Data-Driven Greeks and Market-Path Simulation Project Ideas

Summary

The document responds to a student seeking manageable machine-learning project ideas with two research directions from quantitative finance. The first is deep hedging for risk management, along with data-driven estimates of option Greeks such as delta. Instead of relying entirely on a pricing model, the suggestion is to estimate sensitivity from observed market data and investigate whether that can support hedging or risk decisions.

The second direction is to use data to build more realistic simulations of spot-price paths, with the aim of improving on simple stochastic-volatility descriptions and supporting trading-strategy research. The answer points to the importance of market regime shifts, but offers no implementation plan, dataset, evaluation design, or empirical results. Both ideas are ambitious research themes rather than ready-made beginner projects; the response does not assess their difficulty or establish that data-driven approaches outperform model-based methods.

Key ideas

• Deep hedging is proposed as a machine-learning direction for derivative risk management.
• Observed market data could be used to estimate Greeks such as delta for hedging applications.
• Another proposed direction is to simulate spot paths that better reflect the market’s data-generating process.
• Regime shifts are identified as a challenge for realistic market simulation.
• The response offers research themes without datasets, evaluation methods, or evidence of performance.

Tags

# What are good machine learning projects for a senior student?


# What are good machine learning projects for a senior student?












I'm a senior year student, I study software engineering, I recently started with a machine learning tutorial, I want a machine learning projects that are not hard to make and at the same time good for the community, and people are willing to use it.

## Answer by StackG (score 1)

https://quant.stackexchange.com/a/58260

Two thoughts that I'm interested in at the moment...

- A Deep Hedging-stype approach to risk management (eg. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3355706)

Fundamentally all of derivative pricing quant finance is Model-Driven. Why? Probably mostly because it's easier. However, we now have maybe 30 years of high frequency tick data available globally for many exchange traded instruments, can we come up with data-driven Greeks models (eg. Delta is just the change in price when the underlying moves, why not measure it directly instead of assuming a model?) For risk management and hedging?

- Better understanding of the market's Data Generating Process (aka. the holy grail)

Can we use data to simulate spot paths that are better than 'stoch vol with x vol-of-vol'? If you could, then all of trading strategy research becomes easier. Of course, this involves understanding market regime shifts etc...

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.