Resilience and regime change in refining margins: evidence from cointegration, Markov chains, and topological data analysis
Abstract
This paper analyses the 3:2:1 crack spread over 20 years (2006 to 2025) using three independent methods: Engle-Granger cointegration, Markov chain regime modelling, and persistent homology applied to Takens delay embeddings. The data are publicly available EIA spot prices obtained via the FRED database. The objective is to determine whether the long-run equilibrium relationship between crude oil and refined product prices has survived the structural shocks of the last decade: the oil price collapse of 2014-2016, the COVID-19 pandemic, the war in Ukraine, and the addition of new refining capacity in the Middle East and Africa. Cointegration holds in 68% of rolling five-year windows, confirming long-run equilibrium. The equilibrium margin has shifted from $12/bbl (2006-2021) to $18/bbl (2024-2025). The error correction half-life has increased from 0.9 to 5.8 months, with the adjustment coefficient becoming statistically non-significant after 2017 (p = 0.38). Markov chain analysis discretises the margin into five regimes and provides N-step transition probabilities. The steady-state probability of normal-or-better margins is 55%. Topological data analysis reveals the most striking change: the pre-2020 crack spread exhibited 18 to 24 persistent homological cycles reflecting seasonal and inventory patterns. The post-2020 periods show near-zero topological complexity. The cyclic structure that made margins partially predictable for 15 years has not survived the combined disruptions. The three methods converge on the same diagnosis through independent mathematical frameworks. The practical implication is that prediction of refining margins is not impossible but conditional: it works when the market has topological structure, and it does not work when the structure is absent. The tools presented here provide a way to distinguish between the two regimes in near-real time.
Related
- Topological Flame Bifurcation, Aerodynamic Flashback Margins, and Multi-Pathway NOx Scaling in a 3D Swirl-Stabilized 100% Pure Hydrogen Aero-Engine Combustor
- Do topological and quantum-inspired molecularrepresentations add value? Evidence from Africannatural-product-inspired antimalarial chemical space
- Refining machine learning potentials through thermodynamic theory of phase transitions
- Data Characterization and Multi-Plant Data Integration Strategies for Artificial Neural Network Modeling in Palm Oil Refining
- Refining AMOEBA Electrostatics for d‑Glucose and 1→4-Linked d‑Glucose Oligosaccharides
BibTeX
@misc{larraz2026resilience,
title = {{Resilience and regime change in refining margins: evidence from cointegration, Markov chains, and topological data analysis}},
author = {Rafael Larraz},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010459/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010459/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Resilience and regime change in refining margins: evidence from cointegration, Markov chains, and topological data analysis
AU - Rafael Larraz
PY - 2026
DA - 2026/10/14
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010459/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010459/v1
AB - This paper analyses the 3:2:1 crack spread over 20 years (2006 to 2025) using three independent methods: Engle-Granger cointegration, Markov chain regime modelling, and persistent homology applied to Takens delay embeddings. The data are publicly available EIA spot prices obtained via the FRED database. The objective is to determine whether the long-run equilibrium relationship between crude oil and refined product prices has survived the structural shocks of the last decade: the oil price collapse of 2014-2016, the COVID-19 pandemic, the war in Ukraine, and the addition of new refining capacity in the Middle East and Africa. Cointegration holds in 68% of rolling five-year windows, confirming long-run equilibrium. The equilibrium margin has shifted from $12/bbl (2006-2021) to $18/bbl (2024-2025). The error correction half-life has increased from 0.9 to 5.8 months, with the adjustment coefficient becoming statistically non-significant after 2017 (p = 0.38). Markov chain analysis discretises the margin into five regimes and provides N-step transition probabilities. The steady-state probability of normal-or-better margins is 55%. Topological data analysis reveals the most striking change: the pre-2020 crack spread exhibited 18 to 24 persistent homological cycles reflecting seasonal and inventory patterns. The post-2020 periods show near-zero topological complexity. The cyclic structure that made margins partially predictable for 15 years has not survived the combined disruptions. The three methods converge on the same diagnosis through independent mathematical frameworks. The practical implication is that prediction of refining margins is not impossible but conditional: it works when the market has topological structure, and it does not work when the structure is absent. The tools presented here provide a way to distinguish between the two regimes in near-real time.
ER -