Scholarlink

<p>This paper introduces a new model averaging approach to consider uncertainty in model specification using an asymmetric loss, linear exponential (LINEX) loss function. We are motivated by the existing model-averaging prediction analysis studies being based on symmetric loss functions, which cannot meet practical situations where different weights are needed for over-prediction and under-prediction. The existing approaches cannot be used for the asymmetric loss. The proposed model averaging estimator established via the LINEX model averaging (LMA) criterion is shown to be optimal in achieving the lowest possible LINEX loss. We demonstrate the superiority of the LMA method and its effectiveness in movie forecasting and bitcoin volatility forecasting applications. Compared to other methods, the LMA estimator effectively reduces asymmetric loss and performs reasonably well even in the case of symmetric loss.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Frequentist model averaging under a linear exponential loss

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • Xinmin Li,
  • Hua Liang,
  • Huihang Liu,
  • Tingting Tong,
  • Tian Xie

摘要

This paper introduces a new model averaging approach to consider uncertainty in model specification using an asymmetric loss, linear exponential (LINEX) loss function. We are motivated by the existing model-averaging prediction analysis studies being based on symmetric loss functions, which cannot meet practical situations where different weights are needed for over-prediction and under-prediction. The existing approaches cannot be used for the asymmetric loss. The proposed model averaging estimator established via the LINEX model averaging (LMA) criterion is shown to be optimal in achieving the lowest possible LINEX loss. We demonstrate the superiority of the LMA method and its effectiveness in movie forecasting and bitcoin volatility forecasting applications. Compared to other methods, the LMA estimator effectively reduces asymmetric loss and performs reasonably well even in the case of symmetric loss.