How to extract downturn from a period without downturn?

Details

Supervisors

Saerens, Marco

Faculty

Louvain School of Management

Degree label

Master [120] en ingénieur de gestion

Abstract

Over decades, the current world has experienced different financial crises. The reliability of banks becomes more and more ubiquitous for the overall economic stability according to the European Central Bank. Anticipating and foreseeing sufficient capital provisions is crucial to be prepared to distressed impact leading to a significant loss increase. While the partner company of the thesis AXA Bank Belgium (ABB) has not experienced any downturn for the last 20 years, the challenge of the thesis is to extract downturn from stable macroeconomic data in order to fulfill the European Banking Authority requirements. As house price index is determinant in the economic activity and as the mortgage loans represent the largest part of the ABB total assets, it is used as an input to estimate the expected credit losses from mortgages in the estimation of the Loss Given Default (LGD). Univariate time series models are employed in order to achieve an appropriate fit: Holt's Linear method, Linear regressions, ARIMAs, DLMs. The best univariate appears to be the DLM. Then, multivariate models are reviewed: VARs, VECMs, BVARs, GARCH models. Among those, the best one is the VAR model. A machine learning algorithm, the Neural Network, is also investigated. Theoretical Downturn models are drawn either from DLMs or from VARs, considering the left tail of 95\% prediction interval for each macroeconomic variable. As a result, these models are used to estimate the expected losses in downturn on the ABB internal data. The impact on house price index helps to stress the Loan To Value (LTV) ratio and to estimate LGD leading to compute the expected credit losses.