TL;DR: the answer box
- Cumulative RSI is a sum. Two RSI(2) readings added together run from 0 to 200. "Buy below 40, exit above 25" on an average is buy below 80, exit above 50 on the sum. Same trades.
- The 80/50 rule survives next-open fills. ES 2007 to 2026: 546 trades, a 66.5% win rate and a worst fall of 27.9%. Plain RSI(2) at the same levels: 790 trades, 62.9% and 43.0%.
- Summing is not a general upgrade. Two readings lost the Ret/DD matchup more often than not in 23 of 24 test slices. They win per trade and lose on trade count.
- More readings, lower scores. The typical ES rule fell from a Ret/DD of 4.53 with one reading to 2.24 with seven.
- RSI(2) at 40 / 25 shows no sign of fading. Its profit factor was higher in the second half of all three samples: 1.49 against 1.42 on ES.
- Four losing years on ES, not one. The worst was -$17,550 in 2018, on one contract.
What is Cumulative RSI, and how did I test it?
RSI first. The Relative Strength Index, from J. Welles Wilder's 1978 book, scores recent up days against recent down days on a 0 to 100 scale (StockCharts ChartSchool, RSI). RSI(2) looks back just two sessions, so it swings hard. Larry Connors made it popular as a short-term pullback gauge (StockCharts ChartSchool, RSI(2)). The idea is mean reversion: buy a market that just dropped, betting on a bounce.
Cumulative RSI adds up the last few readings. Connors and Cesar Alvarez defined it in High Probability ETF Trading (2009). Instead of reading only today's RSI(2), you add today's to yesterday's.
Summing and averaging pick the same days. Some tests average the two readings instead of adding them. That gives the same trades, as long as you halve the levels. Averaged 40 and 25 are summed 80 and 50. The trap is copying one scale's numbers onto the other. More on that below.
Here is one real trade from the 80/50 rule on SPY.
The rules. Long only. Buy the next open when Cumulative RSI closes below the buy level. Sell the next open when it closes above the exit level. No time exit. I moved five settings at once: the RSI period, how many readings to add up (one reading is plain RSI), the buy level, the exit level, and a trend filter that only buys above the 200-day average. That makes 9,240 rules per market. The full list is in the reference tables at the end.
The markets. SPY daily bars from 1993-02-02 to 2026-06-12 on a $35,000 account, whole shares, no compounding. E-mini S&P 500 (ES) and E-mini Nasdaq-100 (NQ) futures from 2007-01-03 to 2026-07-01, one contract each. One ES contract controlled $376,750 at the last close, 10.8 times the account (CME Group contract specs). That is 18,480 rules on the S&P 500 and 27,720 with NQ.
The fills: Execution rule v2. Every signal is read at the close and buys the next day's open. Every exit is decided at a close and sells the next open. One position at a time. A close-fill test buys at the close of the signal bar, a price you only know once it has printed.
The scores, in plain words.
- Win rate: the share of trades that made money.
- Drawdown, or worst fall: the biggest drop from a high point.
- Profit factor: dollars won for each dollar lost.
- Ret/DD: net profit divided by the biggest dollar drawdown. Profit per dollar of pain. Higher is better.
- Median, or typical: the middle value.
A rule needs 50 trades before I count it as reliable. Every result also sits next to buying and holding, and next to random entries.
Does Cumulative RSI beat RSI(2) at the same levels?
For the 80/50 rule, yes on SPY and ES, and no on NQ. Here is the 80/50 rule against one RSI(2) reading at the same per-reading levels, with the same exits and the same fills.
On ES the gap is wide. Ret/DD 9.62 against 5.00. The 80/50 rule ranks near the very top of the 5,376 reliable ES rules.
It trades less and earns more per trade. On ES it takes about 31% fewer trades and sits in a trade 22.7% of the time, against 29.2%. A close-fill test of this rule on 2006 to 2025 reported 542 trades and a 67% win rate. Next-open fills give 546 and 66.5%. The rebuild reproduces the trade count and the win rate.
The biggest gain is in the worst fall. Adding the second reading cut the account's worst drop on every market, NQ included.
On NQ, plain RSI(2) still edges it. Ret/DD 6.91 against 6.31, even with the smaller worst fall. So "smoothing beats RSI(2) on every index" fails here: the 80/50 rule had the higher Ret/DD on SPY and ES, and the lower one on NQ.
It clears the random bar. I drew 1,000 sets of random entries on each market. The 80/50 rule beat 95% of them on every market, the random 95th percentile: 9.62 against 4.83 on ES, 8.10 against 5.52 on SPY, 6.31 against 5.18 on NQ.
Is summing better as a rule, or is 80/50 one lucky cell?
One strong cell, not a better rule. This is the test that matters, and I could not find another Cumulative RSI page that runs it. I took every buy and exit level and ran it twice: once with one RSI(2) reading, once with two summed. Then I counted which won.
Summing wins the trade and loses the account. On SPY, two readings gave a bigger average trade in 98.2% of the 165 pairs. They traded less in every pair. They made more net profit in only 31.5%, and had the better Ret/DD in only 30.3%. Bigger trades, fewer of them, and less profit in most pairs.
It is not an RSI(2) quirk. I ran the same count for RSI periods 2 to 5, filter off and on, on all three markets. That is 24 slices.
Most ES rules don't clear the luck bar. On ES, 88.1% of random entry sets made money. Beating 95% of random entry sets is hard. On ES, only 691 of 5,376 reliable rules cleared that bar.
That is why one good cell proves little about the method. On SPY, the top reliable rule scored 16.86. The 27 cells around it had a median of 10.50. The top cell of a big grid looks better than the ground it stands on.
Do more summed readings help?
No. For RSI(2) with the trend filter off, every extra reading lowered the typical Ret/DD on all three markets. Think of it as a pivot table: readings down the side, markets across the top.
Longer sums mean longer holds and fewer trades. Across the whole ES grid, the typical one-reading rule holds 4.13 bars and averages 0.31% a trade over 244 trades. With seven readings it holds 9.94 bars and averages 0.47% over 111 trades. Bigger trades. Less than half as many. Ret/DD 2.08 against 4.08.
Is the 80/50 rule a stable region or a peak?
It sits on a profitable plateau, and on ES it sits at the top of it. Picture the grid as a landscape. On a plateau, nudge a setting and the result barely moves. On a spike, one small nudge drops you off a cliff. A spike is what curve-fitting looks like.
The neighbours all made money. I checked the 18 cells around the rule: buy level one step either side, exit level one step, and one, two or three readings. On SPY, ES and NQ, 18 of 18 were profitable.
But the patch is small. With two readings, only 41 of 164 ES settings beat 95% of random entry sets. With one reading, 57 of 165 did. The 80/50 rule is one of only 41 two-reading settings that cleared it.
Don't plan with the peak. On ES the rule scored 9.62. Its 18-cell neighbourhood had a median of 5.28. I lost roughly $270K early in my career, partly to overfit systems that looked better than they were. That is why I quote the ground around a best cell, not the cell. Plan with 5.28.
If you trade your own version, ask the same question: is your setting on a plateau, and does its trade history survive a reshuffle and a fresh data test? That check is what AlgoChef runs on a strategy you have already backtested.
Why are Cumulative RSI levels like 35 and 65 not RSI levels?
Because they are sums. Connors' published rule buys when two summed RSI(2) readings close below 35. That is an average reading of 17.5, a deep pullback. Levels of 40 and 25 from an averaged test are averages. Type them straight into a summed indicator and you buy at an average of 20 and sell at 12.5. A different, rarer rule.
The trap does not always cost you. On NQ the copied rule happened to score higher, 7.38. It always means you are testing something else.
Connors' own rule trails plain RSI(2). With next-open fills it scored 5.69 on SPY, 3.42 on ES and 1.33 on NQ. One RSI(2) reading at the same per-reading levels and the same filter scored 6.44, 5.74 and 5.78. The plain reading won on all three markets.
Has RSI(2) stopped working because everyone uses it?
Not in these samples. A common reason to switch to a smoothed version is that RSI(2) got popular and its edge faded. I cut each sample at its middle bar and compared plain RSI(2), buy below 40 and sell above 25, in the two halves.
Two halves are a coarse test. On that test, no market showed fading. On SPY the average trade slipped from 0.16% to 0.15% while the profit factor rose from 1.34 to 1.38.
How many losing years did it have?
Four on ES, not one. A close-fill test of this rule on 2006 to 2025 reported one losing year. With next-open fills on 2007 to 2026, it lost money in 4 of 20 calendar years.
Three were small. 2018 was not. 2010 lost -$112, 2022 -$412 and 2023 -$1,162. 2018 lost -$17,550. Plain RSI(2) at the same levels had 3 losing years on ES. On SPY both versions had 5 losing years of 34.
The window and the fills both changed, so this does not prove the fill caused the gap. It does show the ride was not smooth.
Does Cumulative RSI beat buy and hold?
Not on profit. On pain, by a wide margin.
Buy and hold made more money. On SPY it grew 8.82% a year against 3.73%. The rule was only in a trade 24.0% of the time.
The rule hurt far less. Its worst fall on SPY was 10.5%. Buy and hold fell 56.47% by March 2009. Divide the yearly return by the worst fall and the rule scores 0.354 against 0.156.
Costs do not break it. Commission and slippage take the rule's SPY profit from $83,638 to $75,801 over 976 trades. On ES, from $221,300 to $205,466 over 546 trades.
Does it work on ES futures as well as SPY?
Yes, with bigger swings. The same rule wins about as often and spends about as much time in a trade on both. ES traded one leveraged contract, from 2007. SPY bought whole shares, from 1993. Those two changes come together here, so I can't split the bigger swings between them.
One ES contract is 10.8 times a $35,000 account. So the same signal that cost SPY 10.5% at its worst cost the ES account 27.9%. Buy and hold on ES fell 114.35% of the account by March 2009, more than the whole account. If you trade futures, size for that leverage and test on futures data.
The honest limits
- Three US index markets. Nothing here says how Cumulative RSI behaves on single stocks, bonds or commodities. Dow futures (YM) are not in my data.
- I chose the grid and the 80/50 rule. That is selection, which is why I quote neighbourhoods and matched pairs, not the best cell.
- No costs except the cost lines above. No compounding. SPY without dividends. NQ has no cost line because the engine has no NQ tick size.
- ES and NQ are one contract on $35,000. That is a leverage choice, not a recommendation.
- Weighted and uneven versions of the sum were not tested.
What this means for you
- Check which scale your indicator uses. Two summed RSI(2) readings run 0 to 200. A level of 35 on the sum is 17.5 on one reading.
- Don't add readings hoping for a better Ret/DD. On ES the typical score fell from 4.53 with one reading to 2.24 with seven.
- If you trade the 80/50 rule, plan with its neighbourhood. On ES that is a median Ret/DD of 5.28, not the 9.62 peak.
- Don't drop RSI(2) because it is popular. In these samples its profit factor at 40 / 25 was higher in the second half.
- Test with next-open fills. A close-fill test found one losing year on ES. Next-open fills found four.
Summing RSI(2) readings bought bigger trades, fewer of them, and in most matched pairs a lower Ret/DD. The 80/50 rule scored 9.62 on ES. The ground around it scored 5.28. Trade the ground.
Reference tables
The headline rules on all three markets. RSI(2) buys below 40 and sells above 25. Cumulative RSI(2, 2) buys below 80 and sells above 50. Next-open fills, no costs.
| Market | Rule | Trades | Win rate | Avg trade | Profit factor | Ret/DD | Worst fall |
|---|---|---|---|---|---|---|---|
| SPY | RSI(2) | 1370 | 62.2% | $55 | 1.36 | 6.54 | 15.7% |
| SPY | Cumulative RSI(2, 2) | 976 | 65.0% | $86 | 1.57 | 8.10 | 10.5% |
| ES | RSI(2) | 790 | 62.9% | $269 | 1.47 | 5.00 | 43.0% |
| ES | Cumulative RSI(2, 2) | 546 | 66.5% | $405 | 1.75 | 9.62 | 27.9% |
| NQ | RSI(2) | 786 | 62.8% | $468 | 1.59 | 6.91 | 36.2% |
| NQ | Cumulative RSI(2, 2) | 556 | 64.7% | $622 | 1.74 | 6.31 | 23.9% |
Worst fall is the largest percentage drop of a $35,000 account.
The settings I swept.
| Setting | Range | Values |
|---|---|---|
| RSI period | 2 to 5 | 4 |
| Readings summed | 1 to 7 (1 is plain RSI) | 7 |
| Buy level, per reading | 5 to 55, step 5 | 11 |
| Exit level, per reading | 25 to 95, step 5 | 15 |
| Trend filter (close above its 200-day average) | off, on | 2 |
9,240 rules per market. Levels are set per reading, so with three readings a buy level of 40 is a sum below 120.
The 18-cell neighbourhood of the 80/50 rule. Buy level 35, 40 or 45 per reading, exit level 25 or 30 (25 is the grid edge), one, two or three readings.
| SPY | ES | NQ | |
|---|---|---|---|
| Cells profitable | 18 of 18 | 18 of 18 | 18 of 18 |
| Ret/DD range | 5.90 to 12.55 | 2.91 to 9.62 | 3.16 to 7.42 |
| Median Ret/DD | 7.88 | 5.28 | 6.04 |
| The 80/50 rule itself | 8.10 | 9.62 | 6.31 |
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
