The Performance of Credit Rating Systems in the Assessment of Collateral Used in Eurosystem Monetary Policy Operations

Coppens, François; González, Fernando; Winkler, Gerhard

doi:10.2139/ssrn.1687537

Cited by 97 publications

(7 citation statements)

References 101 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, the traffic light test (cf. Coppens et al, 2007) will diagnose the relevance of our model for practitioners. Most of the analyses performed during model development and refining, will be in-sample type followed by out-ofsample tests.…”

Section: Methodsmentioning

confidence: 97%

Firm's default ? new methodological approach and preliminary evidence from Poland

Berent

Bławat

Dietl

et al. 2017

View full text Add to dashboard Cite

Research background: Bankruptcy literature is populated with scores of (econometric) models ranging from Altman’s Z-score, Ohlson’s O-score, Zmijewski’s probit model to k-nearest neighbors, classification trees, support vector machines, mathematical programming, evolutionary algorithms or neural networks, all designed to predict financial distress with highest precision. We believe corporate default is also an important research topic to be identified with the prediction accuracy only. Despite the wealth of modelling effort, a unified theory of default is yet to be proposed. Purpose of the article: Due to the disagreement both on the definition and hence the timing of default, as well as on the measurement of prediction accuracy, the comparison (of predictive power) of various models can be seriously misleading. The purpose of the article is to argue for the shift in research focus from maximizing accuracy to the analysis of the information capacity of predictors. By doing this, we may yet come closer to understanding default itself. Methods: We critically appraise the bankruptcy research literature for its methodological variety and empirical findings. Default definitions, sampling procedures, in and out-of-sample testing and accuracy measurement are all scrutinized. In an empirical part, we use a double stochastic Poisson process with multi-period prediction horizon and a comprehensive database of some 15,000 Polish non-listed companies to illustrate the merits of our new approach to default modelling. Findings & Value added: In the theoretical part, we call for the construction of a single unified default forecasting platform estimated for the largest dataset of firms possible to allow testing the utility of various sources of micro, mezzo, and macro information. Our preliminary empirical evidence is encouraging. The accuracy ratio amounts to 0.92 for t = 0 and drops to 0.81 two years ahead of default. We point to the pivotal role played by the information on firm’s liquidity (alternatively in profitability) and — in contrast to Altman’s tradition — hardly any contribution to predictive power of other financial ratios. Macro data is shown to be critical. It adds, on average, more than 10 p.p. to accuracy ratio. In the future, we hope to integrate listed and non-listed firms data into one model, ideally at higher frequency than annual, and include the information on firm's competitiveness position.

show abstract

Section: Methodsmentioning

confidence: 97%

Firm's default ? new methodological approach and preliminary evidence from Poland

Berent

Bławat

Dietl

et al. 2017

View full text Add to dashboard Cite

show abstract

“…The positive assets correlation has a significant impact on the distribution functionofportfoliolosses.Forthisreason,Coppens [Coppens,Gonzalez,Winkler 2007]consideredtheimplementationofassetcorrelationintotheprocessofassessing thequalityofPDforecasts.Inhisstudyhegeneralizedthebinomialtest,whichled himtoanewformula.Underhisapproachthecriticalvalueforthestatisticaltest wascalculatedasfollows: (12) Whenthenumberofelementsineachgroupincreases,thedistributionofthe above-mentionedstatisticconvergestoχ 2 distributionfunctionwithK−2degreesof freedom. The backtesting analysis according to the Hosmer-Lemeshow statistical testcomesdowntoacomparisonofthevalueoftheteststatisticwiththecritical valuederivedfromχ 2 cumulativedistributionfunctionforgivensignificancelevelα.…”

Section: Statistical Methods Under Backtesting Proceduresmentioning

confidence: 99%

Backtesting analysis. How to assess the quality of PD models in a retail banking

Siarka¹

2019

Prace Naukowe Uniwersytetu Ekonomicznego we Wrocławiu

View full text Add to dashboard Cite

Thepaperreferstotheprobabilityofdefaultmodelvalidationprocedureinretail banking.Theauthorpresentstheideaofbacktestinganalysisfocusingonsensitivityanalysis of capital requirements under stress scenarios. The paper addresses statistical methods which can be applied in credit risk management under the backtesting exercise in retail banking.Theadvantagesanddrawbacksofspecificapproachesarediscussed.Furthermore, theoutcomesoftheempiricalimplementationofselectedmethodsarepresented.Theauthor considerstheimpactofpositiveassetcorrelationonvariousvalidationapproaches,whereno correlationisassumed,andprovesthatthezero-correlationassumptionsmayresultinamore prudentapproach.Thisfindingwasconfirmedbytheempiricalanalysisperformedforretail portfolios. The research concerned PD parameters calculated for car and mortgage loans. The backtesting results revealed that PD forecasts created for mortgage portfolios underestimatedcreditriskduringthecrisisperiodwhichstartedin2008.However,carloan portfoliocreditriskpredictionsappearedtoberobust.

show abstract

“…To assess the degree to which probabilities of default match realized defaults, we perform a binomial-style test for different credit quality buckets, using the Credit Quality Steps (CQS) defined by the Eurosystem for the validation and annual monitoring of credit rating systems. In particular, we use a backtesting strategy outlined in Coppens et al (2017) yellow and red indicating that there is a positive or strongly positive discrepancy between expected and realized defaults respectively.…”

Section: Backtestingmentioning

confidence: 99%

Corporate default forecasting with machine learning

Moscatelli

Parlapiano

Narizzano

et al. 2020

Expert Systems with Applications

View full text Add to dashboard Cite

We analyze the performance of a set of machine learning (ML) models in predicting default risk, using standard statistical models, such as the logistic regression, as a benchmark. When only a limited information set is available, for example in the case of financial indicators, we find that ML models provide substantial gains in discriminatory power and precision compared with statistical models. This advantage diminishes when high quality information, such as credit behavioral indicators obtained from the Credit Register, is also available, and becomes negligible when the dataset is small. We also evaluate the consequences of using an ML-based rating system on the supply of credit and the number of borrowers gaining access to credit. ML models channel a larger share of credit towards safer and larger borrowers and result in lower credit losses for lenders.

show abstract

The Performance of Credit Rating Systems in the Assessment of Collateral Used in Eurosystem Monetary Policy Operations

Cited by 97 publications

References 101 publications

Firm's default ? new methodological approach and preliminary evidence from Poland

Firm's default ? new methodological approach and preliminary evidence from Poland

Backtesting analysis. How to assess the quality of PD models in a retail banking

Corporate default forecasting with machine learning

Contact Info

Product

Resources

About