المفاهيم الأساسية
DeepTraderX, a Deep Learning-based trader, has demonstrated the capability to rival and surpass the performance of established algorithmic trading strategies in a realistic, multi-threaded financial market simulation.
الملخص
The paper introduces DeepTraderX (DTX), a Deep Learning-based algorithmic trader, and presents its performance in a multi-threaded market simulation. DTX was trained on historical Level-2 market data to learn a mapping from market state to trading quotes. The model was extensively tested against well-known trading strategies, including ZIC, ZIP, GDX, and AA, in both balanced group and one-to-many experiments.
The results show that DTX consistently outperforms or matches the profits of these established strategies, highlighting the potential of leveraging simple Deep Learning models to create more efficient financial markets. Key insights include:
- DTX exhibited superior performance in 6 out of 8 experiments, matching profits in 1 out of 8.
- DTX outperformed or matched the profits of 3 out of the 4 traders tested, including those deemed "super-human".
- DTX's performance was particularly strong against GDX and AA, with a win-tie outcome against the latter.
- The results against ZIP were more nuanced, with DTX recording both a victory and a defeat.
The findings underscore the potential of Deep Learning-based trading algorithms to adapt and thrive in realistic, asynchronous market environments, potentially contributing to more efficient and equitable financial markets.
الإحصائيات
The time of the trade when it took place.
The type of customer order used to initiate the trade, either a "bid" or an "ask" order.
The limit price of the trader's quote that initiated the trade.
The midprice of the LOB at the time of the trade.
The microprice of the LOB at the time of the trade.
The LOB imbalance at the time of the trade.
The spread of the LOB at the time of the trade.
The best (highest) bid on the LOB at the time of the trade.
The best (lowest) ask on the LOB at the time of the trade.
The difference between the current time and the time of the previous trade.
The quantity of all quotes on the LOB at the time of the trade.
An estimate of the competitive equilibrium price at the time of the trade.
Smith's α metric using the estimate of the competitive equilibrium price at the time of the trade.
The price of the trade.