Skills Gained (Andrew Green)
- Develop a new XVA platform
- CUDA GPU / C++ (11) CPU Hybrid Compute environment
- Grid Computing
- Adjoint Algorithmic Differentiation (AAD): CPU/GPU implementation using NAG dco/dco_map
- Multi-asset platform: IR & FX, later inflation, commodities and equity derivatives
- Deep Learning applications in XVA
Reduced memory usage and batched Monte Carlo simulation to be suitable for CUDA acceleration [02/06/2020] See Method MonteCarloSimulation::generatePath in header simulation.h
Use boost function boost::math::erf_inv to generate gaussian variates based on a merssene twister uniform random generator [02/06/2020]
Use Intel MKL to generate gaussian variates based on a merssene twister uniform random generator [07/06/2020]
Fixed The Exposure Calculation profile for a given Interest Rate Swap Mark to Market [09/06/2020]
Added new dissertation notes with the Reference used during the study [09/06/2020]
Finished MonteCarlo Convergence for IRS Expected Exposure with a 0.01 accuracy and 4750 simulations on each simulation point.[14/06/2020]
TODO - Estimate the stddev and confidence interval for simulations
TODO - Apply a variance reduction method on HJM model for risk factor evolution to speed up the MC convergence
TODO - Transpose the Exposure matrix and check for memory bandwidth usage improvement
TODO - Modify YieldCurve calculation DF(t, T) and ForwardRates F(t;t, T). Data must be accessed using a linearized array
TODO - Implement a CUDA Accelerated Kernel for HJM simulation
TODO - Benchmark the solution Measure Performance
TODO - Support negatives Interest Rates
TODO - Perform sensitivity Analysis with AAD