XLE 10-day relative strength vs SPY: forward-return predictability and Brent-trend regime flip
The idea was elegant: when Brent is below its 50-day trend, a strong XLE relative to SPY is likely positioning noise that should fade; when crude is above trend, that same strength should reflect real fundamentals and persist. Tested over roughly three years of overlapping daily observations, only half of that story survives.
Below a weak Brent tape, the relationship tip-toed negative — relative strength slightly foreshadowed weaker XLE returns — but the effect was mild, with a slope of -0.216 and a p-value just shy of conventional significance. Above trend, the persistence leg simply did not show up. The slope was basically flat, and the regime difference itself was statistically indistinguishable from zero. The full charts and regression tables are below; the honest takeaway is that this rule is more intuition than edge.
Over the past ~3 years, does XLE's 10-day total return relative to SPY predict XLE's forward 10-day return, and does the relationship flip depending on whether Brent crude is above or below its 50-day moving average? I expect relative strength to mean-revert when Brent is below trend—energy spikes against a weak crude tape are positioning-driven and fade—whereas relative strength above trend reflects genuine supply-demand re-rating and persistence.
How this was measured
Resampled XLE and SPY minute bars to daily closes, then computed 10-trading-day total returns. Relative strength is defined as XLE 10-day return minus SPY 10-day return. The predictor is the lagged relative-strength observation; the outcome is XLE's forward 10-day return. Brent regime is assigned using the PRIOR trading day's Brent close versus its 50-day moving average to avoid same-day settlement look-ahead. Because 10-day returns overlap daily, all regression p-values use HAC (Newey-West with 10 lags) standard errors. The below-trend and above-trend slopes are estimated both separately and inside a single interaction model, with the interaction term measuring the regime flip.
The key numbers
Reading the numbers
Across 694 overlapping daily observations, the predictive slope was negative in both Brent regimes: -0.216 below the 50-day MA and -0.085 above it. Neither is statistically clear (p≈0.08 and p≈0.36), so the data do not confirm a reliable regime flip.
The charts
Each dot pairs a day's prior 10-day XLE-vs-SPY relative strength with the next 10-day XLE return, but only for days when Brent traded below its 50-day moving average. The cloud is wide and scattered, and the estimated downward slope is -0.216, meaning stronger recent relative strength was weakly associated with lower forward returns. The p-value of 0.078 is borderline but not below the usual 0.05 threshold, so this mean-reversion pattern is suggestive rather than decisive.
This is the same scatterplot for the 285 days when Brent was above its 50-day moving average. The slope is also negative at -0.085, but flatter than in the below-trend regime and far from statistically clear with p=0.358. For the user's question, this matters because the expected flip to persistence above trend does not show up: higher relative strength is not followed by reliably stronger forward XLE returns when crude is above trend.
The two bars directly compare the fitted slopes: -0.216 when Brent is below its 50-day MA and -0.085 when Brent is above it. Both bars sit below zero, so the relationship is negative in both regimes; the above-trend bar is merely less negative, not positive. The 0.131 gap between them points in the direction of a regime difference, but the interaction p-value of 0.375 says that gap could easily be noise, so the hypothesized negative-to-positive flip is not supported.
Regime-specific predictability summary
| Regime | N | HAC slope | HAC t-stat | HAC p-value | Pearson r | Spearman rho |
|---|---|---|---|---|---|---|
| Brent < 50d MA | 409 | -0.2163 | -1.765 | 0.0775 | -0.1879 | -0.1623 |
| Brent > 50d MA | 285 | -0.0851 | -0.92 | 0.3577 | -0.1033 | -0.0274 |
The takeaway
Bottom line: over roughly three years and 694 overlapping daily observations, XLE's 10-day relative strength versus SPY is not a dependable predictor of the next 10-day XLE return, and the Brent 50-day trend does not cleanly flip the relationship. When Brent traded below its 50-day MA, the slope was negative (-0.216, p≈0.078), which directionally matches your mean-reversion idea — but with 409 such days, this is a mild lean, not a proven effect. Above trend, the slope was basically flat (-0.085, p≈0.36), so the persistence leg of your hypothesis failed to appear; if anything, the relationship was slightly negative there too. The regime shift itself was small (+0.131 slope difference) and statistically indistinguishable from zero (interaction p≈0.375), meaning the two regimes are not reliably different. Practical takeaway: don't hang a strategy on this rule. The below-trend mean-reversion signal is the only hint of life, and it is too weak to separate from noise and overlapping-window artifacts.
The fine print
- 10-day overlapping windows leave the effective sample much smaller than 694; HAC errors help but a non-overlapping test would be the stricter check.
- Brent regime uses the prior day's close versus its 50-day MA; results could shift if same-day Brent close is used instead.
- The three-year sample is dominated by one energy/commodity cycle, so the below/above split may not generalize to other regimes.
- XLE is an energy-equity ETF, not crude itself; other equity, yield, and positioning factors may be doing part of the work.