Where models earn their keep
We confess something most quant shops hide: our best signals fail about a third of the time in live markets. That failure rate is not a bug. It is the honest baseline that keeps us improving. We publish what we learn, document what breaks, and invite scrutiny because defensible research outlasts impressive claims. Our work spans factor crowding dynamics, machine learning interpretability, and market microstructure anomaly detection.
Collaborate with usPublished insights and white papers
We explore methods for making complex ensemble models interpretable to risk committees. The goal is not simpler models but clearer explanations of why a crowded signal triggered when it did.
Our anomaly detection research identifies microstructure patterns that appear during factor stress periods. These early footprints often precede broader crowding alerts by several trading sessions.
We maintain a living bibliography of factor crowding literature, updated quarterly, that maps academic contributions to practical implementation challenges faced by institutional risk teams.
Research footprint
Numbers that reflect disciplined inquiry rather than marketing ambition. We measure output by reproducibility and peer engagement, not volume alone.
Days of tick data analyzed across equity and futures markets to validate crowding signal stability
Distinct risk factors modeled for concentration dynamics, from standard style factors to custom client overlays
Percent of published backtests that include explicit out-of-sample periods and failure scenario analysis
| Year | Milestone | Area | Details |
|---|---|---|---|
| Academic foundations | Factor | Originated as a research project examining how mutual fund flow data predicted momentum factor drawdowns. Initial findings suggested crowding signals carried predictive power beyond standard risk metrics. | |
| Prototype model | Engineering | Built the first production-grade crowding detection engine using gradient boosting on fund holdings and options market data. Model achieved promising out-of-sample results during the volatility events of that year. | |
| Interpretability framework | Methods | Developed methods for decomposing ensemble predictions into factor-level contributions. This allowed risk teams to understand not just that crowding was elevated but which specific exposures drove the signal. | |
| Client validation program | Adoption | Launched structured validation partnerships with three institutional risk teams. Their feedback reshaped alert thresholds, delivery formats, and the frequency of model retraining cycles. | |
| Microstructure layer | Engineering | Added order book and execution data streams to detect crowding footprints at higher frequency. This layer often provided several days of advance warning before traditional positioning signals triggered. | |
| Live production systems | Operations | Transitioned from batch research environment to real-time signal delivery. Introduced automated model monitoring that flags performance decay and triggers retraining without manual intervention. | |
| Open research initiative | Community | Began publishing methodology notes and failure case studies publicly. The initiative aims to advance industry understanding of crowding dynamics while inviting external critique of our approaches. |
The researchers building these signals
Dr. Amara Osei
Lead Research Scientist, Factor Dynamics
Former postdoctoral researcher in market microstructure. She designed the original crowding detection framework and now oversees model architecture decisions.
James Nakamura
Senior ML Engineer, Model Interpretability
Specializes in making complex ensemble outputs explainable to non-technical stakeholders. He built the factor decomposition layer that risk committees rely on.
Priya Sharma
Quantitative Analyst, Signal Validation
Runs the out-of-sample testing program and maintains the living bibliography. She ensures no model ships without documented failure modes and sensitivity analysis.
Liam O'Brien
Lead Data Engineer, Pipeline Architecture
Responsible for ingestion, normalization, and real-time delivery infrastructure. His systems process billions of records daily with sub-minute latency.
Research collaboration and updates
We work with academic institutions, industry working groups, and client research teams to advance understanding of factor crowding. Whether you want to co-author a study, challenge our methodology, or simply receive our quarterly research digest, we are open to the conversation.
Share your research interest
Tell us what aspect of crowding, microstructure, or interpretability you are exploring. We read every inquiry and route it to the relevant specialist.
Share your research interest
Tell us what aspect of crowding, microstructure, or interpretability you are exploring. We read every inquiry and route it to the relevant specialist.
Connect with the right researcher
Our team matches your interest with the analyst or engineer best positioned to engage. No sales calls, just direct technical conversation.
Connect with the right researcher
Our team matches your interest with the analyst or engineer best positioned to engage. No sales calls, just direct technical conversation.
Establish collaboration terms
We agree on scope, data sharing boundaries, publication rights, and timelines. Every partnership is structured to protect both parties intellectual contributions.
Establish collaboration terms
We agree on scope, data sharing boundaries, publication rights, and timelines. Every partnership is structured to protect both parties intellectual contributions.
Produce and review jointly
Work proceeds through shared repositories, regular check-ins, and joint review sessions. We publish only when both parties are satisfied with the rigor of the output.
Produce and review jointly
Work proceeds through shared repositories, regular check-ins, and joint review sessions. We publish only when both parties are satisfied with the rigor of the output.
Join our research network
Subscribe to our quarterly research digest or propose a collaboration. We respond to every serious inquiry and welcome challenges to our assumptions.