Level decay

The claim, and why it is worth testing

A level is said to decay: roughly 80% of first tests hold, about two thirds of second tests, and only a third of third tests. The figure circulates with no sample size attached to it, which is the usual sign that it has been reasoned rather than counted. It is also the kind of claim that changes how somebody sizes a trade, so it is worth the work of counting.

How this was counted

A level is a confirmed swing pivot with five sessions either side, so it is only a level once the market has agreed it is one. A touch is price coming within 0.15% of it, with a three-session cooldown so one slow grind across a level is not counted as four separate tests. A reaction is price moving 0.30% away from the level, in the direction the level implies, within two sessions. Tests are counted per level, so the fourth tap of a level knows it is the fourth.

The result

Reaction rates by tap count came in at 67.4%, 66.9%, 69.3% and 67.9%, on samples of 1,478, 1,321, 1,196 and 7,797. There is no decay in that series. The third test is the strongest of the four, which no amount of favourable reading turns into a falling curve.

What a null was needed for

A 67% reaction rate means nothing on its own, because price moves 0.30% in some direction most of the time. So each touch was matched against level-free bars on the same ticker, in the same calendar month, in the same direction. Those matched bars reacted 48.1% to 52.2% of the time. The level is worth 15.0 to 19.3 percentage points over nothing, and the interval on that lift excludes zero at every tap count.

What it does not settle

This is a measurement of price pivots, not of dealer gamma nodes, and it is published under that limit rather than past it. It also says nothing about which direction the reaction goes, only that one happened. And it was run on the US index complex; single names were not tested and should not be assumed to behave the same way.

Questions

Do support and resistance levels get weaker each time they are tested?

Not measurably. Across 11,792 touches of confirmed index swing pivots, the share that produced a reaction was 67.4% on the first test, 66.9% on the second, 69.3% on the third and 67.9% on the fourth and beyond. The third test scored higher than the first. Across twenty variations of the measurement the largest fall from first test to third was 1.15 percentage points, which is noise at these sample sizes.

Then why does the 80/66/33 figure get repeated?

Because it is intuitive and nobody publishes a sample size next to it. A level that has been tested three times is, by construction, a level that survived three tests, so the surviving population is not the population the intuition imagines. We could not find the measurement behind the figure, so we ran one.

Does a level still beat no level at all?

Yes, and that is the finding that actually matters. Against a null built from level-free bars matched on ticker, calendar month and direction, a pivot reacted 15.0 to 19.3 percentage points more often than nothing did, at every tap count. Levels work. They just do not wear out on the schedule they are said to.

Is this measuring dealer gamma nodes?

No, and the page would be dishonest if it implied otherwise. This measures confirmed price swing pivots, which is the nearest thing the available history can settle. It is the price-only shadow of a claim made about dealer gamma nodes, and it can neither confirm nor refute that claim.

Related

Other concepts