Modello LGSR forward looking

Abstract

In this work, we propose a hierarchical model to introduce Forward-Looking effects on the Loss Given Default Rate (LGDR) estimate, as required by IFRS9. The Framework consists of two modules: a SURTS satellite model (Seemingly Unrelated Regressions Model Time Series), which analyses the dynamics of the systemic LGSR (bad loans LGDR) and a set of selected macroeconomic factors, and a Beta Inflated-(0,1) model which estimates the LGSR for the single entity. The basic hypotheses for the construction of the hierarchical model will also be illustrated, underlining how this approach is particularly relevant for LSIs (Less Significant Institutions). The theoretical aspects are followed by an application on a series released by the Bank of Italy, presenting the LGDR estimation process on an archive of closed bad loans by a set of banks belonging to the CABEL (ICT Service Provide) network. By way of example, we illustrate the forecast results for the three-year period 2022-2024 for the systemic LGDR. Other aspects related to the construction of LGDR models are addressed, such as the segmentation of the portfolios and the selection of individual attributes. In particular, we introduce the NPL vintage as an explanatory variable in the LGDR model, outlining the interconnections with the effects of macroeconomic projections.

Keywords

• lgsr forward looking