Constructing Scenarios of Time Heterogeneous Series for Stress Testing

Abstract

Heterogeneous global trends in asset prices and savings affect the macro economy. Our challenge is to use limited data to make inference regarding underlying causes. In general, government and business decision makers, FDIC type regulators and risk professionals need quantitative tools to help generate plausible scenarios of state-dependent and time heterogeneous nonstationary time series. We suggest using maximum entropy type bootstraps, recently implemented in an R software package called "meboot." A new modification of meboot divides the data series into blocks and can randomly modify the (down, at or up) direction of series within each block. Our large number of resamples are then available for construction of scenarios for probabilistic stress testing. A simulation study evaluates the performance of our proposal in the context of many types of time-heterogeneity showing that it behaves better than moving block bootstraps. We apply meboot tools to stress test inference regarding Granger-causality between asset prices and world savings rates, and also to the 'Value at Risk' used in Finance.