Econometric Risk Modeling of Evolutionary Systems – ERMES

The proposal is a fundamental research project on the econometric modeling of multivariate, macroeconomic time series. Its global aim is to provide new measures of risk that might be useful for the evaluation of financial stability. Why is it necessary to develop new risk measures' Because traditional measures of risk recently appeared to poorly capture the complexity of financial macroeconomic systems, which are dynamic, nonstationary and interdependent by nature. In this project, we propose to use our expertise in econometric theory and in the statistical theory of stochastic processes in order to expand new macroeconometric models that are better fitted to the complex structure of the market. The recent worldwide crisis of the banking system showed the limitation of the basic macroeconomic models and their associated econometric methods. Very often, models used in risk control are too simple approximations of the underlying complex market mechanisms. Among the complex stylized facts of modern markets that are hardly addressed in econometric models, this project focuses on three interlocking aspects: (i) the stability of the structural models over time, (ii) the appearance of rare events, and (iii) the ability to capture endogenous risks. These three aspects will be discussed in Sections 2 and 3 below. They appear as three global directions and they define our quantitative economic objectives. However, in order to reach those goals, the core of the project will be concerned by the technical aspects given by the econometric theory and statistical challenges that are behind those three tasks. The inherent parameter instability of the observed time series makes necessary the development of statistical tools to detect nonstationarity in a time series (here, the word nonstationarity is understood in the general meaning of a time series with a time-varying correlation structure). The existence of extreme events and correlation between those events leads questions on how to do inference under nonnormal distributions including inference on stochastic processes with extreme values. Finally, modeling endogeneous risks synthesizes the two problems and calls for the development of multivariate models of nonstationarity with tail dependences.