Emotions affect investors’ willingness to take risks during trading sessions. A machine learning & facial recognition experimental research

How to cite

Poggi, Sofia. 2022. "Emotions affect investors’ willingness to take risks during trading sessions. A machine learning & facial recognition experimental research." AEA RCT Registry. AEARCTR-0009681. https://doi.org/10.1257/rct.9681-2.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Sofia Poggi*

* Primary investigator

Registry ID
AEARCTR-0009681
DOI
https://doi.org/10.1257/rct.9681-2.0
First registered
Last updated
Status
On going
Trial dates
2022-06-17 – 2023-10-31
Countries
Private
JEL codes
G41
Registry record
http://www.socialscienceregistry.org/trials/9681

Abstract

Analyzing human emotions is fundamental in decision-making, considering that they determine more than 90% of our behaviors.
Nowadays, it is increasingly essential for companies to measure emotions to deeply understand and optimize people's decision-making processes and avoid wasting economic resources. Theoretically, according to traditional finance researchers such as Sharpe (1964), Modigliani, Miller (1958), Malkiel, and Fama (1970) , investors are considered rational decision-makers regarding trading in stock markets. However, looking into economic literature over the last 30 years, an extensive body of literature has shown the effects of cognitive and behavioral biases on financial decisions. Growing evidence suggests that behavioral factors affect individual economic behavior, and some of these factors demonstrably influence financial markets. Firstly, this study intends to capture emotions during the trading decision-making leveraging a cutting-edge technology: artificial intelligence can transform micro-facial expressions into emotions. Secondly, the study is a field experiment on real trade market rather than a laboratory experiment to directly capture and test the causal impact of emotions on changes in individual risk aversion and whether the observed changes in risk aversion on the real market were caused or empathized by a negative emotion such as fear as tested by Guiso, Sapienza, and Zingale, 2013 in a laboratory experiment.To better understand whether fear could be responsible for the change in risk aversion and to better identify the emotional channel, we rely on a treatment and control framework. Half of the participants will watch a short horror video before a trading session. Since the subject will be randomly assigned to watch the video, the idea is that this difference in treatment should entirely drive the difference in risk aversion and emotion intensity detected between the groups. For example, watching a horror movie triggers an emotional and physical response similar to those produced by a severe financial loss. So differentially from the Guiso et al. experiment, the real experiment should mimic the situation of investors who are emotionally and financially affected by the stock market crash rather than the investors affected only financially. In a few words, the research aims at detecting the impact of emotion on investors’ willingness to take risks during trading sessions through the application of a Machine learning algorithm able to see facial micro-expression and transform it into data.

Intervention

The participants assigned to the treatment group will be asked to watch a short horror video before the trading session. A camera will film each part of the treatment and control group during the trading session. The data collected through computer vision and action unit algorithms will then be analyzed. At least on average, the treatment and the comparison group will be the same in all dimensions at the start of the evaluation.

Experimental design

The evaluation will be conducted through a field experiment among freelance traders (18 to 25 years) part of 30 Italian university finance clubs. The investigation will look like an online trading challenge among the entire finance community. The community counts about 3000 subscribers; each club counts 90 subscribers.
I decided to experiment on non-experienced traders because, as demonstrated by (Dufwenberg, Lindqvist, and Moore, 2005), bubbles are dampened or eliminated when some or all traders are experienced.

The challenge will have standard and specific conditions among all traders.
 Each trader will use the same trading simulator " starting Finance App" with the same virtual amount of money
 Each participant will start the performance on the same day, simultaneously, for the same duration. Friday 17th June 3.30 pm to 5 pm
 Each participant will take part in the challenge by working from their home
 Reference Market: stock (SP 500)

Design details

Randomization method
We assign units to different groups by a random process from this pool of eligible units. Randomly assign which individuals are in the treatment and the control group. In particular, half of the participants will be assigned to the treatment group and half to the control group.
Randomization unit
individual
Clusters
3000 individuals
Observations
100 individuals
Arms
50 individuals
Primary outcomes
Fear impacts investor’s willingness to take risks during trading sessions
Secondary outcomes
Emotion detection and intensity of investor’s decision-making during trading sessions
Partners
Partner_type: private_company Url: https://startingfinance.com/

Cited by

No citation data available for this registration yet.

Keywords