JP on Machine Learning and Trading at Jane Street

ZUSAMMENFASSUNG

In 30 Sekunden

JP, an equities trader in Jane Street’s Hong Kong office, describes how his work moves between model development, research discussions and live trading. He discusses the growth of machine learning infrastructure, low signal-to-noise ratios, evolving markets and the firm’s collaborative culture.

1A variable workdayJP alternates between refining models, discussing new approaches and checking model behavior during busy markets.
2Research infrastructure has grownThe scale of compute and the breadth of modeling techniques have increased since he joined.
3Finance has distinctive constraintsJP points to low signal-to-noise ratios and markets that evolve in response to a strategy.
4Collaboration is centralHe values colleagues sharing ideas and helping one another solve problems.

Das Wichtigste

A trading researcher’s work can range from small model improvements to monitoring live markets.
Jane Street has expanded its research compute and shared modeling infrastructure.
Financial machine learning must contend with low signal-to-noise ratios and markets that evolve as strategies are deployed.
Both modeling and infrastructure offer ways to contribute to machine learning work.

ThemenMachine Learning Fintech Infrastruktur & Deployment Company Culture

VOLLTEXT

Den bearbeiteten Text nach Kapiteln lesen

Transkript zeigen

Nur das Transkript

Trading in Hong Kong

JP describes his varying responsibilities and the pace of change in Asia. Outside work, he enjoys Hong Kong’s trails and beaches.

I’m JP, an equities trader at Jane Street since 2017. My average day is hard to describe because it varies: sometimes I focus on small improvements to existing models, and sometimes we have open-ended discussions on the desk about what approaches to try next.

When markets are busy, I spend more time following the day’s trading and checking whether our models are reacting appropriately to data that may be out of distribution. Asia is Jane Street’s fastest-growing region, and I see that in the pace of innovation and change as we tackle new problems in Hong Kong.

Outside work, Hong Kong is a great base for travel in Asia, but it is also an outdoorsy city. I enjoy trail running: there are hiking trails close to the office and beaches less than an hour away.

Scaling machine learning

Research compute and modeling techniques have expanded. Jane Street is building more CPU and GPU capacity and shared internal libraries.

Data and modeling have always been central to much of our work, but the scale of research infrastructure and compute has grown considerably. The range of modeling techniques we use every day is also very different from when I started.

Continuing to scale our machine learning infrastructure is a major priority. We are increasing the CPU and GPU compute available to researchers and building large internal libraries for sharing modeling techniques across regions and desks.

Financial models and collaboration

JP explains the low signal-to-noise ratio and changing market environment. He describes opportunities in modeling and infrastructure and the collaborative culture he enjoys.

Machine learning in finance has distinctive challenges. Financial problems tend to have a low signal-to-noise ratio compared with many problems in other fields, so we need to account for that when designing architectures. Markets also evolve against you, and you have to consider what changes when you implement a strategy in the market.

We are not necessarily looking for one profile or background for machine learning. We want smart, curious people interested in tackling these problems, whether on the modeling or infrastructure side; there are many ways to contribute.

My favorite part of working at Jane Street is the collaborative environment. People share ideas on the desk, help each other solve problems, and build toward what we will work on next.