Essays on credit quality and macroeconomic environment: non-linear models and forecasting

In all countries, the banking system takes a central role in the economy being the key engine of growth. Financial crises, especially the turmoil of 2007 have underscored that, independently of the stage of development, all countries are susceptible to shocks and understanding the channels between credit risk and business cycle is crucial for evaluating banks robustness and system stability. Actually, every weakness of the sector and the deterioration of bank’s loans quality arises episodes of costly banking system distress, correlated economic crisis, and damage growth prospects. As soon as banks stop functioning normally, they cannot longer provide credit to the economy. Therefore, a prompt recovery is impossible without the awareness and a good understanding about the drivers of credit risk necessary to implement the suitable resolution instruments. Generally, linear models have depicted the relation between credit quality and macro environment, but the strong dependence of bank loan default on the economic cycle, subject to changes in regime, could suggest nonlinear approaches as more appropriate. In this regard, the broad object of this thesis is to extend and apply nonlinear models to this relation by developing three self-contained chapters Chapter 1 lays the groundwork for the next two by reviewing non-performing loans literature along with the literature on Markov Switching and variants model, discussing properties and estimation procedure as well. The work outlines a critical presentation of the extant ways to estimate the nexus between economic activity and asset quality, individuating some critical lacuna as regard the methodological procedures. Findings suggest that the existing literature considers diverse sample, geographical area and time framework but fail to address regime shifts in the data and nonlinearities. Specifically, it recognizes the shortcomings of linear methods and a slightly lower number of applications deviated from those approaches. Although a growing empirical literature is using threshold models, there lack empirical works on the determinant of bad debts that employ the Markov regime-switching framework suggested as an appealing research tool to such demand and as a cross-validation method for the robustness of the results. Chapter 2 focuses on the validity of the Markov switching framework in USA. It attempts to model and forecast three different kinds of bank loan default, detecting non-linearity and asymmetries in their relations with macro variables by the adoption of a Markov Switching approach. By comparing this specification with a classical linear model, empirical results lend support for the validity of the non-linear model in capturing the presence of regimes and asymmetries, changing in correspondence of the major recession periods spanning from 1987 to 2017. Moreover, it gives evidence of a clear outperformance of the Markov switching concerning the linear counterpart, both in modelling and forecasting. Finally, chapter 3 studies the evolution of correlation between total US delinquency loans and macro variables using the Dynamic Conditional Correlation model (DCC). To our knowledge, it makes the first attempt to provide a time varying analysis of correlation in our area of research. Results document that the dynamic correlation does not increase greatly during financial turmoil except for few variables and in lagged terms. Likewise, it details how correlations change over different time of crises (more specifically the Saving and Loans crisis in 1989 and later, the Dotcom in early 2000s and the Subprime crisis in 2007-08).