High Performance Computing and Economic Scenario Generation: Integrating Expert Forecasts into Plane Price Modeling

The problem at hand is the integration of expert forecasts for plane prices into a fully calibrated basic economy. The economy is simulated through an Economic Scenario Generator (ESG), which includes macroeconomic processes, interest rate term structures, etc.. By defining the available best-case, worst-case, and mid-case forecasts to correspond to the 95%, the 50% and the 5% quantiles of the plane price distribution, one could describe the problem with the following optimization setting: min (βc,αc,i,δc′,σc,i) {‖ ( Q ( (ĨT,S) P̃ c,i,T,S(βc, αc,i, δ c, γ c,σc,i),q ) − F̃M ) W‖F } The tilded matrices represent simulation results, i.e. they have the dimension timesteps T and scenarios S. The function Q(M̃T,S ,q) : R ·S × [0, 1] → R ·q is mapping a matrix M̃T,S of simulated scenarios with dimension (T × S) onto each timestep’s quantiles q , resulting in a matrix of dimension T × q. FM is the (T × q) matrix of expert forecasts, and W is a (T × q) weighting matrix. ‖‖F denotes the Frobenius norm, and is the element-wise multiplication. The economy simulation is computation-intensive, for which we take benefit of using GPGPU techniques. The optimisation part is also a high-dimensional computationintensive problem, for which we use a natural computing approach using Differential Evolution.