This study examines how international crude oil price movements are transmitted to refined petroleum product prices, with a focus on asymmetric adjustment—i.e., whether the speed and magnitude of price responses differ between positive (upward) and ...
This study examines how international crude oil price movements are transmitted to refined petroleum product prices, with a focus on asymmetric adjustment—i.e., whether the speed and magnitude of price responses differ between positive (upward) and negative (downward) shocks. Such asymmetry can feed directly into inflation dynamics and consumer fuel expenditures, yet prior studies have not reached a consistent conclusion. A key reason is that empirical findings are sensitive to the structural assumptions of the econometric model and to sample design choices (analysis window, data frequency, market/product coverage, and treatment of structural instability). Rather than stopping at a single binary verdict (“asymmetry exists or not”), this study estimates four representative asymmetric price transmission models on an identical crude–refined product dataset and provides a quantitative, model-to-model comparison.
The dataset consists of West Texas Intermediate (WTI) spot prices and refined product spot prices from major U.S. markets (New York Harbor, U.S. Gulf Coast, and Los Angeles) for gasoline, diesel, jet fuel, and heating oil, spanning June 2006 to September 2025. Both monthly and weekly series are constructed to separate frequency effects. The four models are the Asymmetric Error Correction Model (AECM), the Nonlinear Autoregressive Distributed Lag model (NARDL), the Threshold Autoregressive Error Correction Model (TAR‑ECM), and the Markov‑Switching Error Correction Model (MS‑ECM). To ensure comparability, the same pre-testing protocol is applied across models. Unit root and cointegration tests are used to verify integration orders and long-run equilibrium relationships; because applying the full period as a single estimation window does not yield stable cointegration for some market–product combinations, estimation is conducted on two subsamples where unit-root and cointegration conditions are jointly satisfied: 2006.06–2017.08 and 2017.09–2025.09. In addition, Bai–Perron structural break tests are implemented as a supplementary diagnostic for potential structural instability, which motivates the inclusion of MS‑ECM that endogenously estimates regime changes.
Model performance is evaluated using four criteria: (i) goodness-of-fit (AIC), (ii) long-run and short-run asymmetry tests using Wald F or LR χ² statistics and p-values, (iii) sensitivity to data frequency (monthly vs. weekly), quantified by regression-based comparisons of changes in AIC and test statistics, and (iv) robustness under potential structural instability, assessed via Bai–Perron evidence and MS‑ECM regime-switching estimates (transition probabilities, expected duration, and regime-specific adjustment coefficients).
Empirically, mean AIC differences across the four models are not statistically significant, suggesting that fit alone does not determine model choice. In contrast, asymmetry detection differs sharply by model. For long-run asymmetry tests, MS‑ECM produces significantly larger test statistics than AECM (model-dummy regression coefficient 49.25, p<0.01). For short-run asymmetry, NARDL and MS‑ECM exceed AECM (coefficients 1.415 and 18.0563, respectively, p<0.01). At the 10% significance level, the number of rejections of the symmetry null is 49 for MS‑ECM, 13 for NARDL, 8 for AECM, and 4 for TAR‑ECM. Weekly data tend to improve AIC for all models, but frequency sensitivity in asymmetry statistics remains model-dependent; in particular, MS‑ECM shows a significant rise in the long-run χ² statistic under weekly data (week coefficient 72.32, p<0.01). Importantly, statistically significant asymmetry is not uniformly observed across all markets and products; instead, it concentrates in specific market × product × period combinations, consistent with partial asymmetry.
Overall, asymmetric crude–refined product pass-through cannot be concluded reliably using a single model or a single metric such as AIC. This study therefore recommends a parallel comparison protocol: use AECM as a benchmark, use NARDL to separate short-run responses to positive vs. negative shocks, and use MS‑ECM to identify regime-dependent adjustment and potential structural instability. A limitation is that the analysis is restricted to the spot (wholesale) stage, and does not directly incorporate retail pass-through or tax/distribution components. Future research should integrate inventories, refining margins, utilization rates, exchange rates, and demand indicators in a multivariate framework and link wholesale and retail price data to directly identify the drivers of asymmetric adjustment.