I am sure you have found a strategy before and discovered that it works fine on some symbols, but just awful on others. This learning step tries to deal with that problem. In my experience many trading strategies can work but they need to be adjusted to work with the way a stock symbol trades.
To figure this out we will need a program to generate lots of combinations. The program will loop through thousands of combinations and will try to see if the strategy (indicator combinations) has worked consistently in the past. Rather than using the default settings of indicators we will slightly modify them to find out which settings work with the largest group of symbols we can find.
All of the trading strategies that the system uses requires 3 indicator categories that work together but don't test the same type of category. For each strategy we will select 1 indicator from 3 distinct categories. We will use a Momentum indicator, a Trend indicator and a Volatility indicator.
You can think of Learning Mode as a crude step to just orienting ourselves with our indicator selection. This step tests the strategy overall to see if it's consistent. The whole point of Algorithmic Trading is taking a small piece of profit and then moving on to the next trade. Our goal is not to follow a long dynamic trend, we are getting into the trade for a couple days and getting out. To discover what works we define success as a stock symbol that succeeds at a 1% profit with a 1% stop loss greater than 80 percent of the time.
Learning Mode is time consuming and expensive. To run the below sample with 11,879 indicator combinations over 2 years of data and 3 time periods will take days. It will also cost a lot in cloud computing time, maybe $2 to $5 / hour depending on how fast you want to go. Also, when the program finishes it might not yield any good results with the indicators we have chosen. In my experience most indicator combination don't work. As of November 2019 I have only found 6 that work and I have been at this for 3 years. Most of the time you end up with only a small group of symbols that succeed. Maybe 2% of the stock market. I define a successful algorithm as being compatible with more than 5 ~ 20 Percent of market symbols.
So you might be saying at this point, 1% is small and this whole thing is a waste of time. But is definitely is not.
|Stock Price||1% of Stock Price||Shares Purchased||Profit Per Trade|