Improving Forecast of Binary Rare Events Data: A GAM‐Based Approach

Abstract

This paper develops a method for modelling binary response data in a regression model with highly unbalanced class sizes. When the class sizes are highly unbalanced and the minority class represents a rare event, conventional regression analysis, i.e. logistic regression models, could underestimate the probability of the rare event. To overcome this drawback, we introduce a flexible skewed link function based on the quantile function of the generalized extreme value (GEV) distribution in a generalized additive model (GAM). The proposed model is known as generalized extreme value additive (GEVA) regression model, and a modified version of the local scoring algorithm is suggested to estimate it. We apply the proposed model to a dataset on Italian small and medium enterprises (SMEs) to estimate the default probability of SMEs. Our proposal performs better than the logistic (linear or additive) model in terms of predictive accuracy. Copyright © 2015 John Wiley & Sons, Ltd.

A Joint Scoring Model for Peer-to-Peer and Traditional Lending: A Bivariate Model with Copula Dependence

“…Most of the performance measures in Table show that the approach that is suggested in this paper of using BivGEV outperforms the univariate Logit 1 that is commonly used in industry. In agreement with the results obtained in the univariate framework (Andreeva et al ., ; Calabrese et al ., ; Calabrese and Osmetti, ), using an asymmetric link function improves the predictive accuracy of the scoring model, as can be seen by comparing the results for BivGEV and the bivariate probit model. Finally, we observe that BivGev is more accurate in forecasting defaults than are the two logit models (SMOTE+Logit 1 and SMOTE+Logit 2 ) when a SMOTE approach is applied.…”

“…Instead, the characteristics of the rare events, represented by the defaulted borrowers y i = 1, are more informative than those of the non-defaulters y i = 0, so the default probability for the actual defaults is underestimated (Andreeva et al, 2016;Calabrese et al, 2016). Previous studies (Calabrese and Osmetti, 2015;Calabrese et al, 2016) show these disadvantages for various default percentages (1%, 2%, 5% and 10%).…”

“…Several studies (Andreeva et al ., ; Calabrese and Osmetti, ; Calabrese et al ., ; Calabrese and Giudici, ) have shown that this model is very flexible and outperforms the logit and the probit models in predicting defaults. Moreover, if the sample is imbalanced with a low percentage of y =1, as in credit scoring models, the best link function is obtained for τ <0.…”

The effectiveness of TARP-CPP on the US banking industry: A new copula-based approach

“…Calabrese and Osmetti (2015) propose the BGEVA model by including an additive component to the GEV model. The BGEVA model has been then improved by Calabrese, Marra, and Osmetti (2016) .…”

“…However, the use of probit and logit link is not appropriate when modelling rare events. If we classify the rare events as ones, the major drawback of these approaches is that they underestimate the probability of binary rare events for values close to one, such as bank defaults or bail-outs ( Calabrese & Osmetti, 2015;King & Zeng, 2001;Wang & Dey, 2010 ). To overcome this limitation, a model has been proposed in the operational research literature ( Andreeva, Calabrese, & Osmetti, 2016;Calabrese & Osmetti, 2013;2015;Marra, Calabrese, & Osmetti, 2014 ) -the Binary Generalised Extreme Value Additive model, BGEVA (GEV model in the parametric form).…”