Recency bias meets factor crowding

Thursday night. Friend texts a screenshot of a model basket up 60 percent YTD. "Should I move my SIP into this."

Asked him one thing back. What made it do 60.

The basket ranks its picks by trailing 12 month return. That is a momentum tilt with a friendly name. The screen that picked it is the screen every other investor in the same group chat is reading on the same evening. Once you can see the engine, the question stops being whether to chase the 60 percent and starts being whether you have done this before without realising.

The factor has done it three times in eighteen years. Mar 2009 post GFC junk rally minus 28 percent. Mar 2020 COVID rotation minus 21 percent. Feb 2025 crowding unwind minus 19 percent. A 60 percent year is one tail of a distribution that also contains a minus 28. The brain treats it as the next year. The market treats it as the last one.

The mechanic underneath has a name. McLean and Pontiff 2016 ran 97 published anomalies out of sample after the academic paper landed. Median return fell 26 percent. Once a factor has a label, more capital runs it, more capital narrows the spread, and the SIP that arrived after the win year shows up in time to be the last buyer. Sirri and Tufano 1998 documented the flow asymmetry. Winners attract roughly five times the inflows per point of out performance vs losers shedding the same.

Three checks before the click. Name the factor, because if the screen has a label the crowd has one too. Read the rolling worst 12M window, not the loudest year, because the mean tells you the prize and the tail tells you the toll. SIP into the strategy you can hold while the chart goes red for nine straight months. The cancelled SIP earns the drawdown without the recovery.

The brain reads last year's return. The market is already reading next year's reversion.

What recency bias does to a return table

Recency bias is the habit of treating the latest observation as the most likely next one. A trailing 12 month return is the purest recency number there is: it is one window, the most recent, presented as if it described the strategy. The checks in the opening all say the same thing in different words. Look at every window, not the last one.

So I ran that check on our own model strategies, because I would rather show the problem on numbers I publish than on someone else's. For each strategy I computed every rolling 12 month return in the back test, month by month, then picked out the best window, the worst window, and what happened in the 12 months right after the best one.

The Smallcap model's best 12 months ended in Oct 2023 at plus 161.35 percent. Anyone who saw that number and arrived then got plus 4.32 percent over the next 12 months. The Focused Largecap 20 model's best window ended a month later at plus 119.63 percent, and the next 12 months were minus 8.36 percent. Neither strategy broke. The screenshot simply showed the tail of the distribution, and the next year came from the middle of it.

Read the toll, not the prize

Rule two in the opening has a concrete version. Before the best year, ask for the worst rolling window and the maximum drawdown. In the same back test, the Smallcap model had 7 negative rolling 12 month windows out of 51 and a maximum drawdown of 22.34 percent. Focused Largecap 20 had 11 negative windows out of 49 and a maximum drawdown of 23.56 percent. Those are the numbers that decide whether you will still be holding when the next good year arrives.

Why the crowd always arrives late

The flow asymmetry in the opening is the engine of the crowding problem. Money chases a good trailing number much harder than it leaves a bad one. So the largest amount of money usually sits in a strategy just after its best year, which is exactly when the next year is most likely to be ordinary. The fund's own return, measured from one rupee on day one, can look excellent while the average rupee in it does much worse. The post on herding and social proof works that gap out in rupees.

The three checks, in order

1. Name the factorRanked on trailing return means momentum, whatever the label says.
2. Read the worst windowWorst rolling 12 months and maximum drawdown, side by side with the best.
3. Size for the red stretchPick an amount you would keep investing through months of losses.

The drawdown recovery calculator shows how long a given fall takes to climb back from at a given return. The free Learn module on overconfidence, anchoring and recency bias covers the research behind the checks, with a quiz at the end.