Question
Do two major oil-sector equities (Exxon Mobil Corporation, XOM, and Chevron Corporation, CVX) exhibit statistically significant correlation in daily returns, and what is the magnitude, confidence interval, and temporal stability of their coupling—testing whether same-sector geographic/commodity proximity predicts structural equivalence in return dynamics?
Method
We computed the Pearson correlation coefficient between daily adjusted-close returns for XOM and CVX over the period 2010-01-01 to 2024-12-31 (n = 3,772 trading days), sourced from yfinance. Statistical significance was established via a distribution-free permutation test with 2,000 random shuffles of one return series, yielding an empirical p-value. The 95% confidence interval was constructed via 2,000 bootstrap resamples of the paired return series. To assess temporal stability, we recomputed the Pearson correlation within each calendar year (2010–2024) using the same method, producing a rolling per-year series that reveals time variation in the coupling strength. The Spearman rank correlation was computed as a robustness check against outliers and non-linearity. This design isolates the return co-movement between two firms sharing sector (integrated oil & gas), commodity exposure (crude oil, natural gas, refined products), and overlapping geographic footprints (U.S. operations, global upstream/downstream assets), testing whether such proximity in categorical space—sector, commodity, and partial geographic overlap—manifests as strong structural equivalence in equity returns.
Result
The full-sample Pearson correlation is r = 0.8329 (95% CI: [0.8149, 0.8500], p = 0.0005, n = 3,772). The Spearman rank correlation is ρ = 0.8155, confirming robustness to rank transformations. The permutation test decisively rejects the null hypothesis of zero correlation (p = 0.0005 < 0.001), indicating that the observed coupling is statistically significant at conventional thresholds.
The per-calendar-year Pearson correlations (in-sample within each year) reveal substantial time variation:
- 2010: 0.815
- 2011: 0.903
- 2012: 0.785
- 2013: 0.724
- 2014: 0.799
- 2015: 0.871
- 2016: 0.743
- 2017: 0.607
- 2018: 0.750
- 2019: 0.755
- 2020: 0.866
- 2021: 0.905
- 2022: 0.878
- 2023: 0.840
- 2024: 0.720
The correlation ranges from a low of 0.607 in 2017 to a high of 0.905 in 2021, with a mean of approximately 0.80 and standard deviation of roughly 0.08. The series exhibits clustering: elevated correlations in 2011, 2015, 2020–2022 (periods encompassing the Arab Spring oil price spike, the 2015 oil price collapse, and the COVID-19 pandemic/recovery), and lower correlations in 2013, 2017, and 2024. The 95% confidence interval [0.8149, 0.8500] for the full-sample estimate is narrow, reflecting the large sample size, but the per-year variation demonstrates that the coupling is not constant—it strengthens during periods of sector-wide stress or commodity price volatility and weakens during relative stability or firm-specific divergence.
Interpretation
The result quantifies a strong, statistically significant, and time-varying positive correlation between XOM and CVX daily returns. The full-sample estimate of 0.83 implies that approximately 69% of the variance in one stock's returns is linearly associated with the other's (r² ≈ 0.69), a magnitude consistent with structural equivalence: the two firms occupy nearly identical positions in sector/commodity space, face common oil price shocks, regulatory shifts, and macroeconomic drivers (interest rates, global demand), and exhibit overlapping operational geographies (U.S. refining, Gulf of Mexico production, international upstream portfolios).
The per-year dynamics reveal that this equivalence is not mechanically fixed but responds to the economic environment. The correlation peaks above 0.90 in 2011 and 2021—years marked by acute sector-wide shocks (Libyan civil war supply disruption in 2011; pandemic demand collapse and OPEC+ supply cuts in 2020–2021). During such episodes, idiosyncratic firm differences (management decisions, asset mix, hedging strategies) are swamped by common factors, and the two equities move nearly in lockstep. Conversely, the correlation drops to 0.607 in 2017, a year of relative oil price stability and divergent firm strategies (XOM's aggressive capital spending under CEO Darren Woods vs. CVX's capital discipline). The 2024 decline to 0.72 may reflect renewed firm-specific differentiation (e.g., differing exposure to Permian Basin growth, LNG export capacity, or renewable energy pivots) as the sector exits the post-pandemic volatility regime.
This time variation is economically interpretable: the gravity model metaphor—where "distance" in sector/commodity space predicts correlation—holds on average but is modulated by the salience of common vs. idiosyncratic factors. When commodity price volatility is high, the shared oil-price "gravity well" dominates, pulling the two return series together. When volatility is low, firm-specific "distances" (operational efficiency, reserve quality, financial leverage, ESG positioning) reassert themselves, and the correlation decays toward the lower bound observed in 2017. The result does not support a claim of perfect hedging equivalence (correlation = 1.0 would be required), but it does support strong substitutability for portfolio construction: holding one stock provides substantial exposure to the other's systematic risk, though not complete redundancy.
The result does not establish causality or predict future returns. It is a descriptive bound on historical co-movement, not a forward-looking trading signal. The correlation could weaken if the firms' operational profiles diverge further (e.g., one pivots aggressively to renewables) or strengthen if sector consolidation or regulatory harmonization increases homogeneity. The confidence interval [0.8149, 0.8500] applies only to the 2010–2024 window and assumes the return-generating process is stationary within that period—an assumption the per-year variation challenges.
Relation to the Literature
The result sits at the intersection of gravity models of spatial/categorical proximity and financial return co-movement. [P1], [P2], [P3], and [P4] establish the gravity model framework in non-financial contexts: student university choice [P1], park visitation [P2] [P4], and foreign direct investment [P3] all exhibit distance decay—interaction strength falls with geographic, institutional, or informational distance. [P8] applies this logic directly to cross-border equity flows, finding that "the geography of information heavily determines the pattern of international transactions"—investors overweight proximate (informationally or geographically close) assets. Our result extends this to intra-sector equity return correlation: XOM and CVX are "close" in sector/commodity space (zero distance in GICS classification, near-zero distance in commodity exposure), and their return correlation of 0.83 quantifies the strength of that proximity.
[P6] and [P10] analyze spatial correlation networks in real estate and land use, using modified gravity models to construct correlation matrices and identify clusters. [P6] finds that Chinese real estate price co-movements are "relatively low and stable" except during COVID-19, when synchronization spiked—a pattern mirrored in our per-year series (2020–2021 correlation surge). [P10] applies social network analysis to non-grain land use correlations, finding that "high-degree nodes form control poles"—analogous to our finding that sector-wide shocks (oil price volatility) act as a "control pole" that synchronizes XOM and CVX returns. The difference is that [P6] and [P10] study spatial networks across many nodes, while we study a two-node system in categorical (sector/commodity) space; the gravity logic is the same.
[P9] argues that geographic concentration of production (footwear in the U.S.) persists due to "heterogeneity in entrepreneurial opportunities" and tacit knowledge spillovers, not just efficiency. This suggests that sector clustering (oil majors in Houston, tech firms in Silicon Valley) creates informational proximity that amplifies return correlation—analysts, investors, and executives share information, reinforcing co-movement. Our result is consistent: XOM and CVX are embedded in the same informational and institutional network (energy conferences, analyst coverage, index inclusion), which may sustain the high correlation even when operational "distance" (firm-specific strategies) increases.
[P7] discusses urban sprawl and the difficulty of "a usefully integrated economic model of urban economies"—a challenge that parallels the difficulty of modeling equity return correlations without a unified framework for sector, commodity, geographic, and informational proximity. Our result does not resolve this, but it provides a quantified empirical bound on one dimension (same-sector, overlapping-geography, common-commodity) that future models must match.
The literature does not provide a direct precedent for our specific computation (XOM-CVX daily return correlation with bootstrap CI and per-year dynamics), so the result is a novel empirical contribution. It confirms the gravity model intuition—proximity predicts correlation—but adds the crucial finding that the mapping is time-varying and state-dependent, not a fixed structural parameter.
Limitations
Two-asset system: The result quantifies one pairwise correlation. It does not test whether the gravity model generalizes across the full energy sector (e.g., does correlation decay with firm size, geographic distance, or commodity mix?). A multi-asset network analysis (e.g., all S&P 500 energy stocks) would test distance decay more rigorously.
Sector homogeneity: XOM and CVX are both large-cap integrated oil & gas firms with similar business models. The high correlation may reflect mechanical similarity (both are leveraged bets on crude oil prices) rather than a deep gravity mechanism. Testing pairs with greater categorical distance (e.g., XOM vs. a pure-play E&P firm, or XOM vs. a renewable energy stock) would isolate the distance-decay effect.
No out-of-sample validation: The per-year correlations are in-sample within each year. We do not test whether the 2010–2023 correlation predicts the 2024 correlation, or whether a model trained on early years forecasts later years. Out-of-sample R² would quantify predictive power.
Stationarity assumption: The bootstrap CI assumes the return distribution is stationary over 2010–2024. The per-year variation (0.607 to 0.905) suggests non-stationarity—structural breaks around 2017 (correlation trough) and 2020–2021 (correlation peak) are plausible. A regime-switching model or rolling-window CI would be more robust.
Omitted common factors: The correlation conflates direct co-movement (e.g., operational spillovers, competitive responses) with indirect co-movement via common factors (oil prices, interest rates, market-wide risk appetite). A factor model (e.g., Fama-French plus oil price factor) would decompose the correlation into systematic and residual components, isolating the firm-pair-specific coupling.
Daily frequency: Daily returns include microstructure noise (bid-ask bounce, stale prices, intraday volatility). Weekly or monthly returns might yield a cleaner estimate of fundamental co-movement, though at the cost of sample size.
Survivorship and universe: Both firms survived the full 2010–2024 window. The result does not account for delisted energy firms (bankruptcies during the 2015–2016 oil crash, or 2020 pandemic), which might exhibit different correlation dynamics. The universe is also U.S.-listed large-caps; smaller firms or international peers (e.g., BP, Shell) might show weaker correlations due to regulatory or currency distance.
No causal identification: The result is purely descriptive. We cannot distinguish whether high correlation reflects (a) common shocks, (b) competitive interdependence (one firm's output decision affects the other's price), or (c) investor herding (index funds mechanically buying/selling both). An event study (e.g., firm-specific earnings surprises) or instrumental variable (e.g., exogenous regulatory shock to one firm) would test causality.
Interpretation of "distance": We invoke sector/commodity/geographic proximity as the "distance" metric, but we do not quantify it. A formal gravity model would regress pairwise correlations on explicit distance measures (e.g., Jaccard distance in asset portfolios, geographic distance between headquarters, text similarity in 10-K filings). Our result is a single data point in that hypothetical regression.
Strengthening the result would require: (a) expanding to a multi-asset panel to test distance decay, (b) out-of-sample validation of correlation forecasts, (c) factor decomposition to isolate residual co-movement, (d) event studies to test causal channels, and (e) formal specification of a distance metric in sector/commodity/geographic space. The current result establishes a quantified empirical bound on same-sector equity correlation and its time variation, providing a benchmark for gravity models of financial co-movement.
Research evidence, not investment advice: This is a research finding quantifying historical return co-movement between two equities. It is not a recommendation to buy, sell, or hold either security, nor a prediction of future correlation or returns. Portfolio construction and hedging decisions require consideration of investor-specific risk tolerance, horizon, and constraints, which this analysis does not address.