How are the performance charts generated?

The performance charts are generated by backtesting the networks against bar data of four hour periods for cryptocurrency markets and one day periods for stocks. The networks evaluate each new bar and provide either directional bias or buy/sell bias outputs for the open of the next bar. Each backtest has a simulated portfolio that starts with $1000 and follows the bias with the full portfolio amount at each step. The v1 networks we offer network access to currently have been running live since Summer 2021, while the v2 networks have been running live since the beginning of July 2023 the true start for their out of sample performance is the start august 2021 as nothing beyond that was used in training.

How are the networks trained?

We are always experimenting with new training techniques, initially we utilized a technique known as neuroevolution. Neuroevolution involves optimizing parameters of data by treating them like genes, which are then passed on in a survival of the fittest fashion. This provides a more global search than gradient-based optimizations and allows us to be flexible with our evaluation in training. 

More recentally we have been using gradient descent on attention based neural networks as well.

Are you continuing research with plans to release new networks?

Yes, we plan on continuing to train new trading networks on as many different markets and assets as possible. Setting up the live trading networks and integrating them with the site can be complex, so adding new bundles will often be live on the site several weeks after they have completed training. 

What does subscribing to the network bundles get me?

By subscribing to the bundles, you will get access to the page showing the networks' current directional outputs as well as the ability to receive email notifications every time the outputs are generated, customizable so you’re able to receive emails only when they change direction, no change ie. an email received every period that network outputs would be generated, or no emails if you’d prefer to utilize the webpage to observe the neural networks’ outputs from there.