I wanted to give everyone an update on what I tried over the last 2 years.
-
Wrote some EAs purely based on technical indicators (combinations of 2 indicators). Didn’t even achieve a 55:45 winning-losing rate (tried adjusting entry points, take-profit, stop-loss programatically). Not surprising to me now as there are many more factors to consider when getting in/out of trades.
-
Wrote some Python code with Oanda API and a demo account. Didn’t get anywhere. Btw, Oanda support team does not really respond to questions from demo account holders.
-
Did some research on deep learning models. Don’t think LSTM would be helpful. Quite some researchers published their papers with excellent outcomes based on historical data only.
But according to my personal experience, I still think it possible to let a piece of code learn tick data and figure out what is going on. In April, I opened another demo account and made a 30% profit over 2 months by merely learning price actions of 2 currency pairs. I know two months is too short to convince anyone. But I do feel I got better. Just need to know exactly what made me feel so. Maybe it is the ability to identify reversals better, or to take the pulse of market strength, or to read charts or see price action patterns, or simply by luck in the sense that the market gives me a second chance always.
My idea now is to feed my code with ticks. The piece of code picks information on a non-human level but keeps building/improving an internal model that can make fairly good predictions. This piece of code can be based on existing machine/deep learning models (for time series data) but I strongly doubt it now after my research. Please correct me if I am wrong here. Maybe reinforcement learning is the way to go? like Q-learning? Some other times, I feel statistics based approaches may work better as trading success is all about probability and statistics. What do you guys think? Any thoughts, ideas?
Thanks