A data driven approach for optimal stochastic control in finance

金融领域最优随机控制的数据驱动方法

• 批准号: 530985-2018
• 负责人: Li, Yuying
• 金额: $ 2.8万
• 依托单位: University of Waterloo
• 依托单位国家: 加拿大
• 项目类别: Collaborative Research and Development Grants
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=697902
• 关键词: data; driven; approach; optimal; stochastic

Many problems in finance naturally arise as stochastic optimal control problems, including for example optimal trade execution, pricing and hedging of variable annuities and multi-period asset allocation. We will investigate a new, data driven approach to handling such stochastic optimal control problems arising from finance. In particular, we propose to use a machine learning approach to constructing the optimal controls directly from the observed data. To understand the issues of this new paradigm, we will use this approach on three problems in finance. The first test problem will involve the optimal multi-period rebalancing of portfolios over a long time frame with taxable accounts. At each of these multiple rebalancing times one would need to take into consideration tax scenarios such as marginal tax rates, capital gains and capital losses. The second study will involve the timing of optimal multi-period portfolio rebalancing where we seek both optimal times to rebalance and the amounts to reallocate during these times. Our final study involves strategies for portfolios that combines selling options on single assets balanced by the concept of trend following involving baskets of assets. Here one needs to find the best weighting of each part. Canada is a world leader in Machine Learning and AI. The field of financial technology (FinTech) is one area which will benefit from better data analytics and more effective learning methods. This is particularly true when it comes to managing the risks involved with long term investment allocation problems.

金融中的许多问题自然会以随机最优控制问题的形式出现，例如，包括最优交易执行、可变年金的定价和对冲以及多时期资产配置。我们将研究一种新的、数据驱动的方法来处理由金融产生的这种随机最优控制问题。特别是，我们建议使用机器学习方法来直接从观测数据构造最优控制。为了理解这一新范式的问题，我们将在金融领域的三个问题上使用这种方法。第一个测试问题将涉及在很长一段时间内对有应税账户的投资组合进行最优的多阶段再平衡。在上述多次再平衡中的每一次，人们都需要考虑边际税率、资本利得和资本损失等税收情景。 第二项研究将涉及最优多阶段投资组合再平衡的时机，即我们寻求再平衡的最佳时间和在这些时间内要重新分配的金额。我们的最后一项研究涉及投资组合的策略，这些投资组合结合了单一资产的出售选择，并平衡了涉及一篮子资产的趋势跟随概念。在这里，人们需要找到每个部分的最佳权重。加拿大在机器学习和人工智能方面处于世界领先地位。金融科技领域(金融科技)将受益于更好的数据分析和更有效的学习方法。这一点尤其正确 当谈到管理涉及长期投资配置问题的风险时。