Larry Connors R3 Strategy: 8,067 Backtests on SPY and Index Futures

TL;DR: the answer box

• R3 still works, rarely. As published: 70 trades on SPY, 71 on ES and 81 on NQ in 25.63 years. Win rates 70.0%, 80.3% and 81.5%, with Ret/DD 2.24, 2.25 and 2.60.
• ES is not a better market for R3 than SPY, just a bigger contract. The ES average trade is $853 against SPY's $170. As a share of the position it is 0.50% against 0.49%.
• Loosening R3 buys more trades, and rarely a better average trade. Only 15.2% of relaxed ES versions beat R3's average trade. A relaxed version traded 373 times on ES instead of 71 and lifted Ret/DD from 2.25 to 8.91. Its average trade fell 43%, and its win rate fell from 80.3% to 74.8%.
• "Every version makes money" is not proof. All 2,688 relaxed ES versions made money. So did random entries. The real test is beating random entries by a wide margin: 98.9% of comparable ES versions did, and 82.9% on NQ.
• No stop, and two large losses. The two largest ES losses, -$15,125 and -$15,100, came in 2025 and 2026. Each one cost about 18 average trades.

What are the R3 rules?

R3 comes from Larry Connors and Cesar Alvarez's book High Probability ETF Trading (2009). It buys short, sharp pullbacks in a market that is still in an uptrend.

It is built on RSI(2). RSI, Welles Wilder's Relative Strength Index, is a score from 0 to 100 that compares recent up moves with recent down moves. The (2) means it looks back just two days, so it swings hard. Below 10 is the oversold line R3 waits for. Above 70 is the exit line.

R3 buys when four things are true at the close:

1. The close is above its 200-day average (the uptrend check).
2. RSI(2) has fallen three days in a row.
3. The first of those falls started below 60.
4. RSI(2) closes below 10 today.

It sells when RSI(2) closes above 70. There is no stop loss. Here is the rule on one real signal.

And here is the trade that signal produced, bar by bar. It is the ES trade whose profit sits closest to the middle of all 71. So it shows an ordinary R3 trade, not the best one.

How I tested it

Three markets, one window. SPY, ES and NQ, daily bars from 2001-01-02 to 2026-08-20. That is 25.63 years. It holds the 2001 to 2002 bear market, 2008, 2020 and 2022.

The fills: Execution rule v2. The signal is read at the close. The trade buys the next open. The sell is also placed at a close and fills at the next open. One position at a time. Connors' book trades at the close. This test reads the signal at the close and fills at the next open instead.

Sizing. SPY gets a $35,000 account in whole shares, with no compounding. ES and NQ trade one contract each.

Costs. The headline numbers have no costs. With commission and slippage added, the ES average trade drops from $853 to $824. Costs do not decide the result on ES or SPY. NQ has no cost check.

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 worst drop. Think of it as profit per dollar of pain. Higher is better.
• Profit factor: dollars won for each dollar lost.

The sweep. Then I loosened R3's buy level, falling-bar count, trend filter, exit and time cap, the way traders usually do. The RSI buy level ran from below 5 to below 40. "At least X of the last Y RSI bars falling" came in 14 settings. The 200-day filter went on and off. The sell moved to RSI above 50, 60, 70 or 80, with a time exit of 5 bars, 10 bars or none. That makes 2,688 versions per market, 8,064 in all. Add the published rules on each market: 8,067 backtests.

Does the Larry Connors R3 strategy still work?

Yes, on all three markets. It just does not trade often.

On ES, R3 took 71 trades in 25.63 years. That is about 2.8 a year. It won 80.3% of them, with a Ret/DD of 2.25 and a profit factor of 2.23. Its worst drawdown was $26,925.

A high win rate on 71 trades is a small sample. And R3 spends most of its life waiting.

You wait weeks, sometimes months, for four checks to line up. Then you hold for a median of 4 bars.

Did it still work after the book came out?

This is the test every published strategy has to pass. R3 passed it on ES and NQ. On SPY it wore thin.

Before the book, from 2001 to 2008, R3 fired only 11 to 20 times per market. A profit factor of 29.91 on 13 SPY trades is too few trades to judge. So skip those rows.

After the book, from 2009 to 2026, the profit factor was 2.23 on ES and 3.07 on NQ. SPY slid to 1.78. Since 2021 it fell to 1.21 on SPY, on just 17 trades. Seventeen trades cannot prove the SPY edge is gone. They do show it has been thin for five years.

The cost of having no stop

R3 holds until RSI(2) bounces above 70, however long that takes. The median ES hold was 4 bars. Twice recently it ran far longer.

• ES, bought 2025-02-26, sold 2025-03-18 after 14 bars: -$15,125.
• ES, bought 2026-03-04, sold 2026-04-01 after 20 bars: -$15,100.

The same February 2025 entry cost -$27,920 on NQ. An 80.3% win rate with no stop means losses are rare. The two largest on ES each cost about 18 average trades.

Is R3 better on index futures than on SPY?

Not on the S&P 500. In this test, ES and SPY show near-equal Ret/DD; ES makes more dollars because the contract is bigger.

The usual reason to move R3 to futures is leverage. Leverage is real, but it is sizing, not edge. One E-mini S&P 500 contract is $50 times the index. That is far more than a $35,000 SPY account controls. So the ES average trade is 5.0x SPY's in dollars.

Take the size out and the two match. Ret/DD is 2.24 on SPY and 2.25 on ES, a gap of 0.01. NQ does a little better, at 2.60, on a different index. Trade futures for the contract size and costs you want. Don't trade them expecting a better R3: in this test, its Ret/DD on ES matched SPY's.

What happens when you relax the R3 rules?

The median relaxed ES version trades 298 times, against R3's 71. And on ES, most relaxed versions have a smaller average trade than R3's. In the version I picked, the ES average trade fell from $853 to $486. In the version I picked, the total still grew: 3.0x the profit on 5.3x the trades.

Many traders believe you can loosen R3 to trade far more often "without hurting performance." Half of that is true. On Ret/DD, the relaxed version wins on every market.

Now look at what each trade earns. Here is every one of the 2,688 relaxed ES versions.

On ES, 95.1% of relaxed versions beat R3's Ret/DD. Only 15.2% beat its average trade. Loosening the rules rarely raised the average trade. It found more of them.

Which relaxed version did I pick, and why not the best one?

The single best ES cell scored a Ret/DD of 10.44 on 402 trades. I did not use it. It is the top pick of a 2,688-entry search, chosen on the same data it is scored on. Early on I lost roughly $270,000 over years, partly by buying overfit systems that looked best on paper. So the top cell gets set aside.

Instead I looked for a plateau. Think of the results as a map. On a plateau, nudge a setting and the score barely moves. On a spike, one small nudge drops you off a cliff. A spike's score rests on one exact setting in this one data set. That is the curve-fitting risk I set aside.

I grouped the versions into cells by buy level, sell level, filter and time exit. I kept only cells in the top quarter whose neighbours one RSI level either side were also in the top quarter. Twenty-seven cells qualified. Inside the top one, I took the middle version, not the best. Here it is:

RSI(2) below 25, at least 1 of the last 3 RSI bars falling, close above the 200-day average. Sell at the next open after RSI(2) closes above 70, or 10 bars after entry.

That is a wide, high-scoring shelf. And the same shelf shows up on all three markets.

SPY and ES peak at RSI(2) below 25. NQ peaks at below 20. That I picked the cell on ES and it holds on the other two markets is a good sign. But it was picked on the whole 2001 to 2026 window, so it is not a fresh test.

Does the three-day fall rule matter?

Mostly it removes trades. Once RSI(2) is below 25, at least one falling bar is almost always already true. So "1 of the last 3 falling" gives exactly the same 373 trades as "1 of the last 2, 4 or 5."

Demand all 3 of 3 falling, as R3 does, and you drop to 177 trades at a Ret/DD of 5.42.

The relaxed version against R3, side by side

Here is the catch with that Ret/DD jump. Ret/DD rises with trade count over a fixed window. Profit adds up trade by trade, but the worst drop does not. So 8.91 against 2.25 is not four times the edge.

Look at profit per day in the market instead. On ES, per percentage point of time in a trade, R3 earned $12,307 and the relaxed version $7,277. The relaxed version earns less per day it holds a trade. It earns more in total because it holds trades about five times as often.

That is the trade-off in one line. You give up 43% of the average trade and drop from an 80.3% to a 74.8% win rate. You get 5.3x the trades, 3.0x the profit and a smaller worst drop: $20,362 against $26,925. Costs bite the smaller trade harder: with costs, the relaxed ES average trade is $457 instead of $486.

The relaxed version also held up over time. From 2001 to 2008 it took 84 to 87 trades per market, enough to read. It was weak then, with a profit factor of 1.37 on SPY, 1.57 on ES and 1.21 on NQ. Since 2009 it has been stronger.

Is the edge real, or just the market going up?

Over 2001 to 2026, most comparable relaxed versions beat random entries on all three markets. On SPY and ES, almost all of them did.

Here is why that test matters. All 2,688 relaxed ES versions made money. That sounds like proof. But stock indexes rose over these years. On SPY, ES and NQ from 2001 to 2026, even a random long-only entry held a few days made money.

So I built a random rule. It buys at random, long only, and asks for as many entries as the typical relaxed version (307 on ES). Holding one position at a time, it took 268 trades, each held 3 bars like the typical version. It sells at the next open, like the study. I ran it with ten random seeds. On ES it made $43,296 on average.

The bar to clear is that average plus two of its spreads (spread: how much the ten random runs varied). I compared it only with versions that trade about as often as the random rule.

On SPY, 707 of 714 versions cleared the bar. On ES, 702 of 710. On NQ it is most of them, not nearly all: 614 of 741.

For context, buy and hold made $167,109 on SPY over the same window, with a worst fall of 56.45% in March 2009. The relaxed SPY version made $47,879 with a worst drop of $5,166 on a $35,000 account, while in a trade 25.4% of the time. Less money, a quarter of the exposure. That is the honest shape of this system.

The verdict, and the honest limits

R3 made money on all three markets from 2001 to 2026 with next-open fills. It just traded only 70 to 81 times per market in 25.63 years.

Where the hype is right. The published rules still make money on SPY, ES and NQ, and in the 2009 to 2026 stretch on its own. Loosening them does multiply the trades, 5.3x on ES. The version I picked comes from the top-ranked of 27 ES cells that pass the stable-region rule, and on SPY and ES nearly all comparable loosened versions beat random entries.

Where it is wrong. ES does not hold a better R3 than SPY. It holds a bigger contract. And loosening R3 does hurt one thing: the average trade. In the version I picked it fell 43% on ES, and only 15.2% of relaxed ES versions beat R3's. The higher Ret/DD came with 5.3x the trades and a smaller average trade, not a bigger one.

The limits, in plain words.

• A small sample. R3 as published has 70 to 81 trades per market, and only 11 to 20 before 2009.
• No stop. Both of the largest ES losses are recent. I did not test a stop.
• Next-open fills. Connors trades at the close. This test does not measure that difference.
• One contract on futures. It controls far more index than a $35,000 SPY account. Read the SPY rows for the risk of the idea.
• Three related US stock indexes, long only. Nothing here covers commodities, currencies or single stocks.
• The relaxed version was picked on ES, over the whole window, then applied to SPY and NQ unchanged.

What this means for you

1. Size R3 for its losses, not its win rate. An 80.3% win rate on ES came with a $26,925 worst drawdown.
2. Don't move from SPY to ES expecting a better system. Move for contract size and cost.
3. If you want more trades, move the RSI level. On ES, with the 200-day filter on and a 10-bar cap, every buy level from RSI(2) below 15 to below 40 that has enough trades scores a median Ret/DD of 5.43 to 8.91.
4. Judge a loosened version by profit per day in the market. Ret/DD flatters any version that simply trades more.
5. Check any version against random entries and a fresh stretch of data before you trade it. That is what AlgoChef is for: import the trade list from your platform, then run a Monte Carlo reshuffle and an in-sample versus out-of-sample split before a dollar goes in.

R3 still made money on SPY, ES and NQ through 2026 with next-open fills. It is under-sampled. The loosened version took 373 ES trades. Stay on the plateau, and size for the loss you have not had yet.

Want the next study when it is finished, including the ones that overturn something I believed? That is what the newsletter is for: StatOasis.com/Overfit

Reference tables

For readers who want the detail behind the charts.

Table 1. R3 as published against the relaxed version, 2001 to 2026

Execution rule v2, next-open fills, no costs. SPY on a $35,000 account in whole shares, ES and NQ one contract.

Table 2. All 2,688 relaxed versions per market

Table 3. Relaxed versions against random entries

Methodology and honesty footer: 8,067 backtests on SPY, ES and NQ daily bars, 2001 to 2026, Execution rule v2 (signal at the close, fill at the next open). Flat-only, SPY $35,000 in whole shares, futures one contract, no compounding, frictionless headline figures with a costed view for SPY and ES. Buy and hold and a seeded, frequency-matched random control computed on the same bars. Every figure here comes from the study's own results tables. This is research, not investment advice.