Hyperparameter Configuration for a Transformer-Bayesian Asset Pricing Model

Minute-level high-frequency data for CSI 300 constituent stocks from 2019 to 2024 was used to evaluate a novel Transformer-based Efficiently-Fused Optimized Bayesian Network (Trans-EFOBN) model. The dataset, shared by Qi Fu on figshare, contains hyperparameter configurations for the model, which achieved a mean absolute error of 0.037 and an R² of 0.86. The model was tested in simulated trading scenarios with transaction costs, yielding a 14.2% annualized return and a Sharpe ratio of 0.95.

Use Cases

• Tuning hyperparameters for transformer-based sequential feature extraction in financial time series.
• Optimizing Dynamic Bayesian Network parameters for modeling hierarchical dependencies between macro and micro market variables.
• Benchmarking new asset pricing models against the reported MAE and R² performance metrics.
• Simulating trading strategies using the described model framework to evaluate risk-adjusted returns.

Strengths

• Model performance is quantitatively validated with specific metrics: MAE of 0.037 and R² of 0.86.
• Temporal coverage is explicitly defined as 2019 to 2024 for CSI 300 constituent stocks.
• Trading performance is reported with concrete figures: 14.2% annualized return and a Sharpe ratio of 0.95.
• Dataset is openly shared under a CC-BY-4.0 license, facilitating reuse.

Limitations

• The dataset is extremely small at 5.5 KB, suggesting it contains only configuration parameters, not the underlying market data.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain tasks.

Provenance

Source

Wind database (as cited in the description for the underlying market data).

Collection Method

Likely contains the hyperparameter configuration used for the empirical analysis described.

Time Range

2019 to 2024

Freshness

Last updated 2026-05-08 17:28:08; freshness should be verified.

Geography

China (CSI 300 index constituent stocks)