Rolling Returns
Return measured across many overlapping periods instead of one — it kills the cherry-picked start date that flatters point-to-point numbers.
Why you care
A single "5-year return" depends entirely on the two dates you picked. Start just before a rally and any fund looks brilliant; start just before a crash and the same fund looks poor. Rolling returns fix this by computing the return over every possible window in a period, so no single lucky start date can flatter the number.
Run the numbers
Instead of one 3-year return, you compute the 3-year CAGR starting from every day (or month) over the last 10 years — hundreds of overlapping windows (illustrative). Now you can say "in 85% of 3-year windows the fund returned over 10%," which is a claim about consistency, not one convenient date.
Where this goes
Why you care
Rolling returns measure a fund's return over a fixed holding length (say 3 or 5 years) starting from many different points in time, then look at the whole distribution rather than one figure. Where a point-to-point return picks a single start and end date, rolling returns slide that window across the entire history. They ask "what would an investor have earned no matter when they happened to start."
The problem they solve is real and widely exploited. Trailing returns — "3-year return", "5-year return" — are quoted from today backwards, so they're hostage to exactly where the market sits right now and where it sat on that one start date. A fund can be made to look excellent or ordinary purely by choosing when the clock starts, and marketing does choose. Rolling returns strip out that luck. By averaging across every start date, they answer the more useful questions. How consistent was the fund, how bad was its worst window, how often did it clear a return you'd actually be happy with.
Run the numbers
Suppose you want to judge a fund on 3-year performance over the last 10 years (illustrative). Instead of one number:
- Compute the 3-year CAGR for the window starting each month: Jan 2016–Jan 2019, Feb 2016–Feb 2019, and so on. That's roughly 85 overlapping 3-year windows.
- Now summarise the distribution: the median 3-year return, the worst 3-year window, the best, and the share of windows above some threshold.
You might conclude "across all 3-year windows, this fund's median return was 11%, its worst was +2%, and it beat 10% in 85% of them." That is a far more honest description than "the fund's 3-year return is 14%," which might just mean the last three years happened to be kind. The worst-window number is especially useful, because it's the closest thing to "what could realistically have gone wrong for someone with my holding period."
Where this goes
Rolling returns are mechanically just CAGR repeated across overlapping windows, so they inherit its meaning while removing its date-sensitivity. Their real power shows when they're placed next to the benchmark's rolling returns over the same windows. A fund that beats its benchmark in 8 windows out of 10 is telling a trustworthy story. A fund that won by a mile in a single lucky stretch, but trailed the rest of the time, is telling a different one.