Bayesian Decision Maker Research Articles

Predicting Trust Dynamics With Personal Characteristics

Previous research into trust dynamics in human-autonomy interaction has demonstrated that individuals exhibit specific patterns of trust when interacting repeatedly with automated systems. Moreover, people with different types of trust dynamics have been shown to differ across seven personal characteristic dimensions: masculinity, positive affect, extraversion, neuroticism, intellect, performance expectancy, and high expectations. In this study, we develop classification models aimed at predicting an individual’s trust dynamics type–categorized as Bayesian decision-maker, disbeliever, or oscillator–based on these key dimensions. We employed multiple classification algorithms including the random forest classifier, multinomial logistic regression, Support Vector Machine, XGBoost, and Naive Bayes, and conducted a comparative evaluation of their performance. The results indicate that personal characteristics can effectively predict the type of trust dynamics, achieving an accuracy rate of 73.1%, and a weighted average F1 score of 0.64. This study underscores the predictive power of personal traits in the context of human-autonomy interaction.

Unrealized arbitrage opportunities in naive equilibria with non-Bayesian belief processes

A non-Bayesian decision maker forms posterior beliefs through an – ever so slightly – violation of Bayes’ rule. A naive equilibrium is a competitive equilibrium for a multiperiod complete markets economy such that every economic agent – Bayesian or non-Bayesian – assumes that all economic agents are Bayesian decision makers. If all agents are indeed Bayesian decision makers, the naive equilibrium coincides with the standard concept of an arbitrage-free equilibrium for which dynamic price ratios are comprehensively pinned down as the equilibrium price ratios of Arrow–Debreu securities in a static economy. If at least one agent is a non-Bayesian decision maker, however, some equilibrium price ratios will change over time. These changing price ratios imply the existence of unrealized dynamic arbitrage opportunities in a naive equilibrium with non-Bayesian decision makers.

Forbearance in Institutional Investment Management: Evidence from Survey Data

We survey 218 institutional investors from 22 countries representing over $4.1 trillion in AUM to understand the drivers of forbearance in the termination of external asset managers. Although asset managers are fired for a variety of reasons, including taking on too much or too little risk as well as organizational changes at the investment manager or institutional investor level, poor performance is by far the dominant cause. There is surprising tolerance for underperformance and holding periods for investment managers are unexpectedly long. Forbearance is important and we argue that performance evaluation should be multifaceted, akin to a Bayesian decision-maker who conducts continued due diligence and updates beliefs about returns with process information.

Asynchrony rescues statistically optimal group decisions from information cascades through emergent leaders.

It is usually assumed that information cascades are most likely to occur when an early but incorrect opinion spreads through the group. Here, we analyse models of confidence-sharing in groups and reveal the opposite result: simple but plausible models of naive-Bayesian decision-making exhibit information cascades when group decisions are synchronous; however, when group decisions are asynchronous, the early decisions reached by Bayesian decision-makers tend to be correct and dominate the group consensus dynamics. Thus early decisions actually rescue the group from making errors, rather than contribute to it. We explore the likely realism of our assumed decision-making rule with reference to the evolution of mechanisms for aggregating social information, and known psychological and neuroscientific mechanisms.

Clustering Trust Dynamics in a Human-Robot Sequential Decision-Making Task

In this paper, we present a framework for trust-aware sequential decision-making in a human-robot team wherein the human agent’s trust in the robotic agent is dependent on the reward obtained by the team. We model the problem as a finite-horizon Markov Decision Process with the trust of the human on the robot as a state variable. We develop a reward-based performance metric to drive the trust update model, allowing the robotic agent to make trust-aware recommendations. We conduct a human-subject experiment with a total of 45 participants and analyze how the human agent’s trust evolves over time. Results show that the proposed trust update model is able to accurately capture the human agent’s trust dynamics. Moreover, we cluster the participants’ trust dynamics into three categories, namely, Bayesian decision makers, oscillators, and disbelievers, and identify personal characteristics that could be used to predict which type of trust dynamics a person will belong to. We find that the disbelievers are less extroverted, less agreeable, and have lower expectations toward the robotic agent, compared to the Bayesian decision makers and oscillators. The oscillators tend to get significantly more frustrated than the Bayesian decision makers.

Trust in the Danger Zone: Individual Differences in Confidence in Robot Threat Assessments.

Effective human–robot teaming (HRT) increasingly requires humans to work with intelligent, autonomous machines. However, novel features of intelligent autonomous systems such as social agency and incomprehensibility may influence the human’s trust in the machine. The human operator’s mental model for machine functioning is critical for trust. People may consider an intelligent machine partner as either an advanced tool or as a human-like teammate. This article reports a study that explored the role of individual differences in the mental model in a simulated environment. Multiple dispositional factors that may influence the dominant mental model were assessed. These included the Robot Threat Assessment (RoTA), which measures the person’s propensity to apply tool and teammate models in security contexts. Participants (N = 118) were paired with an intelligent robot tasked with making threat assessments in an urban setting. A transparency manipulation was used to influence the dominant mental model. For half of the participants, threat assessment was described as physics-based (e.g., weapons sensed by sensors); the remainder received transparency information that described psychological cues (e.g., facial expression). We expected that the physics-based transparency messages would guide the participant toward treating the robot as an advanced machine (advanced tool mental model activation), while psychological messaging would encourage perceptions of the robot as acting like a human partner (teammate mental model). We also manipulated situational danger cues present in the simulated environment. Participants rated their trust in the robot’s decision as well as threat and anxiety, for each of 24 urban scenes. They also completed the RoTA and additional individual-difference measures. Findings showed that trust assessments reflected the degree of congruence between the robot’s decision and situational danger cues, consistent with participants acting as Bayesian decision makers. Several scales, including the RoTA, were more predictive of trust when the robot was making psychology-based decisions, implying that trust reflected individual differences in the mental model of the robot as a teammate. These findings suggest scope for designing training that uncovers and mitigates the individual’s biases toward intelligent machines.

Anchored or Not: How Much Information Does 21st Century Data Contain on Inflation Dynamics?

Inflation was low and stable in the United States during the first two decades of the 21st century and broke out of its stable range in 2021. Experience in the early 21st century differed from that of the second half of the 20th century, when inflation showed persistent movements including the "Great Inflation" of the 1970s. This analysis examines the extent to which the experience from 2000-2019 should lead a Bayesian decisionmaker to update their assessment of inflation dynamics. Given a prior for inflation dynamics consistent with 1960-1999 data, a Bayesian decisionmaker would not update their view of inflation persistence in light of 2000-2019 data unless they placed very low weight on their prior information. In other words, 21st century data contains very little information to dissuade a Bayesian decisionmaker of the view that inflation fluctuations are persistent, or "unanchored". The intuition for, and implications of, this finding are discussed.

Predicting Security Returns in the Presence of Upper Price Limits

The objective of this paper is to develop a Bayesian forecasting model of security returns in a security market in which daily trading is subject to price limits. Traders and investors are assumed to be Bayesian decision makers under uncertainty, in the sense that they combine their a-priori information about the security returns with realized stock returns, to form the posterior and predictive distributions, on which trading decisions would be made. It allows traders and investors to adjust their long/short trading positions while considering the presence of upper price limits. In our model, security returns are assumed normally distributed, with an unknown mean and a known variance. Our traders and investors apply Bayes’ rule to update their prior information to derive the security returns’ predictive distribution. Our major contribution is the derivation of the stock returns’ predictive distribution, which is analytically quite complex. To illustrate the usefulness of our forecasting model, we provide five numerical examples of the predictive distributions, in which we alter one of the parameters, ceteris paribus, and demonstrate its impact on the predictions. This would allow traders and investors to upgrade their trading positions in the presence of an upper price limits.

Can Media Pluralism Be Harmful to News Quality?

I study the effect of polarization and competition on information provision. With a single expert who faces decision-makers with het- erogeneous priors, the expert solves a trade-off between persuading sceptics and retaining believers. With high polarization, an expert has incentives to supply low-quality information to leverage believers' credulity. With multiple experts with opposite biases, competition is harmful if attention is limited. Unbiased and Bayesian decision-makers rationally devote attention to like-minded experts. Echo chambers arise endogenously, whereas decision-makers would be better informed in monopoly. My model can rationalize the spread and persistence of conspiracy theories and fake news.

Model Uncertainty in the Cross Section

We develop a transparent Bayesian approach to quantify uncertainty in linear stochastic discount factor (SDF) models. We show that, for a Bayesian decision maker, posterior model probabilities increase with maximum in-sample Sharpe ratios and decrease with model dimensions. Entropy of posterior probabilities represents model uncertainty. We apply our approach to quantify the time series of model uncertainty in North American, European, and Asian Pacific equity markets. Model uncertainty is countercyclical in these markets. It predicts investors’ asset allocation decisions across equity and fixed-income funds. In survey data, investors tend to be more pessimistic about equity performance during periods of higher model uncertainty.

Modeling and Predicting Trust Dynamics in Human\u2013Robot Teaming: A Bayesian Inference Approach

Trust in automation, or more recently trust in autonomy, has received extensive research attention in the past three decades. The majority of prior literature adopted a “snapshot” view of trust and typically evaluated trust through questionnaires administered at the end of an experiment. This “snapshot” view, however, does not acknowledge that trust is a dynamic variable that can strengthen or decay over time. To fill the research gap, the present study aims to model trust dynamics when a human interacts with a robotic agent over time. The underlying premise of the study is that by interacting with a robotic agent and observing its performance over time, a rational human agent will update his/her trust in the robotic agent accordingly. Based on this premise, we develop a personalized trust prediction model and learn its parameters using Bayesian inference. Our proposed model adheres to three properties of trust dynamics characterizing human agents’ trust development process de facto and thus guarantees high model explicability and generalizability. We tested the proposed method using an existing dataset involving 39 human participants interacting with four drones in a simulated surveillance mission. The proposed method obtained a root mean square error of 0.072, significantly outperforming existing prediction methods. Moreover, we identified three distinct types of trust dynamics, the Bayesian decision maker, the oscillator, and the disbeliever, respectively. This prediction model can be used for the design of individualized and adaptive technologies.

SEQUENTIAL EXPERT ADVICE: SUPERIORITY OF CLOSED‐DOOR MEETINGS

AbstractTwo career‐concerned experts sequentially give advice to a Bayesian decision maker (D). We find that secrecy dominates transparency, yielding superior decisions for D. Secrecy empowers the expert moving late to be pivotal more often. Further, (i) only secrecy enables the second expert to partially communicate her information and its high precision to D and swing the decision away from first expert's recommendation; (ii) if experts have high average precision, then the second expert is effective only under secrecy. These results are obtained when experts only recommend decisions. If they also report the quality of advice, fully revealing equilibrium may exist.

Optimal learning before choice

A Bayesian decision maker is choosing among two alternatives with uncertain payoffs and an outside option with known payoff. Before deciding which alternative to adopt, the decision maker can purchase sequentially multiple informative signals on each of the two alternatives. To maximize the expected payoff, the decision maker solves the problem of optimal dynamic allocation of learning efforts as well as optimal stopping of the learning process. We show that the decision maker considers an alternative for learning or adoption if and only if the expected payoff of the alternative is above a threshold. Given both alternatives in the decision maker's consideration set, we find that if the outside option is weak and the decision maker's beliefs about both alternatives are relatively low, it is optimal for the decision maker to learn information from the alternative that has a lower expected payoff and less uncertainty, given all other characteristics of the two alternatives being the same. If the decision maker subsequently receives enough positive informative signals, the decision maker will switch to learning the better alternative; otherwise the decision maker will rule out this alternative from consideration and adopt the currently more preferred alternative. We find that this strategy works because it minimizes the decision maker's learning efforts. We also characterize the optimal learning policy when the outside option is relatively high, and discuss several extensions.

Generalization of results and neoclassical rationality: unresolved controversies of behavioural economics methodology

Many of behavioural economists have succumbed to misconception that economic models must be identified with observable reality. Therefore, behavioural approach enriches the sober neoclassical models with various psychological aspects of decision making in order to describe immediate reality more accurately. However, in contrast to natural sciences, social sciences aim at variance-based phenomena and thus the trade-off between theoretical objectiveness and empirical particularity remains. Behavioural insights are then at risk of being a mere fragmented empirical “evidence” of psychological descriptivism, which in combination with typical value neutrality should keep us vigilant about application in public policy. Secondly, the paper provides a progressive reading of rationality through a neoclassical lens. It is argued that rationality of a typical Bayesian decision-maker is immune to most of behavioural critique. It is shown on various examples that central behavioural concepts like satisficing behaviour or prospect theory cannot fundamentally challenge pillars of the neoclassical framework.

A mechanism for the elicitation of second-order belief and subjective information structure

This paper describes a direct revelation mechanism for eliciting (a) decision makers’ range of subjective priors under Knightian uncertainty and their second-order introspective belief and (b) Bayesian decision makers’ range of posteriors and their subjective information structure.

Statistical mechanics of cell decision-making: the cell migration force distribution

AbstractCell decision-making is the cellular process of responding to microenvironmental cues. This can be regarded as the regulation of cell’s intrinsic variables to extrinsic stimuli. Currently, little is known about the principles dictating cell decision-making. Regarding cells as Bayesian decision-makers under energetic constraints, I postulate the principle of least microenvironmental uncertainty principle (LEUP). This is translated into a free-energy principle and I develop a statistical mechanics theory for cell decision-making. I exhibit the potential of LEUP in the case of cell migration. In particular, I calculate the dependence of cell locomotion force on the steady state distribution of adhesion receptors. Finally, the associated migration velocity allows for the reproduction of the cell anomalous diffusion, as observed in cell culture experiments.

The Performance Effect of Feedback Frequency and Detail: Evidence from a Field Experiment in Customer Satisfaction

ABSTRACTThis paper presents the results from a field experiment that examines the effects of nonfinancial performance feedback on the behavior of professionals working for an insurance repair company. We vary the frequency (weekly and monthly) and the level of detail of the feedback that the 800 professionals receive. Contrary to what we would expect if these professionals conformed to the model of the Bayesian decision maker, more (and more frequent) information does not always help improve performance. In fact, we find that professionals achieve the best outcomes when they receive detailed but infrequent (monthly) feedback. The treatment groups with frequent feedback, regardless of how detailed it is, perform no better than the control group (with monthly and aggregate information). The results are consistent with the information in the latest feedback report being most salient and professionals in the weekly treatments overweighting their most recent performance, hampering their ability to learn.

The Performance Effect of Feedback Frequency and Detail: Evidence from a Field Experiment in Customer Satisfaction

This paper presents the results from a field experiment that examines the effects of non-financial performance feedback on the behavior of professionals working for an insurance repair company. We vary the frequency (weekly and monthly) and the level of detail of the feedback that the 800 professionals receive. Contrary to what we would expect if these professionals conformed to the model of the Bayesian decision maker, more (and more frequent) information does not always help improve performance. In fact, we find that professionals achieve the best outcomes when they receive detailed but infrequent (monthly) feedback. The treatment groups with frequent feedback, regardless of how detailed it is, perform no better than the control group (with monthly and aggregate information). The results are consistent with the information in the latest feedback report being most salient, and professionals in the weekly treatments overweighting their most recent performance, hampering their ability to learn.

Informative Starting Points and Asset-Return Puzzles

Decision-makers typically rely on informative starting points that are somewhat incorrect and then attempt to make appropriate adjustments. Such reliance on informative starting points may be an optimal response of a Bayesian decision-maker who faces finite computational resources (Lieder et al 2013). Professional equity-analysts take the same approach when they co-categorize closely related firms in peer-groups and routinely extrapolate the analysis pertaining to a prominent firm to other firms in the same peer-group. We show that if the representative agent behaves like a professional equity analyst, then a unified explanation for 5 asset-market phenomena emerges. The phenomena explained include high and counter-cyclical equity premium, size effect, value premium, and media-coverage effect. A novel prediction of the model is that stocks with less volatile payoffs outperform stocks with more volatile payoffs. Empirical evidence strongly supports this prediction.

Genes as Cues of Relatedness and Social Evolution in Heterogeneous Environments.

There are many situations where relatives interact while at the same time there is genetic polymorphism in traits influencing survival and reproduction. Examples include cheater-cooperator polymorphism and polymorphic microbial pathogens. Environmental heterogeneity, favoring different traits in nearby habitats, with dispersal between them, is one general reason to expect polymorphism. Currently, there is no formal framework of social evolution that encompasses genetic polymorphism. We develop such a framework, thus integrating theories of social evolution into the evolutionary ecology of heterogeneous environments. We allow for adaptively maintained genetic polymorphism by applying the concept of genetic cues. We analyze a model of social evolution in a two-habitat situation with limited dispersal between habitats, in which the average relatedness at the time of helping and other benefits of helping can differ between habitats. An important result from the analysis is that alleles at a polymorphic locus play the role of genetic cues, in the sense that the presence of a cue allele contains statistical information for an organism about its current environment, including information about relatedness. We show that epistatic modifiers of the cue polymorphism can evolve to make optimal use of the information in the genetic cue, in analogy with a Bayesian decision maker. Another important result is that the genetic linkage between a cue locus and modifier loci influences the evolutionary interest of modifiers, with tighter linkage leading to greater divergence between social traits induced by different cue alleles, and this can be understood in terms of genetic conflict.