I ve used couple of indicators but I did it also trying to understand daily price variation and stocks behaviour
Throw $100 at it and see what happens. It's possible that you're making multiple mistakes, not only regarding test period, but you may also be making other wrong assumptions. You can't even know if your orders can be filled, so test this live first. You'll learn more than asking theoretical questions.
You should repeat your backtesting over a 20 year period to make sure it's profitable always. The market is dynamic and can have some particular characteristics in some years that make a strategy profitable if it happens to take advantage of those characteristics during those years but your strategy might not work for all the years as each year is different. That's why you have to backtest all the years and the more years that you backtest the better.
Welcome to backtesting, you think you’ve got it all figured out then market changes and then it all goes to sh#t
If your system traded multiple times a day rather than just once a day you wouldn't necessarily need to test back so far. I.e if I have a system that trades 50 times a day and it tested well in June, July and August and also in October, November and December of this year and results were consistently profitable in both the summer and the recent months, for myself, it would be worth throwing some money at. How far to back test is often relative to trade frequency. Some systems cannot be accurately back-tested at all. From the minimal information that you provided it looks like there are lots of break-evens or trades being filtered out through curve-fitting.
I do not use automatic systems so I don t practice HFT. I ve just applied simple but logical rules. As trading should be. Indeed I ve learnt from the market. If there is something logical that doesn't work what s the reason of applying it ? The 0s that you see , are not trades. It means that day I didn t trade. My system can be backtested as there is nothing particularly complicated and it regards several important companies. The only thing is that you gotta spend some time. I m throwing on it some money and see how it works, at the same time, already working to make a new one for a 20 y period
Trying to find one rule-set by which to trade a 20 year length (where the market might show 4-5-6 different faces) is very much a fool's errand. It'll kill you, AND your account. "Algorithmic Trading" (simply: "Rule-based Trading") is what you've described. But you're thinking that there must only be one, single, algorithm. What if you had rules by which to choose from different algo's? Or different portions in algos? A second layer of rules? Now you're into "Artificial intelligence." Fancy, fancy terms, for what is really just a slightly different way of thinking. ["Out-of-the-box-spread thinking??" Nah. It's not Bad Pun Thursday.] So, do you, right now, expect the market to change in the next 2 years? I certainly do. So, I would like a one or two year model, for which I might trust it for the next month. That's all. Just one little month at a time.
I m not so naive about thinking to find one rule set by which to trade 20 years. The point is having a view about how stocks quantitatively behave. Would you say that this wouldn't be helpful ? I don't want to apply mechanically but combine quantitative datas with the interpretation of the moment.
Make an example. Of the 10y strategies that I have already elaborated, I cutted the period in half and I ve seen if I could have extracted the same rules that I have extracted over the 10y period, using just the first half. The answer is yes (90%). In the next 5 years they would have been winners, of course with different returns, some would have gained more some other less than the previous 5 years. But the tendency would have been the SAME.