Pre-registered 2026-10-17 (commit 6bccc9f) · 1 run · data snapshot 2026-09-19T090800Z
The idea
Mark Minervini won the US Investing Championship in 1997 with +155% and published the rules he uses to pick stocks. His Trend Template has eight conditions: moving averages stacked and rising, price well above its 52-week low and close to its 52-week high, and strong relative strength. The proposed mechanism combines the 52-week-high anchor described by George and Hwang with the idea, from O'Neil, of only buying established trends.
The prior was negative. Three earlier measurements of long momentum and 52-week-high signals in this universe had found no alpha.
How it was tested
Pre-registered, reviewed and locked on 2026-10-17 (commit 6bccc9f). All eight conditions mechanically, with a relative-strength percentile of at least 70. US stocks, ETFs and funds excluded, at $10 or more with at least $5M of 20-day dollar volume. Buy at the next open, ranked by relative strength with at most 20 new positions a day. Exit after 20 sessions or at a 10% stop, the book's non-negotiable maximum loss. Counted in a family of five long momentum configurations, the Gate 7 bar was raised to t ≥ 2.33.
What was not tested is the discretionary half of the method: the VCP pattern, the pivot breakout, selling into strength.
Where it died
The single run failed all four conditions it faced. Gate 7: t −0.90 and −0.306% per trade over 6,692 trades, beta 0.93. Gate 8: t −1.14 on the captured trades, capture ratio 0.00. Gate F: the alpha per trade is below zero, not merely below 0.30%. Stability: negative in both halves, t −0.40 on 2,487 trades in 2014-2019 and t −1.04 on 4,205 in 2020-2026.
The record
Thesis
Stocks meeting all eight conditions of Mark Minervini's published Trend Template (aligned rising moving averages, near the 52-week high, far from the 52-week low, high relative strength) earn positive Jensen alpha over the next 20 sessions. Mechanism: underreaction anchored on the 52-week high (George & Hwang 2004) plus established-trend filtering (O'Neil / IBD).
Prior: Expected to fail. Three earlier measurements of long momentum / 52-week-high signals in this universe had no alpha.
Minervini, M. (2013). Trade Like a Stock Market Wizard. McGraw-Hill. US Investing Championship 1997 winner (+155%).
George, T. & Hwang, C. (2004). The 52-Week High and Momentum Investing. Journal of Finance 59(5).
Specification
signal
price > MA150 and > MA200; MA150 > MA200; MA200 rising over 21 sessions; MA50 > MA150 and > MA200; price > MA50; ≥ 30% above 52w low; within 25% of 52w high; relative-strength percentile ≥ 70
universe
US stocks (ETFs and funds excluded), price ≥ $10, 20d dollar volume ≥ $5M
entry
next open
exit
20 sessions or 10% stop (the book's non-negotiable max loss)
ranking
relative strength, strongest first, max 20 new positions/day
not tested
The discretionary part of the method (VCP pattern, pivot breakout, selling into strength). What died is "mechanical template + 10% stop + 20-session exit", not Minervini's full method.
Gates
Gate 7 — Jensen alphafail
After regressing out market exposure, is there alpha left?
t ≥ 2.33 (Bonferroni, family of 5 momentum longs)
t (HAC)-0.90alpha / trade %-0.31beta+0.93trades6,692
Gate 8 — Capture-preserved alphafail
Does the subset of trades a real portfolio can actually take keep the alpha?
t_cap ≥ 0.99 and ratio ≥ 0.50
t captured-1.14capture ratio+0.00
Gate F — Frictionfail
Is the alpha per trade large enough to pay for its own execution?
alpha/trade ≥ 0.30%
alpha / trade %-0.31
Stabilityfail
Is the alpha present in both halves of history?
t > 0 in both halves
t 2014 2019-0.40n 2014 20192,487t 2020 2026-1.04n 2020 20264,205
Why it died
Beta 0.93 and negative alpha with the same sign in both sub-periods: the template buys the market and nothing else. Adding moving-average structure and relative strength to the 52-week-high signal made it worse (t +0.53 → −0.90).
What it taught
Contest-winning rules are not edges: a winner-takes-all contest selects the tail of a variance distribution. Five of five long momentum / 52-week-high configurations in this lab have no Jensen alpha.