Scholarlink

<p>This study investigates the out-of-sample predictive power for high-frequency returns of individual Japanese stocks and its implications for the price discovery process. Using 48 predictors across various look-back windows derived from high-frequency data, the analysis identifies trade imbalance, order imbalance, and past returns as the most important variables for forecasting. The magnitudes and signs of the standardized coefficients indicate that more recent look-back windows contain greater predictive information than earlier ones, and the momentum effect dominates the reversal effect. Employing an unconditional prediction strategy over all test periods yields average predictive performance inferior to that of the infeasible benchmark based on future average returns. When the forecasting strategy focuses on stable stocks with many observations where the estimated coefficients of the three key variables remain consistently nonzero and further limits the analysis to trading days that meet these criteria, it achieves a higher average out-of-sample <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> than the unconditional prediction strategy and outperforms the infeasible benchmark. This demonstrates the existence of stocks whose high-frequency returns can be predicted at 5&#xa0;s intervals. For the stocks with the highest number of trades and the highest trading volume, narrowing the look-back window further enhances performance in forecasting 5 second-ahead returns. However, the predictability disappears once the forecasting horizon extends beyond a few minutes. Our results suggest that, compared with representative U.S. large-cap stocks, Japanese large-cap stocks with lower liquidity and less frequent activation of informative signals may constrain short-horizon price discovery.</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

Out-of-sample predictability of high-frequency stock returns on the Tokyo Stock Exchange

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Masato Ubukata,
  • Toshiaki Watanabe

摘要

This study investigates the out-of-sample predictive power for high-frequency returns of individual Japanese stocks and its implications for the price discovery process. Using 48 predictors across various look-back windows derived from high-frequency data, the analysis identifies trade imbalance, order imbalance, and past returns as the most important variables for forecasting. The magnitudes and signs of the standardized coefficients indicate that more recent look-back windows contain greater predictive information than earlier ones, and the momentum effect dominates the reversal effect. Employing an unconditional prediction strategy over all test periods yields average predictive performance inferior to that of the infeasible benchmark based on future average returns. When the forecasting strategy focuses on stable stocks with many observations where the estimated coefficients of the three key variables remain consistently nonzero and further limits the analysis to trading days that meet these criteria, it achieves a higher average out-of-sample \(R^2\) R 2 than the unconditional prediction strategy and outperforms the infeasible benchmark. This demonstrates the existence of stocks whose high-frequency returns can be predicted at 5 s intervals. For the stocks with the highest number of trades and the highest trading volume, narrowing the look-back window further enhances performance in forecasting 5 second-ahead returns. However, the predictability disappears once the forecasting horizon extends beyond a few minutes. Our results suggest that, compared with representative U.S. large-cap stocks, Japanese large-cap stocks with lower liquidity and less frequent activation of informative signals may constrain short-horizon price discovery.