Why Past Performance Is Not Enough to Choose a Fund

The claim being made

Almost every fund is sold on its record. The figures are accurate, the disclaimer is present, and the implied argument is nonetheless that a fund which did well is a fund that will do well.

That is a claim about prediction, and prediction is testable. The question is not whether the past returns were real — they were — but whether knowing them tells you anything useful about the returns you have not yet received.

This page sets out what a track record actually contains, the four specific reasons a published one reads better than the reality it came from, and what evidence would settle the question. The section near the end says why we cannot supply that evidence, and what to do instead.

What a track record is a record of

A fund's return over a period is the outcome of three things mixed together, and the number does not say how much came from each.

Some of it is what the market did. A fund holding Indian shares through a strong decade returns a great deal without anyone at the fund doing anything clever. Compare it against the market rather than against zero and much of the apparent achievement disappears.

Some of it is what kind of thing the fund holds. A fund concentrated in whichever part of the market happened to do best will lead its category, and this is a fact about its mandate rather than about its manager. It also means the same fund will trail badly when that part of the market is out of favour, without anybody having changed their mind about anything.

And some of it is luck. This is the part people find hardest to accept, and the easiest to demonstrate: among enough funds, some will have excellent records for no reason at all. The arithmetic of that is worked through in data mining and backtest overfitting, and the conclusion transfers directly. A record is evidence of skill only to the extent that it is longer, and more consistent, than luck would have produced across the number of funds you were choosing from.

Separating those three is the whole difficulty. Ranking funds by return does none of it.

Four reasons the published record flatters

Beyond the mixing problem, the record you are shown has usually been shaped. None of the following requires anyone to have done anything improper.

The funds that failed are not in the list. Funds that perform badly are closed or merged into better-performing ones, and their records generally go with them. What remains is a set of survivors, and the average of survivors is not the average of what was available to choose from at the time. This is the single largest distortion in fund data, and it is invisible in any published table because the missing entries are missing.

The period was chosen. Since inception, the last five years, since the current manager arrived — each is defensible and each produces a different figure, and you are shown one of them. Our page on rolling against point-to-point returns shows how far apart those choices can land on the same underlying data.

The comparison was chosen too. A fund compared against a benchmark that does not match what it holds will look skilful whenever its actual holdings are in favour. The check is whether the benchmark is one the fund would still be measured against in a year when the comparison flattered it less.

The fund is not the same fund. Managers change. Mandates change. Most of all, size changes: a strategy that worked with a small amount of money often cannot be run with a large amount, because the trades that produced the returns move the price once they are big enough. A record earned small is not a promise that can be kept large.

The test that would settle it

The design is not complicated and it is rarely run.

1. Rank the funds on one period. Then measure them over a later, separate one. Sorting funds by past return and then reporting their past return proves nothing at all, and it is astonishing how much published analysis amounts to exactly that.
2. Include every fund that existed at the start of the ranking period, including the ones that closed or were merged before the end. If they are dropped, the test measures survivors and returns a flattering answer to a question nobody asked.
3. Compare like with like. Otherwise the exercise rediscovers that funds holding whatever did well, did well.
4. Repeat it across several separate periods, because one repetition of the ranking could happen by chance in any set of funds.
5. Decide the measure before looking. Return, risk-adjusted return, consistency, the worst year — they will not agree, and picking the one that supports the conclusion afterwards is the same error in a different place.

The honest gap

Running that test needs the complete history of every fund in a category, including every fund that no longer exists, across several market cycles.

This site does not have that data. We hold index history and inflation history; we hold no fund-level records at all. So this page states no conclusion about whether Indian funds' past returns predict their future ones. Where our other articles carry computed tables, this one deliberately carries none — because the only table we could build from an index would be a table about the index, and presenting it as evidence about funds would be the error this page exists to warn about.

That absence is worth noticing for its own sake. The data needed to check the claim on which the entire fund industry is sold is not something an ordinary investor can obtain.

What to use instead

The useful response is not to ignore the past but to weight it against things that are knowable in advance rather than only in retrospect.

Cost is the one property of a fund you know before you buy and which is nearly certain to persist. It is subtracted every year regardless of performance, and the arithmetic of that subtraction over a long holding period is not small — set out in how investment fees reduce wealth.

Mandate tells you what the fund is obliged to hold, which determines most of what it will do. A fund that must hold a particular slice of the market will follow that slice, and knowing which slice you are buying is worth more than knowing how the slice behaved last year.

Structure — whether returns depend on one person's continuing judgement, how large the fund has become relative to what it invests in, and how much it trades. These are checkable and they bear directly on whether a past record could be repeated even in principle.

And where you do look at a record, look at the whole shape of it rather than the headline: how it behaved in the worst stretch, and whether its good years came from the mandate or from something the manager decided. That is the distinction why good returns do not make a portfolio safe takes apart in detail on index data, where we can actually show our working.

What to take away

The disclaimer is not a legal formality. Past performance is insufficient because a return figure mixes market, mandate and luck without labelling them, because the record shown to you has been selected in at least four ways, and because the test that would settle the matter needs data that is not generally available.

Choose on the things that are true in advance — what it costs, what it must hold, and whether its result depends on somebody staying and staying right. Then, if a fund's record still matters to you, ask the seller which period they chose and what happened to the funds that are no longer in the table.