Backtested the Death Cross on the 41 Biggest ETFs
While the strategy cuts drawdowns in almost every fund, it improved risk-adjusted returns in only 4 of 41.
Hypothetical performance. Backtested results are simulated, do not reflect real trading, and are not indicative of future returns. For research and education only — not investment advice.
The "death cross" — the 50-day moving average falling below the 200-day — is the most-feared pattern in financial media. I backtested it on SPY: hold while the 50-day is above the 200-day, go to cash on the death cross, ~20 years. It does one thing well: it cut SPY's worst drawdown from 55.2% to 33.7% — roughly half the pain of the 2008 collapse. That is a real result, and it's why the signal has a following. The cost is everything else: the strategy returned 479.9% against 784.3% for simply holding, and it didn't buy better risk-adjusted returns either — Sharpe 0.49 vs 0.51, essentially a wash.
If you stopped there, you'd file the death cross under "decent risk-reducer" and move on. So I ran the identical strategy across the 41 largest ETFs by dollar volume — SPY, QQQ, SMH, IWM, SOXX, VOO, and dozens more (sector SPDRs, bonds, gold, single-country and thematic funds). Here's what the data shows.
One backtest tells you almost nothing
- It beat buy & hold on total return in only 3 of 41 ETFs (XLF, FXI, XLC).
- It improved risk-adjusted return (Sharpe) in just 4 of 41. The best of them, XLC, improved Sharpe by 0.11. The other 3 improved it by 0.05 or less.
- The one thing it did do almost everywhere — reduce drawdown — held for 38 of 41 funds, but at a punishing price in return.
The median ETF sacrificed about 198 percentage points of total return for a 16.4% smaller drawdown, and its Sharpe actually fell — risk-adjusted, it got worse, not better. Take SMH: buy-and-hold returned 5162.8% over the window, while the death-cross version returned 2543.8% — you'd have left roughly 2,619 percentage points of return on the table to pull the drawdown from 62.0% down to 33.6%.
Even volatility barely predicts it
The intuitive guess is that the death cross earns its keep on jumpy, high-volatility funds and does nothing on calm ones. The data won't support even that:
- Drawdown reduction vs volatility: r = 0.22 (weak positive) — barely a tendency, nothing you could act on.
- Sharpe improvement vs volatility: r = 0.21 (weak positive) — weak at best.
- Return sacrificed vs volatility: r = -0.36 (moderate negative) — higher-vol funds tend to give up more return.
Ten funds across the range
Ten of the 41, picked to span asset classes rather than to rank by size — broad market, semiconductors, gold, bonds, financials, China, communications. Each runs its own death cross against its own buy & hold. Same pattern across all 41.
| ETF | Vol (1y) | Strategy return | Buy & hold | Return gap | Strategy max DD | Buy & hold max DD | Drawdown reduction | Sharpe Δ | Win rate |
|---|---|---|---|---|---|---|---|---|---|
| SPY | 12.8% | 479.9% | 784.3% | -304.5% | 33.7% | 55.2% | 21.5% | -0.02 | 80% |
| QQQ | 19.6% | 1242.5% | 2225.5% | -983.0% | 28.6% | 53.4% | 24.8% | -0.03 | 83% |
| IWM | 19.1% | 80.2% | 483.6% | -403.3% | 50.3% | 59.0% | 8.7% | -0.28 | 64% |
| SMH | 38.9% | 2543.8% | 5162.8% | -2619.0% | 33.6% | 62.0% | 28.4% | -0.02 | 69% |
| SOXX | 44.9% | 2591.8% | 3579.4% | -987.6% | 34.3% | 66.8% | 32.6% | 0.05 | 75% |
| GLD | 28.5% | 298.0% | 540.9% | -242.9% | 36.3% | 45.6% | 9.3% | -0.10 | 43% |
| TLT | 9.6% | 51.6% | 82.7% | -31.0% | 26.9% | 48.3% | 21.5% | -0.10 | 33% |
| XLF | 14.5% | 265.6% | 226.6% | 39.0% | 45.2% | 82.7% | 37.6% | 0.05 | 54% |
| FXI | 20.0% | 153.9% | 104.8% | 49.1% | 43.3% | 72.7% | 29.3% | 0.00 | 62% |
| XLC | 14.9% | 150.0% | 143.6% | 6.4% | 30.3% | 46.7% | 16.4% | 0.11 | 67% |
So even the most obvious predictor — how volatile the fund is — tells you little about whether the rule will help it. The only reliable way to know is to test the specific fund.
The setting that changes the answer
Every backtest has to decide when your order fills. The crossover appears at Thursday's close — do you get Thursday's closing price, or Friday's open? Filling at the same close that produced the signal is look-ahead: at that moment you couldn't have known. On SPY that single setting is worth about 47 percentage points of return (526.9% the flattering way, 479.9% the honest way) — same rule, same data, same window. Most tools never tell you which one they used. Everything above fills at the next open.
The takeaway
A backtest on a single ticker is an anecdote, not evidence. On SPY the death cross was a defensible risk-for-return swap — less return, but a much smaller drawdown at about the same Sharpe. Across most of the biggest ETFs that swap falls apart: you sacrifice far more return for less (often no) risk-adjusted benefit — and you couldn't have guessed which funds from volatility alone. Before you trust any rule — this one or anything else — run it across the specific things you actually hold and check whether the edge survives. That's a two-minute check, not a research project.
Method & caveats
- All figures as of August 13, 2026, 20-year window ending that date; a live re-run later rolls the window forward and moves the numbers slightly (drawdown is stable, returns creep).
- 41 largest ETFs by 30-day dollar volume (pulled from the screener, deduped by ticker); each ETF's SMA50/200 death-cross strategy vs its own buy & hold, up to ~20 years of daily data.
- All figures are total return — dividends reinvested, prices split-adjusted — measured the same way on both the strategy and buy-and-hold legs.
- Signals are computed on the daily close and filled at the next day's open — no look-ahead. (Filling at the same close that generated the signal is a common backtesting error that flatters the strategy; every figure here uses the honest next-open fill.)
- "Beat return / improved Sharpe / reduced drawdown" measured per fund; ETFs with fewer than 3 crossover signals are excluded from the stats (too little history).
- Returns are gross — no transaction costs or slippage modeled — but turnover is low (a median of ~13 signals per fund over ~20 years), so costs are immaterial and would only make the timing strategy look slightly worse vs buy-and-hold, never better.
- Correlation ≠ causation; volatility(1y) is recent volatility measured against a multi-year backtest; a single snapshot in time; daily data only (no intraday).
Don't trust a backtest you didn't run — try it yourself → quantibot.ai/backtest.