Break Risk – BREAKRISK

The global economy is subject to large and sudden changes in regimes, as evidenced by the severe crashes in the financial system during the last two decades, including the global financial crisis of 2008. A distinctive feature of the last crashes it that these are episodes during which previously unrelated financial markets can become extremely dependent, which makes the global economy very risky for all stakeholders. A better understanding of such episodes of the economy is hence of primary importance.

In this research program, we develop and implement quantitative tools that explicitly take into account the sometimes abruptly changing nature of the economic environment. In particular, our research agenda centers on three axes that are central to our
understanding of risk in times of crises.

The first axis focuses on models in which the economy can oscillate between several regimes, characterized by different laws of motion, as well as models of structural change. These models reflect the notion that the behavior of the economy may differ across periods of crisis and calmer periods. Existing regime switching models have the drawback that all parameters change when there is a regime switch.

First, we propose a methodology for multiple breaks, where we relax the assumption that all parameters need to change when there is a structural break. This is done by shrinking irrelevant break-parameters towards zero. Estimation is based on a sampling scheme tailored to change-point and Markov Switching models. Second, we develop a framework for off-line multiple change-point problems in one-parameter exponential family observation-driven time series models. Third, we propose a monitoring procedure for sequential change-point detection for integer-valued time series. Finally we check for the presence of breaks in an extended version of the heterogeneous autoregressive model of realized volatility and we assess the model's in- and out-of-sample performance.

Our second broad topic is models of correlations in default risk, whose deficiency was at the epicenter of the subprime crisis. We propose two new bottom-up models of credit risk. The first tackles the default dependence of securitized subprime mortgages. On top of a classical multinomial logistic framework, we use composite likelihood to estimate a pairwise copula model of default, where dependence is a function of physical distance between mortgages, pairwise averages and differences in mortgage characteristics, as well as interest rates and past local default rates. The second introduces default correlations in the forward-intensity model of \citeasnoun{duan2012multiperiod}, which predicts corporate defaults over multiple horizons and is the methodological basis for the non-profit credit rating initiative at the National University of Singapore that provides default forecasts for 60000 firms worldwide. Finally we propose a multivariate equidependent copula regime-switching model to analyze how dependence in local house prices in the U.S., and thus the benefits of geographic diversification of real estate portfolios, change with conditions on the credit market.

Finally, we analyze the behavior of the risk premia that investors require as a compensation for carrying variance risk. First, we propose a joint regime-switching model of realized and risk-neutral variances in a state-space framework with three regimes, designed to capture heteroskedasticity, jumps, and time varying dependence between the variance risk premium, the risk neutral and physical variances. Second, we propose a particle-based Bayesian method to estimate a very general model of dynamic volatility with both underlying and derivative observations, that encompasses many specifications that have been proposed in the literature. Finally, we will build a state space model for the joint dynamics of the 'efficient' price as well as for a set of parameters that capture the shape of the limit order book,