Jatin Lamba

Quantitative Research & Development Intern · New York

When a company reports earnings, the news often moves the stocks of related companies too: its suppliers, customers and competitors. I built models that predict which companies will be most affected when a peer reports, and by how much, using data on how businesses are connected and how their stocks have moved together in the past.

Tested on more than two million historical earnings events, the model's predictions tracked what actually happened far better than looking at past behavior alone. It is used to help decide how much to invest ahead of earnings announcements.

Quantitative Researcher, previously Quantitative Developer

Edge Focus uses data and machine learning to decide which loans to invest in. I joined as one of the first five people and stayed as it grew from $50 million to over $2 billion in assets, earning investors 10–12% annual returns.

I started by building the firm's trading systems and data infrastructure, then moved into research. My main work was forecasting how each investment would perform over its life, and deciding in real time which of roughly 30,000 daily opportunities to take on with a limited budget. The approach I designed turns one large, portfolio-wide decision into a simple yes-or-no test for each individual investment. I also stress-tested our models against past recessions and other economic shocks.

MIT Sloan, with Qube Research & Technologies

How much will the prices of oil, metals and other commodities swing tomorrow, and how will they move together? We compared simple and sophisticated forecasting methods across eight commodity futures and found that a simple, well-tuned approach was as accurate as the complex ones while giving steadier forecasts.

Adobe Research · NAACL 2021 · U.S. patent

Most question-answering systems reply with text only. We proposed a system that reads a document and answers with both a passage and the most relevant image, which people found easier to understand.

Bachelor's thesis, IIT Bombay · Interspeech 2021

A model that detects events in a video, such as a dog barking or a car passing, and tells whether each one is seen, heard, or both, while learning from only rough labels.

TU Braunschweig, Germany

A tool that automatically rewrites high-performance code, where many threads share the same data, so it cleans up memory safely without slowing down.

Master's in Financial Mathematics, Financial Engineering concentration

Coursework in econometrics, optimization and quantitative investing. Side projects include forecasting U.S. inflation from its components (28% more accurate than a standard baseline) and studying how the way a model is trained changes its ability to predict which stocks will outperform.

B.Tech with Honors in Computer Science and Engineering

Admitted with an All India Rank of 33 among 1.2 million candidates. Teaching assistant for programming and logic courses.