دانلود رایگان مقاله انگلیسی پیش بینی ورشکستگی با استفاده از ترکیبی از مدل خوشه بندی و MARS - اشپرینگر 2018

Abstract

In this paper, we compare the performance of two non-parametric methods of classification and regression trees (CART) and the newly multivariate adaptive regression splines (MARS) models, in forecasting bankruptcy. Models are tested on a large universe of US banks over a complete market cycle and run under a K-fold cross validation. Then, a hybrid model which combines K-means clustering and MARS is tested as well. Our findings highlight that (i) Either in training or testing sample, MARS provides, in average, better correct classification rate than CART model (ii) Hybrid approach significantly increases the classification accuracy rate in the training sample (iii) MARS prediction underperforms when the misclassification of the bankrupt banks rate is adopted as a criteria (iv) Finally, results prove that non-parametric models are more suitable for bank failure prediction than the corresponding Logit model.

Conclusion

In this paper, we developed a blend model based on two non-parametric classification models to study the bankruptcy of US banks. We provide a comparative approach between CART, MARS and K-means-MARS. Our main objective is to predict bank defaults some time before the bankruptcy occurs, and to build an early warning system based on CAMEL’s ratios. We based our empirical validation on a large panel of US banks gathered from both Bankscope and from the Federal Deposit Insurance Corporation. The main contribution of our paper with regard to the existing literature is twofold:

– Methodological and conceptual: First, we propose, for the first time, a hybrid model that combines K-means and MARS models. We provide a comparative framework not only to non-parametric models but also to parametric models Logit and CDA (Affes and Hentati-Kaffel 2017).

– Empirical validation:

(i) Our study focuses on a large sample of US banks with different size. The paper analyses the behavior of banks over a 6-year period rich in events (it encompasses tow sub-periods, stress period 2008–2009 and a recovery one 2010–2013).

(ii) The comparative approach highlighted the supremacy of the proposed hybrid model in terms of accuracy classification for both training and validation samples.

(iii) The model enhanced the classification accuracy by 1% for the training sample

(iv) We established that MARS underperforms, by the misclassification rate of the bankrupt banks, notably for 2008 and 2009. Also, according to the Area under Curve (ROC), MARS model showed better accuracy results than CART model

(v) The results differ from 1 year to another, but a general behavior for all distressed banks could be conducted. CART classification shows that among the 10 tested ratios, the most important predictors are capital adequacy variables. Also, we note that the asset quality ratios (NPLTA) and (NPLGL) are much more important than the other two components (LLRTA) and (LLRGL). According to MARS the most important variables was also the components of the capital adequacy. The Liquidity variables (TLTD and TDTA) have an importance in detecting bank failure only in 2010 and 2013. We note that, with respect to parametric models (see Affes and Hentati-Kaffel 2017) the asset quality was also an important component to explain the financial conditions of banks (except for 2009 and 2010).