Connors RSI: Three Measurements Averaged
Connors RSI averages three components: a short-period relative strength index of price, a relative strength index of the current up or down streak, and a percentile rank of the latest one-day return. Only the streak component measures something the standard oscillator family does not.
How it works
Three separate calculations, averaged with equal weight into a single 0–100 reading.
Component one is a short relative strength index of price, conventionally 3 periods rather than the usual 14.
Component two is a relative strength index applied to the current streak — the number of consecutive days price has closed up or down, expressed as a positive or negative count.
Component three is a percentile rank of the most recent one-day return against the last hundred returns: where does today’s move sit in the distribution of recent moves?
Taking it apart
The streak component measures something the standard family does not. Every common oscillator measures distance — how far price moved. This counts consecutive days in one direction, regardless of size. Six tiny up days score higher than one enormous one.
That is a genuinely different question, in the same way the Aroon family’s time-counting is different, and it is the reason this indicator is more than a repackaging.
On this site’s shared 576-bar history there are 286 directional runs, averaging 2.01 bars, with the longest at 11. That distribution is what a streak measure is reading: most streaks are two bars, a streak of five is already unusual, and anything past eight is rare.
Which tells you how to read the component honestly. A streak reading at an extreme means something uncommon has happened by the standards of the series — not that a reversal is due, but that the current run is long relative to the distribution.
The percentile-rank component is closer to a volatility reading than a directional one. It asks whether today’s move was large relative to recent moves, which is a question average true range answers more directly and with a comparable number.
And the first component is a relative strength index, which on this history moves at 0.71 with momentum, 0.71 with the moving average convergence divergence histogram and 0.87 with the stochastic. If you already run an oscillator, a third of this indicator is a duplicate of it.
In practice
Three components carry three lookback settings — conventionally 3, 2 and 100 — plus the decision to weight them equally. Nothing derives any of those, and each is a dial that can be tuned until past data looks good.
The equal weighting is the least examined choice of the four. There is no argument for a third of the signal coming from a volatility rank other than that three components divide neatly.
No component uses volume. Three price transformations remain three price transformations.
The short lookbacks make it a fast, mean-reversion tool by design, and that is what it was built for. Applied to a long-horizon trend question it converges on the behaviour of the plain index it contains.
A gap moves all three components in the same direction simultaneously — a large return ranks at the extreme, the short index jumps, and the streak extends or resets. Three components agreeing is one event entering three calculations.
It produces no price, so a stop has to come from structure or volatility, exactly as with every other oscillator.
Each signal traded costs 2% of a typical bar’s range in round-trip costs on this history — and a 3-period oscillator generates signals frequently, which is the point of it and also the bill.
And the order book knows about none of it. Three transformations of one close series is one input, however it is averaged.
What Connors RSI is not
It is not three independent confirmations. All three components come from the same closes, and two of them measure things you probably already have.
It is not a trend tool. The lookbacks are short by design; it was built for mean reversion.
It is not an improvement on the relative strength index in general. It is a different tool aimed at a narrower question.
And it is not a standard. Platforms implement the streak and rank components with small variations, so readings differ.
When it fails
In a sustained trend it stays at an extreme. A 3-period index pins quickly, streaks extend, and large same-direction returns rank high — so all three components push the same way and the composite sits near its bound while price keeps going.
The second failure is treating an extreme as a reversal signal. The indicator says the recent run is unusual by recent standards. Unusual things continue regularly, and on this history the longest run was 11 bars against a mean of 2.
A third is the parameter surface. Three lookbacks plus a weighting is four choices, and any four-way optimisation on a few hundred bars finds something that looks excellent and is noise.
A fourth is running it alongside a plain relative strength index. One is a component of the other; they cannot meaningfully disagree.
And a fifth is applying it to a trending instrument. It is a mean-reversion tool. Used where price persists, it reports extremes constantly and each one is an invitation to fade a move that is not finished.
The original data
On this site’s shared 576-bar history: 286 directional runs, mean length 2.01 bars, longest 11 — and
bar-to-bar change correlations among five price oscillators running from 0.53 to 0.87, with the relative
strength index against the stochastic at the top. Both sets are in
research/series-measurements.json, produced by site/measure_series.py.
The run distribution is the number this indicator actually needs and does not ship with. A streak component is only readable against the distribution of streaks in the instrument you are trading, and that distribution is different for every market and every timeframe. Compute it first — count your own runs, get the mean and the tail — and the streak component becomes interpretable instead of decorative. It is a dozen lines of code, and it is the part of this indicator that is worth having.
Related
RSI is the component you probably already run. Mean reversion is the approach this was designed for. And choosing indicators carries the correlation matrix that decides whether a composite is adding anything.
Connors relative strength index is the first composite indicator I took apart rather than adopted, and doing that is the useful exercise regardless of what you conclude. Two of the three parts were things already on my chart. The third was not, and I could have had it on its own for free.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.