The golden cross, 25 years, 8 names: it works — and still loses to holding
A teardown. Every number in the results table is the tool's saved output; the "open leg" column and the reading at the end are ours and are marked as such.
2026-09-07 · textbook definition, public
The claim. The golden cross is the most quoted signal in retail technical analysis: go long when the 50-day moving average crosses above the 200-day, get out when it crosses back below. It is usually presented as a way to "catch the big trends and sidestep the big drawdowns."
What we ran. The textbook definition with nothing tuned: SMA(50) crossing above SMA(200) on daily closes → long with 100% of equity; SMA(50) crossing below → flat. No stop-loss, no take-profit, no costs modelled (the tool's cost panel re-runs the trades at 5–50 bps afterwards). Eight large US names, the full 25-year window our feed provides (Sep 2001 → Sep 2026, 6,283 daily bars each). Anyone can reproduce this on the free tier from the chart's condition tab — two moving averages, one crossover.
Results — 8 runs.
| Run | Trades | Win rate | Max DD | Strategy return | Buy & hold | Verdict flag | Open leg (ours, not counted by the tool) | Realized + open (ours) |
|---|---|---|---|---|---|---|---|---|
| SPY · 1d · 2001–2026 | 11 | 72.7% | 8.46% | +356.99% | +638.44% | UNDERPERFORMS_HOLD · Read with caution | long since 2025-07-01 · +24.7% | ≈ +470% |
| QQQ · 1d · 2001–2026 | 16 | 62.5% | 10.19% | +1,021.48% | +2,204.36% | UNDERPERFORMS_HOLD · Read with caution | long since 2025-06-24 · +33.2% | ≈ +1,394% |
| AAPL · 1d · 2001–2026 | 10 | 70.0% | 32.4% | +16,867.95% | +105,363.87% | UNDERPERFORMS_HOLD · Read with caution | long since 2025-09-15 · +35.2% | ≈ +22,841% |
| MSFT · 1d · 2001–2026 | 19 | 47.4% | 39.32% | +417.28% | +1,788.87% | UNDERPERFORMS_HOLD · Read with caution | long since 2026-08-31 · −1.5% | ≈ +410% |
| NVDA · 1d · 2001–2026 | 11 | 81.8% | 20.68% | +93,420.05% | +81,227.45% | BEATS_HOLD · Read with caution | long since 2025-06-27 · +46.0% | ≈ +136,439% |
| AMZN · 1d · 2001–2026 | 13 | 61.5% | 26.98% | +1,466.70% | +68,928.04% | UNDERPERFORMS_HOLD · Read with caution | long since 2026-05-06 · −6.0% | ≈ +1,373% |
| JPM · 1d · 2001–2026 | 16 | 50.0% | 32.99% | +422.32% | +917.42% | UNDERPERFORMS_HOLD · Read with caution | long since 2026-06-12 · +11.8% | ≈ +484% |
| XOM · 1d · 2001–2026 | 16 | 43.8% | 40.14% | +140.08% | +297.19% | UNDERPERFORMS_HOLD · HIGH_DRAWDOWN · Read with caution | long since 2025-08-22 · +43.3% | ≈ +244% |
Every run also carries the tool's LOW_SAMPLE flag (10–19 trades in 25 years) and NO_COSTS flag.
The tool's own words.
- SPY: “Underperforms buy & hold by 281.45% — it lost to doing nothing (hold: 638.44%).” · “Only 11 trades — thin sample; treat as preliminary (aim for 30+).” · “No trading costs modeled — real fills will be worse.”
- SPY, cost panel: “Survives realistic costs — Profit holds up to >50 bps of round-trip cost.”
- NVDA: “Beats buy & hold by 12192.6% (hold: 81227.45%).”
- XOM: “Max drawdown 40.14% — could you actually hold through that?”
Our reading (not the tool's).
- It is not a fake. Eight names, eight positive results, and the cost panel can't kill any of them — a 50 bps round trip barely dents a strategy that trades once every two years. This is the opposite of the 93%-win-rate strategy in teardown #1, where the win rate was the artifact and the profit was missing. Here the win rate is unremarkable (44–82%) and irrelevant: a handful of multi-year trends did all the work.
- "Works" is not "worth it." Seven of eight lose to simply holding, most by a wide margin. The cost of the golden cross is not fees — it's the months spent flat after a death cross while the market recovers. Across a 25-year window that was mostly up, that cost compounds into hundreds or thousands of percentage points.
- The one win is the one you'd expect. NVDA beats holding because its history contains crashes deep enough that being out for a while paid for the late re-entries. That is a fact about NVDA's path, not evidence the rule finds trends.
- Thin samples, real drawdowns. Ten to nineteen trades in 25 years is too few for any ratio to mean much (the tool says so on every page), and "sidestepping the drawdowns" still left MSFT at −39% and XOM at −40% peak-to-trough.
- What the table doesn't count. Every one of these runs is currently long — the tool's engine leaves a position that is still open at the end of the data out of every metric. We added that leg by hand (entry close → last close on Sep 4, 2026). It changes the sizes, not the verdicts: with the open leg included, SPY is ≈ +470% vs +638% held.
Honest limits. One 25-year window, one regime sequence (two crashes, one long bull); long-only; no taxes or slippage; entries and exits fill at the signal bar's close. The open-leg column is our arithmetic on the same price series, not tool output.
Receipts
8 share pages, open without an account (each shows the tested window, bar count, verdict and cost panel):
- SPY 1d: https://stlab.app/v/452189b1-646b-4284-b741-4c0587e1b1be
- QQQ 1d: https://stlab.app/v/818872f6-f2fd-4244-b1bf-a70778d7c71b
- AAPL 1d: https://stlab.app/v/4213d1a1-3d89-4341-9372-a4ea218772eb
- MSFT 1d: https://stlab.app/v/7b123a58-2cfa-4a3f-9065-360a04be87b9
- NVDA 1d: https://stlab.app/v/2d9bc7f1-d7c2-448c-bf35-a93cf044e37c
- AMZN 1d: https://stlab.app/v/e6907bc1-d4a8-4354-a7b7-4f98202ae5a8
- JPM 1d: https://stlab.app/v/8312be8b-9534-4492-a641-dbaa68e1e20d
- XOM 1d: https://stlab.app/v/b8642ea3-6687-46d8-86bb-08c58d1855ea
Simulated results on historical data — research output, not financial advice. Past performance does not guarantee future results.