Well, start trading it already! I too am cynical, but then this makes me want to see the real actual trading results. I suspect that some of the trades won't get filled. Also are you including commisions in your analysis? The difference is small but if you make $146 per trade then minus $9 commish would be $137 per trade.
Yes I am including commission. I am also including 1 tick of slippage. All trades are done with market orders, so getting filled shouldn't be a problem, however the 1 tick of slippage may be generous. Even so at well over $100 a trade I have room to allow for plenty of extra slippage. It will be running on a sim account starting tomorrow AM. Averages 3 trades a week, I will post here the results to any trades it makes.
Finding new profitable neural networks after reaching a certain threshold is becoming very difficult. I have 100's of ideas on how this strategy can be improved. Will be a while before I am fully satisfied. However, I do have it running live on SIM account today so that I can start to get a few real time statistics on how it performs.
Is it just EOD data that you are using? Are you interested at all in doing with the U.S. Steel (Symbol: X), so I can run it with my robotic traders, which only work with U.S. Stocks. Thanks, Ed
The strategy uses 5 minute bars. I do not want to get side tracked since I believe I may be on to something. However, once I complete this strategy, I would be open to looking into U.S. Steel.
Hi Frost, As you noted, the NN results depend upon what is your input data. Also it depends upon what do you predict. I wonder how you may find 50 different NNs? You may train NN based on raw prices, or better momentum of prices. All other technical indicators are just derivation of raw prices, do you train it on TA and get different results? Could you give me an example of your inputs? Here is an old article regarding SOM NNs http://www.tssupport.com/support/base/?action=article&id=1806 Best regards, RR
That is an interesting article. I am open to discussing most topics about my strategy, except for my neural networks. The key is what your inputs are and what your predicting as you stated. I turned a corner not too long ago with my NNs when I changed completely the way I was approaching the problem. Had a "light" bulb moment. The way I am structuring my inputs and what each neural network aims to gain about the overall picture is very different than anything I've seen published or discussed on the topic. Its different but at the same time makes so much sense I am unsure how I took so long to "think" about it. Its easy to get confused from all the articles/tutorials/research papers published on the topic. All I can suggest is stop reading those and take a step back. What does a neural network really do... and what do you need to know to make a trading decision? I want to say more, but really can't. The concept is so simple its scary. If I wrote just 1 more sentence on this subject within a week there could easily be a handful of people with results very similar to mine.
Due to an oversight in how I was loading the strategies in NinjaTrader no valid trades will be taken today as the strategy sees insufficient data. Starting tomorrow the strategy should run correctly. I will backtest against today at end of day to see if any trades should have been taken..