Week 40 Day 6: Correlation: Why Diversification Actually Works

If stocks and bonds had a correlation of +1, there would be no point in owning both -- they would crash together. But stocks and bonds historically have approximately 0.0 to -0.3 correlation: when stocks crash, bonds often hold steady or rise. Combining them reduces your portfolio's overall volatility without proportionally reducing your expected return. That is the free lunch of diversification.

Historical correlations between asset classes (approximate): U.S. stocks (VTI) vs. bonds (BND): approximately -0.1 to +0.2. U.S. stocks vs. international stocks (VXUS): approximately +0.8. U.S. stocks vs. REITs (SCHH): approximately +0.6. U.S. stocks vs. gold: approximately +0.0. U.S. stocks vs. TIPS (VTIP): approximately +0.0 to +0.2. Bonds vs. TIPS: approximately +0.5. Key insight: the best diversifiers have LOW correlation with stocks. Bonds and gold are the classic portfolio diversifiers, not because they earn high returns but because they move independently of (or opposite to) stocks. The diversification math: two assets each with 16% standard deviation and 0.0 correlation, held 50/50, create a portfolio with approximately 11.3% standard deviation -- a 29% reduction in risk for free. IMPORTANT: correlations are not stable. During the 2008 crisis, correlations between many asset classes spiked toward +1 (everything crashed together). This is 'correlation breakdown' and is the biggest limitation of diversification. However, bonds (especially U.S. Treasuries) maintained their negative correlation with stocks during the 2008 and 2020 crashes. Notable exception: 2022 was unusual -- both stocks AND bonds declined simultaneously (rising interest rates hurt both). This was the first significant positive-correlation stress event in decades, reminding investors that no risk management framework is guaranteed.

The mathematics of diversification derive directly from portfolio variance: Var(P) = sum_i sum_j w_i * w_j * sigma_i * sigma_j * rho_ij, where w_i are portfolio weights, sigma_i are asset standard deviations, and rho_ij are pairwise correlations. When rho < 1, portfolio variance is less than the weighted average of component variances -- the 'diversification benefit.' This benefit is maximized when rho is negative and vanishes when rho = +1. Markowitz (1952) showed that the optimal portfolio (the one that maximizes return per unit of risk) is found by solving this variance minimization problem subject to a return constraint, producing the 'efficient frontier.' Correlation stability is the critical assumption: DeMiguel, Garlappi, and Uppal (2009) showed that estimation errors in the correlation matrix (which is notoriously unstable) can make the mathematically optimal portfolio perform WORSE than a simple 1/n equal-weight portfolio -- the optimal solution is optimal only for the estimated parameters, which may differ significantly from the true parameters. For practical portfolio construction, this favors simple allocation rules (60/40, 80/20, three-fund portfolio) over optimization-based allocations that depend on precise correlation estimates. The 2022 stock-bond positive correlation episode was explained by Ilmanen (2003, 2022): when inflation is the dominant risk (as in 2022), stocks and bonds are positively correlated (both lose value when rates rise). When growth is the dominant risk (as in 2008, 2020), stocks and bonds are negatively correlated (stocks fall, bonds rally as rates are cut). The regime-dependent nature of correlations argues for multi-asset diversification (including TIPS, which benefit directly from inflation) rather than reliance on the stock-bond correlation alone.