The Better-RSI Showdown: We Tested 4 RSI Upgrades on SPY, QQQ, IWM, and DIA

Methodology & risk note: Backtested event study across SPY, QQQ, IWM, and DIA daily price data, 1,856 strategy variants, frictionless. Results are hypothetical and not investment advice — past patterns don't guarantee future results. Full method and disclaimer below.

TL;DR

• Two variants actually beat plain RSI. Connors RSI and Z-Score RSI modestly outperform the plain-RSI baseline on median risk-adjusted return across all tested configurations. The edge is real, but it is not dramatic.
• Laguerre RSI underperforms. Despite its "lag-free" marketing, Laguerre RSI produced the worst median results of the four families: negative median CAR, the lowest share of profitable variants, and the largest median drawdown at 66.2%.
• The Triple RSI "~90% win rate" doesn't show up here. Actual win rates on our implementation: 56–65% across SPY, QQQ, IWM, and DIA depending on hold period. All configurations were profitable. The ~90% figure, attributed to QuantifiedStrategies, is not reproduced.
• High win rate doesn't mean profit. Five variants in this study had win rates above 60% and still lost money — all on the short side, where small wins absorbed one large directional loss.

CAR is compound annual return; MaxDD is maximum drawdown, the largest peak-to-trough loss. CAR/MaxDD combines them: a higher number means more return for the drawdown taken.

Is there a best RSI indicator?

Every trading YouTube channel runs the same playbook. An eye-catching title. A win-rate claim in the thumbnail. A "secret" indicator that fixes RSI's problems.

The titles alone make the case: "Stop Using RSI, Use This SECRET Indicator Instead" (StatOasis video #15), "Lag-Free Trading? The Hidden Power of Laguerre RSI" (#10), "Z-Score RSI: The Upgrade You Didn't Know You Needed" (#11). And for the boldest claim: "Triple RSI Trading Strategy: Boost Your Win Rate to 90%" from QuantifiedStrategies, covered in StatOasis video #3.

Trading decisions made on thumbnails have a poor track record. So we ran the backtests.

Four RSI variants, four major US index ETFs, 1,856 strategy configurations, all scored the same way. Two variants hold a real edge over plain RSI. One significantly underperforms. The headlining win-rate claim does not survive an honest test.

The plain-RSI baseline used here comes from our earlier RSI vs Stochastic vs Williams %R study, which tested plain RSI as part of a 33,792-variant oscillator comparison.

How we tested: 1,856 backtests across four ETFs

Here is exactly what sits behind every number below.

• Instruments: SPY, QQQ, IWM, and DIA — four major US index ETFs.
• Window: each ETF's full daily history through 2026. SPY runs from February 1993, DIA from January 1998, QQQ from March 1999, and IWM from June 2000.
• Families tested: plain RSI (1,152 variants), Connors RSI (288 variants), Laguerre RSI (128 variants), Z-Score RSI (288 variants) — 1,856 total. Triple RSI was tested separately as a rule-based signal mini-backtest.
• What we varied: indicator settings (lengths, thresholds), trade direction (long / short), and a protective time exit (0, 5, 10, or 15 bars — where 0 means the indicator decides the exit).
• Position rules: $10,000 fixed per position, no compounding, one position at a time.
• Fills: signals computed on the close and filled at the next open, so there's no look-ahead.
• Costs: frictionless. No commission, no slippage. The numbers are base rates — you would need to cover real execution costs on top of them.

Indicators defined:

Connors RSI combines three components in equal weight: RSI(3) of price, RSI(2) of the consecutive-day win/loss streak, and a 100-period percent rank of the 1-day rate of change. Developed by Larry Connors. Recommended thresholds are 90/10 rather than plain RSI's 70/30, because the composite moves faster. We use the canonical formula (rsi_len=3, streak_len=2, rank_len=100).

Laguerre RSI applies a 4-pole Laguerre digital filter, developed by John Ehlers in 2004, to price before computing the RSI ratio. A single parameter, gamma, controls the smoothing. Low gamma (e.g., 0.2) gives faster, more sensitive signals; high gamma (e.g., 0.7–0.8) gives slower, smoother ones. Our implementation uses gamma=0.5 and outputs 0–100. The canonical Ehlers formula outputs 0–1; the math is identical, the scale is multiplied by 100. Thresholds are 80/20.

Z-Score RSI (StatOasis formulation): first compute Wilder RSI(14), then re-express each RSI value as a rolling z-score over the past 20 RSI readings (using population standard deviation). A reading of +2 means RSI is 2 standard deviations above its recent average; −2 means 2 below. Output is unbounded and centered at 0. This is a specific formulation worth noting: some platforms use "Z-Score RSI" to mean RSI applied to z-scored prices, which is the reverse approach. The StatOasis version normalizes the RSI output itself, measuring whether RSI is at a historical extreme relative to its own recent behavior rather than whether raw price is statistically stretched.

Triple RSI fires a long-only entry when all three conditions hold at the close: (1) RSI has declined for 3 consecutive bars, (2) RSI is below 30, (3) price is above SMA(200). We tested RSI(2), our default, and RSI(5), the canonical QuantifiedStrategies version.

Finding #1: Which families were profitable?

The share of profitable configurations gives the first read on each family.

The spread across families is narrow. All four sit between 50% and 54% profitable. No family dominates by a large margin, but Laguerre RSI finishes last. Exactly half its variants lost money — below the plain-RSI rate — which means Laguerre's default parameter space does not reliably produce an edge across this universe.

Connors RSI and Z-Score RSI both land at 53.8%, edging out plain RSI's 52.8%. A small gap, and it shows up in the other metrics too.

Finding #2: Do the upgrades actually beat plain RSI?

Counting profitable variants tells you how many were positive, not how well they performed. Median CAR and median CAR/MaxDD answer that.

Connors RSI and Z-Score RSI beat the plain-RSI baseline on every measure: higher median CAR, better median CAR/MaxDD, higher median net profit. Z-Score RSI is the bigger mover, with a median CAR of 1.05% versus the baseline's −0.15%.

Laguerre RSI goes the other direction. A median CAR of −0.66%, a negative median CAR/MaxDD, and the lowest median net profit of any family at $245 — roughly 26% of plain RSI's $940. Switching from plain RSI to Laguerre RSI across the full parameter space tested here would have made results worse on every dimension.

The improvements from Connors RSI and Z-Score RSI are real. They are also modest. "Modest but real" is a more accurate description than "secret indicator" or "upgrade you didn't know you needed."

Finding #3: High win rate, negative returns — the trap

Win rate is probably the most misunderstood statistic in trading. A strategy can win 60% of its trades and still lose money if the losses on the remaining 40% are large enough to swamp the gains.

This study produced five examples of exactly that — win rates above 60%, negative net profit.

Four of the five are on the short side of RSI mean-reversion strategies on index ETFs. US equity index ETFs trend upward over time. A short mean-reversion strategy wins on small countertrend bounces — the market ticks up briefly, the short closes for a gain — then hits one sustained upward move that wipes out many small winners in a single trade.

The worst dollar loss in the study is not even in the 60%+ cluster. It belongs to Laguerre RSI on QQQ Short: a 55.1% win rate paired with a net profit of −$21,637. Win more than half your trades by a clear margin, still end up significantly negative. This is the win-rate trap in its starkest form.

Win rate alone tells you nothing useful. The ratio of average win to average loss — and whether you are trading with or against the structural drift of the instrument — determines the outcome.

The mirror-image trap: perfect win rates on tiny samples

The same data holds the opposite illusion, and it is just as dangerous. The highest win rates in the entire 1,856-variant sweep look spectacular on paper: a plain-RSI configuration on QQQ Long with a 100% win rate, and one on DIA Long at 94.1%. Sort the sweep by win rate and the top of the leaderboard reads like a holy grail.

Then read the trade counts. The 100% variant won 5 trades out of 5 — total profit $1,900. The 94.1% variant won 16 of 17. Every single one of the ten highest-win-rate variants in the study sits below 50 trades, which is exactly why the engine flags anything under that threshold as statistically unreliable rather than celebrating it. Five coin flips landing heads is not a biased coin.

So the win-rate statistic fails in both directions: a 60% win rate can lose $6,500 (the short-side trap above), and a 100% win rate can be five lucky trades. The two numbers that rescue it are the ones this study leans on everywhere — the trade count, and the win/loss size ratio behind the rate.

Finding #4: Which family draws down the least?

Risk matters as much as return. A strategy with a 90% drawdown is not tradeable for most people regardless of its return figure.

Connors RSI and Z-Score RSI carry lower median drawdowns than plain RSI. At 35.92% and 38.27% against plain RSI's 40.65%, the drawdown numbers line up with the return picture from Finding #2. The same two families win on both dimensions.

Laguerre RSI's 66.22% median drawdown is in a different category. Most traders cannot hold through a two-thirds peak-to-trough loss in a real account. The best individual Laguerre RSI variant — QQQ Long, thresholds 20/80, 5-bar exit — had a 20.86% drawdown on 355 trades and net profit of $28,961. But that is the top of the distribution, not the typical result. At the median, Laguerre RSI draws down roughly 66%.

Finding #5: Does Triple RSI really achieve a 90% win rate?

This is the claim that drew the most scrutiny. QuantifiedStrategies describes Triple RSI as achieving a "90–91% win rate" on SPY since 1993. StatOasis video #3 covers the strategy and common implementation pitfalls.

We tested two versions: RSI(2), our implementation, and RSI(5), the canonical QuantifiedStrategies version. Both used the 3-condition rule described in "How we tested" above. Neither included the "RSI below its value 63 bars ago" volatility filter that appears in the original canonical version.

RSI(2) on SPY — 5-bar hold: 445 trades, 59.1% win rate, $13,844 net profit. RSI(2) on SPY — 10-bar hold: 332 trades, 65.1% win rate, $22,791 net profit. RSI(5) on SPY — 5-bar hold: 185 trades, 58.9% win rate, $7,546 net profit. RSI(5) on SPY — 10-bar hold: 155 trades, 64.5% win rate, $11,541 net profit.

Neither version reproduces a 90% win rate. Win rates of 59–65% on SPY are real and positive — every SPY configuration produced a profit — but 65% and 90% are not close.

Two factors likely explain the gap. First, the canonical version includes a "RSI below its 63-bar-ago value" condition that filters out lower-quality signals during certain market regimes. Removing that filter lets through trades the original version would skip, which pulls the win rate down. Second, the canonical study reports 78 trades on SPY since 1993; our RSI(5) version generates 185 trades over the same period. More signals under the same 3-condition rule means the signal is firing at lower-conviction moments. A tighter filter that takes fewer trades can achieve a higher win rate precisely because it discards the borderline cases.

All tested configurations were profitable. The ~90% win rate is not what a straightforward implementation of the publicly available rules produces.

Across all four ETFs at the 5-bar hold, the RSI(2) aggregate was 1,482 trades at a 56.3% weighted win rate and $36,235 combined net profit. At the 10-bar hold, 1,099 trades at 60.6% and $55,851. Every configuration was profitable. None reached 90%.

The head-to-head: best variant per family

Here is the best reliable variant (minimum 50 trades) from each family, ranked by CAR/MaxDD.

At their best, all four families find profitable configurations. Connors RSI and Z-Score RSI edge out plain RSI at the top. Laguerre RSI's best variant trails considerably on risk-adjusted performance even with the highest trade count in the group.

Treat these figures as upper-bound illustrations, not representative expectations. The median results in Findings #2 and #4 are a more honest picture of what a typical parameter choice delivers.

The verdict, and the honest caveats

Two of the four tested RSI upgrades outperform plain RSI in this 1,856-variant test. Connors RSI and Z-Score RSI beat the plain-RSI baseline on median CAR, median CAR/MaxDD, and share of profitable variants. The advantage holds across all three measures but is not large on any of them.

Laguerre RSI underperforms plain RSI on every measured dimension: negative median CAR (−0.66% vs −0.15% for plain RSI), negative median CAR/MaxDD, the lowest profitable-variant rate (50.0%), and the largest drawdowns. The "lag-free" framing is a marketing claim; the backtested performance here does not support it.

The Triple RSI "~90% win rate" claim, attributed to QuantifiedStrategies, does not hold under our implementation. Actual results run 59–65% on SPY and 56–61% on the four-ETF aggregate. Those are positive, profitable win rates. They are not 90%.

Now the limits, because a study that hides them isn't worth trusting.

• Frictionless. Every number was computed with no commission and no slippage. Real costs would shrink any edge, and the thin ones might not survive. Read these magnitudes as relative, not as take-home returns.
• An event study, not a deployable system. This measures base rates across configurations. It tells you where edges concentrate, not how to build a risk-managed strategy. That is separate work.
• The Triple RSI comparison is not exact. The canonical QuantifiedStrategies version includes a "RSI below its 63-bar-ago value" filter that our implementation does not have. Results are not directly comparable to the original claim.
• Best-variant numbers are cherry-picked optima. The top strategies in each family are the top of their distribution. The median results are the more honest benchmark.
• Laguerre RSI sweeps only gamma=0.5. Different gamma values would produce different results. This is a single-parameter-value test for this family.

Findings at a glance

Key takeaways

• Two upgrades hold up, one doesn't. Connors RSI and Z-Score RSI modestly outperform plain RSI on median risk-adjusted return across 1,856 variants. Laguerre RSI underperforms on every tested dimension.
• The win-rate trap is real on the short side. Strategies with 60%+ win rates that end up losing money are not a hypothetical — five exist in this data, all tied to short mean-reversion strategies on upward-trending index ETFs.
• The canonical Triple RSI win rate is not reproduced. Win rates of 59–65% on SPY are genuine and the strategy is profitable. The ~90% figure requires conditions beyond our implementation, specifically the 63-bar-ago RSI filter.
• Laguerre RSI carries the largest drawdowns in the study. A median MaxDD of 66.22% makes the typical configuration difficult to hold in a real account regardless of win rate.
• Best-variant performance is real but not typical. The top Connors RSI variant hit a CAR/MaxDD of 1.250. The median Connors RSI variant hit 0.010. The gap between "best" and "typical" is wide in every family.

Methodology: how this data was generated

This is a backtested event study, not live trading results. The numbers come from our in-house research engine run over the full daily price history of each ETF.

• Data source: Daily OHLCV (open, high, low, close, volume) price data for SPY, QQQ, IWM, and DIA.
• Date range: SPY from February 1993, DIA from January 1998, QQQ from March 1999, IWM from June 2000, all through June 2026.
• Variants: 1,856 total across four RSI families (RSI 1,152, Connors RSI 288, Z-Score RSI 288, Laguerre RSI 128), sweeping indicator settings, direction (long / short), and time exit (0 / 5 / 10 / 15 bars). Triple RSI tested separately as a rule-based mini-backtest.
• Entry and exit: signals computed on the close and filled at the next open (no look-ahead). One position at a time, $10,000 fixed per position, no compounding.
• Reliability threshold: 50 trades minimum. Variants below this are flagged, not dropped.
• Scoring: each variant measured for percent profitable, median and best net profit, win rate, CAR, MaxDDPct, and CAR/MaxDD.
• Frictionless assumption: computed without commissions or slippage. Real-world costs would reduce any edge shown here.

Disclaimer

All results in this article are derived from historical backtesting using daily ETF price data and do not represent actual trading results. Backtested performance is hypothetical. Past performance of any pattern does not guarantee future results. This article is for educational and informational purposes only and does not constitute investment advice. StatOasis is not a registered investment advisor. Nothing here is a recommendation to buy or sell any security. Please consult a licensed financial professional before making any investment decision.

Transparency: StatOasis sells trading-education products. Our research is produced independently and is not altered to favor a sale.

Freshness: The data is current through June 2026. We re-review these studies when the underlying datasets are extended.

Get the next study in your inbox

This is one study in an ongoing series. If you want the next one — the same kind of head-to-head, myth-busting, frictionless-but-honest backtest — join The Overfit newsletter at StatOasis.com/Overfit. It's where we publish the data behind the trading ideas everyone argues about.

Which RSI you pick matters less than which win rate you refuse to trust.

For the developer — paste-ready: the block below is valid JSON-LD (one @graph). Drop it as-is into a single <script type="application/ld+json"> tag in the page <head>. Before publishing, confirm the URL / @id paths resolve (author page /about/ali-casey, logo.png, the /terms license) and update datePublished / dateModified if the live date differs.