Can Gold and Fear Warn You Before a Nasdaq Crash?
We tested a simple weekly rule that steps out of the Nasdaq-100 when gold stops outrunning fear. It roughly halved the worst drawdown. It also gave up return. Here are both sides, with the numbers.
ApolloAlpha Research | 6 November 2026
The idea: before big equity selloffs, fear tends to rise faster than gold. Can that be turned into a crash filter?
The idea in plain terms
Each week we compare gold with the market's fear gauge, the VIX, and look at how that comparison has been trending over the last couple of months.
- When gold is holding up or gaining against fear, the filter stays in the Nasdaq-100 for the following week.
- When that trend has turned down, the filter steps aside into cash for the following week.
The signal is read at Friday's close and only used for the week after, so no future information leaks into the test. Cash earns 0% in this test, which understates what real cash would earn.
Reading the signal: gold against volatility
The VIX is often called the market's fear gauge: it measures how much movement options traders expect from the S&P 500 over the next 30 days. Gold is the classic place to hide when people are nervous. Comparing the two shows how much gold is worth relative to how much fear is priced in.
- When that comparison is improving, gold is gaining on fear. That has usually meant a calm market.
- When it is deteriorating, fear is rising faster than gold. That is the pattern the filter treats as a warning.
The full written rule, the year-by-year numbers and the robustness tests are part of our Institutional research (see below).
See it for yourself
Top: growth of $100 for the filter and for simply holding QQQ, with a red marker on each week the filter moved to cash. Middle: the gold/volatility filter itself, with a red x on each week it signalled cash. Bottom: the worst peak-to-trough drop of each strategy inside every calendar year, with green marking the years the filter's drop was smaller. Zoom into any period by dragging on the top two panels: both lines restart at 100 at the left edge of your view, and double-click to reset. The chart follows the page's light or dark theme.
Best viewed on a larger screen.
What we measured
| 6 Mar 2015 to 30 Oct 2026 | Crash filter | QQQ buy and hold |
|---|---|---|
| Annualised return | 15.0% | 18.0% |
| Worst peak-to-trough drop | -18.7% | -35.1% |
| Sharpe ratio (4% risk-free rate) | 0.95 | 0.73 |
| Share of weeks held in cash | about half | none |
| Years with a smaller worst drop | 11 of 12 (1 tie) | none |
The filter did what a crash filter should in the worst years. Inside 2022, QQQ fell 34.1% from its peak at the worst point and the filter fell 18.7%. In 2020 the gap was 27.3% against 4.3%, and in 2018 and 2025 it was about 21% against under 4%.
On a risk-adjusted basis the filter came out ahead. Its Sharpe ratio was 0.95 against 0.73 for buy and hold, because it earned most of the return with roughly half the volatility.
The cost was real. The filter was in cash about half the time, and in strong years it captured only part of the rally. Over the whole period it returned less than simply holding.
So, what does this tell us?
A single cheap signal built from gold and the VIX carried real information about when the Nasdaq-100 was about to suffer, and acting on it roughly halved the worst drawdown. But risk reduction is not free: the same rule sat out a lot of good weeks.
Whether that trade is worth it depends on the investor. Someone who could not stomach a 35% drop may value it. Someone who only cares about the highest long-run return may not.
The complete rule with its exact settings, the interactive chart showing every signal week, year-by-year numbers, and the robustness tests we ran on it.
🔒 Read the full strategy →Method and limits
Data. Daily adjusted closes (dividends included) for QQQ and the gold ETF GLD, and the VIX index level, from Yahoo Finance, resampled to the last close of each week (Friday). Study window: 6 March 2015 to 30 October 2026, 609 weeks.
Returns. The weekly QQQ return when the filter is invested, 0% when it is in cash, compounded. Max drawdown inside a calendar year is the largest fall from a peak to a later low on the weekly closes, counting from the prior year-end level. The Sharpe ratio uses a 4% risk-free rate on weekly returns, annualised. No trading costs, taxes or slippage are included.
Limits. The rule was fitted on data up to June 2026, so everything before then is in-sample. June 2026 onward is out of sample: it was not used to build the rule. Because the rule was fitted on past data, we also tested it for robustness. We varied its settings and checked that nearby values gave similar results, so the result is not sitting on a sharp peak that only the exact chosen settings would reach. One of the settings is more sensitive than the other, which is why the in-sample numbers above are still likely flatter than a live version would be. The out-of-sample period is short, about five months, so it can confirm very little yet. It is one market, a single sharp drawdown (2022) drives much of the result, and the test assumes the filter can trade at the Friday close.
Research note: This is an exploratory study of historical data. It is not a trading recommendation or financial advice, and past results do not guarantee future results.
ApolloAlpha Research
Questioning assumptions. Testing what the market actually does.