QQQ capitulation / euphoria monitor

session of 2026-10-09 (New York close) · page rebuilt 2026-10-10 05:04 CST

Today's two picks

Market gate OPEN — SPY is above both its 200-day and 50-day averages. These two are buys under the model. (Breadth is 29%, shown for context: it is reported but no longer gates, because testing showed a breadth filter cost return without reducing drawdown.)

DELLDell Technologies

87

new buy$586.06 20d +3.3% 12-1 mom +208% RS 0.98 of 6m high U/D 1.73 +16% vs 50d+93% vs 200d

HPEHewlett Packard Enterprise

87

already held$73.46 20d +18.3% 12-1 mom +112% RS 1.00 of 6m high U/D 1.81 +26% vs 50d+93% vs 200d

Model: trend (above the 50- and 200-day averages), 12-month momentum excluding the last month, relative strength at a 6-month high, return per unit of its own volatility, up-day vs down-day volume, and a penalty for names already stretched above the 50-day. Two names a day, bought at the next open, held 20 sessions. Backtested on 1996–2015, then checked on 2016–2026: holdout CAGR +24.6% against SPY's +13.3% — but a −47% maximum drawdown against SPY's −34%. Better returns, a rougher ride. It buys only while SPY holds both its 200-day and 50-day averages; an earlier breadth requirement was removed after testing showed it cost about two points of CAGR and prevented nothing. The panel is survivorship-flattered, so read the gap to SPY rather than the level.

Track record — every live pick graded against what happened next

6 live picks graded alongside the replayed history, in absolute terms and against the median stock on the same day. 123 picks over 101 sessions (6 of them live, the rest replayed from cached history), 931 graded outcomes.

What the model was validated on, before any of this ran: parameters chosen on 1996–2015, then checked once on 2016–2026. Holdout CAGR +22.7% against SPY's +13.3%, with a −46% maximum drawdown against SPY's −34%, and 55% of trades profitable. That panel is 295 survivors over 30 years and is survivorship-flattered, so read the gap to SPY rather than the level.

If you had bought them — live picks only

Covering 2026-10-02 to 2026-10-09 (5 days on which the model picked). Live picks only. Replayed history is deliberately excluded here: you could not have bought a pick that was never sent to you, so showing its profit in this table would be a P&L on trades that never existed. Replayed results live in the track record below, which measures the model rather than your account. Entry is the next session's open after a name appears, because the scan runs an hour after the close and that day's close was never available to buy. Equal $10,000 positions, no costs, held to the latest close. Read this table for the individual names, not for a verdict on the screen. Most positions here are only days old, and the backtest puts the edge at 20–60 sessions while finding the 1–5 day excess statistically indistinguishable from zero. A red ten-day total is noise in either direction; the 20- and 60-day rows in the track record above are what the screen should be judged on.