This study investigates how U.S. retail investors’ recollection of past S&P 500 performance relates to their expectations of future performance. Using a Prolific survey, we elicit recalled returns (2025 and 2020–2025), forward-looking expectations (12-month and 5-year), and then reveal actual historical returns before re-eliciting expectations. The within-subjects pre/post design estimates how correcting memory shifts beliefs. Two between-subjects randomizations test (a) whether recalling past returns before forming expectations increases extrapolation, and (b) whether framing nudges on the information page differentially shift post-information expectations. The study is motivated by Jiang, Liu, Peng, and Yan (2025, QJE) and the Bordalo-Gennaioli-Shleifer memory framework.
Experimental design
Just a survey on prolific
Design details
Randomization method
respondents are self-selecting on prolific. I see no demographic or other information about them. A random number generator randomizes further.
1. Order (50/50) — condition: recall_first vs expect_first — whether participants recall past returns before or after stating
their expectations.
2. Nudge (thirds) — nudge: none, extrapolate ("Many other people have extrapolated historical returns…"), or dividends
("Dividends…are near historical lows.")
3. Information on very recent stock returns (50/50 split, then further divided) — recency: half see all 4 time references together (all_pct or all_points), half see a single reference (one of yesterday, week_ago, month_ago, jan1 × pct/points), or none. This controls which S&P 500 performance data points are displayed on the info reveal page.