Background Economic stability depends on the ability to foresee financial risk, particularly in markets that are extremely volatile. Unbalanced financial data is difficult for traditional Support Vector Machine (SVM) models to handle, which results in subpar crisis detection capabilities. In order to improve financial risk early warning models, this study combines Gaussian SVM with stochastic gradient descent (SGD) optimisation (SGD-GSVM). Methods The suggested model was developed and assessed using a dataset from China's financial market that included more than 2,000 trading days (January 2022–February 2024). Missing value management, Min-Max scaling for normalising numerical characteristics, and ADASYN oversampling for class imbalance were all part of the data pretreatment process. Key evaluation metrics, such as accuracy, recall, F1-score, G-Mean, AUC-PR, and training time, were used to train and evaluate the SGD-GSVM model to Standard GSVM, SMOTE-SVM, CS-SVM, and Random Forest. Results Standard GSVM (76% accuracy, 1,200s training time) and CS-SVM (81% accuracy, 1,300s training time) were greatly outperformed by the suggested SGD-GSVM model, which obtained the greatest accuracy of 92% with a training time of just 180 seconds. Additionally, it showed excellent recall (90%) and precision (82%), making it the most effective and efficient model for predicting financial risk. Conclusion This work offers a new method for early warning of financial risk by combining SGD optimisation with Gaussian SVM and employing adaptive oversampling for data balancing. The findings show that SGD-GSVM is the best model because it strikes a balance between high accuracy and computational economy. Financial organisations can create real-time risk management plans with the help of the suggested technique. For additional performance improvements, hybrid deep learning approaches might be investigated in future studies.

This study reconsiders the timing of the equity risk factors by using the flexible neural networks specified for image recognition to determine the timing weights. The performance of each factor is visualized to be standardized price and volatility charts and `learned' by flexible image recognition methods with timing weights as outputs. The performance of all groups of factors can be significantly improved by using these ``deep learning--based'' timing weights. In addition, visualizing the volatility of factors and using deep learning methods to predict volatility can significantly improve the performance of the volatility-managed portfolio for most categories of factors. Our further investigation reveals that the timing success of our method hinges on its ability in identifying ex ante regime switches such as jumps and crashes of the factors and its predictability on future macroeconomic risk.

This article applies natural language processing (NLP) to extract and quan- tify textual information to predict stock performance. Leveraging an exten- sive dataset of Chinese analyst reports and employing a customized BERT deep learning model for Chinese text, this study categorizes the sentiment of the reports as positive, neutral, or negative. The findings underscore the predictive capacity of this sentiment indicator for stock volatility, excess re- turns, and trading volume. Specifically, analyst reports with strong positive sentiment will increase excess return and intraday volatility, and vice versa, reports with strong negative sentiment also increase volatility and trading volume, but decrease future excess return. The magnitude of this effect is greater for positive sentiment reports than for negative sentiment reports. This article contributes to the empirical literature exploring sentiment anal- ysis and the response of the stock market to news on the Chinese stock market.

Salience theory has been proposed as a new stock trading strategy. Therefore, to assess the validity of this proposal, a complex decision trading system was constructed based on salience theory, a variational mode decomposition (VMD) model, a bidirectional gated recurrent unit (BiGRU) model, and high-frequency trading. The system selected 30 Chinese new energy concept stocks, ranked the stocks using salience theory, and selected the top and bottom three stocks for two portfolios. Twelve stages were established, after which the VMD and BiGRU models were applied to the predictions. The final predicted returns for the high ST group A (GA) were 194.06% and for the low ST group B (GB) were 165.88%. This paper validated the powerful utility of salience theory and deep learning to analyze Chinas new energy market. And it explains the issues and questions raised by previous researchers.

The emergence of deep learning has yielded substantial advancements in computer vision, hence offering novel opportunities for the interdisciplinary exploration of finance and computer science. This paper adopts a cognitive dissonance theory viewpoint to investigate the impact of managers face emotion on market performance and risk in Chinese listed companies from 2016 to 2022. We employ deep learning model to analyze managers’ facial emotion. We find that the more positive facial expressions of managers in earnings conference call predict better market performance, lower volatility and share price crash risk. This study deepens the application of cognitive dissonance theory.

This article applies natural language processing (NLP) to extract and quan- tify textual information to predict stock performance. Leveraging an exten- sive dataset of Chinese analyst reports and employing a customized BERT deep learning model for Chinese text, this study categorizes the sentiment of the reports as positive, neutral, or negative. The findings underscore the predictive capacity of this sentiment indicator for stock volatility, excess re- turns, and trading volume. Specifically, analyst reports with strong positive sentiment will increase excess return and intraday volatility, and vice versa, reports with strong negative sentiment also increase volatility and trading volume, but decrease future excess return. The magnitude of this effect is greater for positive sentiment reports than for negative sentiment reports. This article contributes to the empirical literature exploring sentiment anal- ysis and the response of the stock market to news on the Chinese stock market.

We propose a deep reinforcement learning method to improve pairs trading by identifying nonlinear relationships in stock news. Using the CSI 300 index constituents from 2015 to 2022, we integrated cointegration and news co-occurrence analysis in asset pairing and used a threshold-based approach in trading design. Results showed our NEWS-CO-DRL method, fusing deep learning and news co-occurrence, outperformed in return generation and risk control, indicating its potential for the Chinese A-share market.

We track the fundamental informed traders' (FITs) behavior and show the fundamental momentum effect in the Chinese stock market. We train the deep learning model with a set of fundamental characteristics to extract fundamental implied component from realized returns. The fundamental part characterizes the price movement driven by FITs. Fundamental momentum differentiates from the fundamental trend and is not quality minus junk (QMJ) factor. Underreaction bias helps explain the strategy, as it generates stronger profit during periods of low investor sentiment and aggregate idiosyncratic volatility. Fundamental momentum is not sensitive to changing beta and robust in subsamples and machine learning models.

Fund managers’ ability to evaluate risk has important implications for their portfolio management and performance. We use a state-of-the-art deep learning model to measure fund managers’ forward-looking risk assessments from their narrative discussions. We validate that managers’ negative (positive) risk assessments lead to subsequent decreases (increases) in their portfolio risk-taking. However, only managers who identify negative risk generate superior risk-adjusted returns and higher Sharpe ratios, and have better intraquarter trading skills, suggesting that cautious, skilled managers are less subject to overconfidence biases. interestingly, only sophisticated investors respond to the narrative-based risk assessment measure, consistent with limited attention by retail investors.

A deep learning method is applied to predict stock portfolio allocation in the Chinese stock market. We use 6 original price and volume series as benchmark model settings and further explore the model's predictive performance with social media sentiment. Our results show that our model can achieve a high out-of-sample Sharp ratio and annual return. Moreover, social media sentiment could increase the performance for both Sharp ratio and annual return while reducing annual volatility. We provide an end-to-end stock portfolio allocation model based on deep neural networks.