The Wave59 software program contains many improved versions of popular trading techniques. Our Fractal Trend index resembles the behavior of the better known ADX indicator, and our Ultra Smooth Momentum curve is like many of the more widely used momentum curves but without the noise. These two tools are examples of improved algorithms. In other words, our indicators outperform the popular versions because we are calculating them using a better formula. Another way to improve an indicator without having to redesign the algorithm is to use artificial intelligence techniques.

Wave59 contains an advanced AI module that allows it to apply neural networks to a particular problem the user has specified. For example, let's suppose that we use momentum frequently in our own trading. We are happy with our indicator, but can't help but wonder if there isn't some way to make it even better. Wouldn't it be nice if our momentum curve gave us signals earlier? That would result in more accurate entries and bigger profits.

All of this can be accomplished with neural networks. We can take a number of different length momentum curves, pass them to the neural network, and ask it to use those inputs to improve our original momentum curve. Take a look at the chart below:

The pink curve at the bottom is our 9 bar momentum curve, and the green line is what our neural network came up with. Notice how the green line leads the original both to the upside and downside. Zero line crossovers occur 2-3 bars earlier as well. Now, the improvement is not spectacular, but getting a signal 2-3 bars ahead of time is nothing to laugh at. The trader using the neural net on the chart above is effectively 15 minutes ahead of the trader without it. In day trading time, 15 minutes can mean the difference between a profit and loss.

The difference between our nets and everything else we've ever seen lies in the way our nets train. We use a proprietary genetic algorithm to teach the neural net the patterns in the data. Genetic algorithms are optimization tools based on evolutionary theories. Chromosomes containing neural network data are mated, mutated, and sorted in the same way as plant and animal life in nature. It's just survival of the fittest, and the best chromosome at the end of the training period is the one that gets used in the neural net. This is not the place to go into detail about genetic algorithms, but their main strength lies in the fact that they can sort through an enormous number of possible combinations in a very short amount of time. In the 30 seconds that it took to train the network demonstrated above, our genetic algorithms covered a search space of 2.0 x 10^185. That's a 2 with 185 zeros after it, a REALLY big number.

Other networks use genetic algorithms, but only for part of the training, such as selecting inputs. Then they fall back to the old standby algorithms like Back Propagation to actually train the network. We use one genetic algorithm for everything including neuron weights, network structure, and input selection. Our genetic algorithm has complete control over all phases of network training and structure.

Those last two paragraphs might not mean much to you if you're a trader. Who cares what training algorithm is used, as long as the indicator produced is a good one? And if you did ask that, you would definitely have a point. The reason we went into all of this is as a lead in to explain why our nets can do one spectacular thing that others can't. By using genetic algorithms in this way, we can ask our net to perform an incredible task:

Usually when you train a neural net, you need to provide some kind of target indicator for it to try and imitate. This is almost always some kind of momentum curve, shifted 3-4 bars into the future so it is profitable. But since our neural nets don't use conventional training algorithms, we don't actually need a target indicator at all. We can just say, "Here are some inputs. Combine them so that the result is profitable." Below is a graphic detailing all the information Wave59 needs to get started building profitable indicators for you.

Want a good overbought/oversold indicator? Dump a couple Stochastic indicators into the network, set your buy level to -0.8 and your sell level to 0.8 and let the software build you one! How about something to determine the trend in the market? Use the Fractal Trend Index plus a few momentum curves, set the buy and sell levels to 0, and in less than a minute the software will present you with an indicator that has made money on your data by determining the trend. The possibilities are endless! This is what artificial intelligence was meant to be like!

Let's demonstrate this with an example. We fed the neural network our Periodogram indicator (overbought/oversold) and Smart Money Index (trend following). We told it to buy on a cross above -0.65 and sell on a cross below 0.65. We used 5 days of 10 minute data to do the training, which gave us the following results:

These results are really spectacular. 92% accuracy with over $12K profit in only one week! Of course this could never be attained since this was the training sample, but it does demonstrate that the neural network is really good at finding profitable systems. Now let's look at what happened over the next two days when we would have been actually trading:

(Note to sharp-eyed readers: Although this report starts on the 11th, this was done only to allow
enough room to calculate the indicators so the 12th would be complete. So this out of sample set has
not been contaminated by training data. This can be verified by checking the Trade by Trade
results at the bottom of the report.)

These results are more modest, but are still very good. $3,000 profit in only two days trading is quite an accomplishment, even if it was accompanied by 40% accuracy. The important thing is that our new indicator actually makes money by itself. By using other tools in addition to our neural network, we really start to stack the deck in our favor.

Other traders have a tool kit full of indicators that are all unprofitable on their own. Stochastics, RSI, and moving averages all fail miserably when followed exclusively. But this software has the ability to build a tool kit full of enhanced indicators that each make money when used alone. On one hand you have a trader with a bunch of unprofitable tools, and on the other a trader with profitable tools. Who will be more successful in trading the markets? The tools you use are the foundation to your trading success, so why trade with unprofitable tools?

Click here to order Wave59, or here to return home.