Bank Failure Prediction Research Articles

Using boosting algorithms to predict bank failure: An untold story

From a modeling point of view, our work provides a novel approach to better use XGBoost for bank failure prediction, determining the essential technical aspects that can improve the predictive accuracy. Of these technical aspects, the two crucial factors are assigning correct values to target variables and careful predictor selection (through ANOVA, correlation, information value tests, and weight of evidence). We also highlight that bank failure could be predicted four to five quarters earlier when all predictive signals simultaneously appear. Hence, we strongly suggest using quarterly data instead of yearly data. In addition to practical implications, our present work also contributed to the existing literature. We confirm the results of existing studies that emphasized that XGBoost has strong predictive power (Carmona, Climent, and Momparler (2018)). Moreover, we provide evidence that XGBoost outperforms other models in the same boosting family, including gradient boosting and AdaBoost, through an intensive comparison of predictive power. These contributions might facilitate future work on bank failure prediction.

Bank Failure Prediction in the Arab Region Using Logistic Regression Model

Bank Failure Prediction in the Arab Region Using Logistic Regression Model

Relevance of size in predicting bank failures

AbstractEmploying a statistical model‐building strategy, this study aims to analyse the United States' bank failures across different size categories (small, medium, and large). Our results suggest that factors associated with bank failures vary across respective size categories, and the average marginal effects (AMEs) of mutually significant covariates also exhibit significant variability across different size classes of banks. The results are robust to up‐to 3 years of lagged regression estimates, various control variables, interaction between bank size and bank charter, alternative bank size classifications, and macroeconomic crisis periods.

Predicting bank insolvencies using machine learning techniques

Proactively monitoring and assessing the economic health of financial institutions has always been the cornerstone of supervisory authorities. In this work, we employ a series of modeling techniques to predict bank insolvencies on a sample of US-based financial institutions. Our empirical results indicate that the method of Random Forests (RF) has a superior out-of-sample and out-of-time predictive performance, with Neural Networks also performing almost equally well as RF in out-of-time samples. These conclusions are drawn not only by comparison with broadly used bank failure models, such as Logistic, but also by comparison with other advanced machine learning techniques. Furthermore, our results illustrate that in the CAMELS evaluation framework, metrics related to earnings and capital constitute the factors with higher marginal contribution to the prediction of bank failures. Finally, we assess the generalization of our model by providing a case study to a sample of major European banks.

Bank Failure Prediction Model Based on Governance: A Case of Rural Banks in Indonesia

Since it was first operating in 2005 until 2017, Indonesia Deposit Insurance Corporation (IDIC) has liquidated 91 rural banks which were determined as failed banks by supervision authority. The cause of the failing of the bank is mainly due to the incapability of the bank to meet the minimum Capital Adequacy Ratio (CAR). Bank’s capital was shrunk by the loss caused by fraud. The fraud is mostly induced by the lack of good corporate governance implementation. By using the logistic regression, it can be concluded that (1) the incomplete of responsibility letter which will be used in the event of bank failure, submitted by the bank commissioners; (2) the incomplete of responsibility letter which will be used in the event of bank failure, submitted by the bank directors; (3) role duplication between shareholders and board of directors; and (4) bank had classified as special supervision, have impact on the increase of rural banks failure. At the same time, the compliance level of rural banks to a correct premium payment has impacted to decrease of rural banks failure possibilities.

An ordinal classification framework for bank failure prediction: Methodology and empirical evidence for US banks

Bank failure prediction models usually combine financial attributes through binary classification approaches. In this study we extend this standard framework in three main directions. First, we explore the predictive power of attributes that describe the diversification of banking operations. Second, we consider the prediction of failure in a multi-period context. Finally, an enhanced ordinal classification framework is introduced, which considers multiple instances of failed banks prior to failure (up to three years prior to bankruptcy). Various ordinal models are developed using techniques from multiple criteria decision analysis, statistics, and machine learning. Moreover, ensemble models based on error-correcting output codes are examined. The analysis is based on a sample consisting of approximately 60,000 observations for banks in the United States over the period 2006–2015. The results show that diversification attributes improve the predictive power of bank failure prediction models, mainly for mid to long-term prediction horizons. Moreover, ordinal classification models provide a better description of the state of the banks prior to failure and are competitive to standard binary classification models.

Loan Portfolio Risk and Capital Adequacy: A New Approach to Evaluating the Riskiness of Banks

We develop an accounting-based Loan Portfolio Risk (LPR) variable that measures time-varying volatility in default risk for a portfolio of bank loans. An Equity-to-LPR ratio (ELPR) is incrementally important in predicting bank failure up to five years in advance, after controlling for all the CAMELS variables. It is also additive to other fundamental-based risk measures from prior studies. Further, we find publicly-listed banks with higher ELPR have lower market-implied costs-of-capital, and ELPR strongly predicts cross-sectional stock returns under market stress conditions after adjusting for other known risk factors. We conclude ELPR captures key aspects of bank risk that are missing in current Basel Committee risk-weighted-asset calculations.

Bank Failure: A New Approach to Prediction and Supervision

Bank failures are costly to customers and the wider market. Prevention is always better than cure but in light of recent economic downturns, it has become increasingly difficult for regulators to allocate more resources towards in-depth monitoring of banking practices. In this paper, we construct a tool that is able to predict bank failures ahead of time with reasonable accuracy. Through a logistic regression on a matched sample of 536 failed and non-failed US banks, we determine the financial indicators that most accurately predicts bank failure. From the regression, we construct a Bank Health Index that assesses a bank’s propensity to failure. In-sample and out-of-sample tests show that our model is about 90% accurate two years prior to failure, and 95% accurate the year before failure. The accuracy and efficiency of the model and index provides a more efficient and effective tool for assessing a bank’s propensity to failure besides requiring far less resources. With these methods, regulators will be able to take preventive measures at least one year before failure, saving the economy millions if not billions in the process.

Bank Failure Prediction in Relation to the Business Life Cycle

The results suggest that the capital ratio, loan ratio, non-performing loans, provisions for loan losses, fixed assets, return of equity, ratio of interest income to interest expenses, and ratio of non-interest income to non-interest expenses all had different correlations to the financial distress experienced by banks that were at a business life cycle stage. The logistic model employed in this study for predicting bank failure explains most of the banking trends in NIC banks at the declining stage. The accuracy of G8 banks at the growth stage performed well, whereas NIC banks at the declining stage performed poorly.

Comparison of SVM and FSVM for predicting bank failures using chi-square feature selection

Bankruptcy doesn’t happen suddenly, but there are early indications that can be seen by investigating the financial statement of a bank. In this research, we aim to find the best bankruptcy prediction model to give an early warning for regulators so that it can help them to prevent or lessen the negative effects on economic systems. We will be performing SVM and modification of SVM by adding fuzzy membership function called FSVM to analyze bank’s health. We chose machine learning for bankruptcy prediction because it can give faster result rather than traditional statistical method. The prediction accuracy will be measured by using the dataset that consists of 65 Turkish banks of which each of them has an information of 20 financial ratios. Furthermore, to improve the accuracy prediction, we also perform chi-square feature selection (CSFS) to filter any irrelevant features of total 20 features in our dataset. CSFS can sort all 20 features based on chi-square score from the most relevant feature to the least one. After that, we will choose 5, 10, and 15 best features, so that we have four datasets to be classified into healthy and non-healthy banks. We found that using 5 features and SVM classifier gives the highest accuracy prediction, which scores 98.28%. For most cases, SVM gives better performance compared to FSVM.

Systematic risk factors and bank failures

We investigate the impact on bank failures of two systematic risk factors—the stress in the interbank funding market and conditions in the local housing market—in the two banking crises during the period from 1985 to 2011. We find that local housing market conditions had a stronger relation with bank failures during the first episode than during the second episode. By contrast, the interbank funding market condition only weakly helped to predict bank failures before 2005, but it had a very strong impact on bank failures during the latest banking crisis. Therefore, banking crises seemed to be driven by episode-specific risk factors. Accordingly, even though it is common wisdom at present that overall banking risk is more than just the sum of risks at the individual bank level and that macro-prudential overlay is important in bank regulation, finding the right systematic risk factors for the next banking crisis or correctly gauging their impact on bank failures might be challenging.

Forecast bankruptcy using a blend of clustering and MARS model: case of US banks

In this paper, we compare the performance of two non-parametric methods of classification and regression trees (CART) and the newly multivariate adaptive regression splines (MARS) models, in forecasting bankruptcy. Models are tested on a large universe of US banks over a complete market cycle and run under a K-fold cross validation. Then, a hybrid model which combines K-means clustering and MARS is tested as well. Our findings highlight that (i) Either in training or testing sample, MARS provides, in average, better correct classification rate than CART model (ii) Hybrid approach significantly increases the classification accuracy rate in the training sample (iii) MARS prediction underperforms when the misclassification of the bankrupt banks rate is adopted as a criteria (iv) Finally, results prove that non-parametric models are more suitable for bank failure prediction than the corresponding Logit model.

Predicting failure in the U.S. banking sector: An extreme gradient boosting approach

Banks play a central role in developed economies. Consequently, systemic banking crises destabilize financial markets and hamper global economic growth. In this study, extreme gradient boosting was used to predict bank failure in the U.S. banking sector. Key variables were identified to anticipate and prevent bank defaults. The data, which spanned the period 2001 to 2015, consisted of annual series of 30 financial ratios for 156 U.S. national commercial banks. Identifying leading indicators of bank failure is vital to help regulators and bank managers act swiftly before distressed financial institutions reach the point of no return. The findings indicate that lower values for retained earnings to average equity, pretax return on assets, and total risk-based capital ratio are associated with a higher risk of bank failure. In addition, an exceedingly high yield on earning assets increases the chance of bank financial distress.

The Expected Rate of Credit Losses on Banks' Loan Portfolios

ABSTRACT Estimating expected credit losses on banks' portfolios is difficult. The issue has become of increasing interest to academics and regulators with the FASB and IASB issuing new regulations for loan impairment. We develop a measure of the one-year-ahead expected rate of credit losses (ExpectedRCL) that combines various measures of credit risk disclosed by banks. It uses cross-sectional analyses to obtain coefficients for estimating each period's measure of expected credit losses. ExpectedRCL substantially outperforms net charge-offs in predicting one-year-ahead realized credit losses, and reflects nearly all the credit loss-related information in the charge-offs. ExpectedRCL also contains incremental information about one-year-ahead realized credit losses relative to the allowance and provision for loan losses and the fair value of loans. It is a better predictor of the provision for loan losses than analyst provision forecasts, and is incrementally useful beyond other credit risk metrics in predicting bank failure up to one year ahead.

How Early Can Non-Performance Loan Predict Bank Failure? Evidence from US Bank Failure during 2008-2010

Probit model was applied on the non-performance loans (NPL) of eight quarters, quarter 1— quarter 8, in determining the significant quarter before the bank was declared failure. The result of the Probit estimates found that as early as one-year ahead (4th quarter-ahead) bank-failure can be alerted and predicted. The NPL of the 4th quarter was a significant predictor of bank failure. The estimates of the model correctly predicts 89.6 percent of the U.S. banks that failed and 97.6 percent of the banks that survived during 2008-2010. Overall, the estimated model correctly predicts 95.5 percent of the observations (89.6 percent of the failure =0 and 97.6 percent of the survival=1 observations).The paper provides policy prescription that bank managements and bank regulators should pay attention to the early quarter(s) that are significant factor (s) for bank failure.

Predicting US bank failures: A comparison of logit and data mining models

AbstractPredicting bank failures is important as it enables bank regulators to take timely actions to prevent bank failures or reduce the cost of rescuing banks. This paper compares the logit model and data mining models in the prediction of bank failures in the USA between 2002 and 2010 using levels and rates of change of 16 financial ratios based on a cross‐section sample. The models are estimated for the in‐sample period 2002–2009, while data for the year 2010 are used for out‐of‐sample tests. The results suggest that the logit model predicts bank failures in‐sample less precisely than data mining models, but produces fewer missed failures and false alarms out‐of‐sample.

Predicting bank failure: An improvement by implementing a machine-learning approach to classical financial ratios

This research compares the accuracy of two approaches: traditional statistical techniques and machine learning techniques, which attempt to predict the failure of banks. A sample of 3000 US banks (1438 failures and 1562 active banks) is investigated by two traditional statistical approaches (Discriminant analysis and Logistic regression) and three machine learning approaches (Artificial neural network, Support Vector Machines and k-nearest neighbors). For each bank, data were collected for a 5-year period before they become inactive. 31 financial ratios extracted from bank financial reports covered 5 main aspects: Loan quality, Capital quality, Operations efficiency, Profitability and Liquidity. The empirical result reveals that the artificial neural network and k-nearest neighbor methods are the most accurate.

Predicting Bank Failure: Evidence from Turkey and West African Economic and Monetary Union (WAEMU)

The study analyzes the factors that may have an impact on the banking sector, particularly the ones which could be useful to predict the financial failure of banks. The study compares and contrasts the banking systems of Turkey and West African countries where the main actor in financial system occurs to be the banks. First, it has been found that (equity + profit)/total assets, fixed assets/total assets, operating expenses/total assets, (personnel expenses + severance pay)/total assets ratios are effective in estimating of financial failure of banks in Turkey. Second, Net Period profit/Average equity and Interest income/ Average Earning assets ratio are effective in prediction of financial failure of banks in the West African Economic and Monetary Union. The study suggests that to determine financial failure of banks, various financial ratios play a fairly consequential role for a variety of countries.

Building an Early Warning System of Bank Failure with Alternative Methods

An early warning system (EWS) to flag troubled banks has been worked on since the early 1990s. The S&L crisis in the 1980s as well as the financial crisis of 2007-2009 have been extensively worked on to find important variables that help to predict bank failure and the logit model has been the standard workhorse. This paper adopts another strategy to predict bank failure in the machine learning field in order to build a more efficient EWS. Using a logit model and a random forest, I find that the random forest approach significantly outperforms the logit regression in the training (in-sample) data. The random forest also outperforms the logit in bank failure prediction in the testing (out-of-sample) data as the data used to estimate the model is closer in time to the testing sample. One way to make an EWS more effective would be to flag potentially troubled banks a few years in advance using a logit model at the beginning of a crisis, and then quickly update these predictions with a random forest as the crisis evolves and more data readily available.

A Bank Failure Prediction Model for Zimbabwe: A Corporate Governance Perspective

The primary objective of this study was to come up with a bank failure prediction model for Zimbabwe. The research sample comprised five failed commercial banks that were operational in 2003 as well as five non-failed commercial banks that were operational during that same period. The model developed in this research was applied to each of these banks and a failure classification awarded. Out of a sample of ten banks, the model misclassified one bank as failed instead of non-failed and this signified a strong predictive power. Results revealed a distinct pattern of owner managed banks being predicted to fail while those banks run by professional managers, divorced from ownership, were getting high passes, a sign of stability. Some owner managed entities were predicted as non-failing and this was interpreted as emanating from a strong presence of institutional and other outside shareholders with a significant shareholding in the banks and thus eliminating shareholder concentration. The findings from the research showed that owner managers were more likely to commit corporate governance abuses than professional managers. It was concluded that corporate governance factors significantly contributed to the bank failures experienced in Zimbabwe between 2003 and 2004. As a result, banks need to focus more on corporate governance factors to avoid failures in the future.