What each indicator measures and how people fool themselves

The four most-used indicators have public, exact formulas, and none predicts anything on its own. This session covers them one by one with their origins, and closes with the two errors that generate the most false confidence — which are about method, not tools.

What each indicator measures, at a glance

The five most-used indicators measure different things and fail in different places. None indicates direction on its own, and each has one specific market condition where its usual reading misleads. This table sets out what each one measures, what escapes it, and which mistake gets made with it most often.

Indicator	What it measures	What it does not capture	The typical mistake	Where it misleads most
Moving average	The average price over the last N periods	Anything about the present: it describes what already happened	Expecting it to anticipate a turn	In a range, where price crosses it up and down while going nowhere
ATR	How much the asset moves in a typical period, opening gaps included	Direction. It says nothing about up or down	Reading it as a buy or sell signal	Right after a sharp change in behaviour, because it takes fourteen periods to reflect it
RSI	How much gains have outweighed losses over fourteen periods	The strength of the underlying trend	Selling above 70 and buying below 30	In a sustained trend, where it can sit above 70 for weeks without turning
MACD	The distance between a fast average and a slow one	Anything the averages themselves don't capture: it inherits their full lag	Counting the line cross and the histogram as two separate confirmations	In a range, where it produces contradictory crosses one after another
Bollinger Bands	How many standard deviations the price sits from its twenty-period average	The real probability of an extreme: price distributions aren't normal	Treating a tag of the lower band as a buy signal	In a strong trend, where price rides the band for weeks

The sections below cover where each formula comes from and who built it. The table summarises; the detail sits underneath.

What a moving average measures and why it's always late

A moving average is an asset's average price over a fixed number of periods, recalculated each day. It smooths daily noise and reveals the underlying direction. And it's late by construction, not by defect.

The two ways to calculate it

Simple moving average: you sum the last N closes and divide by N. Each day weighs exactly the same.

Exponential moving average: more weight goes to recent prices. It reacts sooner to a change of direction, and in exchange gets thrown off more by moves that turn out to be temporary.

Why the lag is unavoidable

A 200-session average includes, by definition, the previous 200 closes. For it to turn, the new price has to be different enough to shift an average of two hundred numbers, and that takes time.

It isn't a fixable defect: it's what the tool does. A moving average smooths, and smoothing means responding late.

What it does do well is answer a concrete question: is the current price above or below what's been normal lately? That question has an objective answer — and crossing two different averages produces the signal that generates more headlines than any other.

What track record the death cross actually has

When the 50-session average crosses below the 200-session one it's called a death cross and makes headlines. The historical data says something rather different from what the name suggests.

The numbers

An analysis by Fundstrat, authored by its technical analyst Mark Newton, reviewed 48 death crosses on the S&P 500 since 1929. The result:

Subsequent horizon	Average S&P 500 return
3 months	+2.8%
6 months	+4.3%
12 months	+5.0%

The hit rate — share of cases with positive returns — was above 50% at all three horizons. Only at seven days and one month was the average return negative.

Put differently: historically it has paid better to buy after a death cross than to sell.

Why this happens

Because the signal is late by construction, as we just saw. For the 50-day average to fall below the 200-day, the price has to have been falling for quite a while already. By the time the headline appears, much of the damage is usually done.

And the golden cross, to be fair

The complete system — buy on the golden cross, sell on the death cross — has a documented record worth telling in full.

QuantifiedStrategies, a site that specialises in publishing tests of trading systems against historical data, published a backtest of the full pair on the S&P 500 since 1960, using simple 50- and 200-session averages. It found only 33 signals in 66 years, with an average trade lasting around 350 days.

It's worth placing what kind of source this is, because it doesn't carry the same weight as the ones above. Wilder and Bollinger authored the tools themselves; Park and Irwin published in a peer-reviewed academic journal. A backtest published on a specialist site is a concrete test, reproducible in its parameters and without external review. It orients; it doesn't settle an argument. Its annualised return came in slightly below buy-and-hold, but with two important differences: it was invested only about two-thirds of the time, and maximum drawdown was roughly halved.

That's the honest conclusion and it doesn't fit a headline: it isn't a signal for timing the market, it's a system that reduces the shock in exchange for some return. Whether that's a good trade depends on what you're optimising for, which connects directly to the next chapter.

What the ATR measures and why it beats a percentage

The ATR — Average True Range — measures how much an asset moves in a typical period, in real price units. J. Welles Wilder created it and published it in 1978 in his book New Concepts in Technical Trading Systems.

The problem it solved

Measuring volatility with the day's high-minus-low falls short, because it ignores gaps. If a stock opens far below the previous close, that jump was real movement and doesn't appear inside the day's range.

Wilder defined the true range as the greatest of three values:

• Today's high minus today's low.
• Today's high minus yesterday's close.
• Today's low minus yesterday's close.

The last two are what capture the gap.

What the ATR measuresWilder (1978) · New Concepts in Technical Trading Systems

Yesterday closeToday highToday lowGAPYesterdayToday

The true range is the largest of three values

1. Today high − today low
2. Today high − yesterday closeCaptures the gap
3. Today low − yesterday closeCaptures the gap

The third: today low minus yesterday close.

Wilder, J. W. (1978). New Concepts in Technical Trading Systems.

How the ATR is calculated

The 14-period ATR — Wilder's standard — starts as the average of the first 14 true ranges. From there, each new value is calculated as:

Today's ATR = ((Yesterday's ATR × 13) + Today's true range) ÷ 14

That formula weights recent history without discarding earlier data, producing a smooth curve rather than one jumping daily.

Why it beats a fixed percentage

Saying "set the stop at 2%" treats a calm stock and a volatile one identically, and that can't work well for both. Two percent in a stock moving 0.5% a day is an enormous distance; the same 2% in one moving 4% a day gets cleared in minutes.

Measuring distances in multiples of ATR adjusts automatically to each asset's actual behaviour. It's the same idea from Chapter 2 on volatility, with a unit built for day-to-day rather than for the year.

What the RSI measures and where it fails

The RSI — Relative Strength Index — was also created by Wilder and published in the same 1978 book. It measures the speed and magnitude of recent moves, on a 0 to 100 scale.

The formula

It's calculated in two steps.

First, relative strength: divide the average gain by the average loss over the last 14 periods.

Second, convert to a 0-100 scale:

RSI = 100 − (100 ÷ (1 + relative strength))

Wilder considered an RSI above 70 to indicate overbought and below 30 oversold.

Why it gets stuck above 70

In a sustained trend, the RSI can stay above 70 for weeks on end. Anyone selling every time it crosses 70 sells repeatedly against a move that continues.

The same problem, inverted, in downtrends: the RSI can live below 30 for a long time without any floor arriving.

The reason sits in the formula itself. The RSI compares gains against losses inside a fourteen-period window, so in a clean trend the numerator dominates the denominator continuously. It isn't that the indicator is wrong: it's measuring exactly what it claims to measure, and what it claims to measure isn't exhaustion.

It's an indicator that describes well what has happened and gets misused as a reversal signal. The distinction is the same one running through this whole session.

What the MACD compares

The MACD — Moving Average Convergence Divergence — was developed by Gerald Appel in the late 1970s. The histogram usually accompanying it was added later by Thomas Aspray.

The three pieces

The MACD line is the 12-period exponential moving average minus the 26-period one. It measures the distance between a fast average and a slow one.

The signal line is the 9-period exponential moving average of that MACD line.

The histogram is simply the difference between the two lines above.

What it's actually measuring

It's measuring whether recent movement is separating from or converging on the underlying movement. When the fast average pulls away from the slow one, recent momentum is stronger than the established trend. When they converge, it's running out.

The usual reading: when the MACD line crosses above the signal line it's read as bullish, and below as bearish.

Why the histogram confirms nothing

And there's a practical detail that confuses many people: the histogram and the line cross say exactly the same thing. When the MACD line crosses above the signal line, the histogram goes from negative to positive. They aren't two independent confirmations: they're the same fact drawn twice.

From which comes a useful check against any screen crowded with indicators: ask how many of them are calculated from the same input. Three tools built on moving averages aren't three opinions, they're one opinion repeated three times, and the sense of agreement between them is an arithmetic illusion.

What Bollinger Bands actually contain

John Bollinger created them and published them in 1983. The centre band is a 20-period simple moving average, and the upper and lower bands are calculated by adding and subtracting two standard deviations from that same average.

The correction the author himself makes

Here's the most useful item in this section, and it comes from the official rules Bollinger publishes on his own site.

It gets repeated constantly that, using two standard deviations, the bands should contain 95% of prices, following the statistics of a normal distribution.

Bollinger says the literal opposite in his rule number 14: avoid making statistical assumptions based on the standard deviation, because the distribution of security prices is non-normal and the typical sample size is too small for statistical significance. And he adds in parentheses: in practice we typically find 90%, not 95%, of the data inside Bollinger Bands with the default parameters.

That 5% difference isn't an academic nuance. It means extreme moves occur twice as often as the theory suggests, which is exactly the kind of error that ruins anyone sizing their risk from the theory.

And the second correction, which contradicts the most common use

His rule number 7 is equally direct: a tag of the lower band is not, in and of itself, a buy signal.

That's probably the indicator's most widespread use and its creator explicitly advises against it.

Bollinger summarises his tool's purpose in the last of his rules: the bands don't provide continuous advice, they help identify setups where the odds may be in your favour. That's a considerably more modest description than most users give them.

What overfitting is

Overfitting happens when you tune a strategy's parameters over and over on the same historical data until you find the combination that works best on that specific stretch.

Why you always find something

Because with enough attempts, finding a combination that worked in the past is guaranteed by chance. No real edge needs to exist: enough trying suffices.

It's the same mechanism by which someone flipping a coin a thousand times eventually gets ten heads in a row. That streak existed, is verifiable, and says nothing about the coin.

Why this isn't a theoretical problem

Go back to this chapter's first session. Park and Irwin's review of 95 studies flagged precisely this — data snooping and ex post rule selection — as one of the problems preventing the favourable evidence from being conclusive.

Meaning: overfitting doesn't just ruin homemade strategies. It contaminates much of the field's academic literature.

How to recognise it in practice

Three fairly reliable signs:

Overly specific parameters. A strategy using a 37-period average and an 11-period RSI almost certainly reached those numbers by testing, not by reasoning.

Results that collapse on a slight change. If it works with 37 periods and fails with 36 and 38, there's no logic behind it: there's a fitted coincidence.

Many rules for little data. The more conditions a strategy has and the fewer trades it has generated, the more likely it's describing the past rather than explaining it.

The method chapter returns to this in full, because avoiding it has a name and a procedure.

Why looking at a chart knowing the ending deceives

Hindsight bias is even more treacherous than overfitting, because it requires no calculation. Looking at old charts is enough.

The mechanism

Any pattern that preceded a rise looks, in retrospect, obvious and easy to have anticipated. The signal jumps out.

The problem is that that same pattern, on the day it was forming, looked like many others that led nowhere. The difference isn't in the pattern: it's that you now know the ending.

Why it's so hard to avoid

Because it doesn't feel like an error. It feels like understanding.

Reviewing historical charts produces a genuine sensation of learning — "now I see how this works" — that is internally indistinguishable from having actually learned something. And that's why it generates confidence nothing justifies.

The only thing that stops it

Writing the reading down before knowing the outcome, with the invalidation criterion included.

It's exactly the discipline the earnings chapter's final session set out: if you write down what you expect and what would have to happen for you to be wrong, afterwards you only have to check. If you don't write it, the interpretation adjusts itself to what happened and you never learn anything.

It's also, incidentally, why Volatly publishes each event's scenario before it happens and seals it afterwards, with the misses included. An archive built after the fact proves nothing, however good it looks: it only proves someone can read the past, which anyone can do.

What to remember

• A moving average is late by construction, and the death cross has positive average returns at three, six, and twelve months across 48 occurrences since 1929.
• The ATR includes opening gaps, which is why it measures real movement better than a fixed percentage.
• John Bollinger himself corrects two myths about his bands: they contain 90% rather than 95%, and tagging the lower band isn't a buy signal.
• Overfitting and hindsight bias share a root: letting knowledge of the future leak into an evaluation that didn't have it at the time.

Milestone reached

That closes Chapter 6. You can describe a chart out loud without making anything up: what the evidence does and doesn't say about technical analysis, how a trend is defined, why levels form according to real order data, what each indicator measures with its formula, and the two errors that generate the most false confidence.

Chapter 7 shifts ground entirely. It leaves the market and enters the one thing you fully control: how much to risk, and why being right isn't the same as winning.

Related: what "risk" really means · backtesting: how it works and the four ways to fool yourself · structure, levels, and volume