How machine learning is being used in quantitative finance

Ben Lorica

The Data Exchange Podcast: Arun Verma on alternative data sources, machine learning, and model risk management.


Subscribe: iTunes, Android, Spotify, Stitcher, Google, and RSS.

In this episode of the Data Exchange our special correspondent and editor Jenn Webb speaks with Arun Verma, Head of Quantitative Research Solutions at Bloomberg. My first job post-academia was as lead quant in a small hedge fund. Since then, I’ve followed the industry from afar and I’ve long been interested in the role of data and models in financial services. Arun and I discussed quantitative finance when we ran into each other at the O’Reilly AI conference in London last year. He was slated to give a talk on extracting trading signals from alternative data sets, an important subject among quants.

Ray Summit has been postponed until the Fall. In the meantime, enjoy an amazing series of FREE monthly virtual conferences, the next one is scheduled for June 10th: The Road to AutoML – Hyperparameter Tuning and Neural Architecture Search. Go to anyscale.com/events to register.

Jenn and Arun discussed a range of topics including:

    • The quantitative finance landscape.
    • The challenges in identifying and using alternative data sources.
    • Applications of machine learning in finance, including deep learning and reinforcement learning.
    • New natural language models and their applications in finance.
    • Model Explainability and Model Risk Management.
    • Artificial General Intelligence.

(Full transcript of our conversation is below.)

Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.

Download a complete transcript of this episode by filling out the form below:

Short excerpt:
Jenn: ​Let’s start with a big picture question: how would you characterize the current state of quantitative finance?

Arun: ​It’s currently a perfect marriage of traditional quant finance and machine learning (ML) disciplines. The traditional quant finance is based more on prescriptive models—data models that postulate a theory, which you then prove. Now we are intersecting that with the data science world, which is kind of the opposite; it’s completely data-driven modeling. We don’t use any prescriptive modeling. We are essentially learning the models from the data itself. This is a very interesting time, where we have a good marriage of the statistics with the algorithmic modeling approaches.

In finance, we always talk about factors that drive the markets. It’s very good to have a conceptual theoretical framework where we talk about the physics of finance, essentially. But it turns out, finance is not science. So, it helps to flip the picture and forget about your theoretical frameworks and your assumptions that go into those theoretical frameworks. Instead, just learn the underlying structure of the behaviour of markets and market data purely from the data.

Jenn: ​Can you expand on some of the biggest challenges quants are facing in these areas? You mentioned data sources. What issues around alternative data sources are they facing?

Arun: ​It’s a very complex discipline. Quants have to know the financial models, the asset pricing models, all the complexities of different derivative instruments or financial data and so on, but they also need to know a lot about machine learning and data science these days. If that’s not enough, we also need to wear multiple hats in terms of being the data guy, being the technologist and, at the end of the day, being very good at communication. Communication is actually a key skill for quants these days because we are often working with very complex ideas, and it’s important to be able to distill those ideas down to something easy to understand and interpret by a business user. They need to know a lot beyond the quant finance world.

Related content:

[Image from PxHere ]