Range Expansion Index (REI) vs RSI(2): 3,321 Trades on SPY, QQQ, DIA and ES

TL;DR: the answer box

• "REI" is not one rule. On SPY, REI(8) below -60 scored a Ret/DD of 0.06. REI(5) below -45 scored 5.35. Same name, same exit.
• RSI(2) below 10 beat both published REI settings on all four markets. Ret/DD 6.03 on SPY, 4.06 on QQQ, 3.51 on DIA, 4.82 on ES.
• Almost every REI setting made money. 224 of 224 on SPY, 223 after costs. That proves little: random entries made money too.
• Fewer REI settings beat luck. On SPY, 63.9% of reliable fixed-hold REI settings beat matched random entries by two spreads or more. For RSI(2), 94.4% did.
• Different days, same bet. Only 27.1% of REI(5)'s SPY signal days were RSI(2) days. Yet their daily profits moved together, a correlation of 0.63.
• The steady zone is a short sum. REI 2 to 5, bought below -20 to -50: a median Ret/DD of 4.91 on SPY, with 80 of 80 settings profitable after costs.

How I tested the Range Expansion Index

The Range Expansion Index is Tom DeMark's oscillator from 1994. An oscillator is a gauge that swings between two limits. RSI(2), the one I compare it with, looks only at closing prices. REI also looks at highs and lows.

Here is the idea in plain words. Each day, REI takes the change in the high over two days and adds the change in the low over two days. It keeps that move only if today's range overlaps prices from 5 to 8 days back. Then it adds up the kept moves and divides by all the movement. You get a reading from -100 to +100. A deep negative reading means the recent action was mostly down.

The trouble: the published versions disagree on three things.

• The window. DeMark's 1994 version, as Wikipedia sums it up, adds up 8 days and draws lines at +60 and -60. His 1999 options book, as ProRealCode codes it, adds up 5 days with lines at +45 and -45.
• The filter. MetaStock and Infront use one overlap test (method 1). Wikipedia and MQL5 use another (method 2).
• The level. Plus or minus 45, or 60.

So I named mine first. Method 1, checked against TradeStation on 7,269 bars, with 0 bars off by more than 0.0001.

The rules. Long only. When REI closes below a level, buy the next day's open. Sell at the open 5 bars after entry. Both published windows ran as named settings, fixed before I saw any result. RSI(2) below 10 ran under the same rules. RSI(2) is the 2-day Relative Strength Index, the classic short-term dip gauge.

The sweep. Then I tested every nearby setting: REI lengths 2 to 8, buy levels -20 to -80, and four exits. That is 224 REI settings per market and 1,000 backtests in all.

The markets. SPY from 1998, QQQ from 1999, DIA from 1998 and ES (E-mini S&P 500 futures, one contract) from 2006, all to 2026-08-20. ETFs trade a $35,000 account in whole shares, with no compounding.

The fills: Execution rule v2. The signal is read at the close. The trade buys the next open and sells at an open. You can't buy at a close you haven't seen yet.

The scores. A few plain measures carry this article:

• Drawdown: the biggest drop from a high point.
• Ret/DD: net profit divided by the biggest dollar drawdown. Profit per dollar of pain. Higher is better.
• Win rate: the share of trades that made money.
• Median: the middle value, or the "typical" setting.
• Reliable: a setting with 100 or more trades.

Here is one trade. It is the REI(5) trade whose profit sits closest to the median of all 430 on SPY.

REI(5) closed at -46.9 on 2015-03-27. The trade bought the open of 2015-03-30 at 206.98. It sold the open of 2015-04-07 at 207.85. $148. A typical trade, not the best one.

Which REI are you actually using?

On SPY, that one choice moved Ret/DD from 0.06 to 5.35.

Start with how often each version fires. Same chart, same dips, two settings:

Over the full test, REI(5) below -45 fired on 1,316 SPY closes. REI(8) below -60 fired on 262. The shorter, shallower setting fired 5.0 times as often.

Now the results:

The 8-day classic barely broke even. On SPY it made $680 on 116 trades, with a 50.9% win rate and a $10,602 drawdown. After costs, -$86. It ranked 210 of 210 reliable REI settings. Dead last.

The 5-day version did well. On SPY it made $69,164 on 430 trades, a 57.7% win rate, a $12,919 drawdown and a Ret/DD of 5.35. It traded 3.1 to 3.7 times as often as the 8-day version across the four markets.

The filter matters too. Switch REI(5) on QQQ from method 1 to method 2, and its Ret/DD falls from 3.14 to 1.24.

That is four rules under one three-letter name. A backtest of "REI" tells you nothing until it names the window, the level and the filter.

Does REI beat RSI(2) for mean reversion?

Not as published. Mean reversion is the bet that a market that just dropped tends to bounce. Under the same rules, RSI(2) below 10 beat both published REI settings on 4 of 4 markets.

On SPY, RSI(2) made $72,636 on 358 trades with a $12,036 drawdown. Ret/DD 6.03. REI(5) made $69,164 with a bigger drawdown, $12,919.

On ES the gap was widest. RSI(2) scored 4.82, against 2.29 for REI(5) and 1.12 for REI(8).

The whole sweep says the same. The typical reliable REI setting on SPY scored a Ret/DD of 4.08. The typical RSI(2) setting scored 5.89. Only 31 of 210 REI settings beat that RSI(2) median.

The best single REI cell on SPY scored 8.98. The best RSI(2) cell scored 9.63. Don't chase either one. A single best cell is an in-sample pick, so I don't promote it.

Does REI beat a random entry?

This is the claim that fools people. "Almost all settings were profitable" sounds like proof. Early on I lost roughly $270K to overfitting and leverage.

Here is the test. Take random buy days. Use the same number of trades and the same hold as the setting. Sell at the open the same way. Repeat it with 10 different random draws, called seeds. Then ask: how far above the random average did the real setting land? I measure that in spreads. A spread is how much the random results moved from seed to seed. Two spreads above the random average is the bar this study uses.

I ran that test on every SPY setting with a fixed hold and 100 or more trades.

101 of 158 REI settings cleared the bar on SPY. 17 of 18 RSI(2) settings did. On ES the gap held: 34.9% of REI settings against 61.1% for RSI(2).

Now the three named settings, market by market:

On SPY, random entries matched to REI(5)'s 430 trades made $24,280, with a spread of $15,185. REI(5) made $69,164, 3.0 spreads above. RSI(2) sat 4.2 spreads above its own control. REI(8) finished 1.2 spreads below its control's average.

And the yardstick every dip-buying article should print: buy and hold SPY made $238,838 on the same $35,000, with a worst drop of 56.41%. REI(5) made $69,164 and sat in a trade only 29.9% of the time.

Does REI give different signals from RSI?

Different days, yes. Different money, no.

On SPY, REI(5) fired on 1,316 closes and RSI(2) on 723. They agreed on 357. So most REI signal days are not RSI(2) days. Across every REI length and level, the overlap ran from 20.0% to 46.2%.

But look at the money. Correlation measures how closely two things move together: 1 is lockstep, 0 is no link. The daily profit correlation between REI(5) and RSI(2) was 0.63 on SPY and 0.69 on ES. Both rules buy a short dip in the same index and hold it for about a week. On SPY and ES, their daily profits moved together.

Year by year the link was looser, a correlation of 0.37 on SPY. REI(5) made money in 23 of 28 years and RSI(2) in 24.

Does running REI and RSI(2) together improve Ret/DD?

Sometimes. Not as a rule.

I ran each REI setting beside RSI(2), each on its own account, and added the daily profits. The pair beat the better of the two alone in 4 of 8 cases. It beat it by more than 5% in only 2: QQQ with REI(8), 26% higher, and DIA with REI(8), 19% higher. On ES, REI(5) plus RSI(2) scored 3.45, below RSI(2) alone at 4.82.

Both clear wins came with REI(8), the setting least tied to RSI(2). Before you run two oscillators side by side, check how their returns move together, not how their signals look. Put both trade lists into AlgoChef and compare them before real money goes on the pair.

What are the best REI settings?

A short sum, not a magic number. Picture every setting as a cell in a table: REI length down the side, buy level across the top.

The zone: REI 2 to 5, bought below -20 to -50, all four exits. That is 80 settings per market. On SPY their median Ret/DD was 4.91, the lowest was 2.09, and 80 of 80 made money after costs. On QQQ the median was 3.30, on DIA 3.83 and on ES 3.37.

Think of it as a broad profitable region, not a single peak. On SPY all 80 settings made money, and even the weakest scored 2.09. The scores still varied a lot across the region, up to 8.98. None of them lost money.

Longer sums trade less. On SPY with a 5-bar hold, the median trade count fell from 672 at REI(2) to 222 at REI(8). REI(8) below -80 traded 20 times in 28 years. A rule that rare can't tell you much.

I picked this zone after seeing the results. That makes it in sample: it was chosen on the same data it is scored on. So I checked it two ways. First, split each market's history in half.

On SPY, all 80 zone settings made money in each half. REI(8) at -60 did not hold up: Ret/DD 0.40 in the first half, -0.31 in the second. The second check is other markets.

Does REI work on other markets without tuning?

Each rule led somewhere. None led everywhere.

I ran the three named settings, untouched, on eight more markets: IWM, NQ and YM (stock indexes), ZN (10-year Treasury note), GC (gold), RB (gasoline), ZW (wheat) and 6E (Euro FX).

• REI(8) below -60 made money on 8 of 8. The rule that came last on SPY was the steadiest one away from it, with the top score on 6E (3.01), gold and gasoline.
• REI(5) below -45 made money on 7 of 8. It led on IWM and NQ: Ret/DD 6.44 on IWM and 5.49 on NQ. It lost on wheat.
• RSI(2) below 10 made money on 6 of 8. It led on YM, ZN and wheat, and lost on Euro FX and gasoline.

So the rules mostly made money off their home markets. But the winner changed from market to market. These figures are before costs, one contract each, so compare Ret/DD, not dollars.

The honest limits

• Long only. The short side was not tested.
• The zone is in sample. The halves check all 80 zone settings. The eight extra markets check only one of them, REI(5) below -45.
• No stop loss. The largest single loss on one ES contract with REI(5) was -$33,288.
• Costs trim every number. The headline figures leave out costs. With costs, 223 of 224 SPY REI settings still made money. The eight-market check is before costs.
• Method 1 for the sweep. Method 2 ran only for the two named settings, and it moved Ret/DD on all four markets.
• A backtest is not a live edge. It is the best evidence about the past.

What this means for you

1. Name your REI before you test it. Window, level and filter. On SPY the 8-day classic at -60 made $680 in 28 years. The 5-day version at -45 made $69,164.
2. Use a short sum. REI 2 to 5, bought below -20 to -50, was the steady zone here: 80 of 80 settings profitable after costs on SPY. It is in sample, so test it on your own market first.
3. Benchmark against RSI(2) first. Under the same rules, RSI(2) below 10 beat both published REI settings on all four markets.
4. Check the correlation, not the chart. REI(5) and RSI(2) fired on mostly different days and still moved 0.63 together on SPY.
5. Judge a zone, not a cell, and test it against random entries with the same trade count before you trust it.

REI(5) below -45 made money on all four markets and beat random entries on three. It is not a hidden replacement for RSI(2). Testing always trumps theory.

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

Table 1. The three named settings, buy the next open, sell 5 bars later. Ret/DD is net profit divided by the biggest dollar drawdown. Profit factor is dollars won per dollar lost.

Table 2. The full sweep and the steady zone, per market.

Methodology and honesty footer: 1,000 backtests on SPY (1998 to 2026), QQQ (1999 to 2026), DIA (1998 to 2026) and ES (2006 to 2026) daily bars, plus the three named settings untuned on eight more markets. REI is method 1 (DeMark as MetaStock and Infront publish it), matched to TradeStation bar for bar. Execution rule v2 (signal at the close, fill at the next open, time exit at the open 3, 5 or 10 bars after entry). Long only, flat-only per setting, ETFs $35,000 in whole shares, futures one contract, no compounding. Frictionless headline figures with a costed figure on every headline row. Buy-and-hold and seeded, frequency-matched random controls computed on the same bars. Every figure here comes from the study's own results tables. This is research, not investment advice.