Do Bonds Really Protect You When Stocks Crash?

You hold stocks and bonds so that when one has a bad year, the other might not. How often that actually works - and how a simulator should model it - matters more than most people think.

In 2008, U.S. stocks lost 37%. Ten-year U.S. Treasury bonds gained 20% the same year. In 2022, it was a different story: stocks fell 18% and those same bonds fell 18% too. Same two investments, two completely different relationships.

That relationship is called correlation, and it shapes how much protection your mix really gives you.

What Correlation Means

Correlation measures how two investments move together, on a scale from -1 to +1:

• +1: they always move in the same direction. Holding both gives no protection.
• 0: they move independently. Holding both still smooths the ride, because their bad years rarely line up.
• -1: they always move in opposite directions. One cushions the other perfectly.

A common misunderstanding is that investments need negative correlation to diversify. They don't. Even independent investments reduce risk when combined. Take stocks with 16% volatility and bonds with 6%: a 60/40 mix has about 10% volatility if the two are independent - well below the 12% you get from simply averaging them. Only a correlation of +1 leaves you at 12%.

What History Says

Using annual U.S. data from 1928 to 2025 (S&P 500 and 10-year Treasuries), the stock-bond correlation over the whole period is almost exactly zero: +0.02. But the average hides very different eras:

In the high-inflation decades, stocks and bonds tended to fall together. From 2000 until the 2022 inflation shock, bonds were an excellent cushion. Nobody knows which kind of era the next 30 years will be.

How a Simulator Handles It

A Monte Carlo simulator draws a return for each investment in every simulated year. If it draws them independently, a stock crash is paired with a random bond year. With a correlation, a stock crash is more likely to come with a bond year that is also weak (positive correlation) or strong (negative correlation).

The standard way to do this is Cholesky decomposition: the simulator draws independent random numbers, then mixes them with a matrix derived from the correlations, so the results move together in the specified way. The math is routine; the hard part is choosing the correlation.

Retirement Lab uses a fixed, deliberately cautious set on every plan, free or Pro:

A stock-bond correlation of +0.3 is close to the 1970-1999 era: bonds still help, but they are not counted on to rally every time stocks fall. If the future looks more like 2000-2021, real results would tend to be somewhat better than the simulation; if it looks like the inflation years, the simulation is close to the mark.

Why It Matters Most Early in Retirement

The damage from a crash depends on what the rest of your savings does at the same time. In the first years of retirement, when you are withdrawing from a large balance, a crash where bonds hold up lets you spend from bonds while stocks recover. A crash where bonds fall too - like 2022 - leaves nothing safe to sell. That is sequence-of-returns risk at its worst.

What This Means for Your Plan

Don't count on bonds always rallying. They did in 2008, but not in 2022 or through much of the 1970s. A plan that only works if bonds cushion every crash is more fragile than it looks.

Correlation and fat tails solve different problems. Fat tails are about how often extreme years happen; correlation is about whether your investments have those years at the same time. A realistic simulation needs both.

Hold things that behave differently. The point of mixing investments is that their bad years don't all line up. History shows that works most of the time - just not every time.