RSI Explained: What It Measures and What It Misses
RSI is a momentum oscillator that turns recent average gains and losses into a 0-100 reading. See what it measures, how it is calculated, and where it breaks.
By Imperial Analytics
The Relative Strength Index, or RSI, is one of the most widely cited technical indicators in retail trading and one of the most frequently misread. It compresses recent price action into a single number between zero and one hundred, and that compression is both its appeal and its limit. This post defines what RSI actually measures, walks through the math from the source, and names the four conditions where the reading quietly stops being useful.
By Imperial Analytics
What RSI actually measures
RSI is a momentum oscillator. It compares the average size of recent up closes to the average size of recent down closes over a defined lookback window, then maps the ratio to a zero-to-one-hundred scale. A reading near one hundred means the lookback window was dominated by up closes that were also relatively large. A reading near zero means the opposite. The indicator was introduced by J. Welles Wilder Jr. in 1978.1
Two things to hold separate from the start. RSI measures the recent balance of up and down closes. It does not measure absolute price level, distance from a moving average, volume, or anything about market structure outside the lookback window. It is a one-dimensional view of price, narrow on purpose.
The default lookback Wilder published is fourteen periods. On a daily chart that is fourteen sessions. On a five-minute futures chart it is the prior seventy minutes of bars. The choice of period changes the reading materially, which matters for any trader running RSI on more than one timeframe.
How RSI is calculated
RSI is built from two running averages: an average of up-close magnitudes and an average of down-close magnitudes over the lookback period. Wilder's smoothing is a modified moving average that gives the most recent bar a small weight on top of the prior smoothed value. The ratio of average gain to average loss becomes relative strength, and the formula RSI = 100 - 100 / (1 + RS) maps that ratio to the zero-to-one-hundred range.1
The walk-through, in order. For each bar in the window, take the close minus the prior close. If the difference is positive, it is the bar's gain and the loss for that bar is zero. If negative, the absolute value is the bar's loss and the gain is zero. Average the fourteen gains and the fourteen losses separately. Call those AvgGain and AvgLoss. Then RS = AvgGain / AvgLoss and RSI = 100 - 100 / (1 + RS).
On the bar after the first reading, Wilder's smoothing applies. The new AvgGain becomes (prior AvgGain × 13 + current gain) / 14, and the new AvgLoss is the same formula on the loss side. That smoothing is why a RSI series calculated with a simple moving average will not match a platform that uses Wilder's smoothing, and why the two can produce different overbought and oversold signals on the same price series.
Data note
The arithmetic above is the methodology. Any numerical example in this post is illustrative. Imperial Analytics only surfaces pattern claims on a trader's own data when the sample meets the minimums defined in the AI Operating Charter: twenty trades in the matching condition for behavioral patterns, fifteen for time-of-day claims, and ten for day-of-week claims.
A concrete check. If the last fourteen bars show seven up closes averaging six points each and seven down closes averaging two points each, then AvgGain is 7 × 6 / 14 = 3.0 and AvgLoss is 7 × 2 / 14 = 1.0. Relative strength is 3.0 / 1.0 = 3.0. RSI is 100 - 100 / 4 = 75. A reading of seventy-five lands above the conventional overbought threshold of seventy. The up closes were both more frequent in proportion and three times the size of the down closes, on average.
What overbought and oversold really mean
Overbought and oversold are conventions, not predictions. Wilder used seventy and thirty as default thresholds because they captured the upper and lower tails of the RSI distribution on the markets he studied. The terms describe an extreme reading on the indicator, not a forecast of reversal. An overbought RSI on a strong uptrend says recent up closes dominated recent down closes, which is also a fair description of a healthy trend in progress.1
The language is a frequent source of misreads. "Overbought" sounds like a verdict on price, as if the market has gone too far and is due to come back. RSI cannot make that claim. What it can say is that the recent balance of closes is unusually one-sided. Whether that one-sidedness is a sign of exhaustion or of trend strength depends entirely on what is happening outside the indicator's lookback window.
Two trader-relevant adjustments are common. In strong trends, traders shift the thresholds to eighty and twenty to filter out the routine extremes that show up in trending regimes. On shorter lookback periods, such as a RSI computed over seven bars, the reading is more sensitive and crosses the conventional levels more often, so the bands often have to widen with it. None of these adjustments is canonical. They are tuning choices that should be tested on the trader's own log, not adopted on faith.
↳ Note
Overbought does not mean wrong. It means the recent closes leaned one way. Whether that lean is exhaustion or strength is a question the indicator cannot answer.
Where RSI breaks down
RSI fails in four well-known conditions. It stays pinned above seventy through long stretches of strong trend, so the overbought reading is not a sell signal. It oscillates noisily around fifty in chop, so it produces many false crossings. It is sensitive to the lookback period, so different defaults disagree on the same price series. And it is a strict function of past closes, so it can only confirm what price has already done.
The trend case is the most expensive one. A futures contract in a clean trend can hold RSI above seventy for ten, twenty, or more consecutive bars. A trader who takes every cross above seventy as a sell signal is short the strongest part of the move and pays for the position twice, once on the entry and again on the stop. The standard adjustment is to require additional confirmation, such as a swing high, a break of a moving average, or a divergence pattern, before treating an overbought reading as actionable. None of those confirmations is part of RSI itself.
The chop case shows up in the middle of the trading session, especially on lower timeframes around lunch and the post-open consolidation. RSI hovers in the forty-to-sixty band and crosses fifty repeatedly without producing a directional move. Each cross looks like a signal and almost none of them pays. The honest reading of RSI in this regime is that the indicator has no information to add. The trader has to look elsewhere or stand aside.
The lookback sensitivity case is easy to miss. A seven-period RSI on a five-minute ES chart can be at twenty-five while a fourteen-period RSI on the same chart sits at forty. Both are real readings. Neither is wrong. They answer slightly different questions about slightly different windows. A trader who has not picked one period and committed to it will find themselves reading whichever number supports the position they already want to take.
The confirmation-lag case is the structural one. RSI is computed from closes that have already printed. It cannot lead price. The most it can do is summarize what just happened. Any claim that RSI is predictive at the bar level is a claim about how the next close will compare to the prior fourteen closes, which is a hard distribution to forecast from the indicator alone.
How to use RSI without leaning on it
RSI earns its place in a futures trader's process by being one input among several, with the lookback period fixed in advance and the signals measured against the trader's own outcomes. A useful workflow is to tag every trade in the journal with the RSI reading at entry and the RSI reading at exit, then check whether trades taken in different RSI bands have different expectancies. That check turns the indicator from a belief into a measurement.
The principle is the same one that applies to any indicator. The indicator is not the edge. The edge is whatever pattern the trader's own results reveal when the indicator is used inside a defined process. If RSI entries in the thirty-to-forty band produce a positive expectancy on the trader's setup and entries in the sixty-to-seventy band do not, that is information worth keeping. If both bands produce roughly the same expectancy after enough trades, that is also information, and the right response is to stop weighting RSI in the decision.
The minimum journal fields to enable this check are the timestamp and instrument of the trade, the planned setup, the RSI reading and period at entry, the RSI reading and period at exit, and the realized P&L. Without those fields the question cannot be answered from the data. With them, the question can be answered honestly the first time the sample reaches the threshold for a behavioral pattern claim, which is twenty trades in the matching condition under the AI Operating Charter.
Frequently asked questions
Frequently asked questions
- q: Does RSI anticipate price moves or only react to them? a: RSI is a reactive measure because it is derived from closes that have already printed. It does respond quickly to recent changes in close balance, which is why some traders treat it as a faster signal than long moving averages, but it cannot anticipate the next close from its own readings.
- q: What is the difference between Wilder's RSI and a simple moving average RSI? a: Wilder's RSI uses a modified moving average that weights the most recent bar against the prior smoothed value, while a simple moving average RSI uses an unweighted mean of the lookback gains and losses. The two methods will produce different absolute readings on the same price series, and they can disagree on whether a threshold has been crossed.
- q: Does RSI work on intraday futures charts? a: RSI is defined on any series of closes, so it can be computed on a one-minute, five-minute, or hourly futures chart. The interpretation does not change with the timeframe, but the noise level does. Shorter timeframes produce more frequent threshold crossings and more false signals, which is why many intraday traders widen the bands or lengthen the lookback period when applying RSI to a five-minute MES or MNQ chart.
- q: How many bars does RSI need before the reading is reliable? a: Wilder's formula requires the lookback period to seed the first AvgGain and AvgLoss. After that, the smoothing recursion means the reading continues to absorb the influence of the seed for many bars. A common rule is to wait at least three lookback periods past the seed before treating the reading as stable, which for a fourteen-period RSI is roughly forty-two bars of warm-up.