I lead the marketing measurement and incrementality team at Rocket. My team builds the MMM, attribution, and experimentation systems behind how the Rocket Mortgage and Redfin brands decide where their marketing money goes.
I'm an economist by training — Ph.D. from the University of Kentucky, where I still teach the Ph.D. applied econometrics course. Most of what I do is econometrics pointed at a marketing problem: geo-experiments and synthetic control, Bayesian media mix models, incrementality-based attribution, and scenario planning for when the macro picture won't sit still.
The tools I want usually don't exist, or exist and are too slow, so I end up writing them. Two are public: panelkit for panel and geo experiments, and tsecon for time-series econometrics. Both started as things I needed for my own work and got cleaned up afterward.
Two libraries, both written because I needed them and couldn't find a version that did what I wanted. One is for experiments, one is for time series.
A causal-inference library I wrote from scratch for panel and geo experiments — Synthetic Control, Augmented SC, Synthetic DiD, MC-NNM, Callaway–Sant'Anna, and Sun–Abraham — in Rust with Python bindings.
The numerical core has no dependencies (matmul, Cholesky, QR, Jacobi SVD, simplex/Frank–Wolfe solvers). It runs about 60× faster than a NumPy/SciPy synthetic control per fit, and about 1,400× faster on a full placebo test, with reproducible, bit-identical results. It also includes a geo-test design layer for power analysis, market selection, and post-test measurement.
The time-series econometrics toolkit I kept rebuilding by hand — first for my dissertation, then again at work — finally written down properly. It covers a full workflow rather than a single estimator: diagnostics, ARIMA and GARCH, and VARs end to end.
The part I care most about is the structural identification suite: sign restrictions, zero-and-sign, narrative, long-run, proxy / external-instrument, heteroskedasticity, and non-Gaussian identification, all under one interface. It also does local projections, Bayesian VARs, nowcasting, panel methods, and the term structure.
Estimates how monetary-policy shocks affect household consumption by identifying functional monetary-policy shocks. It uses functional local projections across the whole interest-rate term structure and finds that consumption responses vary with household balance sheets and life-cycle stage.
Instructor for Ph.D. Applied Econometrics (University of Kentucky, 2025), a special section of Graduate Macroeconomics, R for Economists, Principles of Macroeconomics, and Economic & Business Statistics. Former Fulbright grantee (Andorra).
Python, PySpark, R, MATLAB, Rust, SQL, PyTorch, PyMC, Scikit-learn, XGBoost
Bayesian MMM, geo-experimentation & incrementality, synthetic control & DiD, attribution, forecasting & time-series econometrics
AWS (SageMaker, S3, Glue, Athena, Lambda), Databricks, Spark, Snowflake, MLflow, Seldon, Argo, agentic AI, data clean rooms
English (native), French & Spanish (working fluency), Japanese (N3)
Email is the fastest way to reach me: chasecoleman93@gmail.com. I'm also on GitHub and LinkedIn. Happy to talk measurement, experiment design, or anything in the libraries.
