How CEO Gender Shapes Interpretation of Financial Information: Evidence from an Experimental Study

Participants complete three rounds of a forecasting task. In each round, they read a short, standardized case about a retail firm and make an initial EPS forecast. The intervention embeds two orthogonal manipulations:
1.CEO gender randomization. In each round, the perceived gender of the CEO is randomized by swapping only the CEO’s first name and corresponding pronouns from a pre-specified list of gender-diagnostic names. All other case content (firm facts, financials, layout, length, tone) is held constant, and no additional gender cues are introduced. Assignment to a female-named vs. male-named CEO occurs with equal probability.
2.Information signal randomization (optimistic vs. pessimistic). After the initial forecast, participants see three noisy signals of the true EPS realization labeled as “experts’ forecasts.” In each round, this set is randomized to be either optimistic (all signals above the true EPS realization known to the researchers) or pessimistic (all below), with equal probability. The three experts’ forecasts are paired with a brief text that contextualizes the signals; these texts are generated by a large language model trained on analysts’ reports to emulate sell-side language and framing.
The two manipulations—CEO gender and the sign of the experts’ signals—are randomized independently in each of the three rounds at the participant × round level.

Nothing to add with respect to what is the public box above.

2025-11-19

2025-11-20

In each of the three rounds, our primary outcomes are the participant’s prior and posterior beliefs for three items: (i) a point EPS forecast for the focal firm, (ii) the CEO’s managerial talent (both the first order belief of the student and the second order belief of what the stuednts thinks is the average in the class), and (iii) the firm’s fundamentals (again, both first order beliefs and second order beliefs of the class average are elicited). For talent and fundamentals, participants place the CEO/firm on a 0–100 relative-rank scale among retail firms of similar size (higher = better). Prior measures are elicited immediately after reading the case; posterior measures are re-elicited after the information display labeled as experts’ forecasts. The prior and posterior EPS forecasts and the second order beliefs for managerial talent and firm's fundamentals) are incentivized with a threshold-plus-lottery incentive scheme: for each incentivized question, any student whose answer falls within ±5% of the realized value is entered into a lottery; 10 winners per question receive £20.

In each round we collect two open-ended responses—one after the prior and one after the information display—asking participants which factors they considered when forming their initial beliefs and when updating. We will derive text-based measures from these answers .

Participants complete three rounds of a forecasting task. Each round begins with a short case about a retail firm that is presented to students as fictional but adapted from a real public company: the financial statements and EPS history are real, while the firm’s name and any identifying details are withheld. After reading the case, participants report a prior incentivized EPS forecast and two relative-rank judgments—managerial talent of the CEO and the firm’s fundamentals—on a 0–100 scale relative to retail peers of similar size. The case embeds two independent manipulations: (i) CEO gender, randomized via first name and pronouns while all other text is held constant, and (ii) an information display consisting of noisy signals of the true EPS realization labeled as “experts’ forecasts,” accompanied by a single paragraph that contextualizes those signals; the signals are randomized to be optimistic or pessimistic in each round. Participants then provide posterior values for the same three outcomes. Each round also includes two brief open-ended questions about the factors behind their prior and their update. Finally, we also elicit incentivized predictions about how classmates will answer the managerial-talent and fundamentals questions. We use a threshold-plus-lottery incentive scheme: for each incentivized question, any student whose answer falls within ±5% of the realized value is entered into a lottery; ten winners per question receive £20. Randomizations occur independently in every round.

Nothing to add with respect to what is the public box above.

Done by computer.

Both the ceo-gender and the news signal randomization are implemented at student-round level and they are independent between each other.

No

317 students.

Each student takes three round of the exercise for a total of 317*3=951 student-round observations.

For each of the 6 firms, we have four treatments arms coming from 2 genders of the CEO for given firm X 2 noisy signals (optimistic vs pessimisitic experts). Hence we have a total of 6x2x2=24 cells and our 951 student-round observations will be evenly distributed in these 12 cells.

N/A