Pipeline capacity utilization explains 45-76% of the mid-continent crude basis spread — but the effect fades toward the coast and evaporates in rail-flexible basins like the Bakken

Summary

Pipeline capacity utilization out of a producing basin is a large and statistically powerful explanation for local crude basis spreads — but only in basins where pipeline is effectively the sole outbound mode, and the relationship is strongly nonlinear and region-dependent, not a single universal coefficient. The most direct quantitative answer comes from Agerton & Upton (Energy Journal, 2019): a regression of each U.S. crude benchmark's discount to Brent on the share of mid-continent-to-Gulf-Coast shipments moving by rail/barge (the inverse of pipeline utilization) explains 72-76% of the variance (pseudo-R²) in the WTI Cushing and WTI Midland discounts, versus under 1% explained by a refining-capacity-mismatch variable — pipeline/shipping constraints dominate refining constraints by roughly two orders of magnitude for these benchmarks. But that explanatory power falls off toward the coast: it drops to ~45-48% for WTS and LLS and to under 20% for the already-coastal HLS and Gulf Coast fuel oil, which are much less dependent on inland pipeline capacity in the first place. McRae (NBER WP 24170, 2017) confirms the mechanism is threshold/nonlinear: within available capacity, added Permian production reduces producer price only mildly (~$1.61/bbl per 100,000 b/d), but once production exceeds combined refinery-plus-pipeline capacity, price falls sharply further (~$11.15/bbl per additional 50,000 b/d) — and a hypothetical pipeline eliminating a real 2014 bottleneck was estimated to raise producer prices by $5.72-$10.86/bbl. Does it hold across basins? Yes for the Permian and for Colorado's DJ/Niobrara Basin, where pipeline is the binding transport mode. No for the Bakken: McRae finds the North Dakota price differential statistically uncorrelated with pipeline-capacity utilization, and Luong (2023, Resources Policy) independently confirms via cointegration analysis that Bakken sweet crude stays linked to national WTI even during pipeline congestion — because the region's large, flexible rail network lets shippers route around a congested pipeline in a way the Permian cannot.

Data note: askamerica's own warehouse (energy, econ, cftc schemas) carries national/state crude production and refinery utilization and the national WTI Cushing benchmark, but has no basin-specific basis-spread series (WTI Midland, Bakken sweet at Williston) and no pipeline-capacity-by-basin table — both "cushing" and "midland" returned zero catalog matches. This is a confirmed coverage gap; every quantitative figure below is drawn from primary academic research (read directly from the source PDFs) and EIA/market reporting instead.

How much does capacity utilization explain? Two independent, converging estimates

Agerton & Upton (2019, published in The Energy Journal; working paper dated Sept. 2017) regress each crude benchmark's monthly discount to Brent on (a) the share of PADD 2/4-to-PADD 3 crude movements by rail and barge — a direct inverse proxy for pipeline capacity utilization — and (b) the average API gravity of PADD 3 refinery inputs, a proxy for refining-constraint mismatch. Results (their Table 3, read directly from the paper):

• WTI Cushing: shipping-constraint R² = 0.720 (total model R² = 0.724); refining-constraint R² = 0.007
• WTI Midland: shipping-constraint R² = 0.742 (total R² = 0.755); refining-constraint R² = 0.006
• WTS (mid-continent sour): shipping R² = 0.448 (total 0.480)
• LLS (Gulf Coast light sweet): shipping R² = 0.480 (total 0.517)
• HLS (Gulf Coast heavy sour): shipping R² = 0.181 (total 0.221)
• Fuel oil, U.S. Gulf Coast: shipping R² = 0.145 (total 0.158)

In dollar terms, they estimate a 10-percentage-point increase in the rail/barge share of mid-continent-to-Gulf-Coast shipments is associated with a $2.01-$3.79/bbl discount for the three mid-continent crudes; at the point where rail/barge carried 60% of volumes (near the peak of the 2012-2014 bottleneck), this implies a discount of roughly $12.58-$22.72/bbl attributable to shipping constraints alone. Their headline conclusion, stated directly in the paper: "scarce pipeline capacity explains half to three quarters of the deviation of mid-continent crude oil prices from their long-run relationship with Brent crude" — and they find no significant contribution from refining constraints, which some contemporaneous commentary (around the 2015 crude-export-ban debate) had blamed instead.

McRae (2017, NBER WP 24170) takes a threshold-regression approach using "excess production" (output minus local refinery input capacity minus pipeline export capacity) for the Permian specifically. All specifications find a positive, statistically significant coefficient on excess production, with a much steeper slope once production exceeds capacity — evidence the relationship is a nonlinear threshold effect, not a smooth linear one. This matches the 2018-19 real-world Permian episode: the WTI Midland-Cushing spread widened to nearly $18/bbl in September 2018 when Permian pipelines saturated, narrowed to under $6/bbl within three weeks of the 350,000 b/d Sunrise Pipeline expansion starting up, and fell to $0.25/bbl by March 2019 after the Seminole-Red repurposing added a further 200,000 b/d (EIA, March 2019).

Does the relationship hold across basins?

Permian Basin — holds strongly. Pipeline is nearly the only outbound mode; both studies above find large, statistically robust relationships between capacity utilization and basis spread, and the mechanism is directly observable in real episodes (2012-2015 and 2018-2019 Permian bottlenecks and their resolutions).

Colorado (DJ/Niobrara Basin) — holds. McRae finds the same qualitative pattern: production exceeded refinery+pipeline capacity from 2013-2015, producing roughly a $2.80/bbl discount to WTI Cushing; three new pipelines (White Cliffs expansion, Northeast Colorado Lateral, Saddlehorn) pushed capacity above production by late 2015-2016 and the discount closed.

Bakken (Williston Basin, North Dakota) — does NOT hold. McRae reports the North Dakota price differential is statistically uncorrelated with excess production. Two reasons converge: (1) Bakken excess production has been persistently positive since roughly 2010-2011 despite pipeline capacity nearly quadrupling (232,000 b/d in 2007 to 833,600 b/d in 2016), leaving too little within-basin variation between "constrained" and "unconstrained" regimes to identify a threshold effect the way Permian/Colorado data allow; and (2) the Bakken developed an unusually large, flexible rail network early — EIA (2013) reports Williston Basin rail takeaway capacity grew from 265,000 b/d in 2011 to roughly 660,000 b/d by December 2012, allowing Bakken crude to bypass a congested Cushing gateway entirely and reach East, West, and Gulf Coast refineries. This is independently corroborated by Luong (2023, Resources Policy, "Crude oil pipeline constraints: A tale of two shales"): using a cointegration framework, Bakken sweet crude at Williston remains cointegrated with national WTI essentially throughout the sample even during pipeline congestion, while WTI Midland only re-cointegrates with WTI after a Permian bottleneck clears. Two independent studies, different methods and periods, same conclusion: the relationship between pipeline utilization and local basis spread is powerful specifically where pipeline is the only real transport option, and it disappears where a flexible substitute (rail) exists.

Current relevance (2026): East Daley Analytics' March 2026 note on a sustained-$100/bbl WTI scenario finds the same dynamic still governs Permian pricing today: with corridors like Cactus II already operating above nameplate capacity and Gray Oak/Cactus III near full utilization, an incremental ~500,000 b/d of 2027 Permian supply is expected to widen the Midland-to-Houston (MEH) differential further — the identical utilization-to-basis-spread mechanism documented in the 2012-2019 academic literature, still operative fourteen years into the shale boom.

Caveats and confounds

Neither study isolates capacity utilization from every other driver of basis spreads. Agerton & Upton's shipping-share variable is itself an outcome of the capacity/demand balance rather than a pure exogenous "capacity utilization" measure, and both studies' sample periods (roughly 2006-2016 for McRae, 1990-2015 for Agerton & Upton) predate the 2015 crude-export-ban repeal and the more recent 2018-19 and 2026 episodes cited qualitatively above from EIA and market sources. McRae explicitly controls for the WTI Cushing-LLS differential and storage dynamics at Cushing (also documented by Büyükkahin, Lee, Moser & Robe 2013, whose paper shows Cushing storage conditions independently move the WTI-Brent spread in a nonlinear fashion) and finds the capacity-utilization terms remain large and significant net of those controls. Agerton & Upton similarly control for refinery-input API gravity and find it explains far less than shipping constraints. Both papers' R²/coefficient estimates should be read as historical, sample-specific magnitudes (particularly strong during the 2011-2015 and 2018-19 constrained episodes) rather than a fixed universal elasticity that applies unchanged in every period or price environment.

What This Report Does Not Answer

• Use askamerica's own data where possible: search_catalog confirmed no basin-level basis-spread (Midland/Cushing/Bakken-specific) or pipeline-capacity-by-basin table exists in the corpus; the energy schema has only national/state crude production and refinery utilization, so the quantitative answer relies entirely on primary academic literature (read directly) and EIA/market sources, as disclosed in the Summary and sources list.