Brad Lasater.

Brad Lasater

Applied Scientist · Quantitative Research & Engineering

Building decision systems under uncertainty.

I develop forecasting, optimisation, and machine-learning systems that turn noisy observations into operational decisions. My current independent project applies that experience to systematic volatility: building and validating the infrastructure from market data and pricing through to portfolio construction and execution.

GitHub LinkedIn brad@bradlasater.com

I build production decision systems for noisy, constrained environments. Over ten years I have built machine-learning, forecasting, state-estimation, and optimisation systems across consumer marketplaces, live events, aerospace, and defense — real-time auction and ranking systems serving millions of users, live-event demand forecasting that cut WAPE by roughly 60% against a prior-season-average baseline, particle filters and constrained real-time sensor scheduling, and a patented stochastic production-planning model.

The volatility work is where that experience meets a domain I have not worked in before, so it is built in public with an evaluation protocol fixed before any result existed, and a build-status page that says what runs and what does not. Estimating what uncertainty is worth, then trading the difference — eventually. Today it estimates, and the estimates are the thing worth reading.

Explore the system View CV Get in touch

Relocating to New York in January 2027 and open to quantitative research and research-engineering roles.

Build phase
4 of 10 stages run today; the rest are designed and not built
Latest validated evidence
No component has a published validation date yet
Next milestone
Volatility forecasting — Reproduce the baseline forecast evaluation from a frozen dataset and configuration, with a documented target horizon and temporal split, and report the HAR-RV error before any learned model is fitted.

Counted from the component manifest, which is what the architecture and build status pages render from.

In development

An end-to-end volatility system, under active construction

A research and execution platform for defined-risk index-volatility strategies, beginning with SPY and SPX/SPXW options at roughly 5–45 days to expiry, with VIX options captured for the term structure. The system is designed to link point-in-time options data, arbitrage-aware surface construction, physical-variance forecasts, risk-neutral variance extraction, a no-trade gate, Greek-constrained sizing, event-driven backtesting, and execution under observed spreads and slippage. That is the target architecture, not a description of what runs today.

The evaluation rules were fixed and published before any results existed, and they bind the work as it lands: every trial logged, validation purged and walk-forward, costs charged rather than assumed away, live observations append-only, and a post-mortem written whenever the system loses money or a hypothesis dies. None of it has produced a result yet. What runs today is the ingest and pricing box: options chains, trades and day bars land on a schedule with a daily coverage audit; implied volatility is inverted per contract with own Greeks, checked every evening against put–call parity and reprice identities; and those are reduced to an ATM implied-volatility term structure — one row per date, root and expiry, back to 2022-08-31 — with the forward recovered from put–call parity, because no index level is entitled on this data tier. On top of that curve a raw-SVI smile is fitted per expiry over OTM strikes for SPX and SPXW, with butterfly and calendar arbitrage repaired inside the fit rather than checked afterwards; SPY is refused there because its strikes are American. A HAR-RV baseline forecast is written and tested but not yet scheduled or evaluated out of sample. Everything downstream of that — sizing, risk authorisation, execution — is designed and not built.

System overview Evaluation protocol Research log GitHub
Rationale

Why volatility, why now

Every system I have built professionally solves the same problem: estimate a hidden state from noisy, incomplete observations, then allocate a scarce resource against that estimate before the world moves on. Satellite position from sparse radar returns, with sensor time allocated by a binary linear program. Maintenance demand under stochastic failure, against finite production capacity. Live-event demand months out, against inventory and pricing.

The variance risk premium is that problem in a different coat. Implied volatility is the market’s forecast of a hidden state, and a persistently biased one, because someone is paid to carry the risk of being wrong. The bias is not itself an edge — it is well documented, and harvesting it blows up the moment you stop asking how rich the premium is relative to the risk being assumed right now, and what it actually costs after spreads and hedging error to hold the position.

What is new here is the domain, not the method. That is why the build is public, why the evaluation protocol was fixed before a single result existed, and why the honest answer to “does it work” today is that nothing has been evaluated yet.

Proof

Three systems I have shipped

Each is the same shape as the volatility problem: a hidden state estimated from noisy observations, and a scarce resource committed against that estimate under a hard constraint. Problem, contribution, outcome.

Marketplace decision systems

Nift · Ticketmaster

Problem. Real-time auction and ranking decisions for millions of users, where the value of a match is unobserved at decision time and the decision cannot wait.

Contribution. Built and owned the production ranking and allocation systems end to end — features, models, serving, and the monitoring that catches a silent degradation before a business metric does.

Outcome. Systems in continuous production making live decisions under latency budgets, with the offline-to-online gap measured rather than assumed.

Forecasting

Ticketmaster

Problem. Forecast live-event sales months ahead, where demand is non-stationary, seasonal, and event-specific, and the incumbent baseline was a prior-season average.

Contribution. Temporally dynamic gradient-boosted trees with time-respecting validation, so the reported improvement was measured out of sample rather than fitted.

Outcome. Cut forecast WAPE by roughly 60% against that baseline — a defensible number because the comparison was against the method actually in use, not against zero.

State estimation & optimisation

Lockheed Martin · Boeing

Problem. Recover the true position of an object in space from sparse, noisy radar returns, and decide where to point a sensor next — under a hard latency constraint, with sensor time the scarce resource.

Contribution. Particle filters for the estimation, a binary linear program for the scheduling, and at Boeing a stochastic production-planning model for maintenance demand under random failure.

Outcome. Deployed in special programs; the production-planning work issued as U.S. Patent 10,204,323.

Background

Selected experience

The through-line is decision-making under uncertainty: estimating hidden state from noisy observations, and allocating scarce resources against those estimates in real time.

Oct 2023 – Present

Senior Data Scientist

Nift · Remote
Part-time, concurrent with Ticketmaster
  • Own modelling and optimisation for a multi-brand auction and ranking system serving millions of users per month — online allocation under budget and inventory constraints.
  • Built the data pipelines and feature store backing both real-time inference and offline training, keeping serving-time and training-time features consistent.
  • Established model-health and retraining frameworks covering calibration drift, data quality, and segment-level performance decay.
Aug 2021 – Mar 2025

Lead Data Scientist

Ticketmaster
  • Cut live-event sales forecast WAPE by roughly 60% against a prior-season-average baseline, using temporally dynamic gradient-boosted trees.
  • Led the team delivering data-science products, translating commercial strategy into technical roadmaps.
  • Developed feature-selection optimisation that shortened the path from candidate model to production deployment.
Oct 2018 – Aug 2021

Lead Space Algorithm Developer

Lockheed Martin · Special Programs
  • Implemented particle filters producing actionable ephemeris for space surveillance, recovering latent state from sparse, noisy, irregularly sampled observations.
  • Built a binary linear program for real-time sensor scheduling, allocating a constrained set of sensors across competing targets under hard latency requirements.
  • Designed Kalman and Singer filter instantiation procedures that improved convergence on late threat detections.
Feb 2016 – Oct 2018

Data Scientist, Defense Analytics

Boeing
  • Patented the F/A-18 Fleet Upgrade Model — a discrete-event simulation optimising production plans under stochastic demand and resource constraints.
  • Developed genetic algorithms for multi-echelon inventory optimisation; the work secured production funding.
  • Led the data-science arm of the T-45 physiological events investigation, driving root-cause identification.
Full history, including military and teaching service
Credentials

Patents & education

U.S. Patent 10,204,323 — F/A-18 Fleet Upgrade Model & Discrete Event Simulation
Production-planning optimisation under stochastic demand, resource, and mission-availability constraints.
Presenter — Boeing Technical Excellence Conference
SherFire: multi-echelon inventory optimisation via genetic algorithms.
M.Eng., Systems Engineering — Arizona State University, 2015
B.S., Applied Mathematics — University of Texas at Arlington, 2013.
Ph.D. coursework completed, Stochastic Multi-Objective Optimization — Missouri S&T
Taken part-time while working full-time at Boeing.
Availability

What I am looking for

I am relocating to New York in January 2027 and am open to quantitative research and research engineering roles. My ideal role combines statistical research with ownership of the full data-to-deployment lifecycle. I work best in small teams with high autonomy and clear risk constraints.

Contact

Get in touch

If you are hiring or would like to discuss my research, email me directly or connect on LinkedIn.

brad@bradlasater.com LinkedIn GitHub

This page was last updated .