A paper recently went up on Quantpedia: “Commodity portfolio strategy for a potential 2026 inflationary and supply shock regime”.
Out of 10 commodity ETFs — gold, silver, copper, crude oil, and the like — buy only the 3 that have been strongest over the past year. Add volatility targeting on top and, the paper claims, you get better returns with lower drawdowns.
So I ran the backtest myself, on 14.6 years of data.
1. The paper’s results do replicate.
7% annualized, max drawdown of -45%. Clearly better than a plain equal-weighted basket (2% annualized). So far, so good.
2. But once you slice it by period, the cracks show.
- 2013–2016 (commodity bear market): -11% annualized, MDD -45%. Four straight years in the red.
- 2017–2021 (middling): +8% annualized.
- 2022–2026 (inflation regime): +26% annualized.
The numbers that look good are entirely propped up by the last four years. The nine years before that were either mediocre or downright brutal.
3. What that “last four years” of returns actually is
Of the +26% annualized from 2022–2026, roughly 80% came from a single year — 2025.
- Silver (SLV): +162%
- Platinum (PPLT): +139%
- Palladium (PALL): +76%
- Gold (GLD): +65%
A one-off event where precious metals all ripped at the same time. It wasn’t the strategy being clever — it was just that anyone holding precious metals during that stretch made money.
4. Compared to U.S. equities (QQQ)
Over the same 14.6 years, QQQ returned 20% annualized. The commodity strategy: 7%. And if you slice the timeline into 4 windows and compare,
QQQ wins in every window but one.
Verdict: I’m not running this in real money.
- Drawdown is larger than the annualized return (disqualifying as a standalone strategy).
- The whole track record hinges on a single precious-metals rally inside a 14.6-year window.
- Recovery from a bear market takes four years.
The chart a paper puts in front of you and what the actual time series looks like, sliced apart, are often this different. If I had just followed the paper without backtesting it myself, I’d have had to sit through a 2013–2016-style stretch for four years.
※ Verified on 14.6 years of daily data with monthly rebalancing. Transaction costs not included.
댓글