Recency Bias

Recency bias is a systematic tendency to over-weight recent observations when forming an expectation about the future, so a forecast comes to look like a continuation of the recent past. In investing it is the mechanism behind performance chasing: money flows toward whatever has done well lately and away from whatever has not, on the implicit assumption that the recent pattern carries information about the next period. A note on the name, because it collides with a different term. In cognitive psychology "recency effect" means something narrower, the tendency to recall the last items in a list better than the middle ones. The finance sense described here is about forecasting, not about memory for lists, and the two should not be conflated.

The best evidence for recency bias in financial markets comes from the markets themselves rather than from experiments. Robin Greenwood and Andrei Shleifer, in "Expectations of Returns and Expected Returns" in the Review of Financial Studies 27(3), analyzed six separately collected time series of investor expectations of future stock market returns spanning 1963 to 2011. In the working-paper version of that study the authors summarize the result this way: the six measures were "highly positively correlated with each other, as well as with past stock returns and with the level of the stock market," and, critically, "investor expectations are strongly negatively correlated with model-based expected returns." In plain terms: when investors were most optimistic, the models that attempt to estimate future returns from valuations were at their least optimistic, and vice versa. That is about as clean a statement as the literature offers that expectations extrapolate rather than anticipate.

A complementary result concerns how far back the relevant experience runs. Ulrike Malmendier and Stefan Nagel, in "Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?" in the Quarterly Journal of Economics 126(1), found that the market conditions people have personally lived through shape their willingness to take financial risk for decades afterward. Recency bias, on that account, is not only about last quarter. A cohort that entered adulthood during a long bear market carries it, and one that entered during a long bull market carries the opposite.

One number that circulates in this territory should be treated with care. The gap between the return a fund reports and the return the average dollar in it actually earned is real and regularly measured, and it is often presented as a measurement of what recency bias costs investors. How much of that gap is caused by mistimed buying and selling, as opposed to arising mechanically from when money happened to flow in, is genuinely disputed in the professional literature. The existence of a gap between fund returns and dollar-weighted investor returns is well established. A specific figure for the cost of bad timing is not, and this page prints none.

The mechanism is an availability shortcut applied to a forecasting problem. Asked what returns to expect, a person reaches for the most accessible sample of returns, which is the recent one, and treats it as representative of the distribution. It usually is not, because financial returns are noisy over short windows and the recent stretch is a small and non-random slice.

Four places it shows up in a household's decisions:

• Performance chasing between funds. Selecting a fund on its trailing three-year number, which is by construction the funds that just did well, and rotating again when a different one takes the lead.
• Drift in the target allocation. After a long run in one asset class, a portfolio that was set at a deliberate mix now holds far more of it, and the run makes that feel like a decision rather than a drift.
• Risk tolerance that moves with the market. The same person describes themselves as comfortable with volatility after two good years and as conservative after a drawdown, without anything about their circumstances changing.
• Assuming the recent rate environment is normal. Savings rates, mortgage rates and inflation all get built into plans as though the last few years define the range.

A hypothetical, resolving a design question rather than an amount. Two investors hold the same three funds. The first reviews performance each quarter and moves money toward whichever has done best. The second rebalances on a fixed schedule back to a written target allocation, regardless of which fund led. Neither is forecasting. But the first has built a process that mechanically buys what has risen and sells what has fallen, which is recency bias implemented as a policy, while the second has built one that does the opposite by construction. This is the substantive reason a written policy and a mechanical contribution schedule are useful, and it does not depend on anyone predicting anything.

One caution about how to state the conclusion. "Recent returns are a poor input to a forecast" is what the evidence supports. "Recent returns will reverse" is a market prediction, and a different claim entirely. Only the first is defensible, and the difference matters, because a reader who takes recency bias as a reason to bet against recent performance has simply adopted the opposite forecast.

Pros

• Recency bias explains a well-documented pattern, the gap between fund returns and investor returns, without needing to assume anyone is foolish.
• It has an unusually strong finance-specific evidence base, so it does not rest on laboratory results transplanted into an investing context.
• The countermeasures are mechanical and cheap: a written allocation, a scheduled contribution, and a scheduled rebalance.

Cons

• The concept is easy to turn into its own forecast. Concluding that whatever just happened will now reverse is not a correction, it is the same error pointed the other way.
• Recent information is sometimes genuinely the most relevant information, and a permanent policy of discounting it would be wrong.
• The commonly quoted dollar cost of bad timing is disputed, so the concept is often sold with a number that is weaker than the concept itself.
• Naming the bias does not counteract it. Only a rule decided in advance does.