The 2-period RSI trade, tested the hard way
Buy a large-cap when its 2-day RSI collapses while it's still above its 200-day average. It's one of the most widely published retail setups of the last twenty years, and it is almost always sold on one number: the win rate. This page is the rule, frozen, before we have collected a single trade.
On 2026-11-02 this experiment had not yet reached both pre-registered floors (at least 30 graded trades and at least 20 different companies). Under the rule fixed in advance, the verdict is computed automatically, once, at the first daily run where both floors are met, and published here either way. Until then the page shows no verdict language. The running count is in the live experiments report.
Signals are logged automatically each session and graded five sessions later. The verdict is written once, on or after 2 November 2026, and published whether it passes or fails.
Why not September, as this page first said. When we registered this test we said "around September, once 30 trades exist." On 28 August, before opening any outcome file, we noticed that wording left the verdict date open: we could have declared it on whichever week the result first looked good. That is the multiple-looks problem this site exists to catch in other people's records. So we fixed one evaluation date in advance (2 November, the date already set for Experiments 03 and 06–09, so the whole family can be corrected for multiple testing together) and added a floor of at least 20 distinct companies. Both changes can only make a pass harder. The amendment is dated in HYPOTHESES.md. This page should have said so at the time; it now does.
Why this one
Experiment 01 taught us something we'd rather have learned cheaply: on our own data, a rule that took profits at +5% produced a 67% win rate and an expectancy of −4.4% per trade. High win rate, money-losing strategy, no contradiction. That's the trap this strategy is most exposed to, because the win rate is the number on the box.
There's a second reason. Experiment 01's screen operated in low-float microcaps where the raw daily drift was under 1% and a realistic round-trip cost was around 2%. Nothing could have worked there — the friction was bigger than any available signal, in either direction. Liquid large caps invert that: a round trip costs about a tenth of a percent, so a real effect would actually be detectable. If this fails, it fails on its merits.
The rule, frozen
- Universe
- 40 liquid US large caps across sectors, fixed at registration. Adding or removing a name voids the test.
- Entry condition (evaluated after the close)
- Close is above the 200-day simple moving average and Wilder 2-period RSI is below 5. If more than five names qualify, take the five most oversold — so no single day can dominate the sample.
- Entry price
- The opening price of the next session. Signals are written to an append-only log before that open.
- Exits graded (all three, every trade)
- The same-day close; the fifth session's close; and the published exit — first close where 2-period RSI rises above 70, otherwise the five-session time exit.
- Costs
- 0.10% round trip, applied to every trade. Stated up front so it can't be tuned afterwards.
- Benchmark
- A day-matched control: the equal-weight return of the entire 40-name universe over the identical window. The question is not "did it make money" — that mostly measures whether the market went up — but "did it beat holding the same universe on the same day."
What it has to clear
Frozen at registration, so it cannot be softened later:
1. At least 30 post-registration graded trades.
2. The mean and the median day-matched excess return positive, net of costs. Not either — both. Experiment 01 taught us that means on financial data can be one lucky trade.
3. A 95% confidence interval on that excess that excludes zero, computed treating each company as the unit of evidence rather than each trade.
4. The direction holding across at least three consecutive weekly snapshots.
5. At least 20 distinct companies behind those trades. Added 28 Aug 2026, before any outcome was read; it can only make a pass harder.
The win rate is reported but is explicitly not a pass criterion. That is frozen here, in advance, precisely so it cannot be quietly substituted for the result if the expectancy disappoints.
What we expect to happen
Stating a prediction before the data is part of the method, and it's uncomfortable on purpose.
Short-horizon mean reversion in liquid equities is a real, documented effect — we are not expecting nonsense. But this specific formulation has been publicly known and heavily traded since roughly 2008–09, and effects that get published tend to get arbitraged. So our registered prediction is: the high win rate replicates (probably somewhere around 65–75%), and the day-matched excess return is small or zero after costs. We'd put the chance it clears the full bar at roughly one in three.
A high win rate with no excess return would be the most useful outcome we could get — because that is exactly what the strategy is sold on.
How you'll be able to check it
Every signal is written to rsi2_signals.csv before its entry session, and every graded outcome to rsi2_outcomes.csv, both public and append-only. The complete frozen rule — universe, thresholds, costs, pass bar and this prior — is in HYPOTHESES.md, dated. The verdict, whichever way it goes, will be recorded in the audit log and written up here.
If we later change any constant in that rule, the test is void and we'll say so. That's the deal.
What this experiment is not
It is not a recommendation to trade this or anything else, and nothing here is advice. It is not a claim about any particular author or service — we are testing a published technique, not grading a person. And whichever way it lands, it will describe this rule, on these 40 names, over this window. It will not be evidence about mean reversion in general.
Get every verdict by email
Verdicts land on dates fixed in advance; each experiment page shows its date. We'll email you each one, free, the day it publishes — and nothing else. No signals, no picks, no offers. Verdicts stay free and public for everyone either way; this just means you don't have to remember to check back.
One confirmation email first — you're not on the list until you click it. Privacy
Educational and informational only — not investment advice, not a recommendation, and not a broker-dealer. ThePickLog is operated by AMD Ventures, LLC (Florida).
← Experiment 01: our own screener, and why it failed
Evidence Validation · Score & metrics · Trust & Data
Tools Calculator · Vetting Guide · 7-Check Scorecard
Legal Disclaimer · Privacy · Terms