Behavioral FinanceTrading Psychology

Memory Bias in Trading: Why Investors Recall Better Returns Than They Earned

Summary by Robert Gorak · Published July 21, 2026 · Last reviewed July 21, 2026

Daniel J. Walters and Philip M. Fernbach·2021·Proceedings of the National Academy of Sciences
Sample: 411 (Study 1); 144 completed of 151 recruited (Study 2); 366 (Study 3)Period: 2019 (Study 1); January–June 2020 (Study 2); 2018 (Study 3)

Positive memory bias in investing is investors' tendency to recall past returns as more favorable than what they achieved. In "Investor memory of past performance is positively biased and predicts overconfidence" (2021), Walters and Fernbach documented this bias among real investors. They surveyed 411 U.S. investors who recalled a 44.1% return on their most consequential 2019 trade versus an actual 39.8%. Positivity bias predicted overconfidence with a Cohen's d of 0.71 in Study 1's regression analysis.

What the Study Found

Investors overestimated the return on their second-most consequential trade too, recalling 40.6% versus an actual 33.5%, t(410) = 3.43, P < 0.001. In Study 2, 144 participants recalled a 29.6% average return on their top 10 trades versus an actual 21.6%. Participants forgot 39.7% of losing trades compared to 29.9% of gains in Study 2's selective forgetting analysis. Distortion and selective forgetting each predicted overconfidence, with effect sizes of d = 0.67 and d = 0.39 respectively. In Study 3, overconfidence fell from 9.2% in the control condition to 5.8% in the treatment condition, t(364) = 2.91, P = 0.004.

Methodology

Study 1 surveyed 411 U.S. investors on Pollfish about their two most consequential trades from 2019. Study 2 recruited 151 Prolific Academic participants, 144 of whom completed trades bought after January 1 and sold before July 1, 2020. Study 3 recruited 366 online-forum investors, randomly assigning them to look up 2018 stock trades or report investment sectors. Regressions controlled for age, gender, income, education, total investment assets, number of stocks owned, and financial advisor status.

Key Statistics

Metric Finding Context
Memory bias, Trade 1 44.1% recalled vs. 39.8% actual Study 1, t(410) = 2.14, P = 0.033, n = 411, 2019 trades
Memory bias, Trade 2 40.6% recalled vs. 33.5% actual Study 1, t(410) = 3.43, P < 0.001, n = 411, 2019 trades
Overall positivity bias 29.6% recalled vs. 21.6% actual Study 2, t(143) = 3.54, P < 0.001, n = 144, 2020 trades
Selective forgetting rate 39.7% of losses forgotten vs. 29.9% of gains Study 2, logistic regression, n = 144
Distortion effect on overconfidence d = 0.67 Study 2, Table 4 column 1
Selective forgetting effect on overconfidence d = 0.39 Study 2, Table 4 column 1
Positivity bias effect on overconfidence d = 0.71 Study 1, Table 2 column 1, n = 411
Overconfidence by condition 9.2% control vs. 5.8% treatment Study 3, t(364) = 2.91, P = 0.004, n = 366
Trading frequency by condition 16.4 trades control vs. 13.2 trades treatment Study 3, t(364) = 2.26, P = 0.024, incentive-compatible measure

Why This Matters

Financial advisors and platforms could reduce investor overconfidence by regularly displaying account statements rather than relying on client recall. Because traders who overestimate past success intend to trade more often, correcting memory bias may also lower unnecessary transaction costs. The gap between distortion (misremembering magnitude) and selective forgetting (omitting losses entirely) suggests debiasing tools need to address both memory failures separately. Self-reported performance reviews used in coaching or robo-advisory nudges are vulnerable to the same biases documented here.

Frequently Asked Questions

44.1% and 40.6% were the average returns 411 investors recalled for their two biggest 2019 trades, compared to actual returns of 39.8% and 33.5%. Walters and Fernbach (2021) define this positive memory bias as investors recalling past trading performance as better than what actually occurred.

9.2% was the average overconfidence level among investors in Study 3's control condition, who did not look up their actual returns. Overconfidence fell to 5.8% in the treatment condition, where 366 participants looked up their actual 2018 trades, t(364) = 2.91, P = 0.004.

16.4 trades was the trading intention in Study 3's control condition versus 13.2 in the treatment condition, t(364) = 2.26, P = 0.024. In Study 1, positivity bias in recalled returns predicted higher trading frequency with an effect size of d = 0.20.

39.7% of losing trades were forgotten by Study 2 participants versus 29.9% of gains, the study's test for selective forgetting. Distortion, the tendency to misremember a trade's magnitude, had a larger effect on overconfidence (d = 0.67) than selective forgetting (d = 0.39).

Source

Daniel J. Walters and Philip M. Fernbach (2021). Investor memory of past performance is positively biased and predicts overconfidence. Proceedings of the National Academy of Sciences.

Read the full paper