Competitive effects of implicit auction on interconnectors: evidence from Japan

Abstract

Interconnectors play a crucial role in electric power systems. They contribute to balancing demand and supply in real-time, guaranteeing efficient dispatch in wide geographic regions, and increasing competition by creating large markets. However, interconnector capacity is a scarce resource because vertically integrated utilities were required to have generating capacity enough to supply most customers within their operating region under a regulated monopoly. Hence, identifying the efficient allocation method is essential, particularly after recent electricity market restructuring. This study evaluates the competitive effect of the implicit auction on the interconnector transmission capacity. The implicit auction allocates all the interconnector capacity simultaneously with electric energy in the day-ahead market. This method prevents market participants from strategically withholding the physical interconnector capacity ex ante to exercise market power, as allowed under the first come, first served rule. This study empirically shows how the capacity was withheld from the day-ahead market under the first come, first served rule using detailed reservation data. Next, I show that the implicit auction increases interconnector capacity available at the day-ahead market and trade volume. I use machine-learning methods, such as random forest and deep neural networks, to predict the counterfactual market outcomes without implicit auction. I find that the gain from trade under the implicit auction is more than US$55 million per year in Japan, which is more than 100 times the implementation cost of the implicit auction.

1 Introduction

Interconnectors play a crucial role in electric power systems. They contribute to balancing demand and supply in real-time, guaranteeing efficient dispatch in wide geographic regions, and integrating intermittent renewable energy. Interconnector also increases competition in the generation and retail sectors by creating large markets. However, interconnector capacity is a scarce resource because vertically integrated utilities were required to have generation capacity enough to supply most customers within their operating region under a regulated monopoly in many countries. Hence, identifying the efficient allocation method is essential, particularly after recent electricity market restructuring. Interconnector capacity is generally allocated through market- or nonmarket-based congestion management methods (European Transmission System Operators, 2004). What is the economic advantage of market-based congestion management methods compared with nonmarket-based ones?

This study empirically evaluates the competitive effect of the implicit auction relative to first come, first served (FCFS) allocations. The implicit auction allocates interconnector capacity simultaneously with electric energy in the day-ahead market, whereas the FCFS rule allows incumbents to reserve the capacity before the energy market. The FCFS rule effectively allocates interconnector capacity as a free physical transmission right (PTR), a right to physically schedule flow on an interconnector across bidding zones. In theory, this rule provides rights holders with a tool to exercise market power by creating the appearance of congestion where none exists (Bushnell, 1999). They can withhold PTR from the subsequent day-ahead market to create market splitting and increase local market prices within the import constraint zone by overestimating their capacity needs. The implicit auction prohibits interconnector capacity from being reserved in advance, and all capacity is released to the day-ahead market, allowing additional cross-zonal trade in the energy market and narrowing the price gap between the import and export zones. The implicit auction also affects the liquidity of the day-ahead market because firms that previously reserved interconnector capacity for bilateral contracts are required to participate in the market.

This study focuses on two of Japan’s most congested interconnectors: the Kitahon interconnector connecting the Hokkaido and East zones and the Kanmon interconnector connecting the West and Kyusyu zones. The empirical challenge is that market outcomes are unobservable if the FCFS rule is not abolished. I use several machine-learning methods, such as random forest (RF) and deep neural networks (DNNs), to predict the counterfactual market outcomes in the post-treatment period. I evaluate the predictive performance and choose the best model to predict the counterfactuals by comparing the loss functions. Moreover, I estimate the gain from trade under implicit auctions by comparing actual and counterfactual market outcomes.

I find that the implicit auction increases the efficient allocation of interconnector capacity and prevents the exercise of market power. Furthermore, I find that the implicit auction has a significant economic impact on welfare. The capacity release effect for the Kitahon and Kanmon interconnectors respectively reduces generation costs by approximately $10 million and $17 million over 6 months. The total annual cost saving of the implicit auction is estimated at more than $55 million/year. This gain is more than 100 times the one-time implementation cost of the implicit auction.

My contributions are threefold. First, this study presents the first empirical evidence of the impact of an implicit auction compared with the nonmarket-based FCFS rule. Numerous studies have examined the inefficiency of an explicit auction, another market-based method used in Europe, compared with implicit auction (e.g., Brunekreeft et al., 2005; Bunn & Zachmann, 2010; Creti et al., 2010; Ehrenmann & Neuhoff, 2009; Füss et al., 2020; Gugler et al., 2018; Keppler et al., 2016; Newbery et al., 2016). These studies have shown that implicit auction can avoid the economically inefficient use of interconnector capacity (i.e., export energy from a high- to a low-price zone) attributable to ex-ante interconnector capacity trade under the explicit auction. The present study reveals that implicit auction can reduce the underutilization of interconnector transmission capacity caused by the cancelation of reserved interconnector capacity under the FCFS rule. Furthermore, I show that the implicit auction has not only a price-converging effect by releasing the interconnector capacity previously reserved under the FCFS rule (capacity release effect) but also a price-diverging effect by forcing those who reserved the capacity to participate in the market (volume effect).

Second, I show empirical evidence that the FCFS rule allowed PTR holders to exercise market power, as predicted by the theoretical literature (e.g., Borenstein et al., 2000; Bushnell, 1999; Joskow & Tirole, 2000). I also show how an implicit auction could prevent PTR holders from exercising market power by simultaneously allocating the PTR in the day-ahead market. Previous studies have examined market power through energy withholding in the electricity market given transmission capacity. On the contrary, this study examines market power, explicitly allowing market participants to withhold interconnector capacity through ex-ante reservation. Detailed hourly interconnection capacity reservation data allow me to observe how much capacity is withheld from the day-ahead market under the FCFS rule.

Third, I can avoid the potential omitted variable bias that can arise when studying two European countries (e.g., Germany and France), which are interconnected by other neighboring countries and for which relevant data are often unavailable. By contrast, the Japanese power system is not connected to other countries, and all zonal data are available. An additional advantage is that the Japanese network is effectively radial, and the analysis does not suffer from loop-flow issues.

The remainder of this paper is further organized as follows. Section 2 briefly describes the Japanese electricity sector, and Sect. 3 explains Japan’s interconnector congestion management methods. Section 4 defines the two effects of an implicit auction: interconnector capacity release and volume effects. Section 5 presents the empirical strategy, and Sect. 6 explains the data. Then, Sect. 7 presents the estimates of the overall impact of the implicit auction, capacity release effect, and volume effect. Section 8 conducts the welfare analysis through the counterfactual prediction, and finally, Sect. 9 concludes the study.

2 Japanese electricity sector

The Japanese electricity sector is characterized by a vertical supply structure comprising generation power plants, wholesale markets, transmission and distribution grids, and retail markets. Historically, vertically integrated private utilities provided generation, transmission and distribution, and retail services under cost-of-service regulation. These utilities are responsible for ensuring the security of supply for each operating geographical region. In addition, they built the generation capacity, transmission network, and distribution network facilities necessary for this purpose. Interconnector capacity was infrequently used except for unplanned outages of power plants. Three major regulatory reforms were introduced following the Fukushima nuclear accident after the great earthquake and tsunami in 2011. First, the government fully opened the retail market to all households in April 2016. Second, implicit auction replaced the FCFS rule in October 2018 to create a level playing field for generation and retail competition. Third, the government required the vertically integrated utilities to unbundle their transmission and distribution networks legally by April 2020.

Although the wholesale market has gradually opened up to competition since the 1990s, the generation capacity remained highly concentrated. The Electricity and Gas Market Surveillance Commission (2018a) noted that 10 former vertically integrated utilities and two incumbent independent power producers generate 83% of national electricity. Therefore, incumbent companies in each region are still dominant players in the generation market after the reform. Their generation capacity primarily consists of large-scale thermal (fueled by coal, liquefied natural gas [LNG], and oil), nuclear, and hydropower plants. All nuclear power plants ceased operation after the Fukushima nuclear accident for safety reasons. Hence, they increased electricity production from thermal power plants, particularly from imported LNG power plants. Conversely, new entrants mainly invested in renewable energy under the feed-in tariff scheme introduced in 2011, particularly in solar photovoltaic generation.

In the wholesale stage, most of the electricity generated by vertically integrated incumbents is sold internally to their retail affiliates without using an organized exchange market. In 2014, 77.7% of their electricity supply was procured from their generation power plants (Ministry of Economy, Trade and Industry, 2016). They also procure electricity through bilateral contracts with independent power producers for approximately 20% of the retail customer’s demand (Electricity and Gas Market Surveillance Commission, 2016). Only 1% of their electricity demand was procured from the centralized exchange market operated by the Japan Electric Power Exchange (Ministry of Economy, Trade and Industry, 2016). As Bushnell et al. (2008) reveal in the US electricity markets, the vertical integration between generation and retail segments suggests a weakened incentive to exploit market power in the day-ahead market.¹

The government introduced several regulatory measures to increase the liquidity of the exchange market. First, the government has encouraged vertically integrated utilities to sell all “surplus electricity” based on marginal costs to wholesale markets since 2013. Surplus electricity is defined as the maximum electricity to be generated minus the retail demand, the sales demand for bilateral contracts, and the reserved generating capacity for balancing power. This initiative is called “net bidding.” The government also required vertically integrated utilities to submit purchase bids to the market, reflecting the marginal costs of the generators they used to supply their customers. High buy bid prices were only allowed when their energy supply was at risk of shortages. Second, the government introduced “gross bidding” in April 2017 to further force the incumbent vertically integrated firms to increase the quantity of sell offers with marginal costs into the market (Kanamura & Bunn, 2022). The initial target was at least 10% of their electricity sales volume. The incumbent vertically integrated firms later committed to increasing the volume of the gross bidding to 20–30% of their electricity sales volume by March 2019. Notably, these measures are implemented to increase the liquidity of the day-ahead market, but they also work to prevent them from exercising market power (Electricity and Gas Market Surveillance Commission, 2021a).

Third, electricity generated from renewable energy under the feed-in tariff has been sold at the day-ahead market since 2017. This event increased the sell offer quantity in the day-ahead market. Finally, the government introduced the implicit auction, which obliged those who used to reserve interconnector capacity for their bilateral contract to participate in the day-ahead market, as explained in Sect. 3.

After these changes, the amount of electricity traded in the day-ahead market gradually increased (Fig. 1). In April 2017, the day-ahead market contracted volume accounted for less than 5% of total final electricity consumption. However, the share of contracted volume reached 30% after October 2018. The retail market share by new entrants has gradually increased with the improved liquidity in the wholesale market. Before 2016, all incumbent vertically integrated utilities had more than 90% retail market share in their geographic region (Electricity and Gas Market Surveillance Commission, 2016). However, based on the volume of electricity sold at the retail stage in March 2018, the market share of new entrants reached approximately 12.7% (Electricity and Gas Market Surveillance Commission, 2018b). Advances in retail competition have also increased the surplus electricity sold by vertically integrated retailers at the exchange and increased the sell offer volume.

Fig. 1

Source: Japan Electric Power Exchange and Transmission System Operators

Volume of sell offer and buy bid in the day-ahead market and share of contracted volume in total electricity consumption. The blue line is the daily moving average of the sell offer volume, whereas the red line is the daily moving average of the buy bid volume in the day-ahead market (left y-axis). The black line is the monthly average share of contracted volume in total final electricity consumptions (right y-axis).

As depicted in Fig. 2, the Japanese day-ahead market covers nine fixed regions connected by interconnectors. The tenth region, Okinawa, is an unconnected island and is not shown in the figure. Each region corresponds to the control region operated by the former vertically integrated regional monopoly. I classify these nine regions into four zones based on the frequency of congestion, that is, Hokkaido, East, West, and Kyusyu.

Fig. 2

Day-ahead market price zones in Japan

Hokkaido is located in the northernmost part of Japan. During the study period, Hokkaido had the highest day-ahead market price partly because oil-fueled thermal power plants are used to compensate for lost electricity from a nuclear power plant that ceased operations after the Fukushima nuclear accident. The East zone includes Tokyo and Tohoku. A high-voltage direct current Kitahon interconnector connects the Hokkaido and East zones. The operating capacity of the Kitahon was set to 600 MW in both directions in case the 600 MW generating unit at the Tomatou-atsuma coal power plant in Hokkaido was shut down. A local vertically integrated utility, Hokkaido Power Electric, was considered a pivotal supplier as it owns more than 90% of the generation capacity in the Hokkaido region (Electricity and Gas Market Surveillance Commission, 2022). The East zone includes the Tokyo region, which has the highest demand center. I define the West zone as the combination of Chubu, Kansai, Hokuriku, Shikoku, and Chugoku regions and regard it as a single zone, as congestion is limited and the market price is almost the same within the zone. The East and West zones are connected by high-voltage direct current frequency converters with 1200 MW of operating capacity in both directions.² Kyusyu is located in the southwestern part of Japan. A high-voltage alternate current Kanmon interconnector connects the West and Kyusyu zones. Kanmon has an operating capacity of 2800 MW from the Kyusyu zone to the West zone and 500 MW from the West zone to the Kyusyu zone. The large capacity reflects that J-Power, a former state-owned power producer established by the Japanese government in 1952, developed the Kanmon interconnector together with the development of the Matsushima coal power plant in Kyusyu to export some of its electric energy to the West zone.

J-Power developed and owned all three interconnectors. J-Power has played a unique role in the history of Japan’s power industry in complementing private vertically integrated utilities. It developed not only hydro and coal power plants but also most interconnectors, including Kitahon and Kanmon. J-Power supplied electricity to vertically integrated utilities through bilateral contracts, sometimes through its interconnectors, but did not engage in retail business.

3 FCFS rule and implicit auction

In Japan, the day-ahead market uses a blind single-price auction and a market splitting method. Market participants submit a pair of price-quantity functions, representing the minimum willingness to sell for a sell offer or the maximum willingness to pay for a buy bid per half hour on the day before the delivery, specifying the region. Block bid is also allowed to ensure the continuous operation of inflexible power plants, such as coal or nuclear power plants. It submits sell offer price-quantity pairs with a requirement for simultaneous sale over a span of more than 2 h. At 10:00 a.m. every day, the Japan Electric Power Exchange aggregates the bidding information, constructs the sell offer curve and the buy bid curve, and determines the single market price and the contracted volume at the intersection of the two curves. When the initial auction allocates the scheduled flow above available interconnector capacity (this is “congestion” in the power system), the Japan Electric Power Exchange divides the national market into subzonal markets. In addition, auctions are repeated within the subzonal markets.³ Moreover, the market splitting algorithm allocates all net available interconnector capacity (NAIC) to market participants who successfully contract in the auction (i.e., who submit the sell offer price below the zonal price and submit the buy bid price above the zonal price) in the presence of congestion. This is how market splitting manages congestion on interconnectors in the day-ahead market (Marmiroli et al., 2009).

Before October 2018, the FCFS rule allowed incumbents to reserve their capacity for free before the start of the day-ahead auction at 10:00 a.m. Under the FCFS rule, the NAIC in the day-ahead market is calculated as follows:

$$ {\text{NAIC}} = {\text{Operating}}\;\;{\text{capacity}} - {\text{Margin}} - {\text{Reserve}}\;\;{\text{capacity}}{.} $$

(1)

The operating capacity is the maximum amount of electric energy that can flow through an interconnector. The limit is determined as the minimum value that satisfies all four constraints: thermal capacity, synchronization stability, voltage stability, and frequency. The margin is the reserve capacity to maintain system security in the event of an emergency. The Organization for Cross-regional Coordination of Transmission Operators, an association of regulators and transmission system operators, calculates the operating capacity and margin annually. The operating capacity is stable except during planned maintenance or outage.⁴ The margin varies between 400 and 520 MW for Kitahon and zero for Kanmon. Note that, an implicit auction was already used in the market under the FCFS rule, but it was applied only to the remaining NAIC. FCFS-based reservation is allowed for up to 10 years and can be renewed exclusively.

Reserved interconnector capacity under the FCFS rule was a non-tradable PTR (Bushnell, 1999; Gilbert et al., 2004; Joskow & Tirole, 2000). The FCFS rule gives PTR holders an instrument not only to avoid the risk of market price change due to market splitting but also to exercise market power (Twomey et al., 2005). During market splitting, the market clearing algorithm accepts a sell offer with a high price, referring to the right side of the sell offer curve in the import zone, and the sell offer price determines the zonal price in the import zone. This creates an opportunity for PTR holders to increase market prices.

Figure 3 depicts a simplified numerical example of how PTR holders can increase their market revenue in market splitting. Players A–D in zone A and players E and F in zone B submit a sell offer or buy bid function to the Japan Electric Power Exchange. The sell offer price colored in red determines the market prices. The left figure shows that an interconnector has no congestion and that the day-ahead market price equates between two zones. A total of 40 kWh of scheduled flow is allocated on the interconnector, with a market price of 7 yen/kWh in both zones. Notably, the sell offer price of player B (7 yen/kWh) determines the market price, whereas player E in zone B, who sends a sell offer at 11 yen/kWh, cannot sell at all. The right figure illustrates the scenario when the available interconnector capacity is limited to 10 kWh. The market prices diverge between zones A and B due to market splitting. The sell offer price of player A determines the market price of 5 yen/kWh in zone A, whereas the sell offer price by player E determines the market price of 11 yen/kWh in zone B, selling 30 kWh of energy. If player E in zone B can reserve interconnector capacity to import energy from zone A before the day-ahead market auction, the likelihood of market splitting and thus the market price in zone B will increase.

Fig. 3

Example of market splitting in the presence of congestion. Source: Organization for Cross-regional Coordination of Transmission Operators (2016)

This example demonstrates that the resulting market price depends on the available interconnector capacity in the day-ahead market, even though all market participants do not alter their sell offer or buy bid functions. As the market split increases local market price in the import zone, creating congestion could be profitable for PTR holders in the import zone. The incentive to exercise market power would increase as the amount of inframarginal generating capacity the player has (Davis & Wolfram, 2012). Thus, the FCFS rule may encourage the incumbent PTR holders to strategically withhold the interconnector capacity from the market through ex-ante reservations to increase the market revenue in an importing zone. By contrast, creating congestion is not profitable for PTR holders in export zones because it declines local market price and hence market revenue.

Figure 4 shows that most interconnectors were used for bilateral contracts under the FCFS rule. The 70–100 TWh/year of electric energy flowed interconnectors for bilateral contracts until 2017, corresponding to approximately 7–10% of the annual national power consumption. After the implicit auction was implemented in October 2018, the amount of trade across interconnectors through bilateral contracts dropped substantially to approximately 0.25 TWh/year. Meanwhile, the amount of electric energy flowing interconnectors allocated in the day-ahead market skyrocketed, which indicates that the trade volume for bilateral contracts now flows into the day-ahead market. Intraday and forward markets exist, but these trade volumes were quite low. Almost all electricity trade flowing interconnectors have occurred in the day-ahead market since 2019.

Fig. 4

Cross-zonal trade volume via interconnectors by trading methods (TWh/fiscal year). Source: Organization for Cross-regional Coordination of Transmission Operators (2017a, 2017b, 2018, 2019)

Most reserved capacities were withheld from the day-ahead market under the FCFS rule. Figure 5 shows the daily average reservation rates for the Kitahon interconnector (panel A) and Kanmon interconnector (panel B) from October 2016 to September 2018. The red circle shows the reservation rate seven days before delivery, and the black dot shows the reservation rate two days before delivery. The direction of the reserved flow is from East to Hokkaido in panel A and from Kyusyu to West in panel B. This case shows that both interconnectors are reserved in one direction. The reservation rate in both interconnectors is almost 100% seven and two days before delivery, which indicates that no available capacity exists for the day-ahead market.

Fig. 5

Source: Organization for Cross-regional Coordination of Transmission Operators

Daily average reservation rate under the FCFS. Red circles show the reservation rate 7 days before delivery, and black dots show the reservation rate 2 days before delivery.

Notably, the Kitahon interconnector has 600 MW of operating capacity, and the margin was set between 400 and 520 MW. Thus, the maximum reservable capacity was approximately 80–200 MW. Although some reserve capacity is released 2 days before delivery for Kitahon, approximately 80% of the capacity is already allocated before the day-ahead market. The data on which firms reserved interconnector capacity are not publicly available; however, anecdotal evidence shows that one of the incumbent retailers in Hokkaido reserved Kitahon’s capacity to import energy from the East zone. When the predicted demand in Hokkaido 2 days before delivery was lower than that of 7 days ago, the retailer may have canceled part of the reservation.

Based on the historical knowledge of how the Kanmon interconnector was developed, I can suppose that at least two vertically integrated regional utilities in the West zone reserved their capacity to import electricity from Kyusyu. J-Power has two coal power plants in Kyusyu for bilateral contracts with three vertically integrated utilities. Matsushima coal power plant has a rated capacity of 1 GW and supplied 370 MW of electricity to Kyusyu Power Electric, 468 MW to Chugoku Power Electric, and 70 MW to Shikoku Power Electric. Matsuura coal power plant has a rated capacity of 2 GW and supplied 756 MW of electricity to Kyusyu Power Electric, 754 MW to Chugoku Power Electric, and 400 MW to Shikoku Power Electric. Therefore, Chugoku Power Electric and Shikoku Power Electric in the West zone reserved 1690 MW of Kanmon interconnector to import the electricity from J-Power in the Kyusyu zone. The Kanmon interconnector has 2700 MW of operating capacity and no margin set. Thus, panel B indicates that other retailers may reserve the rest of 1 GW of the capacity in the West zone by May 2017, reaching 100% of the reservation rate. The differences between the reservation rates on the second and seventh days are small in Kanmon, arguably reflecting the large size of the reservable capacity. In sum, Fig. 5 shows that the most capacity of both interconnectors was reserved and not released to the day-ahead market under the FCFS rule.

Previous studies have discussed a “capacity release” mechanism to alleviate the market power concern when transmission capacity is traded as PTR (Bushnell, 1999; Joskow & Tirole, 2000). In Japan, however, reserved capacity is not allowed to trade under the FCFS rule, and no secondary market exists for interconnector capacity. Those who could reserve interconnector capacity can cancel the reservation. If they cancel before the day-ahead market, then it effectively becomes a capacity release. However, the cancelation was allowed even after the day-ahead market clearing, leaving unused capacity. The cancelation itself does not imply exercising market power, but rather indicates that they could initially reserve more than necessary to withhold the capacity from the market.

Those who canceled the reservation were subject to a penalty under the FCFS rule. This penalty was intended to deter excessive reservation. The penalty was set at 10 yen/MW (less than 25% of the marginal cost of coal) of canceled capacity when they canceled more than 10% of the initial reserve capacity (Organization for Cross-regional Coordination of Transmission Operators, 2017a). In other words, no penalty was imposed for cancelations within 10% of the initial reservation. The low penalty suggests that they could still strategically withhold unnecessary interconnector capacity from the market at a minimal cost.

Figure 6 highlights how much the reserved interconnector capacity was canceled after the day-ahead market. Panel A plots the relationship between the market price gap between Hokkaido and the East zones in the day-ahead market and the reservation rate of the Kitahon interconnector at 5:00 p.m. on the day before delivery. A negative reservation rate indicates that the interconnector capacity is reserved to transmit electricity from the East to the Hokkaido zone. The two graphs on the left in each panel show the 2-year observations from October 2016 to September 2018 under the FCFS rule. The two graphs on the right show the 2-year observations after the complete implicit auction was implemented from October 2018 to September 2020.

Fig. 6

Source: Japan Electric Power Exchange and Organization for Cross-regional Coordination of Transmission Operators

Market price gap and interconnector capacity reservation rate at 5:00 p.m. on the day before delivery. A negative reservation rate indicates that the capacity is reserved for transmitting electricity from the East zone to Hokkaido in panel A and from Kyusyu to the West zone in panel B.

Panel A shows that although the day-ahead market price in Hokkaido was higher than that in the East zone, the reservation rate of the Kitahon interconnector was greater than − 100% at 5:00 p.m. in many observations under the FCFS rule. The positive price gaps indicate that the market splitting occurred in the Kitahon interconnector during day-ahead market clearing, with all available capacity allocated to the market, reaching a − 100% reservation rate at 10:00 a.m. This is because Japan Electric Power Exchange uses a market splitting algorithm that allocates all NAIC to market participants who successfully contract in the day-ahead auction in the presence of congestion, as explained in Sect. 3. However, Kitahon’s capacity reservation rate at 5:00 p.m. frequently records values greater than − 100%, implying that those who reserved the Kitahon interconnector canceled their capacity reservations after 10:00 a.m. The information regarding canceled capacity is not available. However, it can be calculated when the price gap is non-zero by subtracting the reserved capacity at 5:00 p.m. from the reserved capacity at 10:00 a.m.

Panel B illustrates a similar situation in the Kanmon interconnector. The market price in the West was higher than that of Kyusyu for some hours. However, the Kanmon interconnector capacity had not been fully allocated by 5:00 p.m. under the FCFS rule. Hence, Fig. 6 confirms that the reserved interconnector capacity was frequently canceled after the day-ahead market splitting.

Under a complete implicit auction, incumbents could no longer reserve capacity, and all NAIC was allocated to the day-ahead market. In the two graphs on the right side in panel A, there are still positive price gaps, but the reservation rate at 5:00 p.m. is − 100% in most observations. It indicates that Kitahon's all available capacity is now fully allocated (i.e., reserved) in the day-ahead market. The allocation of Kanmon interconnector also improved after introducing the implicit auction. In summary, the FCFS rule allowed FTR holders to cancel reservations after the day-ahead market in both interconnectors. This is significant because those who reserve interconnector capacity could wait to release the capacity until after the market splitting.

Price gaps across congested interconnectors did not necessarily reduce after the implicit auction. Figure 7 shows the half year average market price gap. I divided the price gap into four groups. The first group is 0–100 yen/MWh (i.e., market coupling). The other three groups show the size of the price gap in the market splitting: 100–1000, 1000–10,000, and over 10,000 yen/MWh. The left figure shows the price gap between Hokkaido and the East zone between the Kitahon interconnector. Before October 2018, the price gap was large. Over 1000 yen/MWh of price gap was recorded over half of the period. Market coupling (i.e., the price gap was less than 100 yen/kWh) was observed in less than one-third of the period. Interestingly, the price gap has further widened since October 2018. The occurrence of market integration dropped by 3.8% after the implicit auction. The right figure shows the price gap between the West and Kyusyu zones between the Kanmon interconnector. Markets are coupled in 70–80% of the hours before and after the implicit auction, and almost no hours exist when the price gap records more than 10,000 yen/MWh. The difference in price gaps between the North and West zones may be caused not only by the different amounts of the interconnector capacity available at the market, but also by the extent of the interconnector capacity withholding under the FCFS rule.

Fig. 7

Source: Japan Electric Power Exchange

Market price gaps by half year (April 2016–March 2019). I categorized the price gap into four groups: 0–100 (market integration), 100–1,000, 1000–10,000, and over 10,000 yen/MWh. The left figure shows the price gap between the Hokkaido and East zones across the Kitahon interconnectors. The right figure shows the price gap between the West and Kyusyu zones across the Kanmon interconnectors.

4 Effects of implicit auction

I set up a hypothesis that a complete implicit auction (i.e., abolishing the FCFS rule) has two effects on the day-ahead market in Japan. First is the capacity release effect. The implicit auction releases the entire interconnector capacity previously reserved for bilateral contracts. This increases the NAIC and the trade quantity in the day-ahead market across zones, similar to an interconnector capacity upgrade investment. Consequently, this event will narrow the market price gap between the import and export zones. Figure 8 shows how an increase in the NAIC narrows the price gap between zones in the day-ahead market. ${P}_{Autarky}^{export}$ (${P}_{Autarky}^{import}$) is the price in the export (import) zone under autarky. A price gap exists between ${P}_{Autarky}^{export}$ and ${P}_{Autarky}^{import}$. As the trade quantity increases, the price in the export zone increases to ${P}_{Trade}^{export}$, and the price in zone B decreases to ${P}_{Trade}^{import}$, converging the gap.

Fig. 8

Capacity release effect

Second is the volume effect. The complete implicit auction prohibits reserving interconnector capacity for bilateral contracts across zones and may increase the volume of the day-ahead market. After implementing a complete implicit auction in October 2018, those who used to reserve the interconnector capacity for bilateral contracts were required to bid in the day-ahead market to obtain the interconnector capacity to trade across zones. Moreover, the regulator gave them temporary financial transmission rights to alleviate the unexpected risk of a price gap derived from market splitting. This transitional measure allowed retailers who could successfully bid a day-ahead market to hedge the price gap by March 2026. Thus, incumbents were strongly incentivized to submit bids in the day-ahead market. Consequently, the complete implicit auction increased the volume of sell offer and buy bid quantities in the day-ahead market. Figure 1 shows that both volumes indeed increased discontinuously after October 2018.

This increase in the volume may have affected the market price gap. If the additional sell offer price in the export zone is lower than the marginal cost of the marginal generator, then the additional sell offer lowers the resulting zonal price. The left panel of Fig. 9 depicts the former scenario. Otherwise, the prices in the export zone will not be affected. Similarly, the right panel of Fig. 9 depicts that if the additional buy bid price in the import zone is higher than the former equilibrium price, then the additional volume raises the market price in the import zone. Otherwise, the prices in the import zone will not be affected. Note that both the amount of additional bids and the bidding price determine whether the price gap is affected. In summary, the volume effect may widen the price gap between the import and export zones.

Fig. 9

Volume effect. The left figure shows the new market equilibrium when the additional sell offer price in the export zone is lower than the marginal cost of the marginal generator. The right figure shows the new market equilibrium when the additional buy bid price in the import zone is higher than the former equilibrium price

5 Empirical strategy

First, I model the day-ahead market price gap between the import and export zones as a function of the implicit auction dummy IA, supply controls, demand controls, and fixed effects to estimate the overall impact of the implicit auction:

$$ \Delta p_{, t} = \alpha_{0} + \beta_{1} IA + \beta_{2} Supply_{t} + \beta_{3} Demand_{t} + FE_{t} + \varepsilon_{t} . $$

(2)

$\Delta p_{, t}$ is the day-ahead market price difference between the import and export zones (Hokkaido–East zones and West–Kyusyu zones) measured by yen/MWh. ${\beta }_{1}$ is the parameter of interest to measure the effect of the implicit auction. Supply control variables include solar photovoltaic and wind generation in each zone. Nuclear power generation is supplied as a baseload and included as an exogenous variable. Daily coal prices, monthly LNG import prices, and daily Brent oil prices are used to control fossil fuel costs. The demand variables are the actual hourly electricity consumption in each zone, are proxies for forecasted demand, and are assumed to be exogenous in the short term.

FE_t includes hour, day of the week, month, and hour × month fixed effects. These fixed effects are intended to control for unobservable progress of the retail competition, gross bidding, growing renewable energy sold at the day-ahead market, and prediction errors in renewable energy generation and demand. Standard errors are clustered by day to address autocorrelation.

I assume that the effect of the implicit auction appears immediately. Data collection spans 6 months from the introduction of the implicit auction. The data collected at intervals of 1 week, 2 weeks, 1 month, 2 months, 3 months, and 6 months following October 2018 are utilized to assess the short-term impact of the implicit auction.

Next, I decompose the effect of the implicit auction into the capacity release and volume effects, holding other control variables constant. I then estimate the following equation:

$$ \Delta p_{, t} = \alpha_{0} + \beta_{1} NAIC_{t} + \beta_{2} Volume_{t} + \beta_{3} Supply_{t} + \beta_{4} Demand_{t} + FE_{t} + \varepsilon_{t} . $$

(3)

Equation (3) replaces the IA dummy variable in Eq. (2) with NAIC and volume variables to estimate the capacity release and volume effects separately. NAIC is the interconnector capacity available in the day-ahead market. The NAIC has two different values depending on the flow direction. I use the value from the export to the import zone (i.e., East to Hokkaido zones for Kitahon interconnector and Kyusyu to West zones for Kanmon interconnector).

Volume is the additional sell offer quantity in the export zone after an implicit auction. The sell offer price-quantity function in the export zone and the buy bid price-quantity function in the import zone are not observable. Hence, I use the amount of counterfactual reserve capacity as a proxy, assuming that the added sell offer quantity in the export zone would be equal to the volume of counterfactual reserve capacity if the FCFS rule was maintained. This assumption seems plausible because those who used to reserve capacity were not only required to submit sell offer or buy bid functions in the market but also given the transitional financial transmission right conditional on making a successful bid in the market, as discussed in the previous section. Note that I assume that those who used to reserve the interconnector capacity would not change the sell offer price but increase the sell offer quantity. This assumption is based on the fact that the sell offer price by incumbent vertically integrated utilities was quasi-regulated to be equal to marginal costs under the net bidding, as explained in Sect. 2. ${\beta }_{1}$ and ${\beta }_{2}$ are the parameters of interest to measure the capacity release and volume effects, respectively.

6 Data

The Japan Electric Power Exchange publishes half-hourly day-ahead market price data. I calculate the average hourly market prices (yen/MWh) to merge with the hourly covariates. The half-hourly data on operating capacity, margin, and reserved interconnector capacity are available at the Organization for Cross-regional Coordination of Transmission Operators. I use the reserved interconnector capacity at 3:00 p.m., 2 days before delivery, as the reserved interconnector capacity under the FCFS rule. This implicitly assumes that incumbents could have canceled the reservation after the day-ahead market on day t − 1 but did not cancel from 3:00 p.m. on day t − 2 to 10:00 a.m. on day t − 1. This assumption seems plausible because they had to keep the reservation until the day-ahead market clearing to cause market splitting. The volume variable equals the predicted counterfactual reserve capacity and is estimated in Sect. 8. Supply and demand variables are gathered from transmission system operators’ websites. The daily coal price data are obtained from the globalCOAL NEWC Index, whereas the Energy Information Administration assembled the daily Brent oil price. The monthly LNG import price is calculated based on the Trade Statistics of Japan in the Ministry of Finance. I used the data from June 4, 2016, to March 31, 2019. I exclude the data after April 1, 2019, because the financial transmission rights market was introduced that day, which may have changed the bidding behavior of the market participants, particularly new entrants. Additionally, there is an increasing risk of omitted variable bias due to unobservable events as the period extends.

Table 1 presents the descriptive statistics. The number of observations in the price gap between the Hokkaido and East zones is slightly smaller than that between the West and Kyusyu zones because the day-ahead market in Hokkaido was closed from September 7 to 26, 2018, due to the Hokkaido Eastern Iburi earthquake. The NAIC of the Kanmon interconnector is much larger than that of the Kitahon interconnector. The Kanmon interconnector has an operating capacity of 2800 MW, whereas the Kitahon interconnector has an operating capacity of 600 MW. Solar power generation in Japan is much greater than wind power generation, except in the Hokkaido zone. Nuclear power plants were present in all four zones but stopped operation after the Fukushima nuclear accident in 2011. Only some nuclear power plants in the West and Kyusyu zones resumed operation during the study period.

Table 1 Descriptive statistics (June 4, 2016–March 31, 2019)

7 Results

Table 2 shows the short-run effect of the implicit auction on the price gap between the Hokkaido and East zones. The first column shows the result of the sample which includes 1 week after the implicit auction. The implicit auction has a negative but insignificant effect on the price gap. The lack of significance may be because of the lack of variation in the dummy variable. The second column includes 2 weeks after the implicit auction. The price gap decreases by 1300 yen/MWh on average 2 weeks after the implicit auction. The third, fourth, fifth, and sixth columns include periods 1, 2, 3, and 6 months after the implicit auction, respectively. The implicit auction increases the price gap by 975–1,372 yen/MWh after 3–6 months. The change in the sign of the effect may be related to the relative size of the capacity release and volume effects, which are further examined in Table 2.

Table 2 Short-run effect of the implicit auction on the price gap among North zones

Table 3 shows the short-run effect of the implicit auction on the price gap between the West and Kyusyu zones. The implicit auction increases the market price gap after 2 weeks, 1 month, and 2 months by 400–600 yen/MWh. I decompose the implicit auction into interconnector capacity release and volume effects to better understand the impact of the implicit auction on market prices.

Table 3 Short-run effect of the implicit auction on the price gap among the West zones

Table 4 denotes the short-run effect of capacity release and volume on the price gap between the Hokkaido and East zones. NAIC of Kitahon is negatively associated with the price gap and is statistically significant in the sixth column. One MW increase of NAIC of the Kitahon interconnector relates to an average decline of 17 yen/MWh in the outcome. By contrast, the volume variable, a proxy for the additional sell offer quantity in the East zone, is positively associated with the market price gap. Then, a 1 MWh increase in the volume relates to a 40–60 yen/MWh increase in the outcome. I can confirm that capacity release and volume have a counteracting effect on the market price gap. The magnitude of the volume effect appears to be larger than the capacity release effect. Note that the volume variable is an incomplete proxy (sell offer and buy bid functions added in the export and import zones after the implicit auction are unobservable). The measurement error makes it difficult to interpret the result as it indicates a causal relationship.

Table 4 Short-term effect of capacity release and volume on the price gap among North zones

Table 5 denotes the short-run effect of capacity release and volume on the price gap between the West and Kyusyu zones. NAIC of the Kanmon interconnector is negatively associated with the price gap and is statistically significant in all the columns. One MW increase in NAIC relates to an average decline of 0.4 yen/MWh in the outcome. The volume variable is positively associated with the market price gap in the second to sixth columns. A 1 MWh increase in the volume relates to a 0.6–0.7 yen/MWh increase in the outcome. This result confirms that the capacity release and volume have counteracting effects on the market price gap in the western zones.

Table 5 Short-term effect of capacity release and volume on the price gap among the West zones

The price gap will vary throughout the day. Table 6 reports the interaction model that the NAIC of each interconnector and the volume variables interacted with each of the 23 h of the day to investigate the heterogeneous effect. The increase in NAIC for Kitahon has a price gap-lowering effect, but this effect varies by hour. The effect of lowering the price difference weakens from 6:00 a.m. to 1:00 p.m. but becomes even stronger from 4:00 p.m. to 6:00 p.m.

Table 6 Heterogeneous effect of capacity release and volume throughout a day

The volume effect varies in magnitude in terms of its impact on the increase in the price gap within a day. The effect on the price gap weakens from 5:00 a.m. to 6:00 a.m., whereas the effect of increasing the price gap strengthens to nearly two to three times from 3:00 p.m. to 10:00 p.m. These heterogeneous effects within a day may reflect the supply and demand conditions. Evening hours correspond to the peak demand when incumbents and new entrants tend to send buy bids with high prices and sell offer quantity is less because incumbents have less surplus energy.

The result between the West–Kyusyu zones shows a different pattern. The capacity release effect of Kanmon NAIC has a price gap-lowering effect, which weakens between 1:00 a.m. and 6:00 a.m. as well as between 5:00 p.m. to 11:00 p.m. The volume effect has a price gap-increasing effect, weakening during 1:00 a.m.–6:00 a.m., 10:00 a.m.–12:00 p.m., and 5:00 p.m.–11:00 p.m. The different patterns compared with eastern Japan may be attributed to increased solar power generation and restarted nuclear power operations in the western zones, which affect the sell offer and buy bid price quantity pairs in those areas.

8 Welfare analysis

8.1 Gain from trade

Figure 10 illustrates the gain from trade under the FCFS rule and implicit auction. The blue dotted upward-sloping curve represents the sell offer curve in the export zone. I assume the sell offer curve approximates the linear marginal cost function. This assumption is plausible for two reasons. First, the vertically integrated incumbents have been required to sell their surplus electricity in the day-ahead market at marginal costs of their generators since 2013, as stated in Sect. 2. This quasi-regulation on sell offer price applied not only to their surplus electricity but also to the gross bidding started in 2017. The sell offer quantity by the incumbents amounts to more than 80% and 70% before and after the implicit auction, respectively (Electricity and Gas Market Surveillance Commission, 2019a, 2019b). Therefore, the quasi-regulation effectively prevented the incumbents from exercising market power by setting a high sell offer price, making the market price primarily reflect the marginal cost of fossil fuels. Second, the regulatory agency, the Electricity and Gas Market Surveillance Commission, monitored their sell offer prices and found no manipulation (Ofuji & Tatsumi, 2016; Electricity and Gas Market Surveillance Commission, 2021b).

Fig. 10

Gain from trade under the FCFS rule and implicit auction

The buy bid curve in the export zone is shown in the black solid vertical line and is assumed to be perfectly inelastic in the short-run. The trapezium bounded by the sell offer curve, the buy bid line, the x-axis, and the y-axis is the total generation cost in the export zone under autarky. The red solid upward-sloping curve represents the sell offer curve from the import zone shown in the mirror image: The x-axis of the import zone starts from the right to the left. The buy bid curve in the import zone is shown in the same vertical line as the export zone, and the sum of the buy bid quantity in the export zone is equal to the width of the x-axis. The trapezium surrounded by the sell offer curve, the buy bid line, the x-axis, and the y-axis is the total generation costs in the import zone under autarky. The vertical line DG denotes the price gap between the two zones under autarky.

Electricity is traded across zones when interconnector capacity is available at the day-ahead market. q_FCFS denotes counterfactual trade volume if the FCFS rule was in effect after October 2018. When the trade is realized, the export zone increases electricity production, whereas the import zone reduces electricity production, substituting it with inexpensive electricity from the export zone. Consequently, the market price gap between export and import zones converges from DG to EF. The trapezium EFGD surrounded by the two sell offer curves denotes the gain from trade under the FCFS rule relative to autarky.

The implicit auction has a volume effect. When cheap electricity previously contracted through a bilateral agreement is offered into the day-ahead market at the export zone, the sell offer curve shifts rightward (shown by the blue solid line in the figure), lowering the autarky price in the export zone. Therefore, the volume effect increases the price gap under autarky. The interconnector capacity release and volume effects can increase trade quantity to q_IA relative to the counterfactual trade quantity q_FCFS under the FCFS rule. The trapezium ABCD shows the gain from trade under the implicit auction. It is interpreted as production cost savings resulting from efficient merit order dispatch across zones and reduced market power (Mansur and White 2012; Ryan, 2021; Cicala, 2022).

The base of trapezium ABCD is CD, which is a counterfactual market price under autarky after the introduction of implicit auction. I cannot estimate the CD and the overall gain of the implicit auction because individual sell offer and buy bid data are unavailable. Instead, I estimate part of the gain from the implicit auction. The trapezium AHFE is the gain from the interconnector capacity release under the implicit auction. The trapezium AHFE is calculated as follows:

$$\Delta W=\frac{1}{2}\left(AH+EF\right)\times \left({q}_{IA}-{{q}}_{FCFS}\right).$$

(4)

8.2 Counterfactual prediction

I need to predict AH, a counterfactual market price gap under the FCFS rule if trade quantity were ${q}_{IA}$; EF, a counterfactual market price gap under the FCFS rule; and ${{q}}_{FCFS}$, a counterfactual trade quantity if the complete implicit auction had not replaced the FCFS rule after October 2018, to estimate the gain from the interconnector capacity release under the implicit auction. I use the least absolute shrinkage and selection operator (LASSO), RF, DNN, and linear regression to compare the mean square error (MSE) of the predicted counterfactuals to obtain the best prediction model. These machine-learning methods can flexibly approximate the conditional mean of the dependent variable and predict counterfactuals better than linear regression.

LASSO adds a regularizer to the linear regression, which penalizes the sum of the absolute values of the coefficients, helping the model avoid in-sample overfitting (Mullainathan & Spiess, 2017). I use tenfold cross-validation to select the optimal value of the tuning parameter and the R package “gmnnet” to train the model.

RF combines a regression tree model with bagging. The regression tree sequentially divides the sample based on the explanatory variable’s threshold value and predicts an outcome variable. The RF takes the average of several hundred trees constructed by bootstrapping the training sample with randomly chosen subsets of explanatory variables. One of the advantages of RF is that it requires relatively little tuning (Athey & Imbens, 2019). I use the R package “randomForest” to train the model.

For LASSO and RF, I divide the pre-treatment sample into training and test data in a ratio of 8:2. The sample is not randomly divided to ensure that the training data are always older than the test data. Pre-treatment data are collected from June 4, 2016, to September 30, 2018. I use the training data to estimate the model by minimizing the MSE. The test data are used to evaluate the predictive performance of the models with the loss function.

For DNN, the sample is divided into training, validation, and test data in a ratio of 6:2:2 as DNN requires hyperparameter tuning. Again, the sample is not divided randomly. The validation data are used to select the optimal hyperparameters to avoid overfitting. The test data are reserved for evaluating the model performance. The training data are standardized by extracting the mean and dividing it by the standard deviation. The validation and test data use the same mean and standard deviation values for standardization. A DNN is trained using training data with a rectified linear unit activation function and the Adam optimizer, which is a stochastic gradient descent algorithm. Hyperparameters include several hidden layers, the number of units, dropout rate, and learning rate. A “hyperband” is used to find the best set of hyperparameters efficiently (Li et al., 2018). The number of epochs is set to 800, and the batch size is 64. Early stopping is introduced to avoid overfitting, which completes training when the MSE does not decline consecutively over five epochs. The model is trained using the “TensorFlow” and “Keras” libraries in Python. Table 7 lists the combinations of the search space of the hyperparameters (Table 7).

Table 7 Hyperparameters for tuning

I first predict the counterfactual reserved interconnector capacity if the FCFS rule was in effect after October 2018 to calculate the gain from trade under the implicit auction. I model the reserved interconnector capacity under the FCFS rule as a function of the gross available interconnector capacity (GAIC) and time-fixed effects:

$${q}_{reserve,t}={\alpha }_{0}+{\beta }_{1}{GAIC}_{t}+{FE}_{t}+{e}_{t},$$

(5)

where ${q}_{reserve,t}$ is the reserved interconnector capacity under FCFS. GAIC is operating capacity minus the margin. This equation indicates that the reserved interconnector capacity is restricted by GAIC. FE_t denotes the hour, day of the week, and month effects.

Table 8 summarizes the predictive performance of each model. The first to fourth columns are the result of the linear regression, LASSO, RF, and DNN, respectively. I predict the counterfactual reserved interconnector capacity by RF as the RF model has the smallest test MSE for both interconnectors.

Table 8 Test MSE of reserved interconnector capacity

Based on the model estimated by RF, I predict a counterfactual reserved interconnector capacity ${\widehat{q}}_{reserve, t}^{post}$, if the FCFS rule was maintained after October 2018, using the data in the post-treatment period. Figure 11 shows the observed and counterfactual reserved interconnector capacity. The black dots indicate the actual reserve capacity before the implicit auction, and the red dots show the counterfactual reserve capacity after October 2018.

Fig. 11

Reserved interconnector capacity under FCFS rule

Next, I predict the counterfactual price gap (corresponding EF in Fig. 10) if the FCFS was not replaced by the implicit auction, as follows:

$$ \Delta p_{, t} = \alpha_{0} + \beta_{1} NAIC_{t} + \beta_{2} Supply_{t} + \beta_{3} Demand_{t} + FE_{t} + \varepsilon_{t} . $$

(6)

Equation (6) is similar to Eq. (3). However, the model has no volume variable because it only appears after the introduction of the implicit auction. I only use the data from the pre-treatment period (i.e., before October 2018). FE includes hour, day of the week, and month fixed effects. Table 9 shows that RF can predict the day-ahead market price gap between Hokkaido and the East with the lowest MSE, whereas OLS produces the lowest MSE for the West and Kyusyu compared with other machine-learning methods.

Table 9 Test MSE of day-ahead market price gaps

Thus, I use RF and OLS to respectively predict the counterfactual price gap between Hokkaido and the East zones and between the West and Kyusyu zones in the post-treatment period. Figure 12 presents the observed and counterfactual market price gaps. The black dots show the daily mean actual price gap, and the red circles represent the daily mean counterfactual price gap after October 2018.

Fig. 12

Day-ahead market price spreads

Next, I estimate the counterfactual trade volume through each interconnector ${{q}}_{FCFS}$ allocated in the day-ahead market if the FCFS rule was not abolished, as follows:

$${\widehat{q}}_{FCFS}={\widehat{NAIC}}_{t} \;\, if \;\, \Delta {\widehat{p}}_{, t}\ne 0.$$

(7)

Note that the estimated $\widehat{NAI{C}_{t}}$ and counterfactual price gap $\Delta {\widehat{p}}_{, t}$ are used. ${\widehat{q}}_{FCFS}$ is equal to $\widehat{NAI{C}_{t}}$ as long as the market price gap is nonzero because the market splitting algorithm always allocates all the remaining capacity in the wake of congestion (Marmiroli et al., 2009).

Finally, I calculate the counterfactual day-ahead market price if the FCFS rule was in effect after October 2018 and the trade quantity was q_IA (denoted as AH in Fig. 10), as follows:

$$AH=EF+{\beta }_{Capacity\,\,Release}\left({q}_{IA}-{\widehat{q}}_{FCFS}\right).$$

(8)

I need to estimate the effect of capacity release on the price gap: ${\beta }_{Capacity Release}$, to evaluate the gain from trade under the implicit auction. The potential concern to estimate the effect of capacity release is endogeneity. NAIC could be endogenous when, for example, incumbents determine the amount of interconnector capacity reservation and cancelation based on the expected price gap in the coming day-ahead market. Hence, I regress Eq. (3) but with the data from October 1, 2018–March 31, 2019, to address this concern. NAIC would no longer be an endogenous variable in this period because incumbents could not reserve the interconnector capacity after October 1, 2018.

Table 10 presents the estimation results. Column (1) is the baseline estimation with the hour and day of the week fixed effects. This result indicates that a 1 MW increase in the NAIC of the Kitahon line reduces the day-ahead market price gap between Hokkaido and the East zone by 13.81 yen/MWh. The volume variable is statistically insignificant to the outcome. Columns (2) and (4) introduce additional fixed effects. Column (2) includes month fixed effects, and column (3) includes hour × month fixed effects. Column (4) includes both month and hour × month fixed effects and is our preferred specification. The result shows that a 1 MW increase in the NAIC of the Kitahon line reduces the day-ahead market price gap between Hokkaido and the East zone by 14.09 yen/MWh.

Table 10 Capacity release effect of the Kitahon interconnector

Table 11 shows the estimation results for the Kanmon interconnector across the West and Kyusyu zones. A 1 MW increase in the NAIC of the Kanmon interconnector decreases the market price gap by 1.7 yen/MWh between the West and Kyusyu zones across all specifications. The coefficient size is much larger than that in Table 5, which indicates that an endogeneity may bias the previous estimates. Column (4) is my preferred specification and is used to estimate trade gains by capacity release. The coefficient of Kanmon’s capacity release is much smaller than that of Kitahon. I argue that the difference comes from the difference in marginal cost between trade zones.

Table 11 Capacity release effect of the Kanmon interconnector

Figure A in the Appendix presents the hourly zonal market price and the estimated marginal costs of fossil fuels. The estimated marginal costs help us identify the marginal generator’s fuel type, determining the zonal market price (Germeshausen and Wolfing, 2020). For example, when the zonal market price was higher than the estimated marginal cost of oil-fired generation, oil-fired generation was the marginal generator. Similarly, when the zonal market price is lower than the oil-fired generation but higher than the estimated marginal cost of the LNG generation, the LNG generation was assumed to be the marginal generator in that hour.

In the Hokkaido zone, oil-fired generation seemed a marginal generator for many hours. The Hokkaido zone had no LNG generation power plant until the end of February 2019. By contrast, LNG generation seemed to be marginal generators in the East zone for many hours. This result suggests a significant gain from trade among these zones through the Kitahon interconnector because the marginal cost of oil is much larger than that of LNG generation or coal. Conversely, the West and Kyusyu zones have similar types of marginal generators (i.e., LNG or coal generation), resulting in a small capacity release effect (Gugler & Haxhimusa, 2019).

8.3 Estimates of gain from trade from the capacity release effect under implicit auction

Table 12 summarizes the result of welfare analysis under FCFS rule and implicit auction. I estimate that the counterfactual average price gap between Hokkaido and the East zones under the FCFS rule after October 2018 would be 3196.13 yen/MWh, compared with the 3116.39 yen/MWh 6 months before the introduction of the implicit auction. The counterfactual average price gap between the West and Kyusyu zones under the FCFS after October 2018 would be 485.80 yen/MWh for Kanmon, compared with the 635.17 yen/MWh before October 2018. These estimates indicate that if the FCFS rule was in effect after October 2018, then the level of the price gap would be similar to the value observed in the past 6 months.

Table 12 Gain from trade under FCFS and implicit auction

The counterfactual average price gap after October 2018 if the implicit auction had only capacity release effect is 2049.59 yen/MWh between Hokkaido and the East zones and 4.39 yen/MWh between the West and Kyusyu zones. This result indicates that the average price gap after the introduction of the implicit auction would be lower than the counterfactual price gap under the FCFS rule by 1146.54 yen/MWh for Hokkaido and East zones and 481.41 yen/MWh for West and Kyusyu zones.

The actual hourly average trade quantity under the FCFS rule is 38.26 MW for Kitahon and 31.00 MW for Kanmon. The counterfactual hourly average trade quantity after October 2018 is 28.53 MW for Kitahon and 436.65 MW for Kanmon. Then, the actual hourly average trade quantity after the implicit auction increased to 127.90 MW for Kitahon and 1953.34 MW for Kanmon. Thus, I estimate that implicit auction increased trade quantity on average by 99.37 MW per hour for Kitahon and 1516.69 MW per hour for Kanmon, relative to the FCFS rule.

Finally, using Eq. (4), I estimate that the capacity release effect of the implicit auction reduced 1.09 billion yen (10 million USD⁵) of generation cost for Hokkaido and the East zones and 1.68 billion yen (17 million USD) for West and Kyusyu zones for 6 months. The total annual generation cost reduction becomes approximately 5.54 billion yen (55 million USD). Although this gain only captures the gain from the capacity release effect and thus provides conservative estimates, it is significantly larger than the one-time implementation cost of the implicit auction, which, according to a member of the Japan Electric Power Exchange, is 20–30 million yen.

9 Conclusion

The difference in the day-ahead market price between the Hokkaido and East zones was very large under the FCFS rule. The reservation rate for the Kitahon interconnector was almost 100%, and the available capacity in the day-ahead market was nearly zero. Hence, the congestion was the norm. Analysis of the interconnector reservation data with the market data illustrates that the scarce interconnector capacity was frequently canceled after the day-ahead market clearing. A similar situation occurred in the Kanmon interconnector in a small magnitude. This may have been an attempt by those who could reserve the capacity to strategically cause market splitting and increase the market prices in the import zone. This result indicates that the FCFS rule created a favorable condition for incumbents at the cost of new entrants.

This study evaluates the impact of one of the new market-based congestion management methods of interconnector transmission capacity: the implicit auction. The implicit auction has two effects on the day-ahead market: the interconnector capacity release and volume effects. The implicit auction increases the interconnector capacity allocated in the day-ahead market and allows additional trade between the export and import zones. The volume effect arises because the implicit auction requires those who used to reserve the capacity to submit a sell offer or buy bid in the market to acquire the capacity. Counterfactual market outcomes are predicted by combining linear regression with machine-learning methods to estimate increased gain from trade under the implicit auction. The partial gain from trade produced by the capacity release effect under the implicit auction is estimated to be approximately $10 million for the Kitahon and $17 million for the Kanmon interconnector for 6 months. The annual national production cost savings from the capacity release effect is approximately $55 million.

This study reveals the competitive advantages of the implicit auction relative to the FCFS rule. The implicit auction not only prevents incumbents from exercising market power by withholding interconnector capacity in the day-ahead market but also promotes efficient resource allocation across zones. However, it seems that the exercise of market power was not a significant issue under the FCFS rule because I show that the implicit auction does not decrease the price gaps in the northern zones and western zones. One reason is the estimated size of the capacity release effect is smaller than the estimated size of the volume effect. Another possible reason is the vertical integration between generation and retail segments, or the quasi-regulations requiring the vertically integrated incumbents to send sell-offers based on marginal costs and to increase sell-offer volume under the gross bidding initiative could mitigate their market power of energy withholding in the day-ahead market.

This study has two limitations owing to the lack of available data. First, there are no firm-level interconnector capacity reservation data and bidding information available. Therefore, further verification of whether incumbents exercise market power using capacity reservation is impossible. Second, I cannot estimate the effect of the implicit auction on the reduction of CO₂ emissions because transmission system operators only publish aggregate fossil fuel generation data. The implicit auction may reduce CO₂ emissions by decreasing generation from oil- and gas-fired power plants with high marginal costs in an import zone while increasing the CO₂ emission by dispatching cheaper coal-fired power plants more frequently in an export zone.

Appendix

1.1 Estimates marginal costs of fossil fuels by zone

I calculate marginal costs for each region (i), each fuel type (f) (i.e., coal, LNG, and oil), and each month (m).

$$m{c}_{ifm}=\frac{CIF\,\,pric{e}_{fm}+other\,\,fuel\,\,related\,\,cos{ts}_{f}}{Heat \;rat{e}_{fm}\times Heat\,\,efficienc{y}_{if}\times (1-self\,\,consumption\,\,rat{e}_{if} )}+O\&{M}_{f}.$$

mc is the marginal cost (yen/MWh). CIF price (yen/t) is the cost, insurance, and freight price. CIF data are from the Energy Information Center (2024). Other fuel-related costs (yen/t) include domestic transport costs and landing charges. I use the data from “kakudengen no syomoto ichiran” of the Agency for Natural Resources and Energy (2024b). Heat rate (MJ/t) is collected from “denryoku chousa toukei” of the Agency for Natural Resources and Energy (2024a). Heat efficiency (%) is from “denryoku zyukyuu no gaiyou 2012” published by the Agency for Natural Resources and Energy (2012). Self-consumption rate (%) is from the Federation of Electric Power Companies of Japan. O&M is operation and maintenance costs, which come from “kakudengen no syomoto ichiran” of the Agency for Natural Resources and Energy. Figure 13 shows the hourly market price and the estimates of each zone’s monthly average marginal costs. The marginal cost of the East zone is calculated as the mean of the marginal cost of the Tohoku and Tokyo regions. The marginal cost of the West zone is calculated as the mean of the marginal cost of the Chubu, Kansai, Hokuriku, Shikoku, and Chugoku regions.

Fig. 13

Hourly zonal price and estimated marginal costs of fossil fuels. Zonal day-ahead market price is shown in black dots. The marginal cost of oil-fired, LNG, and coal power plants is shown in brown, pink, and gray lines, respectively