Forex Trading Using Intermarket Analysis

oUTPUT layer
The output layer is where the network’s forecasts are made. During 
training, the network makes its forecasts, errors are computed and 
“connection weights” between neurons are adjusted prior to the next 
training iteration. Connection weights are altered by an algorithm—the 
“learning law,” including the back-propagation method—to minimize 
output errors. Lots of adjustments may be necessary at any point along 
the way to get the desired results.
Two types of real number outputs in financial analysis include price 
forecasts, such as the next day’s high and low, and forecasts of forward-
shifted technical indicators, such as the five-day moving average value 
for two days in the future. The network developers have to decide not 
only what output to forecast but also how far into the future to make 
the forecast.
Then comes extensive testing to verify the accuracy of the network’s 
forecasts. Testing is performed by creating an independent test file 
of data not used during the training process. In the testing mode the 
neural network is given these new inputs and uses the representation 
that it had previously learned to generate its forecasts so the network 
can be evaluated under real-time conditions. This is analogous to 
“walk-forward” or “out-of-sample” testing of rule-based trading strat-
egies. The developers can compare performance results from various 
networks and decide which network to use in the final application.

---

69
ForeX trading using interMarket anaLysis
As with other aspects of neural network and intermarket analysis 
research, there are a number of ways to evaluate performance of a neu-
ral network-based trading strategy. Traders should not attempt to tweak 
it by making human “adjustments” without going through the whole 
development cycle as such changes could undermine the accuracy 
and integrity of the network’s forecasts and results. That is one reason 
why traders are not given the option within VantagePoint to make any 
change in parameters because the best parameter choices have already 
been defined after more exhaustive research than most traders could 
ever accomplish.
The result is a trading tool that is not only highly accurate but also 
very simple to use even by novice forex traders. Traders do not have 
to be rocket scientists to apply the forecasting capabilities of neural 
networks in trading the forex markets.