03 — Cognitive Biases

Recency Bias#

Statistical1 min read
◆ The PatternWeighting recent events more heavily than historical ones.

Recency bias — weighting recent events disproportionately when forming expectations.

Mechanism:
• Bull market for 5 years → "stocks always go up" (extrapolation)
• Crash last month → "the market is too dangerous" (fear persistence)
• Recent quarter's earnings → overshadows decade of data

Historical base rates are more reliable than recent anecdotes. The recent past feels like the permanent future — it isn't.

Worked example

Ten years of returns average 9.6%. The last three averaged 22.3%, and an expectation that halves the weight of each older year lands at 18.4% — nearly double the long-run figure, just after the best stretch in the sample.

Where it misleads
  • Recent data is not worthless: volatility really does cluster, so recent volatility is a better guide to next month’s than a ten-year average is. The bias is extrapolating returns.
  • Surveys show investors expect high returns after high returns (Greenwood & Shleifer 2014) — and those are the periods followed by lower returns on average.
  • A long-run mean is itself an estimate; ten years is a short sample, and the “base rate” can be just as wrong in a different way.
In code
r = pd.Series([0.12, 0.08, -0.05, 0.15, 0.10, 0.07, -0.18, 0.22, 0.25, 0.20])   # ten years of returns, made up
long_run = r.mean()
last_three = r.tail(3).mean()
felt = r.ewm(alpha=0.5).mean().iloc[-1]  # each year counts half as much as the one after it
Python, in your browser — edit the code and run it again

Assumes import numpy as np and import pandas as pd. Each snippet carries its own example numbers; the comments say which are made up.

Sources
  • R. Greenwood & A. Shleifer, “Expectations of Returns and Expected Returns”, Review of Financial Studies 27(3), 2014
  • U. Malmendier & S. Nagel, “Depression Babies: Do Macroeconomic Experiences Affect Risk Taking?”, Quarterly Journal of Economics 126(1), 2011
  • W. F. M. De Bondt & R. Thaler, “Does the Stock Market Overreact?”, Journal of Finance 40(3), 1985

Last reviewed October 2026 · How this site checks its content · Report a mistake

Pattern bridge: Overweighting recent events is the human version of exponential moving averages. In ML, cross-validation exists specifically to prevent this — don’t just test on the latest fold.
Linked from 5
← Previous
Anchoring
Open in the full reader, with the topic sidebar