20 Jan 2022
This paper studies how to use sample data to improve decisions robustly when the datagenerating process (DGP) is only known to belong to a set of independent but possibly nonidentical distributions. It proposes two achievable notions of how decisions based on inference from data can improve upon those without using the data no matter which possible DGP governs the uncertainty. When decisions are made according to the maxmin expected-utility criterion, either of these notions is guaranteed if and only if the updated set of DGPs accommodates (contains) the true DGP. In the current setting, common inference methods (e.g., maximum likelihood and Bayesian updating) are shown to often fail this property. This paper proposes two novel and tractable updating rules that accommodate the true DGP either asymptotically almost surely or in finite sample with a pre-specified probability. Finally, it explores implications for applications such as asset pricing under ambiguity.
18 Jan 2022
Hirshleifer (1971) famously argued that the public disclosure of socially useless information hurts welfare because it creates unwanted economic fluctuations. We show that this logic can fail if the disclosed information concerns the medium of exchange. We consider an economy where agents gradually learn about the quality of a new asset and coordinate to adopt it as a medium of exchange or abandon it. The demand of this money-like asset can be partially convex, and the convexity translates more economic fluctuations into higher asset prices, making the asset a more useful payment device. Therefore more information disclosure sometimes raises welfare, even when information is not socially useful, i.e. when new information does not affect agents’ adoption decisions. When there are competing monies, the aggregate liquidity and welfare can be non-monotone in beliefs and hence a good news about a new money can be a bad news for the aggregate economy. In an extension with heterogenous agents we illustrate that the presence of some hodlers can change the allocation substantially.
06 Dec 2021
Our world is networked where everyone and everything is connected. Networks are ubiquitous, including social networks, human interaction networks, gene regulatory networks, to name just a few. Machine learning and data analytics methodologies have been found useful in extracting hidden interaction and communication patterns from related network data. Applications include online user behavior characterization, user alignment across social networks, epidemic risk prediction, disease characterization, etc. In this talk, how related applications can be formulated as machine learning and data analytics problems will be presented and discussed.