5 / 7 · 8 min
Indicators: they know nothing beyond what is already on the chart
Any indicator is a formula over the same price and volume you already see. There is no new information in it; it is useful in exactly two ways, and both are worth knowing in advance.
Start with what is rarely said out loud. An indicator does not observe the market — it recomputes a chart that is already drawn. RSI, MACD, stochastic and moving averages are all built from the same four numbers of each candle. Nothing that is absent from price and volume can appear inside them: this is arithmetic, not a data source. From that follows both the main limitation and the real usefulness.
Two honest uses
- Measuring what the eye judges badly. "Price is far from its average" depends on the zoom of your screen; expressed in numbers it does not.
- Replacing feelings with a rule. An indicator with a fixed parameter answers the same way yesterday, today and in a panic. Discipline matters more here than accuracy.
And one limitation that cannot be removed
Anything that averages the past lags behind the present — that is a property of the definition, not a flaw in the implementation. A moving average of 50 candles cannot turn before those 50 candles have turned. So an indicator is unfit for the role of a forecaster and fit for the role of a gauge. Demanding an early signal from it is like demanding that a thermometer predict the weather.
Example: price 2,480, the 50-candle average 2,340, ATR 70. The stretch equals two ATR — price has travelled two ordinary daily ranges away from its own recent norm. By itself this signals nothing: in a strong trend the stretch stays high for weeks. The value is elsewhere — the number is comparable between instruments. Saying "it went far" about bitcoin and about a small coin is only possible this way, because "far" is of a different size for each.
A moving average as support
A very common and very convenient delusion: the line is drawn, price sometimes bounces off it, the conclusion suggests itself. But an average has no orders. It is the arithmetic mean of the last N closes, a number recomputed every candle that crawls along after price. There is nothing there to push off. Coincidences happen because in a trend the average runs roughly where price runs — and the longer the trend, the more convincing the coincidence looks. Testing it is easy: change the period from 50 to 45 and the "support" ends up at a different price.
"RSI above 70 means sell"
RSI compares average gains with average losses over the last N candles. A high reading means exactly one thing: recent candles rose faster than they fell. In a strong trend that state persists for weeks — bitcoin's daily chart has had stretches where RSI did not drop below 70 for more than three weeks running, and the "sell signal" stood there rising the whole time. Overbought does not mean expensive, it means fast. Fast growth is a normal condition at the start of a trend, not at its end.
Three indicators instead of one
It feels as if agreement among three oscillators is more reliable than the opinion of one. In fact RSI, stochastic and MACD are computed from the same price series and almost always agree by construction — their agreement adds no confirmation, it adds confidence. That is the worst possible trade. If you want an independent check, take something computed from OTHER data: volume, open interest, funding, the order book. Three formulas over one price are one opinion repeated three times.
What each one actually measures
| Indicator | What it measures | What it cannot do |
|---|---|---|
| moving average | the average price of the last N candles | be a level: there are no orders behind it |
| RSI | how many times faster gains were than losses | tell "expensive" from "fast" |
| ATR | the ordinary range of a candle | say anything about direction |
| MACD | the gap between two averages | get ahead of price: it is derived from it |
Fitting the period to history
The most expensive mistake in this subject, because it looks like work. You try RSI with a period of 14, 12, 11 and find that on last year 11 worked best. That is not a discovery: with a dozen attempts something is bound to come out best on any stretch, including a random one. Testing must happen on data you did not see while fitting, and you should try as few parameters as possible. If the difference between 11 and 14 is large, that is a sign of curve-fitting rather than a find — a robust rule should not fall apart when one number is swapped for its neighbour.
The bar "RSI above 70" is the same for everyone, but instruments are not. Instead of a common bar we use a percentile from each coin's own year of history: a reading is compared not with somebody else's norm but with its own. "Funding in the top percent for the year" is a statement that for THIS coin this happens about once in a hundred days; "funding 0.05%" is just a number.
Analytics and percentilesOne indicator, twenty charts, zero exceptions
Pick one indicator and write the rule down in a single sentence, with no "usually" and no "if the situation allows". For example: enter when price closes above the 50-period average, and only then. Run it in bar replay across twenty different coins without changing anything as you go. Write down the outcome of every case. The point is not for the rule to be profitable — it almost certainly will not be. The point is to see how many times you wanted to make an exception: that number is your real problem with indicators, not the choice of period.
The 50-period average held price perfectly five times in a row. Does that prove it works?
No, and you can check it in a minute. Change the period to 45 or 55 and see whether those same five touches still held. If the picture fell apart, you were looking at a coincidence: in a trend the average runs beside price at any period, and some value is bound to line up with a few pullbacks. A robust observation survives a swap to the neighbouring parameter; curve-fitting does not.
Why is agreement between RSI, stochastic and MACD not a confirmation?
Because all three are computed from the same price series and differ only in the smoothing formula. They agree almost always — not because the market is telling them something, but because they are looking at the same thing. Confirmation only comes from an independent source: volume, open interest, funding, the order book, the behaviour of other coins. Three views of one price are still one price.