Factor Research

Discover, evaluate, and manage reusable quantitative factors with a guided research campaign

🟢 Research only: Factor Research produces hypotheses and evaluated factors. It does not deploy strategies, place trades, or move funds.

Factor Research turns a market hypothesis into a traceable set of factor candidates. Start from the Factor Research page or ask the agent in natural language. The same campaign, evaluation history, and factor library remain available after you close the page or restart the app.

Start a research campaign

Describe the signal you want to investigate. For example:

Research whether unusually strong 24-hour price moves persist over the next 12 hours across supported crypto markets.

When a request is underspecified, the agent asks only for choices that can materially change the result:

the asset or requested universe;
the data interval and prediction horizon;
the research direction or economic hypothesis;
the resource budget.

You can accept the recommended defaults. The agent then creates the same campaign used by the Factor Research page, rather than a separate chat-only workflow.

Research flow

Research setup → Research & Alpha Discovery → Factor construction → Screening and formal evaluation → Factor list

Research setup freezes the request, data settings, playbook, model, and budget for the round.
Research & Alpha Discovery searches market information, research evidence, and possible Alpha leads. A lead is a hypothesis to test, not a validated factor.
Factor construction converts supported hypotheses into executable Factor DSL expressions.
Screening and formal evaluation removes invalid candidates and evaluates the survivors with the selected qualification algorithm.
Factor list groups the results by status and keeps their source campaign, metrics, and evaluation evidence.

A campaign can run multiple rounds. You can adjust its budget, pause, resume, or cancel it. Completed work and evidence remain persisted. Expired work leases can be recovered after a process restart.

Requested and evaluated assets

Three asset scopes can appear in a campaign:

Requested universe is the broad scope described in the research request, such as all supported crypto assets.
Representative set is a bounded sample selected for a practical first evaluation. A user-selected set is preserved.
Evaluated assets are the exact symbols present in the frozen evaluation snapshot.

The task and evaluation report show the actual evaluated assets. Results from a representative set describe that set; they do not prove that the factor works for every asset in the requested universe.

Factor statuses

Evaluation passed means the latest valid formal qualification report is PASS.
Evaluating means automatic formal evaluation is still running.
Evaluation failed includes INVESTIGATE and REJECT results that did not meet the selected qualification standard.
Screened out means the candidate did not reach formal evaluation.

Empty status groups are hidden in the conversational result. The full task library keeps all groups and their counts.

Evaluation methods and OOS Rank IC

Basic Qualification Evaluation checks the minimum admission requirements: data integrity, direction-adjusted out-of-sample Rank IC, and uniqueness against the factor library. Comprehensive Robustness Evaluation adds stricter statistical, stability, novelty, transaction-cost, and holdout checks.

OOS Rank IC measures how well the factor ranks future outcomes on data outside the fitting sample. The admission IC shown in the approved factor library is the frozen, direction-adjusted OOS Rank IC from the exact passing report. Higher is better after direction adjustment, but IC alone does not guarantee a useful or profitable strategy.

Inspect and reuse a factor

Open a factor to inspect its DSL expression, economic hypothesis, initial and formal metrics, expected direction, asset coverage, evaluation history, and source research task. The complete factor library prioritizes evaluated factors that passed and preserves the unsuccessful candidates as research evidence.

Approved factors can be selected in Strategy Studio. Strategy Studio freezes the exact factor definition and passing evaluation used when the strategy is generated.