working papers

2024

1. Emotions in hybrid financial markets

   Lorenzo Cominelli, Gianluca Rho, Caterina Giannetti, and 6 more authors

   Discussion Papers, Sep 2024

   Number: 2024/311 Publisher: Dipartimento di Economia e Management (DEM), University of Pisa, Pisa, Italy

   Abs Website

   We investigate whether human traders experience milder emotions when participating in a financial market populated by artificial agents as opposed to a market comprising solely humans. In particular, by manipulating across conditions the number of artificial players, we assess how much emotions vary along with price dynamics (i.e. the occurrence of price bubbles). Notably, to ensure robustness, we evaluate emotions using three distinct methods: self-reporting, physiological responses, and facial expressions. Results show larger bubbles and milder emotional reactions in conditions with a higher count of artificial agents. Furthermore, negative emotions indirectly contribute to the mitigation of price bubbles. Ultimately, we observe a moderate degree of consistency across emotional measurements, with self-reported data being the least consistent among them.

2. Teaming Up with Artificial Agents in Non-routine Analytical Tasks

   Lorenzo Cominelli, Federico Galatolo, Caterina Giannetti, and 4 more authors

   Discussion Papers, Nov 2024

   Number: 2024/314 Publisher: Dipartimento di Economia e Management (DEM), University of Pisa, Pisa, Italy

   Abs Website

   Using a real-life escape room scenario, we investigate how different levels of embodiment in artificial agents influence team performance and conversational dynamics in non-routine analytical tasks. Teams composed of either three humans or two humans and an artificial agent (a Box, an Avatar, and a hyper-realistic Humanoid) worked together to escape the room within a time limit. Our findings reveal that while human-only teams tend to complete all tasks more frequently, they also tend to be slower and make more errors. Additionally, we observe a non-linear relationship between the degree of agent embodiment and team performance, with a significant effect on conversational dynamics. Teams with agents exhibiting higher levels of embodiment display conversational patterns more similar to those occurring among humans. These results highlight the complex role that embodied AI plays in human-agent interactions, offering new insights into how artificial agents can be designed to support team collaboration in problem-solving environments.

3. From WEIRD to GREAT? Exploring data quality of global platforms for online research

   Philipp Chapkovski, Eyal Peer, and Elena Brandt

   Jul 2024

   Abs DOI

   Online research relies heavily on platforms and panels, who sample predominantly from wealthy, educated, industrialized, rich and democratic countries (a.k.a., WEIRD), raising concerns for biases. Some platforms recently aim outside the West, to more GREAT (Growing, Rural, Eastern, Aspirational, and Transitional) areas. We explore how two such global platforms (Toloka and Besample) perform on data quality, in key aspects of attention, compliance, honesty, reliability, naivety and replicability, compared to West-centric platforms including Amazon Mechanical Turk (MTurk), Prolific, CloudResearch and Connect. We find that global platforms’ participants fail attention-check questions more than Prolific, CloudResearch and Connect, but not more than MTurk participants. Most attention failures were among participants with relatively low English proficiency. In the other aspects of compliance, honesty, reliability, naivety and replicability, the global participants perform comparably or somewhat inferiorly to Prolific, CloudResearch and Connect, but perform better than MTurk. This demonstrates an interesting phenomenon of global participants providing valid responses even after failing attention-check questions. This, we find, can be partially explained by language proficiency. We also examine how differences in data quality can be explained by location, usage patterns, demographic patterns and the time of day across time zones.

4. Teaming up with social artificial players in non-routine analytical task

   Federico Galatolo, Lorenzo Cominelli, Caterina Giannetti, and 2 more authors

   Aug 2024

   Publisher: OSF

   Abs DOI

   We investigate human-robot interaction within multiple-member collaborative teams, focusing on an escape room scenario to challenge participants with diverse tasks that mimic non-routine analytical tasks. This interdisciplinary research aims to uncover communication dynamics and understand how different social artificial agents impact on team dynamics and performance.

2023

1. Does Voluntary Disclosure of Polarizing Information Make Polarization Deeper? An Online Experiment on Russo-Ukrainian War

   Philipp Chapkovski, and Alexei Zakharov

   Feb 2023

   Abs DOI

   Does the animosity toward a holder of an opposite political opinion or the behavior toward someone whose opinion on a divisive issue is unknown depends on whether that opinion was disclosed or withheld voluntarily? In order to study this question, we conducted a pre-registered study in Russia, measuring the pro-war dictators’ behavior towards their partners with aligned or conflicting views on the war in Ukraine using give-or-take modification of Dictator Game. In the presence of a large polarisation gap (outgroup discrimination), we did not find that intentional vs. unintentional disclosure of the recipients’ positions affected the transfers of the dictators; at the same time, dictators’ beliefs about the share of war supporters among experiment participants and the donations made by other dictators were causally affected. Our study is the first one to consider this dimension of social interactions, and contributes to the quickly growing literature on political polarisation.

2. Cross-Impact and Price Bubbles: A Two-Asset Lab-Experiment

   Philipp Chapkovski, Francesco Cordoni, Caterina Giannetti, and 1 more author

   Aug 2023

   Abs DOI

   We investigate cross-impact in a hybrid experimental market, featuring both human and artificial agents. We exogenously vary across two treatments the available liquidity for trading. In treatment Separated participants have one distinct portfolio for each one of two stocks (markets are segmented), while in Unique participants have a unique portfolio, i.e., they can freely move their capital from one asset to the other. We observe larger financial bubbles for the speculative asset and larger (asymmetric) cross-market impact when participants have a unique portfolio. By comparing experimental with synthetic data and distinguishing between types of player activity, we can attribute cross-impact to human players, especially when the fundamental values of the two stocks are closer to each other, as well as to artificial players’ reaction to human presence.

3. Unintended Consequences of Corruption Indices: An Experimental Approach

   Philipp Chapkovski

   Mar 2023

   Abs DOI

   Using the results of a pre-registered online experiment, this paper examines how information about a corruption in a group can affect intergroup relations. Corruption indices are not only a valuable tool for investors and policymakers to make informed decisions, but can also lead to statistical discrimination: Individuals from a more "corrupt" region may be perceived as less trustworthy or more prone to dishonest behavior.To test this hypothesis, we manipulated the amount of information participants had about their potential partners’ regions of origin and asked them to (a) estimate the proportion of participants in each region who report a more profitable outcome in a coin toss game and (b) transfer money to a partner in each region in a trust game. The presence of a regional corruption index led participants to significantly overestimate the degree of dishonesty by participants from more corrupt regions and to reduce trust towards them. The results show how corruption indices can be a source of statistical discrimination against outgroups despite the well-meaning intentions of their creators.