Seminario 22/09 – Debrah Meloso (Toulouse Business School) Algorithm choice in experimental markets: individual and aggregate effects

Información

We use high-performance Continuous Double Auction trading software and algorithms to study the effects of algorithmic trading on pricing and allocative efficiency in a laboratory environment. In addition to trading manually, participants can deploy algorithms that bid marginal valuations modulo a spread and a market-making (maker) or liquidity-taking (taker) parameter. The spread and maker/taker parameters are controlled by the participants who also decide whether to deploy an algorithm or not. With no manual trading, the algorithms are such that in equilibrium there are efficiency losses stemming from a prisoner-dilemma-like argument, which diminish with the number of traders but do not vanish in the limit. Pure strategy equilibrium requires coordination with all buyers/sellers choosing the same parameters, and with maker robots having large spreads, while taker robots have small spreads. Data from six laboratory experimental sessions support many of the theoretical findings. Most traders deploy an algorithm whenever available (82% average across all rounds), and while coordination is not achieved, traders who deploy makers (takers) significantly increase (decrease) the chosen spread with experience. While these algorithm choices—consistent with the theory—are efficiency reducing, as high spread depletes gains from trade and taker algorithms deplete liquidity, efficiency is increased with respect to rounds without algorithms. Realized gains from trade increase from 58% to 91% and, while the allocative efficiency increases across the board, those who most benefit are the traders who perform poorly in manual trading.