Backtest · S&P 500 ETFs and index futures · 1993 → 2026

Does Averaging Down Work? Buying the Dip on S&P 500 ETFs and Index Futures, Tested

Buy one more each time the price falls, sell everything once the position is up, start again. Every buy lowers the average cost, so a small bounce is enough to close at a profit — and it almost always does. We tested whether that profit is worth the cash it ties up: on S&P 500 funds back to 1993, on Nasdaq-100 and S&P 500 futures over two years, inside a single trading day, and as a “time average” that buys on a clock.

Published October 14, 2026 · Stable Income Strategy research

The short answer

Averaging down — buying one more slice each time the price falls X% and selling everything when the position is up T% — almost always closes at a profit: on S&P 500 funds since 1993 every cycle closed, even on a 3× fund through 2000–2009. The catch is the cash, which you learn only after the worst fall: up to $315,000 per $1,000 slice, earning 1.3–6% a year on that money, and holding the fund sized to the same worst fall earned more in almost every setting (6.4% against 2.5% in that case). On Nasdaq-100 and S&P 500 futures it returned about what holding did, and in a test of profit targets from +0.5% to +3% every target did worse than simply holding to 3 PM. The intraday versions that made money in 2024–26 all lost on SPY minute data from 2016 to 2024.

most cash tied up per $1,000 slice (3× S&P 500 fund, 1993 →): $315,000
a year on that cash, against holding at the same fall: 2.5% vs 6.4%
of days a +0.5% target was hit — and it still lost money: 66%
years up for the guarded intraday finalists on SPY, 2016–2024: 2–6 of 9

Key takeaways

Cycles close: with unlimited buying, every setting on SPY, SSO, UPRO, SPXL and a 3× S&P 500 fund since 1993 made money, and every cycle closed — the longest took 3.4 years (August 2000 → January 2004).
The cost is the cash: up to $315,000 tied up per $1,000 slice, known only after the worst fall. On that money it earned 1.3–6% a year.
Holding a fixed share of the fund, sized to fall no further, beat it in almost every setting: 7.3% against 6.0% a year on UPRO (2% steps, +2%), 5.7% against 2.5% on the 3× fund (1% steps, +5%).
With a fixed budget the slices run out and it turns into plain holding: falls of −73% to −98% on the 3× funds, and one cycle stayed open from March 2000 to 2017.
On NQ and ES futures (2024–26) every step size made money, but 1% steps on NQ needed 27 contracts and $2.6 million; per dollar needed it returned about what holding did.
Profit targets cut winners: buying MNQ on a schedule and selling at +0.5% averaged −$8.5k over two years, against +$5.3k for holding the same buys to 3 PM.
The intraday versions were luck of 2024–26: guarded MES rules that made $3.4k–$4.4k a year there lost $533 to $1,554 a year on SPY 2016–2024.

What we tested

Four tests of the same idea: buy more as the price falls, sell everything at a profit, start again. All four use real prices (with 2× and 3× funds built from the S&P 500 before the real ones existed); the method at the end lists the fills and costs.

1S&P 500 funds since 1993

Tested: A $1,000 slice every 1–10% lower, everything sold at +1% to +10% on the average cost, then start again — on SPY, SSO (2×), UPRO and SPXL (3×), and 2× and 3× funds built back to 1993. Two ways to fund it: unlimited slices, or a fixed account split into 5, 10 or 20 slices.

Found: Every setting made money and every cycle closed, even on 3× through 2000–2009. But it needed up to $315,000 per $1,000 slice and earned less than holding at the same risk. With a fixed budget, once the slices ran out, the 3× funds fell 73–98%.

2Nasdaq-100 and S&P 500 futures, held for days

Tested: One more NQ or ES contract every 1, 3, 5, 10 or 20% below the high since the last exit; everything sold when the price is back at that high. Real futures prices, October 2024 → October 2026.

Found: Every step size made money — the one big drop, April 2025, recovered in about three months. But 1% steps on NQ needed 27 contracts and $2.6 million, and per dollar needed it returned about what holding did.

3Inside one day

Tested: A contract every 0.1–2% below the day’s high on 1-minute bars, flat at the close; then buying in the morning only, a day filter, and a whole-position day stop. MNQ and MES (the micro Nasdaq-100 and S&P 500 futures) and MNG (micro natural gas).

Found: The all-day version made at best $14k in two years on MNQ, with $43k of cash needed on its worst day. Guarded morning versions on MES made $3.4k–$4.4k a year in 2024–26 — and every one lost money on SPY 2016–2024.

4The time average

Tested: One micro contract every 30, 60, 90 or 120 minutes for 1–4 hours, starting between 9:30 and 1:00; everything sold at +0.5%, +1%, +2% or +3% on the average cost, or at 3 PM. MNQ and MES, October 2024 → October 2026.

Found: Every profit target did worse than no target. What the best schedule earned was the Nasdaq-100’s usual rise from 9:30 to 3 PM, times the number of contracts.

Unlimited buying: every cycle closes, at a price

$1,000 a slice and no limit on slices. “Money needed” is the most it ever had in at once — set by the worst fall, so it is known only afterwards. “Yearly on that money” is how fast an account of exactly that size grew. The last column keeps a fixed share of the same account in the fund, sized to fall no further than the strategy did.

Averaging down with unlimited $1,000 slices on S&P 500 funds, against holding at the same worst fall
Fund · step down, sell at	Profit a year	Money needed	Worst open loss	Yearly on that money	Its worst fall	Holding at the same fall
UPRO (3×) since 2009 · 2%, +2%	$3,807	$38,000	−$9,153	6.0%	−19%	7.3%
UPRO since 2009 · 1%, +10%	$12,935	$151,000	−$65,792	5.4%	−25%	9.5%
3× S&P 500 fund since 1993 · 1%, +5%	$11,008	$280,000	−$181,490	2.5%	−40%	5.7%
3× fund since 1993 · 1%, +10%	$11,999	$315,000	−$214,427	2.5%	−46%	6.4%
3× fund since 1993 · 5%, +5%	$1,908	$54,000	−$35,214	2.4%	−42%	6.0%
SPY since 1993 · 2%, +2%	$562	$36,000	−$10,212	1.3%	−23%	4.0%

Profit is per $1,000 slice. Buy and hold, for scale: SPY 10.9% a year since 1993 (worst fall −55%), the 3× fund 14.7% (−98%), UPRO 33.5% since 2009 (−77%). The longest cycle on the 3× fund ran 3.4 years, August 2000 → January 2004, and still closed at a profit: buying lower drags the average cost down, so a bounce is enough.

A fixed budget: once the slices run out, it is just holding

The realistic version: a fixed account split into 5, 10 or 20 equal slices, no buys after the last one, profits reinvested, idle cash earning nothing. When the fall goes past the last slice, the account rides it down like the fund.

Averaging down with a fixed account split into slices, against buy and hold
Fund · step down, sell at	Slices	Yearly	Worst fall	Longest cycle	Holding: yearly	Holding: worst fall
UPRO since 2009 · 5%, +2%	10	13.5%	−33%	3 months	33.5%	−77%
Same on the 3× fund since 1993	10	11.4%	−87%	4.3 years	14.7%	−98%
3× fund since 1993 · 2%, +2%	5	15.9%	−98%	17.3 years	14.7%	−98%
3× fund since 1993 · 10%, +2%	20	2.9%	−20%	2 months	14.7%	−98%
SPY since 1993 · 2%, +2%	10	4.4%	−46%	2.7 years	10.9%	−55%
UPRO since 2009 · 1%, +10%	5	34.2%	−77%	2.6 years	33.5%	−77%

UPRO’s smooth first row (18 of 18 years up) came from a bull market; the same setting on the 3× fund since 1993 fell 87%. Settings that kept every fall small (10% steps, +2%, 20 slices: worst falls −2% to −21%) made only 0.3–3.2% a year. Keeping idle cash in Treasury bills added 1–2.5% a year; buying again right after each sale, instead of waiting for a dip, was no better.

Futures: for days, then inside the day

Each step buys one contract. MNQ and MES, the micro Nasdaq-100 and S&P 500 futures, are a tenth the size of NQ and ES, so their dollar results are a tenth too.

NQ and ES, held for days

Every step size from 1% to 20% made money on both, because the one big drop (April 2025: NQ −27%, ES −22%) came back in about three months. The cash needed grew with the ladder: at 1% steps on NQ, 27 contracts, a $1.67 million open loss and $2.6 million needed. On the cash it needed it returned 102–135% (193% on ES at 20% steps), against 112% for holding one NQ contract and 82% for one ES — no real edge, and a slow bear market like 2022 would have needed far more contracts.

Natural gas: the price did not come back

On micro natural gas (MNG) at 1% steps, counted on the position’s own value, the ladder has been stuck since March 10, 2025 (as of the data’s end in October 2026), holding 68 contracts and needing $5.37 with the price at $3.22. Holding one MNG for the two years made only $275, though the quoted price rose 20%.

All day, flat at the close

The first dip below the day’s high came back to that high the same day 87% of the time at 0.1% steps on MNQ, but only 10% at 2% steps (MES: 83% → 6%). Selling at the average buy price lost money at every step. The best, MNQ at 0.25% selling one step above the average, made +$14k in two years — about $28 a day — with $760 of cash needed on a typical day but up to $43k, a −$24k day (April 8, 2025) and a −$32k worst run. MNG lost every way.

Morning only, with a day stop

Buying only until 11:00 cut the cash needed: MES at 0.25% steps made money in both years (selling three steps above the average: $26 a day, worst day −$2.7k, worst run −$5.2k). Trading only when the overnight range (6 PM to 9:30, high minus low) was at least 0.5% of the price lifted it to $9.0k and $8.5k a year, with 25 days worse than −$500. A $450 whole-position day stop kept every day above −$500 but cut the profit: the best that held in both years (0.5% steps, out by 2 PM) made $3.7k and $3.4k (worst day −$454, worst run −$2.0k).

Profit targets cut winners

The time-average test: buy one MNQ on a schedule through the day, then sell everything at a profit target over the average cost — or at 3 PM if the target never comes. Each row is the average two-year result over every start time, length and buying interval.

Time averaging on MNQ, average two-year result by profit target
Sell everything at	MNQ, average over all schedules
+0.5% over the average cost	−$8.5k
+1%	−$4.4k
+2%	+$0.9k
+3%	+$1.4k
No target: sell at 3 PM	+$5.3k

A +0.5% target was hit on 66% of days and still lost: it fires on up days after one to three buys and stops buying, while down days get the whole schedule held to 3 PM. Targets of +2% and +3% were hit on 1% of days or fewer. On MES every target lost on average, and no target made −$0.1k. The best schedule — no target, buying at 9:30 and every 30 minutes for 3 hours — made +$27.7k on MNQ ($15.1k and $12.6k by year, worst day −$9.4k, worst run −$23.9k); one MNQ held from 9:30 to 3 PM every day made +$4.9k.

Was the intraday edge real? How we checked

Two years of minute bars and thousands of settings will always produce a winner. These checks ask whether it survives years it was not picked on.

Many settings, two years

The morning test had 54 settings per market, so its best row is partly luck. A sweep of 107,856 MES settings put the top group at: start 9:00, buy until 1:00 PM, 0.25–0.35% steps, 2–3 contracts at most, a $450 day stop — whose best made $4.4k and $4.2k by year.

Picked on one year, run on the other

The top 20 settings picked on the first year averaged −$582 on the second; picked on the second year, −$2,448 on the first. With the day stop, the best pick on one year made about $0 on the other.

Outside test: SPY, 2016–2024

The guarded finalists run on SPY 1-minute bars from January 2016 to November 2024 all lost money: −$533 to −$1,554 a year, worst runs of −$7.6k to −$14.7k, 2–6 of 9 years up, and 2022 alone −$4.6k to −$8.5k.

Not the adaptation, the years

The same adapted rules still made $2.1k–$3.8k a year on ES in 2024–26, so the switch to SPY is not what broke them: 2024–26 was unusually kind to buying dips.

Without the stop

The uncapped 0.25% version made +$682 a year on SPY with a VIX filter, but had 84 days worse than −$500, a −$19k worst run and −$13.9k in 2022.

Why it falls short

The worst fall sets the cash

How much money the ladder needs is decided by the deepest fall, and you learn it only afterwards: up to $315,000 per $1,000 slice on the 3× fund, $2.6 million at 1% steps on NQ. Sized for that, the account earned 1.3–6% a year.

Holding does the same job

Holding a fixed share of the fund, sized to fall no further, earned more in almost every setting — 7.3% against 6.0% on UPRO, 6.4% against 2.5% on the 3× fund. Most of the cash the ladder must keep ready is used only in the rare deep fall.

Targets sell the winners early

Once up 0.5% on the average, the Nasdaq-100 tended to keep rising into the afternoon. On the 167 days a +0.5% target was hit (MNQ, five buys every 30 minutes from 9:30), it booked $201k where holding to 3 PM made $234k — and the losing days were unchanged.

Buying more costs money

In the MES sweep, wider steps and fewer contracts did better on average. Buying two contracts per step, or growing the size, did not help reliably.

A fixed budget becomes buy and hold

When the slices run out, the rest of the fall is plain holding: −73% to −98% on the 3× funds, and a cycle opened in March 2000 stayed open until 2017.

Holding longer is just the drift

Holding five MNQ buys (every 30 minutes from 9:30) past 3 PM into the evening raised the total from $22.2k to $55.6k by 9 PM, but $46.9k of it came in the second year and the worst run grew from −$19k to −$42k. That is the Nasdaq-100’s overnight rise; averaging added nothing, and no target beat no target there either.

The risks

The money needed is known only after the worst fall. A deeper or slower bear market than the test saw needs more slices — or the account runs out first.
Leveraged funds fall far: UPRO fell 77% at its worst since 2009, and a 3× S&P 500 fund built back to 1993 fell 98%. A fixed budget rides those falls once its slices are spent.
Futures losses grow with every contract added: at 1% steps on NQ the open loss reached $1.67 million before the April 2025 drop recovered.
Prices do not have to come back. On micro natural gas a 1% ladder has been stuck since March 2025, holding 68 contracts.
The futures tests cover two years with one big drop, which recovered in about three months — a kind window for buying dips.
Intraday results are fragile: the best settings from 2024–26 lost money on SPY 2016–2024, $4.6k to $8.5k in 2022 alone.

Questions

Does averaging down work?

It closes trades at a profit almost every time, but it did not beat simply holding at the same risk. Buying one more $1,000 slice every X% lower and selling at +T% on the average cost made money in every setting on SPY, SSO, UPRO, SPXL and a 3× S&P 500 fund since 1993, and every cycle eventually closed. But it needed up to $315,000 of cash per $1,000 slice and earned 1.3–6% a year on that cash, while holding the fund sized to the same worst fall earned more in almost every case.

Is buying the dip a good strategy?

Piling into a falling market was not better than holding in our tests. The averaging-down ladder earned less per dollar than holding the fund at the same worst fall, and the intraday dip-buying versions that made money in 2024–26 lost $533 to $1,554 a year on SPY from 2016 to 2024. The dip rules that held up in our other S&P 500 studies (such as buying when RSI(3) is under 20) buy one fixed amount and are out within days, so they never pile up.

How much money do you need to average down?

Far more than it looks, and you learn the amount only after the worst fall. With $1,000 slices the most cash in at once was $36,000 on SPY at 2% steps, $38,000 on UPRO at 2% steps and up to $315,000 on a 3× S&P 500 fund since 1993 at 1% steps. On Nasdaq-100 futures at 1% steps the ladder reached 27 NQ contracts and needed $2.6 million (a tenth of that with MNQ micro contracts).

Does taking profit at +1% work for day trading?

Not in our test. Buying one MNQ (micro Nasdaq-100 future) on a schedule through the day and selling everything at +1% over the average cost averaged −$4.4k over two years, against +$5.3k for holding the same buys to 3 PM with no target. A +0.5% target averaged −$8.5k even though it was hit on 66% of days. Every target did worse than no target, on MNQ and on MES: the target sells the up days early, while the down days are held to the afternoon anyway.

Is averaging down the same as a martingale?

It is a close cousin. A martingale doubles the bet after each loss; this ladder adds one equal slice each step, so the size grows more slowly — but the risk still grows as the price falls. At 1% steps on Nasdaq-100 futures it held 27 contracts with a $1.67 million open loss before the April 2025 drop recovered. Buying two contracts per step, or growing the size, did not help reliably in our intraday tests.

Does averaging down work on UPRO or other leveraged ETFs?

It closes cycles, but holding did better at the same risk. On UPRO since 2009, 2% steps with a +2% target made $3,807 a year per $1,000 slice but needed $38,000 — 6.0% a year on that money, against 7.3% for holding a fixed share of UPRO sized to the same fall. With a fixed budget in 10 slices, 5% steps and +2% looked smooth on UPRO (13.5% a year, −33% worst fall, 18 of 18 years up), but on a 3× fund since 1993 the same setting fell 87%.

Does averaging down work for day trading futures?

It looked good only in the years it was tuned on. Buying MES (micro S&P 500 futures) on each 0.25–0.5% dip in the morning, with a $450 day stop, made $3.4k–$4.4k a year in 2024–26. Run on SPY 1-minute bars from 2016 to 2024, every guarded finalist lost $533 to $1,554 a year, with 2022 alone costing $4.6k to $8.5k. Settings picked on one of the two years also lost on average on the other.

What is a better way to buy dips?

In our research, two approaches did better than a ladder: buying one fixed amount on a short-term dip and getting out within days (for example when RSI(3) on the S&P 500 is under 20, or holding only overnight above the 200-day average), and sizing a leveraged fund by its volatility. Both keep the position from piling up as prices fall. Their numbers are in the related studies.

Is this investment advice?

No. It is a backtest of past prices for education. Leveraged funds and futures can lose a large part of an account quickly, and past results do not promise future ones.

How the test works

S&P 500 funds: SPY since 1993, SSO since 2006, SPXL since 2008, UPRO since 2009, and 2× and 3× funds built from the S&P 500 back to 1993 so the 2000–02 and 2008 bear markets count (they track the real funds within 0.2% a year).
The ladder: the first $1,000 slice X% under the highest price since the last sale, one more equal slice each further X% down, everything sold at the average cost + T%; steps of 1, 2, 3, 5 and 10% and targets of +1, +2, +5 and +10%. Resting limit orders on daily bars with dividends included, read the cautious way (sale first, no buys after a sale that day), no commission; 1-minute bars since 2016 gave the same numbers except at 1% steps with +1% targets.
“Money needed” is the most cash ever in the position at once — for futures, margin plus the worst open loss. “Yearly on that money” is the growth of an account exactly that size from day one, idle cash earning nothing. A cycle still open at the end is valued at the last close.
“Holding at the same fall” keeps a fixed share of the same account in the fund, the rest in cash, rebalanced daily, sized so its worst fall is no deeper than the strategy’s.
Futures: real NQ, ES and natural-gas futures prices from October 11, 2024 to October 7, 2026 (498 sessions), joined across contract rolls so profit is exact, with 1-minute bars inside the day. Limit orders fill exactly at their level (at the open after a gap); $2.50 a side plus rolls on full-size contracts; in the time-average test $0.80 a side and one tick on each buy and on the 3 PM sale; day stops pay one tick of slippage.
Outside test: SPY 1-minute bars from January 2016 to November 8, 2024, with SPY × 10 standing in for the ES price, and a VIX ≥ 16.5 filter (or none) in place of the overnight-range filter.

Not investment advice

This is a study of past prices, published for education. Leveraged funds, options and futures can lose most of their value quickly, and a pattern seen in the past can break. Nothing here is a recommendation to buy or sell anything. Decide with your own judgment, or with a licensed adviser.