Paper: arXiv 2309.11979

Authors: Jiashu Lou

Abstract

With the rapid development of big data and computing devices, low-latency automatic trading platforms based on real-time information acquisition have become the main components of the stock trading market, so the topic of quantitative trading has received widespread attention. And for non-strongly efficient trading markets, human emotions and expectations always dominate market trends and trading decisions. Therefore, this paper starts from the theory of emotion, taking East Money as an example, crawling user comment titles data from its corresponding stock bar and performing data cleaning. Subsequently, a natural language processing model BERT was constructed, and the BERT model was fine-tuned using existing annotated data sets. The experimental results show that the fine-tuned model has different degrees of performance improvement compared to the original model and the baseline model. Subsequently, based on the above model, the user comment data crawled is labeled with emotional polarity, and the obtained label information is combined with the Alpha191 model to participate in regression, and significant regression results are obtained. Subsequently, the regression model is used to predict the average price change for the next five days, and use it as a signal to guide automatic trading. The experimental results show that the incorporation of emotional factors increased the return rate by 73.8% compared to the baseline during the trading period, and by 32.41% compared to the original alpha191 model. Finally, we discuss the advantages and disadvantages of incorporating emotional factors into quantitative trading, and give possible directions for further research in the future.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs only partially modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.0)

Why this score: The paper applies fine-tuned BERT and regression models to stock sentiment data with a reported backtest showing a 73.8% return increase, indicating strong empirical implementation, but the mathematics is primarily application of existing models with minimal novel theoretical derivation.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Quantify market sentiment impact on trading"] --> B["Data Acquisition & Cleaning<br>East Money user comments"]
  B --> C["Method: BERT Fine-tuning<br>vs Baseline Models"]
  C --> D["Process: Sentiment Polarity Labeling<br>for comment data"]
  D --> E["Integration: Combining Sentiment<br>with Alpha191 Model"]
  E --> F["Outcome: Regression &<br>5-Day Price Prediction"]
  F --> G["Findings:<br>73.8% return increase vs baseline<br>32.41% vs Alpha191 alone"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a BERT-based sentiment analysis model for stock forum posts and uses its polarity signals in an automatic trading system.

Source sentence
paper uses the BERT model to perform sentiment analysis on the post titles, and uses the sentiment polarity values as trading signal factors, and develops an automatic trading system.

Assumes 5

  • A transaction fee of 0.5% is applied to each trade. execution
    Source sentence
    We set the transaction fee to 0.5%, and the initial capital to 100000.
  • Each purchase must be held for at least 10 days before the next decision. execution
    Source sentence
    we also stipulate that each purchase must be held for at least 10 days before making the next decision
  • Buy signals are triggered when the predicted price change exceeds the 90th percentile of the test set; sell signals when below the 10th percentile. model
    Source sentence
    buy 1000 shares of the target when the prediction is greater than the 90th percentile of the test set, and sell 500 shares of the target when it is less than the 10th percentile.
  • The THUCTC Chinese text classification dataset provides appropriate sentiment labels for fine‑tuning BERT. data
    Source sentence
    we have to rely on existing labeled datasets, here we use THUCTC Chinese text dataset.
Show 1 more
  • The 90,000 East Money comment titles from 2019‑01‑01 to 2020‑01‑01 are representative of market sentiment. data
    Source sentence
    about 90,000 comment titles from January 1, 2019 to January 1, 2020 are selected as experimental data.

Data

Backtest sample: equities

  • THUCTC Chinese text dataset · Tsinghua University text-news public
    2005-2011
    Source sentence
    THUCTC (THU Chinese Text Classification) is a Chinese text classification toolkit
  • East Money Stock BBS comment titles · East Money alternative public
    2019-01-01 to 2020-01-01
    Source sentence
    about 90,000 comment titles from January 1, 2019 to January 1, 2020 are selected as experimental data.

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Partial

Paper text extraction: Modeled - 0.5% transaction fee per trade

Source sentence (the authors' words)
We set the transaction fee to 0.5%, and the initial capital to 100000.

Finds 3

  • Incorporating emotional factors increased the return rate by 73.8% compared to the baseline strategy. performance
    Source sentence
    the incorporation of emotional factors increased the return rate by 73.8% compared to the baseline during the trading period
  • The emotional‑enhanced model outperformed the original Alpha191 model by 32.41% in return rate. performance
    Source sentence
    and by 32.41% compared to the original alpha191 model
  • The fine‑tuned BERT model achieved performance improvements over the original BERT and baseline models. predictive-accuracy
    Source sentence
    the fine‑tuned model has different degrees of performance improvement compared to the original model and the baseline model

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.