📊 Principal Component Analysis — Multi-Asset ETFs

SPY, QQQ, TLT, GLD, TIP, DBA, DBC, USO | Weekly returns | 5 Years (262 obs)

Variance Explained by PC

Correlation Matrix

SPYQQQTLTGLDTIPDBADBCUSO
SPY1.000.940.110.150.290.160.170.11
QQQ0.941.000.130.110.230.120.110.05
TLT0.110.131.000.120.68-0.11-0.18-0.23
GLD0.150.110.121.000.280.160.320.15
TIP0.290.230.680.281.000.090.190.12
DBA0.160.12-0.110.160.091.000.430.24
DBC0.170.11-0.180.320.190.431.000.88
USO0.110.05-0.230.150.120.240.881.00

PC Loadings (first 4 components — 91.99% of variance)

Interpretation:
PC1 (53.1%): Oil/Commodities factor — USO (0.87), DBC (0.45), DBA (0.13) dominate. This is the energy/real assets dimension.
PC2 (22.3%): Equity risk factor — QQQ (0.77), SPY (0.58). Pure equity beta.
PC3 (9.2%): Gold/inflation hedge — GLD (0.81), TLT (0.37), DBA (0.27). Real asset + duration.
PC4 (7.4%): Duration/rates — TLT (0.61), DBA (-0.73). Long bonds vs agriculture (inflation regime switch).

PC Scores Over Time

PC scores show the market's exposure to each factor at each point in time. Large moves = factor shock.

PC1 vs PC2 — Factor Space (each dot = 1 week)

X = oil/commodities factor, Y = equity factor. Quadrants: top-right = risk-on + oil up, bottom-left = risk-off + oil down.
PCA on weekly return covariance matrix. Source: Yahoo Finance | Chart: Marshall 🔶