Indicators & Oscillators
Oscillators and their lies
"RSI is above 70, it's overbought, sell." That sentence has cost retail traders more money than any other single idea in technical analysis — not because RSI is broken, but because "overbought" does not mean what the word sounds like. It does not mean expensive. It means the recent buying has been unusually one-sided, which in a range is a warning and in a trend is a description of strength.
Oscillators are normalised: they take the last N bars and express today's position INSIDE that window as a number from 0 to 100. Stochastic %K asks literally "where did we close within the recent high-low range?" — 90 means near the top of the last N bars, 10 near the bottom. RSI asks a similar question about the balance of up-closes against down-closes. Both are therefore relative to a moving window, which produces the property that trips everybody: in a sustained trend the window keeps moving with price, so the reading can sit at 85 for weeks without the market being "too high" by any measure that would put money in your account.
- RSI (Wilder, usually 14) — the smoothest of the three, built from average gains versus average losses. Bands at 70/30. Slow to reach extremes, so its extremes carry more weight.
- STOCHASTIC (%K with a %D smoothing) — the most sensitive, since it uses only the high-low range of the window. Bands at 80/20. Excellent for timing inside a range; unbearable as a standalone trend tool.
- WILLIAMS %R — arithmetically the stochastic's mirror, running from −100 to 0. If you already use Stochastic it adds nothing; it is here so you recognise it on someone else's chart.
- In all three, a SHORTER period means more signals and more noise. The period is not a quality dial — it is a trade between being early and being right.
Run the lab above on Range and then on Uptrend without changing anything else. In the range, entries into overbought are followed on average by a fall — the oscillator is doing exactly what it was designed for, telling you the swing is stretched inside a stable band. In the uptrend, the identical signal is followed on average by a further RISE, because each overbought reading is simply the trend continuing to make new highs within its own moving window. One indicator, one rule, two opposite outcomes — and the only thing that changed was the regime the market was in.
Price makes a new high at 120 after an earlier high at 118. The RSI, however, reads 62 at the new high against 78 at the old one. That is bearish divergence, and here is what it literally means: the second advance had a weaker balance of buying pressure than the first, even though it went further. Fewer participants pushed price higher, which is the participation story from module 1 expressed as a number. Bullish divergence is the mirror at a bottom — a lower low in price on a higher oscillator low — and tends to be the more reliable of the two, because declines end in exhaustion more cleanly than advances end in euphoria.
RSI, Stochastic and Williams %R are all computed from the same closes over similar windows, so they are heavily correlated by construction. Stacking them does not give you three opinions; it gives you one opinion printed three times, at the cost of screen space and the illusion of confirmation. If you want a genuine second opinion, take it from a different KIND of information — a trend measure, volume, a longer timeframe, market breadth. Confirmation only means something when the two sources could have disagreed.
Elder does not use oscillators to find reversals at all. He uses them for one narrow job: once a longer timeframe has established the direction, an oscillator moving AGAINST that direction on the shorter timeframe marks the discount at which to enter. In an established uptrend, an oversold stochastic is not a warning — it is the pullback you were waiting for. In a downtrend, an overbought reading is where you sell. The oscillator never chooses the direction; it only chooses the moment, inside a direction chosen elsewhere. That inversion is what module 5 is built on.
An oscillator does not measure price; it measures price's position inside a window that moves with price. That single sentence explains every one of its famous failures. In a range the window is stable, so the reading is meaningful and mean reversion pays. In a trend the window travels, so an extreme reading is self-referential — it says "we are near the top of where we have recently been", which is the definition of a trend, not a reason to bet against it.
You see RSI at 82 on a daily chart. Before deciding anything, what one question determines whether this is a sell signal or a buy signal? (Which regime is the market in — which in practice means: what is the longer timeframe doing? In a range, 82 is a stretched swing and a fade is reasonable. In an established uptrend, 82 is strength, and the tradeable event is not this reading but the PULLBACK that follows it, which you would buy, not sell.)
Oscillators are bounded because they measure position inside a moving window, which makes them excellent range tools and dangerous trend tools. RSI is the smoothest, Stochastic the most sensitive, Williams %R a mirror of Stochastic — pick one, not three, since they are all reading the same data. Divergence is the genuinely valuable signal because it measures participation, and it is best used to defend rather than to enter. Above all, adopt Elder's inversion: let a longer timeframe choose the direction, and let the oscillator choose only the moment.
On one daily chart, mark every RSI reading above 70 in the last year. Beside each, write whether the weekly trend at that moment was up, down or sideways, and what price did over the next ten sessions. Sort the list by that middle column. The three groups will tell different stories, and the difference between them is the only thing you actually needed to learn about "overbought".
Indicators & Oscillators