pith. sign in

arxiv: 2606.29591 · v1 · pith:ICXHLFX2new · submitted 2026-06-28 · 💱 q-fin.TR

The Bounce Has No Direction: Sign, Magnitude, and the Microstructure of Equity Return Predictability

Victoria Portnaya This is my paper

Pith reviewed 2026-06-30 01:28 UTC · model grok-4.3

classification 💱 q-fin.TR
keywords return autocorrelationbid-ask bouncemicrostructureequity predictabilityvariance ratiosign testmagnitude channelFourier-Residue Identity
0
0 comments X

The pith

SPY's negative lag-1 return autocorrelation arises entirely from magnitude shrinkage rather than directional reversal.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops the Fourier-Residue Identity to split the autocorrelation of returns into separate sign and magnitude channels that can be tested independently. When applied to SPY and other instruments, the lag-1 negative correlation disappears once the magnitude channel is isolated, leaving an insignificant sign channel. This pattern matches the expected signature of bid-ask bounce and stale prices rather than any tendency for prices to reverse direction. The same decomposition uncovers a directional reversal at lag 3 and shows that mean reversion is present only in equities and bonds across a broad panel of assets. Additional results give the variance-ratio test a spectral reading and supply a subsample ratio that distinguishes persistent autocorrelation from noise.

Core claim

The Fourier-Residue Identity decomposes return autocorrelation into a sign channel (k=2) and a magnitude channel (k=4). In SPY the lag-1 autocorrelation is carried entirely by the magnitude channel: the sign test yields p=0.11 while the joint test reaches p<10^{-12}. A large absolute return yesterday therefore forecasts a smaller absolute return today irrespective of sign, the statistical fingerprint of bid-ask bounce and non-synchronous constituent staleness rather than directional reversal. At lag 3 a separate directional reversal becomes detectable. Mean reversion is confined to exchange-traded equities and sovereign bonds; credit ETFs, commodities, FX and crypto are statistically indisti

What carries the argument

The Fourier-Residue Identity (FRI), which decomposes autocorrelation into independent sign (k=2) and magnitude (k=4) channels.

If this is right

  • Standard variance-ratio tests receive an exact spectral interpretation through the proved Fejer identity VR(q)=1+2C_q.
  • A subsample diagnostic R_N=G_{N/2}/G_N distinguishes structural autocorrelation (R_N approaching 1) from sampling noise (R_N approaching sqrt(2)).
  • Directional reversal appears at lag 3 even though it is invisible to the ordinary autocorrelation function.
  • Mean reversion is limited to equities and sovereign bonds; other asset classes behave as random walks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The decomposition could be applied at intraday frequencies to separate liquidity-driven effects from any slower behavioral predictability.
  • If magnitude effects dominate daily horizons, strategies that treat negative autocorrelation as reversal may be trading against liquidity provision rather than exploiting mispricing.
  • Extending the sign-magnitude split to volatility or order-flow series might reveal whether the same channels operate in those data.

Load-bearing premise

The Fourier-Residue Identity decomposes return autocorrelation into independent, non-redundant sign and magnitude channels that can be tested separately.

What would settle it

A statistically significant contribution from the sign channel to lag-1 autocorrelation in a larger sample or different market regime would falsify the claim that magnitude alone drives the observed correlation.

read the original abstract

SPY's lag-1 return autocorrelation ($\hat\rho(1)=-0.081$, $z=-7.4$) is among the most significant regularities in empirical equity finance, yet the standard variance-ratio (VR) test cannot determine whether it reflects directional reversal or magnitude shrinkage - phenomena with entirely different trading implications. We develop the Fourier-Residue Identity (FRI), which decomposes return autocorrelation into a sign ($k=2$) and a magnitude ($k=4$) channel, each independently testable and neither redundant. Applied to six US instruments over 1993--2026 and a 21-instrument cross-asset panel, the FRI delivers a sharp microstructure diagnosis. The lag-1 autocorrelation in SPY is driven entirely by magnitude: the FRI sign test is insignificant ($p=0.11$) while the full test achieves $p<10^{-12}$. A large move yesterday predicts a smaller move today regardless of direction - the fingerprint of bid-ask bounce and non-synchronous constituent staleness, not directional reversal. At lag 3, a significant directional reversal ($p=0.02$) invisible to the scalar ACF reveals a separate partial-price-adjustment channel. We prove the Fejer identity VR(q)=1+2C_q (confirmed to <0.001 on all series), giving the Lo-MacKinlay test a spectral interpretation, and introduce a subsample diagnostic R_N=G_{N/2}/G_N that classifies equity autocorrelation as structural (R_N->1) rather than sampling noise (R_N->sqrt(2)). The cross-asset panel shows mean reversion confined to exchange-traded equities and sovereign bonds; credit ETFs, commodities, FX, and crypto are indistinguishable from random walks. All estimators pass 27 unit tests; Monte Carlo confirms correct 5% size under GARCH.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper introduces the Fourier-Residue Identity (FRI) to decompose equity return autocorrelations into independent sign (k=2) and magnitude (k=4) channels. It applies this to SPY and other assets from 1993-2026, concluding that SPY's lag-1 autocorrelation (ρ̂(1)=-0.081) is entirely due to magnitude effects (sign test p=0.11, full test p<10^{-12}), indicative of bid-ask bounce rather than directional reversal. It also proves the Fejer identity relating variance ratio to autocovariances, introduces a subsample diagnostic R_N, and analyzes cross-asset panels showing mean reversion only in equities and bonds.

Significance. If the FRI decomposition is valid and the channels are orthogonal, the paper provides a valuable microstructure interpretation of return predictability, distinguishing magnitude shrinkage from sign reversal with different trading implications. The proof of the Fejer identity offers a spectral view of the Lo-MacKinlay VR test. Strengths include passing 27 unit tests, Monte Carlo size checks under GARCH, and the subsample diagnostic for distinguishing structural autocorrelation from noise. The cross-asset findings add to understanding of where predictability exists.

major comments (1)
  1. [Abstract] Abstract: the central claim that lag-1 autocorrelation in SPY 'is driven entirely by magnitude' rests on the assertion that the FRI 'decomposes return autocorrelation into a sign (k=2) and a magnitude (k=4) channel, each independently testable and neither redundant'. The insignificant FRI sign test (p=0.11) is used to rule out directional reversal, but this requires that the k=2 statistic fully isolates the sign component with no residual cross terms from E[sign_t sign_{t-1} |r_t| |r_{t-1}|] that are not captured by the separate magnitude channel; without an explicit verification of orthogonality (e.g., via the full derivation or a cross-term bound), the 'entirely magnitude' diagnosis is not yet load-bearing.
minor comments (1)
  1. [Abstract] The abstract states the Fejer identity VR(q)=1+2C_q is 'confirmed to <0.001 on all series'; adding the precise numerical check (e.g., maximum deviation across the six instruments) and its location in the text would strengthen the reproducibility claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive report. The single major comment concerns the need for explicit verification that the FRI sign (k=2) channel fully isolates directional effects without residual cross terms from the magnitude process. We address this below and commit to strengthening the exposition.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that lag-1 autocorrelation in SPY 'is driven entirely by magnitude' rests on the assertion that the FRI 'decomposes return autocorrelation into a sign (k=2) and a magnitude (k=4) channel, each independently testable and neither redundant'. The insignificant FRI sign test (p=0.11) is used to rule out directional reversal, but this requires that the k=2 statistic fully isolates the sign component with no residual cross terms from E[sign_t sign_{t-1} |r_t| |r_{t-1}|] that are not captured by the separate magnitude channel; without an explicit verification of orthogonality (e.g., via the full derivation or a cross-term bound), the 'entirely magnitude' diagnosis is not yet load-bearing.

    Authors: We appreciate the referee drawing attention to this foundational point. The FRI identity is obtained by applying the residue theorem to the Fourier transform of the joint characteristic function of consecutive returns; the k=2 term extracts the contribution of the sign process while the k=4 term extracts the contribution of the magnitude process. By construction the cross term E[sign(r_t) sign(r_{t-1}) |r_t| |r_{t-1}|] is absorbed into the magnitude channel because the even-powered residue isolates the absolute-value dependence. Asymptotic orthogonality of the two statistics follows from the fact that their covariance is zero under the maintained regularity conditions (finite fourth moments and mixing), which is confirmed by the Monte Carlo size checks reported in the paper. To make the separation fully transparent we will add, in the revised manuscript, an appendix that (i) states the precise orthogonality condition, (ii) derives the cross-term bound, and (iii) shows that any leakage is o_p(1) and does not affect the sign-test p-value at the reported sample sizes. With this addition the claim that the lag-1 autocorrelation is entirely magnitude-driven will rest on an explicit rather than implicit separation. revision: yes

Circularity Check

0 steps flagged

No circularity: FRI decomposition and tests are presented as independent empirical channels with external validation

full rationale

The paper introduces the Fourier-Residue Identity as a new decomposition into sign (k=2) and magnitude (k=4) channels, states they are independently testable and non-redundant, and applies them to data with reported p-values. It also proves the Fejer identity VR(q)=1+2C_q (externally verifiable) and validates estimators via 27 unit tests plus Monte Carlo size checks under GARCH. No equations reduce a prediction to a fitted input by construction, no load-bearing self-citations, and no ansatz smuggled via prior work. The central claim rests on direct application to observed series rather than definitional equivalence.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Review based on abstract only; full details on assumptions unavailable.

axioms (2)
  • ad hoc to paper The Fourier-Residue Identity decomposes autocorrelation into independent sign and magnitude channels
    Introduced and applied in the paper
  • standard math Fejer identity VR(q)=1+2C_q
    Stated as proved in the paper

pith-pipeline@v0.9.1-grok · 5869 in / 1120 out tokens · 31739 ms · 2026-06-30T01:28:00.685044+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

19 extracted references

  1. [1]

    Amihud, Y., & Mendelson, H. (1987). Trading mechanisms and stock returns: An empirical investigation.Journal of Finance, 42(3), 533–553

    1987

  2. [2]

    Andrews, D.W.K. (1991). Heteroskedasticity and autocorrelation consistent co- variance matrix estimation.Econometrica, 59(3), 817–858

    1991

  3. [3]

    Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing.Journal of the Royal Sta- tistical Society, Series B, 57(1), 289–300

    1995

  4. [4]

    (1991).Time Series: Theory and Methods, 2nd edn

    Brockwell, P.J., & Davis, R.A. (1991).Time Series: Theory and Methods, 2nd edn. Springer-Verlag

    1991

  5. [5]

    (1997).The Econometrics of Financial Markets

    Campbell, J.Y., Lo, A.W., & MacKinlay, A.C. (1997).The Econometrics of Financial Markets. Princeton University Press. 28

    1997

  6. [6]

    Chow, K.V., & Denning, K.C. (1993). A simple multiple variance ratio test. Journal of Econometrics, 58(3), 385–401

    1993

  7. [7]

    Fama, E.F. (1970). Efficient capital markets: A review of empirical work.Journal of Finance, 25(2), 383–417

    1970

  8. [8]

    French, K.R., & Roll, R. (1986). Stock return variances: The arrival of infor- mation and the reaction of traders.Journal of Financial Economics, 17(1), 5–26

    1986

  9. [9]

    Glosten, L.R., & Milgrom, P.R. (1985). Bid, ask and transaction prices in a specialist market with heterogeneously informed traders.Journal of Financial Economics, 14(1), 71–100

    1985

  10. [10]

    Ho, T., & Stoll, H.R. (1981). Optimal dealer pricing under transactions and return uncertainty.Journal of Financial Economics, 9(1), 47–73

    1981

  11. [11]

    Jegadeesh, N. (1990). Evidence of predictable behavior of security returns.Jour- nal of Finance, 45(3), 881–898

    1990

  12. [12]

    Lo, A.W., & MacKinlay, A.C. (1988). Stock market prices do not follow random walks: Evidence from a simple specification test.Review of Financial Studies, 1(1), 41–66

    1988

  13. [13]

    Lo, A.W., & MacKinlay, A.C. (1989). The size and power of the variance ratio test in finite samples: A Monte Carlo investigation.Journal of Econometrics, 40(2), 203–238

    1989

  14. [14]

    Lo, A.W., &MacKinlay, A.C.(1990).Aneconometricanalysisofnonsynchronous trading.Journal of Econometrics, 45(1–2), 181–211

    1990

  15. [15]

    Madhavan, A. (2000). Market microstructure: A survey.Journal of Financial Markets, 3(3), 205–258

    2000

  16. [16]

    Roll, R. (1984). A simple implicit measure of the effective bid-ask spread in an efficient market.Journal of Finance, 39(4), 1127–1139

    1984

  17. [17]

    Scholes, M., & Williams, J. (1977). Estimating betas from nonsynchronous data. Journal of Financial Economics, 5(3), 309–327

    1977

  18. [18]

    Stoll, H.R. (1978). The supply of dealer services in securities markets.Journal of Finance, 33(4), 1133–1151

    1978

  19. [19]

    Stoll, H.R. (1989). Inferring the components of the bid-ask spread: Theory and empirical tests.Journal of Finance, 44(1), 115–134. 29

    1989