Interpretable Intraday Signals for Real-Time Open-to-Close Directional Prediction: Evidence from Chinese Equity Indices

Abstract

This study tests whether a practitioner-style intraday heuristic can be formalized into reproducible, statistically testable signals for real-time stock-index directional prediction. At 13:30 each trading day, the primary task predicts whether the final close will exceed the daily open. Although the final close remains unknown at the prediction time, part of the open-to-close path has already been observed, so the task differs from forecasting only the remaining-session direction. The heuristic links two 30-minute windows, Zone 2 (10:01–10:30) and Zone 5 (13:01–13:30), to the final open-to-close direction of Chinese equity indices and treats a Zone 5 breakout of the morning range as a confirmation signal. We encode this judgment as three discrete variables, ZONE2_SIGNAL, ZONE5_SIGNAL, and BREAK_SIGNAL, and estimate an interpretable logistic regression using only information available by 13:30. The model is trained on 2015–2024 observations for the Shanghai Stock Exchange (SSE) Composite Index and the China Securities Index (CSI) 300 Index, evaluated on their 2025 observations, and applied without re-estimation to the Shenzhen Stock Exchange (SZSE) Component Index. The formalized framework improves 2025 holdout accuracy from 65.43% for the deterministic rule to 69.55%, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.768. The same-cutoff 13:30 price-position benchmark remains stronger, with 83.54% accuracy and an AUC of 0.893. As a supplementary remaining-session boundary test, the same specification is re-estimated to predict whether the final close exceeds the 13:30 close. Its 2025 holdout accuracy is 54.12%, with an AUC of 0.481, and it does not outperform the majority-class benchmark. The evidence therefore supports a transparent signal-decomposition framework for real-time open-to-close directional prediction, but not reliable post-13:30 forecasting, trading profitability, causal mechanisms, or market-wide generality.