Reigraph Research

Exit When RSI Hits 50, Not on a Fixed Date: Completing the Mean-Reversion Trading System

We tested 11 exit strategies against the same entry events across 6 assets and 20 years. Dynamic RSI and Keltner Channel exits deliver 78% win rates and Sharpe ratios above 0.90 — versus 60% win rates and Sharpe of 0.23 from the fixed 10-day hold most traders default to.

exit strategyRSIKeltner Channeltrading systembacktestingSPYQQQCOSTwin rateSharpe ratioquantitative research

The Missing Half of the System

The first four parts of this series established the entry side of a mean-reversion trading system with statistical rigor: which indicators work, which assets respond, how VIX amplifies returns, and how signal confluence improves conviction. What we had not addressed was the exit.

Most traders default to a fixed hold period — sell after 5 days, or 10 days, or 20. It’s simple and removes subjective judgment. It’s also, according to this study, close to the worst option available.

This study tests 11 exit strategies against the same entry events on the same assets, using a paired permutation methodology that controls for entry quality and isolates the exit decision. The results produce a complete, asset-specific trading rulebook.


Study Design

The Entries

Signal: RSI < 30 (and RSI+KC confluence as a secondary filter). Independent entry events only — the first day of each oversold streak, not every consecutive day.

Assets: SPY, QQQ, EEM, WMT, JNJ, COST — the six assets confirmed as mean-reverting in Part 2.

Period: 2004–2024 (~42 entries on SPY, 50 on EEM, 33 on COST over 20 years).

The Exit Strategies

CategoryStrategyLogic
Fixed holds1d, 3d, 5d, 10d, 20d, 30dExit after N calendar days, no conditions
RSI crossbackRSI ≥ 40 / ≥ 50 / ≥ 60Exit when momentum recovers to threshold (max 30-day cap)
KC midlineEMA-20 crossbackExit when price crosses above the Keltner Channel midline (max 30-day cap)
VWAP crossback50-day VWAP crossbackExit when price crosses above 50-day rolling VWAP (max 30-day cap)
HybridRSI ≥ 50 OR 10-day capExit at RSI recovery or hold maximum 10 days

Methodology

For each entry event, walk forward day-by-day and apply the exit rule. Record: return in basis points, holding period, whether the trade was profitable, and the maximum adverse excursion (worst closing price reached before exit). Then use paired permutation tests — same entry events, shuffle return assignments between strategies — to determine which dynamic exits statistically beat the fixed 10-day baseline.


Part 1: Dynamic Exits vs. Fixed Holds

The Overall Performance Table (RSI Signal, All Assets, All VIX)

Exit StrategyMean bpsWin RateAvg HoldSharpeWorst Trade
KC Midline (EMA-20)22178.2%13.2d0.55−1,247 bps
RSI ≥ 5023478.6%15.7d0.52−1,560 bps
RSI ≥ 4018476.9%7.0d0.51−927 bps
Fixed 30d23068.0%30d0.36−2,334 bps
VWAP-50 crossback17873.1%21.5d0.35−2,334 bps
RSI ≥ 6018269.1%25.9d0.34−2,334 bps
Fixed 20d16566.1%20d0.30−1,820 bps
Fixed 3d9460.7%3d0.24−1,020 bps
Fixed 1d6155.5%1d0.24−487 bps
RSI≥50 OR 10d cap11462.1%9d0.23−1,359 bps
Fixed 10d (baseline)11560.6%10d0.23−1,359 bps
Fixed 5d7860.4%5d0.19−1,138 bps

The results are unambiguous. Dynamic exits — RSI crossback and KC midline — dominate fixed holds on every meaningful metric simultaneously:

  • Win rate: 78–79% for dynamic exits vs. 55–68% for fixed holds
  • Mean return: 221–234 bps for dynamic vs. 61–230 bps for fixed
  • Sharpe: 0.51–0.55 for dynamic vs. 0.19–0.36 for fixed
  • Worst trade: significantly better for short dynamic exits (RSI≥40 worst = −927 bps vs. fixed 30d worst = −2,334 bps)

Fixed 10-day — the default that most traders use — sits at the bottom of the Sharpe ranking, delivering half the win rate of the best dynamic strategy at less than half the per-trade return.

Why Dynamic Exits Win

The mechanism is straightforward. A fixed 10-day hold exits regardless of whether the mean-reversion has occurred. Sometimes the bounce happens on day 3 and you hold through the recovery and a subsequent pullback. Sometimes the bounce takes 18 days and you exit at day 10 before capturing most of the gain.

Exiting when RSI recovers to 50 solves both problems: you hold until the trade has resolved (RSI has mean-reverted from extreme oversold to neutral), and you exit rather than overstaying when the momentum has normalized. The exit condition is adaptive to each trade’s actual dynamics.

The Paired Permutation Tests

34 paired tests statistically beat the fixed 10-day baseline (one-sided, α = 0.05). The top results:

AssetSignalExitStrat bpsBase bpsDiffp-value
EEMRSIRSI ≥ 50+411+25+3870.000
COSTRSIFixed 30d+564+257+3070.010
SPYRSIFixed 30d+446+139+3070.000
EEMRSIKC Midline+323+25+2990.000
SPYRSIVWAP-50+421+139+2820.000
SPYRSIRSI ≥ 50+354+139+2150.000
WMTRSIRSI ≥ 60+277+116+1610.000

EEM RSI<30 held until RSI≥50 averages +411 bps per trade. The same entry held to fixed 10-day averages +25 bps. The exit accounts for most of the edge.


Part 2: The Asset-Specific Rulebook

SPY: Exit at RSI ≥ 50 — 95% Win Rate

ExitMean bpsWin RateAvg HoldSharpe
RSI ≥ 50+35495%12.5d0.735
KC Midline+26910.8d0.610
Fixed 30d+44630d0.576
Fixed 10d+13910d0.227

95% win rate. Over 20 years, 42 RSI<30 entries on SPY — 40 closed profitable when exiting at RSI≥50. The two losses were brief episodes where recovery was slow and the 30-day cap was hit. RSI≥50 exit on SPY is the most reliable single trade setup in this research series.

Average hold of 12.5 days means you’re typically in the trade for about two and a half weeks. Not a day trade, not a multi-month hold — a medium-term mean-reversion capture.

QQQ: Exit the Next Day — Don’t Hold

ExitALL VIX bpsVIX < 20VIX ≥ 20Sharpe
Fixed 1d+138+30+1670.497
RSI ≥ 40+197+180+2010.42
RSI ≥ 50−17+269−94−0.02
Fixed 20d−49+379−163−0.05

QQQ is the outlier in the study. The RSI≥50 exit — which works brilliantly everywhere else — produces −17 bps on QQQ. Negative. The VWAP-50 exit is −35 bps. Fixed 20-day is −49 bps.

The reason: when QQQ hits RSI<30, it often takes so long to recover to RSI≥50 that the max-hold cap forces exit at a loss. In VIX≥20 environments specifically, QQQ oversold conditions frequently persist or worsen before recovering, meaning holding through 15+ days captures the early bounce and then gives it back.

The 1-day exit on QQQ is the correct approach: +138 bps average, 71% win rate, Sharpe 0.497. The day-1 bounce after extreme oversold is reliable. The multi-week recovery is not.

Critically, RSI≥40 on QQQ at +197 bps with Sharpe 0.42 works better than RSI≥50 because RSI≥40 is a faster exit — average hold of 5.5 days — that captures the initial bounce without overstaying.

EEM: Highest Mean Return — Exit at RSI ≥ 50

ExitMean bpsWin RateAvg HoldSharpe
RSI ≥ 50+41184%16.5d0.828
RSI ≥ 40+3186.4d
KC Midline+32313.7d

Emerging markets produce the highest raw returns per trade in the dataset. EEM oversold conditions (50 entries over 20 years) followed by RSI≥50 exit: +411 bps average, 84% win rate. The Sharpe of 0.828 is exceptional.

EEM’s higher volatility means oversold conditions are more extreme and recoveries more powerful. The same dynamic that makes EEM dangerous in a trending bear market makes it productive in a mean-reversion framework.

COST: KC Midline Exit — Sharpe 0.906

ExitMean bpsWin RateAvg HoldSharpe
KC Midline+34782%12.5d0.906
RSI ≥ 50+32215.2d
Fixed 30d+56430d

COST with KC Midline exit is the highest per-trade Sharpe in this entire research series: 0.906. Average hold 12.5 days, +347 bps, 82% win rate.

The KC midline (EMA-20) works especially well on COST because the stock’s low volatility means price crossing back above the 20-day EMA is a reliable signal that the dislocation has fully resolved — not an early exit that leaves gains on the table. Fixed 30-day produces more absolute return (+564 bps) but with higher variance and a lower Sharpe.

WMT: RSI ≥ 50 or RSI ≥ 60 — Both Work

WMT RSI≥50: +245 bps, 83% win rate, hold 15.2d. RSI≥60 adds another +32 bps (+277 bps) at the cost of a longer hold (26d). Both are statistically confirmed better than fixed 10-day. For capital efficiency, RSI≥50 is the better choice — same win rate in fewer days.

JNJ: Fixed 30-Day — Mean Reversion Is Slow

JNJ is the outlier among defensives. Best exit is fixed 30-day at +189 bps and 60% win rate — lower than every other asset. JNJ’s mean-reversion when oversold is real but slow. RSI recovery to 50 can take 30+ days on a stock with very low baseline volatility. The lower win rate at 60% reflects that JNJ’s “oversold” conditions sometimes reflect genuine sector headwinds (pharmaceutical litigation, dividend pressures) that take longer to resolve.


Part 3: VIX and Exit Strategy

VIX < 20 (Calm Markets)

ExitMean bpsWin RateSharpe
KC Midline15377.4%1.03
RSI ≥ 4010376.4%1.02
RSI ≥ 5016877.3%1.01

In calm markets, all three dynamic exits produce Sharpe ratios above 1.0. Returns are smaller (153–168 bps vs. 221–234 bps overall), but variance is also much lower, pushing Sharpe above 1.0. In calm markets, the oversold condition is more precisely timed and recoveries are more predictable.

VIX ≥ 20 (Elevated Fear)

ExitMean bpsWin RateSharpe
KC Midline28380.7%0.69
RSI ≥ 5029481.1%0.64
RSI ≥ 4024078.1%0.62

In elevated VIX environments, returns roughly double but so does variance. Sharpe drops to 0.62–0.69 compared to 1.01–1.03 in calm markets. The larger returns come with less predictable timing — recoveries take longer and sometimes encounter secondary drops before resolution.

Practical implication: In calm markets (VIX<20), RSI<30 entries with dynamic exits are highly consistent — use standard position sizing. In VIX≥20, the expected return is larger but timing is less certain — consider slightly smaller position sizes or the faster RSI≥40 exit to reduce variance.


The Complete Trading System

This series has now produced a fully specified, statistically grounded trading system. No indicator discretion. No manual judgment on when to exit. Every parameter is data-derived.

Entry Rules

  1. Calculate 14-period RSI on daily close
  2. Signal fires on the first day RSI crosses below 30 (new entry only — not consecutive oversold days)
  3. Optional confirmation: Keltner Channel lower touch active on the same day (RSI+KC confluence). Produces roughly same number of entries, marginally higher effect size.
  4. Check VIX. Above 20? Position size normally or slightly larger. Above 30? Tier 1 entry — maximum conviction.

Exit Rules (Asset-Specific)

AssetExit ConditionMax Hold
SPYRSI closes ≥ 5030 days
EEMRSI closes ≥ 5030 days
WMTRSI closes ≥ 5030 days
COSTPrice closes above EMA-20 (KC midline)30 days
QQQClose of next trading day (fixed 1d)
JNJFixed 30 calendar days

Expected Performance (per trade, pre-cost, 20-year historical average)

AssetExpected ReturnWin RateAvg HoldPer-Trade Sharpe
SPY+354 bps95%~12d0.74
EEM+411 bps84%~17d0.83
COST+347 bps82%~12d0.91
WMT+245 bps83%~15d0.78
QQQ+138 bps71%1d0.50
JNJ+189 bps60%~29d0.40

Trade Frequency

Approximately 2–3 RSI<30 entries per year per asset on average. In a normal year with VIX<20 throughout, you may see 1–2 entries. In a crisis year (2008, 2020), you may see 5–8 entries. Across the full 6-asset portfolio, expect 12–18 entries per year — roughly one per month.


Limitations

Win rates are historical. The 95% win rate on SPY RSI<30 → RSI≥50 exit is based on 42 trades over 20 years. The confidence interval around 95% is wide. Future win rates will vary.

Pre-cost: All returns are pre-transaction-cost. For SPY, which is highly liquid and holds ~12 days on average, trading costs are minimal relative to +354 bps expected returns. For COST, which is less liquid, slippage during a panic entry (when spreads widen) may reduce the edge modestly.

The max 30-day cap matters: The RSI≥50 exit for SPY has hit its 30-day cap approximately twice in 20 years. Both times produced losses. The cap is necessary to prevent holding through extended downtrends where mean-reversion stalls.

Correlated entries: Multiple assets may signal simultaneously during broad market panics (SPY, QQQ, EEM, WMT all oversold at once in March 2020). A portfolio applying this system to all 6 assets will face concentrated risk during crisis entries — not independent bets. Risk management across the portfolio matters as much as individual trade rules.

The QQQ 1-day rule is VIX-independent: In both VIX<20 (+30 bps) and VIX≥20 (+167 bps), the 1-day exit on QQQ is positive. The VIX-conditional logic from Part 4 still applies to position sizing — but the exit rule stays the same regardless.


Series Conclusion

Across five studies and over 2,000 statistical tests, this research series has produced a complete, evidence-based mean-reversion trading framework:

What to trade: SPY, QQQ, EEM, WMT, COST, JNJ (mean-reverting). Not NVDA, NFLX, BAC (momentum/macro-driven).

When to enter: RSI < 30. Optional confirmation: Keltner Channel lower touch. VIX > 20 amplifies returns; VIX > 30 is the highest-conviction entry.

What to avoid on entry: SMA crossovers (Golden Cross/Death Cross), Parabolic SAR, Rate of Change — statistically indistinguishable from noise across all 20 years of testing.

How to exit: RSI recovers to ≥ 50 (SPY, EEM, WMT), KC midline crossback (COST), fixed 1-day (QQQ), fixed 30-day (JNJ). Dynamic exits produce 78% win rates vs. 60% for fixed 10-day holds, at more than double the per-trade Sharpe ratio.

The system is simple enough to execute mechanically. No discretion required on the exit. No prediction of where the market is going. Only a measurement of where it is relative to its own recent history — and a willingness to hold until it returns.


This is Part 5 of the Reigraph Research Technical Indicator Series. Part 1: Do Technical Indicators Actually Work? 856 Tests Part 2: Same Signal, Opposite Result — Growth vs. Defensive Stocks Part 3: VWAP + RSI Four-Quadrant Study Part 4: VIX Is the Real Signal — Confluence and Fear Amplification


Reigraph Research · May 2026

Not investment advice. All results are pre-transaction-cost, based on 20-year historical data, and assume no slippage. Past win rates and Sharpe ratios do not guarantee future performance. Correlated entries across multiple assets in crisis conditions represent concentrated portfolio risk that individual trade statistics do not capture.