Can the standard EBIT-based structural model replicate credit ratings? : an empirical study on S&P500 non-financial firms

• Autor(a) principal: Madsen, Simen Bjølseth
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.14/31316
• topic: Credit risk; Default prediction; Structural models; Credit ratings; Risco de crédito; Previsão de insolvência; Modelos estruturais; Classificações de crédito; Domínio/Área Científica::Ciências Sociais::Economia e Gestão

The objective of this thesis is to analyze whether the default measures obtained through the standard EBIT-based structural model are comparable to those obtained through credit ratings. This study covers all non-financial companies present on the S&P500 throughout the 2004-2018 period. Credit risk measures coming from the two approaches were found to be broadly comparable. Nevertheless, it was found that on average the structural model under predicts the credit-ratings probability of default by 0,68 p.p. and over predicts the distance to default by 0,57 standard deviations. This under prediction of credit risk was observed across all sectors, though with different degrees of intensity depending on the economic sector. The underprediction was found in all years of study except the financial crisis period. This dissertation proceeded by analysing the relation between the model and rating agencies default measures. The two estimates show a relatively strong correlation, notably 44% in the case of the probability of default and 52% in the case of the distance to default. The relation between the distances to default measures has been further studied through panel data regressions both on levels (with and without firm fixed effects) and on time differences. Under all approaches the coefficient for the model distance to default measure was found to be relatively small but significant at all the usual confidence levels. This result suggests that the structural model tends to overreact on all new information, while rating agencies act more smoothly.