Hi, I have just terminated a test on a artificial intelligence distributed system that provides 3-days-in-the-future forecasts. What do you think about these results?
About the test:
1) Systems doesn’t know any kind of data coming from the future
2) There’s not overfitting in the test beacause I have not tuned the system on a
particular time window in order to obtain good results.
Index: DowJones (DJI)
Test window: 01-08-2003 to 30-08-2005 (540 market days)
Total number of forecasts: 77
Total number of correct BUY operations: 41
Total numerr of wrong BUY operations : 13
Total number of correct SELL operations:11
Total number of wrong SELL operations: 3
Allocation of the capital: 26,66% (days in which capital is invested)
Forecasts correctness: 76,47%
Average gain per operation: 0,987%
Length of longest correct operation series: 9 (from 30/10/2003 to 11/02/2004)
Length of longest wrong operation series:2
Gain in the window: 45,91% (with capital not invested in 74,4% of the days)
Hello, it’s good result. I work on the same issue it’s very interessant. More over you can have an infinite utilitie with this tool. But I’m interrest to know what kind of language do you use? Thank you
The real question is how much did you outperform against the market. I’ve been programming on bikresearch.si project for about 4 years and we have always use that metric. If you can do a backtesting simulation for any S&P100 stock for last 100 days we can compare results.
our system is formed by about 700 pages of java code and works on four QuadCore servers. What do you mean for “infinite utilitie”?
thanks for your observation, the difference of the DJI in the same period is 30,4%. But our system had a gain of 45,91% investing capital only in 26,6% of the days. In the remaining time we can invest in other Indexes (our tests show similar results on others indices).
—NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!