Investing
The one research class that survives walk-forward — shown honestly, with its caveat and live forward result.
Walk-forward research path
The signal is only interesting after the holdout.
ROBUST
Point-in-timefiled data
HoldoutOOS first
Walk-forwardrolling folds
Forwardlive clock
Backtest window2016-06-13 → 2026-06-12
Evidence-first research
Rank the universe by a point-in-time fundamental each quarter, hold the top K, equal-weight. IS = in-sample fit · OOS = held-out · Forward = live since the last rebalance.
| Screen | Top K | OOS net | OOS win | OOS maxDD | Walk-forward | Forward | Status |
|---|---|---|---|---|---|---|---|
| EPS Acceleration | 15 | +51.4% | 83% | −13.8% | 6/6 folds | +16.2% | Robust |
| Revenue Growth | 10 | +39.7% | 58% | −19.1% | 2/4 folds | +0.0% | NOT_ROBUST |
| Return on Equity (ROE) | 5 | +29.3% | 58% | −20.0% | 4/6 folds | -3.1% | Robust |
| Momentum ∧ Quality | 10 | +21.4% | 67% | −23.0% | 4/6 folds | +5.3% | Robust |
| Net Margin | 5 | -1.0% | 33% | −23.7% | 1/4 folds | +0.0% | NOT_ROBUST |
Honest caveat: PIT-clean on fundamentals (filed_at<=decision date), but universe = today's filers → survivorship-limited; not a tradeable edge on magnitude alone
Method: real 1d candles, quarterly rebalance, rank by point-in-time fundamental (latest accession filed<=date), next-open fills, 1 bps/side, IS-rank/OOS-holdout
Why this is
the honest one
A signal that looks great in-sample usually dies out-of-sample — that’s the rule, not the exception, and our
trading research shows it openly. EPS acceleration, ROE and momentum∧quality are different:
they keep beating buy-and-hold when we re-fit on rolling windows we never saw during selection. That’s as close to
a real edge as we’ve found. We still show the negative live forward window above — because magnitude on a
survivorship-limited universe is not a tradeable promise. This is research, not investment advice; we don’t
manage money and we don’t tell you what to buy.
Out-of-sample by default
Headline results are out-of-sample unless explicitly marked otherwise.
Walk-forward, not backtest
We re-fit at each step and test on the next window to reflect how it would have behaved.
Caveats, shown
Drawdowns, regime shifts and limits are reported, not buried.
Test an investing signal before you trust it
Run the same out-of-sample and walk-forward checks on your own tickers — free, no account needed.