Can DeepSeek-R1 Price Options? An Experiment in LLM-Driven Derivatives Analytics

Hypothesis

Large language models can generate correct financial code. But can they reason about stochastic calculus and produce correct pricing for path-dependent options?

The Test

We gave DeepSeek-R1 three progressively harder tasks:

1. Implement Black-Scholes with Greeks (closed-form)
2. Price a barrier option via Monte Carlo with antithetic variates
3. Calibrate a Heston model to market swaptions

Results