Connors RSI: the 3-component formula, oversold levels and a test

Contents

1. What is Connors RSI, and does it beat plain RSI for short-term reversals?
2. How is Connors RSI calculated? The three components
3. How does Connors RSI differ from a standard RSI?
4. What counts as oversold on Connors RSI?
5. Which trend filter and exits do people pair with it?
6. What does a stand-in backtest on NQ daily bars show?
7. How should you test Connors RSI on your own market?
8. Frequently asked questions
9. References

What is Connors RSI, and does it beat plain RSI for short-term reversals?

Connors RSI is the average of three 0–100 readings: a 3-period RSI of the close, a 2-period RSI of the up/down streak, and the percent rank of today's return over 100 bars. It hits extremes in days, not weeks, so it gives more short-term reversal signals than plain RSI; whether they pay after costs is decided by your own test.

Connors RSI is a short-term oscillator built for one job: to flag the moment a market has fallen (or risen) too far, too fast, over the last two to five bars. The RSI indicator it builds on was published by Wilder (1978) and measures the balance of up and down closes over a lookback window. Connors RSI takes a very short version of that measure and adds two more readings, then averages the three. Think of three judges scoring the same dive: one judges the last three closes, one judges how many days in a row the market has fallen, and one judges how large today's drop was compared with the last hundred days. The final score is the mean of the three cards.

Does it beat plain RSI? Not by construction. A faster oscillator reaches its oversold zone more often, so it trades more often. More trades only help if the average trade after commission and slippage stays positive, and a faster signal is also more often wrong, because a two-day dip is less information than a three-week decline. Park and Irwin (2007) reviewed decades of studies on technical rules and found that results depend heavily on costs and on how many rules were tried before one was reported. The honest answer to "does it beat plain RSI" is therefore a measured one: run both with the same filter, the same exit and the same costs, and compare the per-trade result. This article explains the formula and the common rule, then shows how to run that comparison. A general view of which technical analysis tools and indicators can be tested as exact rules is in a separate article.

How is Connors RSI calculated? The three components

Each of the three components is a number from 0 to 100, and Connors RSI is their plain average, so a reading of 10 means all three parts agree that the last few bars were unusually weak. The default settings are 3, 2 and 100: a 3-period RSI, a 2-period RSI of the streak, and a 100-bar percent rank.

1. RSI(3) of the close. This is the standard RSI with a 3-bar lookback instead of the usual 14. It measures how much of the last three bars' movement was up versus down. Three up closes in a row push it near 100; three down closes push it near 0. The RSI indicator, its formula and 70/30 levels are explained in their own article.
2. RSI(2) of the streak. A streak is the count of consecutive closes in one direction: +1, +2, +3 for three up closes in a row, −1, −2, −3 for three down closes, and 0 on a bar that closes unchanged, which resets the count. The streak is itself a series of numbers, so you can run an RSI on it. A 2-period RSI of the streak drops toward 0 as the streak grows more negative and rises toward 100 as it grows more positive.
3. Percent rank of the 1-day return over 100 bars. The percent rank is the share of the last 100 one-bar returns that were smaller than today's return, expressed from 0 to 100. If today's drop is the worst of the last 100 bars, the rank is 0. If today's gain is the largest, the rank is 100. A rank of 16 means only 16 of the last 100 returns were worse than today's.

The one formula to remember:

Connors RSI = (RSI(3) of close + RSI(2) of streak + percent rank of 1-day return) ÷ 3

Here RSI(3) of close is the 3-bar RSI of price, RSI(2) of streak is the 2-bar RSI run on the up/down streak counter, and percent rank of 1-day return is the 0–100 position of today's return among the last 100 returns. Dividing by 3 keeps the result on the same 0–100 scale as each part.

How does Connors RSI differ from a standard RSI?

A standard 14-period RSI measures only the balance of up and down closes over about three weeks of daily bars, while Connors RSI adds streak length and how extreme today's return is, which is why it reaches extremes in two or three days instead of two or three weeks. A 14-period RSI is like a room thermometer: it tells you the average temperature over a while and moves slowly. Connors RSI is like a smoke alarm: it reacts to the last few minutes and goes off often, sometimes for burnt toast.

Property	Standard RSI (14)	Connors RSI (3, 2, 100)
Inputs	Up and down closes only	Up and down closes, streak length, rank of today's return
Lookback	14 bars (about three weeks on daily bars)	3 bars, 2 bars of streak, 100 bars for the rank
Time to reach an extreme	Usually several weeks of one-sided movement	Usually two to four consecutive bars in one direction
Usual thresholds	30 oversold, 70 overbought	10 or 20 oversold, 80 or 90 overbought
Typical use	Trend strength, divergence, swing timing	Short-term pullback entry inside a trend
What it misses	Sharp two-day drops that do not move a 14-bar average much	Slow, grinding declines with small daily moves and no long streak

The practical difference shows up in a normal pullback. Suppose a market in an uptrend falls for three days after a long run of gains. The 14-period RSI still carries eleven earlier up days in its window, so it may dip from 65 to about 38 and never reach 30. Connors RSI sees three down closes (low RSI(3)), a streak of −3 (low streak RSI) and, if the third day was a large drop, a low percent rank. It can read 9 on that third day. Same price action, two very different readings, because they ask different questions.

Speed is also the source of the extra false signals. In a market that falls for ten days, Connors RSI will read oversold on day three, day five and day eight, and the first two of those entries lose. A 14-period RSI may give one signal near the end of the decline, or none. Which behaviour you want depends on your exit rule and your costs, which is why the next two sections matter more than the oscillator itself.

What counts as oversold on Connors RSI?

Common thresholds are below 10 or below 20 for oversold and above 80 or 90 for overbought, far stricter than the 30 and 70 used for plain RSI, and the stricter the level the fewer but cleaner the signals, which you check by counting trades per year at each level. Because Connors RSI is an average of three fast readings, it visits 30 very often; a 30 line would produce a signal every few days and most of them would be noise. The useful range starts near 20 and ends near 5.

Threshold	What it requires	Signal frequency	Typical use
Below 20	Two or three weak bars; one component may be mild	High	Looser entry, more trades, more small losers
Below 15	All three components clearly low	Medium	A middle setting many testers start with
Below 10	Three down bars and a large drop ranked near the bottom of 100 bars	Low	Strict entry, fewer trades, larger average bounce expected
Below 5	Near-worst return of the last 100 bars plus a long streak	Very low	Rare; a few signals per year on daily bars

The overbought side mirrors this: above 80 is loose, above 90 is strict, above 95 is rare. For a short-selling rule the same logic applies in reverse. The right level is not a matter of taste. Count how many signals each level produces per year on your data, and check whether the count gives you enough trades to judge. A level that fires six times a year gives you sixty trades in ten years, which is a thin sample. A level that fires forty times a year gives you four hundred, but many of them will be shallow dips that cost more in slippage than they earn.

This counting step is the same one that turns any discretionary reversal idea into a testable rule. A double bottom chart pattern only becomes testable once its exact rules are written down and its trades can be counted; a Connors RSI threshold is the same decision in one number.

Which trend filter and exits do people pair with it?

Most published rule sets buy an oversold reading only when price is above a long moving average and exit when Connors RSI rises back above 50 to 70 or price closes above a short moving average, so the rule has no fixed stop and the exit does most of the risk control. A trend filter is a condition that only allows trades in the direction of the longer trend; here it stops the rule from buying every dip in a falling market. The exit is time-based in effect: a bounce of one to five bars usually lifts Connors RSI above 70, and if no bounce comes, the trade stays open until one does.

1. Trend filter: is the close above the 200-day moving average? If not, skip the signal. The 200-day line is the common choice because it is slow enough that the filter rarely flips during a pullback. Some testers use 100 or 50 days; shorter filters let more trades through in bear markets.
2. Entry: is Connors RSI below 10 at the close? Some rule sets use 5, 15 or 20. If yes, buy at the next bar's open. The signal is read at the close and acted on at the next bar, which is the realistic fill.
3. Hold: wait until Connors RSI closes above 70, or the close rises above the 5-day moving average. Either condition signals that the bounce has arrived. Some rule sets use 50 or 60 as the exit level; a lower exit shortens the hold and reduces the average win.
4. Exit: sell at the next bar's open. Again, decide on the close, fill on the next bar.

The missing fixed stop is the part to think about. A mean-reversion entry buys weakness, so a stop placed a little below the entry is hit on exactly the bars when the market keeps falling, which is often. Many testers therefore run the basic rule without a stop and let the trend filter and the exit do the work, then test a wide stop separately to see what it costs. This is the opposite design from a Donchian channel breakout, which buys strength and relies on a stop below the channel; comparing the two on the same market shows how differently reversal and breakout rules spend their risk. The short side of the Connors RSI rule mirrors everything: price below the 200-day line, Connors RSI above 90, exit below 30 or below the 5-day line.

What does a stand-in backtest on NQ daily bars show?

The run below uses a DMI rule as a stand-in for a short-term reversal entry, kept as one candidate out of 1,024 for its return against its drawdown, and its numbers show what a daily-bar NQ rule with costs looks like, not what Connors RSI earned. The run tested Nasdaq-100 futures (NQ) daily bars from 2019-01-02 to 2025-06-30, long and short, one contract, with $20 commission and $10 slippage charged on every trade. The DMI indicator was its only entry and exit; there was no stop, no target and no trend filter. It is a stand-in built from a standard indicator, not Connors RSI itself.

Three things in the box are worth reading the way you would read your own Connors RSI test. First, the search tried 1,024 candidates and kept this one for its return against its drawdown; choosing the best after the fact flatters the kept candidate, because some of its edge over the others is luck that lined up with this history. Bailey et al. (2014) show that the more variants a search compares, the larger this selection bonus grows. Second, 76 trades in six and a half years is about a dozen a year, a thin sample; the uncertainty around the $662.76 average trade is wide. Third, time in market of 7.72% is the profile a short-hold rule has: in the market for a few days at a time, flat the rest. A Connors RSI rule on daily bars shows a similar profile, which means most of its return comes from a small number of bars, and a small change in fills or costs moves the result a lot.

The cost line also deserves a check against the contract. On NQ one tick is 0.25 index points, worth $5.00 at $20 per index point, so $10 of slippage per trade is two ticks. Exchanges can change contract values, so confirm them on the exchange's contract page, or put the contract into the free futures contract calculator to see tick value and point value side by side. None of this says a DMI rule or a Connors RSI rule earns these numbers going forward; it says what a daily-bar NQ rule looks like once commission and slippage are inside every trade.

How should you test Connors RSI on your own market?

Set costs and next-bar fills first, run the plain RSI and the Connors RSI rule with the same filter and exit on the same data, compare average trade after costs and trade count, and only then judge whether the extra components add anything. A fair head-to-head test changes exactly one thing: the entry oscillator. Everything else stays identical, otherwise you will not know whether the difference came from the oscillator or from a different exit.

1. Fix the market, the bar size and the period. Daily bars are where this rule is usually described. Write down the start and end dates and keep the most recent block out of every research run for one final test.
2. Set commission and slippage per trade before the first run. Use your broker's commission plus at least one tick of slippage per side for a liquid future. Use next-bar fills: the signal is read at the close, the fill is at the next open.
3. Write both rules with the same filter and exit. Rule A: RSI(2) below 10 at the close, price above the 200-day moving average, exit when the close rises above the 5-day moving average. Rule B: identical, except the entry reads Connors RSI below 10.
4. Run both and compare three numbers. Average trade after costs, number of trades, and maximum drawdown. A rule that trades twice as often with half the average trade has gained nothing and paid more in costs.
5. Run the neighbouring levels. Try 5, 15 and 20 for both rules. If the result holds only at one level, you found a lucky number, not an edge.
6. Judge on the untouched block once. Run the winner of steps 4 and 5 on the reserved data with the rule you wrote in step 3, unchanged.

Step 2 is where most informal comparisons go wrong. A Connors RSI rule trades more often than an RSI(14) rule and somewhat more often than an RSI(2) rule, so a comparison without costs favours it automatically; the extra trades look like extra profit until each one pays its two ticks and its commission. In AlphaPrime (Valynox) the backtesting of each candidate already includes the commission, slippage and next-bar execution you set, so when two oscillator rules sit side by side in the results the gap between them is net of what the extra trades cost. Whether you run the comparison there or in a spreadsheet, the order is the same: costs first, identical rules second, per-trade numbers third.

The next thing to do is small. Load one year of daily bars for the market you trade, compute Connors RSI with the 3, 2, 100 defaults, and count how many closes fall below 20, below 15 and below 10. That count tells you which threshold gives you enough trades to judge, and it is the number you need before any of the rule tests above makes sense.

Frequently asked questions

What are the default Connors RSI settings?

The usual defaults are 3, 2 and 100: a 3-period RSI of the close, a 2-period RSI of the up/down streak, and a 100-bar lookback for the percent rank of the one-bar return. The three readings are averaged into one 0–100 value. Changing the first two numbers changes how fast the indicator reacts; changing the third changes how far back today's return is compared.

Does Connors RSI work on futures or only on stocks?

The formula applies to any series of closes, so it can be computed on index futures, commodities or stocks. Whether a rule built on it holds after costs is a separate question that only a backtest on that market answers. On futures, decide which session's close counts as the daily close before computing it, because a full-session close and a day-session close give different streak counts and different returns.

Can Connors RSI be used for short trades?

Yes, by mirroring the long rule. The common short version requires price below the long moving average, enters when Connors RSI closes above 90 (or 80), and exits when it falls back below 30 or the close drops below the 5-day moving average. Test the long side and the short side separately, because index futures have spent more time rising than falling and the two sides rarely behave alike.

Does Connors RSI repaint?

No. All three components use closed-bar data, so a finished bar's value does not change later. The value of the bar that is still forming does move until that bar closes, so a rule should read the signal at the close and act on the next bar. That is also how a realistic backtest fills the trade.

Is Connors RSI better than RSI(2)?

Not automatically. RSI(2) is already a fast oscillator, and Connors RSI adds the streak and the return rank on top of a similar short RSI. On some markets the extra parts filter out shallow dips; on others they add trades that cost more than they earn. Run both with the same filter, exit and costs on the same data and compare average trade after costs, trade count and drawdown.

References

• Wilder, J. W. (1978). New Concepts in Technical Trading Systems. Trend Research.
• Park, C.-H., & Irwin, S. H. (2007). What Do We Know About the Profitability of Technical Analysis? Journal of Economic Surveys, 21(4), 786–826.
• Bailey, D. H., Borwein, J., López de Prado, M., & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.
• CME Group. E-mini Nasdaq-100 Futures contract specifications (contract size $20 × index, minimum tick 0.25 index points = $5.00). cmegroup.com/markets/equities/nasdaq/e-mini-nasdaq-100.contractSpecs.html