TL;DR: the answer box
- The trades really are different. Only 28% of Ultimate C%'s SPY trades start on an RSI(2) entry day, and 24% on ES. For RSI(3), a near twin of RSI(2), it is 54% and 56%.
- Different days, not unrelated profits. Their daily profits still move together, with a correlation of 0.57 on SPY and 0.61 on ES.
- Beside RSI(2), it earned its place on SPY. One account of each made $173,895 with a $15,704 worst drawdown, a Ret/DD of 11.07. Two accounts of RSI(2) made $185,492 with $20,325, a Ret/DD of 9.13. On ES the gain was thin: 4.47 against 4.37.
- On SPY the two never lost money in the same calendar year. Their yearly profits correlated at -0.28.
- On its own, each has its own numbers. Buy below 30, sell above 65 or after 5 bars: a win rate of 60.7% with a Ret/DD of 4.91 for Ultimate C% on SPY, and 68.8% with 9.13 for RSI(2).
- Its settings have a stable block. All 8 settings from buy below 25 to 40 and sell above 65 to 70 passed a stability test on both markets. The 30/65 setting sits inside it.
What is Ultimate C%, and how is it different from RSI(2)?
A mean reversion indicator tries to spot a market that has dropped too far, too fast, so you can buy the bounce. The one most traders reach for is RSI(2). That is Welles Wilder's Relative Strength Index over just two days: a 0 to 100 score of recent gains against recent losses.
Ultimate C% asks a different question. Where does today's percent change rank among recent days? It checks three windows: the last 2, 8 and 32 days. In each window, the weakest day scores 0 and the strongest scores 100.
The three answers are blended with weights 16, 4 and 1, so the short window counts most. Then the blend is smoothed over 3 days. The final reading can also sit a few points outside 0 to 100. That is part of how it is built.
Think of RSI(2) and Ultimate C% as two different thermometers. They both measure "cold", but they are not the same instrument, so they don't read cold on the same days. That is the point of having two.
Here is what that looks like on a real trade.
At the signal close, Ultimate C% read 22.0. RSI(2) read 92.0. Same close, same market, opposite ends of the scale. The trade bought the open of 2014-11-07 at 2729.25 and sold the open of 2014-11-14 at 2735.00. That is +5.75 points, or $288 on one contract.
It is not a cherry-picked winner. On 277 of Ultimate C%'s 484 ES trades, RSI(2) was at or above 30 at the signal close, so it had no signal at all. This one is the trade closest to their median profit.
How I tested it
I designed Ultimate C%, so the test uses the same rule I launched it with (buy below 30, sell above 65 or after 5 bars) and Ultimate C%'s published settings.
- The rule: buy when the indicator closes below 30. Sell when it closes above 65, or after 5 bars if it has not.
- Three indicators on that rule: Ultimate C% at its published settings, RSI(2), and RSI(3). RSI(3) is the yardstick. It shows what a near twin of RSI(2) looks like: almost all of its signal days are RSI(2) signal days.
- Two markets: SPY from 1994-02-01 and ES (E-mini S&P 500 futures) from 2006-01-03, both to 2026-08-20. That is 32.55 years of SPY and 20.63 years of ES.
- Every setting, not just one: buy levels from 10 to 40, sell levels from 50 to 80, and a time cap of 5 bars, 10 bars or none. That is 147 settings per indicator per market, and 594 backtests in all.
- Execution rule v2: the signal is read at the close and the trade buys the next open. The exit is ordered at a close and sells the next open. One position at a time.
- Sizing: SPY on a $35,000 account in whole shares, no compounding. ES on one contract.
- No costs in the headline numbers. With commission and slippage, RSI(2)'s ES net profit goes from $190,838 to $174,278, and Ultimate C%'s from $138,375 to $124,339. RSI(2) still makes more after costs.
The scores, in plain words:
- Win rate: the share of trades that made money.
- Drawdown: the biggest drop from a high point, in dollars.
- Ret/DD: net profit divided by that biggest drop. Profit for each dollar of pain. Higher is better.
- Profit factor: dollars won for each dollar lost.
- Correlation: how closely two sets of profits move together. 0 means unrelated, 1 means in lockstep.
Does Ultimate C% trade different days from RSI(2)?
Yes. This is what it was built for, and it holds up.
On SPY, Ultimate C% took 758 trades and RSI(2) took 909. Only 216 started on the same day. That is 28% of Ultimate C%'s trades.
Now look at RSI(3), the near twin.
98% of RSI(3)'s signal days are RSI(2) signal days too. Switching from RSI(2) to RSI(3) adds almost no new signal days.
Ultimate C% is a different animal. RSI(2) agrees on only half of its signal days. The two sit in a trade together on a third of the days either one is in. On ES the picture is the same: 47% of signal days shared and 24% of trades.
Now the honest half. Different days is not the same as unrelated profits. Both indicators buy weakness in the same index.
Their daily profits correlate at 0.57 on SPY and 0.61 on ES. That is lower than the twin's 0.80 and 0.79. It is still well above zero.
Does adding Ultimate C% beside RSI(2) diversify the strategy?
On SPY, yes. On ES, barely.
This is the test that matters if you already trade RSI(2). Say you have two accounts. You can run RSI(2) in both. Or you can run RSI(2) in one and Ultimate C% in the other. Same money, same rule.
On SPY, the pair gave up 6% of the profit and cut the worst drawdown by 23%. The pair made $173,895 with a worst drop of $15,704. Two accounts of RSI(2) made $185,492 with a worst drop of $20,325.
The pair's Ret/DD of 11.07 is higher than either indicator alone. That is what diversification is supposed to buy you: a little less return for a lot less pain.
On ES, the gain was thin. The pair gave up 14% of the profit to cut the worst drawdown by 16%. Ret/DD moved from 4.37 to 4.47. Real, and small.
The yearly view backs up the SPY result. Over 32 calendar years, Ultimate C% lost money in 6 and RSI(2) in 4. Never in the same year. Their yearly profits correlated at -0.28. On ES it is weaker: a yearly correlation of 0.76, and 1 shared losing year out of 20.
How do Ultimate C% and RSI(2) compare on their own?
Each one has its own numbers. On the launch rule, RSI(2) has the higher Ret/DD on both markets.
On SPY, the profit factor is the same: 1.47 each. RSI(2) made more, $92,746 against $81,148. Its worst drawdown was smaller too, $10,162 against $16,532. Ultimate C% holds longer, 4.3 bars on average against 3.2. It sits in the market 40.0% of the time against 35.4%. More time exposed, no more profit per dollar risked.
On ES, the drawdowns flip. Ultimate C%'s worst was $35,262. RSI(2)'s was $43,700. But RSI(2) made more money and has a profit factor of 1.42 against 1.32, so its Ret/DD is the higher one: 4.37 against 3.92.
Why one undated test is not enough
When I launched Ultimate C%, a test on ES futures put its Ret/DD at 9.4 against RSI2's 4.7. That test gave no trade count, no dates and no fill rule.
On 20.63 years of ES with next-open fills, that order does not hold: 3.92 against 4.37. I can't rebuild the launch test, so I won't guess why the numbers moved. The lesson is simple. A ranking from one undated test of one setting did not reproduce on ES from 2006 to 2026 under Execution rule v2. It was my own ranking.
Is it just the 30 and 65 setting?
No. That is why I tested all 147 settings in the sweep, on both markets.
I only compared settings where both indicators traded at least 100 times. On SPY, RSI(2) had the higher Ret/DD in 112 of 116. The typical setting scored 8.61 for RSI(2) and 4.30 for Ultimate C%. On ES it is a coin flip: Ultimate C% was higher in 55 of 105, with typical scores of 4.53 and 4.52. RSI(2) still had the higher profit factor in 67 of those 105.
One thing to keep in mind. Ultimate C% fires less often at the same level: a typical 582 trades against 794 on SPY. Over a fixed window, more trades tend to lift Ret/DD.
Which Ultimate C% settings are stable?
A good setting sits on a plateau, not a spike. On a plateau, you can nudge the setting and the result barely moves. On a spike, one small nudge drops you off a cliff. A spike picked from many tries is the best draw of a search, so I don't use one.
So I fixed a test before looking at the answer. A buy and sell pair passes if it has at least 100 trades and a profit factor of 1.3 or more at every time cap (5 bars, 10 bars and none), on SPY and on ES.
16 of 49 Ultimate C% pairs passed. The stable block is buy below 25 to 40, sell above 65 to 70. All 8 of 8 pairs there passed, and the 30/65 launch setting is inside it.
With a 5-bar cap, that block scored a Ret/DD of 4.09 to 7.28 on SPY, on 584 to 973 trades. On ES it scored 2.83 to 6.02.
Deep buy levels, below 10 to 20, mostly fail, for two different reasons. Below 10 and 15, all 14 pairs miss on both counts: too few trades and too low a profit factor. Below 20, 6 of 7 pairs trade often enough but miss on profit factor.
For comparison, RSI(2) passed in 47 of 49 pairs. Its stable ground is wider. Ultimate C% has a smaller, clear stable block: the one that passed on both SPY and ES over these years.
Don't chase the peak. The single best Ultimate C% setting on ES scored a Ret/DD of 9.71, on 304 trades. It is the top of 147 tries, picked after the fact. Quote the peak and you quote the best draw of a search. Quote the block and you describe the surface.
Is the RSI(2) edge fading because everyone uses it?
Not on this evidence. The usual story is that RSI(2) is crowded, and crowded edges fade. I can't measure crowding. I can measure wear: if the edge were wearing out, its profit factor should drop from the first half to the second.
RSI(2)'s profit factor went from 1.47 to 1.48 on SPY, and from 1.45 to 1.42 on ES. No sign of wear.
Ultimate C% had the weak start: a profit factor of 1.12 on ES from 2006 to 2016, then 1.41. Two halves are not a trend, so I would not read a comeback into it.
Do the signals beat random entries?
Mostly. Ultimate C% on ES is the weak spot.
The average SPY bar was up 0.0407% over these years, and random long trades made money too. That is why the sweep and the 30/65 rules are checked against a random control: random entries asked for as often as the typical setting trades (628 times on SPY, 420 on ES), held 4 bars and sold at the open the same way. Of those, 476 of 628 became trades on SPY and 320 of 420 on ES. It ran 10 times with fixed seeds. On SPY it made $21,080 on average, with a spread of $11,994. On ES, $81,625, with a spread of $39,707.
The "luck bar" is that average plus two spreads. A setting that clears it made more than the random average by over two spreads.
On SPY, 100.0% of comparable RSI(2) settings cleared the bar, and 95.3% of Ultimate C%'s. On ES, 87.5% and 59.0%.
The 30/65 rule tells a similar story, with one limit: it trades more often than the control did, so this is a looser match. On SPY, Ultimate C% finished 5.0 spreads above the control's average. On ES, only 1.4. RSI(2) on ES finished 2.8 spreads above.
And the bigger picture, which every mean reversion article should print:
Buy and hold SPY made $518,686 on the same $35,000 over the same 32.55 years, with a worst fall of 56.45%. The best of the three indicator rules made $92,746, while in the market about a third of the time. On these markets and years, the indicator rules earned far less than holding, with far less time exposed.
The verdict, and the honest limits
What held up. On the 30/65 rule, Ultimate C% picks mostly different entry days from RSI(2) on SPY and ES. Only 28% of its SPY trades start on an RSI(2) entry day, against 54% for a near twin. Paired with RSI(2) on SPY, it lifted Ret/DD from 9.13 to 11.07 and cut the worst drawdown by 23%. The two never lost money in the same SPY year. Its 30/65 setting sits inside a stable block.
What did not. One undated launch test ranked it far above RSI(2). Over 20 years with next-open fills, RSI(2) has the higher Ret/DD on both markets. On ES, Ultimate C%'s edge over random entries is thin.
The limits:
- Two related markets, long only, daily bars. SPY and ES follow the same index. Nothing here covers intraday charts, other markets or short trades.
- No costs in the headline numbers. At the launch rule, costs cut net profit for both. RSI(2) still makes more on SPY and ES.
- No stop loss. The largest single loss was -$29,150 on one ES contract for Ultimate C% and -$20,450 for RSI(2).
- Ultimate C%'s own lookbacks were not changed. It ran at its published settings. Different lookbacks might trade differently.
- Correlated profits. Different days, but a daily correlation of 0.57 to 0.61. I did not test the pair during S&P 500 declines on their own.
- Wear, not crowding. The halves test shows whether RSI(2)'s profit factor dropped. It can't count how many traders use RSI(2).
- The launch test can't be checked line by line. The launch test gave no trade count, dates or fill rule. I matched its rule and settings, not every detail.
- The pair test does not prove why. The mixed SPY book scored higher while the two mostly entered on different days. I did not test which part of that did the work.
- A backtest is not a live edge. It is the best evidence about the past.
What this means for you
- Use Ultimate C% for what it is: a different oscillator. On the 30/65 rule on SPY and ES, it took many entries RSI(2) did not. On 277 of its 484 ES trades, RSI(2) had no signal.
- On SPY, it did best beside RSI(2), not instead of it. In the two-account test, swapping one RSI(2) account for it raised the book's Ret/DD from 9.13 to 11.07, for 6% less profit. On ES the gain was small, 4.47 against 4.37.
- Pick from the stable block. Buy below 25 to 40, sell above 65 to 70 passed on SPY and ES over these years. Don't chase the single best cell.
- Judge a second system by the whole book. On its own Ultimate C% scored 4.91 on SPY. In the pair, the book scored 11.07.
- Pair every win rate with a risk number. Ultimate C%'s 60.7% win rate on SPY came with a Ret/DD of 4.91. RSI(2)'s 68.8% came with 9.13.
- Get it free. Ultimate C% is a free download in the Bonus section of the StatOasis community, so you can run this test on your own markets.
Early on, I spent years hunting for the best indicator. It cost me. In this test, the SPY book with a second signal that mostly trades on different days had the higher Ret/DD and the smaller worst drop.
On SPY, the book that scored higher did not use a better indicator. It mixed two: one RSI(2) account plus one Ultimate C% account scored a Ret/DD of 11.07, against 9.13 for two RSI(2) accounts. On ES the gain was small, 4.47 against 4.37.
Want the next study when it is finished? That includes the ones that overturn something I believed. The newsletter is for that: StatOasis.com/Overfit.
Reference tables
The launch rule in full. Buy below 30, sell above 65 or after 5 bars. Next-open fills, no costs.
| Market | Indicator | Trades | Net profit | Average trade | Average hold | Largest loss |
|---|---|---|---|---|---|---|
| SPY | Ultimate C% | 758 | $81,148 | $107 | 4.3 bars | -$4,648 |
| SPY | RSI(2) | 909 | $92,746 | $102 | 3.2 bars | -$4,333 |
| SPY | RSI(3) | 611 | $74,457 | $122 | 4.0 bars | -$4,831 |
| ES | Ultimate C% | 484 | $138,375 | $286 | 4.3 bars | -$29,150 |
| ES | RSI(2) | 571 | $190,838 | $334 | 3.1 bars | -$20,450 |
| ES | RSI(3) | 385 | $125,362 | $326 | 4.0 bars | -$29,150 |
First half against second half. Same rule.
| Market | Period | Indicator | Trades | Win rate | Ret/DD | Profit factor |
|---|---|---|---|---|---|---|
| SPY | 1994 to 2010 | Ultimate C% | 375 | 60.3% | 2.86 | 1.54 |
| SPY | 1994 to 2010 | RSI(2) | 468 | 66.5% | 5.58 | 1.47 |
| SPY | 2010 to 2026 | Ultimate C% | 383 | 61.1% | 5.41 | 1.40 |
| SPY | 2010 to 2026 | RSI(2) | 441 | 71.2% | 4.22 | 1.48 |
| ES | 2006 to 2016 | Ultimate C% | 242 | 59.9% | 0.97 | 1.12 |
| ES | 2006 to 2016 | RSI(2) | 295 | 70.5% | 4.80 | 1.45 |
| ES | 2016 to 2026 | Ultimate C% | 241 | 64.3% | 3.52 | 1.41 |
| ES | 2016 to 2026 | RSI(2) | 276 | 73.6% | 3.03 | 1.42 |
Methodology and honesty footer: 594 backtests on SPY (1994 to 2026) and ES (2006 to 2026) daily bars, Execution rule v2 (signal at the close, fill at the next open). Flat-only, SPY $35,000 in whole shares, ES one contract, no compounding, frictionless headline figures with a costed view. Buy-and-hold and a seeded random control, matched on requested entries and hold, computed on the same bars. Every figure here comes from the study's own results tables. This is research, not investment advice.
