RSI Trading Strategy Tested: Does Overbought Really Mean Sell?

What RSI measures

J. Welles Wilder Jr. introduced the Relative Strength Index in his 1978 book New Concepts in Technical Trading Systems, the same book that gave traders the average true range (ATR) and the parabolic SAR. RSI compares the size of recent gains with the size of recent losses over a lookback period, usually 14 bars, and scales the result from 0 to 100.

• A high RSI means recent gains have been much bigger than recent losses.
• A low RSI means the opposite.
• By convention, above 70 is "overbought" and below 30 is "oversold".

Those labels come with a built-in forecast, and the textbook RSI strategy is built on it: sell when overbought, buy when oversold. That forecast is what we tested.

How we tested it

We locked in the rules before looking at the results:

• Data: QQQ weekly closes, adjusted for the March 2000 split, from June 2000 (after a 30-week warm-up so the indicator settles) to September 2026.
• Indicator: 14-week RSI with Wilder's original smoothing.
• Signal: a weekly close with RSI above 70, or below 30.
• Horizons: the next week and the next four weeks.
• Benchmark: every week in the sample, and 10,000 random sets of weeks of the same size.

Weekly bars show the higher-timeframe trend a futures trader watches, and our data source doesn't have decades of intraday prices.

What we found

• RSI above 7047
• RSI below 309

Exact values from table 2 in this article. Bar lengths share a zero baseline; no new calculation or simulated price path. Read the table and its stated sample and limits in this article.

Overbought was common, and slightly negative. RSI was above 70 in 11% of weeks. Over the next four weeks, the average return was −0.65% against +0.81% for all weeks. In random draws, a gap that big came up about 0.1% of the time.

But overbought didn't mean down. Even after those weeks, the index went up over the next four weeks 54.3% of the time. A trader who sold every overbought reading would have been wrong more often than right.

Oversold was rare, and it bounced. RSI dropped below 30 in only 1.7% of weeks. The next four weeks averaged +4.64%, up 65.2% of the time.

The next-week result was small both ways. The one-week numbers, −0.13% and +0.15%, are within normal weekly noise. Whatever information RSI carried showed up over a month, not a week.

Why the sample is smaller than it looks

The 23 oversold weeks weren't 23 independent events. Oversold readings come in runs: once RSI drops below 30, it often stays there for several weeks. Grouped into separate episodes, there were only nine.

They were bunched up in a few crashes:

Nineteen of the 23 weeks came in 2001, 2002 and 2008. A result built on three bear markets tells you how those bear markets ended, not how the next one will.

The four-week returns also overlap. A run of five oversold weeks gives you five four-week windows that share most of their data. That makes the random-draw comparison look stronger than it really is. We report it, but we don't treat it as proof.

Level reached versus direction

The most useful lesson here is the difference between two questions:

1. Has the move gone far? RSI answers that well. A reading above 70 tells you recent gains have been unusually one-sided.
2. Will it reverse? RSI answers that badly. After an overbought reading, the index kept going up more often than not.

Strong trends produce overbought readings and stay overbought. In a bull market, RSI can sit above 70 for weeks while prices keep climbing. The 151 overbought weeks came in 47 episodes, so a typical episode lasted about three weeks.

That's why a lot of technicians use RSI as a measure of momentum, not a reversal signal: a high RSI describes strength, not exhaustion.

What this means if you trade NQ

• Don't fade overbought on its own. On weekly data, an overbought Nasdaq-100 went up over the next month more often than it fell. Shorting because "RSI is above 70" worked less than half the time.
• Treat oversold as context, not a trigger. The average bounce after oversold readings was big, but it came from nine episodes bunched up in crashes, and the ride to get there was violent.
• Pair it with trend. In our test of big down weeks, the long-term trend separated drops that bounced from drops that didn't. In our 10-month moving average test, a simple trend filter cut the worst drawdown by two-thirds. An extreme RSI reading means more when you know which regime you're in.
• Expect more volatility, not a direction. Extreme readings tend to show up in fast markets. The size of the next move is easier to anticipate than its direction.

Why so many RSI strategies look better in backtests

Search "RSI strategy" and you'll find results way better than ours. Three things usually explain the gap:

1. Parameter search. Testing RSI(2), RSI(5), RSI(14), thresholds of 10, 20, 25 and 30, then reporting the best combo. In their 2007 review of the technical analysis literature in the Journal of Economic Surveys, Cheol-Ho Park and Scott Irwin pointed to this kind of data snooping as a central problem in the field.
2. Short samples. A test that starts in 2010 never sees 2000–2002 or 2008, the periods that produced most extreme readings.
3. No benchmark. "The index went up after oversold readings" doesn't mean much on an index that went up most of the time anyway.

We used the textbook settings, 14 periods and 70/30, and didn't try others. If you test variations, write them all down and report all of them.

How to check it yourself

1. Open a weekly chart of QQQ or NQ and add RSI with the default 14 periods.
2. Mark every week that closes with RSI above 70 or below 30.
3. For each mark, write down the close four weeks later.
4. Compare with what the index did over all four-week periods in the same span.

Limits

• Weekly bars only. RSI on 5-minute or daily charts is a different signal, and this test doesn't cover it.
• Overlapping windows. Four-week returns from back-to-back weeks share data, which overstates the statistical strength.
• Few oversold episodes. Nine episodes can't establish a reliable rule.
• Price-only four-week returns. The four-week numbers leave out dividends. At QQQ's yield, the effect is small.
• One index, one period. The Nasdaq-100 recovered from every decline in this sample. That's not guaranteed.

How we did it

• QQQ weekly prices from Alpha Vantage (TIME_SERIES_WEEKLY), adjusted by us for the two-for-one split of March 20, 2000.
• RSI: 14 weeks, with Wilder's smoothing (an exponential average with a weight of 1/14 on gains and losses).
• Episodes: consecutive weeks in the same zone counted once.
• Random comparison: 10,000 draws of the same number of weeks, with a fixed seed.

Sources

• Wilder, J. W., Jr. (1978). New Concepts in Technical Trading Systems. Trend Research.
• Park, C.-H., & Irwin, S. H. (2007). What do we know about the profitability of technical analysis? Journal of Economic Surveys, 21(4), 786–826. IDEAS/RePEc
• Alpha Vantage. TIME_SERIES_WEEKLY (QQQ). Data retrieved October 2026.

FAQ

Does the RSI trading strategy work? Not reliably. After weekly RSI above 70, the Nasdaq-100 went up over the next four weeks 54% of the time, even though the average return was slightly negative.

Does RSI oversold mean the market will go up? After weekly RSI below 30, the next four weeks averaged +4.64%, but there were only nine episodes, mostly in 2001, 2002 and 2008.

What's the best RSI setting? There's no proven best. We used Wilder's default of 14 periods. Picking the setting that worked best in the past usually overstates future results.

Is RSI better on weekly or daily charts? They answer different questions. This test only covers weekly bars.

How often is the Nasdaq-100 overbought on weekly RSI? About 11% of weeks from 2000 to 2026, usually in runs of a few weeks.

---

Research and education only. Not financial advice. Past results do not predict future returns. A backtest is not real performance.

How this article was made: drafted with AI assistance from public data and the sources listed above; facts and figures checked by Gabriel Décarie and Karel Panneton.