From Data Acquisition to Lag Modeling: Quantitative Exploration of A-Share Market with Low-Coupling System Design

Fang Jianyong · 2025

Read From Data Acquisition to Lag Modeling: Quantitative Exploration of A-Share Market with Low-Coupling System Design by Fang Jianyong (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-19255.txt.

Introduction

Jianyong Fang and co-authors hunt lead-lag patterns in Chinese stocks in twenty twenty-five. Some stocks move first, and coupled partners follow after a delay. Finding those links can inform trading and reveal market structure. Their framework works in two stages, built as a low-coupling modular system.

First, long-term coupling between stocks is measured from daily data. Correlation, dynamic time warping, and rank-based metrics do that job. Then high-frequency data at one, five, and fifteen minute scales is mined. Cross-correlation, Granger causality, and regression models test for real lead-lag effects. Strongly coupled pairs often showed such effects, above all at finer time scales. The modular design supports scale and reproducibility.

Catalog record

CREATORFang Jianyong
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-19255

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

From Data Acquisition to Lag Modeling: Quantitative Exploration of A-Share Market with Low-Coupling System Design Fang Jianyong Abstract This paper presents a comprehensive two-stage approach to detect and analyze lead-lag effects in the Chinese A-share market. Our methodology first identifies highly coupled stock pairs through long-term data analysis, then verifies the presence of lead-lag relationships using high-frequency data. We implement a modular, low-coupling system to process daily data for coupling detection and 1-minute, 5-minute, and 15-minute data for lag verification. The core implementations, fixed_multi_downloader.py and two_stage_lag_analyzer.py , showcase our system design principles and methodological approach. Through detailed empirical analysis using cross-correlation, Granger causality tests, and regression models, we identify significant lead-lag effects primarily in stock pairs with established long-term coupling. Our results demonstrate that: (1) among coupled stock pairs, lag effects are more pronounced and shorter in higher frequency data; (2) certain industry leaders consistently influence follower stocks; and (3) the lead-lag pattern sensitivity varies across different market conditions. The system’s modular architecture enables independent development, testing, and maintenance of components, significantly enhancing research reproducibility and extensibility. This research contributes to both the understanding of A-share market microstructure and the methodological approach to financial data processing systems, providing insights for quantitative trading strategies based on cross-stock information flow. 1 Introduction The A-share market, as an important component of China’s financial system, has been a focus of both academic research and industry practice. Information dissemination and price adjustment in the market are often not instantaneous but subject to certain time delays, a phenomenon known in financial literature as the lead-lag effect [ 1 , 2 ] . The lead-lag effect suggests that price movements of certain securities lead those of others, revealing not only the microstructure characteristics of the market but also providing potential trading opportunities for investors. This research introduces a novel two-stage approach to studying lead-lag relationships in the Chinese A-share market. Unlike conventional studies that directly search for lead-lag effects across all possible stock pairs—a computationally expensive and often inefficient approach—our method first identifies highly coupled stock pairs through long-term data analysis, then verifies the presence of lead-lag relationships in these pre-screened pairs using high-frequency data. This ”screening and verification” framework significantly improves the efficiency of lead-lag detection while reducing false positives. Our study constructs a low-coupling data processing and analysis system to implement this two-stage approach. The system collects and processes data at different time granularities: daily data for long-term coupling identification, and high-frequency data (1-minute, 5-minute, 15-minute) for short-term lead-lag verification. This methodology enables us to systematically explore how stocks that exhibit synchronized price movements over longer time frames may display leader-follower relationships at shorter intervals. In the context of the Chinese market, the coupling and subsequent lead-lag effect may be driven by multiple factors. First, the A-share market is dominated by retail investors, which may lead to heterogeneity in information response [ 3 ] . Second, China’s special trading system (such as price limits) may delay the full reflection of information in prices [ 4 ] . Furthermore, stocks of different industries and sizes have differences in liquidity and information transparency, which may also cause asynchronous price adjustments [ 5 ] . Our research unfolds in three aspects: first, we build a low-coupling modular system to achieve an efficient data processing workflow; second, we apply our two-stage methodology to identify truly meaningful lead-lag relationships; finally, by analyzing result differences across time granularities, we explore the temporal scale characteristics of lead-lag effects in previously identified coupled stock pairs. The main contributions of this research are: (1) introducing a more efficient ”coupling screening followed by lag verification” methodology; (2) demonstrating a low-coupling modular approach to processing large-scale financial time series; (3) providing multi-time-scale evidence of lead-lag effects in China’s A-share market; and (4) exploring potential trading strategies based on identified lead-lag relationships. 2 Literature Review Research on lead-lag effects can be traced back to the pioneering work of Lo and MacKinlay [ 1 ] , who found that price movements of large-cap stocks often lead those of small-cap stocks. Since then, numerous studies have confirmed this phenomenon in different market contexts. Badrinath et al. [ 6 ] found that stocks with higher institutional investor ownership often lead stocks dominated by retail investors. Chordia and Swaminathan [ 7 ] emphasized the key role of trading volume in lead-lag relationships, noting that high-volume stocks typically respond faster to market information than low-volume stocks. In recent years, with increased availability of high-frequency trading data, research on lead-lag effects has become more refined. Huth and Abergel [ 8 ] studied lead-lag relationships at different time scales using high-frequency data, finding that these relationships are more significant at shorter time scales. Cai et al. [ 9 ] used deep learning methods to detect non-linear lead-lag relationships between stocks, improving predictive performance. In the context of the Chinese market, research on lead-lag effects has also made progress. Lin et al. [ 5 ] found that intra-industry lead-lag effects in the A-share market are significant and related to company size and analyst coverage. Yang et al. [ 15 ] studied the asymmetry of information transmission in the A-share market, finding that good news and bad news propagate at different speeds. Zhou and Zhu [ 10 ] discovered significant lead-lag relationships between upstream and downstream enterprises in the industrial chain, possibly related to business connections between companies. Despite the rich achievements of existing literature, current research still has several limitations: first, most studies focus on a single time scale (usually daily or longer periods), lacking comprehensive examination of multiple time scales; second, research methods tend to rely on single techniques, lacking cross-validation with multiple methods; finally, discussions on system design and implementation are relatively sparse, limiting the reproducibility and extensibility of research. This study attempts to fill these research gaps by developing a low-coupling modular system, combining multiple analysis methods and multi-time-scale data, providing more comprehensive empirical evidence for understanding the information transmission mechanism in China’s A-share market. 3 Data and System Design 3.1 Data Sources and Multi-granularity Design This research collected A-share market data from 2019 to 2024 at four granularity levels: • 1-minute data (stored in dedicated directory /data_1min_fixed ) • 5-minute data (stored in dedicated directory /data_5min_fixed ) • 15-minute data (stored in dedicated directory /data_15min_fixed ) • Daily data (stored in dedicated directory /data_daily_fixed ) Data was obtained via the Akshare API, which provides access to OHLCV (Open, High, Low, Close, Volume) data from the East Money database. The multi-granularity approach serves several purposes: • Enables detection of lead-lag effects at different time horizons; • Allows comparison of effect strength and persistence across time scales; • Mitigates the impact of microstructure noise prevalent in very high-frequency data; • Provides a more complete picture of information transmission dynamics. We collected data for 1,283 stocks in the A-share market, excluding stocks with listing dates after January 1, 2019, and those with extended trading halts (more than 20 consecutive trading days). 3.2 System Architecture and Low-Coupling Design The system was designed with modularity and low coupling as primary architectural principles. Figure 4 illustrates the high-level architecture. Figure 1 : Data Acquisition Module Detailed Design Figure 2 : System architecture diagram showing the low-coupling design

Read the complete text