What a trailing return actually measures
A trailing return is the annualised return over a period ending today. Three-year trailing, five-year trailing — one window, one end date, one number.
The end date is doing far more work than it appears to. Consider a scheme that was ordinary for thirty months and then ran hard for six. Its three-year trailing figure will look excellent, and almost all of it was earned in the final fifth of the period. Compute the same figure six months earlier and the fund looks unremarkable.
Nothing about the fund changed between those two calculations. Only the date you happened to run them on. That sensitivity has a name — end-point bias — and it is not a minor caveat: it is the dominant influence on every trailing figure you will be shown.
What rolling returns do instead
Rather than one three-year window, compute every three-year window available in the history. Start in January 2015 and measure to January 2018. Start in February 2015 and measure to February 2018. Continue to the present.
The result is not a number. It is a distribution — hundreds of three-year outcomes, one for every start date an investor could actually have chosen. No single end date can dominate it, because every end date is in it.
That reframes the question usefully. Instead of “what did this fund return?” you are asking “across every three-year period an investor might have lived through, what happened?” — which is the question a person deciding today is actually facing, since they do not know which window they are about to get.
Which figures in the distribution matter
Most presentations of rolling returns lead with the average. The average is the least informative number in the set.
| Figure | What it tells you | Weight |
|---|---|---|
| Worst window | The outcome of the unluckiest possible entry date | Highest — this is your realistic downside |
| Share of windows below zero | How often this holding period failed to make money at all | High |
| Median | The typical outcome, unaffected by a few extreme windows | High |
| Spread between best and worst | How much of the outcome was down to timing rather than the fund | Medium |
| Average | Little that the median does not say better | Low |
The reason for that ordering: a wide distribution means your result depended heavily on a start date you did not choose deliberately. A narrow one means the fund delivered something similar whenever you arrived. Two funds with identical averages and different spreads are not comparable investments, and only the distribution shows it.
Why this matters most for SIP investors
A lumpsum investor experiences exactly one start date. A SIP investor experiences a new one every month.
So a SIP is, structurally, a bet on the whole distribution rather than on one draw from it. The relevant question is not “what did this fund return over the last three years” but “across all the entry points I am about to accumulate, what has the spread of outcomes looked like”.
That makes rolling returns the natural lens for the way most Indian investors actually invest, and makes a single trailing figure close to irrelevant to them — a point developed further in SIP or lumpsum?
The trap inside rolling returns
Rolling windows overlap, and this is where careful analysis goes wrong.
The three-year window starting in January 2015 and the one starting in February 2015 share thirty-five of their thirty-six months. They are not two independent observations of the fund. They are very nearly the same observation, counted twice.
Two consequences worth holding on to. Do not treat the count of windows as a sample size — a decade of monthly three-year windows gives you roughly eighty-odd overlapping windows but only about three genuinely independent three-year periods. And be suspicious of any statistical claim built on overlapping windows: standard deviations, confidence intervals and information ratios computed this way are systematically overstated, because the arithmetic assumes an independence the data does not have.
None of that makes rolling returns less useful. It makes them a description of history rather than a statistical estimate — which is what you wanted from them anyway.
What rolling returns still cannot tell you
Three honest limits.
They are history. A consistent past is not a promise. Fund managers change, mandates drift, and the market regime that produced a decade of results does not renew itself on request.
They do not account for the fund's own changes. A ten-year rolling study of a scheme that changed manager and mandate five years ago is measuring two different funds and reporting one distribution.
They say nothing about why. A narrow distribution could reflect genuine process discipline or a mandate so constrained that the manager had no room to differ from the index. The distribution cannot distinguish those, and the fee question turns on which one it is.
Running the distribution yourself
This is arithmetic on a public NAV series, so it should not require trusting anybody's summary of it.
FNOTrader's Mutual Funds app computes rolling returns across the full AMFI history — around 34 million NAV rows, refreshed nightly at 22:30 IST — and reports the median, the worst window and the share of negative windows rather than leading with the average, alongside the scheme's performance against its own benchmark over the same windows. Formulas are published in the user guide.
Common questions
What are rolling returns?
The return computed from every possible start date across a history, rather than from one window ending today. Instead of a single three-year number you get hundreds of three-year outcomes — one for each start date an investor could actually have chosen.
Why are trailing returns misleading?
Because they depend heavily on where the window happens to end. A fund that was ordinary for thirty months and then ran hard for six will show an excellent three-year trailing figure, and moving the end date back six months makes the same fund look unremarkable.
What should I look at in a rolling return distribution?
The worst window first — that is your realistic downside — then the share of windows that lost money, then the median. The average is the least informative figure in the set, and the spread between best and worst tells you how much of the outcome was timing rather than the fund.
Why do rolling returns matter more for SIP investors?
Because a SIP accumulates a new entry date every month, so it is exposed to the whole distribution of outcomes rather than to one draw from it. A single trailing figure describes one start date, which is not the situation a SIP investor is in.
Can I treat each rolling window as a separate data point?
No. Consecutive three-year windows share thirty-five of their thirty-six months, so they are nearly the same observation counted repeatedly. A decade of monthly windows yields many overlapping windows but only about three genuinely independent three-year periods.
Are statistics computed on rolling windows reliable?
Standard deviations, confidence intervals and ratios built on overlapping windows are systematically overstated, because the arithmetic assumes an independence the overlapping data does not have. Rolling returns are best read as a description of history rather than as a statistical estimate.
Do good rolling returns mean a fund will perform well?
No. They describe what happened, not what will. They also cannot detect that a scheme changed manager or mandate midway, in which case the distribution is blending two different funds into one picture.
Continue reading
More in Mutual Funds · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article