Title: Financial networks via conditional autoregressive expected shortfall
Authors: Giovanni Bonaccolto - University of Enna Kore (Italy) [presenting]
Massimiliano Caporin - University of Padova (Italy)
Abstract: Several approaches have been developed to build financial networks, emphasizing the distress state of the involved companies to assess their systemic relevance. Building on this strand of the literature, we propose a new method that connects the expected shortfalls of $N$ financial institutions. In particular, the network structure is retrieved from $N$ expectile regression models, each of them is calibrated to estimate the expected shortfall of a company at the level $\theta$ conditional to the expected shortfalls of the $N-1$ remaining institutions. Our method is designed to deal with large values of $N$, as we impose an $l_1$-norm penalty on each expectile regression model, an effective tool that allows us to select the relevant connected institutions. In our model, we capture the temporal persistence of expectiles by including latent autoregressive components. As a last exercise, we interpret the obtained network as a financial portfolio and compare it with the optimal portfolio with the minimum expected shortfall. We implement our method on the STOXX 600 financials constituents, highlighting interesting empirical findings, especially during distress periods.