Senior researcher on a systematic global-equities platform, conducting quantitative research across broad equity universes.

Graduate-level teaching in quantitative finance, data science, and systematic investing at Columbia IEOR and New York University Tandon. Details in the Teaching section.

Founding member of a mid-frequency systematic US-equities pod. Developed production alpha signals and proprietary datasets across a range of alternative data types.

Led teams of researchers and engineers delivering production-grade AI/ML systems across trading, research, and strategy. The work involved large-scale consumer-transaction data, forecasting and anomaly-detection systems, satellite-derived indicators, and other proprietary datasets.

Research at Lamont's Ocean and Climate Physics division with Richard Seager and Isla Simpson, on large-scale climate dynamics and monsoon variability.

Research with William Boos on monsoon dynamics, moist baroclinic instability, and atmospheric waves.

A graduate course I developed on the full research lifecycle behind systematic equity investing: defining prediction problems, constructing usable data, evaluating backtests, accounting for transaction costs, and translating research into investment decisions. Students work with institutional WRDS data to reproduce, update, and challenge published strategies. An organizing theme runs through everything: a large fraction of published trading strategies either no longer work or never did, due to capacity constraints, poor data, improper sampling, or data-handling errors. Understanding why is at least as important as being able to spot it.

Student final papers are published on arXiv. Papers from the Fall 2024 iteration:

• Refining and Robust Backtesting of A Century of Profitable Industry Trends
• Hunting Tomorrow’s Leaders: Using Machine Learning to Forecast S&P 500 Additions & Removal
• Productivity of Short Term Assets as a Signal of Future Stock Performance
• Volatility-Volume Order Slicing via Statistical Analysis
• S&P 500 Trend Prediction
• Market-Neutral Strategies in Mid-Cap Portfolio Management: A Data-Driven Approach to Long-Short Equity
• AI-Enhanced Factor Analysis for Predicting S&P 500 Stock Dynamics
• A Multi-Factor Market-Neutral Investment Strategy for New York Stock Exchange Equities
• An Application of the Ornstein-Uhlenbeck Process to Pairs Trading
• Parameters Optimization of Pair Trading Algorithm
• Enhanced Momentum with Momentum Transformers
• A Deep Learning Approach for Trading Factor Residuals

The course runs annually. Presentation slides from past iterations are available on request.

Graduate operations research covering linear and integer programming, network flows, dynamic programming, and stochastic processes, with an emphasis on modeling choices and quantitative problem solving.

During the PhD I taught undergraduate mathematics including Calculus, Analysis I and II, Methods of Applied Mathematics, and Ordinary Differential Equations. That experience shaped the precise quantitative communication I still try to bring to both research and teaching.

The methodological standard for systematic equity research is predictive power. That is insufficient. A signal can show strong historical information coefficient, survive an honest backtest, and still tell you almost nothing about the structural mechanism producing the excess return, or whether it will survive the transaction costs, risk contamination, and governance failures that end most live signals. These lecture notes provide the framework for asking harder questions.

The book is structured around fourteen questions, each naming a genuine decision point every systematic signal must pass through: What are you predicting? Can you trust the data? How much evidence genuinely survives model selection? Where does the signal’s edge actually live? What will it cost to trade? When should it be retired? These are operational questions with specific, checkable answers.

This book is a distillation of lecture notes I developed over more than a decade of teaching quantitative finance at Columbia University and New York University. That teaching was always conducted with the explicit knowledge and authorization of my concurrent employers. The material draws entirely on published academic and finance research. No proprietary strategies, signals, or firm-specific methods appear anywhere in it.

A practitioner-focused treatment of the full research lifecycle in systematic equity investing, from defining a prediction problem and constructing usable data, through factor modeling, backtesting, transaction cost analysis, and portfolio construction, to signal governance and research hygiene. The book draws on years of teaching. Its central argument, that predictive performance is necessary but insufficient for evaluating systematic strategies, runs through everything and motivates the methodological standards it develops in its place.

The starting point was something that bothered me about how practitioners talk about beta. The OLS regression of excess returns on the market factor is treated as if it identifies a structural parameter: the causal sensitivity of an asset to market movements. It doesn't.

The conventional estimator conflates causal exposure with correlation induced by omitted risk sources. Under textbook asset pricing assumptions the two coincide in expectation, so the distinction is easy to wave away. This paper works out when identification actually holds, when it doesn't, and what IV and graphical causal models can do about it.

Time-series forecasting reframed as an image-to-image regression problem: encode a series as a 2D image, train a model on such images, and predict the next one. Distributional forecasting falls out naturally. The method works well for cyclic patterns and is the precursor to the inpainting patent.

The idea: technical analysts read charts visually, so train a CNN to do the same thing. We encoded financial time series as candlestick images and found that standard convolutional networks recover complex, multi-scale trading patterns from the visual representation.

How orographically forced Rossby waves and gravity wave drag affect rainfall far from the mountains that generate them, including shifts in the tropical ITCZ and monsoon rainfall over Asia. Both wave types drive vertical motion through the omega equation, and the dynamical response is nonlocal in ways the literature had underappreciated.

Tests whether moist baroclinic instability explains monsoon-depression growth. Developing storms lack the required upshear tilt of potential-vorticity anomalies, suggesting greater similarity to tropical-depression spinup and leaving the intensification mechanism unresolved.

The MJO modulates stratospheric gravity waves generated by flow over the Tibetan Plateau. A remote connection: tropical convection phasing the vertical distribution of gravity wave momentum flux far into the extratropical stratosphere.

The observed decline in Indian monsoon depression frequency over recent decades turns out to be partly an artifact. Changing observational practices account for much of the apparent trend. I've thought about this finding often since, in contexts well beyond atmospheric science.

The second paper from the dissertation. Three mechanisms by which resolved and parameterized waves interact to drive the Brewer-Dobson circulation, and why the conventional additive decomposition of their contributions can be misleading even in an idealized model.

The central result of the PhD: perturb the parameterized gravity wave drag in a climate model, and the resolved Rossby wave response partially cancels it. Not by coincidence; by necessity. A stability constraint makes the Brewer-Dobson circulation more robust to gravity wave parameterization assumptions than the literature had suggested. Conventional sensitivity experiments may therefore mischaracterize the role of gravity wave schemes in stratospheric climate models.

A book chapter extending the master's thesis work to landfalling hurricanes, examining how aerosol loading affects both lightning activity and overall storm intensity as a cyclone makes landfall.

My first paper, from the Hebrew University master's. Aerosol effects on hurricane electrification and lightning, simulated with an explicit spectral bin microphysics model. High aerosol concentrations substantially alter where and how much lightning forms within a tropical cyclone.

A time series becomes a partial image with a blank region representing the future; an inpainting model fills in the pixels, yielding a distributional forecast. The natural extension of the image-to-image forecasting framework.

Summarizes application usage across a device using blurred or pixelated screen captures that log which applications are open without exposing document content.

The granted patent underlying the Trading via Image Classification research: converts time-series data into images and applies image classification to identify recurring patterns.

Uses historical data and Monte Carlo tree search to compute the probability that a user's financial goals are achievable and propose an optimal action sequence to achieve them.

Uses eye-tracking to identify and demarcate what human readers attend to in financial and regulatory documents, producing annotated versions that highlight key sections, tables, and clauses.

Filters subjective response uncertainty in ordinal survey data by constructing a simulated model that matches the observed response distribution with controlled variability.

Given a library of known signal patterns, converts both the library and incoming series into images and matches them to identify whether the current data contains any known pattern.

Two consecutive years co-organizing the competitions track. The goal was benchmark environments that practitioners would recognize as realistic financial problems, not just clean academic datasets.

A workshop on a question often left outside purely technical ML work in finance: how AI adoption affects financial stability, governance, and institutional decision-making.

Graph neural networks in industrial settings: finance, supply chains, knowledge bases.

Graph methods applied to financial data: firm relationship networks, supply chains, knowledge graphs extracted from SEC filings.

Model risk, regulatory compliance, fairness, governance: the parts of AI deployment in finance that technical papers rarely address.

Forecasting, anomaly detection, representation learning in finance.

• Frontiers in Quantitative Finance, Philadelphia, September 25, 2026
• Wolfe Research 10th Annual Global Quantitative & Macro Investment Conference, NYC, October 14–15, 2026
• Battle of the Quants, London, October 22, 2026
• Columbia MAFN Portfolio Management & AI Conference, NYC, November 5, 2026
• ICAIF 2026, Milan, November 14–17, 2026
• Neudata NY Winter Data Summit, NYC, December 3, 2026
• NeurIPS 2026, Sydney, December 6–12, 2026
• Future Alpha 2027, Javits Center, NYC, March 16–17, 2027
• Data+AI Summit 2027, San Francisco, June 21–24, 2027

• Practitioners’ Seminar, Columbia MAFN, Wednesdays 6:10–7:25 pm, 207 Mathematics Building (Fall 2026)
• Financial Engineering Practitioners Seminar, Columbia IEOR, Mondays 7:00–9:00 pm, Schermerhorn 501 (ongoing)

• Bloomberg-Columbia ML in Finance Conference, NYC, September 2026
• Cornell / Rebellion Research AI & Finance Conference, NYC, September 2026
• Future Alpha (Quant Strats) 2026, NYC, April 2026
• ICAIF 2025, Singapore, November 2025 (Senior Program Committee)
• Quant Strats 2025, NYC, March 2025
• NeurIPS 2024, Vancouver, December 2024
• ICAIF 2024, NYC, November 2024 (Competition Chair)
• Frontiers in Quantitative Finance 2024, Philadelphia, September 2024
• Bloomberg-Columbia ML in Finance Workshop, NYC, September 2024
• ICML 2024, Vienna, July 2024
• Data + AI Summit 2024, San Francisco, June 2024
• ICAIF 2023, NYC, November 2023 (Competition Track Organizer)

Recognition for sustained invention activity at JPMorgan. Translating research into patent applications became a useful complementary discipline: it requires stating with unusual precision what is technically new and how it differs from prior approaches.

Selective fellowship for data scientists and researchers transitioning from academia to industry.

Presented at the SPARC General Assembly for the wave-mean-flow work.

Early-career award for the same cluster of work, from SPARC's Southern Hemisphere program.

Research fellowship at UCLA's Institute for Pure and Applied Mathematics during a long program on geophysical fluid dynamics.

NYU's main PhD fellowship, covering four years of doctoral study at Courant. It made the PhD possible.

Fellowship to attend the 15th International Conference on Clouds and Precipitation, Cancun, where I presented work from the master's thesis on aerosol effects on hurricane electrification.

Coverage of a student-run competition I organized at ICAIF on financial reinforcement learning. Several participants ended up with internship offers at Optiver, IMC Trading, and Capital One.

A piece on quant researchers at banks and hedge funds who hold academic appointments. I was one of the examples.

A profile of JPMorgan AI Research at the time I was running part of it: who we recruited and what the research culture was like. The hybrid model of doing publishable academic work inside a major bank was still unusual enough to be interesting to journalists.

Coverage of the group growing to ~50 people. The piece got the general picture right: it was a real research organization, not a PR operation.

An interview about our research on Indian monsoon depression trends: whether the apparent decline was climate change or bad data. Published during my research stay at Yale.

Doctoral work at NYU's Courant Institute, in the Center for Atmosphere Ocean Science, at the intersection of applied mathematics, geophysical fluid dynamics, and large-scale numerical simulation. Advisors: Edwin Gerber and Oliver Bühler.

The dissertation, What Drives the Brewer-Dobson Circulation? Wave-Mean-Flow Theories, Interactions Between Resolved and Unresolved Waves, and the Limits of Downward Control, established that large-scale Rossby waves and parameterized gravity waves are coupled in their joint driving of the stratosphere's global overturning circulation. The central result: changes in parameterized gravity wave drag induce a compensating response in the resolved Rossby wave field, a consequence of a fundamental stability constraint. This coupling makes the Brewer-Dobson circulation more robust to parameterization assumptions than the literature had recognized, and reframes how sensitivity experiments involving gravity wave schemes should be interpreted. Two papers in Journal of the Atmospheric Sciences: Compensation (2013) and What Drives the BDC (2014).

A master's in Alexander Khain's group, focused on cloud microphysics and its interaction with atmospheric dynamics. The thesis used the Hebrew University Cloud Model (a spectral bin model with explicit droplet and ice particle size distributions) to simulate how aerosol loading affects hurricane electrification and lightning. High aerosol concentrations substantially alter where and how much lightning forms within a tropical cyclone. Published in Journal of the Atmospheric Sciences, 2008.

The work introduced me to high-performance numerical simulation of atmospheric systems, background that carried directly into the modeling work at Courant.

Undergraduate degree in atmospheric sciences and chemistry. The dual major gave me a foundation in both the dynamical and chemical aspects of the atmosphere: thermodynamics, radiation, and fluid dynamics on one side; physical and analytical chemistry on the other. A natural combination for the aerosol-microphysics research that followed in the master's.

Interviewers want to see what you’ve done with real data and real constraints, not just coursework. If you’re a student, getting an internship or research project on your resume matters more than an additional class. The work doesn’t have to be at a top firm; it just has to be substantive.

Send more applications than feels comfortable. The downside of a rejection is small; the expected value of an additional application is positive. Don’t disqualify yourself from roles before the employer has had the chance to.

Most applications won’t lead anywhere. That’s the structure of the market, not a verdict on you. The candidates who get where they’re going are usually the ones who kept going after the 15th or 20th rejection.

After each interview, good or bad, write down what you stumbled on. Most people repeat the same mistakes because they never diagnose them. Treat each attempt as a data point.

The technical bar in quantitative finance keeps moving. Dedicate time, regularly, to staying current. Showing up with a stale skillset is a harder problem to fix in the middle of a job search than before it.

Figure out what makes you a specific candidate rather than a generic one. The resume that reads like everyone else’s doesn’t give the interviewer a reason to remember you. Lead with what you actually do well.

Technical interviews have predictable structure. Know your own resume cold: everything on it is fair game. Practice the fundamentals until they’re reflexive. Cognitive load under pressure is real, and preparation is the only mitigation.

Enthusiasm for the work matters, but it has to be real. Firms can tell the difference between someone who cares about the research problem and someone performing interest. If the role genuinely excites you, let that come through, and if it doesn’t, that’s also worth knowing before you take it.

Know what the group actually does. For quant firms, that means reading their papers, talks, and public research. For banks, it means understanding the specific desk or team, not just the brand. Interviewers notice when you’ve done this, and when you haven’t.

“What does success look like in the first six months?” and “What does the team find hardest?” are almost always good questions. Asking nothing, or asking questions you could have answered from the website, is a missed opportunity.

Don’t relax when you think an interview is going well. Overconfidence in the final questions is a common failure mode. Finish with the same attention you brought to the first question.

The people who schedule your interviews, the recruiter, and the engineer you have lunch with all contribute to the hiring decision. Be the same person in every conversation.