Sharpe ratio above 2.90. LSTM learns to trade momentum… no indicators, just data.

A classic paper from Oxford researchers shows how to supercharge time series momentum strategies using deep learning. Instead of relying on handcrafted trend signals (like MACD or moving averages), the Deep Momentum Network trains an LSTM to learn both trend and position sizing, directly optimized for Sharpe ratio.

• Tested on 88 global futures (commodities, FX, equity indices, bonds),

• Sharpe ratio jumps from 1.39 (classic TSMOM) to 2.91,

• Works even with 2-3 bps of transaction costs,

• And improves further with turnover regularization for illiquid assets.

Forget thresholds and indicators. This LSTM learns how to size positions directly from raw data.