Valen0604

Macroeconomic Policy and Sector ETF Returns

Provost Undergraduate Research Scholarship (PURS), University of Houston, Fall 2025. Revised August 2026.

Question

Do macroeconomic policy variables (interest rates, inflation, federal spending, tariffs, trade balance) explain U.S. sector ETF returns beyond what standard equity risk factors already capture?

Data

Series	Source	Frequency
11 SPDR sector ETFs (XLB, XLC, XLE, XLF, XLI, XLK, XLP, XLRE, XLU, XLV, XLY), adjusted close	Yahoo Finance	Monthly, Dec 1998 – Sep 2025 (XLC from Jun 2018, XLRE from Oct 2015)
Fama-French three factors and risk-free rate	Kenneth French Data Library	Monthly
Effective federal funds rate (DFF)	FRED	Daily
CPI (CPIAUCSL) and inflation rate	FRED / BLS	Monthly
Federal government expenditures (FGEXPND)	FRED	Quarterly
Customs duties receipts	FRED	Quarterly
Trade balance, goods and services (BOPGSTB)	FRED	Monthly
Average effective tariff rate	USITC	Annual
VIX (open)	CBOE via Yahoo Finance	Monthly

All data are in PURS Data.xlsx, one sheet per series.

Method

For each ETF, the excess real log return is regressed on the Fama-French factors plus changes in the macro variables:

r_t − RF_t = α + β·Mkt_RF + s·SMB + h·HML
          + γ₁ ΔFedFunds + γ₂ ΔInflation + γ₃ Δlog(Spending) + γ₄ Δlog(Receipts)
          + γ₅ Δlog(TradeBalance) + γ₆ ΔTariff + γ₇ ΔVIX + ε

Nominal series are deflated by CPI. Standard errors are Newey-West (lag 3 monthly, lag 1 quarterly).

The model is estimated at monthly frequency for all 11 ETFs, and at quarterly frequency for the 9 ETFs with a full history, because three regressors (spending, receipts, trade balance) are quarterly and would otherwise be zero-padded two months in three.

An influence check (Cook’s distance, re-estimation excluding 2020) is run on the Energy model, which is the only sector with material macro exposure.

Findings

• Sector returns are dominated by the market and value/growth factors. Market betas range from ~0.5 (Staples, Utilities) to ~1.2 (Financials, Materials); HML loadings separate Energy and Financials (value) from Tech (growth). R² is 0.5–0.85 for most sectors, almost entirely from these three factors.
• Energy is the only sector with robust macro exposure. Its return rises with inflation (quarterly β ≈ 2.7–3.0, p < 0.002, robust to excluding 2020) and with federal spending growth (β ≈ 0.3–0.4, p ≈ 0.03 excluding 2020). A negative relationship with customs receipts is present but driven largely by the 2025 tariff shock and should not be read as a general effect.
• Interest rates explain little at monthly frequency. At quarterly frequency Materials and Industrials show negative rate sensitivity; the result does not appear monthly and is reported as suggestive only.
• Tariff effects are unstable. The annual tariff series changes once a year, so estimates depend heavily on a handful of observations. Health Care is the only sector with a consistent (positive) coefficient across frequencies; other sector results flip between monthly and quarterly.

  Limitations

• Mixed data frequencies: quarterly and annual regressors are forward-filled at monthly frequency. The quarterly specification addresses this for the quarterly series but not for the annual tariff rate.
• ~180 coefficient tests across the two frequency specifications; individual results at p ≈ 0.05 should be discounted accordingly.
• The federal funds rate on the first day of each month is used as the month’s rate; a monthly average would be less noisy.
• XLC (86 months) and XLRE (119 months) have short histories and are excluded from the quarterly model.

  Revision history

The original Fall 2025 version contained two data-handling errors: the fed funds series was assigned synthetic dates rather than its actual dates, and ETF returns were computed from monthly open prices, which misaligned them with the month-end factor returns by one period. Both are corrected here. The original headline result (strong negative Energy sensitivity to real interest rates) does not survive the corrections; the inflation exposure reported above is the corrected finding.

Running

install.packages(c("tidyverse", "readxl", "sandwich", "lmtest", "broom"))
source("PURS.R")

Requires PURS Data.xlsx in the working directory.