How Stochastic Finance Models Relate to Machine Learning

Summary

The document raises questions about the present and future roles of stochastic modeling in finance as machine learning tools become more capable. It considers whether the wider use of Black–Scholes may make it harder to find profitable opportunities by applying familiar pricing models, and asks whether machine learning can replace equation-based model building or whether the approaches can complement one another.

It also asks what stochastic calculus contributes to modern financial practice, whether hybrid stochastic and machine learning methods exist, and how these tools may evolve over the following decade. The text offers background and motivation but does not answer these questions, present a strategy, or provide empirical evidence. Its claims about the declining usefulness of familiar models are posed as a concern rather than established findings, so readers should treat the document as a research agenda rather than guidance on model selection or trading.

Key ideas

• The document questions whether familiar pricing models still offer exploitable trading opportunities.
• It asks whether machine learning can substitute for stochastic equation-based models.
• It raises the possibility of combining stochastic modeling with machine learning.
• The discussion poses questions but supplies no answers or empirical evidence.

Tags

# stochastic modeling and machine learning


# stochastic modeling and machine learning












For a little bit of background, I've been studying stochastic calc and a few of it's applications (currently I'm still at the early stages of learning applications) and have been curious as to whether or not trading strategies using stochastic modeling are still relevant in the modern day age (late 2017 as I'm writing). One example might be: seeing as the familiarity with Black-Scholes has grown so much over that past ~30 years, using it as a strategy to find and capitalize off of what a trader might deem a mis-pricing no longer seems do-able (or at best, has become extremely difficult) due to it's popularity (i.e. since everybody has been trained on how to apply it/where to use it, hence it's become less effective at doing what it was able to do in the past).

My main questions are as follows: has the evolution of tools in Machine Learning replaced the role of model building using stochastic equations (can more complex models built from ML perform the same role as stochastic modeling without the uniformity of logic/formulae/precise derivations of models)? What is the main role of present day stochastic modeling in finance? Are there any hybrid stochastic machine learning models/methodologies? What role will stochastic calc/modeling play in finance within the next 10 years or so?

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.