Systematic · Diversifying alpha · U.S. equities
A multi-agent quantitative investment platform: independent research pods under principal-investigator orchestration, feeding one long-only U.S. equity book. 42 signals in production today, each screened to add alpha the book doesn't already hold.
Thesis
Every large multi-manager converged on the same design: independent research pods hunting alpha in parallel, with a shared data, risk, and execution platform underneath. The design's limit has always been people. There are only so many world-class pods you can field.
Polyvaria removes that limit. Its research pods are agent teams directed by centralized principal-investigator orchestration: priorities and risk budgets flow down, results and attribution flow up. A pod develops its hypotheses from idea through implementation against the shared backtester and submits the survivors for validation. The PI layer moves compute toward the families earning it, retires lines of research that stall, and keeps the shared record of what has already failed, so no pod re-runs a dead hypothesis. Adding a pod is a deployment rather than a hire, and research breadth grows with compute; the gates still decide what earns weight in one book.
The platform underneath is staffed the same way. Risk modeling, portfolio optimization, execution, data onboarding, and backtesting are each owned by an independent agent team working continuously on that subsystem's quality and fidelity. None of it was practical until reasoning models could carry a research loop end to end and have the result clear the same bar as human work. Now that they can, breadth is an orchestration problem.
Approach
The same code runs the live daily rebalance and the multi-year historical backtest. The backtest pins the as-of date and simulates its fills; everything else is the same path. Each stage is owned by one of the platform's independent agent teams.
01
Point-in-time by construction.
The investable universe is rebuilt for every historical date from delisting records, so the backtest never sees a name the live book couldn't have traded. Survivorship discipline is enforced in code, not promised.
point-in-time universe
02
42 signals in production.
A research pipeline ingests price, fundamental, event, and news data and produces cross-sectional rankings of the universe. Fundamentals enter as first reported and news as it arrived, never as later revised. Signal families span value, momentum, quality, capital discipline, defensive, reversal, liquidity, post-earnings drift, and sentiment. Every candidate has to clear an orthogonality and decay-resistance bar before it earns a weight in the live book.
cross-sectional alpha ranks
03
Regime-conditioned sector and factor rotation.
A regime classifier across volatility state, yield-curve shape, growth/inflation quadrant, and breadth maps the current market to a weighting overlay. Momentum, defensive, value, and reversal families lean in or out of the book as the regime shifts, and sector exposures rotate with them, rather than being statically blended. The overlay clears the same holdout and decay gates as any signal before it touches live weights.
regime overlay · sector tilts
04
Five structural factors.
Market, size, value, profitability, and investment exposures are estimated with rolling regression. Raw signals are winsorized, z-scored within sector, and residualized against the factor structure, leaving an expected-return vector the factor model can't explain. The same factor covariance enters the optimizer's objective directly, with shrinkage on residual variance.
residual alpha · factor covariance
05
Constrained mean-variance.
A conic solver maximizes expected return net of risk and transaction cost, subject to long-only, sector, market-cap, single-name, and turnover constraints. Cost and turnover are charged against current holdings, which keeps the trade list stable from one rebalance to the next.
target weights · trade list
06
Scheduled against the cost model.
A scheduling layer slices parent orders across the day against a spread-and-impact model, with participation caps and child-order placement tuned to liquidity and to the broker's observed fill behavior. Post-trade TCA feeds realized costs back into the optimizer's transaction-cost term, so the cost curve the book trades against reflects what it has actually been paying.
fills · realized cost
Research Library
Each family is owned by a research pod that works it for new signals against the platform's backtester and historical archive. More candidates means more accidental discoveries, which is exactly why the validation gates sit outside the pod: a candidate stays on the bench until it clears them.
Principles
Signals, costs, configurations, and the platform code itself all sit on feedback loops. As the underlying models improve the loops close faster, and every team inherits each new model generation as soon as it passes the platform's own evaluation harness. What reaches the book still clears the gates.
Every cached signal is evaluated daily on its information coefficient over the prediction horizon, so decay gets flagged early. The machine-learning overlay refits on a walk-forward schedule with an embargo against leakage, and each evaluation cycle adds new hypotheses to the cache while reweighting the old ones.
The factor model, exposure estimator, covariance assembler, constraint set, and cost model are all hot-swappable from configuration. Backtest-replay tooling lets the pods run counterfactual configurations against history and see which choices actually move attribution.
The linear and square-root impact coefficients in the cost model are fitted to the platform's own broker fills and refitted as trade history accumulates, so slippage assumptions track what the book pays to trade. The same fills drive the post-trade TCA that order scheduling is tuned against.
Pipelines, monitors, data ingest, the backtesting harness, and the platform code itself are built and improved by the same agent pods that operate the system day to day; validation stays with gates the builders don't own. There is no separate engineering organization for research to wait on.
Controls
The agents operate the platform; people set its boundaries. The mandate, the constraint stack, and the gate thresholds belong to the platform's human principals, and nothing reaches the live book without clearing them.
Stack
The interesting risks belong in the research, not the plumbing, so components are picked for determinism and auditability first and speed second. The heavy intelligence lives in the research layer, not in the models that trade.