Countless sets of cosmetic detail. But no matter how good these are, trying to trade in the wrong style for the individual leads to failure.
From a non professional retail trader's perspective: Books, publications, academic papers, web sites..., most popular and well known strategies didn't work, those that worked, just didn't generate enough profits to be worth trading full time for a living. Unless I am a PM with lots of OPM? Bottom line: Develop your own strategy.
Never tried this, since I developed my own model. Read his blogs. Seems like a pretty sharp guy. http://www.priceactionlab.com/
"Ok, ... but ... how does a strategy get into my noggin?" For me, Reading. I read books and websites regarding trading and strategies. As I read, I would think, "That sounds like a good strategy." But I continued reading. Code: Do As I continued reading, I would think, "That sounds like a better strategy!" But I continued reading. Eventually, I thought I understood enough to (now) have a strategy in my noggin. Then, I tested it. Code: Loop
You find strategies to test by gaining experience in actual trading. By trading small, this will give you ideas that can be tested going forward. Without some experience as to how markets move, routing, order placement, risk management and on and on, you will have little/no idea how to build strategies, and thus what to backtest. Personally, I have no use for backtesting as it assumes the market going forward will be like prior markets. That, and these things are usually done with unrealistically-small sample sizes Like the 50 trades you mention (I know people who do 100 + trades per day). Backtesting is usually done on sim programs, Demo platforms, or papertrading which often have little relationship to executing and trade management in the real world. Should you not be dissuaded by this argument, and insist on this pursuit, you need to start with an idea (again, where do you get these?), and test it over numerous market conditions - say the entire period from 1999 until now - involving many hundreds if not thousands of trades.
Reading academic papers,articles etc. will probably give good idea what is already throughly tested and what not to repeat exactly the same way. It is always nice to google how popular an idea is. Finding some order from past chaos(price data) is pretty easy(just some coding skills and curve fitter is ready), but to expect same to repeat in future is wrong and the more complex(adaptive to past) method the more it is likely to be bad idea. What i have found in my years of experimenting with datamining and emulating retail side conditions with custom well optimized c++/cuda software is that to overcome trading costs algos need to be waiting most of the time and exit levels should be many times larger than costs of trade. But someone else may have found completelx different ways trough their experiences. To go against other market participants,ie firms who get ultra low spreads and can scalp like mad at higher frequencies with much more favoralble rules/conditions and have data mining farms with gigantic power bills it cant be easy task.