Rolling Returns Against Point-to-Point Returns
A point-to-point return is a single number produced by two dates somebody chose. A rolling return is what happened over every period of that length. Run both over the same Indian data and the headline figure turns out to be one draw from a distribution wide enough that two people doing the same thing for the same ten years ended up more than four times apart.
What is the difference between rolling and point-to-point returns?
A point-to-point return takes a start date and an end date, compares the value at each, and expresses the difference as a rate per year. Almost every return you are ever quoted is one of these. Our own record produces +13.21% a year measured this way, from 30 June 1999 to 31 August 2026.
A rolling return does the same calculation over and over, starting a new period on every trading day and following each to its end. Instead of one number it produces thousands, and what you get is not a figure but a distribution.
The two are not competing estimates of the same quantity. The point-to-point number is a member of the set the rolling calculation produces. The question is never which is right — it is what a single member of a wide set can tell you.
How widely do rolling returns vary?
| If you held for | Periods tested | Worst | Poor | Typical | Good | Best | Lost money |
|---|---|---|---|---|---|---|---|
| 1 year | 6,511 | -55.31% | -12.75% | +12.41% | +48.97% | +107.61% | 23.6% |
| 3 years | 6,016 | -15.23% | +2.40% | +13.40% | +34.34% | +61.68% | 6.3% |
| 5 years | 5,519 | -1.03% | +6.31% | +13.60% | +27.36% | +47.64% | 0.1% |
| 7 years | 5,023 | +4.90% | +9.27% | +13.71% | +24.17% | +30.47% | 0.0% |
| 10 years | 4,282 | +5.13% | +9.43% | +13.87% | +19.04% | +22.27% | 0.0% |
| 15 years | 3,043 | +8.59% | +11.14% | +13.46% | +17.50% | +19.37% | 0.0% |
| 20 years | 1,804 | +9.68% | +12.03% | +14.78% | +16.96% | +17.99% | 0.0% |
Take the ten-year row. Someone who held for ten years got somewhere between +5.13% and +22.27% a year, depending on nothing except the day they happened to start.
As money that is: ₹100 left alone for ten years became ₹165 in the worst period and ₹747 in the best. Two people who did exactly the same thing, for exactly as long, in exactly the same index, ended more than four times apart. Neither of them made a decision the other did not.
The middle case turned ₹100 into ₹367, which is a perfectly reasonable outcome and describes neither of them.
Where does a headline return sit in the range?
Show these numbers as a table
| Return a year, over every 10-year period in the record | Number of periods | Share |
|---|---|---|
| +5.1% to +6.0% | 7 | 0.2% |
| +6.0% to +6.9% | 41 | 1.0% |
| +6.9% to +7.8% | 88 | 2.1% |
| +7.8% to +8.7% | 127 | 3.0% |
| +8.7% to +9.6% | 240 | 5.6% |
| +9.6% to +10.5% | 289 | 6.7% |
| +10.5% to +11.4% | 234 | 5.5% |
| +11.4% to +12.4% | 285 | 6.7% |
| +12.4% to +13.3% | 445 | 10.4% |
| +13.3% to +14.2% | 551 | 12.9% |
| +14.2% to +15.1% | 460 | 10.7% |
| +15.1% to +16.0% | 334 | 7.8% |
| +16.0% to +16.9% | 223 | 5.2% |
| +16.9% to +17.8% | 171 | 4.0% |
| +17.8% to +18.7% | 227 | 5.3% |
| +18.7% to +19.6% | 242 | 5.7% |
| +19.6% to +20.5% | 165 | 3.9% |
| +20.5% to +21.4% | 112 | 2.6% |
| +21.4% to +22.3% | 41 | 1.0% |
The dashed line is +13.21% — the single figure our whole 27.2-year record produces, and the one a headline would quote. It is a different kind of number from the bars above it: those are annualised returns over ten-year holds, many of them, while the whole-record figure comes from a span that occurs exactly once and cannot itself be one of the ten-year draws. What can still be asked is where a return like it would land if the same history were instead cut into ten-year slices — and the answer is about the 40.5% percentile, meaning a majority of the ten-year periods in the very same data annualised slightly better than the number that supposedly summarises the whole of it.
That is worth pausing on, because the natural assumption is that a long-record figure is a good average of the experiences inside it. It is not an average of anything. It is what you get from two particular dates: the first day we have data for and the last. Move either and the figure moves.
Why does the start date of a return figure matter?
Once you see the spread, the way point-to-point figures are usually presented starts to look different.
Nobody has to lie to produce a flattering number. They only have to choose the period, and every period is a choice — the last five years, since inception, since the strategy changed. Each is defensible, each produces a different figure, and only one of them will be printed. A reader shown the winning choice cannot tell it apart from a figure that was not chosen at all.
This is the reason a rolling table is the harder thing to publish and the more useful thing to read. A distribution cannot be improved by choosing where to start it, because it starts everywhere. That property is worth more than any individual number in it.
How do you read a rolling returns table?
Three cautions, all of which apply to the table above.
The periods overlap, heavily. A ten-year row built from every trading day contains thousands of periods but only a couple of genuinely independent ones, because a period beginning today and one beginning tomorrow share almost all their history. The "periods tested" column counts windows, not evidence, and treating it as a sample size makes the record look far stronger than it is.
The range narrows with length, and the narrowing is real but limited. It also narrows at both ends: the spectacular outcomes disappear along with the disastrous ones. What barely moves as the holding period lengthens is the typical result, which is the finding worked through in why long-term equity returns remain uncertain.
And a rolling table is still one country over one stretch. Every period in it is drawn from the same 27.2 years, so it cannot tell you anything about a market whose history looked different from India's.
How do you use rolling returns when judging a fund?
When you are quoted a return, ask what the two dates were. If the answer is a period somebody selected, treat the number as one observation rather than as the answer, and ask what the range looked like.
When you are planning, do not plan against the middle of the distribution as though it were guaranteed. The reason to hold a goal's money in something safer as its date approaches is precisely that the range at short horizons is wide — set out in the glide path to a goal date.
And when you see a rolling table, the honest reading is the whole row, not the "typical" column. The typical column is a point-to-point figure again, wearing a better hat.
Educational content only. This is not personalised financial, investment or tax advice. Figures quoted are historical or illustrative and are not forecasts. Consult a qualified professional before acting on anything you read here.