An Artificial Neural Network and Bayesian Network model for liquidity risk assessment in banking
Introduction
Banks are subject to many different potential risks that range from those related to the technological and financial structure, affecting also their reputation, to those derived from the institutional and social environment. These risks are not mutually exclusive and have some intersections that make them hard to isolate and identify.
Liquidity risk, together with credit risk, operational risk and market risk, is categorized as a financial risk. However, a full consensus on the definition of liquidity risk is still to be reached mostly due to its ambiguity and vagueness. The ambiguity of the term liquidity risk follows from the multiple probable meanings that it can be given according to the context; the vagueness is given by the fact that the term “liquidity” can refer to different dimensions at the same time especially when used together with market liquidity risk or systemic liquidity risk (SLR) [78].
There are diverse viewpoints on what the definition of liquidity risk should be, all of them referring mainly to whether or not liquidity risk considers (1) solvency, (2) cost of obtaining liquidity or (3) immediacy [98]. For example, liquidity risk could be interpreted as the “capability to turn an asset quickly without capital loss or interest penalty”, or as the risk of being unable to raise funds on the wholesale financial market [98]. In this paper, we follow the first of these two approaches, that is, we assume that liquidity risk arises because revenues and outlays are not synchronized [51].
The commitments of banks to shareholders to maximize the profits lead to a development in the volume of investments, while the commitments to depositors to refund make necessary to retain adequate liquidity especially considering depositors’ stochastic behavior. Such a conflict between shareholders and depositors impels the bank directors to make a balance between profitability by long term investment and risk due to short term commitments. Liquidity management and surveillance of maturity mismatch of deposits and loans can be considered the main concerns of bank managers. Management's task becomes even more critical when the bank faces early withdrawals. The reason of this challenge is that short term deposits are the main funding resources for banks. In addition, loans are usually invested in weak liquidation assets. Too much liquidity causes an inefficient allocation of resources, while low liquidity can lead to a reduction in the deposits interest rate, a loss of market and credit, an increase of debt and, finally, to the bank's failure. In other words, insufficient liquidity can kill the bank suddenly, but too much liquidity will kill it slowly [75]. Thus, it is extremely important to handle liquidity risk prudently and evaluate it correctly by an efficient and systematic method.
Liquidity risk relates to a complex set of factors such as significant operational risk loss, deteriorating credit quality, overreliance on short-term borrowing, overreliance on borrowing from very confidence sensitive funds providers, market risk and so on. Also, banks that are part of financial groups or bank holding companies need to identify key risk indicators that are indicative of the group's risk and reputation [75]. Each bank has to select a set of indicators that is most relevant to its funding situation and strategies (bank specific indicators). In particular, a bank primarily funded by insured deposits has far less need for a risk indicator of liability diversification than a wholesale funded bank [75]. In addition, liquidity risk may be affected by global factors usually described via macroeconomic variables.
The standard framework to measure liquidity risk compares expected cumulative cash shortfalls over a particular time horizon against the stock of available funding sources [44]. This requires assigning cash-flows to future periods for financial products with uncertain cash-flow timing. However, on one side, there still is a lack of consensus on how to assign such cash-flows [98]. On the other side, plurality, multiplicity and diversity of accounts make the calculation of net cash-flows so difficult and time consuming that accessing such data in a short period of time is impossible.
The minimum liquidity standards under Basel III [13], [14], [15], [16], [17] are based on two complementary ratios: liquidity coverage ratio (LCR) and net stable funding ratio (NSFR). Although these ratios reflect the concept of liquidity risk correctly, implementing them in a banking system is not practical. In fact, both the numerators and denominators of these ratios include some weights related to inflows and outflows that must be conveniently estimated (and sometimes manually adjusted). The complexity of calculations of these coefficients together with the problem of an actual classification for the concept of “stable assets”, make LCR and NSFR useless for many practical purposes. Moreover, banks do not usually make available their information/datasets to external researchers.
The main goal of the current study is the design of a simple, practical, easy to control and analyzable system capable of warning about probable liquidity risk based only on raw data available in the book or balance sheet of the bank without any predefined function.
Nowadays machine learning methods can solve problems like this quite easily and applications of these methods to large databases, data mining, can lead to accurate results. Fortunately, banks are a very rich source of historical data. Thus, we can implement these techniques for measuring a bank liquidity risk and analyzing its key factors and the interconnections among them. More precisely, Artificial Neural Networks (ANNs) and genetic algorithm can be used for an approximate measure of liquidity risk and Bayesian Networks (BNs) to estimate and analyze the distribution function of liquidity risk.
Despite the capacity of machine learning methods to model real situations where future results must be predicted starting on imprecise or missing data, their applications to bank liquidity risk measurement remain very sporadic in the literature.
Liquidity scenarios are modeled differently depending on the fact that they use bank-specific factors or market-specific factors. In this study, we focus on the definition of liquidity risk determined by the concept of solvency. As a consequence, we focus on endogenous factors to construct a model whose characteristics will allow us to specifically address loan-based liquidity risk prediction issues.
The proposed model is flexible and can be applied to any loan-based scenario. However, its main purpose is to promote a systematic analysis of bank specific measurements based on the balance sheet ratios. The choice of using the balance sheet data is justified by the fact that the balance sheets are the most accessible, reliable and official reports that any bank is obliged to compile and safely retain.
The current model uses ANNs and BNs to analyze and assess liquidity risk and its key factors. The resulting assessment method comprises the use of several genetic algorithms and numerous tests to train a suitable ANN and learn the optimal BN to analyze the data.
The ANN and BN approaches represent two complementary phases: while ANN is used to approximate the general trend of the risk and find the two most influential factors in a non-efficient way, BN finds the most influential factor and determines the probability that liquidity risk occurs even in situations where it is not possible to measure all the indicators. The liquidity risk results obtained by ANN complement and are complemented by those obtained by BN. Since, the data implemented in both phases are the same, the numerical results can be used to confirm one another.
A case study based on a real bank dataset is performed to show the validity of the proposed assessment method.
The numerical results of the case study show that loan-based factors are inevitable given their key role in the model. Both the case study and the dataset were carefully chosen to reflect the solvency-based definition of liquidity risk. Some complementary factors (see also Section 2) can be added like credit rating, downgrade, significant operational loss and so forth, but there may be no enough data available to use these factors in practice.
The loan-based constraint imposed by the definition adopted for liquidity risk represents a limitation of the model which should be compensated by its applicability to an already large number of banks (all those whose main funding strategy consists of loans and deposits) and the efficient implementation of data derived from the balance sheet ratios into a two-phase ANN-BN intelligent schema whose results complement and relatively confirm one another.
The remainder of the paper proceeds as follows. Section 2 reviews some of the most recent and relevant liquidity risk assessment methods in the literature. Section 3 provides a description of the problem, including main goals and model variables. Section 4 presents the proposed model including a brief theoretical overview of the main general features of ANNs and BNs. Section 5 presents the numerical results obtained by implementing the model at a U.S. bank to demonstrate applicability and efficacy of the proposed method. In Section 6, we present our conclusions and future research directions.
Section snippets
Literature review
Different definitions of liquidity risk lead to different risk measurements. Conceptually, this risk is related to the mismatch of cash inflows and outflows and unfortunately a significant portion of bank financial products have uncertain cash-flow timing [56]. Thus, to measure bank liquidity risk, one idea is to assign uncertain cash-flows to future periods by different methods like surviving models and lifetime models [78].
Assessing liquidity risk is scenario specific [14]. The occurrence of
Research questions
Imperfect information and shortage of knowledge about a phenomenon makes it vague and uncertain. Thus, unavoidably, most crucial decisions must be made under nondeterministic conditions while any negligence may lead to irrecoverable ending. Hence, the following natural question arises: how can we make rational decisions to minimize expected risk? This is the most important question for bank managers and also the main motivation for this study.
We aim at defining an analytical but practically
Introducing the two-phase ANN-BN approach
The proposed model comprises two complementary phases, an ANN phase and a BN phase.
These two phases deal with the two different tasks described by Questions 1 and 2 in Section 3.1. More precisely, while an ANN can be used to approximate the general trend of the risk and find the two most influential factors in a non-efficient way, a BN is capable to find the most influential factor and confirm the trend of ANN. Moreover, as it will be shown later, the BN approach allows to determine the
Case study: implementation of the proposed method
In this section, we show the results obtained by applying the proposed liquidity risk measurement method to a set of real data provided by a large U.S. bank focusing mainly on loans.
The collected dataset refers to a period of almost eight consecutive years, from 2005 to 2011 plus a couple of months of 2004, and were extracted from monthly reports. All ratios (i.e., our input and output variables) were already normalized but had to be increased in number via a standard averaging technique. The
Conclusion
This study has addressed the problem of defining an efficient and systematic method to prudently handle and correctly evaluate bank liquidity risk.
The ambiguity and vagueness that characterize the liquidity risk concept make difficult the formulation of an unquestionable definition for it. Identifying the factors that determine and/or influence liquidity risk and formulating an appropriate functional form to approximate and predict its value represent an even more involved task. At the same
Acknowledgment
The authors would like to thank the anonymous reviewers and the editor for their insightful comments and suggestions.
Madjid Tavana is Professor and Distinguished Chair of Business Analytics at La Salle University, where he serves as Chairman of the Business Systems and Analytics Department. He also holds an Honorary Professorship in Business Information Systems at the University of Paderborn in Germany. Dr. Tavana is Distinguished Research Fellow at the Kennedy Space Center, the Johnson Space Center, the Naval Research Laboratory at Stennis Space Center, and the Air Force Research Laboratory. He was recently
References (107)
- et al.
Artificial neural network development by means of a novel combination of grammatical evolution and genetic algorithm
Eng. Appl. Artif. Intell.
(2015) - et al.
Asset commonality, debt maturity and systemic risk
J. Financ. Econ.
(2012) - et al.
Multi-objective optimization for building retrofit: a model using genetic algorithm and artificial neural network and an application
Energy Build.
(2014) - et al.
Bank-specific, industry-specific and macroeconomic determinants of bank profitability
J. Int. Financ. Mark. Inst. Money
(2008) - et al.
The asymptotic number of acyclic digraphs, II
J. Comb. Theory Ser. B
(1988) - et al.
A survey on feature selection methods
Comput. Electr. Eng.
(2014) - et al.
Improving generalization of MLPs with sliding mode control and the Levenberg–Marquardt algorithm
Neurocomputing
(2007) - et al.
Periodic attractor for reaction–diffusion high-order Hopfield neural networks with time-varying delays
Comput. Math. Appl.
(2017) - et al.
Hybrid control on bifurcation for a delayed fractional gene regulatory network
Chaos Solitons Fractals
(2016) - et al.
Bifurcations in a delayed fractional complex-valued neural network
Appl. Math. Comput.
(2017)
The arbitrage-free valuation and hedging of demand deposits and credit card loans
J. Bank. Financ.
Measuring systemic risk-adjusted liquidity (SRL): a model approach
J. Bank. Financ.
Levenberg–Marquardt methods with strong local convergence properties for solving nonlinear equations with convex constraints
J. Comput. Appl. Math.
Performance of the Levenberg–Marquardt neural network training method in electronic nose applications
Sens. Actuators B
An improved method for the design of quadrature mirror filter banks using the Levenberg–Marquardt optimization
Signal Image Video Process.
Automatic and quantitative evaluation of attribute discovery methods
A survey of deep neural network architectures and their applications
Neurocomputing
Some research on Levenberg–Marquardt method for the nonlinear equations
Appl. Math. Comput.
Improved reliability modeling using bayesian network and dynamic discretization
Reliab. Eng. Syst.
Neural network modeling of energy use and greenhouse gas emissions of watermelon production systems
J. Saudi Soc. Agric. Sci.
Factors Influencing the profitability of domestic and foreign commercial banks in the European Union
Res. Int. Bus. Financ.
A theory of inferred causation
Stud. Logic Found. Math.
Neural networks: an overview of early research, current frameworks and new challenges
Neurocomputing
An improved algorithm for building self-organizing feedforward neural networks
Neurocomputing
Deep learning in neural networks: an overview
Neural Netw.
Chaos time-series prediction based on an improved recursive Levenberg–Marquardt algorithm
Chaos Solitons Fractals
Finite-time synchronization of fractional-order memristor-based neural networks with time delays
Neural Netw.
Bayesian networks as a decision support tool in credit scoring domain
Numerical solution of systems of second-order boundary value problems using continuous genetic algorithm
Inf. Sci.
Application of continuous genetic algorithm for nonlinear system of second-order boundary value Problems
Appl. Math. Inf. Sci.
Liquidity risk and islamic banks: evidence from Pakistan
Interdiscip. J. Res. Bus.
Transformation Risk and its Determinants: A New Approach Based on Basel III Liquidity Management Framework
Liquidity risk and performance in the banking system
J. Financ. Regul. Compliance
Bankruptcy prediction for credit risk using neural networks: a survey and new results
IEEE Trans. Neural Netw.
A cross-country analysis of the bank supervisory framework and bank performance
Financ. Mark. Inst. Instrum.s
Principles for Sound Liquidity Risk Management and Supervision
Basel III: International Framework for Liquidity Risk Measurement, Standards and Monitoring
Revisions to the Basel II Market Risk Framework
Basel III Liquidity Standard and Strategy for Assessing Implementation of Standards Endorsed by Group of Governors and Heads of Supervision
Basel III: The Liquidity Coverage Ratio and Liquidity Risk Monitoring Tools
Pattern Recognition and Machine Learning
An analysis of the determinants of bank rating: comparison across rating agencies
Aust. J. Manag.
Liquidity Risk in Banking: Is there Herding?
Bayesian Belief NetworksFrom Construction to Evidence
Concentration and other determinants of bank profitability in Europe, North America and Australia
J. Bank. Financ.
Deciphering the liquidity and credit crunch 2007–2008
J. Econ. Perspect.
Learning Bayesian Networks with Mixed Variables
Deal: A Package for Learning Bayesian Networks
The bank liquidity smile across exchange rate regimes
Int. Econ. J.
MSBNx: A Component-Centric Toolkit for Modeling and Inference with Bayesian Networks
Cited by (95)
Systemic risk prediction using machine learning: Does network connectedness help prediction?
2024, International Review of Financial AnalysisNeural network-based liquidity risk prediction in Indian private banks
2024, Intelligent Systems with ApplicationsData-driven Bayesian network approach to maritime accidents involved by dry bulk carriers in Turkish search and rescue areas
2023, Regional Studies in Marine ScienceA counterfactual thinking perspective of moral licensing effect in machine-driven communication: An example of natural language processing chatbot developed based on WeChat API
2023, Technological Forecasting and Social ChangeOperational research and artificial intelligence methods in banking
2023, European Journal of Operational ResearchCitation Excerpt :Hybrid NN systems have also been used in other studies, such as by López Iturriaga and Sanz (2015), combining a multilayer perceptron and an SOM to display the classification results visually. Tavana et al. (2018) combined an NN with a Bayesian network, whereas Dželihodžić et al. (2018) integrated bagging and NNs. In terms of innovation, some studies contribute to the literature by developing NNs based on different data types.
A conceptual design decision approach by integrating rough Bayesian network and game theory under uncertain behavior selections
2022, Expert Systems with Applications
Madjid Tavana is Professor and Distinguished Chair of Business Analytics at La Salle University, where he serves as Chairman of the Business Systems and Analytics Department. He also holds an Honorary Professorship in Business Information Systems at the University of Paderborn in Germany. Dr. Tavana is Distinguished Research Fellow at the Kennedy Space Center, the Johnson Space Center, the Naval Research Laboratory at Stennis Space Center, and the Air Force Research Laboratory. He was recently honored with the prestigious Space Act Award by NASA. He holds an MBA, PMIS, and PhD in Management Information Systems and received his Post-Doctoral Diploma in Strategic Information Systems from the Wharton School at the University of Pennsylvania. He has published 12 books and over 250 research papers in international scholarly academic journals. He is the Editor-in-Chief of Decision Analytics, International Journal of Applied Decision Sciences, International Journal of Management and Decision Making, International Journal of Knowledge Engineering and Data Mining, International Journal of Strategic Decision Sciences, and International Journal of Enterprise Information Systems.
Amir-Reza Abtahi is assistant professor and chair of Information Technology Management department at Kharazmi University in Tehran, Iran. He received his MSc and PhD in Operations and Production Management from ATU, Tehran, Iran. He has published several papers in international journals such as Expert Systems with Applications, Reliability Engineering & System Safety, Annals of Operations Research, Optimization Methods and Software, Journal of Mathematical Modelling and Algorithms in Operations Research, and Journal of Industrial Engineering International. His research interest includes applied operations research, applied computational intelligence, and operations management.
Debora Di Caprio is visiting research fellow at the Department of Mathematics and Statistics of York University (Canada), researcher in the INTBM International Business and Markets Group at the Universidad Complutense de Madrid (Spain) and tenured instructor of Mathematics at the Polo Tecnologico IISS “G. Galilei” of Bolzano (Italy). She holds a MS and PhD in Pure Mathematics from York University (Canada) and has been researcher and contract professor in Italy at the Seconda Università di Napoli, the Free University of Bolzano and the Università di Trento. She has over 80 papers in international journals on mathematics, operational research, and decision theory.
Maryam Poortarigh is a Machine learning researcher. She recieved her Bachelor's degree in Applied Mathematics from Guilan University, Guilan, Iran. After that, she continued her education in knowledge Engineering and recieved a MS degree from Kharazmi University, Tehran, Iran. She has done some researches in the field of risk, using mathematical optimization techniques, and had a paper published in this area in Production Research jounal. Her main fields of interest are AI, machine learning, modelling and optimization.