Algorithm Reactance in Online Investment Communities

Project: Research

View graph of relations


Algorithmic predictions have become prevalent within online investment communities (OICs), enabling investors to assess future stock performance through quantitative stock ratings. Meanwhile, the key contributors comprise mostly amateur analysts who generate stock analyses. In this way, algorithmic predictions and human analysts share the goal of producing stock predictions yet are different in the presence of such predictions. In turn, algorithmic predictions bring identity threats to these analysts whose image as proficient investment experts may be challenged. Amidst the prevalence of algorithmic predictions in OICs, a puzzle hence arises—whether and how does algorithmic prediction affect analyst-generated content? This proposal predicts an algorithm reactance effect, where analysts cope with identity threats by targeting the algorithmic predictions and themselves. Targeting algorithmic predictions, analysts may exhibit algorithm resistance. Given their lack of control over predictive algorithms (e.g., they cannot conceal algorithmic predictions), analysts seek to distinguish themselves by deviating from their algorithmic counterparts. That is, they strive to draw attention to their distinct stock opinions. Nonetheless, this may “backfire”—analysts’ irrational deviation from algorithms with demonstrated predictive power may lead to deteriorated forecast quality. Targeting themselves, we further posit that analysts will adopt a self-elevation strategy by working both harder and “smarter”. By working harder, analysts improve their productivity by writing more analysis articles to signal their stock analytics capabilities. By working “smarter”, analysts may enhance self-mention with more first-person pronouns to bolster credibility. We maintain that algorithm performance moderates the algorithm reactance effect. The more accurate algorithmic predictions are, the stronger identity threats they may bring to human analysts. In turn, it yields stronger algorithm reactance such that analyst forecast quality will be further deteriorated due to stronger algorithm resistance; self-elevation will be intensified manifested by a higher level of productivity and self-mention. From a managerial standpoint, the potential findings may cast light on how to make the most of algorithmic predictions when they are incorporated into online communities. If analysts are found to exhibit algorithm resistance, one possible intervention is to alert those who make predictions drastically different from their algorithmic counterparts to avoid poor forecasts. Online communities may also educate their users about their algorithms to alleviate derogation by enhancing analysts’ algorithm literacy. Similarly, analysts may be notified when their analysis manifests excessive subjective tones when increasing self-mention that may deteriorate analysts’ credibility and professional images.


Project number9043750
Grant typeGRF
StatusNot started
Effective start/end date1/01/25 → …