Why Good Returns Do Not Make a Portfolio Safe

The reasoning being tested

Open a statement, find a strong three- or five-year number, and a conclusion follows almost without being noticed: what I hold is sound. The return is evidence of quality, and quality is a kind of safety.

The step from the first to the second is the one nobody examines. A trailing return is a description of a period that has finished. Whether it says anything about the period that has not begun is a question with an answer, and the answer can be looked up rather than argued about.

We ran it two ways round, deliberately, because either test on its own would support a conclusion we do not think the evidence justifies.

Test one: what the number looked like before each fall

Take every fall of 20% or more in the record. For each one, go to the day of the high point — the last good day, before anything went wrong — and ask what the return over the previous 5 years was on that day. Then rank that figure against the same measure on every other day in the record.

High point, just before the fall	How far it then fell	Return over the previous 5 years	Out of 100 days, how many looked worse
11 February 2000	-50.2%	record too short	—
14 January 2004	-29.8%	record too short	—
10 May 2006	-29.7%	+29.33%	91
8 January 2008	-59.5%	+44.65%	100
5 November 2010	-27.2%	+22.25%	84
3 March 2015	-21.7%	+13.41%	48
14 January 2020	-38.3%	+9.75%	24

The last column is the ranking. A figure of 90 would mean that on only one day in ten did the trailing return look better than it did on the eve of that fall.

Of the 7 large falls, 5 have enough history behind them to be ranked at all. Ranked against every other day in the record, 3 of them looked better than a median day and 3 sat in its top quarter. The middle case ranked 84 out of 100. Repeating the whole exercise on a 3-year trailing window, which reaches back far enough to rank 6 of the falls, gives a similar picture — so the pattern is not an artefact of the window we happened to pick.

And then the exception, which matters more than the pattern does. The fall that began on 14 January 2020 went on to take 38.27% off the index, and the trailing return going into it was +9.75%, ranking 24 out of 100 — well below an ordinary day. A large fall did not need a good run in front of it.

Test two: the control, and why it is the important half

If we stopped there, a reader would take away something we would then have to spend the rest of the site correcting: that a good trailing return is a warning sign. So the second test asks the question in the direction a person would actually use it.

Sort every day in the record by the return over the previous 5 years. Split into four equal groups. Ask what the following twelve months did.

Show these numbers as a table

Strategy	Poor outcome	Typical	Good outcome
Worst quarter of past records	+5.13%	+20.51%	+55.83%
Second quarter	-8.78%	+14.13%	+38.24%
Third quarter	-2.15%	+10.14%	+37.46%
Best quarter of past records	-29.94%	+11.03%	+43.89%

The groups overlap almost completely, which is the first finding and the one that kills the warning-sign reading. But they are not identical, and the difference is entirely in the shape rather than the middle.

After the worst quarter of past records, the typical next year returned +20.51%, the worst single outcome was -10.50%, and 1.6% of those days were followed by a losing year. After the best quarter of past records, the typical next year returned +11.02% — less, not more — the worst outcome was -55.38%, and 30.9% of days were followed by a loss.

So the honest summary is this. A good trailing return did not predict a fall, and using it as a sell signal would have cost you money in most years. What it did do was sit in front of the widest range of outcomes in the record, including every one of the worst. The past being good did not make the future bad. It simply carried no information about safety in either direction, while feeling as though it carried a great deal.

Why this is what you should expect

None of it is mysterious once you separate two things a return statistic mixes together.

A return tells you the size of the move. It says nothing about how much had to be risked to produce it, and two portfolios with the same return can have arrived by wholly different routes — one steadily, one through a fall its holder very nearly did not sit through. The number is identical and the two holdings are not the same object.

There is also a mechanical reason a good run makes a portfolio riskier rather than safer, and it has nothing to do with the market's mood. Whatever rose fastest is now a larger share of what you own. A portfolio that has done well is, by arithmetic alone, more concentrated in the thing that did well than it was when you set it up — which is the entire argument for rebalancing, and the reason a strong year is the moment to check an allocation rather than the moment to relax about it.

And finally, a trailing return is measured over a period that has already been selected: it ends today, and today is a day you noticed the number because it was good. The distinction between a period chosen and a period not chosen is worked through in rolling against point-to-point returns.

What this test cannot tell you

5 ranked falls is not a sample. The backward test rests on the handful of large falls in one market's record, and the earliest of them are too early to rank at all. It can establish that the days before a crash have often looked good. It cannot put a probability on anything, and none is quoted.

The days overlap almost entirely. The forward test reports thousands of days, and consecutive days share nearly all of their trailing and forward windows. It is far less evidence than the counts imply, which is why the finding is stated as a shape and not as a rate.

It is one country over one stretch, on a total-return index, before costs and before tax.

It says nothing about individual funds or shares. Every figure here describes the index. A test of whether a particular fund's good record predicts its future one needs fund-level history including the funds that closed, which we do not hold — the subject of does past downside protection predict future protection.

What to take away

A good trailing return is worth having and is not evidence of safety. It describes a period that has ended, and on this record it has been just as present before the worst falls as before the ordinary years.

The things that do bear on how safe a portfolio is can all be checked without reference to the return: how much of it is in shares, how concentrated it has become since you last looked, and whether the money you will need soon is somewhere a fall cannot reach. None of those questions is answered by the number at the top of the statement, and all three are worth more than it is.

If a strong run prompts anything, let it be the annual review rather than a decision that nothing needs doing.