Analysis of past week:
Last week was a bit of up and down. The network strategy was suboptimal.
I am comparing three strategies, first hold fiat and make a small profit elsewhere, hold bitcoin, or follow the network predictions. Since I started documenting my predictions on the blockchain these strategies compare as
Fiat with a weekly profit of 0.5%: 16727.17$ (up 0.5 %)
Holding BTC: 11523.58$ (up 4.9 %)
Network strategy: 11273.71$ (up 0.5 %)
Starting capital: 11497.40$
I have been doing this for more than a year now and I think it is time to draw some conclusions.
The bitcoin price is now pretty much where I started. In-between we had a fair amount of ups and downs. And the neural network is exactly on par with hodling bitcoin. There was nothing gained by using the network and (almost) nothing lost.
And of course since we are not in a raging bull market, fiat with 30% profit is superior.
I am not convinced that this is just bad luck on part of the network, but that there is a more fundamental problem. Neural networks find patterns. But are these patterns just statistical flukes that will never be repeated in the future, or are they actual based on real dynamics. A network (at least the type I am using) can never distinguish those two. In addition the training data is extremely limited and it is not simple to build much more sophisticated networks.
Therefore I am no longer convinced that using neural networks is the way to go. Instead I am now pursuing more statistical methods. I will post more when I am finished, but since I no longer believe in the network I will stop posting these forecasts.