Rolling returns — how much the answer depends on when you started

For a window of N years the tool compounds each run of N consecutive calendar returns and takes the Nth root, plotting the result at the window's final year. The shortest window offered is three, because a one-year window would simply replay the annual returns table and a two-year one is barely a window at all. Every figure comes from the shared annual-series helper rather than being recomputed here, which is deliberate: it is the one place the annualisation factor is pinned by tests.

The windows overlap heavily, so neighbouring points share most of their underlying years and the line is far smoother than independent observations would be — do not read a run of similar values as confirmation of anything. With roughly a decade of published data a five-year window produces only a handful of observations, which is too few to describe a distribution, and the page reports the count rather than implying a dense series. Currency, tax and trading costs are all outside the figures.

Why only three and five year windows?

Because of how much data actually exists. Issuer-published calendar returns go back about a decade for most of these funds, so a five-year window already yields only about six overlapping observations and a ten-year window would yield one or two — a number, not a distribution. A one-year window is excluded for the opposite reason: it would just be the annual returns table with extra steps.

What should I actually take from the spread?

That the range is wide enough to swamp most fund-selection decisions. If the same fund annualises at very different rates across windows, then a comparison between two funds measured over one particular window is telling you as much about the window as about the funds. It is an argument for choosing on cost and structure, which are knowable in advance, rather than on a past return that moves when you shift the start date.

Why is the line smoother than the annual returns?

Because each point averages several years and consecutive points share most of those years. A three-year window moving forward by one year keeps two of its three inputs, so the series cannot jump the way single-year returns do. The smoothness is an artefact of the overlap rather than evidence that returns became more stable.

• Annual returns calendar — which fund led which year
• Portfolio backtest — what a mix actually returned
• Sequence of returns risk — the same returns in a different order