Applications of the Bootstrap in Indirect Inference and Scenario Generation

Bootstrap在间接推理和场景生成中的应用

• 批准号: 9813097
• 负责人: Thomas Willemain
• 金额: $ 19.89万
• 依托单位: Rensselaer Polytechnic Institute
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-09-01 至 2003-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9813097&HistoricalAwards=false
• 关键词: Applications; Bootstrap; Indirect; Inference; Scenario

9813097 Willemain The bootstrap is a method of computational inference that creates pseudodata called bootstrap samples by resampling with replacement from the historical data. This research will use the bootstrap to convert one historical input series into many similar samples, thus allowing the use of independent replications for output analysis. The investigation will include both the moving blocks bootstrap, which is the standard bootstrap variant for time series data, and a new alternative, the threshold bootstrap, that seems well-adapted to the demands of indirect inference. This approach is referred to as indirect inference because the bootstrap is not applied directly to the output series but instead is applied directly via the input series. Traditional direct inference focuses on the properties of estimators derived from bootstrap samples (e.g., unbiasedness, consistency), whereas indirect inference focuses on the properties of the samples themselves. In this regard, the most important quality of the bootstrap samples is that they be functionally indistinguishable from independent samples of the same stochastic process. If successful, one benefit of this research would be expanding the number of available inputs to trace-driven simulations, which achieve realism by driving system models with actual historical input series. Another benefit would be the automatic generation of lifelike scenarios for risk assessment in financial planning and for training human operators in complex decision making tasks.

9813097 Willemain Bootstrap 是一种计算推理方法，通过对历史数据进行替换重采样来创建称为 Bootstrap 样本的伪数据。 这项研究将使用引导程序将一个历史输入系列转换为许多相似的样本，从而允许使用独立的复制进行输出分析。 调查将包括移动块引导程序（时间序列数据的标准引导程序变体）和新的替代方案阈值引导程序，它似乎很好地适应了间接推理的需求。 这种方法称为间接推理，因为引导程序不直接应用于输出序列，而是通过输入序列直接应用。 传统的直接推理侧重于从引导样本导出的估计量的属性（例如，无偏性、一致性），而间接推理则侧重于样本本身的属性。 在这方面，引导样本最重要的品质是它们在功能上与同一随机过程的独立样本没有区别。 如果成功，这项研究的一个好处将是扩大跟踪驱动模拟的可用输入数量，通过使用实际历史输入序列驱动系统模型来实现真实性。 另一个好处是自动生成逼真的场景，用于财务规划中的风险评估以及培训人类操作员执行复杂的决策任务。