'SURPRISE main collection -- Dec 18'
(AsPredicted #205,080)
Download PDF

---

---

1) Have any data been collected for this study already?
No, no data have been collected for this study yet.

2) What's the main question being asked or hypothesis being tested in this study?
How do cognitive processing constraints affect how investors respond to earnings surprises? We hypothesize that stricter processing constraints increase people's sensitivity to the sign of earnings surprises – whether announced earnings exceed or fall below the consensus analyst forecast – while decreasing sensitivity to the size of surprises. Stricter processing constraints thus lead to excess sensitivity around zero surprise and attenuation away from zero.

3) Describe the key dependent variable(s) specifying how they will be measured.
Our primary outcome measure is the normalized, predicted percent change of the stock price between the time of the firm's public earnings announcement and the end of the day of the announcement.

We normalize a participant's predicted price change using their prior belief about the average realized returns. Predictions of price changes for positive surprises are divided by the prior beliefs about average returns to positive earnings surprises, while predictions for negative surprises are divided by the prior beliefs about average returns to negative earnings surprises.

4) How many and which conditions will participants be assigned to?
Participants predict the change between the stock price at the time of the earnings announcement and the same-day closing price for 5 different companies (presented in random order). The 5 companies were selected as U.S. companies who have a quarterly earnings announcement within the week following the data collection. For each company, participants see one possible earnings scenario, which is characterized by a standardized earnings surprise (SUE, calculated as [announced earnings-consensus forecast]/stock price). Across subjects, we randomly vary which SUE scenario is presented to them. We randomly draw SUE from +/-{0.0001, 0.0005, 0.001, 0.005, 0.01}. The firm's hypothetical earnings mentioned in the scenario are then calculated based on the firm's real consensus forecast of earnings, its stock price at the time of the experiment, and the SUE.

Participants are randomly assigned (with equal probability) to one of the following two between-subject conditions:

*Baseline:* Respondents receive basic information on the SUE scenario for the firm's earnings announcement in the form of a short text containing announced earnings per share, consensus analyst expectations and the latest stock price. There is no time limit for a prediction.


*High Constraints:* Respondents receive the same basic information on the earnings announcement scenario as well additional background information on the company's history (that is neutral and irrelevant for the price movement on the announcement day). Additionally, there is a time limit of 40 seconds for each prediction. If a participant times out on a given round, they will not be eligible for a bonus payment for that round.

5) Specify exactly which analyses you will conduct to examine the main question/hypothesis.
First, we will test for the basic hypothesized shape of the response function (steeper around zero surprise than far away from zero) in the Baseline condition using piecewise linear specifications. We compare the slope estimate in a regression of normalized predictions on SUE right around 0 (including SUE of either +/-{0.0001} or +/-{0.0001, 0.0005, 0.001} to the slope estimated on SUE values higher than those values (>0.0001 or >0.001, respectively) and the slope estimated on SUE values lower than those values (<-0.0001 or <-0.001, respectively). Our hypothesis is that the estimated slope on the narrow window around zero is larger than the slopes estimated on either side of that window.

Second, we test for the effect of the treatment (High Constraints) on the sensitivity to SUE. We run regressions of normalized predictions on SUE, an indicator for the High Constraints treatment, and the interaction of both. We run this regression repeatedly for expanding windows around zero SUE, i.e. for SUE in the window [-0.0001, 0.0001], for SUE in [-0.0005, 0.0005], for SUE in [-0.001, 0.001], for SUE in [-0.005, 0.005] and for SUE in [-0.01, 0.01], where the latter window corresponds to the full sample. We examine the interaction coefficient (normalized by the baseline coefficient for SUE) and hypothesize that (i) it is positive for SUE right around zero, (ii) negative on the full sample, and (iii) decreases as the SUE window expands around zero.

We conduct these analyses using median regressions given the potentially skewed distribution of normalized predictions induced by priors very close to 0.

6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations.
We exclude those respondents from the survey who fail to pass a comprehension test on the instructions within the first two attempts. Moreover, we exclude all respondents from the analyses who state a prior about average historical stock price changes in response to earnings announcements with a wrong sign. In particular, we exclude respondents who state a belief indicating that average same-day price changes in response to positive earnings surprises were non-positive, as well as all respondents who state a belief indicating that same-day price changes in responses to negative earnings earnings surprises were non-negative.

For all price change predictions, we restrict the entry range to a window ranging from -15% to + 15%.

7) How many observations will be collected or what will determine sample size?
No need to justify decision, but be precise about exactly how the number will be determined.
We collect data from 1,000 participants who successfully complete the survey using a sample of investors. Data is collected on the platform Prolific. We recruit participants who state having an account on a trading platform, are resident of the U.S., and at least 18 years of age.

8) Anything else you would like to pre-register?
(e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?)
We conduct the following additional analyses:
-- Excluding observations reflecting predictions with a sign that is opposite to that of the earnings surprise, i.e. a negative predicted price change for positive SUE and vice versa.
-- Excluding observations in the "High Constraints" group that timed out.
-- Excluding observations from subjects who indicate that they looked up additional information on any company online.
-- We conduct the same analyses using the raw price change predictions (i.e., predictions not normalized by priors) for completeness.
-- We show the interaction coefficients not normalized by the SUE slope of Baseline condition.
-- For completeness we run report OLS regressions (fitting conditional means instead of medians) and analyses on means instead of medians. To account for the potential skew in the normalized prediction measure, we winsorize at normalized predictions of +/-3.

In addition to the statistical tests of our hypothesis specified in (5), we also plot the median response by SUE in each treatment, both excluding and including predictions with a sign opposite to the earnings surprise.