What 5,000 rolling 3-year periods say about discipline.

The case for staying invested in Indian equities usually arrives as a number. Twelve per cent compounded. Fifteen, if the seller is keen. The number is true, in a narrow accounting sense. What it conceals is more useful than what it states. The historical mean is the average outcome across all the experiences a real investor could have had — not a forecast of the experience they will have, nor a description of any one path through the data. The mean is a summary. The distribution is the truth.

Re-tile the NIFTY 500 Total Return record into rolling 3-year windows — start a fresh window every trading day from 1 April 2003 and let it run until the present — and you generate roughly 5,000 overlapping 3-year experiences. Each one is what a different investor would have lived through, depending only on when they happened to start. Read those 5,000 experiences together and a picture appears that the single mean cannot draw.

What the distribution actually looks like

Of the rolling 3-year windows that close before this writing, the median annualised return is in the mid-double digits — close to where the headline mean sits. So far so headline-shaped. The interesting part is the spread. The 90th percentile window earned more than twice the 10th percentile window. The 10th percentile window, in turn, was negative.

Five thousand windows is a lot of windows. Roughly 11–14 per cent of them end negative. Roughly the same share end above 30 per cent annualised. The difference between an investor who picked a good start date and one who picked a bad one was, in the worst pairing, more than 40 percentage points per year — for the same fund, with the same manager, holding the same securities for the same length of time.

This is the part the headline mean conceals. The mean treats every window as equally likely. In a real investor’s life, only one window happens. The window that happens is determined by when capital arrives, which is itself determined by when income arrives, when liabilities clear, when the family event that triggered the lump sum landed. The discipline question is: what do you do across the window you happened to draw?

The discipline penalty, visible

Now layer in switching behaviour. Filter the same 5,000 windows for those that began near a local high — the moments the news cycle was loudest and an investor would have felt most tempted to delay. The median window beginning at one of those tops earned roughly six percentage points per year less than the median window across the full sample. Six percentage points compounded over three years is a 19 per cent gap on the same starting capital.

The same exercise, run for windows that began after sharp 20 per cent corrections (the moments the news cycle says “sit out for now”), produces the inverse: the median window earned roughly five percentage points per year more than the full-sample median. Same fund, same manager, same securities. Different start.

These are population statistics, not investment instructions. No one can reliably identify a top or a bottom in advance, and starting capital usually doesn’t wait for one anyway. The point is narrower and more uncomfortable: the dispersion across start dates is wide enough that the conventional “just stay invested for the long run” framing under-states the cost of letting the news cycle decide when you arrive and when you leave.

The dispersion across start dates is wide enough that the conventional ‘just stay invested’ framing under-states the cost of letting the news cycle decide when you arrive and when you leave.

What this implies for an investor who reads this

Three things, none of them clever.

First, the average return is not your return. Your return is one window. The honest planning question is not “what is the average?” but “what is the 10th-percentile window, given my likely start, and can I tolerate that?”

Second, the cost of switching out of a window mid-flight is not the broker fee. The cost is the part of the window you miss. In the 5,000-window distribution above, switching at the wrong moment converts a typical window into a near-bottom window. The probability of switching at the wrong moment, conditional on the news cycle being loud, is high.

Third, the antidote to the dispersion is not better timing — nobody has that — it is a longer holding period that lets the next window absorb the previous one. The 7-year and 10-year rolling distributions, run on the same data, are dramatically tighter than the 3-year distribution. Holding longer does not change the average. It changes the spread.

That is the closest thing to a strategic implication you can pull from this exercise. It is not a fund recommendation. It is not an asset-allocation prescription. It is an observation about the dispersion of outcomes in a public benchmark, computed from public data, repeatable by anyone with access to AMFI’s daily NAV record and a spreadsheet that can roll a window.

The point of writing it down is the third item above. The investor who internalises that holding longer compresses the spread will, in our experience, be a steadier holder of any fund they pick. Not because they trust the manager more. Because they have looked at the distribution and understood that the news will look loud in some windows and quiet in others, and that the loudness is largely uncorrelated with their personal outcome.

A note on what this isn’t

This analysis is descriptive. It uses public benchmark data; it makes no statement about any specific scheme; it does not constitute investment advice. Past distributions are not promises about future ones. Indian equity markets have evolved meaningfully across the 2003–2026 window — deeper liquidity, broader category mix, different macro regimes — and the rolling-period distribution will continue to evolve. We re-run this exercise quarterly and will revise the piece if the tails change shape materially.

The single useful instruction this piece supports is: when you read a headline number, ask for the distribution underneath it. We will keep publishing those distributions.