>The reason that your system was on the correct side of those moves
>is that it was trained on them. This is good and bad. Obviously
>those moves account for a lot of prediction error, so when the net
>is training it gives a lot of weight to getting that week right,
>possibly at the expense of the rest of the time frame tested and
>trained.
Actually, the test was completely out-of-sample, meaning the training stopped on 06/30/97. The results you see should be close to what would have been achieved if I followed the system religiously and had perfect executions.
The NS Trader nets have mechanisms built in to weight-down the outliers (rare occurances) in the data, so even if I did use this period it should affect the nets significantly. This is one of the reasons I am such a devoted customer to their nets. In my opinion, they are the best in the world (at the price).
>Only experience, luck, and lots of research can possibly position
>you properly for a market event. Best bet, if you make good money
>trading on normal days, take a vacation on abnormal ones, you wont
>make THE BIG trade, but you wont be on the wrong side either.
The idea behind trading systems (neural or otherwise) is to distill a traders experience of the market into an objective and mechanical approach. The research comes into play during the design period. If you have a working system and have put the time and effort into designing a robust system, you shouldn't second guess it. Every system I run is either long or short. I am never out of the market. This may by a little riskier than most, but it fits my trading style and I am comfortable trading it.
Sorry for the delay in response, as I somehow missed this message.
Optim |
