Curious where this will go - I'm building a similar dip/peak detection algorithm. My results show that the dip was 2 hours ago (backtesting). From now, it might still increase, but not massively. Perhaps only around 3-4%. But of course it's just based on historical data, not a prediction.