Scanner

Settings & methodology

Current weights, thresholds, and stage gates. To change values, edit the source code and re-run scans. Until backtest validation, all weights are research-grounded judgment.

Reversal signal weights
Weights now reflect boththe academic literature AND the first backtest's empirical findings. ✓ = empirically validated direction; ◇ = literature-only (not yet backtested).
✓Bullish RSI divergence
✓✓ BACKTEST SIGNIFICANT (t=7.49, μ=+4.62%, +3.64% vs SPY, hit=57%). Only signal to clear Bonferroni. Lit: Pring/Murphy.
20%
✓RSI crossed 50
BACKTEST 2nd strongest (t=2.76, +0.77% vs SPY)
18%
✓Volume accumulation
BACKTEST 3rd (t=2.25, +0.46% vs SPY). Cooper 1999, Lee/Swaminathan 2000
18%
◇Analyst upgrades (30d)
Literature: Stickel 1991, Womack 1996
14%
✓MA50 reclaimed
BACKTEST mixed (μ=-0.08%, +0.62% vs SPY). Lowered from 18%.
10%
✓Higher-lows pattern
BACKTEST suggestive (t=1.67). Lo/Mamaysky/Wang 2000
8%
✓Golden cross
BACKTEST WORST (μ=-0.34%, t=-0.37, -0.77% vs SPY). CUT from 18%.
6%
◇Earnings within 14d
Bernard/Thomas 1989 — PEAD is post-event, not pre
6%
Quality gate (hard filter)
Pre-scoring filter. Fails → automatic SKIP regardless of other signals.
Gross margin≥ 40%
Operating margin≥ 15%
ROIC≥ 15%
5-year revenue CAGR≥ 8%
FCF positive each year5/5 years
Composite weights per mode
oversold
valuation 30%fundamentals 30%technicals 25%sentiment 15%
acceleration
reasonable valuation 10%growth acceleration 40%sustained quality 20%capital return 30%
quality
fair valuation 30%elite quality 40%compound smoothness 20%shareholder returns 10%
Stage classifier
How the reversal layer + mode score combine to label setup maturity.
1
Forming
Score ≥60, weighted signals < 20. Setup taking shape — no turn signals yet.
2
Bottoming
Score ≥60, signals ≤3 and weighted < 45. Early reversal — aggressive investors can start a position.
3
Confirming ★
Score ≥60, weighted 45-70. Multiple weighted signals + still discounted. The high-conviction window.
4
Confirmed
BUY-rated + weighted ≥70. Trend confirmed, partially moved. Enter on multi-mode confirmation.
5
Late
Weighted ≥70 but BUY rating not present (setup decayed). Most of the move has happened.
Pre-validation framework — handle with care

All weights and thresholds are research-grounded judgment, not empirically backtested. The first backtest run (5yr, 22,534 events) showed individual signals have hit rates of 44-48% — essentially coin flips — and no signal clears Bonferroni-corrected significance.

Next priority is the stage-transition backtest (multi-signal confluence on quality-filtered names) to test whether the framework's actual claim has historical edge.

Until then, treat outputs as a disciplined screener for further investigation, not as a prediction engine.