|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- finance |
|
|
- trading |
|
|
- lstm |
|
|
- differential-privacy |
|
|
- turkish-market |
|
|
- bist |
|
|
language: |
|
|
- tr |
|
|
- en |
|
|
library_name: pytorch |
|
|
pipeline_tag: tabular-classification |
|
|
--- |
|
|
|
|
|
# BIST DP-LSTM Trading Model |
|
|
|
|
|
Differentially Private LSTM ensemble for Turkish stock market (BIST) price prediction with sentiment analysis integration. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Developed by:** rsmctn |
|
|
- **Model type:** PyTorch Differential Privacy LSTM Ensemble |
|
|
- **Language:** Turkish, English |
|
|
- **License:** MIT |
|
|
- **Repository:** [BIST_AI001](https://github.com/RSMCTN/BIST_AI001) |
|
|
|
|
|
## Model Architecture |
|
|
|
|
|
This model combines multiple approaches: |
|
|
|
|
|
1. **DP-LSTM Core**: Multi-task LSTM with differential privacy (Opacus) |
|
|
2. **Temporal Fusion Transformer**: Advanced attention mechanisms for financial sequences |
|
|
3. **Simple Financial Transformer**: Lightweight transformer for rapid inference |
|
|
4. **Ensemble Weighting**: Dynamic model combination with confidence estimation |
|
|
|
|
|
## Training Data |
|
|
|
|
|
- **BIST Historical Data**: 2019-2024 (BIST 30 stocks) |
|
|
- **Technical Indicators**: 131+ features across multiple timeframes (1m, 5m, 15m, 60m, 1d) |
|
|
- **News Sentiment**: Turkish financial news corpus with VADER + FinBERT |
|
|
- **Privacy Protection**: ε=1.0 differential privacy with adaptive noise calibration |
|
|
|
|
|
## Performance Metrics |
|
|
|
|
|
- **Direction Accuracy (MVP)**: ≥68% |
|
|
- **Direction Accuracy (Production)**: ≥75% |
|
|
- **Sharpe Ratio**: >2.0 |
|
|
- **Max Drawdown**: <15% |
|
|
- **Signal Confidence**: 65-95% range |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
# This is a demo model - full implementation in production system |
|
|
import torch |
|
|
from transformers import AutoModel |
|
|
|
|
|
# Load model (demo) |
|
|
model = AutoModel.from_pretrained("rsmctn/bist-dp-lstm-trading-model") |
|
|
|
|
|
# Production usage requires full system: |
|
|
# https://github.com/RSMCTN/BIST_AI001 |
|
|
``` |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
**Primary Use Cases:** |
|
|
- Turkish stock market research |
|
|
- Algorithmic trading signal generation |
|
|
- Financial sentiment analysis |
|
|
- Academic research in privacy-preserving ML |
|
|
|
|
|
**Limitations:** |
|
|
- Demo version for research purposes |
|
|
- Requires full system for production use |
|
|
- Not financial advice |
|
|
|
|
|
## Ethical Considerations |
|
|
|
|
|
- **Privacy**: Differential privacy protects individual trader data |
|
|
- **Bias Mitigation**: Diverse training across market conditions |
|
|
- **Transparency**: Open-source implementation |
|
|
- **Responsible AI**: Clear disclaimers about financial risks |
|
|
|
|
|
## Citation |
|
|
|
|
|
```bibtex |
|
|
@misc{bist_dp_lstm_2024, |
|
|
title={Differential Privacy LSTM for Turkish Stock Market Prediction}, |
|
|
author={rsmctn}, |
|
|
year={2024}, |
|
|
url={https://github.com/RSMCTN/BIST_AI001} |
|
|
} |
|
|
``` |
|
|
|
|
|
## Contact |
|
|
|
|
|
- **GitHub**: [rsmctn](https://github.com/RSMCTN) |
|
|
- **Repository**: [BIST_AI001](https://github.com/RSMCTN/BIST_AI001) |
|
|
- **HF Spaces Demo**: [Trading Dashboard](https://huggingface.co/spaces/rsmctn/bist-dp-lstm-trading-dashboard) |
|
|
|
|
|
--- |
|
|
|
|
|
⚠️ **Disclaimer**: This model is for research and educational purposes only. Past performance does not guarantee future results. Always consult financial advisors before making investment decisions. |
