Working Paper: CEPR ID: DP13561
Authors: Alex Chinco; Vyacheslav Fos
Abstract: This paper proposes that computational complexity generates noise. In modern financial markets, it is common to find the same asset held for completely different reasons by funds following a wide variety of threshold-based trading rules. Under these conditions, we show that it can be computationally infeasible to predict how these various trading rules will interact with one another. Formally, we prove that it is NP hard to predict the sign of the net demand coming from a large interacting mass of funds at a rate better than chance. Thus, market participants will treat these demand shocks as random noise even if they are fully rational. This noise-generating mechanism can produce noise in a wide range of markets and also predicts how noise will vary across assets. We verify this prediction empirically using data on the exchange-traded fund (ETF) market.
Keywords: noise; thresholds; complexity; indexing
JEL Codes: G14; G00; G02
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Many funds with different threshold-based trading rules (G23) | Computational infeasibility in predicting net demand (C69) |
| Computational infeasibility in predicting net demand (C69) | Treating demand shocks as random noise (C22) |
| Stocks near rebalancing thresholds (G12) | Experience more noise (C99) |
| Number of neighbors in the ETF rebalancing network (D85) | Higher probability of impact from rebalancing cascades (F65) |