The 20/50/200 cluster, and what it actually tells you.

Three averages, two jobs

A moving average does exactly one thing: it reports where price has been over a window, smoothed. Everything else people read into it comes from comparing one window to another. With three averages there are only two comparisons worth making, and they answer different questions.

The 20 against the 50 answers what is the market doing now, relative to the last few weeks. Both are short enough to turn inside a single move, so their relationship is a statement about the current leg.

The 200 answers something else entirely: what is the background. It is far too slow to time anything. Its value is as a reference the fast pair can be measured against — it does not tell you when, it tells you which direction has the wind behind it.

Why the 20 and 50 are one object, not two lines

Treating the 20 and 50 as separate signals produces the classic problem: you get a crossover, you take it, and the two lines immediately tangle again. The crossover was real; it just carried no information, because the two averages were sitting on top of each other when it happened.

The fix is to stop watching the crossing and start watching the separation. Two averages that are 0.8 ATR apart and widening describe a market that is moving directionally. The same two averages 0.05 ATR apart describe a market that has no direction at all, and any crossover they produce is noise dressed as a signal.

Once you are reading separation rather than crossings, the pair behaves like a single object with a width and a position. That is what "cluster" means here — the 20 and the 50 taken together, with the gap between them treated as a measurement in its own right.

What the 200 adds

The 200 contributes two separate pieces of information, and it is worth keeping them apart because they disagree more often than you would expect.

Position is where the cluster sits relative to the 200. A 20/50 cluster above a rising 200 is a different proposition from the same cluster above a falling one, even though both are "above the 200".

Slope is how far the 200 has actually travelled over some lookback. This is the part most setups leave out, and it is the part that does the most work. A 200 that has moved a tenth of an ATR over the last fifty bars is, for practical purposes, a horizontal line — price will cross it repeatedly, in both directions, and every crossing will look like a regime change and be nothing of the sort.

Requiring a minimum slope before the 200 counts as sloped at all is what stops a drifting market from manufacturing trend conditions.

Alignment is a state; a crossover is an event

Crossovers are popular because they are easy to code and easy to point at after the fact. Their weakness is that they fire exactly once, at the moment two lines touch, which is also the moment the two averages agree least about what is happening.

Alignment is the more useful frame. On any given bar the cluster is either stacked in your favour and separated, or it is not. That is a state you can check continuously, it degrades gradually rather than flipping, and it does not require you to have been watching at the instant of the cross.

The two conditions that make all of it meaningless

Both are common enough that any moving-average setup which ignores them will spend a large share of its time producing confident-looking nonsense.

• Compression. The 20/50 separation collapses. Every reading is technically still computable and none of it means anything — the market is balanced and the averages are describing the same price from two slightly different angles.
• A flat 200. The background reference has no direction to lend. Position relative to it stops being informative, because price is oscillating around a level rather than trending away from it.

The useful move is to treat these as gates rather than as inputs to be averaged in. A compressed cluster should not be able to produce a strong reading no matter how good the rest of the picture looks. If compression is just one factor among several, a sufficiently favourable set of other factors will outvote it, which is precisely the failure you were trying to prevent.

Does the averaging method matter?

Less than the arguments about it suggest, but not nothing. The choice is a straight trade between lag and noise:

Pick one and keep it. Switching methods after a run of bad trades is a way of fitting the tool to the recent past, and the thresholds you set under one method will not mean the same thing under another.

How Traffic Lights reads it

Traffic Lights is one of the five indicators bundled with FlowRegime, and it is the one built directly on this cluster. It scores each side of the market out of five, from the alignment of the cluster against the 200, how far the 200 has genuinely travelled over your lookback, and how much room there is inside the cluster. The count decides the colour: green at or above your green minimum, yellow at or above your yellow minimum, red below both. Compression is a gate, not a vote — when the cluster is tighter than your threshold, green is blocked on both sides.

Every one of those numbers is a setting, including the two thresholds. The point is not that our defaults are correct; it is that you can see what is being counted and change what it takes to earn a colour.

FlowRegime reaches its verdict a different way. There is no moving-average cluster in it and no crossover anywhere in the scoring: it reads absorption, exhaustion and efficiency from order flow and the Level 2 book, and decides whether the tape rewards fading extremes or following displacement. Averaging appears only as plumbing — an ATR on a secondary series to set the scale, and an EMA to smooth the finished score — never as the signal itself. Read this guide as background for the bundled indicator rather than as a description of how the flagship works.

What it does not tell you

A cluster reading is context, not an entry. It describes the background a trade would be taken into — it says nothing about where you get in, where your stop belongs, or what the trade is worth risking. Nor does it anticipate: every one of these averages is computed from bars that have already closed, so a change in the reading follows the move that caused it. Any tool built on them inherits that, and no amount of parameter tuning removes it.