What the Oscillator Is Measuring
RSI is a momentum oscillator measuring the speed and change of price movements, and it is bounded: the value always sits between zero and 100.
The calculation compares the average magnitude of recent gains against the average magnitude of recent losses over a lookback window. When gains dominate, the value rises toward 100. When losses dominate, it falls toward zero. A stock rising steadily by small amounts and one rising violently can produce quite different readings.
Wilder's default lookback is 14 periods. On a daily chart that is roughly three trading weeks.
Why the smoothing method is not arbitrary
Wilder specified an exponential smoothing of the up and down closes rather than a simple average. The reason was practical rather than theoretical: in 1978 the calculation had to be done without a computer, and exponential smoothing requires only the previous value rather than storing the full history.
That detail matters because it means two charting packages can produce slightly different RSI values for the same stock if one uses simple averaging. The number is a convention, not a physical measurement.
Working the Calculation Through
The mechanics are worth seeing once, because they explain why the indicator behaves as it does.
Over the lookback window, each period's close is compared with the previous one. Periods that closed higher contribute their gain; periods that closed lower contribute their loss. The two are averaged separately, and the ratio between them becomes the reading.
What the midpoint actually means
Fifty is the level at which average gains and average losses over the window are equal. It is not a neutral reading in any deeper sense, and crossing it is not an event — it simply means the balance of recent closes has tipped.
Two consequences follow from the structure. The indicator is bounded, so it cannot express "extremely strong" beyond 100 and compresses at the extremes: the difference between 85 and 95 represents a far larger shift in the underlying ratio than the difference between 45 and 55. And it uses closes only, so intraday range is invisible to it. A day that fell sharply and recovered to close flat contributes nothing at all.
The Thresholds Are the Weakest Part
The 70 and 30 levels are conventions Wilder proposed, not properties of markets. Their weakness is structural.
Those last two rows undo the first two. If RSI in an uptrend typically ranges between 40 and 90, then a reading of 75 in an uptrend is ordinary rather than extreme, and selling on it means selling into strength. Fidelity notes the traditional levels can be adjusted for securities that repeatedly reach them, which is an admission that fixed thresholds do not transfer between stocks.
The label "overbought" implies a judgment the arithmetic does not make. RSI knows nothing about value, earnings or news. It reports that recent up-closes have outweighed recent down-closes over fourteen periods. Everything beyond that is interpretation layered on top.
Divergence Is the More Defensible Signal
The reading practitioners take more seriously is divergence: price making a new high while RSI fails to, or price making a new low while RSI does not follow.
The logic is internally consistent. If a stock reaches a higher price on weaker momentum than the previous advance, the buying pressure behind the move has thinned even though the price has not yet turned.
Fidelity treats divergences and swing failures as potential reversal indications, and notes separately that divergence is a stronger signal when it confirms other signals rather than standing alone. That qualification is the operative part.
What divergence cannot tell you
It gives no timing. Momentum can weaken for weeks while price continues higher, and a divergence that eventually resolves into a reversal is indistinguishable, at the moment it appears, from one that resolves into a continued advance.
Changing the Lookback Changes the Indicator
The 14-period default is a choice, and shortening or lengthening it produces a different tool.
A shorter window reacts faster and generates far more threshold crossings, most of which lead nowhere. A longer window produces fewer signals, each arriving later. There is no setting that gives early signals without also giving false ones, because the tradeoff is arithmetic rather than a matter of finding better parameters.
This is where testing different settings until one fits past data becomes self-defeating. A lookback chosen because it worked on the last two years describes those two years.
One Indicator Is Not a Method
The deeper problem with any single oscillator is that it observes one dimension of price and nothing else.
RSI cannot see volume, so it cannot distinguish a move made on heavy participation from one made on almost none. It cannot see the balance sheet, the industry, or why the stock is moving. It cannot see whether an advance followed an earnings release or happened for no discoverable reason. Every one of those distinctions changes what a price move means, and none of them enters the calculation.
This is why practitioners who use it seriously treat it as one input among several rather than a decision rule. Fidelity makes the same point about its sibling indicators, noting they are not intended to be used on their own and work best confirming signals given by others.
What Regulators Say About Trading on Signals
Any indicator used to trade frequently runs into the costs and the base rates that FINRA requires firms to disclose.
Under FINRA Rule 2270, firms must warn customers that day trading can be extremely risky, that they should be prepared to lose all funds used for it, and that it should not be funded with retirement savings, student loans or emergency funds. Customers must be told they will compete with professional, licensed traders employed by securities firms.
The rule also requires caution about advertisements emphasizing the potential for large profits, and its own example illustrates the cost hurdle: a trader in that scenario would need $111,360 of annual profit merely to cover commissions.
Before acting on any oscillator reading, compute what the round-trip costs. The spread is charged on entry and exit regardless of whether the signal was right, and a strategy generating frequent signals pays it repeatedly. An indicator that is correct slightly more often than not can still lose money once execution costs are subtracted.
Where the Reading Becomes Unreliable
- Strong trends. The condition under which the thresholds mislead most is precisely the condition traders most want a signal for.
- Thin securities. In stocks with little liquidity and wide spreads, a few trades can move the reading without reflecting genuine pressure.
- Gaps and halts. RSI is computed from closes; a stock that gaps on news produces a reading describing an event the indicator cannot see.
- Volatility bands. Limit up-limit down constrains what prices can print at all, so the input series is bounded by rules rather than sentiment alone.
- Charting differences. Different smoothing implementations give different values for the same stock.
Why It Persists Despite the Objections
Given the caveats, the indicator's durability is worth accounting for rather than dismissing.
Part of it is genuine: a bounded, normalized momentum measure lets one stock's current state be compared against its own history without regard to price level, which raw price change does not allow. A $12 stock and a $1,200 stock produce comparable RSI readings.
Part of it is self-fulfilling. Enough participants watch the same levels on the same default settings that reactions cluster around them, which makes the levels somewhat real regardless of whether they had any predictive basis to begin with. That is a genuine market phenomenon rather than an argument that the indicator measures something fundamental.
And part of it is simply that it is legible. A single number between zero and 100 is easier to act on than an assessment of a business, which is a reason for popularity rather than a reason for confidence.
Using It Without Overreading It
The defensible uses are narrower than the popular ones. RSI describes recent momentum compactly, which is useful as context. It flags divergence, which is a real observation about the internal consistency of a move. And it allows one stock's current momentum to be compared against its own history.
What it does not do is identify tops and bottoms. The most common way to lose money with this indicator is to treat 70 as an instruction to sell in a market that is rising for reasons the oscillator cannot observe.
The Bottom Line
RSI is a bounded momentum measure with a 14-period default and two conventional thresholds that fail in trending markets — which is where most people reach for them.
Treat it as context rather than instruction, use divergence in preference to threshold crossings, adjust the levels to the security rather than assuming 70 and 30 transfer, and price in the transaction costs before acting on anything. The indicator answers a narrow question about recent price behavior, and answers it honestly. The overreach is entirely in the interpretation.
For a trend-following counterpart, see MACD; for what each signal costs to act on, see the spread and order book depth.
