Rolling Returns — Removing Entry-Timing Bias

Statements like "if you'd invested 10 years ago, it would have doubled" are common. But this is a value calculated by picking exactly one start date and one end date. This is called a 'point-to-point return' or a 'trailing return.'

The problem is that if that start date happened to be the market bottom, the return looks wonderful, whereas if you started at a peak, it looks dismal. It's the same asset, yet the number swings wildly on the single coincidence of 'when you pressed the button.'

This phenomenon, where a particular entry point distorts the result, is called 'timing bias' (point-in-time bias). It's also why ads and testimonials tend to show only especially good numbers.

Rolling returns are a method for removing this timing bias. The idea is simple: "Instead of looking at just one window, calculate every possible start point and lay them all out."

For example, if you look at '10-year rolling returns,' you calculate every overlapping 10-year window—2000–2010, 2001–2011, 2002–2012, and so on—shifting the start date one step at a time. That produces hundreds of '10-year holding report cards.'

Now we see not a single number but a 'distribution.' At a glance you can see what the worst 10 years looked like, what the best 10 years looked like, and what most of them were. If a trailing return is a single snapshot, a rolling return is the whole album shot from many angles.

When you actually compute rolling returns, an interesting pattern emerges: the longer the holding period, the more dramatically the 'gap between best and worst' shrinks.

Using historical data for the U.S. S&P 500 to sketch a rough picture, on an annualized basis a 1-year holding ranges extremely widely—from a worst of about -37 to -44% to a best in the +50% range. A 5-year holding narrows to roughly -12% to +28%, and a 10-year holding to roughly -4% to +20%.

What especially stands out is the 20-year horizon. Historically, several sources note that every 20-year rolling window of the S&P 500 produced a positive annualized return (on a nominal, dividend-included basis). Even the worst 20 years came to roughly +1 to 6% per year, and the best narrowed to around +13 to 17% per year. In short, over short horizons, entry-timing luck greatly determines the result, but over long horizons, the influence of that luck fades.

The real value of rolling returns is not in boasting a 'good average' but in 'not hiding the bad windows.'

There clearly are negative windows in 1-year and 5-year rolls. That means someone who entered at a peak had to endure years below their principal. For example, a 10-year window starting near the 1999–2000 peak hit the dot-com collapse and the financial crisis back to back, and is often cited as a window whose annualized return was close to 0% or negative.

So when you look at rolling returns, you must look not only at 'what the average % is' but also at 'how bad the worst window was, and how long you had to endure losses during it.' Only when you check this loss duration and the maximum drawdown together do you get closer to the real investor experience.

If someone says "this asset returned 15% a year," ask them back: "Is that just one window from a start date to an end date, or a rolling return that looks at many start points?"

Boasting about a single trailing window may hide timing bias. Conversely, if they show even the worst-to-best range of the 5-, 10-, and 20-year rolls, that's far more honest information.

'The Return of Almost Everything' is a site operated by the author, built so you can try this kind of 'timing-bias-free comparison' yourself. Change the start date, compare lump-sum and recurring investing, and see even the loss duration during crash periods with your own eyes. The habit of looking at 'the experience across many start points' rather than a single number is what gives you the strength to hold good assets for the long term.