Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
This code accompanies the the paper Trading with the Momentum Transformer: An Intelligent and Interpretable Architecture (https://arxiv.org/pdf/2112.08534.pdf).
| Date | Stars |
|---|---|
| 2026-07-24 | 632 |
| 2026-07-25 | 632 |
| 2026-07-28 | 632 |
| 2026-07-30 | 632 |
| 2026-08-06 | 632 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Trading with the Momentum Transformer > [!IMPORTANT] > ## Latest Work > [**DeePM: Regime-Robust Deep Learning for Systematic Macro Portfolio Management**](https://github.com/kieranjwood/deepm) extends the Momentum Transformer to end-to-end portfolio construction, re-implemented in PyTorch. Key contributions: > > 1. **Graph neural networks** encoding macroeconomic priors across assets > 2. **Multi-asset cross-sectional attention** with a causal lag (Directed Delay) mechanism > 3. **Portfolio-level loss** — optimises on a pooled portfolio Sharpe ratio rather than univariate per-asset objectives > 4. **Regime-robust minimax optimisation** — a SoftMin proxy for Entropic Value-at-Risk (EVaR) that penalises the worst historical subperiods > 5. **Realistic transaction costs in the loss** — asset-specific costs baked directly into the training objective > 6. **Two-pass exact gradient accumulation** — correct gradients for the coupled Sharpe-ratio objective at scale > > In backtests from 2010--2025, DeePM roughly doubles the net risk-adjusted returns of classical trend-following and improves upon the Momentum Transformer by approximately fifty percent. See the [paper](https://arxiv.org/abs/2601.05975) and [GitHub](https://github.com/kieranjwood/deepm) for full details. ## About This code accompanies the paper [Trading with the Momentum Transformer: An Intelligent and Interpretable Architecture](https://arxiv.org/pdf/2112.08534.pdf) and additionally provides an implementation for the paper [Slow Momentum with Fast Reversion: A Trading Strategy Using Deep Learning and Changepoint Detection](https://arxiv.org/pdf/2105.13727.pdf). ## Using the code 1. Create a Nasdaq Data Link account to access the [free Quandl dataset](https://data.nasdaq.com/data/CHRIS-wiki-continuous-futures/documentation). This dataset provides continuous contracts for 600+ futures, built on top of raw data from CME, ICE, LIFFE etc. 2. Download the Quandl data with: `python -m data.download_quandl_data <<API_KEY>>` 3. Create Momentum Transformer input features with: `python -m examples.create_features_quandl`. In this example we use the 100 futures tickers which have i) the longest history ii) more than 90% of trading days have data iii) data up until at least Dec 2021. 4. Optionally, run the changepoint detection module: `python -m examples.concurent_cpd_quandl <<CPD_WINDOW_LENGTH>>`, for example `python -m examples.concurent_cpd_quandl 21` and `python -m examples.concurent_cpd_quandl 126` 5. Create Momentum Transformer input features, including CPD module features with: `python -m examples.create_features_quandl 21` after the changepoint detection module has completed. 6. To create a features file with multiple changepoint detection lookback windows: `python -m examples.create_features_quandl 126 21` after the 126 day LBW changepoint detection module has completed and a features file for the 21 day LBW exists. 7. Run one of the Momentum Transformer or Slow Momentum with Fast Reversion experiments with `python -m examples.run_dmn_experiment <<EXPERIMENT_NAME>>` ## Trading with the Momentum Transformer: An Intelligent and Interpretable Architecture > Deep learning architectures, specifically Deep Momentum Networks (DMNs) , have been found to be an effective approach to momentum and mean-reversion trading. However, some of the key challenges in recent years involve learning long-term dependencies, degradation of performance when considering returns net of transaction costs and adapting to new market regimes, notably during the SARS-CoV-2 crisis. Attention mechanisms, or Transformer-based architectures, are a solution to such challenges because they allow the network to focus on significant time steps in the past and longer-term patterns. We introduce the Momentum Transformer, an attention-based architecture which outperforms the benchmarks, and is inherently interpretable, providing us with greater insights into our deep learning tradi
Excerpt of 8,258 characters
Read on GitHub
17
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:99685d28816c231f, topic:deep-learning
matched fp:99685d28816c231f, topic:transformer
matched fp:99685d28816c231f, topic:quantitative-finance