We present results from an experiment where participants have access to automated trading algorithms, which they may deploy at will while still trading manually. Treatments differ in whether robots must not be halted, deployment is compulsory, or robots can be halted and replaced at will. We hypothesize that robot trading would reduce mispricing, and that the effect would be more pronounced as commitment degree increases. Yet, compared to manual trading only, we observe equally large and frequent mispricing and, in early trading, significantly higher bid–ask spreads and more frequent flash crashes/price surges. Participants earn more, provided they combine robot and manual trading. Compared to evidence from archival data, we find significantly higher use of liquidity-taking robots. We attribute this to the inability, in the field, to identify the presence of liquidity takers when they happen not to trade.

Prediction markets are a successful information aggregation structure, however the exact mechanism by which private information is incorporated into the price remains poorly understood. We introduce a novel method based on the “Kyle model” to identify traders who contribute valuable information to the market price. Applied to a large field prediction market dataset, we identify traders whose trades have positive informational price impact. In contrast to others, these traders realize profit (on average) in excess of a theoretical expected informed lower bound. Results are replicated on other field prediction market datasets, providing strong evidence in favor of the Kyle model.