RLTrade · ongoing research
Learning how different parts of the market behave, then testing whether that knowledge improves trading decisions.
Live RLTrade portfolio
~+25–30%
Approximate gain · owner-reported
The live figure is Jacob’s approximate reported gain since Schwab launch, dated above. A matched-period S&P 500 comparison is not shown for this estimate.
Median return across completed full-window backtests for each sector’s latest tested stock group. SPY is the S&P 500 proxy recorded in the same test. These are individual simulated strategies; there is no combined live-portfolio return here.
+90.9%median backtest
S&P 500 proxy · SPY +12.0%
Jan 2 – Aug 31, 2026
Retrospective stock selection
+29.3%median backtest
S&P 500 proxy · SPY +12.0%
Jan 2 – Aug 31, 2026
Retrospective stock selection
−12.6%median backtest
S&P 500 proxy · SPY +12.0%
Jan 2 – Aug 31, 2026
Retrospective stock selection
The full range includes losing tests. Group dates differ because sectors finish testing at different times. The stock groups were selected after the displayed evaluation period, and multiple configurations have been evaluated on the same holdout. Selection bias may inflate these exploratory results. Figures are cumulative, not annualized.
Compare market conditions and relative sector strength. Test stock groups in technology, energy, healthcare, real estate and utilities as results become available.
Train reinforcement-learning policies to act on market observations. Compare different inputs and training runs while recording the resulting portfolio values.
Evaluate on dates withheld from model training, alongside SPY and sector benchmarks. Stock-group selection is retrospective in the results shown. Record returns, drawdowns and simulated trading costs. Completed tests that fail integrity checks are excluded.
Keep the failed tests as well as the successful ones. Use the results to decide what needs replication or a different research approach.
The study finds a same-day association with prices, but no next-day lead. The article includes the charts, data sources and method.
Returns and drawdowns come from saved offline evaluation metrics, with a matching completion record and no later integrity invalidation. Each sector uses its newest completed stock group and the largest set of tests sharing exact evaluation dates and the same SPY return. Only scientifically valid tests covering the full evaluation window are included. The median is calculated across those tests; it is not the return of an investable allocation. Returns use each simulation’s recorded fees and execution assumptions; they are not independently audited account results.
SPY figures are the benchmark series saved with the research run, not a separately sourced S&P 500 total-return series. About the SPY benchmark ↗︎
Snapshot: . Public result summary ↗︎
AI · Résumé & projects