In SlothTrading we don’t like to play with data either, but we like to obtain the statistical probability of where a value will be found after a given time interval, as we don’t want to compare ourselves with Mr. Einstein, we don’t even want to know its exact value, we are only interested in whether that value is above or below the current value.
What do we get with this?
If we know the relative position of a security at time t1 with respect to an initial time t0 with a probability equal to 52%, we can be sure that if we open 1000 positions in 520 we will have made a profit.
Is this the Holy Grail of investments?
No, obviously not. There is always the risk that the losses obtained in the 48% of failed positions are higher than the profit obtained in the successful positions. This effect can be minimized by entering the market only when we think that volatility is not excessive.
How can we achieve this probability of success?
With the help of AI and convolutional neural networks. After analysis with different types of assets, we developed a common model that partly met the requirements. We observed that the probability of being right on each security depended on the values provided to train the network and these in turn depended on data intrinsic to the security itself (type of asset, geographic dependence, etc.). These facts led us to create a farm of RNNs with the same genetic basis, but with different epigenetic patterns.
What is the result?
At present, on a volume of 200 stocks, we achieve a 55% accuracy rate in our predictions, obtaining a profit of 88% in the last 12 weeks.
Technical description of the project
The whole project is done with GNU applications:
- MySQL 8.0.27 DB.
- Python 3.10.5
- TensorFlow 2.9.1
- Apache 2.4.51
Both the web server and the database are hosted on AWS so donations are welcome.
All this data can be analyzed on our SlothTrading website.