Hello everyone, I'm new to this forum and now I'm facing a problem in machine learning. Right now I'm writing my own algorithm which has a set of parameter/factors to optimize. And I do want this algorithm is universal, i.e., it doesn't only apply to one or a few stocks but applied to the entire market. So now the only algorithm come up to my head is the neural network, but I don't know how to set up. I can easily use sharpe ratio as my fitness function, but how to I change the parameters? Can someone give me a hand? Thanks.
%%%%%%%%%%%%%%%%%%%%%%%%%%%% Yes,Nick; forget nnetworks + put a 50 day ma,[which is a 50 week ma on weeklys] 200day moving average......................................................... Study that for as many years as you can; wisdom is profITable TO DIRECT
Although I do not wish to dampen your enthusiasm, as regardless of the outcome I believe the exercise shall be to your benefit, I will recommend that you stay away from NNs. If you wish to persist with the NN quest to (no?)cash, then google is your friend.
I tried Neural Networks in trading and it works reasonably well. However I was getting best results when I trained NN on a particular currency pair and particular timeframe. I was using EURUSD 15 minutes timeframe. The NN works for a while and then its performance deteriorate. So I have to keep retraining it every month.
What you are attempting to do is not entirely clear, but it sounds like you are trying to solve an optimization problem. If this is the case, a neural network is not what you want. A neural network is essentially a universal function approximator. In other words, (typically) given a set of input and its corresponding output, a neural network will approximate (perhaps exactly) the function that maps the input to the output. It sounds like you already know the function though, and simply want to find the parameters that maximizes the Sharpe ratio. If a neural network is indeed what you want, I have written open source (free under the GNU General Public License) C++ software that you might find helpful: http://www.deepalgorithms.com/software/machine-learning/ You would need to know C++ though to write an interface specific to your problem. If you are in fact trying to solving an optimization problem (as I suspect), and depending on the complexity of your problem, I can instead refer you to other open source C++ software that I have written to solve optimization and search problems using genetic algorithms.