Deep learning on the order book (with a reality check)
Modelling high-frequency limit order book dynamics with support vector machines
Hand-crafted LOB features (spreads, imbalances, arrival intensities) fed to SVMs. The feature list is still worth borrowing.
Benchmark dataset for mid-price forecasting of limit order book data with machine learning methods
FI-2010: a public 10-level LOB dataset for five Nasdaq Nordic stocks. The standard (if small) benchmark for LOB models.
DeepLOB: Deep Convolutional Neural Networks for Limit Order Books
CNN plus inception module plus LSTM on the top 10 book levels to predict mid-price direction. The most-cited deep LOB model and the usual baseline.
Universal features of price formation in financial markets: perspectives from deep learning
One deep model trained on hundreds of stocks predicts price direction better than stock-specific models: price formation looks universal.
Deep order flow imbalance: Extracting alpha at multiple horizons from the limit order book
Deep nets fed stationary order-flow-imbalance features beat nets fed raw book states, with predictability strongest at the shortest horizons.
LOB-based deep learning models for stock price trend prediction: a benchmark study
Re-runs 15 published deep LOB models on common data: most fail to generalise beyond FI-2010. A reality check before trusting DeepLOB-style results.