I study how futures order books behave around scheduled news, and how well backtests model queue position.

Where does liquidity go around EIA releases? Order-level evidence from CME energy futures, with implications for backtest queue models

Working paper, version 1, on SSRN (DOI 10.2139/ssrn.7609738). It covers the two studies below.

Order-level liquidity around the EIA releases

After EIA releases, CME crude and natural-gas books are missing the equivalent of 33.8 and 48.3 extra seconds of normal near-mid depth over five minutes, against matched Monday, Tuesday and Friday sessions (releases fall on Wednesdays and Thursdays, so a mid-week 10:30 effect is not ruled out).

Working Paper (SSRN)

How wrong are backtest queue models?

Scored against the true CME order queue, the queue-position estimators used by open-source backtesters miss real fill times by 0.31 to 1.07 seconds on average in crude and 0.68 to 1.98 in gas, averaged over all tracked orders, most of which their owners cancel first.

Working Paper (SSRN)Code

Quantitative Research – University of Chicago Project Lab, WH Trading

Building realized-volatility forecasts across futures horizons and benchmarking their accuracy against implied volatility, and evaluating regime-aware hedging in momentum and mean-reverting futures against fixed thresholds on P&L, risk and turnover.

Quantitative Modeling Intern (AI/ML), Endless Moments

Revenue and expense forecasts with ARIMA and LSTM models, about 15% lower out-of-sample RMSE than the baseline, validated walk-forward over two to three years.

Quantitative Research Intern (Econometrics, USDA Grant Funded), Morrison School of Agribusiness

Jacobian-corrected Box-Cox likelihood profiles that reproduce OLS exactly at λ = 1, in Python and Stata/Mata, and a parallelized estimation pipeline on more than 100 million forest-visitation records that cut runtime from about 10 hours to 3.

Optiver × UChicago Explaining Markets: 2nd Globally

An earnings-reaction model explained an extra 0.150 of R² beyond the earnings surprise on 517 events (95% CI 0.10–0.20), 27% above the strongest LLM baseline. Every model change had to pass out-of-sample replication, and the live LLM pipeline processed 584 of 584 events with no errors.

IMC Prosperity 4

Top 1.4% of 18,800+ (#256 Overall, #194 Algorithmic). Market making around Ornstein–Uhlenbeck fair values, with quotes skewed on book imbalance to limit adverse selection; on a 10-strike call chain the edge was only near the money.

mbo-queue

An open-source C++20/pybind11 library for rebuilding order books and tracking queue position from CME market-by-order data. Current version 0.7.0.

GitHubPyPI

C++ limit order book and market-making simulator

My C++20 order book gave the same fills as Liquibook at about 2.5 times its throughput on a deep-queue synthetic stream; in an exploratory near-empty-book test, Liquibook was faster.

GitHub

Honors thesis: Bayesian sequential decision-making under regime switches

At my thesis's settings, a pre-registered simulation re-run found the volatility feature cut false alarms from 26.4 to 2.1 per 1,000 pre-break steps but did not detect breaks faster.

GitHub

Cross-venue price discovery in BTC-USD

Using quote data only, Coinbase BTC/USD mid-price changes Granger-predicted Kraken's in 6 of 9 usable sessions after multiple-testing correction, against 1 of 9 in reverse.

GitHub

SPY volatility surface research

Found and fixed an expiry-clock error (UTC midnight vs 16:00 New York close) that cut the shortest expiry's full-surface SABR RMSE from 0.364 to 0.039; on 273 liquid quotes from one snapshot, SABR fit better than Heston in-sample on every expiry.

GitHub