Stock prediction using Deep Learning methods

I was on and off with trading for approximately 15 years, most of the time it was my side activity, Information Technology was my primary occupation.

Some years ago I took a deep dive into machine learning/deep learning area, as soon as it was required by my ongoing projects and also very interesting and promising.
At one point, more than 2 years ago, idea of application of neural networks/deep learning hit my mind.
I've started experimenting with different implementations and architectures, gradually getting more and more promising results.
Then I started to apply predictions made by my system to my own real trades, trying to improve the models, data processing, etc.
Simple strategy based on the neural networks' predictions shown APR of ~40% on US stock market.
I decided that I could scale it up and make web service in order to provide access to the my system's predictions and supporting information (like performance metrics, etc.).

It took some time to scale up the system to cover most of S&P500 index constituents, but recently I've launched beta of my service, it is available on http://stocksneural.net/.

I was able to obtain following results:

• Accuracy of trend prediction during forecast period - up to 97%
• APR for simple strategies - up to 80% with percent of profitable deals for some instruments up to 100%.

There are still a lot to improve, and we are working on it, but it will be great if real traders like you will take a look on it and share your impressions and feedback.
Could service like this be useful for you?