Capturing time-varying financial contagion: the case of Change-point and Markov-chain volatility models

In finance, risk measures like Value-at-Risk and volatility are essential for
taking decisions and managing risks. These measures depend on forecasts
from econometric models, the parameters of which are typically estimated
using historical data. However, most of existing models ignores the impact
of abrupt changes in the series and this may lead to bad forecasts. In
addition to that, taking into account structural breaks are economically
appealing since one can investigate why these switches occur. The project
will make two empirical contributions to the financial literature by developing
two statistical processes able to take into account abrupt changes in the
parameters.
The first study will model the daily realized variances of indices with a
change-point Vector AutoRegressive (CP-VAR) process. Understanding the
realized variance dynamics is crucial for risk management. The project will
provide results on turning points and study their relationships with economic
data. Additionally, the CP-VAR model innovates by being able to detect
which series dynamics share the same parameter set. This leads to two
relevant improvements over the current models: i) it solves the overparametrization
issue of CP-VAR models and ii) it highlights which series
are leading the others. The latter feature stands for a new statistical tool for
inferring networks.
Secondly, the project will focus on a systemic risk measure, called SRISK.
In fact, the SRISK indicator builds a ranking of the riskiest financial
institutions in the sense that they will need the highest amounts of capital if
the financial market collapses. Nevertheless its estimation requires accurate
long-term volatility predictions. The second study will develop a bivariate
volatility model that produces accurate long-term forecasts (thanks to a
mathematical result on Markov-chains). Ultimately, I will compute SRISK
measures of many European financial institutions and rank them according
to their systemic risk.