I have recently felt despair regarding my own trading abilities and judgment, especially with meme coins or coins controlled by large holders. It's too risky for retail investors to make money.
After reflecting on this painful realization, since I couldn't trust my own judgment in trading and couldn't trust small-cap coins, I wrote a machine learning (Machine Learning, ML) simulation trading program to walk forward through the price fluctuations of ETH over the course of 4 years, opening both long and short positions, and provided it with 42 parameters to let it hit walls on its own. In the first few dozen rounds, it failed to complete the 4 years and went bankrupt, but later it slowly stopped dying, although in the end, it was still losing. It wasn't until about the 170th round that it started to show an 8% positive return.
But 8% is really too poor. After evaluation, I felt that it had become too timid in order to survive, so I unlocked some parameter limits and aimed for higher returns. I then ran it again for 201 rounds, and among those, 187 rounds achieved a +170% ROI. Last night, I connected it to my contract to see how well it can perform.
In the background, I continued running ML until the profits reached a plateau, which should be the most stable model. So 170% is not the endpoint, just a starting point!