AI, Investor Expertise, and Investment Decisions

Last registered on October 22, 2024

Pre-Trial

Trial Information

General Information

Title
AI, Investor Expertise, and Investment Decisions
RCT ID
AEARCTR-0014588
Initial registration date
October 18, 2024

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
October 19, 2024, 9:49 PM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated
October 22, 2024, 9:26 PM EDT

Last updated is the most recent time when changes to the trial's registration were published.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Affiliation
University of Chicago

Other Primary Investigator(s)

Additional Trial Information

Status
In development
Start date
2024-10-17
End date
2024-12-31
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
In this study, we aim to examine the extent to which generative AI is helpful for experts and non-experts in financial markets.
External Link(s)

Registration Citation

Citation
Muhn, Maximilian. 2024. "AI, Investor Expertise, and Investment Decisions." AEA RCT Registry. October 22. https://doi.org/10.1257/rct.14588-2.0
Experimental Details

Interventions

Intervention(s)
We aim to examine whether providing generative AI tools improve individual investment decisions
Intervention Start Date
2024-10-17
Intervention End Date
2024-12-31

Primary Outcomes

Primary Outcomes (end points)
Accuracy of Earnings Prediction
Accuracy of Sentiment
Portfolio returns and Sharpe ratio of short-term investment decisions
Portfolio returns and Sharpe ratio of long-term investment decisions
Primary Outcomes (explanation)
Accuracy of Earnings Prediction:
To assess the accuracy of earnings prediction, we compare the participants' answers with the actual archival data. The accuracy of participant $j$'s earnings prediction after reviewing firm $i$'s year $t$ conference call is:
\begin{align}\label{eq:epspred}
\text{Accuracy}_{ijt} = -1 \times \left| \mathbbm{1}(\text{Increase})_{it} - \left(\frac{\text{Answer}_{ijt}}{10}+0.5\right) \right|
\end{align}
$\mathbbm{1}(\text{Increase})_{it}$ is an indicator that equals one when firm $i$'s year $t+1$ EPS is larger than that of year $t$ and zero otherwise. As a participant's answer ($\text{Answer}_{ijt}$) ranges from $-5$ to $+5$, we scale it with ten and add back 0.5 to make it distributed between 0 and 1. As a participant makes a more correct prediction, the term $\left| \mathbbm{1}(\text{Increase})_{it} - \left(\frac{\text{Answer}_{ijt}}{10}+0.5\right) \right|$ becomes closer to zero. If a participant makes a completely incorrect prediction, this expression takes a value of 1. We multiply it by $-1$ to make higher values indicate a higher accuracy.

Additionally, we also use AI to evaluate the text quality.
*****
Accuracy of Sentiment:
Along with earnings prediction, participants are asked to assess whether the news is positive or negative for the company. Their answers span from $-5$ to $+5$, identical to the earnings prediction. We use the same formula as Equation \ref{eq:epspred} to evaluate the accuracy. However, instead of $\mathbbm{1}(\text{Increase})_{it}$ being an indicator for EPS improvement, we use cumulative stock returns around the earnings announcement dates (from one day before to one day after the announcement) as a benchmark. $\mathbbm{1}(\text{Increase})_{it}$ takes a value of one when announcement returns are positive and zero otherwise.
****
Portfolio returns and Sharpe Ratios for investments:
We measure investment efficiency by comparing the participants' answers with hypothetical maximum realized returns and Sharpe ratios. Participants provided their asset allocation plan for the stocks (Stocks A and B) and cash holdings. Their allocation amounts should add up to \$1,000. We compute relative investment weights $w_A$, $w_B$, and $w_C$ by scaling the amount invested in Stock A, Stock B, and cash by \$1,000 ($w_A+w_B+w_C=1$).
Participants must provide their asset allocation plans for both short-term and long-term. We define short-term returns as the daily returns measured from when the market opens on the trading day immediately following the earnings conference call until it market closes on the same day. Long-term returns are measured as cumulative returns from when the market opens on the trading day immediately following the earnings conference call, extending up to 252 trading days thereafter. We obtain participant $j$'s realized portfolio ($m$) return and Sharpe ratio after reading time $t$ earnings call information in both investment scenarios.

Secondary Outcomes

Secondary Outcomes (end points)
Confidence in Earnings Prediction
Confidence in Sentiment
Timeliness in Earnings Prediction
Timeliness in Sentiment
Secondary Outcomes (explanation)
Confidence is the confidence self-assessed by the participants. Timeliness is measured the time taken (in seconds) to complete each section of the survey..

Experimental Design

Experimental Design
We conduct experiments with more and less financially knowledgeable people.
Experimental Design Details
Not available
Randomization Method
Randomization done in office by a computer or by Prolific directly. Each individual will be assigned to one of three conditions (raw transcript, expert summary or non-expert summary). They then will see the same treatment condition for two matched firms (i.e., the matched company pair company a and company b). These company pairs are randomly assigned to participants as well.
Randomization Unit
Individual-level
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
No clustered assignment
Sample size: planned number of observations
First Experiment: 1800 in total planned Second experiment: 400 in total planned
Sample size (or number of clusters) by treatment arms
First Experiment:
600 per treatment arm

Second experiment:
200 per treatment arm
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
IRB

Institutional Review Boards (IRBs)

IRB Name
University of Chicago
IRB Approval Date
2024-04-30
IRB Approval Number
296