QQQ Premarket Oversold Bounce — Backtest

Cumulative net P&L for 100 shares, by exit date. The dashed line marks Jan 1, 2023: the rules were chosen on data to its left only. Prices are split- and dividend-adjusted, so early-year dollar amounts reflect QQQ's lower price then.

• Bars: 15-minute QQQ bars built from SIP 1-minute bars between 04:00 and 20:00 ET. Indicators: RSI(14), MACD(12,26,9) histogram, Bollinger Bands(20, 2), 20-bar volume average.
• Signal: RSI < 25 and a confirmation score of at least 1.5 (MACD histogram rising +1; close at or below the lower band +1, else at or below the middle band +0.5; volume at least 0.8× its average +1). The signal stays armed while RSI is below 35.
• Entry: buy 100 shares on the first bar that closes higher than the one before it, only if that bar closes between 04:00 and 09:30 ET (pre-market).
• Exit: at the close of any 15-minute bar from 04:00 to 20:00 ET when RSI > 60, or the close reaches the upper band, or the MACD histogram turns down with RSI > 55. No stop; a position can be held overnight.
• Fills in the backtest: the volume-weighted price of the minute after the signal bar closes, plus a per-side cost of 4¢ pre-market, 2.9¢ regular hours and 4.75–5.5¢ after hours. These costs are the quoted half spread plus a delay cost, calibrated so the original 1026 rules reproduce their real-quote result on 2025–2026 (−8.5¢ a share per round trip).

Median hold 4.8 hours; 32% of trades are held more than 6.5 hours, and the position is in the market 3.1% of the time. Exits: RSI 144, upper band 120, MACD fading 35.

This strategy came out of a study of strategy 1026 (QQQ mean reversion on 1-minute bars, 04:00–20:00 ET). With realistic fills, 1026's own rules lose money in every period, so 11,520 variations were tested: bar length (1, 5, 15 min), RSI threshold, entry session, stop, exit rule, time limit, overnight holding, a trend filter and up-bar confirmation. Only 428 were profitable in 2016–2022. The rules above had the best 2016–2022 Sharpe ratio; 2023–2026 was not used to choose them.

• Market drift. QQQ rose about 19.6% a year over the sample, which explains about 2 bp of each trade over these holding times, against +26.5 bp earned.
• Random entries. 300 sets of random pre-market entries with the same exits and costs averaged -1.5 bp a trade (best 1%: +6.2 bp). None did as well as the signal.
• Real quotes. The 62 trades from 2025–2026 were re-priced at the actual bid and ask one second after each signal (buy at the ask, sell at the bid): +231¢ a share per trade, $14,319 for 100 shares, against +231¢ in the bar model. Median spread at entry 6¢, at exit 3¢.
• Selection. With 11,520 variations tested, a 2016–2022 Sharpe of 1.26 could arise by luck alone. The evidence is the out-of-sample period, the checks above, and the neighbouring settings below, all of which stayed profitable from 2023 on.

• Few trades. About 28 a year, so a year of live results is a small sample, and returns are lumpy: 2020, 2022 and 2025 made most of the money, and 2021 lost slightly.
• No stop. Positions are held until an exit signal, sometimes for days. The worst trade lost 3.0%. A 2% stop did not reduce the worst losses, which came from overnight gaps.
• Pre-market fills. Entries happen as early as 04:00, when QQQ trades thinly. Before 2020 pre-market liquidity was lower than the cost model assumes.
• Selection. The rules are the best of 11,520 tested variations; out-of-sample results are usually weaker than the backtest.