Market Making with informed traders

点击翻译按钮获取中文摘要

Nowadays, many types of financial contracts are traded in electronic markets. When designing algorithmic trading strategies to act in these exchanges it is important to understand the different types of trading behaviour one will encounter. An important type of market participant is the market maker, sometimes also called dealer. Her role is to facilitate trade and provide liquidity to the exchange. She does provide liquidity by posting orders on both the bid and the ask side. Typically, market makers make money by earning the spread between bid and ask prices. In addition, depending on the market venue dealers might earn a fixed/variable rebate for providing liquidity.The optimal market making problem has been extensively studied over the years and it has been tackled with different approaches. Until recently, stochastic control is one of the widely used techniques. Lately, Machine Learning techniques have been used to find optimal market making solutions and the role of Artificial Intelligence in this field is rapidly expanding.There are two main sources of risk that a dealer might face which have been identified in the literature. The first one is called inventory risk and arises from the fact that most of the time the dealer's inventory is not zero and it is therefore exposed to changes in the asset's valuation. The second important risk factor is called adverse selection and arises from the fact that among the crowd of market participants there are traders which have private information and try to exploit it at the dealer's expenses.Our research project focuses on the adverse selection problem and on how to identify informed traders among market participants. Order flow is said to be toxic when it adversely selects market makers, who might not know that they are providing liquidity at a loss. Flow toxicity has been studied extensively and researchers have provided over the years two different ways of measuring it, which however have been criticised and are object of a dispute. Our main goal is to develop a market making model where the dealer makes use of Deep Reinforcement Learning techniques to detect informed traders and adjust her quotes according to the order flow. Typical market making models assume that the dealer does not use her accumulated experience to detect patterns in participants' behaviour. Indeed, she could use her memory of the past to adjust her quotes and protect herself from exogenous information that she might not be aware of.Our research should improve the quality of liquidity provision for both the market maker and clients which do not have any private information. Indeed, knowing that there are traders who exploit private information might help the dealer to adjust her quotes and not to trade at a loss too often. Another interesting problem that we would like to tackle is the broker perspective. A broker act in the same way as a market maker but he can decide who she wants to trade with. She could even decide arbitrarily not to trade with informed clients, and this would result in more fair quotes for other clients, who are normally trading for exogenous reasons. This however gives rise to a trade-off between losing money and finding out private information. Indeed, informed traders might be our only source of information.This project falls within the following EPSRC areas: 1) artificial intelligence technologies2) non-linear systems and 3) statistics and applied probability research area. The industry partner/collaborator for this project is BNP Paribas.