Deep Learning Models of High Frequency Financial Data
Using a Deep Learning approach applied to a large dataset of high frequency financial data, we find evidence for a universal and stationary price formation mechanism relating the supply and demand for a stock, as revealed through the order book, to price dynamics. We build a ‘universal price formation model’ which demonstrates stable accuracy across a wide range of stocks from different sectors and for long time periods. The universal model, trained on data from all stocks, outperforms asset-specific linear and nonlinear models trained on time series of any given stock. This shows 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. We also find that price formation has path-dependence over long periods of time (‘long memory’). Joint work with Rama Cont.