Machine learning for systematic trading.
Quantizi researches, builds and operates machine-learning models that trade US equities. We run them on our own capital and work with funds and trading firms that want the same rigor applied to their research.
Most ML trading research fails quietly. Our process is built around that fact.
A model that looks good in a backtest is the default outcome, not an achievement. The work is in making sure the result survives contact with data it has never seen, and keeps surviving once it is live.
Data before models
Point-in-time datasets, survivorship-free universes, corporate-action handling and explicit treatment of delistings. We spend more time on data lineage than on architectures.
Leakage and overfitting control
Purged and embargoed walk-forward validation, strict feature timestamps, and a fixed research protocol that limits how many times we are allowed to look at the test set.
Out-of-sample first
Every strategy earns its place through held-out periods and paper trading before capital is allocated. In-sample performance is not reported internally and is never used to make decisions.
Costs and capacity
Transaction costs, slippage, borrow and market impact are modeled from the start, not added at the end. A signal that only works at zero cost is not a signal.
Risk as a constraint, not a dial
Position limits, factor and sector exposure caps, and drawdown-based de-risking are set before a model goes live and are not relaxed to improve returns.
Monitoring in production
Feature drift, prediction decay and execution quality are tracked continuously. Models are retired on evidence, and every live decision is logged and reproducible.
Two ways we work.
We trade our own capital, and we partner with institutions that want to build or stress-test ML-driven strategies in US equities. We do not manage outside capital and do not sell signals to individuals.
Our models, our capital
Quantizi operates systematic strategies in US-listed equities with its own balance sheet. This is where our methodology is tested daily and where every tool we build is used in anger.
- Fully systematic, model-driven execution
- No external investors, no fund structure
- Research and infrastructure built in-house
Research and model development
For hedge funds, family offices and proprietary trading firms, we take on defined research engagements: from auditing an existing ML pipeline to building a strategy from hypothesis to production-ready code.
- Alpha research and feature engineering on your universe
- Independent validation of existing models and backtests
- Research infrastructure: data pipelines, backtesting, monitoring
- Knowledge transfer so your team owns the result
A small research team, not a product company.
Quantizi is a Menlo Park, California research group focused on one problem: applying machine learning to US equities without fooling ourselves. The name comes from quant and quantize: turning noisy market data into discrete, testable decisions.
Our work sits at the intersection of statistical learning, market data engineering and systematic risk management. We prefer well-understood methods, documented failures and small teams that read each other's code.
Tell us what you are trying to build or verify.
We reply to every serious inquiry, usually within two business days. If you are not sure whether your problem fits, write anyway.
info@quantizi.com- Your firm and role, and the market or universe you trade.
- The problem: a new strategy, a model to audit, or infrastructure to build.
- What data and research you already have.
- Timeline and how you would like to engage.