QuantNet: transferring learning across trading strategies

Systematic financial trading strategies account for over 80% of trade volume in equities and a large chunk the foreign exchange market. In spite availability data from multiple markets, current approaches rely mainly on learning per individual this paper, we take step towards developing fully end-to-end global that leverage systematic trends to produce superior market-specific strategies. We introduce QuantNet: an architecture learns market-agnostic use these learn Each model is composed encoder-decoder pair. The encoder transforms into abstract latent representation processed by shared all while decoder strategy based both local information model. QuantNet uses recent advances transfer meta-learning, where parameters are free specialize problem at hand, whilst driven capture signals markets. By integrating idiosyncratic market can general transferable dynamics, avoiding overfitting with returns. evaluate historical across 3103 assets 58 equity Against top performing baseline, yielded 51% higher Sharpe 69% Calmar ratios. addition, show benefits our approach non-transfer variant, improvements 15% 41% A link code made available appendix.