Microstructural Account of the Demise of Short-Term Trend-Following (Kurth et al.) Fast trend signals (days to weeks) lost their edge after 2009 for equities (IDX) and FX but not for commodities and government bonds/yields. (Table 1, and Figures 3, 4) Why? Crowding doesn't seem to be the explanation, as the timing of CTA AUM and futures liquidity doesn't line up with the loss of edge. The implied return drag is also too small. (Figure 5) It's the signal itself that degraded: the same edge loss is evident when the execution lag is removed from simulations. (Figure 6) Other papers have come to similar conclusions (see comments). Interestingly, when contracts are divided into two groups by (tick size)/volatility ratio, the small-tick contracts exhibit loss of edge, while the large-tick contracts don't. This is not explained by liquidity differences. (Figures 10, 11) The loss of edge mostly shows up for small-tick contracts on high-volatility days. The other three combinations of tick size and volatility were fine. (Figure 18) The authors suggest that these patterns can be explained by the transition to high-frequency trading in market making. They suggest that trend followers pile into the same trade on high-volatility days, creating self-reinforcing loops that push the trend and volatility further. But they can only do this if market markers are willing to take the other side to their own detriment. HFT-dominated market making tries not to absorb predictable directional flow in small-tick markets, while for large-tick markets, spreads are wide enough to compensate market makers for adverse selection. Now it's short-term trend followers' turn to deal with adverse selection. Chasing fast trends means being willing to accept bad fills, but only up to a certain point that doesn't negate profits. Limit orders are problematic because of missed opportunities when the trend runs away from the limit price. If trend followers piling in is what helps sustain short-term trends on high-volatility days and if trend followers are no longer willing to participate, the feedback loops that used to create short-term trends no longer exist, and the signals themselves no longer work even with zero execution lag. The drivers of long-term trends may be different (underreaction, accumulation of large positions, gradual information diffusion). The paper doesn't explore this, but if the conclusions are correct, it may make more sense for traders to accumulate positions gradually today than pre-2009, as market makers are less inclined to stand in the way when prices start moving quickly.