Predicting Default

3 Key excerpts on "Predicting Default"

• 

  eBook - PDF

  Advances in Credit Risk Modelling and Corporate Bankruptcy Prediction

  • Stewart Jones, David A. Hensher(Authors)
  • 2008(Publication Date)
  • Cambridge University Press

    (Publisher)

  They and we reach several similar conclusions. However, in one of the central issues in this study, we differ sharply. Since the studies are so closely related, we will compare their findings to ours at several points. This paper is organized as follows. Section 2 will present models which have been used or proposed for assessing probabilities of loan default. Section 3 will describe an extension of the model. Here, we will suggest a framework for using the loan default equation in a model of cost and projected revenue to predict the profit and loss from the decision to accept a credit card application. The full model is sketched here and completed in Section 5. Sections 4 and 5 will present an application of the technique. The data and some statistical procedures for handling its distinctive character- istics are presented in Section 4. The empirical results are given in Section 5. Conclusions are drawn in Section 6. 1.2. Models for prediction of default Individual i with vector of attributes x i applies for a loan at time 0. The attributes include such items as: personal characteristics including age, sex, number of dependents and education; economic attributes such as income, employment status and home ownership; a credit history including the number of previous defaults, and so on. Let the random variable y i indicate whether individual i has defaulted on a loan (y 1 ¼ 1) or has not (y 1 ¼ 0) during the time which has elapsed from the application until y 1 is observed. 15 A statistical model for credit scoring We consider two familiar frameworks for Predicting Default. The technique of discriminant analysis is considered first. We will not make use of this technique in this study. But one of the observed outcome variables in the data that we will examine, the approval decision, was generated by the use of this technique. So it is useful to enumerate its characteristics. We then consider a probit model for discrete choice as an alternative.

  Sign up to read

  Learn more about book
• 

  eBook - ePub

  Risk, Value And Default

  • Oliviero Roggi(Author)
  • 2015(Publication Date)
  • WSPC

    (Publisher)

  1 Basel II has modified the method for calculating the capital requirements that banks should respect in order to sustain the credit risk entailed in a lending portfolio. Basel II foresees the implementation of rating systems to evaluate the credit standing of corporate clients.

  The accuracy of models to predict insolvency is of interest to researchers — and indeed to all others “obliged” to take “internal rating models” into consideration (banks and enterprises). In particular, for enterprises, self-evaluation of credit standing enables the entrepreneur to understand which variables will be most significant for deciding its banking rating, so that optimization of financial structures becomes both possible and desirable.

  This research therefore proposes to verify and compare the predictive efficacy of the most commonly employed statistical methodologies used to predict corporate insolvency, using a parity sample of selected enterprises and employing the Basel II concept of default. The definition given by the Basel Committee, namely the proxy of a default, identifies a credit event in the life of an enterprise, taking place before the legal declaration of failure (bankruptcy), namely a situation of insolvency regarding any financial obligation to a bank that is overdue by at least 90 days. The statistical methodologies to be compared are as follows: Multivariate Discriminant Analysis (MDA) or Discriminant Analysis by Generalized Estimating Equations approach (DA-GEE), Logistic Regression (LR), and Discriminant-Partial Least Squared Regression Analysis (DA-PLS). Although not yet much used to estimate default, this last methodology offers useful elements for methodological reflection by the research community. Evaluating the applicability of PLS regression to the research problem and to the question of the efficacy of prediction models for insolvency “dynamics” allows us to consider the dynamic of deterioration in the financial situation of an enterprise (financial variables of three consecutive years). This enables methodological comparison of classification results, so as to identify the best analytical instrument to apply.

  Sign up to read

  Learn more about book
• 

  eBook - PDF

  Credit Risk

  Models, Derivatives, and Management

  • Niklas Wagner(Author)
  • 2008(Publication Date)
  • Chapman and Hall/CRC

    (Publisher)

  We believe there are four main contributions to this chapter. First, we analyze and quantify the impact of fi nancial markets, business cycle indicators, and credit market indicators on default probabilities of various rating classes. Second, we show that although all economic variables help explaining default likeli-hoods, their explanatory power di ff ers greatly between investment-grade (IG) and non-investment-grade (NIG) classes. As issuer quality decreases, the dominant systematic factors change from fi nancial markets to business cycle and endogenous default indicators. Relying solely on fi nancial information to explain default probabilities leads to poor results for NIG fi rms. Default peaks are underestimated, whereas default probabilities predicted by the model overshoot realized probabilities during stable and low-default periods. Third, we establish the critical importance of combining past information with contem-poraneous factors. We show that default probability changes are the joint e ff ects of past shocks and subsequent economic trends. Resulting lead – lag e ff ects between the economy and the default cycle partially explain the lower speed and higher persistence of the default cycle. Fourth, we introduce a semi-parametric framework which explicitly takes into account the term structure of hazard rates. Firm ageing e ff ects as well as calendar-time e ff ects are explicitly taken into account. Models are set up in continuous time using the natural time-to-default framework and taking care of censoring. In particular, default probabilities can consistently be computed at all horizons. A speci fi cation test is also introduced. * We abusively use indi ff erently the terms of hazard rates and intensities in this chapter. We only point out the di ff erences when outlining our framework. Business and Financial Indicators & 237 The chapter is organized as follows. Section 14.2 brie fl y presents the default data.

  Sign up to read

  Learn more about book