In many investment banks and major fund-management companies, automated "robot" trading systems now do work that 20 years ago would have required large numbers of human traders to perform: the rise of robot traders is a major success-story for artificial intelligence (AI) research. Although the technical details of currently profitable automated trading systems are closely guarded commercial secrets, the rise of robot trading can be traced back to a sequence of key AI research papers. Each of these key papers relied on minimal abstract simulation models of real financial markets: the simulators provide test-beds for trials in which the performance of different trading strategies could be evaluated and compared. Recent studies have revisited these seminal results, using more realistic simulations of contemporary financial markets, and have cast major doubts on core conclusions drawn in the original publications. Therefore, it seems reasonable to argue that present-day simulation methods are exposing significant problems in past research on automated trading. This position paper presents no new empirical results but instead presents a review of key past papers and an argument, a manifesto, for establishing a shared market-simulator test-bed that adequately reflects current real-world financial markets, for use in future evaluation and comparison of trading strategies.
|Number of pages
|Published - 20 Sept 2019
|International Conference on Modeling and Applied Simulation - Lisbon, Portugal
Duration: 18 Sept 2019 → 20 Sept 2019
Conference number: 18
|International Conference on Modeling and Applied Simulation
|18/09/19 → 20/09/19