Systemic Approach to Risk Estimation Using DSS

Abstract

AbstractA concept is proposed for solving the problem of adaptive risk estimation that is based on the system analysis methodology and combined use of preliminary data processing techniques, mathematical and statistical modeling, optimal state estimation of the nonlinear nonstationary processes widely met in risk analysis and forecasting. The cyclical adaptation of a model structure and its parameters on the basis of a set of statistical criteria of the process being studied provides a possibility for computing high quality estimates of conditional variance forecasts under condition that data available is informative. To identify and take into consideration possible stochastic, structural and parametric uncertainties it is proposed to use optimal and digital filtering and the methods of intellectual data analysis such as Bayesian networks, adaptive Bayesian networks, various approaches to Kalman filtering, particle filters and other necessary instruments. Possible parametric uncertainties of the model being developed are minimized with application of several alternative parameter estimation techniques such as LS, RLS, ML and Markov chains Monte Carlo sampling. The models constructed are used for short term volatility (conditional variance) forecasting to be further used for market risk estimation. Bayesian networks allow for the formal description of other risk types such as operational, credit and actuarial risks, etc. The system proposed has wide possibilities for risk analysis application and further enhancement of its functional possibilities with new methods and computational procedures directed towards refining risk estimation and forecasting procedures.KeywordsRisk estimationNonlinear nonstationary processesUncertaintiesFilteringHeteroscedastic processesVolatility forecastingBayesian networks