Paper: arXiv 2307.08649
Authors: Lili Wang, Chenghan Huang, Chongyang Gao, Weicheng Ma, Soroush Vosoughi
Abstract
In the pursuit of accurate and scalable quantitative methods for financial market analysis, the focus has shifted from individual stock models to those capturing interrelations between companies and their stocks. However, current relational stock methods are limited by their reliance on predefined stock relationships and the exclusive consideration of immediate effects. To address these limitations, we present a groundbreaking framework for financial market analysis. This approach, to our knowledge, is the first to jointly model investor expectations and automatically mine latent stock relationships. Comprehensive experiments conducted on China’s CSI 300, one of the world’s largest markets, demonstrate that our model consistently achieves an annual return exceeding 10%. This performance surpasses existing benchmarks, setting a new state-of-the-art standard in stock return prediction and multiyear trading simulations (i.e., backtesting).
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper employs advanced mathematical concepts including LSTMs, attention mechanisms, and Tanimoto coefficients for latent topic discovery and expectation modeling, indicating high mathematical density. It demonstrates strong empirical rigor by testing on the large CSI 300 dataset, reporting specific metrics like >10% annual return, and comparing against 16 established baselines on the Qlib platform.
Research Flowchart
flowchart TD
A["Research Goal: Improve Stock Return Prediction<br>by modeling latent relationships & expectations"] --> B["Data Input: CSI 300 Market Data<br>(Stock Prices, Volumes, Financial Reports)"]
B --> C["Methodology: Joint Latent Topic Discovery<br>& Expectation Modeling Framework"]
C --> D["Computational Process: Deep Learning<br>Model Training & Optimization"]
D --> E{"Backtesting & Evaluation"}
E -->|"Performance Metrics"| F["Key Findings & Outcomes"]
F --> G["Annual Return > 10%"]
F --> H["New State-of-the-Art Standard<br>Surpasses Benchmarks"]
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 an attention‑based framework that jointly discovers latent stock topics and models investors' subjective expectations for stock trend prediction.
Source sentence
we present a groundbreaking framework for financial market analysis. This approach, to our knowledge, is the first to jointly model investor expectations and automatically mine latent stock relationships.
Assumes 4
- Investors’ subjective expectations influence stock prices. model
Source sentence
Model the influence of investors’ subjective expectations on stock prices.
- Dynamic topics can be discovered without predefined relationships or external knowledge. model
Source sentence
Automatically identify dynamic topics between stocks without making assumptions or requiring additional knowledge.
- Trading is simulated using a top‑k dropout strategy with k=50 and d=5. execution
Source sentence
the top‑k dropout strategy is a straightforward quantitative investment approach: for each trading day, we hold k stocks, sell d stocks with the worst predicted 1‑day return, and buy d unheld stocks with the best‑predicted 1‑day return. In our experiments, k is set to 50, and d is set to 5.
- Data are split into training (2008‑01‑01 to 2014‑12‑31), validation (2015‑01‑01 to 2016‑12‑31) and test (2017‑01‑01 to 2022‑07‑10) periods. data
Source sentence
We use the data from 01/01/2008 to 12/31/2014 as the training set, the data from 01/01/2015 to 12/31/2016 as the validation set for hyper‑parameter fine‑tuning, and the data from 01/01/2017 to 07/10/2022 as the test set.
Data
China CSI 300 · equities
- China’s CSI 300 market-prices unknown
Source sentence
We run comprehensive evaluations of our framework on the China’s CSI 300 financial markets, from 2008 to 2022.
Costs
Not modeled
Finds 3
- The model consistently achieves an annual return exceeding 10%, surpassing all baselines. performance
Source sentence
our model consistently achieves an annual return exceeding 10%.
- In trading simulation, the method attains an annualized return of 10.63% (0.1063±0.0187) and outperforms all 16 baselines in annualized return and information ratio. performance
Source sentence
Our method surpasses all 16 baselines in annualized return and information ratio.
- The method significantly outperforms all 16 baselines across IC, ICIR, Rank IC, and Rank ICIR metrics with about a 10% improvement over the second‑place model. predictive-accuracy
Source sentence
Our method significantly outperforms all the 16 baselines across all four metrics (IC, ICIR, Rank IC, and Rank ICIR) with around 10% enhancement over the second‑place model for each metric.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 34 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.