Does buying 3% dips in Nasdaq-100 stocks beat holding QQQ?

Buy a stock that fell 3% today, only if it falls a little further tomorrow, and sell on the bounce.

Every day, look at the Nasdaq-100 stocks. A stock qualifies when it closed more than 3% below the day before, was still above its 200-day average price, and had been moving more than 3% a day over the last five days. The next morning, place a limit order to buy it 0.9 times its recent daily range below that close, which in practice means roughly 2–4% lower. If the stock never falls that far during the day, there is no trade.

Each purchase uses a tenth of the account and at most ten stocks are held at once; if more qualify than there are free slots, take the ones moving the most. Sell at the first of three exits: the stock bounces to yesterday's close plus half a day's range; it closes above the previous day's high, in which case sell at the next open; or ten trading days have passed. Every buy and sell costs 0.10%, and idle cash earns three-month T-bill interest. Position size, cash interest and the exact fill and exit timing were all fixed before any result was seen.

In practice 93% of trades ended at the bounce exit, so the ten-day limit almost never mattered.

The rule stayed far behind QQQ in every period, with a much shallower worst fall.

The chart starts every account at $10,000 on 31 January 2007. The two T-bill mixes show what QQQ would have earned at the rule's own exposure: the rule beat holding 5% QQQ and 95% T-bills, and roughly matched 15% QQQ and 85% T-bills.

Show on graph · ending value from $10,000

Holding QQQ$187,800Nasdaq-100 equal weight$114,200Dip rule$32,70015% QQQ + 85% T-bills$21,300QQQ at the rule's 5% exposure + T-bills$15,300Treasury bills$13,500

Log scale, so equal percentage moves take equal vertical space. Every series includes 0.10% per side on trades and T-bill interest on idle cash. Unticking a series hides it without changing the scale.

The 2014–2019 and 2020–2026 periods were reserved before any return was calculated. The last eighteen months, from February 2025, were the rule's best stretch in a volatile year: 1.67% per trade over 121 trades.

The gain needs both halves of the rule: the 3% dip and the lower limit order.

Buying the same dips at the next morning's open, without waiting for a further fall, earned almost nothing per trade. Placing the same limit order on random days that were not dips earned nothing either. Removing the uptrend filter kept most of the edge but doubled the account's worst fall.

The edge also repeated year after year rather than coming from one lucky stretch. Each bar is the average trade after costs for trades signalled in that year, with the number of trades underneath.

No. All 145 nearby settings made money, and the base settings sit in the middle of the pack.

If someone tunes numbers to fit the past, two things usually show: small changes break the result, and the chosen numbers sit at the top of their neighbourhood. Neither happened here. We reran the rule with 2%, 3%, 4% and 5% dips, limit orders 0.5, 0.9 or 1.3 ranges below the close, targets of half or one range, 5, 10 or 20-day limits, with or without the volatility filter, and with a 100-day average. Every one of the 145 versions made money per trade from 2014 to 2026. The base settings rank 18th of 145 by return per unit of risk and 71st by growth. The best-growing version made 11.1% a year, still well behind QQQ.

Two standard statistics say the same in numbers. If you picked the best-looking settings from this neighbourhood using past data, there is a 65% chance they would rank below the middle afterwards. That sounds bad but means the opposite of tuning: the settings all behave alike, so there was little to tune. And the base version's return per unit of risk is far above what the luckiest of 145 random tries would produce (deflated Sharpe ratio 0.99).

What these checks cannot rule out is that the idea itself survives because it worked on past data. Buying dips in strong stocks is an old idea precisely because it has worked before.

Costs, fills and cash interest move the number; none of them changes the conclusion.

With no trading costs the account grew 7.3% a year; at 0.25% per side, 4.8%. Without interest on idle cash, 4.7%. Placing orders for every qualifying stock instead of only the free slots, 7.0%. Twenty slots of 5% instead of ten of 10%, 4.5%. Requiring the price to trade 0.2% through the limit before counting a fill, 4.8%. Every version stays far behind QQQ.

Daily prices cannot show whether a resting limit order would actually have filled at the day's low, so the strict rule (fill only if the low is below the limit) and the trade-through row bound that uncertainty. Actual broker fills were not available.

Licensed daily prices, public index membership, rules frozen before any result.

Every rule and every control was fixed in a registration before a single return was calculated, and no setting was changed afterwards. The test uses daily prices from a licensed market-data provider whose terms permit derived research like this, the Nasdaq-100's public membership history, and Federal Reserve T-bill rates. A private test on a different price source reached the same conclusion with slightly more trades; this one has about 13% fewer trades because some delisted companies' price histories are no longer available.

That gap is concentrated in the early years. The share of listed members with prices in this test rises from about 73% in 2007 to 90% by 2014 and 100% from 2022. Missing names were acquired or delisted companies, so the early years lean toward survivors; the 2014–2026 periods that carry the conclusion are better covered.

Published 17 September 2026. Market history ends 8 July 2026. Returns are historical simulations, not a forecast.