Range mean reversion: a boundary and RSI recovery

Research status: untested hypothesis. No backtest, paper or live results exist. Every threshold, period, R value and risk fraction is an untested candidate parameter, not a recommendation.

The idea

“RSI is oversold, buy” is one of the most common and least tested ideas in trading. Oversold can stay oversold. This playbook asks a narrower question: when a range is objectively flat and its edges are already established, does a confirmed recovery off the lower edge behave better than blindly buying a low RSI? The comparison is built into the test.

Freeze the range at 00:00 UTC

• From the previous 96 completed 15-minute bars (24 hours), record high H, low D, width W = H − D (must be positive) and midpoint M. These don’t move for the day.
• Flat filter: the 50-bar SMA changed by no more than 0.05 × W over the last 5 bars. This is a candidate filter, not proof of a stable range.
• Established edges: at least two touches of each boundary (within 0.10 × W), separated by at least 4 full bars, all known at freeze time.
• RSI(14) uses Wilder’s smoothing, with a warm-up period before use.

IF–THEN rules

1. IF the completed signal candle’s low is within the lower zone (between D − 0.05W and D + 0.10W)
2. AND it closes above D + 0.10W but below M
3. AND RSI was below 30 on the previous bar and is 30 or higher now
4. THEN enter at the next bar’s first executable price, if shared checks pass.

• Stop: D − max(2 ticks, 0.05W). Frozen.
• Target: the midpoint M. Frozen. Net reward must be at least 2R after costs.
• Range-break exit: a completed close above H or below D exits at the next open (if the stop hasn’t already filled).
• Time stop: 16 bars (4 hours), or before the UTC day ends, whichever comes first.
• One attempt per day. No automatic second entry.

When NOT to trade

• A completed bar has already closed outside [D, H] since the freeze: the range is retired for the day.
• Boundaries missing or ambiguous, flat filter fails, or RSI hasn’t confirmed.
• Not enough room to the midpoint for 2R.
• Any shared no-trade filter fails.
• Shorting the upper boundary isn’t covered; it would need its own spec and test.

Hypothetical example (made-up prices)

Item	Value
Frozen range	H 210, D 200, W 10, M 205
Signal candle	low 199.80, close 201.10
RSI	27.5 → 31.2 (crossed up through 30) ✓
Entry (next open)	201.15
Stop: 200 − 0.50	199.50
Risk per unit incl. 0.10 costs	1.75
Net reward to 205	3.75 → about 2.1R, just passes

If it works: price reaches 205: about +2.1R. If it fails: the range turns into a downtrend; price trades through 199.50 and fills at 199.40: about −1.06R. The range is retired, with no second attempt today.

Notice how thin this margin is: a slightly later entry would fail the 2R gate. That’s the rule working, not a reason to loosen it.

Ratigan Capital rules in this playbook

• Early Is Wrong: buying while RSI is still falling is a different, untested idea.
• Never Average Down: a range break is information. Exit, don’t add.
• Patience as a Position: a non-flat market means no setup.
• Price Is the Only Truth: RSI is calculated from price; it doesn’t override a close outside the range.

Why it can fail

A range can become a trend without warning. The flat filter lags. Stacking correlated filters (boundary + RSI + flatness) can shrink the sample without adding independent evidence. The test should compare the RSI version against an otherwise identical price-only version and keep both records.

Pre-trade checklist

1. Range frozen at 00:00 UTC, flat filter and touches confirmed?
2. Range not already retired today?
3. Completed candle recovered and RSI crossed back above 30?
4. Stop, target, time stop and size written; net ≥ 2R?

Sources

• StockCharts ChartSchool: Relative Strength Index (RSI)
• StockCharts ChartSchool: Support and resistance
• StockCharts ChartSchool: Moving averages