Data and code for: Contagion risk prediction with Chart Graph Convolutional Network: Evidence from Chinese stock market

This dataset accompanies the study “Contagion risk prediction with Chart Graph Convolutional Network: Evidence from Chinese stock market”, which proposes a framework for contagion risk prediction by comprehensively mining the features of technical charts and technical indicators.
The utilized data include the closing prices of 28 sectors in Shen wan primary industry index, the closing price of CSI-300 Index, and eight classes of trading indicators that include Turnover Rate, Price-to-Earnings Ratio, Trading Volume, Relative Strength Index, Moving Average Convergence Divergence, Moving Average, Bollinger Bands, and Stochastic Oscillator. The sample period is from 5 Jan 2007 to 30 Dec 2022. The closing prices of 28 sectors are downloaded from the Choice database. The closing price of the CSI-300 Index and eight classes of trading indicators are downloaded from the Wind database.
This dataset includes two raw data files, one predefined temporary file, and eighteen code files, which are described as follows:
Sector_data.csv stores the closing prices of 28 sectors.
CSI_300_data.csv includes closing price of CSI-300 Index, and eight classes of trading indicators.
DCC_temp.csv is a predefined temporary file used to store correlation results.
Descriptive_code.py is utilized to calculate the statistical results.
ADF Test.py is utilized to test the stationarity of the data.
Min-max normalization.py is utilized to standardize data.
ADCC-GJR-GARCH.R is utilized to calculate dynamic conditional correlations between sectors.
MST_figure.py is used to a construct complex network that illustrates the inter-sector relationships.
Correlation.py is used to calculate inter-industry correlations.
Corr_up.py, corr_mid.py and corr_down.py are used to calculate dynamic correlations in upstream, midstream, and downstream sectors.
Centrality.py is used to quantify the importance or influence of nodes within a network, particularly across distinct upstream, midstream, and downstream sectors.
Averaging_corr_over_a_5-day_period.py calculates 5-day rolling averages of correlation and centrality metrics to quantify contagion risk on a weekly cycle.
Convert technical charts using PIP and VG methods.py extracts significant nodes and converts them into graphical representations, and save them in Daily Importance Score.csv, Daily Threshold Matrix.csv, and Daily Technical Indicators.csv.
Convert_CSV_to_TXT.py converts Daily Importance Score.csv, Daily Threshold Matrix.csv, and Daily Technical Indicators.csv into TXT files for later use.
Four files included in the folder of Generating and normalizing the subgraphs to generate subgraphs and then normalize them. The receptive_field.py serves as the main program, which calls the other three files. The stock_graph_indicator.py calculates topological structure data for subsequent use.
Predictive_model.py takes normalized subgraphs and Y-values defined by contagion risk as inputs and performs parameter tuning to achieve optimal results.