Working papers results

We provide two characterizations, one axiomatic and the other neuro-computational, of the dependence of choice probabilities on deadlines, within the widely used softmax representation (see below picture) where pt (a; A) is the probability that alternative a is selected from the set A of feasible alternatives if t is the time available to decide, is a time dependent noise parameter measuring the unit cost of information, u is a time independent utility function, and a is an alternative-specific bias that determines the initial choice probabilities and possibly reflects prior information. Our axiomatic analysis provides a behavioral foundation of softmax (also known as Multinomial Logit Model when a is constant). Our neuro-computational derivation provides a biologically inspired algorithm that may explain the emergence of softmax in choice behavior. Jointly, the two approaches provide a thorough understanding of soft-maximization in terms of internal causes (neurophysiological mechanisms) and external effects (testable implications).

We study agents in a social network who receive initial noisy signals about a fundamental parameter and then, in each period, solve a robust non-parametric estimation problem given their previous information and the most recent estimates of their neighbors. The resulting robust opinion aggregators are characterized by simple functional properties: normalization, monotonicity, and translation invariance. These aggregators admit the linear DeGroot's model as a particular parametric specification. However, robust opinion aggregators allow for additional features such as overweighting/underweighting of extreme opinions, confirmatory bias, as well as discarding information obtained from sources perceived as redundant. We show that under this general model, it is still possible to link the long-run behavior of the opinions to the structure of the underlying network. In particular, we provide sufficient conditions for convergence and consensus and we offer some bounds on the rate of convergence. In some parametric cases, we derive the influence of the agents on the limit opinions and we stress how it depends on their centrality as well as on their initial signals. Finally, we study sufficient conditions under which a large society learns the true parameter while also highlighting why this property may fail.

Recent studies argue that major crises can have long lasting effects on individual behavior. While most studies focused on natural disasters, we explore the consequences of the global pandemic caused by a lethal influenza virus in 1918-19: the so-called "Spanish Flu". This was by far the worst pandemic of modern history, causing up to 100 million deaths worldwide. Using information about attitudes of respondents to the General Social Survey (GSS), we find evidence that experiencing the pandemic likely had permanent consequences in terms of individuals' social trust. Our findings suggest that lower social trust was passed on to the descendants of the survivors of the Spanish Flu who migrated to the US. As trust is a crucial factor for long-term economic development, our research offers a new angle from which to assess current health threats.

This paper describes price discovery and liquidity provision in a dynamic limit order market with asymmetric information and non-Markovian learning. Investors condition on information in both the current limit order book and also, unlike in previous research, on the prior order history when deciding whether to provide or take liquidity. Our analysis shows that the information content of the prior order history can be substantial. Surprisingly, the information content of equilibrium orders can differ from order direction and aggressiveness.

We study the role of perceived threats from cultural diversity induced by terrorist attacks and a salient criminal event on public discourse and voters' support for far-right parties. We first develop a rule which allocates Twitter users in Germany to electoral districts and then use a machine learning method to compute measures of textual similarity between the tweets they produce and tweets by accounts of the main German parties. Using the dates of the aforementioned exogenous events we estimate constituency-level shifts in similarity to party language. We find that following these events Twitter text becomes on average more similar to that of the main far-right party, AfD, while the opposite happens for some of the other parties. Regressing estimated shifts in similarity on changes in vote shares between federal elections we find a significant association. Our results point to the role of perceived threats on the success of nationalist parties.

We study the interplay between information acquisition and signaling. A sender decides whether to learn his type at a cost prior to taking a signaling action. A receiver responds after observing the signaling action. In the benchmark model where the sender's information acquisition decision is observed the sender does not acquire information and, therefore, does not signal. A rationale for signaling is provided by the model in which information acquisition is covert. There, in the unique equilibrium outcome surviving a form of never weak best response refinement the sender does acquire information and signals when the information is cheap.