Efficiently Reusing Monte Carlo Simulation Output in Repeated Experiments for Financial and Actuarial Applications

在金融和精算应用的重复实验中有效地重用蒙特卡洛模拟输出

• 批准号: RGPIN-2018-03755
• 负责人: Feng, Mingbin
• 金额: $ 1.17万
• 依托单位: University of Waterloo
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 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=712542
• 关键词: Efficiently; Reusing; Monte; Carlo; Simulation

Consider a setting where simulation experiments are performed repeatedly, using the same simulation model but different input values. For example, in finance and insurance, simulation models support pricing and risk management decisions that are made periodically. In these settings, the standard practice is that each experiment is performed in isolation without using the output of previous experiments. We advocate a new simulation design paradigm, called green simulation, where one reuses output from previous experiments to answer new questions or to enhance the quality of new answers. Green simulation entails a new perspective on the management of simulation experiments. In standard practice, running simulation is a computational expense. In green simulation, running simulation is a computational investment that provides future benefits. The proposed research program involves designing, analyzing, and testing new green simulation experiment designs. The impact of green simulation extends beyond the design and analysis of computer experiments to enterprise risk management, financial engineering, data science, machine learning, and artificial intelligence. The benefit of green simulation is greater computational efficiency. Our preliminary findings show that green simulation can yield estimators whose variance converges to zero in settings where the standard practice yields estimators whose variances do not converge. Experiments on complex actuarial and financial applications show that green simulation can achieve substantially higher accuracy than standard practice. Such improvement can greatly reduce data and model uncertainty in enterprise risk management for insurance companies. The innovation in green simulation lies in its temporal view of a sequence of repeated experiments, as opposed to the standard experiment design that views each experiment in isolation. The proposed green simulation algorithms will improve the efficiency of later experiments by storing and reusing the output of earlier experiments. The proposed research program has numerous promising ventures, some of them include: development of novel green simulation experiment designs; theoretical analysis of the proposed experiment designs; application of green simulation in financial and actuarial applications; integration of machine learning and artificial intelligence methods in green simulation experiments. The ultimate goal of this research is to enable simulation users to conduct simulation experiments faster and cheaper. Simulation modeling is widespread in many businesses and in many scientific and engineering research fields. It often requires intensive use of high-performance computing; this occupies a scarce resource and consumes electricity. Green simulation offers a new venue to conduct simulation experiments more efficiently and more environmentally friendly.

考虑使用相同的仿真模型但不同的输入值重复执行仿真实验的设置。例如，在金融和保险领域，仿真模型支持定期做出的定价和风险管理决策。在这些设置中，标准的做法是每个实验都单独进行，而不使用以前实验的输出。我们提倡一种新的模拟设计范式，称为绿色模拟，其中一个重复使用以前的实验输出来回答新的问题或提高新答案的质量。绿色仿真为仿真实验管理提供了新的视角。在标准实践中，运行模拟是一项计算开销。在绿色仿真中，运行仿真是一项计算投资，可提供未来收益。 拟议的研究计划包括设计，分析和测试新的绿色模拟实验设计。绿色模拟的影响超出了计算机实验的设计和分析，扩展到企业风险管理、金融工程、数据科学、机器学习和人工智能。 绿色模拟的好处是更高的计算效率。我们的初步研究结果表明，绿色模拟可以产生的估计，其方差收敛到零的设置中，标准的做法产生的估计，其方差不收敛。复杂的精算和金融应用程序的实验表明，绿色模拟可以实现比标准实践高得多的准确性。这种改进可以大大降低保险公司企业风险管理中的数据和模型不确定性。 绿色模拟的创新在于它对一系列重复实验的时间视图，而不是孤立地看待每个实验的标准实验设计。所提出的绿色仿真算法将通过存储和重用早期实验的输出来提高后续实验的效率。建议的研究计划有许多有前途的风险，其中包括：开发新的绿色仿真实验设计;拟议的实验设计的理论分析;应用绿色仿真在金融和精算应用;集成机器学习和人工智能方法在绿色仿真实验。 本研究的最终目标是使仿真用户能够更快、更便宜地进行仿真实验。仿真建模在许多企业和许多科学与工程研究领域中广泛存在。它通常需要密集使用高性能计算;这占用了稀缺的资源并消耗电力。绿色仿真为更高效、更环保地进行仿真实验提供了新的途径。