Code for paper "Gradient-assisted calibration for financial agent-based models", which appeared in the Proceedings of the Fourth International Conference on AI in Finance. In this paper, we consider how agent-based models can be constructed in a differentiable manner, and how the differentiability of a differentiable agent-based model might be useful for accelerating certain (although not all) parameter inference procedures. See the INET Oxford YouTube Channel for a recording of a talk I gave about this paper.
To run this code, use python 3.10 and issue
python3.10 -m pip install blackbirds==1.2 jupyter==1.0.0 scienceplots==2.1.0 pygtc==0.4.1 tensorflow==2.14.0
@inproceedings{dyer2023a,
edition = {},
number = {},
journal = {},
pages = {},
publisher = {},
school = {},
title = {Gradient-assisted calibration for financial agent-based models},
volume = {},
author = {Dyer, J and Quera-Bofarull, A and Chopra, A and Farmer, JD and Calinescu, A and Wooldridge, MJ},
editor = {},
year = {2023},
organizer = {4th ACM International Conference on AI in Finance ((ICAIF 2023)},
series = {}
}