Memory and Markets

In many environments, including credit and online markets, past records about participants are collected, published, and erased after some time. We study the effects of erasing past records on trade and welfare in a dynamic market where each seller's quality follows a Markov process and buyers leave feedback about sellers. When the average quality of sellers is low, unlimited records always lead to a market breakdown. Appropriately deleting past records, instead, fosters experimentation and can sustain trade in the long run. Positive and negative records play opposite roles with different intensity, and welfare is maximized for short positive records and long but bounded negative ones. Analogous results obtain within an information design approach, in which an information intermediary simply recommends to uninformed buyers whether to buy from each seller. Our findings have implications for the design of privacy regulations, credit bureaus, and feedback systems in online platforms