Transparency and Learning: Evidence from Defi Markets

Abstract

Using data from one of the first and most popular decentralized lending protocols, MakerDao, we study whether computer-language-based information lends itself for the efficient use of information in a market that features real-time transparency. We first find that there is persistent cross-sectional difference in performance, where persistence increases with investors sophistication. We then study how different types of processing costs affect the extent to which investors use past loan performance to mimic experts in real time (i.e., efficient mimicking). Our results show that awareness costs, proxied by loan activity level, hinder efficient mimicking. More importantly, acquisition and integration costs associated with translating code-based information into useful trading signals impedes investors’ ability to take advantage of information embedded in smart contracts. Our paper has important implications for regulators and practitioners aiming at more efficient use of information in smart contracts and on blockchain.