Scholarlink

<p>Accurate stock market forecasts are important for both investors and regulators. Prior research has argued the effectiveness of denoising-prediction models on stock market time series. However, the one-time denoising process may cause the in-sample series used for model training to be affected by the out-of-sample series, leading to overestimated accuracy. To test whether the denoising-prediction strategy contributes to improving stock market forecasts, this study introduces a sliding window to address the one-time denoising bias. The empirical mode decomposition (EMD) and variational mode decomposition (VMD) techniques are employed for denoising, and a non-iterative machine learning model, extreme learning machine (ELM), is employed as the predictor. Thus, a sliding denoising (SD)-EMD-ELM model and an SD-VMD-ELM model are developed. A comparison of the two models with individual and conventional denoising (D)-EMD-ELM and D-VMD-ELM models reveals that the superior performance of the denoising-prediction models is actually an overestimation due to the one-time denoising bias. This conclusion is corroborated by the technical implementation of the denoising process and the weak-form efficient market hypothesis.</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

Overestimated stock market forecasts resulting from the one-time denoising

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Mengyuan Xiong,
  • Kunliang Xu,
  • Jiayi Yin,
  • Weiqing Wang

摘要

Accurate stock market forecasts are important for both investors and regulators. Prior research has argued the effectiveness of denoising-prediction models on stock market time series. However, the one-time denoising process may cause the in-sample series used for model training to be affected by the out-of-sample series, leading to overestimated accuracy. To test whether the denoising-prediction strategy contributes to improving stock market forecasts, this study introduces a sliding window to address the one-time denoising bias. The empirical mode decomposition (EMD) and variational mode decomposition (VMD) techniques are employed for denoising, and a non-iterative machine learning model, extreme learning machine (ELM), is employed as the predictor. Thus, a sliding denoising (SD)-EMD-ELM model and an SD-VMD-ELM model are developed. A comparison of the two models with individual and conventional denoising (D)-EMD-ELM and D-VMD-ELM models reveals that the superior performance of the denoising-prediction models is actually an overestimation due to the one-time denoising bias. This conclusion is corroborated by the technical implementation of the denoising process and the weak-form efficient market hypothesis.