Currently I am working on FinAgent, leverage LLM to analyze historical price, LLM agent, reflection, and reinforcement learning to forecast signal for stock. After doing a lot experiments using about 10 stocks such as QQQ, SPY, SQQQ, NVDA, TSLA, KO, it is found that the performance is not stable, i.e. in terms of sharpe ratio and profit-and-loss, sometime beat the simple strategy buy-and-hold, sometimes worse than it. Among the evaluated tickers, QQQ performs a little consistent better comparing with buy-and-hold. From today, I will share the forecast signal of my model, just as a log, to monitor it.
| Date | Predict signal - QQQ |
| 2024-11-12 00:00:00-05:00 | buy |
|---|---|
| 2024-11-13 00:00:00-05:00 | hold |
| 2024-11-14 00:00:00-05:00 | hold |
| 2024-11-15 00:00:00-05:00 | buy |
| 2024-11-18 00:00:00-05:00 | buy |
| 2024-11-19 00:00:00-05:00 | buy |
| 2024-11-20 00:00:00-05:00 | hold |
| 2024-11-21 00:00:00-05:00 | hold |
| 2024-11-22 00:00:00-05:00 | buy |
| 2024-11-25 00:00:00-05:00 | buy |
Backtest performance
Start,2024-01-02 00:00:00
End,2024-11-11 00:00:00
Period,218 days 00:00:00
Start Value,10000.0
End Value,13602.3055009464
Total Return [%],36.0230550095
Benchmark Return [%],28.1853615183
Max Gross Exposure [%],100.0
Total Fees Paid,0.0
Max Drawdown [%],7.9068725639
Max Drawdown Duration,29 days 00:00:00
Total Trades,21
Total Closed Trades,20
Total Open Trades,1
Open Trade PnL,-7.941237337
Win Rate [%],70.0
Best Trade [%],9.9766533408
Worst Trade [%],-3.2657928882
Avg Winning Trade [%],2.9705119048
Avg Losing Trade [%],-1.5992107848
Avg Winning Trade Duration,9 days 20:34:17.142857142
Avg Losing Trade Duration,5 days 12:00:00
Profit Factor,4.2586034362
Expectancy,180.5123369142
Sharpe Ratio,2.8644294013
Calmar Ratio,8.5219676323
Omega Ratio,1.5696015704
Sortino Ratio,4.6592903051
