Interpretable selective learning in credit risk

Forecasting credit default risk has been an important research field for several decades. Traditionally, logistic regression has been widely recognized as a solution because of its accuracy and interpretability. Although complex machine learning models may improve accuracy over simple logistic regressions, their interpretability has prevented their use in credit risk assessment. We introduce a neural network with a selective option to increase interpretability by distinguishing whether linear models can explain the dataset. Our methods are tested on two datasets: 25,000 samples from the Taiwan payment system collected in October 2005 and 250,000 samples from the 2011 Kaggle competition. We find that, for most of samples, logistic regression will be sufficient, with reasonable accuracy; meanwhile, for some specific data portions, a shallow neural network model leads to much better accuracy without significantly sacrificing interpretability.     

Predicting credit card delinquencies: An application of deep neural networks

The objective of this paper is twofold. First, it develops a prediction system to help the credit card issuer model the credit card delinquency risk. Second, it seeks to explore the potential of deep learning (also called a deep neural network), an emerging artificial intelligence technology, in the credit risk domain. With real-life credit card data linked to 711,397 credit card holders from a large bank in Brazil, this study develops a deep neural network to evaluate the risk of credit card delinquency based on the client's personal characteristics and the spending behaviours. Compared with machine-learning algorithms of logistic regression, naive Bayes, traditional artificial neural networks, and decision trees, deep neural networks have a better overall predictive performance with the highest F scores and area under the receiver operating characteristic curve. The successful application of deep learning implies that artificial intelligence has great potential to support and automate credit risk assessment for financial institutions and credit bureaus.     

Stochastic volatility options pricing with wavelets and artificial neural networks

Abstract

The paper describes an alternative options pricing method which uses a binomial tree linked to an innovative stochastic volatility model. The volatility model is based on wavelets and artificial neural networks. Wavelets provide a convenient signal/noise decomposition of the volatility in the nonlinear feature space. Neural networks are used to infer future volatility from the wavelets feature space in an iterative manner. The bootstrap method provides the 95% confidence intervals for the options prices. Market options prices as quoted on the Chicago Board Options Exchange are used for performance comparison between the Black‐Scholes model and a new options pricing scheme. The proposed dynamic volatility model produces as good as and often better options prices than the conventional Black‐Scholes formulae.     

信用评分模型开发中的抽样技术

信用评分是运用数据挖掘技术对已知客户的信息进行分析，建立能预测未来客户信用表现的模型。数据准备是评分模型开发过程中非常重要的步骤，数据质量的好坏直接决定了模型的成败。由于银行内部的数据量非常庞大，为了使分析更加有效率，需要对数据进行抽样。因此，如何进行抽样，如何保证样本能够充分代表总体就非常重要。根据信用评分模型的开发经验以及数据挖掘中的抽样理论，现提出如下建立评分模型时应用的抽样技术以及注意事项。     

Information reduction via level crossings in a credit risk model

This paper provides an alternative credit risk model based on information reduction where the market only observes the firm’s asset value when it crosses certain levels, interpreted as changes significant enough for the firm’s management to make a public announcement. For a class of diffusion processes we are able to provide explicit expressions for the firm’s default intensity process and its zero-coupon bond prices.      

Predicting sovereign debt crises using artificial neural networks: A comparative approach

Recent episodes of financial crisis have revived interest in developing models able to signal their occurrence in timely manner. The literature has developed both parametric and non-parametric models, the so-called Early Warning Systems, to predict these crises. Using data related to sovereign debt crises which occurred in developing countries from 1980 to 2004, this paper shows that further progress can be achieved by applying a less developed non-parametric method based on artificial neural networks (ANN). Thanks to the high flexibility of neural networks and their ability to approximate non-linear relationship, an ANN-based early warning system can, under certain conditions, outperform more consolidated methods.     

Black–Scholes versus artificial neural networks in pricing FTSE 100 options

This paper compares the performance of Black–Scholes with an artificial neural network (ANN) in pricing European‐style call options on the FTSE 100 index. It is the first extensive study of the performance of ANNs in pricing UK options, and the first to allow for dividends in the closed‐form model. For out‐of‐the‐money options, the ANN is clearly superior to Black–Scholes. For in‐the‐money options, if the sample space is restricted by excluding deep in‐the‐money and long maturity options (3.4% of total volume), then the performance of the ANN is comparable to that of Black–Scholes. The superiority of the ANN is a surprising result, given that European‐style equity options are the home ground of Black–Scholes, and suggests that ANNs may have an important role to play in pricing other options for which there is either no closed‐form model, or the closed‐form model is less successful than is Black–Scholes for equity options. Copyright © 2005 John Wiley & Sons, Ltd.     

Portfolio creation using artificial neural networks and classification probabilities: a Canadian study

Financial Markets and Portfolio Management - This study aims to verify whether using artificial neural networks (ANNs) to establish classification probabilities generates portfolios with higher...     

Modelling small‐business credit scoring by using logistic regression,neural networks and decision trees

Previous research on credit scoring that used statistical and intelligent methods was mostly focused on commercial and consumer lending. The main purpose of this paper is to extract important features for credit scoring in small‐business lending on a dataset with specific transitional economic conditions using a relatively small dataset. To do this, we compare the accuracy of the best models extracted by different methodologies, such as logistic regression, neural networks (NNs), and CART decision trees. Four different NN algorithms are tested, including backpropagation, radial basis function network, probabilistic and learning vector quantization, by using the forward nonlinear variable selection strategy. Although the test of differences in proportion and McNemar's test do not show a statistically significant difference in the models tested, the probabilistic NN model produces the highest hit rate and the lowest type I error. According to the measures of association, the best NN model also shows the highest degree of association with the data, and it yields the lowest total relative cost of misclassification for all scenarios examined. The best model extracts a set of important features for small‐business credit scoring for the observed sample, emphasizing credit programme characteristics, as well as entrepreneur's personal and business characteristics as the most important ones. Copyright © 2005 John Wiley & Sons, Ltd.     

Predicting direction shifts on Canadian–US exchange rates with artificial neural networks

The paper presents a variety of neural network models applied to Canadian–US exchange rate data. Networks such as backpropagation, modular, radial basis functions, linear vector quantization, fuzzy ARTMAP, and genetic reinforcement learning are examined. The purpose is to compare the performance of these networks for predicting direction (sign change) shifts in daily returns. For this classification problem, the neural nets proved superior to the naïve model, and most of the neural nets were slightly superior to the logistic model. Using multiple previous days' returns as inputs to train and test the backpropagation and logistic models resulted in no increased classification accuracy. The models were not able to detect a systematic affect of previous days' returns up to fifteen days prior to the prediction day that would increase model performance. Copyright © 2001 John Wiley & Sons, Ltd.     

European systemic credit risk transmission using Bayesian networks

The analysis of systemic credit risk is one of the most important concerns within the financial system. Its complexity lies in adequately measuring how the transmission of systemic default spreads through assets or financial markets. The transmission structure of systemic credit risk across several European sectoral CDS is studied by dynamic Bayesian networks. The new approach allows for a more advanced analysis of systemic risk transmission, including long-term and more complex relationships. The modelling reveals as relevant only relationships between the original series and one- and two-lagged series. Network structure learning displays a robust and stationary underlying risk transmission structure, pointing to a consolidated transmission mechanism of systemic credit risk between CDSs. Between 5 % and 40 % of sectoral CDS series variances are explained by the network relationships. The modelling allows us to ascertain which relationships between the CDS series show positive (amplifier) and negative (reducer) effects of systemic risk transmission.     

The efficacy of red flags in predicting the SEC's targets: an artificial neural networks approach

This paper illustrates the application of artificial neural networks (ANNs) to test the ability of selected SAS No. 53 red flags to predict the targets of the SEC investigations. Investors and auditors desire to predict SEC targets because substantial losses in equity value are associated with SEC investigations. The ANN models classify the membership in target (investigated) versus control (non-investigated) firms with an average accuracy of 81%. One reason for the relative success of the ANN models is that ANNs have the ability to ‘learn’ what is important. The participants in financial reporting frauds have incentives to appear prosperous as evidenced by high profitability. In contrast to conventional statistical models with static assumptions, the ANNs use adaptive learning processes to determine what is important in predicting targets. Thus, the ANN approach is less likely to be affected by accounting manipulations. Our ANN models are biased against achieving predictive success because we use only publicly available information. The results confirm the value of red flags, i.e. financial ratios available from trial balance in conjunction with non-financial red flags such as the turnover of CEO, CFO and auditors do have predictive value. © 2000 John Wiley & Sons, Ltd.     

A flow-based corporate credit model

The main purpose of this paper is to develop a flow-based corporate credit model. This model can concurrently and endogenously generate a firm’s multi-period probabilities of liquidity crunch and expected liquidity shortfalls. This study builds a state-dependent internal liquidity model that incorporates both systematic and idiosyncratic shocks into corporate internal liquidity dynamics. The flow-based credit model differs from structural form credit models in that it considers a flow-based insolvency rather than a stock-based one, and has a potential to capture short-term credit information. Additionally, it differs from both reduced form and traditional accounting-based bankruptcy prediction models in that it is able to provide multi-period expected liquidity shortfalls endogenously.     

Bankruptcy prediction of financially stressed firms: an examination of the predictive accuracy of artificial neural networks

This is an extension of prior studies that have used artificial neural networks to predict bankruptcy. The incremental contribution of this study is threefold. First, we use only financially stressed firms in our control sample. This enables the models to more closely approximate the actual decision processes of auditors and other interested parties. Second, we develop a more parsimonious model using qualitative ‘bad news’ variables that prior research indicates measure financial distress. Past research has focused on the ‘usefulness’ of accounting numbers and therefore often ignored non‐accounting variables that may contribute to the classification accuracy of the distress prediction models. In addition, rather than use multiple financial ratios, we include a single variable of financial distress using the Zmijewski distress score that incorporates ratios measuring profitability, liquidity, and solvency. Finally, we develop and test a genetic algorithm neural network model. We examine its predictive ability to that of a backpropagation neural network and a model using multiple discriminant analysis. The results indicate that the misclassification cost of the genetic algorithm‐based neural network was the lowest among the models. Copyright © 2001 John Wiley & Sons, Ltd.     

A simple model of credit contagion

We propose a simple model of credit contagion in which we include macro- and microstructural interdependencies among the debtors within a credit portfolio. The microstructure captures interdependencies between debtors that go beyond their exposure to common factors, e.g., business or legal interdependencies. We show that even for diversified portfolios, moderate microstructural interdependencies have a significant impact on the tails of the loss distribution. This impact increases dramatically for less diversified microstructures.     

Tempered stable structural model in pricing credit spread and credit default swap

In this paper, we explore the features of a structural credit risk model wherein the firm value is driven by normal tempered stable (NTS) process belonging to the larger class of Lévy processes. For the purpose of comparability, the calibration to the term structure of a corporate bond credit spread is conducted under both NTS structural model and Merton structural model. We find that NTS structural model provides better fit for all credit ratings than Merton structural model. However, it is noticed that probabilities of default derived from the calibration of the term structure of a bond credit spread might be overestimated since the bond credit spread could contain non-default components such as illiquidity risk or asymmetric tax treatment. Hence, considering CDS spread as a reflection of the pure credit risk for the reference entity, we calibrate it in order to obtain more reasonable probability of default and obtain valid results in calibration of the market CDS spread with NTS structural model.     

网络支撑农信蝶变--华为3Com为农村信用社打造自适应网络

近年来，基于向质量效益型转轨的战略，各大国有商业银行纷纷缩减在农村的服务网点。在深化农村经济体制改革和适应农村社会经济发展两大目标驱动下，作为中国农村金融服务业的主体之一，农村信用合作社（以下简称“农信”）的体制改革层层深化，发生了令人瞩目的化蛹为蝶之变：从县农信联社统一法人，到确立新的管理体制、组建省农信联社，再到股份制改造、试点成立农村商业银行……     

Real option valuation with neural networks

We propose to use neural networks to value options when analytical solutions do not exist. The basic idea of this approach is to approximate the value function of a dynamic program by a neural net, where the selection of the network weights is done via simulated annealing. The main benefits of this method as compared to traditional approximation techniques are that there are no restrictions on the type of the underlying stochastic process and no limitations on the set of possible actions. This makes our approach especially attractive for valuing Real Options in flexible investments. We, therefore, demonstrate the method proposed by valuing flexibility for costly switch production between several products under various conditions. © 1998 John Wiley & Sons, Ltd.