📅 Originally Published: · Last Updated: · Educational analysis, not personalized investment advice.
An RSI overbought signal above 70 is not an automatic sell order. It tells you that recent gains have been strong relative to recent losses. Whether that strength is about to reverse depends on the asset, trend, holding period, and the exact trading rule being tested.
A five-market study of the canonical RSI(14, 30/70) rule found negative buy-minus-sell ten-day returns in three markets and positive results in two. Only the negative results for Milan and Germany were statistically significant, both at the 10% level. That is evidence against treating 70 as a universal rule, not proof that RSI is useless everywhere.
You open a chart, see RSI at 72, and get an uncomfortable feeling that you are late. Selling can feel prudent because the platform has already labeled the reading “overbought.” That label describes momentum, while the decision also depends on your time horizon, tax cost, position size, and the strength of the broader trend.
The decision is therefore narrower than “Does RSI work?” A better question is: what additional evidence would justify changing this position now? The answer starts with understanding what the 70 line measures and what the studies actually tested.
What RSI 70 actually means
The Relative Strength Index compares average gains with average losses over a chosen lookback period and maps the result to a 0-to-100 scale. A 14-period lookback is the common default. Readings above 70 are traditionally labeled overbought, while readings below 30 are labeled oversold.
Plain meaning: RSI above 70 says recent upward momentum has been unusually strong. It does not estimate fair value, predict a specific decline, or tell you how large any reversal might be.
That distinction matters because a strong trend can remain strong. TradingView’s own RSI documentation presents 70 and 30 as traditional reference ranges, notes that traders alter them, and warns against making a directional assumption from one number alone.
The signal also depends on how a rule is defined. In Chong, Ng, and Liew’s 2014 study, the RSI(14, 30/70) rule did not sell the instant RSI moved above 70. A sell signal occurred after RSI had moved above 70 and then crossed back below it. The researchers then measured the following ten-day return and ignored additional signals during that ten-day window.
📚 Sources: TradingView RSI documentation · Chong, Ng, and Liew (2014)
This analysis is for: investors and traders who are considering a sale because a standard chart shows RSI above 70.
It does not settle: intraday systems, options strategies, individually optimized thresholds, or rules that combine RSI with trend, volatility, valuation, or risk controls.
Mistake 1: selling as soon as RSI crosses 70
A move above 70 marks a condition. The canonical academic rule discussed here waited for RSI to fall back below 70 before recording a sell signal.
That timing difference is easy to miss. Many investors see the word “overbought” and act on the first crossing. The study most often cited in this article tested something more specific: a reversal back through the threshold, followed by a ten-day holding window for measuring returns.
Its Table 3E reports the average ten-day return after buy signals minus the average ten-day return after sell signals. A positive number favors the rule’s separation between buy and sell conditions. A negative number means the returns following the supposed sell condition were stronger than the returns following the buy condition.
| Market | Buy minus sell, 10-day return | Statistical result |
|---|---|---|
| Milan Comit General | -1.025% | Significant at 10% |
| S&P/TSX Composite | +0.393% | Not significant |
| DAX 30 | -0.914% | Significant at 10% |
| Dow Jones Industrials | +0.650% | Not significant |
| Nikkei 225 | -0.083% | Not significant |
The table weakens the idea that one threshold works consistently across developed markets. It does not support the stronger claim that the rule lost to buy-and-hold in every market. The paper’s abstract even describes RSI(14, 30/70) as profitable in the Dow Jones Industrials, although the Table 3E buy-minus-sell spread did not reach conventional statistical significance.
Do not read “three of five negative” as an accuracy rate. The table does not say that RSI was wrong on 60% of trades. It compares average ten-day returns following two types of signals.
Mistake 2: turning one study into a universal verdict
The five-market evidence is useful because it shows instability across markets. Its sample ended in 2002, used ten-day signal windows, tested fixed thresholds, and reported the core Table 3E results before transaction costs. Those design choices limit what the result can prove about a current portfolio.
The safest conclusion is modest: the 70/30 default did not produce consistently positive buy-minus-sell ten-day return spreads across the five markets in that sample. A reader should not jump from that finding to “RSI never works,” just as they should not jump from a single positive market to “RSI always works.”
A 2023 cryptocurrency study adds a useful warning, but it must be described accurately. Its long-only 177.70% cumulative result came from buying in the oversold 0-to-30 range. It was not a long-only strategy that sold at 70. The study’s short-only and combined long-short versions each ended at a 100% loss in that sample, while a separate trend-oriented RSI approach produced better results in that particular crypto sample.
Those experiments support a broader lesson: the framing of RSI matters. Mean-reversion rules, short-selling rules, and trend-following rules are different strategies. Combining their returns under one “RSI 70/30 performance” label creates a comparison that the source does not make.
📚 Source: Zatwarnicki, Zatwarnicki, and Stolarski (2023), especially the strategy definitions and Tables 3 and 7.
Research check: Before relying on an indicator result, write down the entry rule, exit rule, execution delay, asset universe, sample period, costs, and benchmark. If one of those changes, you are looking at a different strategy.
Mistake 3: treating a signal spread as an annual portfolio drag
A ten-day buy-minus-sell return cannot be converted directly into a permanent annual return penalty without a complete trading simulation. The study does not provide an annual drag estimate, so any long-horizon wealth projection needs a separately defined and tested trading path.
To estimate a portfolio effect, you need the actual sequence of entries and exits, the time spent in cash, reinvestment rules, transaction costs, taxes, dividends, and what happens when signals overlap. A fixed annual haircut skips all of those steps. The future-value calculation may be arithmetically correct while its most important input remains unsupported.
As a separate reproducibility check, TheFinSense ran a simple RSI(14, 30/70) simulation on a pinned, price-only daily SPY dataset from January 2010 through December 2019. The rule began invested, moved to cash one trading day after RSI crossed back below 70, and re-entered one trading day after RSI crossed back above 30. Cash earned 0%, dividends were excluded, and each position change incurred 0.10% in the cost-aware version.
| Test | Result |
|---|---|
| Buy-and-hold cumulative multiple | 3.46x |
| RSI rule cumulative multiple, no trading cost | 2.13x |
| RSI rule cumulative multiple, 0.10% per position change | 2.08x |
| Position changes | 21 |
| Fresh crosses above 70 followed by a positive 10-day return | 34 of 46 events, or 73.9% |
The strategy underperformed buy-and-hold in this sample, but the event study also found that prices were higher ten trading days after a fresh cross above 70 in 34 of 46 non-overlapping events. That is the practical danger of treating “overbought” as “must fall now.”
This check serves as a transparent reproduction alongside the peer-reviewed study. It uses one ETF, one decade, one execution convention, and no tax model. Its value is narrower: it reproduces the core decision problem and makes the assumptions visible. The script, dataset hash, and result hash are listed in the method package below.
A practical RSI decision framework
When RSI moves above 70, pause before changing the position. Confirm the trend, define the actual sell rule, and compare the expected benefit with taxes, costs, and the role the holding serves in your plan.
1. Separate a warning from an order
“Overbought” can prompt a review. It should not decide the trade. Decide in advance whether your rule acts on the first cross above 70, a cross back below 70, divergence, a trend break, or something else. Those signals are not interchangeable.
2. Check the dominant trend
An RSI reading above 70 during a broad uptrend can reflect persistent demand rather than an exhausted move. At minimum, compare the reading with price structure and a trend measure you already understand. Adding indicators after seeing the outcome invites data snooping, so keep the rule simple and define it before testing.
3. Ask what would make the sale correct
A sale may still be sensible because the position has become too large, your thesis changed, you need the cash, or the holding no longer fits your risk plan. Those are portfolio reasons. RSI can support the timing discussion, but it should not replace the reason.
4. Include friction
In a taxable account, a sale can create capital-gains tax. A rapid re-entry can add spread and execution costs. Frequent rule changes also raise the chance that you abandon the process after a bad sequence. Test the net decision, not only the chart signal.
5. Write the rule into your process
An investment policy statement can specify when technical signals are allowed to influence position size. For a deeper look at test design, see the technical analysis backtest guide. The related MACD evidence review shows why indicator names matter less than the exact rule and sample.
Reasonable default for a long-term investor: treat RSI above 70 as a prompt to review concentration and thesis risk, not as a standalone reason to sell a diversified holding.
Reasonable default for a systematic trader: backtest the complete rule with lag, costs, an untouched out-of-sample period, and a benchmark before risking capital.
Frequently asked questions about the RSI overbought signal
Is RSI above 70 always a sell signal?
No. RSI above 70 shows strong recent momentum relative to recent losses. It can precede a pullback, but it can also persist during a strong trend. The five-market evidence is mixed, and the canonical academic rule waited for RSI to cross back below 70 before recording a sell signal.
Can RSI stay overbought for weeks?
Yes. A persistent trend can keep RSI elevated for an extended period. The label does not contain a timer, price target, or probability of reversal.
What did the five-market RSI study actually find?
For RSI(14, 30/70), the buy-signal return minus sell-signal return was negative in Milan, Germany, and Japan, and positive in Canada and the Dow. Only Milan and Germany were statistically significant, both at the 10% level.
What is the best RSI setting?
There is no universal best setting. The appropriate lookback, threshold, and confirmation rule depend on the asset, horizon, market regime, and costs. A setting chosen after reviewing the same data can look impressive for the wrong reason, so preserve an out-of-sample test.
Should long-term investors use RSI?
RSI can be a secondary review tool, but portfolio fit, valuation, diversification, taxes, and the investment thesis usually matter more. A long-term investor should not let one momentum reading override a written plan.
Bottom line: what to do when RSI moves above 70
The evidence supports a narrower conclusion: the 70 line is an incomplete decision rule. The five-market study found inconsistent results, and its exact rule was more specific than the red “overbought” label shown on a chart.
When RSI crosses 70, do one useful thing before touching the position: state the sell rule in a full sentence. Include the trigger, execution timing, trend condition, and reason the holding no longer deserves its current weight. If you cannot do that, the chart has raised a question but has not answered it.
Let RSI prompt a review, then keep the final decision tied to your written rule.
YOUR TURN
What evidence, besides RSI crossing 70, would have to change before you sold the position?
Sources, method, and evidence
- Foundational equity study: Chong, T. T.-L., Ng, W.-K., & Liew, V. K.-S. (2014). Revisiting the Performance of MACD and RSI Oscillators. Table 3E and the Rule 4 methodology were checked against the full text.
- Platform definition: TradingView RSI documentation, checked July 22, 2026.
- Crypto evidence: Zatwarnicki, M., Zatwarnicki, K., & Stolarski, P. (2023). Effectiveness of the Relative Strength Index Signals in Timing the Cryptocurrency Market. Strategy definitions were kept separate rather than merged into one 70/30 claim.
- In-house SPY check: daily data from January 4, 2010 through December 30, 2019. The rule, lag, costs, and limitations are stated in the body.
- Limits: the in-house test is a single-asset, single-decade historical simulation. It does not model dividends, taxes, cash interest, slippage beyond the stated cost, or alternative RSI calibration. Past results do not predict future returns.
Reproduction hashes
Dataset SHA-256: f1682f176f9db69ab654405b0126b5c9840f272ee12c6c5bd5c0fec2345fd6f0
Script SHA-256: f40e2125d02d7d574292da939e337587e4110f7fd20c0d0b90ca09c0df02ae9b
Result-series SHA-256: 7c3923b35c8eb93f3fe55196e2153a62544389c911b5fd2337264b13ebe284f7
📋 Update History
- July 22, 2026: Rebuilt the article around the source-supported conclusion; corrected the interpretation of Chong, Ng, and Liew Table 3E; separated the crypto study’s oversold, short, and trend strategies; removed the unsupported annual-drag scenario, calculator, and sensitivity table; added a reproducible SPY check and a decision-focused framework.
- May 12, 2026: Original publication.
Educational quantitative analysis based on published data. Not investment, tax, or legal advice. Consult a licensed professional before acting on any calculation. About TheFinSense.