Forecasts and Forecasting Models: Prediction, Evaluation, Estimation, and Selection Using the Relevant Loss Function

预测和预测模型：使用相关损失函数进行预测、评估、估计和选择

• 批准号: 9520966
• 负责人: Francis Diebold
• 金额: --
• 依托单位: University of Pennsylvania
• 依托单位国家: 美国
• 项目类别: Continuing grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-11-01 至 1998-10-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9520966&HistoricalAwards=false
• 关键词: Forecasts; Forecasting; Models; Prediction; Evaluation

Francis Diebold University of Pennsylvania SBR-9520966 The proposal consists of two projects. The first project proposes the development of methods for estimating and evaluating forecasting models when the loss function is asymmetrical. Most estimation methods in current use are optimal for a quadratic loss function, but not for others. In particular, asymmetric loss functions will generally call for the use of biased estimators. The second project extends frequency domain methods to the estimation of parametric models. The P.I. proposes using bootstrap procedures to find confidence tunnels for estimated spectra.

宾夕法尼亚大学弗朗西斯·迪博尔德SBR-9520966该提案由两个项目组成。第一个项目提出了当损失函数不对称时估计和评估预测模型的方法的发展。目前使用的大多数估计方法对二次损失函数都是最优的，但对其他估计方法不是最优的。特别是，非对称损失函数通常需要使用有偏估计器。第二个项目将频域方法扩展到参数模型的估计。P.I.建议使用Bootstrap程序来为估计的光谱寻找置信度隧道。