Over the last 6-7 months, I have created an algorithm that had an accuracy as high as 97 % over 80 stockmarket/commodity trades using CFD. During that time I managed to achieve 700% profit.

This was an amazing feat, and also a dangerous one. The accuracy rate was high such that the risks involved were masked.

These are highly leveraged (x10) stocks which means a fluctuation of 10% would mean that the stocks would hit stop-loss (SL) levels. Given the accuracy at that time, it was safe to say splitting the trades over 3 different volatile stocks would also provide adequate split, and x50 leverage would allow for larger profits. Unfortunately, all 3 stocks fluctuated too much, and stop losses were hit despite the predictions being correct. The issue was the volatility, not the predictions.

The project was still profitable overall (70-80% return), but I have decided to scrap the algorithm and rewrite the code altogether.

Take home messages for version 1:

1. First I would create artificial fail data by using a virtual portfolio.
2. I would then add the data of failed trades to the pool of successful trades.
3. It will be best to achieve 50 % successful and 50 % failed trades in the generated data.
4. It is not the predictions that matter but the ability to survive the tides of volatility. That has to be taken into account but I am debating whether to use AI or common sense.
5. Pure TA is not good enough. When stocks such as oil are manipulated by production amount, it makes other investable commodities such as cryptocurrencies much more favourable.
6. To those who are following me, be ready for version 2. Meanwhile, I will be trading cautiously using x 10 leverage. Perhaps with the new profits, I may decide to spread risk.
7. No software lasts forever. That is why the human touch exists.

Trade data is available publicly. The software is only available for internal use.

Interesting read:

AI has become so popular in picking stocks that it’s become ineffective