Predictions, Profits, and Delays in Prices
56 Pages Posted: 7 Apr 2023 Last revised: 27 Oct 2023
Date Written: March 24, 2023
Abstract
Using machine learning algorithms, we predict 1-minute returns in the US (and international) stock market from 2005 to 2012 and in 2020. We provide three tests that indicate that return predictability arises due to fragmentation: we only detect return predictability when returns are estimated from prices across exchanges, and return predictability is much lower in less fragmented international markets as well as in the US before Reg NMS. Finally, we provide three tests showing that predictability is partially due to delays in prices caused by the SIP and individual exchanges. In particular, we exploit an exogenous technology upgrade to Nasdaq listed stocks that significantly decreased delays. Using a difference-in-difference setup we find this event decreases return predictability and improves market quality. We propose to use predictability as a proxy for random delays across signals and argue that high-frequency return predictability can be explained by delays in prices, providing another explanation for why paper profits often do not materialize.
Keywords: return predictability, delays, LASSO, limits-to-arbitrage
JEL Classification: G12, G14, C12, C55
Suggested Citation: Suggested Citation