Deep Learning Finds Shared Order-Book Patterns in Equity Price Formation
Summary
This study uses deep learning on a high-frequency database of US equity quotes and trades to model how order-book supply and demand relate to later price changes. It tests predictions of price-move direction using histories of prices and order flow, evaluating performance across stocks and time periods. The model is trained across many stocks rather than separately for each asset.
The authors report stable out-of-sample accuracy across sectors and periods, including for stocks excluded from training. The pooled model outperforms asset-specific linear and nonlinear models in their comparisons, supporting shared patterns in price formation and the value of combining data across stocks. Standard normalizations and splitting training data into sectors or tick-size groups did not improve the reported results, while including more past observations did. The findings indicate path dependence, but the document gives no numerical accuracy, sample dates, or trading-cost analysis. Directional prediction results therefore do not establish that the model would yield profitable trades after execution costs.
Key ideas
- A deep learning model relates order-book supply and demand to subsequent equity price changes.
- Pooling data across stocks outperforms the tested asset-specific linear and nonlinear models.
- The reported predictive accuracy is stable across stocks and periods, including stocks outside the training sample.
- Longer histories of prices and order flow improve forecasting, suggesting path dependence.
- The document reports directional prediction results but does not establish profitability after trading costs.
Tags
Full text
# Universal features of price formation in financial markets: perspectives from Deep Learning # Universal features of price formation in financial markets: perspectives from Deep Learning Using a large-scale Deep Learning approach applied to a high-frequency database containing billions of electronic market quotes and transactions for US equities, we uncover nonparametric evidence for the existence of a universal and stationary price formation mechanism relating the dynamics of supply and demand for a stock, as revealed through the order book, to subsequent variations in its market price. We assess the model by testing its out-of-sample predictions for the direction of price moves given the history of price and order flow, across a wide range of stocks and time periods. The universal price formation model is shown to exhibit a remarkably stable out-of-sample prediction accuracy across time, for a wide range of stocks from different sectors. Interestingly, these results also hold for stocks which are not part of the training sample, showing that the relations captured by the model are universal and not asset-specific. The universal model --- trained on data from all stocks --- outperforms, in terms of out-of-sample prediction accuracy, asset-specific linear and nonlinear models trained on time series of any given stock, showing that the universal nature of price formation weighs in favour of pooling together financial data from various stocks, rather than designing asset- or sector-specific models as commonly done. Standard data normalizations based on volatility, price level or average spread, or partitioning the training data into sectors or categories such as large/small tick stocks, do not improve training results. On the other hand, inclusion of price and order flow history over many past observations is shown to improve forecasting performance, showing evidence of path-dependence in price dynamics.
Shown in full with attribution under the source's licence. Licence: abstract CC0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.