The impacts of carbon pricing on the electricity market in Japan

The Paris Agreement targets limiting global temperature increases, motivating countries to curb greenhouse gas emissions, of which CO2 is the dominant component largely arising from fossil fuel combustion. Electricity generation is a leading source of emissions, making decarbonization of the power sector pivotal for Japan’s 2050 carbon neutrality goal. Japan’s prior initiatives include the voluntary JVETS (2005) and a national carbon tax introduced in 2002, with the tax rate reaching JPY 289/tCO2 and no increases since April 2016. Japan’s power sector underwent reforms with the establishment of JEPX (2005) and retail market liberalization (2016). This study investigates how carbon pricing affects wholesale electricity prices regionally in Japan by estimating the carbon cost pass-through rate (CPTR) across nine regions, addressing: (1) the correlation between electricity prices and carbon costs, and (2) the extent to which carbon costs are passed through to electricity prices.

Carbon pricing internalizes pollution externalities (Pigou, 1920) by raising the marginal cost of fossil-based generation and incentivizing a shift toward low-carbon sources. Prior studies find electricity prices may track CO2 prices (Bauer & Zink, 2005) or reflect both fuel and carbon cost variations via linear models (Sijm & Chen, 2006; Chen, 2008). Electricity market liberalization is often a prerequisite for effective carbon pricing (Fan, 2014). Under competitive market theory, price equals marginal cost, implying carbon costs should be passed through proportionally; however, empirical CPTRs deviate from 100%. For instance, Australian regions often show CPTR > 1 (Nazifi, 2016; Nazifi et al., 2021; Simshauser, 2007), and near-full pass-through has been documented in Spain under inelastic demand and frequent auctions (Fabra, 2014). Electricity prices also display nonlinear relationships with demand (Barlow, 2002; Kanamura, 2007), motivating flexible functional forms when estimating CPTR.

Design and scope: The study estimates regional carbon cost pass-through rates (CPTR) to wholesale electricity prices for Japan’s nine regions (Hokkaido, Tohoku, Tokyo, Hokuriku, Chubu, Kansai, Shikoku, Chugoku, Kyushu) using hourly data primarily spanning April 2016–August 2021, a period with a constant carbon tax rate. Key indicator: CPTR is defined as the ratio of proportional changes in electricity price to proportional changes in carbon cost (ρ = (dP/P) / (dC/C)). A higher CPTR indicates a greater share of carbon costs borne by consumers via higher prices; lower CPTR implies generators absorb more costs, potentially incentivizing efficiency and low-carbon investments. Data: Hourly day-ahead spot electricity prices and demand are from JEPX. Fuel costs, comprising natural gas, light oil, kerosene, and coal, are converted to JPY/kWh using fuel prices (domestic sources for oil products; imported coal prices reflecting ~99% import reliance) and plant heat rates (MMBtu/kWh). The dependent variable is the fuel spread (spark/dark spread) defined as electricity price minus marginal fuel cost (JPY/kWh). Carbon cost is non-observable and proxied as the product of the fixed carbon tax rate (JPY 289/tCO2) and hourly regional carbon intensity (kg-CO2/kWh) derived from thermal generation emissions over output. All prices and costs are expressed in JPY/kWh. Assumptions include: (i) fuel costs are fully and directly passed to electricity prices; (ii) other generation costs (labor, O&M) are constant and negligible relative to fuel; (iii) technology and market structure are approximately constant in the short run; (iv) electricity markets are regionally segmented with limited interregional price effects. Models: 1) Polynomial OLS regression (degree two) with k-fold cross-validation to justify quadratic demand terms: Fuel spread = α0 + α1*C + β1*D + β2*D^2 + ε, where C is carbon cost and D is hourly demand. The model captures nonlinear demand-price relationships observed in scatterplots. The study also discusses log transformations linking α1 to elasticities (d ln P / d ln C), though main reported coefficients are in level regressions on fuel spread. 2) Linear mixed model pooling regions: Fuel spread = α0 + α1*C + α2*D + ε, to estimate an average national CPTR across regions. 3) Generalized Additive Model (GAM): Fuel spread = α0 + α1*C + f(D) + ε, where f(D) is a smooth function capturing flexible nonlinearities. Effective degrees of freedom (edf) indicate smoothness complexity. Descriptive analysis: The study reports summary statistics for prices and carbon intensity by region and documents the January 2021 cold-wave price spike, validating strong demand-related nonlinearity. Estimation and inference: Coefficients are reported with standard errors and significance levels; model fits (R^2) are provided by region and specification.

- Regional CPTRs (Polynomial OLS, α1 coefficients; all significant at 1%): • Kansai: 1.937 (R^2=0.365) • Chubu: 2.157 (R^2=0.318) • Chugoku: 2.995 (R^2=0.291) • Hokkaido: -0.055 (R^2=0.066) • Hokuriku: 0.446 (R^2=0.280) • Kyushu: 0.531 (R^2=0.293) • Shikoku: 0.226 (R^2=0.312) • Tohoku: 0.585 (R^2=0.252) • Tokyo: 0.698 (R^2=0.321) - Interpretation: Kansai, Chubu, and Chugoku exhibit CPTR > 1, implying consumers bear more than the full carbon cost and generators capture windfall margins as carbon costs rise. Hokuriku, Kyushu, Shikoku, Tohoku, and Tokyo show partial pass-through (CPTR < 1), suggesting generators absorb a larger share of carbon costs. Hokkaido displays a small negative CPTR, indicating a negative association between carbon cost and prices. - Robustness (GAM results; all α1 significant at 1%): CPTRs broadly align with polynomial OLS: Kansai 1.962; Chubu 2.153; Chugoku 3.004; Hokkaido -0.051; Hokuriku 0.458; Shikoku 0.225; Tokyo 0.728. Notable differences: Kyushu 0.077 (vs. 0.531 in OLS) and Tohoku 0.360 (vs. 0.585), suggesting stronger nonlinear demand effects in those regions (Kyushu GAM R^2=0.788). - National average (linear mixed model): Average CPTR = 0.596 (p=0.016), indicating that, on average, a 1-unit increase in carbon cost raises electricity prices by about 0.6 units; R^2=0.026; N=258,290. - Descriptive context: Electricity prices spiked above 140 JPY/kWh in January 2021 during a cold wave, then normalized within a month. Average prices (JPY/kWh) over 2016–2020: Hokkaido 5.140 (highest); Tokyo 3.144; Tohoku 3.095; others below ~2 (e.g., Kansai 1.996; Chubu 1.967; Hokuriku 1.964; Shikoku 1.961; Kyushu 1.596; Chugoku 0.737). Carbon intensity means (kg-CO2/kWh): Hokkaido 0.43 (largest, high variability); others range ~0.13–0.17.

The study’s primary question—how carbon pricing affects wholesale electricity prices—finds that pass-through is heterogeneous across Japanese regions. Regions with CPTR > 1 (Kansai, Chubu, Chugoku) indicate that consumers more than fully absorb carbon costs, potentially dulling generators’ incentives to decarbonize under the current tax level. Regions with CPTR < 1 (Hokuriku, Kyushu, Shikoku, Tohoku, Tokyo) suggest generators bear a substantial share of carbon costs, which could encourage efficiency gains and investment in low-carbon generation to protect margins. The negative CPTR in Hokkaido is plausibly linked to the region’s wind resource and the feed-in tariff (FIT) regime, which may offset fuel cost pressures and lead to negative fuel spreads in the data. The national average CPTR near 0.6 implies carbon costs are not fully internalized into wholesale prices, consistent with a relatively low and flat carbon tax rate and an electricity market still transitioning toward full liberalization. Limited competition and incumbent dominance at JEPX may weaken the link between marginal cost (including carbon) and market-clearing prices. Regional policy differences (e.g., Tokyo’s mandatory TMG ETS vs. national voluntary schemes) also create uneven carbon cost exposures, risking carbon leakage as firms shift activity toward less regulated regions. Overall, the findings underscore that both stronger, more uniform carbon pricing and deeper electricity market liberalization are needed to align price signals with decarbonization objectives.

Using hourly data from 2016–2021 and multiple modeling approaches (quadratic OLS, linear mixed model, and GAM), the paper estimates regional carbon cost pass-through rates in Japan’s wholesale electricity markets. It documents substantial heterogeneity: CPTRs above one in Kansai, Chubu, and Chugoku; partial pass-through in Hokuriku, Kyushu, Shikoku, Tohoku, and Tokyo; and a small negative CPTR in Hokkaido. A pooled estimate suggests that, on average, about 60% of carbon costs are passed through nationally. These results indicate incomplete internalization of carbon costs under Japan’s current, relatively low carbon tax and an only partially liberalized electricity market. Policy recommendations include raising the national carbon tax rate and continuing electricity market liberalization to enhance cost-reflective pricing while addressing regional disparities to avoid carbon leakage. Future research should incorporate renewable efficiency metrics and evolving market structures as renewable penetration and policy settings change.

Key limitations include: (1) Carbon cost is proxied as carbon tax rate times carbon intensity, with a constant national tax rate over time; other carbon pricing instruments (e.g., TMG ETS) are not fully integrated into the cost metric. (2) The model assumes full and direct pass-through of fuel costs and treats other generation costs as constant, which may not hold universally. (3) Technology and market structure are assumed constant in the short run; evolving generation mixes and strategic bidding could affect results. (4) Electricity markets are treated as regionally segmented with limited interregional effects; actual transmission constraints and interties could create spillovers. (5) Retail price effects and end-user demand elasticities are not modeled; the analysis focuses on wholesale prices. (6) Renewable efficiency is excluded despite potentially material impacts in some regions (e.g., Hokkaido). (7) Model fit varies by region and specification (low R^2 in some OLS cases), and GAM/OLS discrepancies (e.g., Kyushu) indicate sensitivity to nonlinear demand dynamics.