深度学习与台指期交易-打造天才交易机器人.pdfVIP

深度學習與台指期交易- 打造天才交易機器人 凱德資本 Kevin Agenda ● Learning Process for Trading ● Innovative Thinking Deep Learning ● Our Trading System ● Backtesting Result Real-World Performance ● Future Work ● Our Team Learning Process for Trading ● Observations ○ K bar ○ Volume ○ Technical signals ● Paper trading ○ Time period ○ Number of transactions Learning Process for Trading ● Trading ○ Revise strategy based on performance ○ Emotional behaviours Automated Trading System ● Concept ○ Establish specific rules ○ Run automatically with computers ○ Usually based on the trader’s observations and experiences Automated Trading System ● Pros ○ Minimize emotions ○ Ability to backtest ○ Preserve discipline ○ Improved order entry speed Automated Trading System ● Cons ○ Mechanical failures ○ Monitoring ○ Over-optimization ○ Still limited to the trader itself Innovative Thinking Deep Learning ● Successful trading: Long at low, short at high ○ But, where is the market bottom or top? ● Deep learning: 21st century’s statistics ○ Scalability ○ No need of manual feature extraction ○ Effectively handling time-series Innovative Thinking Deep Learning ● Technical analysis: market is self-explanatory ○ By analyzing the market data with deep learning, we can tell precisely when to long and when to short Our Trading System ● Pros ○ Capacity ○ Low human effort in quantitative research work ○ Unlimited number of trading signals ● Cons ○ Black box Backtesting Result ● Commodity: TXFF ● Interval: 2012/5 - NOW ● Assumption: ○ Initail Capital ：NTD 3M ○ transaction cost ：NTD 500 per trade ○ No switching cost Backtesting Result ● Outcome ：