Point-in-time discipline
The research process uses annual expanding-window fits and preserves out-of-sample score artifacts.
Why Salarium exists
Modern investment research is fragmented across data acquisition, signal generation, model validation, portfolio construction, and risk management. Salarium explores whether those layers can be unified into a governed research system.
Release evidence
Aug 11, 2026, 11:14 PM UTC
Regenerated from committed reports
Ranking artifact
Jun 10, 2026
25 names · committed · not live
Candidate artifact
Jul 10, 2026
20 monitored names · not live
Execution status
Research only
No brokerage connection or live order routing
Project identity
Salarium is a systematic equity research platform, a governed research workflow, and a project built in public. It combines data, alpha research, portfolio construction, risk, and validation without presenting itself as a hedge fund or treating simulation as live performance.
The project is built by Niall Gillen, a finance student at Indiana University's Kelley School of Business, as a serious demonstration of quantitative research, software architecture, and the ability to turn iterative analysis into a coherent product.
The name references the Roman salarium—historically associated with compensation and the linguistic root of “salary.” The modern project uses that idea as a symbol for disciplined capital allocation rather than as a claim about Roman payment practices.
Automation boundary
Automation supports data validation, macro analysis, model tournaments, backtest review, risk review, and research orchestration. Model outputs still pass through explicit quantitative artifacts, portfolio rules, validation periods, and human-inspectable release gates.
The product is not positioned as an AI stock picker. Its credibility rests on governed evidence and reproducibility, not on anthropomorphizing the model.
What makes the work credible
The research process uses annual expanding-window fits and preserves out-of-sample score artifacts.
Portfolio, horizon, breadth, covariance, and signal-weight hypotheses are compared under shared score streams.
Experiments that degraded performance remain archived rather than disappearing from the narrative.
Tests, audits, static builds, route validation, and committed JSON evidence gate the public release.
Research universe
Liquid-500
Locked portfolio-research population
Prediction / rebalance
20D / 10D
Horizon and trading cadence tested separately
Portfolio breadth
Top-10
Rank-15 persistence buffer
Leverage ceiling
1.25x
Permission ceiling, not a usage objective
For technical reviewers
The repository contains model-generation scripts, portfolio evaluators, experiment reports, release-snapshot exporters, governance tests, and the Next.js public product. The public interface is intentionally linked back to committed source evidence.
For non-technical visitors
Release boundary
It does not connect to a brokerage, execute orders, provide personalized advice, or claim that simulated returns will repeat. The value of the release is the architecture, evidence discipline, and reproducibility of the research process.
Read full disclosures