Rolling Returns Show What a Single CAGR Hides

A “5-year return” quoted today is one number, produced by exactly one start date and one end date. Move either date and the number changes - sometimes sharply. A rolling return asks a different question: instead of one window, recompute the same 5-year CAGR starting from every possible date in the data, and report the full spread. It doesn’t answer “what was the return” - it answers what the return could have been, depending only on which year an investor happened to start.

Run that calculation on the Nifty 50’s 35-year history (1991-2026, BSE/NSE data) at every standard holding period, and the spread does something specific: it doesn’t just shrink, it collapses toward a fairly stable centre.

Nifty 50 rolling CAGR by holding period, 1991-2026. Minimum, maximum and average annualised return across every rolling window of each length; the 35-year figure is a single fully elapsed window, not a range. Source: BSE/NSE historical data via BMS Money.

• 1-year holds: ranged from -47.1% to +134.1%, averaging 17.4% - 9 of 35 rolling years were negative.
• 5-year holds: -6.2% to +37.5%, averaging 11.2% - only 1 of 31 rolling windows was negative.
• 7-year holds: 0.0% to +24.5% - zero of 29 rolling windows were negative.
• 10-year holds: never fell below +2.5%, averaging 11.3%.

The popular version of this finding is “stay invested longer and you earn more.” The data doesn’t actually say that. Look at the average column: it runs 17.4% at 1 year, then 12.2%, 11.2%, 11.2%, 11.3%, 11.8% - essentially flat from year 3 onward. The average return doesn’t keep climbing with time. If anything, the 1-year average is the highest number in the table, inflated by outsized single years (+134.1% in one window) that get diluted the moment they’re compounded into a longer holding period.

What actually moves with time is the floor, not the average. The minimum recorded return rises from -47.1% at 1 year to -6.2% at 5 years to a floor that never dips below zero from year 7 onward. Holding longer hasn’t made this index return more, on average - it has made the worst plausible outcome dramatically less bad. That is a narrower, more specific claim than “the power of long-term investing,” and it’s the actual mechanism a rolling return table is built to show.

None of this guarantees the next 35 years will resemble the last 35 - one market’s history is still a single sample, however long. It does mean that a single trailing CAGR, quoted as of one date, can never show this. Only recomputing the same window across every date the data allows can.