Banking Risk Management: A Literature Review. Risks, 7(1), 29. https://doi.org/10.3390/risks7010029
Aziz, S., Dowling, M. M. (2018). AI and machine learning for risk management. Disrupting Finance: FinTech and Strategy in the 21st Century, 33-50. https://doi.org/10.1007/978-3-030-02330-0_3
Artificial Intelligence in Risk Management [Online] Available at https://kpmg.com/ae/en/home/insights/2021/09/artificial-intelligence-in-risk-management.html (Accessed on 29 October 2023)