The aim of this repository is to translate few standard PDE problems of economics and finance (HJB; Kolmogorov-forward/Fokker-Planck) to Physics-informed neural networks. There will be a total of 6 Matlab examples.
- Introductory lecture on Pinns
- 6 benchmark models, implemented in Matlab
- 01: Partial equilibrium model, explained in this note here
- 02: Partial equilibrium model with discrete choices, explained in this note here
- 03: General equilibrium model; HJB plus Kolmogorov Forward, explained in this note here. Honestly Ben’s numerical appendix I sent above is perfect documentation of model 3. Read sections 1-3
- Replicate all the results from the Matlab benchmark with high accuracy with PINNs.
- Carfully check what boundary conditions work fine (hard/soft boundary conditions).
- Also look at BSDE solver methods, and solve models also with this deep learing approach. See also this notebook.
- Look at the website of J. Han.
- Keep in mind that we want to build up know-how so we can add dimensions later to those models; boundary conditinons should not suffer from the curse of dimensinality.
- IMPORTANT: we are interested accurate policies, that is, the slope of the function of the problems.