The Bid-Ask Bounce Manufactures Mean Reversion and Inflates Small-Cap Returns
Can bid-ask bounce imitate profitable short-horizon mean reversion?
Compare price changes measured at trades with changes measured at midquotes, then account for the bid-ask spread in simulated fills. Alternating trades at bid and ask can produce negative short-lag autocorrelation without an executable reversal profit.
Roll model gives a mechanism; bar-level diagnosis is a heuristic and does not prove the mechanism.
Evidence map
| Aspect | Finding |
|---|---|
| What it is | The family of biases that arise because a recorded price is a transaction price, which alternates between the bid and the ask, and because different securities trade at different moments. |
| Key result / formula | Roll's model makes the first mechanism explicit: if trades arrive randomly at the bid or the ask around an unchanged efficient price, successive recorded price changes are negatively correlated purely from the alternation, so a series of closing prices exhibits mean reversion that no trading rule can capture, since buying at the bid is not available to the user of the data. |
| Why it matters for backtesting | These mechanisms can explain some apparent short-horizon predictability, and they deserve testing when a fast reversal strategy looks profitable. |
What it is
Both produce structure in returns that is unrelated to the value of any asset.
Key result / formula
Blume and Stambaugh show the consequence for portfolio returns: because the bias enters each security's return independently, an equally weighted portfolio rebalanced frequently accumulates it, which inflates the measured returns of small and illiquid securities and, by extension, the size effect computed that way. Lo and MacKinlay analyse non-synchronous trading and show it produces the opposite artefact at the portfolio level, spurious positive autocorrelation in an index, because slower-trading constituents incorporate common information with a lag, along with spurious cross-autocorrelation between liquid and illiquid names.
Why it matters for backtesting
The tells are diagnosable on bars. A reversal effect concentrated at the shortest horizon, strongest in the least liquid instruments, and disappearing when the strategy trades at the next bar's open rather than at the signal bar's close, is consistent with this artefact and does not alone identify the cause. Compare a midquote series where available and recompute returns after an explicit spread and feasible fill. The general rule is that the closing price is a transaction price and not a value, so a strategy that buys at a recorded low and sells at a recorded high is assuming it could transact on the favourable side of a spread it did not pay (see [Estimating the Bid-Ask Spread from OHLC Bars (Roll, Corwin-Schultz, Abdi-Ranaldo)]).
Source
Roll, "A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market", Journal of Finance 39(4), 1984, 1127-1139; Blume & Stambaugh, "Biases in computed returns: An application to the size effect", Journal of Financial Economics 12(3), 1983, 387-404; Lo & MacKinlay, "An econometric analysis of nonsynchronous trading", Journal of Econometrics 45(1-2), 1990, 181-211. Primary source