Why gas prices
feather down.

01 / Pump price decomposition

Separating sovereign fuel taxes from oligopoly ex-tax retail markups across 12 countries.

Raw international pump comparisons confuse fiscal taxation with corporate pricing power. In Denmark ($2.35/L) and the Netherlands ($2.26/L), excise, carbon, and VAT taxes account for over 60% of the retail price. However, once we strip out taxes and wholesale Terminal Gate Prices (TGP) using Pandas, a stark industrial-organization pattern emerges: countries where 3 to 5 vertically integrated refiner-marketers or supermarket chains control 65%–89% of stations (Germany, the UK, Australia, Chile, and Canada) sustain ex-tax retail markups of 24 to 28 cents/L — more than double the 12 cents/L spread in unconcentrated US markets anchored by high-volume maverick discounters (Costco, QuikTrip, Wawa).

■ Wholesale Crude & Refining TGP ($/L)■ Ex-Tax Retail & Oligopoly Markup Spread ($/L)■ Sovereign Tax Wedge (Excise + Carbon + VAT)

02 / Rockets, feathers & Edgeworth cycles

How tacit focal points and AI pricing replace price wars with asymmetric pass-through.

In an oligopoly without an explicit cartel ring, firms coordinate dynamically through two signature price trajectories:

• Asymmetric Pass-Through ("Rockets & Feathers"): When wholesale crude/TGP jumps by +10 c/L, stations pass through 93% of the hike within 3 days so no firm is left selling below replacement cost. When wholesale cost drops by −10 c/L, stations hold their posted street signs flat and let retail prices feather down over 21–28 days — because any unilateral price cut risks being misread by rivals as a defection, triggering a price war.
• From Edgeworth Sawtooth Wars to AI Sticky Margins: In competitive markets, stations undercut by 0.2–0.5 c/L daily until margins hit zero, followed by a coordinated 10–12 c/L relenting jump led by a major brand (purple dashed curve). Byrne & de Roos (2019) proved that BP used focal-point signaling in Perth to transition stations to a flat collusive margin (+75% profit jump, green curve), while Assad et al. (2024) showed that AI pricing algorithms in Germany raised duopoly margins by +28% by punishing undercutting in seconds.

03 / Game theory: why gas colludes and other industries don't

Repeated Bertrand supergames: comparing effective δ = exp(−rT) against critical threshold δ* = 1 − 1/N.

In a one-shot Bertrand game with homogeneous 87-octane fuel, price collapses to marginal cost \((P = MC)\). In an infinitely repeated supergame, tacit collusion is sustainable as a Subgame Perfect Nash Equilibrium (SPNE) if and only if the effective discount factor \(\delta = e^{-rT}\) exceeds the critical defection threshold \(\delta^* = \frac{\pi^{\text{deviate}} - \pi^{\text{collude}}}{\pi^{\text{deviate}} - \pi^{\text{punish}}} \approx 1 - \frac{1}{N}\). Click any industry below to inspect why retail fuel, cement, and 3-carrier telecom sit deep inside the collusive zone while B2B SaaS, custom manufacturing, and aircraft procurement cannot sustain tacit collusion:

"""Reproducible Pandas analysis for ST-06: Gas Price Collusion & Oligopoly Supergames."""
import numpy as np
import pandas as pd

# 1. Load the cross-country pump price decomposition & concentration table
countries = pd.read_csv("studies-gas-countries.csv")
countries["ex_tax_price_usd_l"] = countries["wholesaleUsdL"] + countries["markupUsdL"]
countries["tax_share_pct"] = (countries["taxUsdL"] / countries["pumpUsdL"] * 100).round(1)
countries["lerner_ex_tax_pct"] = (countries["markupUsdL"] / countries["ex_tax_price_usd_l"] * 100).round(1)
countries["excess_markup_vs_us_cpl"] = ((countries["markupUsdL"] - 0.12) * 100).round(1)

# Exclude state-capped producer monopoly (DZ) when estimating market oligopoly HHI gradient
market_oecd = countries[countries["code"] != "DZ"].copy()
corr_cr4 = market_oecd["cr4Pct"].corr(market_oecd["markupUsdL"])
corr_hhi = market_oecd["hhi"].corr(market_oecd["markupUsdL"])
slope_cpl_per_10pct_cr4 = np.polyfit(market_oecd["cr4Pct"], market_oecd["markupUsdL"] * 100, 1)[0] * 10

print("=== Cross-Country Oligopoly Markup Summary (Pandas) ===")
print(market_oecd[["code", "pumpUsdL", "wholesaleUsdL", "taxUsdL", "markupUsdL", "cr4Pct", "hhi", "lerner_ex_tax_pct"]])
print(f"Correlation(CR4, Ex-Tax Markup): r = {corr_cr4:.3f} | Correlation(HHI, Markup): r = {corr_hhi:.3f}")
print(f"OLS Gradient: +{slope_cpl_per_10pct_cr4:.2f} cents/L markup per +10 percentage points in CR4 concentration\n")

# 2. Asymmetric Pass-Through ("Rockets & Feathers" Error Correction Model)
pt = pd.read_csv("studies-gas-pass-through.csv")
rocket_half_life = pt.loc[pt["rocketPassPct"] >= 50, "day"].min()
feather_half_life = pt.loc[pt["featherPassPct"] >= 50, "day"].min()
cumulative_wedge_cpl_days = np.trapezoid(pt["rocketPassPct"] - pt["featherPassPct"], pt["day"]) / 10.0
print(f"Rocket 50% pass-through lag: {rocket_half_life} day(s) vs. Feather 50% decay lag: {feather_half_life} days")
print(f"Asymmetric tacit rent per 10 c/L wholesale cycle: +{cumulative_wedge_cpl_days:.1f} cent-days/L\n")

# 3. Repeated Bertrand Supergame Incentive-Compatibility (delta >= delta* = 1 - 1/N)
ind = pd.read_csv("studies-gas-industries.csv")
ind["collusive_spne"] = ind["deltaEff"] >= ind["deltaStar"]
print("=== Supergame Incentive-Compatibility (delta_eff vs. delta_star) ===")
print(ind[["id", "industry", "firmsN", "deltaStar", "deltaEff", "slack", "collusive_spne"]])