NEW:
ADVANCED
TRADESTATION TECHNIQUE
Introduction Notes for NeuroShell Trader Users
We once had in mind to build a collection of indicators for NeuroShell Trader. This project is suspended on account of lack of resources. We however have coded a few inds listed here .
-
GPF Notes for NeuroShell Trader Users
The NeuroShell Trader Pro certainly includes very powerful optimization techniques. Building recursive indicators is virtually limitless, and genetic algorithms can optimize almost everything in the program. Having said that, the NeuroShell Trader has been designed with a holistic approach in mind, and on specific issues, a more focused approach may be more appropriate:
1. Most important, the G.P.F. takes trading factors like stop loss orders in early consideration whereas the Trader optimizes inputs and then trains them in a prediction and / or trading strategy. We believe that the G.P.F. better segregates winning patterns in applying signal filters at an earlier stage regardless of the overall trading strategy. This allows basically to segregate much better entries, and short term traders often do not have the time to put a complex money management strategy in place, and rely mostly on entries.
2. The G.P.F. gives a better control over the genetic algorithm process. They is virtually no chance a good pattern can be missed. For performance reasons, early stopping techniques may cause the NeuroShell Trader to miss them.
3. The G.P.F records not only the best pattern but lists the n best patterns (by default n=10). The best patterns are actually often the second or third listed, from a statistical stability viewpoint.
4. The G.P.F. allows you to build consensus based scoring system from your patterns. For instance, you can build a composite indicator based on the top 3 patterns, eventually weighted by their position, or the equity generated by each. This can certainly alleviate the common problem with predictive neural nets which are generally poor with boolean inputs.
5. The NeuroShell Trader cannot (or tell us how!) simulate our pyramiding function. It cannot filter outliers or noise, or at the very least, to be conservative, we have no idea how.
It must be said that it is virtually impossible to fully mimic the GPF within the NS Trader, even though the actual pattern construction can be simulated. That's what the GPF Add-on for NS Trader does as a matter of fact. It is recommended NOT to use the Trader Optimizer with the GPF, if fed with data already optimized by the GPF.
Having said that, we could be wrong. We agree that the GPF added value for a NeuroShell Trader user is a little lower than for other traders. It may extremely tedious to replicate parts of the GPF in the Trader, but it may not be quite impossible. This is why we offer a substantial discount for NS Trader registered users. See our Sales Page for details.
The new GPF Indicator is now available. See details below.
A GPF Add-on for NSTrader is now available. It replaces the former GPF Interface for the Trader, and can now work on its own, (i.e. without the GPF) as a NS Trader Professional Add-on. This means that you can now use either the genetic optimization embedded in the Trader, or the GPF. Our experimentations show that the patterns detected by the Trader are different from the GPF's, which makes both approaches complementary.
This implementation cannot provide all the levers including in the GPF. On the contrary, the GPF pattern formation has been kept simple like for instance:
| IF | Open of yesterday > High of 2 days ago |
| AND | Close of 4 days ago > High of 3 days ago |
| THEN | 1 |
| ELSE | 0 |
It is a boolean indicator which purpose is here to provide a new uncorrelated input to the NSTrader neural net. Boolean inputs are usually not great inputs. It is recommended to mix with other inputs, and possibly combine several of those boolean into a scoring system.
Here is an example achieved with the NS Trader Pro and the GPF Add-on. The data used is E-Mini Nasdaq-100 March 2002 (Symbol NQH2). It speaks for itself! This model uses a $20 round commission. Minimum training size = 1 month, using 1 walk-forward of 4 days.
Trading Statistics:
Not too bad, isn't it? Makes money on longs and shorts, good success rate...
But guess what: this prediction only uses ONE input, a simple GPF pattern!
Of course, this is very different from the original GPF patterns which are optimized for longs or shorts, but why not being a little inventive sometimes...
Please note that performance may be lower on other instruments, other time frames, other periods, but this powerful indicator should in any case improve your trading models.
Read about our special pricing on the Sales page.