Forecasting Stock Price Crash Risk in the Tehran Stock Exchange Using Graph Deep Learning and Liquidity Indicators

Forecasting Stock Price Crash Risk in the Tehran Stock Exchange Using Graph Deep Learning and Liquidity Indicators

Pouria Goodarz1

1) Faculty of Accounting and Finance, University of Tehran, Tehran, Iran -

IOI Code : AAHC-ABHDGG
Publication : National Conference on Management, Commercialization, Knowledge-Based Economy and Investment for Production(impeb.ir)
Abstract :
    Stock price crash is one of the most important asymmetric risks in capital markets, as it can lead to severe losses for investors, reduced liquidity, intensified herding behavior, and risk transmission among related stocks. In the Tehran Stock Exchange, due to the presence of price limits, closing price mechanisms, trading halts, and buy and sell queue mechanisms, stock price crashes often emerge as gradual and cumulative declines over several trading days. Therefore, conventional price prediction models are not sufficient for designing an early warning system. The objective of this study is to design and empirically evaluate a model for predicting the probability of short-term stock price crashes in the Tehran Stock Exchange using graph deep learning and market microstructure indicators. To this end, daily trading data of listed stocks in the Tehran Stock Exchange from 2019 to 2025 were extracted. After stock screening and trading calendar alignment, the final research sample consisted of 102,600 stock-day observations. In the proposed framework, each stock is considered as a node in the market graph, and relationships among stocks are defined based on return correlation, industry affiliation, and liquidity similarity. The input variables of the model include return, daily price range, the gap between the last traded price and the closing price, volume shock, trading value shock, Amihud illiquidity, order imbalance, and sell queue pressure. The out-of-sample test results show that the Temporal-GAT model outperforms classical and non-graph-based models in predicting short-term stock price crashes. In addition, sell queue pressure, order imbalance, and illiquidity were identified as the most important early warning signals of stock price crashes. The findings of this study can provide a basis for designing early warning systems, ranking high-risk stocks, and actively managing portfolios in the Tehran Stock Exchange.

Keywords : stock price crash tehran stock exchange graph deep learning temporal gat order imbalance liquidity