Hello @wahnker, you ask very sharp questions…this one has a two part answer. 🙂
This slippage can occur:
1. When you get a big red candle
2. While our systems are processing the data
The first cause is standard and will impact any trader and any system.
The second is specific to us, so let me dive into that one.
Our application pulls data and calculates against it fairly quickly, but it’s not pulling and responding every second.
The good news is we have spent the last few weeks focused exclusively on performance. The ML and calculation portion has been almost entirely refactored and we now have performance that is orders of magnitude better.
What this means is we can turn up the rate of data pulls. But we need to do this cautiously and with the right quality assurance. We will get a new roadmap update posted with this info.
This is where doing ML right gets tricky. We are running a proprietary data set. So if we change the rate, we need to wait until we have enough data in this new schema before we can begin training. This isn’t something we can do overnight. Because of this, there is a gap between changes in the foundation of data structure and when we can start making a new model generally available.
With all of that said, for the current model (considering both price and frequency slippage), the “sell” call is excepted to be within +/-0.3% of the stop loss point. So for BTC currently, the expected actual sell range is -0.7% to -1.3%.
This -4.93% definitely exceeded the expected range.
The speed of the drop was unfortunately very exceptional. Looking at the charts on Coinbase Pro, there were significant gaps on the 1 minute chart. I’d be very curious to see where -1% stops actually filled.
Finally, to answer your question–yes, you could manually enter a stop loss. It shouldn’t be required (given the expected range), but in this situation it could have helped.