Stochastic Control and Games in Intraday Markets

日内市场中的随机控制和博弈

• 批准号: RGPIN-2018-05705
• 负责人: Jaimungal, Sebastian
• 金额: $ 2.99万
• 依托单位: University of Toronto
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=654444
• 关键词: Stochastic; Control; Games; Intraday; Markets

In this era of electronic markets, there are a number of new challenges faced by institutional investors (e.g., pension plans & mutual funds, and hence individuals), including how to: efficiently utilize large data feeds for making trading decisions; account for actions from a large number of heterogeneous traders to mitigate risks; incorporate latent information that drives markets; and understanding how to deal with model misspecification. As well, regulators need to study how to best manage and regulate traders to avoid, e.g., market manipulation and/or mini-flash crashes.*** This proposal aims to provide much needed insight into intraday financial markets by looking at empirical & computational aspects, and by studying mathematical problems arising in the context of intraday trading.***Large Stochastic Games*** Electronic markets are essentially large uncooperative games. The mean-field game (MFG) approach solves such problems by approximating the large finite game with the limit of infinite number of players. This proposal aims to generalize MFGs to make the results applicable to real intra-day markets by including features such as latent factors, heterogeneous agents, differing information sets, and prior assumptions. The goal is to understand how large number of interacting agents form markets, and the focus will be on applicable results that can be applied to inform traders, as well as, regulators on how to mitigate risks.***Machine Learning & Games*** Important inter-relationships across markets and assets, as well as the role of latent states, have been largely ignored in the academic literature. I propose to develop data-driven approaches by applying techniques from, and developing new ones in, machine learning. Specifically, I aim to develop reinforcement learning (RL) approaches that combine computational approaches with model-based approaches taken by financial mathematicians. RL uses the reaction of a system to an agent's action in an attempt to optimize some objective (such as a risk-return trader off). Generally, RL produces results that are difficult for regulators and traders to interpret. Model-based approaches, however, produce financially sound results, but are too rigid. I propose to combine these two completely separate lines research so that regulators and traders can be sure that recommendations are data-driven, but financially sound.***Anticipated Impact *** This research agenda will have significant impact in our understanding of intraday markets and, simultaneously, developments in MFGs, model uncertainty, and reinforcement learning. Industrial practitioners, PhD students, and other academics, will benefit from the research agenda I propose here. Regulators will benefit from the insights stemming from the results, as I aim to highlight what rules can mitigate risks such as market manipulation and mini flash crashes.

在这个电子市场时代，机构投资者（例如，养老金计划和共同基金，以及个人）面临许多新的挑战，包括：有效利用大型数据供稿来制定交易决策；解释来自大量异构交易者减轻风险的行动；结合驱动市场的潜在信息；并了解如何处理模型错误指定。同样，监管机构需要研究如何最好地管理和调节交易者，例如市场操纵和/或迷你刷新崩溃。平均场游戏（MFG）方法通过近似具有无限玩家数量的大型有限游戏来解决此类问题。该建议旨在通过包括潜在因素，异构代理，不同的信息集和先前假设等功能，使得MFGS概括为将结果适用于日期市场。目的是了解大量相互作用的代理形成市场，重点将放在适用的结果上，这些结果可用于告知交易者以及有关如何减轻风险的监管机构。***机器学习与游戏*** ***重要的跨市场和资产之间的重要相互关系，以及潜在国家的作用，在学术文献中已被忽略了。我建议通过在机器学习中应用和开发新技术来开发数据驱动的方法。具体而言，我旨在开发增强学习（RL）方法，将计算方法与金融数学家采用的基于模型的方法相结合。 RL使用系统对代理的行动的反应，以优化某些目标（例如，返回风险交易者）。通常，RL产生的结果很难使监管机构和交易者解释。但是，基于模型的方法会产生经济上的合理结果，但过于严格。我建议将这两个完全独立的线路研究结合在一起，以便监管机构和交易者可以确保建议是数据驱动的，但在财务上是合理的。工业从业人员，博士生和其他学者将受益于我在这里提出的研究议程。监管机构将受益于结果的见解，因为我的目标是强调哪些规则可以减轻市场操纵和迷你闪存崩溃等风险。