compoundsimDoes Diversification Work in a Crash? Stocks, Bonds and Gold
Diversification can't prevent a crash, but it can change how deep the fall goes. See how stocks, bonds and gold moved, and when bonds fell too.
Published
Diversification can't prevent a crash, but it can change how deep the fall goes. See how stocks, bonds and gold moved, and when bonds fell too.
Published
Diversification can't prevent a crash, but it can change how deep the fall goes, and how much depends on what you diversify into. Over 1,000 simulated 20-year paths, the median (the middle result) of the max drawdown, the largest fall from a previous high along the way, was -29% for the S&P 500 alone and -30% for stocks spread across regions, but -18% for 60% stocks and 40% bonds. Bonds don't always help, though: in 2022 they fell along with stocks. And at the median in this simulation, the 60/40 mix grew less than the S&P 500 alone.
Correlation measures, from -1 to 1, how much two assets move in the same direction; the closer to 1, the more they move together. Diversification helps when one asset holds up in the months another falls.
As the article on Monte Carlo simulation explains, the model on this site switches between a calm regime and a wild regime, in which prices swing more and fall more often. In the model fitted to the full history, correlations with the S&P 500 look like this:
| Correlation with the S&P 500 | Calm regime | Wild regime |
|---|---|---|
| Developed ex-US stocks (EFA) | 0.80 | 0.91 |
| Emerging market stocks (EEM) | 0.67 | 0.82 |
The wild regime is not the same thing as a crash: it is a stretch of larger swings that lasts 3.4 months on average. Both correlations start high and rise further in it: stocks around the world tend to be sold at the same time. Calendar-year total returns (dividends reinvested) in dollars show the same thing. In 2008 the S&P 500 returned -36.8%, developed stocks -41.0% and emerging market stocks -48.9%: all three fell sharply.
Other assets have behaved differently. Here is the annualized growth of each in the same model's wild regime:
| Asset | Correlation with the S&P 500 (wild) | Growth in the wild regime (annualized) |
|---|---|---|
| S&P 500 (SPY) | - | -8.1% |
| US aggregate bonds (AGG) | 0.30 | 3.7% |
| US long-term Treasuries (TLT) | -0.01 | 7.5% |
| Gold (GLD) | -0.07 | 13.4% |
In the model's wild regime, stocks fell on average while bonds and gold were positive. An annualized rate is the pace if it lasted a full year; wild spells usually end within months, so these are not one-year results. Bonds and gold also have shorter histories than the S&P 500, so their figures rest on fewer wild months, and even the fact that they were positive is uncertain.
The calendar year 2008 agrees. Against the S&P 500's -36.8%, US bonds returned 7.9%, long-term Treasuries 34.0% and gold 4.9%.
Bonds don't always protect a portfolio. In 2022 the S&P 500 returned -18.2%, US aggregate bonds -13.0% and long-term Treasuries -31.2%, the last a bigger fall than stocks. Gold returned -0.8%.
That year, high inflation and fast-rising interest rates pushed stocks and bonds down together. Bond prices fall when rates rise, and the longer the maturity, the further they fall. When a weakening economy brings rates down, bonds tend to cushion stocks. When rising rates are the cause of the fall, bonds can go down with stocks.
The shift shows up in correlations too:
| Correlation with the S&P 500 (monthly, actual) | 2005 to 2020 | 2021 to 2025 |
|---|---|---|
| US aggregate bonds (AGG) | 0.04 | 0.63 |
| US long-term Treasuries (TLT) | -0.31 | 0.57 |
| Gold (GLD) | 0.07 | 0.14 |
Correlation is not a fixed property. Compare the correlation matrix for 2021 to 2025 with the matrix for 2005 to 2020 to see the change.
To see what this means for your money, we compared four mixes under these conditions:
A path's max drawdown is its largest fall from a previous high, measured at month ends.
Median10th percentile
Spreading stocks across regions did not make the drawdown shallower, because correlations are high in the wild regime, annualized volatility (the size of the swings) in that regime is larger, at 24% for developed stocks and 32% for emerging market stocks against 21% for the S&P 500, and their growth in the calm regime is lower. Adding 40% bonds took the median from -29% to -18%, and the 10th percentile from -45% to -30%. That 10th percentile of -30% means that on one path in ten, the portfolio fell at least that far below its high at some point.
Swapping a tenth of the bonds for gold made the drawdown slightly shallower in this comparison, but the step is not as large as the one from adding bonds.
Adding bonds has a price. Here is the same experiment's final balance after 20 years, as a multiple of the starting sum:
| Balance after 20 years (multiple) | 10th percentile | Median | 90th percentile |
|---|---|---|---|
| S&P 500 only | 2.6 | 8.8 | 22.2 |
| Stocks, 3 regions | 2.2 | 7.2 | 20.4 |
| 60 stocks / 40 bonds | 2.3 | 5.0 | 9.1 |
| 60 / 30 bonds / 10 gold | 2.9 | 6.1 | 10.8 |
The 60/40 mix has a lower median than the S&P 500 alone, and the gap is wider at the 90th percentile. At the 10th percentile the gap is small. What bonds change in this experiment is less the bad outcome at the end than the upside and the depth of the falls along the way.
The mix that swaps a tenth of the bonds for gold beat 60/40 at all three levels in this comparison. That reflects gold's strong run in the data period; it is not a rule that swapping into gold raises growth.
The numbers alone don't say which is better. The right mix depends on how large a fall you could sit through without selling, and when you will need the money. If you are already withdrawing, the depth of a fall also affects how long your money lasts, through sequence of returns risk. You can try your own mix in the portfolio simulator.