Indicator library · Composites
Technical Rating, One Score, Many Inputs
One score standing in for a dozen indicators. It is the last page in this library on purpose: a rating is what happens when everything here is collapsed into a single number, and what it destroys in the process is the reason the other pages exist.
How one is built
The construction is simple enough that it is worth writing out, because the simplicity is where the judgement hides.
- Pick a set of indicators, typically several moving averages, two or three oscillators, and a trend measure.
- Reduce each to a vote by comparing it with a threshold: +1, 0 or −1.
- Sum or average the votes, sometimes with weights.
- Map the total onto a label, strong sell, sell, neutral, buy, strong buy.
Step two carries almost everything that matters and is almost never published. Does RSI at 68 vote positive or neutral? Does a crossover from four days ago still count? Is a falling ADX bearish, or directionless as its own page insists? Two ratings built on identical indicators can disagree entirely on those choices alone.
Cancellation is not consensus
The figure is the whole argument. A summed score cannot distinguish agreement from cancellation, and those are the two most different things a set of indicators can be doing.
It matters because of what disagreement is worth. This library keeps returning to one point: the value of a second indicator lies entirely in its ability to contradict the first. Momentum strong while volume thins, price at new highs while breadth narrows. Those are the readings worth having, and every one of them is a disagreement between components. A rating averages them away by design.
The independence problem
There is a second, quieter failure. Most published ratings combine eight or ten measures that are nearly all computed from the same closing prices: several moving averages of different lengths, two or three momentum oscillators, a trend measure derived from the same series.
Averaging those is not a consensus. It is one opinion counted many times, and the resulting score is stable: which reads as robustness and is nothing of the kind. A rating that includes a breadth measure and a volume measure alongside price-derived ones is doing something genuinely different, and it is the exception rather than the rule.
Building one that is worth reading
If a composite is wanted — and for scanning a large universe there is a case for one — three rules make it defensible, and all three are about disclosure rather than cleverness.
Pick components that read different inputs. One price measure, one volume measure, one breadth measure beats eight oscillators, because only the first set can disagree for reasons that matter. Publish the mapping. Every threshold that turns a reading into a vote is a judgement, and an undisclosed judgement makes the score unreproducible. Report the spread alongside the total, so a reader can see whether the number came from agreement or from cancellation.
A composite built that way is a genuine summary. One built the usual way is a label, and the difference is entirely in what its author was willing to show.
Where it misleads
| Situation | What goes wrong |
|---|---|
| Cancellation read as consensus | A split set of components and a unanimous one can produce the same total. The score cannot distinguish them. |
| Components not independent | Eight measures of the same closing prices averaged together is one opinion counted eight times. |
| Thresholds unpublished | Whether RSI at 68 votes positive is a decision made for you and rarely disclosed. It changes the score. |
| Weights unexplained | A weighted average with undisclosed weights is not reproducible by anyone, including its author six months later. |
| Stability read as accuracy | Averaging reduces variance whether or not it adds information. A steadier score is not a better one. |
| Sorted screens built on it | Ranking a universe by a composite ranks it by whichever component dominates the mapping, usually trend. |
The screener case, stated fairly
There is one job a composite does that its components cannot: it sorts. Ranking four thousand instruments by five separate readings is not a thing a screen can do, and a single number makes the list orderable in a way that is genuinely useful as a first pass.
The failure is not the sorting, it is stopping there. A composite ranking is a way of shortening a universe to a few dozen names worth opening the chart on, the same role every screen on this site is described as filling. Treated as a verdict it replaces the work; treated as a queue it saves an afternoon, and the difference costs nothing but the willingness to look at what the components were doing once the list is short.
What to ask for instead
A more honest summary fits in the same space. Three numbers (how many components positive, how many negative, and the extremes) keep the disagreement visible while remaining sortable. Better still is naming which measures disagreed, because "momentum positive, volume negative" is a specific claim someone can check and "moderate buy" is not.
That is the note this library ends on, and it is deliberate. Every page here states what its measure can see and what it cannot, so that two of them disagreeing tells you something. A rating is the one construction that takes all of that apparatus and returns a word, and the word is worth exactly as much as the components you are allowed to inspect behind it.
Frequently asked questions
What is a technical rating?
A single score, often "strong buy" to "strong sell", or a number between −1 and +1, produced by combining the readings of several indicators. Each component is reduced to a vote, the votes are averaged or summed, and the total is presented as an overall verdict. Almost every charting platform publishes one, and the components are rarely listed beside the score.
How are the components combined?
Usually by a simple rule: each indicator is mapped to −1, 0 or +1 by comparing it with a threshold, and the votes are averaged, sometimes with weights. The mapping is where the judgement hides. Whether RSI at 68 counts as neutral or positive, and whether a moving-average crossover two days old still votes, are decisions made by whoever built the rating and almost never published with it.
What does the score hide?
The disagreement, which is the most informative thing the components contain. A market where every indicator points the same way and one where they are split down the middle can produce the same total, and they are completely different conditions. A rating throws away exactly the signal that made a second indicator worth consulting in the first place.
Are the components independent?
Rarely, and this is the deeper problem. Most published ratings mix several momentum oscillators, two or three moving averages and a couple of trend measures, nearly all computed from the same closing prices. Averaging eight views of one input is not a consensus, it is one opinion counted eight times, and it produces a score that looks robust because it is stable rather than because it is informed.
Does a rating perform better than its components?
Sometimes, and for a mundane reason: averaging reduces variance, so a combined score is steadier than any single component. Steadier is not the same as more accurate. Whether the combination adds anything depends entirely on whether the components read genuinely different inputs, which most do not, and any published performance figure inherits every choice of threshold and weight that produced it.
How should a rating be read if I am given one?
As a starting question rather than an answer, and only if the components are visible. The useful reading is not the total but the spread: unanimous components tell you the obvious thing the price chart already showed, while a split score is a flag that something in the instrument is contested, which is where the work actually is. If the platform will not show the components, the score is not checkable and should be treated accordingly.
What would a more honest summary look like?
Three numbers rather than one: how many components read positive, how many negative, and what the extremes were. That preserves the disagreement while still fitting in a table cell. Better still is a short list of which measures disagreed, since knowing that momentum and volume are pointing in opposite directions is a specific, checkable statement, and "moderate buy" is not.
Why do platforms publish them?
Because a single label fits in a list, sorts, and requires no explanation, which are real virtues for a screener and none of them are about accuracy. There is nothing dishonest in that as long as the components are available; the failure is presenting the label alone, since a reader then has no way to distinguish agreement from cancellation.