@matthew55 and everyone, here is the update on this:
Backgound and Root Cause
The 4.x Ethereum model does prefer to issue a relatively higher volume of trades, in about the 4-5 trades/month range. By artificially lowering that, we see drastic and unacceptable reductions in results.
Despite expecting higher volume, the ultra-fast trading seen over the last week or so was unusual. The data coming in put us in “edge case” scenario, so we saw unusual trading behavior. In general, most financial markets have been in a highly-unusual state.
The rapid trading has slowed down without us changing anything.
We will soon post information on each model’s expected trade frequency as well as some technical/statistical data on technology used for prediction.
We will upgrade ETH to the 5.x model series within a week.
I’ve alluded to 5.x briefly before. It’s a natural progression from 4.x and are in fact very related. But they require a complete overhaul to our infrastructure. The 5.x machine learning models are massively more complex and can chug many more inputs. This takes a lot of compute power. The result is another large leap forward in prediction precision.
Once this upgrade to 5.x is complete, we will have moved to the top of the ML stack. I would just expect refinements from that point forward, but you never know–this is an evolving game.
I will note that the 5.x ETH also issues a larger quantity of trades.