If you read my work you always start with a system that makes money. The neural network,SVM are used to improve existing systems that already make money. For example you can use these technologies to predict a momentum indicator which was used in the system Then substitute out the original one with the new predicted one to improve the system's performance.
All this assumes NN 'predictions' are actual improvements. Most predictions are crap. Where's the evidence that NN predictions, especially regarding something as mercurial as a financial market, are worth a damn? More to the point, if neural nets can accurately predict momentum, then they can also predict future prices. Why do you even need a working system with a price predictor at hand? Admit it, this is just another scam to separate suckers from their money.
i agree. the only one I have seen that openly takes its own medicine is www.aiforecasts.com who cares what it uses, as long it actually helps when fighting the monster markets
Interesting. How do you take the output of the NN/SVM and convert that into a trading system with rules? If you were using a random forest or a decision tree, then you could measure the variable importance and then use the significant variables. But how do you trust the predictions of an opaque NN or SVM?
When you predict momentum using neural networks /SVM the improvement comes from two places , first they can filter noise and remove some of the lag of the moving averages /filters in the indicator. For example if we are predicting MACD, we would feed preprocessed raw price data into it, so all we are doing is creating Moving average oscillator with less lag than a standard EMA. Not hard to believe. Since it's a non linear model it can do a better job than regression. It can also find patterns for example predict turns when you statistically move too far in a given direction. The other side is true prediction, this is more complex and requires understand what drives a given market. For example if we were predicting the S&P500 , using things like advance/decline ratios,COT data, sentiment data, plus all the preprocessed price data could lead to predictions with some skill score.
Do you realize how hilarious the phrase "predicting MACD" is? MACD is a crap indicator. I call it Moving Average Crossover Delusion. Why anyone would want to predict MACD or similar clowndicators is baffling. There are some good technical indicators but MACD sure ain't one of them. Why not predict straight-up momentum? No EMAs, no 'histogram' nonsense, just a real important indicator. I'll ask for the last time and let your response speak for itself: Where is the evidence that these NN predictions are worthwhile? Actual evidence, not just vendor claims of glory. Real products have real product reviews. It's not hard to find vehicle performance data. Consumer Reports reviews hundreds of different types of products and makes available their pros and cons. Only trading software vendors expect their potential customers to just 'take our word for it.' One has to wonder why.