This post is sequel to this post. If you haven’t read it, please do read it first.
This is my opinion at the time I write this post, which is based of 3+ years playing around with different types of strategies. As written in the post linked above, all we are doing right now with indicators are trying to predict the future using past data. Freqai or not, it’s still the same game, just being played on different ways.
Wise man used to say “All roads lead to Rome”, which assumes that Rome really exist. Is it really possible to predict the future? Or we have to admit that “All roads lead to dead end”?
Trying to predict the future
Yes, I believe we can kinda predict the future. It won’t be 100%, but with some setup, it can lead to some sustainable profit, but it won’t be the amount of profit that you imagined or wanted. As I have said here, you need to set a realistic target. Having realistic target will make your job easier because you don’t need to have 100% success rate of predicting future to have a long-term profit.
So what is the setup I mentioned above? No, it’s not about list of “best” indicators. That part is ranked at the middle of things. If you want to predict behavior of something, first of all, you need to make sure that the “something” isn’t erratic. In the example previously, if Frog is an impulsive person, then you will have very low chance of predicting their next choice, no matter how advanced you think your method is. Same thing for trading. You will want to choose pairs that have predictable movements and not “impulsive” (which means isn’t easily pumped-and-dumped). Those suitable pairs usually are pairs with high marketcap ranks. Choosing the right pairs will make your trading life way easier. Trust me.
So after choosing the pairs, time to pick the indicators, right? No. You now need to set the direction of the strategy, which I wrote here. Is it important? Yes, because that will influence how aggresive or relaxed is your entry and exit logics. You need to choose how often you want to enter a trade, and how high/low are your profit and loss targets.
Why are they important? First, because as said above, we won’t have 100% prediction rate. So do you want to go quantity route (plenty of trades and hoping that you get more right than wrong), or quality route (less trades but better chance of having more right than wrong)? That’s the entry side. Now for exit side, how high should we aim the profit? Using indicators for exit decisions, or using freqai to predict exit targets, same thing applied, which is they won’t be able to predict 100%. Personally, since the entries already having the uncertainty issue (won’t be 100% accurate), why add more risk by subjecting exit to the same uncertainty? That’s why I set fixed ROI on my strats, and not having conditional exits. Of course I also need to observe the market to see the realistic ROI to be used. For example, expecting 10% ROI on BTC trade is almost impossible. Do your own research to find the good ROI to be used. My advice, don’t be greedy. Faster exit is way better than trying to maximize the profit. In crypto, “trend” can be broken in seconds. This way, I only need to focus on the entry logics.
After that, then we can start looking at indicators. Personal opinion, there is no such thing as best indicator. What you need to do is to have variety of indicators that look at different things. And it might be better to have different length/period variations on those indicators, preferably looking at more bigger picture. So for example, if you are using RSI, then calculate RSI of 14, 28, and 56. Don’t use small length/period, they will make your predictions less accurate.
Doing those things above personally helped me in this journey. If you have been trying other ways and keep failing, give my method a try. Don’t be afraid of trying new ideas, no matter how dumb it might sound. You can always run the bot in dry mode to test those new ideas. Don’t subject yourself to unnecessary risk.
The path I’m currently taking
Despite what I wrote above, currently I’m testing an idea that requires no indicator at all. All I need is looking at price actions and some averages of OHLCV. The main point of this idea is (once again) picking the right pairs for it. The pairs for this idea is completely different than the pairs needed for the strategy above. What’s behind this idea? Just some observations that there are recurring patterns on some pairs that might not be detectable by indicators. So instead of relying on indicators, I looked at how market reacted on certain conditions and whether they were repeatable. This type of idea can’t be backtested, and you can’t find any book or scientific paper discussing it. Would it be successful? I don’t know, but as I said above, don’t be afraid to try new idea.
Conclusion
As you might have notice, both routes that I took started with same thing, which is pairs selection. Each pairs will have their own characteristics, which means the strategy suited to trade them will be different as well. For example, if you want to trade low marketcap pairs, then it might not be suitable to trade them using mean reversion, especially when they are being pump-and-dump’ed. Most likely you will get caught trying to catch falling knife (opening a long trade on a pair that is still mid-way of its way down).
Please spend some time to really look into your pair selection and try to pick the right pairs for your strategy. Good pairs selection is more important than choosing the right indicators.