1.4.4 The Value–Momentum Interaction

Basic Prerequisites: Probability & Measure Theory (correlation, variance) and Linear Algebra.

---

1. Intuition & Practical Objective

Value and momentum are the two best-documented return premia in finance - and they are negatively correlated. Value buys cheap, beaten-down assets; momentum buys recent winners. When one is working, the other tends to be struggling: after a long value drawdown the winners momentum wants are exactly the “expensive” names value wants to short, and vice-versa. This negative correlation is the single most useful practical fact in the momentum literature: combining them is not a compromise, it is a hedge.

Asness, Moskowitz & Pedersen (2013) document the premia in eight markets - US, UK, Europe, and Japan individual stocks plus global equity-index, bond, currency, and commodity futures. Key verified results:

• A momentum and a value premium in every market.
• Correlation between value and momentum residual returns is strongly negative: $\approx -0.60$ across stock markets (Japan $-0.64$), $\approx -0.49$ within non-stock asset classes.
• The equal-weighted $50/50$ combination $r^{\text{COMBO}}=0.5r^{\text{VALUE}}+0.5r^{\text{MOM}}$ outperforms either strategy alone in every market, and across all asset classes produces an annualized Sharpe near 1.45.
• Momentum is insignificant alone in Japan, yet the value–momentum combination there still improves the efficient frontier - the negative correlation does the work even where momentum alone “fails.”

---

2. Mathematical Ground Truth & Derivations

Combination variance. Let value and momentum have monthly means ${\mu }_V,{\mu }_M$, vols ${\sigma }_V,{\sigma }_M$, and correlation $\rho =\mathrm{corr}(r_V,r_M)$. The equal-weight combination $r_C=\frac{1}{2}{r}_V+\frac{1}{2}{r}_M$ has

$${\mu }_C=\frac{1}{2}{\mu }_V+\frac{1}{2}{\mu }_M,{\sigma }_C^2=\frac{1}{4}{\sigma }_V^2+\frac{1}{4}{\sigma }_M^2+2\cdot \frac{1}{4}\rho {\sigma }_V{\sigma }_M.$$

With $\rho <0$ the covariance term is negative, shrinking ${\sigma }_C$ below either leg while keeping ${\mu }_C$ as the average of two positive means. The Sharpe of the combination, ${\text{SR}}_C={\mu }_C/{\sigma }_C$, can exceed both ${\text{SR}}_V$ and ${\text{SR}}_M$ - the hallmark of a genuine diversifier.

Why is $\rho$ negative? Intuitively, both strategies are reversion bets on mispricing but on opposite timing: value assumes price over-reacts to bad news and will revert (buy the fallen), momentum assumes price under-reacts to good news and will continue (buy the risers). They are the two halves of the behavioral “over-/underreaction” tension (Daniel, Hirshleifer & Subrahmanyam 1998; Barberis, Shleifer & Vishny 1998). Empirically the negative correlation is robust across markets and asset classes (AMP 2013).

Optimal weights. The unconstrained tangency weights for the two-asset portfolio are ${\mathbf{w}}^{\ast }\propto {\Sigma }^{-1}\mathbf{\mu }$, which put meaningful weight on both legs precisely because $\rho <0$ raises the diversification benefit. Equal-weighting ($0.5/0.5$) is the robust, simple rule that captures most of this.

---

3. Computational Implementation - why the 50/50 works

Standard library only. Generates two positively-expected-return, negatively-correlated return streams (momentum and value), computes the $50/50$ combination, and verifies the combination Sharpe exceeds each leg and the combination variance identity holds exactly.

The combination’s Sharpe ($+0.67$) nearly doubles ($1.95\times$) the better single-leg ($+0.34$): the negative correlation roughly halves the combo’s volatility (1.89% vs the 3.4–4.9% legs) while keeping the mean near the average. The variance identity is reproduced exactly.

---

4. Failure Modes & First-Principles Breakdowns

1. Correlation is state-dependent. Value–momentum correlation is not constant: it is most negative precisely when it matters (after market declines, when momentum crashes and value rebounds). A naive assumption of constant $\rho$ overstates the steady-state diversification benefit.
2. “Add value to anything” is a myth. The benefit is specific to the negative correlation. Pairing momentum with a positively-correlated momentum-like factor (e.g. two trend signals on the same universe) does nothing for diversification and can concentrate crash risk.
3. Japan’s lesson. Momentum alone is weak/insignificant in Japan, yet the combination still helps - but only because the negative correlation is present. The empirical rule is “combine for correlation reasons,” not “combine because each premium always works.”
4. Re-balancing and crowding. A 50/50 that is never rebalanced drifts into the higher-return leg over time; and both legs being crowded compress the premia the combination was built to harvest (see 05 · Failure Modes on crowding).

---

5. References

• Asness, Moskowitz & Pedersen (2013), Value and Momentum Everywhere, J. Finance 68(3)
• Asness, Frazzini, Israel & Moskowitz (2014), Fact, Fiction, and Momentum Investing
• Fama & French (1993, 1996)
• Daniel, Hirshleifer & Subrahmanyam (1998) and Barberis, Shleifer & Vishny (1998)

---

6. Connected Graph Bridges

• Base: Probability & Measure Theory (correlation, variance) · Linear Algebra
• Continue: 05 · Failure Modes · 06 · Advanced Extensions
• Factor view: Fundamental Multi-Factor Models (value HML + momentum UMD)
• Portfolio construction: Mean–Variance (combo weights) · Risk Parity