Cyclic RSI vs RSI: 11,640 Backtests on S&P 500 Futures and SPY

TL;DR: the answer box

• Typical rule: plain RSI wins. Ret/DD (profit divided by the biggest dollar drop) of 4.40 against 3.65 on SPY, and 3.37 against 2.47 on ES futures.
• Best stable rule: Cyclic RSI wins, until a filter is added on SPY. 11.04 against 6.69 on SPY. Give both the same 200-day trend filter and RSI leads, 11.39 against 10.08.
• Only a little different. The two best rules move together at 0.79 on SPY. 93.4% of the Cyclic RSI buys came within 5 trading days of an RSI buy.
• No extra accuracy. Fixed RSI lines had the higher win rate in 9 of 9 groups of rules sorted by trades a year.
• One setting moved the median most. Dominant cycle. Median Ret/DD was 10.22 at a cycle of 4 and 3.33 at the default 20.
• Less pain, less money than buy and hold. 4.23% a year against 8.82%, with a worst fall of 13.3% against 56.47%.

What is Cyclic RSI?

RSI is the classic dip gauge. It compares recent gains with recent losses on a 0 to 100 scale. J. Welles Wilder built it. By convention its lines sit at 30 and 70 (StockCharts ChartSchool, RSI).

Cyclic RSI (cRSI) is Lars von Thienen's twist on it, from his book Decoding the Hidden Market Rhythm (TradingView script). It has three settings:

• Dominant cycle sets the speed. The RSI inside runs over half of it, so a cycle of 20 gives a 10-bar RSI.
• Vibration smooths the line.
• Leveling sets how far out the buy and sell lines sit.

The big change is the lines. They are not fixed at 30 and 70. They move with the last two cycles of readings. Here is what that looks like on the same prices.

How I tested Cyclic RSI against RSI

One rule, two indicators. Long only. Buy when the line crosses below its lower band. Sell when it crosses above its upper band. No stop, no time exit. For Cyclic RSI the bands move. For RSI they are fixed. That way the only thing that changes is the indicator.

Every grid setting, not just the defaults. 1,950 Cyclic RSI settings and 960 plain RSI settings. Each ran with a 200-day trend filter off and on, on two markets. That is 11,640 backtests. The fastest Cyclic RSI uses a 2-bar RSI, the same as the fastest plain RSI, so neither side gets a speed the other cannot reach.

Two markets. SPY from 1993 to 2026 on a $35,000 account, whole shares. E-mini S&P 500 futures (ES) from 2007 to 2026, one contract. One contract is worth 50 times the index (CME Group contract specs), $376,750 at the last close. That is why ES dollar figures are big.

The fills: Execution rule v2. The signal is read at a day's close. The order fills at the next day's open. One trade at a time.

Three plain terms.

• Ret/DD: net profit divided by the biggest dollar drop from a high point (the drawdown). Higher is better.
• Tradable: a setting with at least 50 trades. Fewer is too thin to trust.
• Best stable rule: one fair pick per family. Among settings whose neighbours also trade 50 times, take the one whose neighbours have the highest median Ret/DD. The rule was fixed before the run. It still picks from the same data, so treat it as a best case.

Is Cyclic RSI better than RSI?

It depends on which rule you compare. That is the whole answer.

The typical rule: plain RSI wins. Take the middle tradable rule from each family. Plain RSI scores 4.40 on SPY, Cyclic RSI 3.65. On ES, 3.37 against 2.47. Plain RSI also wins slightly more often, 76.6% against 74.5% on SPY.

Cyclic RSI gives you far more settings that trade. Its lines sit close to the action, so almost every setting gets used. A fixed line at 10 or 90 is rarely reached.

The best stable rules: Cyclic RSI wins. Cyclic RSI(4, 2, 24) scored 11.04 on SPY. RSI(2) 35/60 scored 6.69. The Cyclic RSI rule also fell less: a worst fall of 13.3% against 22.7%. Win rates were almost the same, 68.0% against 68.5%. So the two won about as often, but Cyclic RSI made more and fell less: $104,367 net against $93,097.

Costs do not change the order. With commission and slippage, the Cyclic RSI rule's net profit goes from $104,367 to $99,075. RSI's goes from $93,097 to $85,865.

Then add one filter. Give both rules the same 200-day trend filter. It only allows a buy when price is above its 200-day average. On SPY the winner flips.

On ES, Cyclic RSI still leads with the filter, 6.05 against 5.68. Two markets, two different answers.

What the winning Cyclic RSI actually is

Here is the catch. The best stable Cyclic RSI pick uses the fastest setting the test allows.

Both best stable Cyclic RSI rules use a dominant cycle of 4. That is a 2-bar RSI. Vibration 2 barely smooths it. What is left of the "cyclic" idea is the moving lines. The best plain RSI rule is RSI(2). Both families ran to the same corner.

Dominant cycle moved the median the most. On SPY the median Ret/DD jumps from 3.33 at the default 20 to 10.22 at a cycle of 4. The other two settings barely move the median. These are medians across all tradable rules, not a test that holds the other two settings still.

The heatmap shows the same thing from above. One bright row. Everything else sits in a narrow blue band.

The defaults rank near the bottom. The published Cyclic RSI(20, 10, 10) beat only 23.1% of tradable Cyclic RSI settings on SPY. Classic RSI(14) 30/70 beat only 5.1% of its family. It took 25 trades in 33.36 years.

Here is one real trade from the best stable Cyclic RSI rule, with RSI(2) underneath.

Does Cyclic RSI give a different equity curve than RSI?

A little. Not a lot.

The main pitch for Cyclic RSI is that it gives you a second strategy, one that does not rise and fall with RSI. I measured that with a move-together score (correlation) of daily profit and loss. A score of 1 means their daily profit and loss rise and fall in perfect step. A score of 0 means no measured tendency to move together.

They buy the same dips. The two best stable rules scored 0.79 on SPY and 0.82 on ES. 93.4% of the Cyclic RSI rule's 678 SPY buys came within 5 trading days of an RSI buy. On ES, 92.2%.

There is some difference. Across the top 25 rules of each family, Cyclic RSI against RSI scored 0.77 on SPY. Two RSI rules scored 0.90 with each other. So Cyclic RSI is a little different. Just not much.

Where a low number comes from. The two defaults score only 0.65. But they trade at very different speeds. RSI(14) 30/70 took 25 trades on SPY. Cyclic RSI(20, 10, 10) took 128. Only 6.2% of the default Cyclic RSI buys came within 5 trading days of an RSI(14) buy. So a low score between a busy rule and an idle one mixes two things: how different the indicators are, and how differently often they trade. This test cannot split the two.

Early on I lost money on systems that looked better in testing than they were. With overlap this high, two strategies that look different on paper are close to one bet placed twice. If you combine strategies, measure the overlap first. That is what AlgoChef is for: you bring your backtested strategies, it validates each one and builds the portfolio from them.

Do adaptive bands improve accuracy?

Not as Cyclic RSI packages them.

The two families trade at different rates. Win rate moves with trade rate. So I sorted the tradable rules into groups by trades a year, such as 0 to 2 or 6 and up. Then I compared the median results inside each group.

Plain RSI rules won more often in all 9 groups. They also had the higher Ret/DD in 7 of 9. Cyclic RSI changes the RSI length and smoothing as well as the lines, so this tests the whole package. As a package, it was not more accurate.

Does Cyclic RSI beat buy and hold?

Not on return. On pain, yes.

Every tradable rule made money, and that proves little. 100.0% of tradable rules in both families were profitable on both markets. The S&P 500 rose over the whole test, so a buy-only rule earns some of that rise for free. The yardstick is random buys with the typical Cyclic RSI rule's holding time. They asked for its 103 buys, but only 65 fit, because each one waits for the last to close. So it matches the hold, not the trade count.

Buy and hold makes more money. $551,738.32 against $104,367 for the best stable Cyclic RSI rule. That rule was in the market only 50.5% of the time.

The Cyclic RSI rule hurts far less. Divide yearly return by the worst fall and you get a pain score. Higher is better. Cyclic RSI scores 0.318. Buy and hold scores 0.156. Random buys score 0.068.

On ES, one contract on a $35,000 account is a leveraged bet. Buy and hold fell 114.35% of the account by March 2009. The best stable Cyclic RSI rule fell 37.1%.

The verdict, and the honest limits

What holds. On daily SPY and ES, Cyclic RSI is a tradable dip-buying tool. Its best stable rule beat RSI's on both markets with no filter, 11.04 against 6.69 on SPY, and it survived costs. Its equity curve is a little different from RSI's: 0.77 across the top rules, against 0.90 between RSI rules.

What does not.

1. "Better than RSI" depends on the comparison. The typical RSI rule won on both markets. The best rules swapped places on SPY once both had the same filter.
2. "Low correlation" did not hold for the best rules. They scored 0.79 and 0.82, and at least 92.2% of the Cyclic RSI buys landed within 5 trading days of an RSI buy.
3. "More accurate" did not show up. Plain RSI rules won more often in 9 of 9 like-for-like groups.
4. The best stable pick is the fastest one. A dominant cycle of 4, far from the default 20.

The limits, in plain words.

• One market family. SPY and ES are both the S&P 500. Stocks, forex and crypto are untested.
• The win-rate test compares the whole Cyclic RSI package with plain RSI. I did not put the moving lines on a plain RSI by themselves.
• The best rules on both sides sit at the fastest corner of the test. Neither can go faster, so their neighbourhood only has one side.
• The best stable rule was chosen from the same data it is scored on. Split in half, the SPY Cyclic RSI rule's average trade was 0.42% before 2010 and 0.46% after.
• No costs except the one check above. No compounding. SPY without dividends, for every rule and buy and hold alike.

What this means for you

1. Treat Cyclic RSI as a close cousin of fast RSI. Its stable picks on SPY and ES are a 2-bar RSI with moving lines. Like most cousins, it shares more than it differs.
2. If you use it, look at the dominant cycle first. Across the SPY grid, the median Ret/DD of rules ran from 3.33 at the default 20 to 10.22 at a cycle of 4. Grouped by vibration or leveling, the medians stayed within 3.07 to 4.00.
3. Check diversification at the same speed. Two rules that look unrelated may just trade at different rates. Count shared buys before you call it a second strategy.
4. Test both with the same filter. The SPY winner changed when both got the 200-day filter.
5. Judge it on pain, not return. 4.23% a year against 8.82% for buy and hold, with a 13.3% worst fall instead of 56.47%.

Cyclic RSI is a decent indicator. Its best rule is only a little different from RSI's best, at 0.79 on SPY. It is RSI's cousin, and it buys mostly the same dips.

Reference tables

What was tested.

1,950 Cyclic RSI settings and 960 RSI settings, filter off and on, on SPY and ES: 11,640 backtests.

The best stable rules, filter off.

Profit factor is gross profit divided by gross loss. Execution rule v2, next-open fills.

Win rate and Ret/DD at matched trade frequency.

Medians across rules with 50+ trades in each group, filter off. ES had no rule trading under 2 times a year.

Everything here is research, not investment advice. It is a backtest on historical data with the assumptions stated above. It is not a recommendation to buy or sell anything, and past results do not predict future ones.

Every couple of weeks I test one strategy everyone repeats, properly, and leave the parts that failed in. StatOasis.com/Overfit