{"id":{"repo_id":"adelaide","oai_identifier":"oai:digital.library.adelaide.edu.au:2440/114273"},"canonical_url":"https://search.dev.ndltd.org/etd/adelaide/oai:digital.library.adelaide.edu.au:2440/114273","repository":{"repo_id":"adelaide","name":"University of Adelaide","base_url":"https://digital.library.adelaide.edu.au/server/oai/request"},"display":{"title":"Identification of relevant predictors of loan default using the Elastic Net model","abstract":"The timely prediction of loan default plays an important role in lending decisions and monitoring loans. However, there has been little development of models for the selection of relevant variables for the prediction of loan default. This study identifies financial and economic indicators for the forward-looking prediction of loan default by the application of a penalised regression approach, namely the Elastic Net model. The study employs a sample of US firms with 162 loan default events in total between 1998 and 2013. The sample is sub-divided to form a Test sample and two holdout samples: one drawn from the same period as the Test sample; and one drawn from a subsequent period. The sample of non-defaulting firms is constructed using prior probabilities based on the bond default rate for each year. The 278 potential variables, including the ten economic indicators and 268 financial ratios or summary indicators, are regularised with the application of the Elastic Net model. This process results in the extraction of the ten predictor variables, thus identified as relevant to distinguishing between defaulting and non-defaulting firms. Only one economic indicator, the interest rate, is identified as relevant to the prediction of loan default. The prediction-usefulness of identified predictor variables are tested using the two most widely used conventional prediction models, multiple discriminant analysis (MDA) and logistic regression (Logit). The resulting MDA and Logit models are compared with Altman’s Z-score model and Ohlson’s O-score model, respectively. Both the Elastic Net prediction models provide more logical explanations of the distinctive characteristics of loan defaulting firms than the Altman’s Z-score and Ohlson’s O-score models. The Elastic Net prediction models outperform the Altman’s Z-score and Ohlson’s O-score models in the accuracy of the Type I, the Type II and the overall classification. When applied to a holdout samples within and outside the same periods, the prediction accuracy of the Elastic Net models is maintained for both defaulting and non-defaulting firms. This thesis contributes to the loan default literature by introducing the Elastic Net model for variable selection which enhances the predictive ability of the loan default prediction model. The findings of this thesis are potentially useful to financial institutions. Identification of financial and economic predictor variables of loan default can also facilitate assessment of the credit risk of loan applicants. The findings of this thesis may also facilitate better loan default prediction for purposes of monitoring loans. Lastly, the identification of relevant predictor variables may be useful for the classification of loans in the application of the expected loss model in the preparation of financial statements.","abstract_html":"The timely prediction of loan default plays an important role in lending decisions and monitoring loans. However, there has been little development of models for the selection of relevant variables for the prediction of loan default. This study identifies financial and economic indicators for the forward-looking prediction of loan default by the application of a penalised regression approach, namely the Elastic Net model. The study employs a sample of US firms with 162 loan default events in total between 1998 and 2013. The sample is sub-divided to form a Test sample and two holdout samples: one drawn from the same period as the Test sample; and one drawn from a subsequent period. The sample of non-defaulting firms is constructed using prior probabilities based on the bond default rate for each year. The 278 potential variables, including the ten economic indicators and 268 financial ratios or summary indicators, are regularised with the application of the Elastic Net model. This process results in the extraction of the ten predictor variables, thus identified as relevant to distinguishing between defaulting and non-defaulting firms. Only one economic indicator, the interest rate, is identified as relevant to the prediction of loan default. The prediction-usefulness of identified predictor variables are tested using the two most widely used conventional prediction models, multiple discriminant analysis (MDA) and logistic regression (Logit). The resulting MDA and Logit models are compared with Altman’s Z-score model and Ohlson’s O-score model, respectively. Both the Elastic Net prediction models provide more logical explanations of the distinctive characteristics of loan defaulting firms than the Altman’s Z-score and Ohlson’s O-score models. The Elastic Net prediction models outperform the Altman’s Z-score and Ohlson’s O-score models in the accuracy of the Type I, the Type II and the overall classification. When applied to a holdout samples within and outside the same periods, the prediction accuracy of the Elastic Net models is maintained for both defaulting and non-defaulting firms. This thesis contributes to the loan default literature by introducing the Elastic Net model for variable selection which enhances the predictive ability of the loan default prediction model. The findings of this thesis are potentially useful to financial institutions. Identification of financial and economic predictor variables of loan default can also facilitate assessment of the credit risk of loan applicants. The findings of this thesis may also facilitate better loan default prediction for purposes of monitoring loans. Lastly, the identification of relevant predictor variables may be useful for the classification of loans in the application of the expected loss model in the preparation of financial statements.","abstract_has_math":false,"creators":["Yi, Jessica Moonhee"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Davidson, Robyn","Loftus, Janice"],"committee_chairs":[],"committee_members":[],"year":2017,"date_issued":"2017","date_published":"2017","updated_at":"2026-07-24T00:50:34Z","subjects":["Loan default","Elastic Net","selection of variables","prediction of loan default","relevant predictor variables"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/2440/114273","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Davidson, Robyn","Loftus, Janice"]},{"key":"dc:creator","label":"Author","values":["Yi, Jessica Moonhee"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2017"]},{"key":"dc:type","label":"Dc Type","values":["Theses"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Loan default","Elastic Net","selection of variables","prediction of loan default","relevant predictor variables"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["http://hdl.handle.net/2440/114273"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["The timely prediction of loan default plays an important role in lending decisions and monitoring loans. However, there has been little development of models for the selection of relevant variables for the prediction of loan default. This study identifies financial and economic indicators for the forward-looking prediction of loan default by the application of a penalised regression approach, namely the Elastic Net model. The study employs a sample of US firms with 162 loan default events in total between 1998 and 2013. The sample is sub-divided to form a Test sample and two holdout samples: one drawn from the same period as the Test sample; and one drawn from a subsequent period. The sample of non-defaulting firms is constructed using prior probabilities based on the bond default rate for each year. The 278 potential variables, including the ten economic indicators and 268 financial ratios or summary indicators, are regularised with the application of the Elastic Net model. This process results in the extraction of the ten predictor variables, thus identified as relevant to distinguishing between defaulting and non-defaulting firms. Only one economic indicator, the interest rate, is identified as relevant to the prediction of loan default. The prediction-usefulness of identified predictor variables are tested using the two most widely used conventional prediction models, multiple discriminant analysis (MDA) and logistic regression (Logit). The resulting MDA and Logit models are compared with Altman’s Z-score model and Ohlson’s O-score model, respectively. Both the Elastic Net prediction models provide more logical explanations of the distinctive characteristics of loan defaulting firms than the Altman’s Z-score and Ohlson’s O-score models. The Elastic Net prediction models outperform the Altman’s Z-score and Ohlson’s O-score models in the accuracy of the Type I, the Type II and the overall classification. When applied to a holdout samples within and outside the same periods, the prediction accuracy of the Elastic Net models is maintained for both defaulting and non-defaulting firms. This thesis contributes to the loan default literature by introducing the Elastic Net model for variable selection which enhances the predictive ability of the loan default prediction model. The findings of this thesis are potentially useful to financial institutions. Identification of financial and economic predictor variables of loan default can also facilitate assessment of the credit risk of loan applicants. The findings of this thesis may also facilitate better loan default prediction for purposes of monitoring loans. Lastly, the identification of relevant predictor variables may be useful for the classification of loans in the application of the expected loss model in the preparation of financial statements."]},{"key":"dc:title","label":"Title","values":["Identification of relevant predictors of loan default using the Elastic Net model"]}]}],"canonical_facts":{"dc:contributor.advisor":["Davidson, Robyn","Loftus, Janice"],"dc:creator":["Yi, Jessica Moonhee"],"dc:date.issued":["2017"],"dc:description.abstract":["The timely prediction of loan default plays an important role in lending decisions and monitoring loans. However, there has been little development of models for the selection of relevant variables for the prediction of loan default. This study identifies financial and economic indicators for the forward-looking prediction of loan default by the application of a penalised regression approach, namely the Elastic Net model. The study employs a sample of US firms with 162 loan default events in total between 1998 and 2013. The sample is sub-divided to form a Test sample and two holdout samples: one drawn from the same period as the Test sample; and one drawn from a subsequent period. The sample of non-defaulting firms is constructed using prior probabilities based on the bond default rate for each year. The 278 potential variables, including the ten economic indicators and 268 financial ratios or summary indicators, are regularised with the application of the Elastic Net model. This process results in the extraction of the ten predictor variables, thus identified as relevant to distinguishing between defaulting and non-defaulting firms. Only one economic indicator, the interest rate, is identified as relevant to the prediction of loan default. The prediction-usefulness of identified predictor variables are tested using the two most widely used conventional prediction models, multiple discriminant analysis (MDA) and logistic regression (Logit). The resulting MDA and Logit models are compared with Altman’s Z-score model and Ohlson’s O-score model, respectively. Both the Elastic Net prediction models provide more logical explanations of the distinctive characteristics of loan defaulting firms than the Altman’s Z-score and Ohlson’s O-score models. The Elastic Net prediction models outperform the Altman’s Z-score and Ohlson’s O-score models in the accuracy of the Type I, the Type II and the overall classification. When applied to a holdout samples within and outside the same periods, the prediction accuracy of the Elastic Net models is maintained for both defaulting and non-defaulting firms. This thesis contributes to the loan default literature by introducing the Elastic Net model for variable selection which enhances the predictive ability of the loan default prediction model. The findings of this thesis are potentially useful to financial institutions. Identification of financial and economic predictor variables of loan default can also facilitate assessment of the credit risk of loan applicants. The findings of this thesis may also facilitate better loan default prediction for purposes of monitoring loans. Lastly, the identification of relevant predictor variables may be useful for the classification of loans in the application of the expected loss model in the preparation of financial statements."],"dc:identifier.uri":["http://hdl.handle.net/2440/114273"],"dc:subject":["Loan default","Elastic Net","selection of variables","prediction of loan default","relevant predictor variables"],"dc:title":["Identification of relevant predictors of loan default using the Elastic Net model"],"dc:type":["Theses"]},"updated_at":"2026-07-24T00:50:34Z"}