From Theoretical to Applicable to AI: A Walk Through Quant Finance and Education with Baruch’s Professor Andrew Lesniewski

Andrew Lesniewski began his career in pure mathematics, completing early work on quantum field theory (CMP, 1983; 1987; 1989), before turning toward quantitative finance. He went on to co-author the SABR volatility model, serve as Head of Financial Engineering at the Depository Trust Company, lead Quantitative Research at Ellington Management, and lead the Fixed Income Quantitative Research at BNP Paribas. Today he is Professor of Mathematics at Baruch College and Curriculum Director of the school’s Master’s in Financial Engineering (MFE) program.

From Pure to Applicable Math

Professor Lesniewski describes his career as a journey from the most abstract mathematics to methods that directly shape financial markets.

“I’ve always thought of myself first as a mathematician. What I try to do is mathematics at the best level it can be done — but applied to real situations.

Pure math is fascinating and important. Applied math is also theoretical, people prove things, but it isn’t always driven by real problems. What I call applicable math is different: it’s math that turns out to be directly useful. That’s what happened with the SABR volatility model, with the LIBOR Market Model, with mortgage prepayment and credit models. These are examples of math that moved from theory into tools that people actually use.”

Generative AI in Finance

That same mindset — making abstract models useful — informs Lesniewski’s current focus on generative AI. He believes it is already beginning to change the way finance handles data.

“A more recent example of applicable math is generative AI. These are fairly straightforward models, but they’re used in very smart ways, trained on huge amounts of data, and they’ve exceeded expectations in many areas.

One project I did last year was about generating synthetic market data. Normally, for financial models, you simulate data with Monte Carlo — you pick a distribution for returns and sample from it. But real markets don’t follow neat distributions. Generative AI lets you skip that step. You give it real data, and it generates new market data that looks the same statistically, passes the tests, but isn’t just resampling.

That’s very powerful for risk management, portfolio optimization, and strategy backtesting. Actual historical data may not give you enough scenarios to work with. Synthetic data gives you a much bigger background for stress testing. And it’s creative in a way — it produces patterns you didn’t feed into it directly. Some of them are wrong, sure, but many are useful enough to explore.”

Creativity and Risk

For Professor Lesniewski, the significance of AI goes beyond efficiency. He sees genuine creativity emerging from the models, though not without risk.

“I don’t agree with people who say AI only reproduces what it was trained on. In my research I’ve seen it come up with approaches I didn’t expect. At first I thought, ‘No, that’s wrong,’ and then I checked and it was right, or at least pointed me in a direction worth following.

That’s creativity in a practical sense. It’s not exotic mathematics — at heart it’s probability, predicting the next word or number — but the scale, billions of parameters, is what gives it the surprising behavior.

At the same time, you need supervision. Someone has to be able to turn off the system if it’s going the wrong way. Governance and validation are critical.

And the old truths of finance don’t go away. Alphas are short-lived. Once others find the same edge, it disappears. AI may help find edges faster, but it doesn’t stop them from being crowded out.”

Teaching and Student Projects

As a teacher, Lesniewski sees technology like AI and paper trading platforms as tools for students to learn how models play out in real markets.

“Almost everything I teach is in the MFE program. I don’t run trading classes, but I supervise a lot of master’s projects. Some students want to work on high-frequency strategies. I recommend they use paper trading to backtest. I’ve had several students do this already, and I expect more this semester. The Interactive Brokers Student Trading Lab platform has been very useful for that. I appreciate the access to paper trading environments.”

Career Opportunities and Curriculum

Professor Lesniewski also notes how market cycles shape both opportunities and curriculum. When interest rates were near zero, fixed income was quiet. As rates rose, students rediscovered bonds while also diving into equities and the buyside.

“There’s a change in the trend over the past fifteen years. I started as a fixed income guy, but when rates were near zero or negative, fixed income became very quiet. As rates went up again, that area came back to life. At the same time, students are very interested in high-frequency trading, equities, and the buyside.

As a program, we try to keep the curriculum relevant. We added courses on crypto when that was emerging, though with the scandals it’s less central now. We have NLP courses — useful for alternative data, but also for fraud detection. Scanning millions of documents is hard, and algorithms help. Next, we’ll add a course on generative AI.”

Programming Languages

In the classroom and in his research, the Professor continues to press toward a finance shaped not just by elegant theory, but by methods that prove themselves in the real world – and in the real world, no MFE discussion is complete without programming. Lesniewski emphasizes the rise of Python, noting how community support and libraries pushed it past rivals.

“Python has become dominant. MATLAB is too expensive and has lost support. R is very good for quick desktop-style tasks — ten minutes and you get an answer — but it’s hard to embed in production. C++ is still extremely efficient, but it’s difficult.

Python hit the sweet spot. Open source, lots of community support, packages like pandas for data, NumPy and SciPy for numerics, PyTorch and TensorFlow for AI. Pandas in particular was a turning point — it let people manipulate data almost like a database. That, plus the move from Python 2 to 3, is when Python really locked in — for now.”