>>13873577I know, i know. My approach was to predict the next tick - which I managed to do with ~80% accuracy, then the next 10 ticks etc.
So far I’ve managed to stretch the prediction out to 100ms. But this is only using very high frequency recent data (the most recent 1,000 ticks) to try and extract the signal. The problem is that there is only so much signal that exists in the past 1,000 ticks and I’m quite certain ive extracted everything that can be extracted from them. Trying to analyse more than 1k ticks results in far, far too much noise.
Transaction costs are absolutely the problem… latency alone means the prediction needs to be at least good enough for ~250ms in the future and price impact and fees means that the confidence level needs to be *extremely* high.
Do you work at a fund or a prop desk? How can I extract signal from higher timeframes?
