RSI and Stock Price Correlation, Tested With a Real Backtest
I tested the common advice — "buy when RSI drops below 30, sell when it rises above 70" — against real Korean stock market data to see how strong the RSI–price correlation actually is. Pushing the thresholds to more extreme levels lowered the win rate but actually raised the average return, and "waiting to confirm a turnaround before buying" showed no clear edge on daily bars. Apply the exact same logic to 5-minute bars, though, and most of that edge disappears.
00 How I Tested This
Using RSI (Wilder's method, 14-period), I tested a mean-reversion strategy that buys in oversold territory and sells in overbought territory, rotating across many stocks. Every time a signal fired for any stock, I assumed the same fixed amount (1,000,000 KRW) was invested, holding only that one position until it sold.
- Universe: only stocks/dates with a market cap of 100 billion KRW or more, and an average trading value over the last 5 sessions of 500 million KRW or more
- Capital: every signal invests the same fixed 1,000,000 KRW — not a percentage of total assets. Each trade is completely independent with no capital constraint, so every signal that fires always executes
- Immediate Entry: buy the instant RSI crosses down through the buy
threshold, then sell the instant RSI crosses up through the sell threshold
(e.g. RSI 20/80 — buy the moment RSI falls from above down through 20, then sell the moment RSI rises from below up through 80) - Turnaround Confirmation: buy on the day RSI, having dropped below
the buy threshold, climbs back above it; sell on the day RSI, having spiked above the
sell threshold, falls back below it
(e.g. RSI 20/80 — buy the moment RSI dips below 20 and then climbs back above 20, sell the moment RSI spikes above 80 and then falls back below 80. Unlike immediate entry, this waits for one confirmed reversal before acting) - Any position that never hit its sell condition by the end of the test period was force-liquidated at the last available closing price (fees included, reflected in every statistic — win rate, average return, and so on)
- A 0.35% fee was applied only on the buy side; the sell side was assumed to carry no fee
01 Do More Extreme RSI Thresholds Work Better?
Immediate entry, comparing 20/80, 30/70, 40/60 (daily bars, 2020–2026)
From here on, thresholds are written as a pair like "20/80" or "30/70" — the first number is the buy threshold, the second is the sell threshold. So "RSI 20/80" means "buy when RSI falls from above down through 20, and sell when it rises from below up through 80" (immediate-entry rule).
The first question worth asking: does buying at a more extreme oversold level, like RSI below 20, actually produce better results? I compared all three thresholds under identical conditions, using the immediate-entry rule throughout.
| Threshold | Trades | Forced Liq. | Win Rate | Avg. Return | Avg. Hold |
|---|---|---|---|---|---|
| RSI 20/80 | 1,619 | 328 (20%) | 46.94% | +2.73% | 565.4 days |
| RSI 30/70 | 7,749 | 902 (12%) | 56.94% | +2.10% | 174.6 days |
| RSI 40/60 | 22,596 | 1,212 (5%) | 60.86% | +0.80% | 68.9 days |
02 Buy Immediately, or Wait for a Turnaround — Is There a Winner?
Immediate entry vs. turnaround confirmation, compared at all three thresholds (daily bars)
Instead of buying the instant RSI first touches the threshold, what changes if you wait for RSI to touch the threshold and then cross back the other way before acting? I applied the same "confirm the reversal first" logic to the sell side too — waiting for overbought RSI to touch its threshold and turn back down.
| Combo | Style | Win Rate | Avg. Return |
|---|---|---|---|
| RSI 20/80 | Immediate Entry | 46.94% | +2.73% |
| RSI 20/80 | Turnaround Confirmation | 46.09% | +2.83% |
| RSI 30/70 | Immediate Entry | 56.94% | +2.10% |
| RSI 30/70 | Turnaround Confirmation | 54.92% | +1.37% |
| RSI 40/60 | Immediate Entry | 60.86% | +0.80% |
| RSI 40/60 | Turnaround Confirmation | 58.54% | +0.61% |
03 Does What Works on Daily Bars Also Work on 5-Minute Bars?
Same RSI(14), same 6 combinations, applied to 5-minute bars (5-min sample ~1 yr vs. daily sample ~6 yr)
I kept the RSI period at 14 and changed only the bar size, from daily to 5-minute. Same number, wildly different time window — 14 daily bars span roughly 3 weeks, while 14 five-minute bars span just 70 minutes.
| Combo | Daily Avg. Return | 5-min Avg. Return |
|---|---|---|
| 20/80 Immediate | +2.73% | +0.03% |
| 20/80 Turnaround | +2.83% | -0.43% |
| 30/70 Immediate | +2.10% | -0.08% |
| 30/70 Turnaround | +1.37% | -0.44% |
| 40/60 Immediate | +0.80% | -0.12% |
| 40/60 Turnaround | +0.61% | -0.44% |
04 Takeaways
- Win rate and average return move in opposite directions. Extreme thresholds like RSI 20/80 have the lowest win rate (46.94%) but the highest average return (+2.73%) — a structure of winning rarely but winning big.
- The more extreme the oversold level, the higher the risk of a position that "never recovers." Stocks that fall to RSI 20 have the highest forced-liquidation rate (20%) and the longest average holding period (565 days).
- "Waiting to see a turnaround" isn't always better on daily bars. Turnaround confirmation only won at 20/80 — immediate entry won at both 30/70 and 40/60.
- Changing the timeframe erases most of the edge. Applying the same RSI(14) logic to 5-minute bars shrinks the daily-bar gains toward zero or negative, and turnaround confirmation was worse than immediate entry at all three thresholds on 5-minute bars.
댓글 쓰기