The Prediction Wizard automatically builds an abbreviated trading strategy for you when you train a net. However, more advanced users may want to insert neural nets made with the Prediction Wizard into a more complicated trading strategy. Some reasons for doing this might include:
1. You would like to add trailing stops.
2. You would like to enhance the buy and sell rules with non-neural net conditions, i.e., hybrids with nets and traditional rules in the same strategy.
3. You would like to have strategies with several nets in them. You can insist that all nets be giving a signal before you buy/sell or 1 net, etc.
There are several ways you can put your nets into the Trading Strategy Wizard:
1. The easiest way is illustrated in Example 20 that is distributed with NeuroShell. There we built a Trading Strategy in which we inserted 3 nets. The rules are already made for you ready to insert. They look like this long entry rule:
Pred > Long Entry Threshold: Predicted 10 bar %Change in Open
These are threshold rules already coded for you the way the Prediction Wizard used them, and you’ll find similar rules for long exit, short entry, and short exit. Just put them in the Trading Strategy where you put the other rules. But NOTE – you must put in all 4 of the rules.
2. You can insert your own threshold rules for your neural nets using the A>B and A<B indicators. Here, you will make A the neural net and B will be some threshold you pick, or you will let the optimizer pick it. A proper threshold was already picked by the Prediction Wizard, but another one might be more appropriate after you add more nets or non-neural rules. For the A part, you will want to use the “Prediction Signal:…” data stream.
Instead of thresholds like A>B you can use the crossabove/below indicators. That way if you decide to add a trailing stop or non-prediction exit rule you will avoid the problem of entering the position again immediately after you exit.
3. You can forget the Prediction Wizard all together and use neural nets already built into indicators. In this way, the training can be optimized at the same time all other parameters are optimized. We sell these as add-ons and they include:
Adaptive Net Indicators