Ventures of an ex indie game developer

Crypto trading status

BitForex aims to be first man in the race to the bottom. By allowing maker trades at zero fees, they've grown massively. My simulation tells me I should be able to make 60% more profit per year if I could execute the same trades using them instead of Binance (which have a 0.1% maker fee). It's not impossible, but BitForex is still small fish in the alt-coin market, so I'll stick with high quality, large quantity and good stability for now. (Besides, BitForex is super-Chinese, with lousy API and old-fashioned instruments;  it wouldn't surprise me if they're hacked.)

On the other hand, I took a stab at implementing RSI deviation-recognition, which proved very simple. That change alone earns me 40% more profit per year.
I've probably done my last AI experiments. For the most straight-forward solutions, such as those convolutional and LSTM I tried, it's not possible to cross the 60% accuracy border. For any coin on any time-period. The price pattern is fractal with regards to time, so it might be possible to mix different time periods and get something from that, but on the other hand, that could be said about a hundred different ideas as well.

The ANN solution looks very tempting for a rational engineer, as there are no fixed borders in finance. But I've realized a couple of things that makes me trust more in 30 hard hand-written rules, than fuzzy logic. First and foremost, the patterns never repeat themselves. Sometimes they come close, often they do not. Secondly, the best indicators are developed by highly intelligent and successful senior researchers and traders. By definition, they are hard to beat. Connors RSI, for instance, took decades to develop and refine.

So instead of throwing everything at a set of weights, I use a combination of indicators, which I tweak using a dumb optimizer. I should implement something like the Adam optimizer used in ANN's, but it's currently working too well for me to make the effort.

The hard part for me is trying to cram in as many good indicators and thresholds as I can, while keeping the code clean.


Finally, this project is coming to an end. It might be possible to live off of the passive income it will produce. Since January, when everything crashed, I'm behind 60%. You'd think that would make me a pessimist, but not me. :)

Anyway, somewhere between beginning of October and end of November 2018, big things are going to happen to the cryptocurrency finance world. My guess is a huge drop in price in mid-October. Then a bull market again in 2019.


About the author

Mitt foto
Gothenburg, Sweden