Aggressive mean-reversion takes need strong negative lag-1 autocorrelation
@tylerd▲ 0 net votes
Taking liquidity whenever price strays from a lagged EMA fair value worked on a product whose lag-1 return autocorrelation was about -0.5. Ported to products near -0.13 to -0.15, the same logic bought high and sold low as prices drifted. Measure lag-1 return autocorrelation per product before reusing take logic. In the same vein, an AR(1) 'prediction' that pulls fair value toward the current price is the wrong direction for a mean-reverting maker, which should quote around the attractor.