Price Impacts of Energy Transition on the Interconnected Wholesale Electricity Markets in the Northeast United States

Abstract

Our regression analysis documents that energy policies to promote renewable energy development, as well as hydroelectric imports from Canada, lead to short-run reductions in average electricity prices (also known as merit-order effects) throughout the Northeast United States. Changes in the reliance upon renewable energy in one of the Northeast’s three interconnected electricity markets will impact wholesale prices in the other two. The retirement of a 1000 MW nuclear plant can increase prices by about 9% in the Independent System Operator of New England market and 7% in the New York Independent System Operator market in the short run at reference hubs, while also raising prices in neighboring markets. Some proposed large-scale off-shore wind farms would not only lower prices in local markets at the reference hubs modeled but would also lower prices in neighboring markets.

1. Introduction

The states in the Northeast region of the United States (U.S.) are among those leading the nation’s energy transition [1,2]. State-level environmental policies in this region have accelerated the development of variable renewable energy (VRE) generated by solar systems and wind farms. These policies have also increased the import of renewable hydroelectric generation from Canada and hastened the retirement of aging power plants with high carbon emissions. Further, the retirement of aging nuclear power plants has altered the generation profile in this region of the U.S. As the wholesale electricity markets in the Northeast are linked by North America’s Eastern Interconnect, actions taken in one state toward energy transition impact electricity prices throughout the entire Northeast.

1.1. State-Level Actions

Within the Northeast, New York has established a goal of 70% renewably sourced electricity by 2030 and zero emissions from its electricity grid by 2040 [3]. Maryland’s Climate Pollution Reduction Act seeks to reduce greenhouse gases by 60% by 2031 and achieve zero emissions by 2045 [4]. New Jersey aims to achieve 100% reliance on clean energy by 2050 [5]. Vermont’s goal is net-zero emissions by 2050 across all sectors [6]. Massachusetts’ Clean Energy and Climate Plan targets a reduction in greenhouse gas emissions of 85% below 1990 levels by 2050 [7]. Connecticut aims to cut 45% of its 2001 emissions levels by 2030 [8]. Maine strives to achieve an 80% reduction in greenhouse gases below 1990 levels by 2050 [9]. Rhode Island’s goal is 100% reliance on renewable energy by 2030 [10]. Pennsylvania’s goal is a 26% reduction in greenhouse gases below 2005 levels by 2025 and an 80% reduction by 2050 [11]. The Virginia Clean Energy Economy Act of 2022 seeks to transition the state’s electric grid to 100% clean energy by 2050 [12].

In New York and New England, CO₂ emissions reduction is achieved in part through the purchase of hydroelectric power generated in Eastern Canada [13]. The importance of these electricity imports to the Northeast was highlighted when the Trump administration threatened to impose tariffs on Canadian imports, and the Premier of Ontario threatened to retaliate by curtailing hydroelectric exports to the U.S. [14].

1.2. Scope

Here, we examine how prices in a market (as well as the prices in a neighboring interconnected market) change as the dependence upon renewable energy increases. Our paper does not separate the impacts of individual state-level energy policies.

The rising market penetration of renewable energy reduces energy prices in a wholesale electricity market via their merit-order effects—a well-documented finding reported in numerous literature reviews [15,16,17,18]. VRE capacity expansion exacerbates the missing money problem [19,20] and the VRE cannibalization problem [21,22,23]. However, a contrary result is possible under certain circumstances [24].

Many of the Northeastern states participate in the Regional Greenhouse Gas Initiative (RGGI), through which carbon trading is enabled by a cap-and-trade program for CO₂ emissions permits. The effect of carbon trading on wholesale energy prices has been extensively studied for California and the Pacific Northwest [25,26]. However, the effects of carbon trading on prices in the PJM Interconnection (PJM), the New York Independent System Operator (NYISO), and the market operated by the Independent System Operator of New England (ISO-NE) have received less academic attention. When retail electricity pricing captures the wholesale energy price increase attributable to the marginal cost of carbon, it improves the economic efficiency of retail electricity consumption and the cost-effectiveness of distributed energy resources.

Antecedent studies have explored the spillover effects of renewable energy policies in one region or state on neighboring regions or states. Zhou et al. [27] conclude that, because of these spillover effects, research confined to quantifying the impacts of renewable portfolio standards in one state may understate the holistic impact of state-level policies. The regression analysis of Abrell and Kosch [28] examines how VRE development in Germany impacts prices and emissions in other Central European countries via trade flows and merit-order effects. Phan and Roques [29] document that the growth of VRE in Germany affects electricity prices in France, while Frauendorfer et al. [30] focus on how Germany’s energy transition affects prices in Switzerland. Unger et al. [31] report that Denmark’s wind generation affects prices throughout the Scandinavian electricity market. Finally, Stiewe et al. [32] find spillover effects of VRE generation throughout Western Europe.

Here, we examine how the shift toward a low-carbon electricity future impacts the energy prices of the three wholesale electricity markets operated by the regional transmission organizations (RTOs) that serve the Northeast U.S. Specifically, we explore renewable energy’s short-run merit-order effects, including the price reduction attributable to the import of hydroelectric generation from Canada. We are unaware of any previous attempts to model the impacts of these supply-side changes, as well as the RGGI, on prices in this region using a transparent regression approach that captures relationships among wholesale prices between these three interconnected markets.

1.3. Geography

Figure 1 portrays the Northeast wholesale electricity markets operated by ISO-NE, NYISO, and PJM. Our geographic choice of the Northeast U.S. reflects that all three RTOs have similar market designs based on the concept of locational marginal pricing, and electricity trading is active among the three wholesale markets. Further, all three markets largely reside in the same time zone, thus obviating the difficulty of properly matching hourly market data across interconnected markets.

A better understanding of the process of price formation in the region depicted in Figure 1 is important, considering the environmental commitments made by the heavily populated Northeast states. However, the impact of renewable energy development on wholesale energy prices in these markets has received less attention than in the markets in Texas and California, likely because the Northeast’s wind and solar generation development has been lagging due to a less favorable climate with fewer sunny and windy days.

1.4. Overview of Findings

Our regression-based approach enables us to test hypotheses related to the merit-order effects of renewable energy (wind, solar, and hydroelectric imports from Canada) and the wholesale markets’ passthrough of the marginal costs of CO₂ emissions. Using a large sample of over 74,000 hourly observations of real-time market data in the 9-year period from 2016 to 2024, we find the following:

• The correlation in wholesale prices between the New England and New York markets is far higher than between either market and the PJM market;
• Increases in wind, solar, and nuclear generation, as well as hydroelectric imports from Canada into New England, lead to a short-run reduction in wholesale electricity prices in the three Northeast wholesale markets;
• The marginal cost of CO₂ emissions is passed through to prices in the three markets;
• There are many challenges inherent in using a regression approach to model price formation in these markets. The results are somewhat sensitive to the estimation method selected. Particularly in NYISO, wholesale electricity prices are not strongly linked to fuel costs, as reflected by natural gas and coal prices at trading hubs.

The rest of this paper proceeds as follows. Section 2 develops our econometric specification and testable hypotheses. Section 3 describes our large sample of hourly market data and provides descriptive statistics and regression results. Section 4 provides conclusions, policy implications, and caveats.

2. Materials and Methods

We adopt a regression-based approach to model wholesale prices in the three markets, allowing us to estimate energy price relationships amenable to hypothesis testing. While the use of a production costing model or a resource planning model would enable a far more detailed analysis of the impacts of changes in the supply side of these markets, such models tend to lack transparency, and the results from such models are often difficult to replicate. Moreover, we are interested in relationships in wholesale prices between the three markets, which would be difficult to explore with separate production costing or resource planning models established for the three separate markets. For our purposes, regression modeling suffices to provide some general insights into the questions that we raise here.

2.1. Specification

While the three wholesale electricity markets in the U.S. set prices on a 5 min basis, our specification uses hourly real-time data at the system level to balance data availability, estimation transparency, and inter-RTO granularity. PJM does not publicly provide data pertaining to the output of certain types of generating units with granularity of less than one hour, impairing our ability to model wholesale price changes on a 5 min basis. Using real-time data circumvents the difficulty of matching day-ahead energy prices with day-head forecasts of the fundamental drivers or determinants of prices, such as the hourly load, solar generation, and wind generation, which may not be readily available. The use of system-level data, rather than data at a nodal or zonal level, enables the modeling of a manageable set of statistical relationships.

Let P_jt denote the hourly wholesale real-time prices (USD/MWh) of RTO_j in hour t, where j = 1 for ISO-NE, 2 for NYISO, and 3 for PJM, and t = 1 for the first observation, while T for the last observation. By modeling a single price in each market, we are ignoring intra-market congestion and line losses, which is a topic for further analysis.

We construct a fuel index for each of the three markets to approximate the cost of the fuel that was used in the power plant on the margin in each hour in each of the three RTOs. To construct the fuel indices, we first calculate the percentiles of hourly demand in each year. We assume that if energy consumption in an hour exceeds the 40th percentile, a natural gas plant is the marginal generation that sets the system price. This is consistent with our review of documents (identified in the Appendix A) suggesting that, at least in PJM and NYISO, a natural gas plant is on the margin about 60% of the time. If energy consumption in an hour is between the 30th and 40th percentiles of the hourly demand, we assign a 0.75 weight to the natural gas price and a 0.25 weight to the coal price relevant to that hour. If energy consumption is between the 20th and 30th percentile, a 0.75 weight is assigned to the coal price and a 0.25 weight is assigned to the natural gas price. If energy consumption is below the 20th percentile, we assume that coal is the marginal generational fuel and we simply use the coal price.

For each market, we use the North Appalachian coal price. Since this is reported with a weekly frequency, the same coal price is used for all hours in a week. It was converted from USD per short ton to USD/MMBtu by dividing the reported price by 25, since the heat content of coal is around 25 MMBTU/short ton.

Different natural gas prices are used for each of the three markets. The Algonquin price is used for ISO-NE, the Transco NY spot natural gas price is used for NYISO, and the TETCO price is used for PJM. For the Algonquin and TETCO prices, we apply the price set in the day-ahead market to reduce endogeneity concerns. We were unable to find a day-ahead price for Transco NY, so a spot market price is used. Hourly data are not publicly available, so values at the end of a trading day are repeated each hour until a new value is reported. We tested the use of the Henry Hub price for each of the markets, since it is a widely used indicative price for wholesale natural gas in the U.S. and a market price set in Louisiana would be clearly exogenous—i.e., not influenced by the electricity or natural gas demand in the Northeast states. However, its correlation with any of the other three natural gas price series was below 0.55, and we found it to be a poor predictor of electricity prices in the Northeast states. Moreover, the three natural gas price series that we decided to use are weakly correlated with each other, suggesting the need to use different natural gas prices for the three markets.

The natural gas and coal prices were multiplied by typical heat rates for natural gas and coal units in the three markets, thus allowing the prices of the two fuels to be used in the same index. Our review of various reports (discussed in the Appendix A) suggested that the use of 7.2 MMBtu/MWh was appropriate for ISO-NE and NYISO and 7.1 was appropriate for PJM. For coal heat rates, 10 MMBtu/MWh was used for NYISO and PJM, while 10.25 was used for NE-ISO. While the last coal plant in NYISO was retired in 2020 and the last coal plant in ISO-NE retired in 2023, we use the coal price in our fuel index since our estimation starts in 2016, and the coal price may serve as a proxy for other, less expensive generation sources.

The creation of the final fuel indices, F_jt for j = 1, 2, 3, involves multiplying the fuel prices by the assumed heat rates and then applying the weights (described above) based on the level of consumption in the hour. The fuel price index provides a very crude approximation to a merit-order curve, with the cost of serving higher levels of demand reflecting the cost of generating electricity with natural gas plants, the cost of meeting lower levels of demand reflecting the cost of generation from coal plants, and some mix of generating units on the margin in between. Ideally, the estimated coefficient on the fuel price index, β_jF, should be around 1.0, as the marginal price of electricity should be approximately equal to the marginal production cost after controlling for all other variables.

C_t represents the daily RGGI carbon price (USD/ton of CO₂ emissions) that does not vary by time of day. Its coefficient is β_jC > 0 for j = 1, 2, 3, which measures the price effect of a USD 1 increase in C_t. If P_jt’s passthrough of the marginal cost of carbon emissions is 100%, β_jC = 0.053, β_jF because natural gas’s carbon content is 0.053 ton per MMBtu [30].

Q_1t and Q_2t represent the total imports of (generally) hydroelectric energy from Quebec and New Brunswick into New England and from Ontario into New York in MWh, respectively. The associated negative coefficients are β_jQ1 and β_jQ2, which measure the price reduction effects of a 1 MWh increase in hydroelectric imports. We treat these variables as exogenous, since the imports are generally based upon purchased power agreements with predetermined scheduled quantities, as discussed in Appendix A.

D_jt represents RTO j’s forecasted system demand (MWh) for j = 1, 2, 3. Its coefficient is β_jD > 0 and measures the increase in price resulting from a 1 MWh increase in D_jt. We treat this as exogenous since the forecasts are generally provided on a day-ahead basis.

Because renewable energy generation in the Northeast remains quite small relative to other regions of the U.S. with more favorable climates for wind and solar generation, we combine wind and solar electricity generation into a single variable. Indeed, our modeling attempts to use separate variables for solar and wind generation provided insignificant or counterintuitive results. As a result, we define R_jt = RTO j’s solar + wind generation (MWh) for j = 1, 2, 3. For NYISO, grid-connected solar generation is not reported. Consequently, “other renewable generation” is used in this calculation for NYISO, which may include some landfill gas or other renewable energy in addition to solar. If its coefficient, β_jR, is below zero, this would provide support for a merit-order effect, whereby an increase in wind or solar generation would reduce the wholesale prices. However, it has been hypothesized that increases in renewable energy generation may increase (rather than decrease) market prices if such generation assets are operated by companies that also have fossil fuel generation assets [31]. Price increases may result if energy companies strategically reduce their conventional energy supplies when the renewable supply is high, thus negating any merit-order effect. Indeed, most of the providers of renewable energy in the Northeast also own and operate fossil fuel generation sources, including NextEra Energy Resources, Calpine, Dominion Energy, Avangrid, LS Power, ENGIE, NRG, Vistra, and Cogentrix Energy. Thus, the existence of an economic merit-order effect from renewable energy might not be a foregone conclusion.

We treat R_jt as exogenous, as solar and wind generation is based on the weather and the renewable energy resource and is not normally dispatched (i.e., curtailed, in this context). According to ISO-NE, solar curtailments undertaken to manage transmission congestion in the region have totaled only 312 MWh in recent years. Similarly, NYISO suggests that current solar curtailments remain low. In NYISO, the percentage of wind generation curtailed by the ISO has ranged from 1.4% to 3.4% in recent years. ISO-NE’s 2022 Annual Markets Report notes that curtailments of renewable energy, including wind, are infrequent and generally insignificant. Citations are provided in Appendix A. We were unable to obtain information about the frequency of wind generation curtailment from PJM.

The next regressor of interest is N_jt = RTO j’s nuclear generation (MWh). Its coefficient is β_jN < 0 for j = 1, 2, 3, measuring the price reduction effect of a 1 MWh increase in N_jt.

As power plant outages could raise prices under conditions of scarcity, we include O_jt = day-ahead forecasts of powerplant outages (in MW) as regressors in the NYISO and PJM models. Unfortunately, a similar variable is not available for ISO-NE.

As NYISO lies between ISO-NE and PJM, we postulate the following energy price regressions with time-dependent intercepts β_jt, which are a linear function of the time of day, day of week, and month of year and random error ε_jt:

P_1t = β_1t + β_1F F_1t + β_1C C_t + β_1Q Q_1t + β_1D D_1t + β_1R R_1t + β_1N N_1t + θ₁₂ P_2t + ε_1t

(1)

P_2t = β_2t + β_2F F_2t + β_2C C_t + β_2Q Q_2t + β_2D D_2t + β_2R R_2t + β_2N N_2t + β_2O O_2t + θ₂₁ P_1t + θ₂₃ P_3t + ε_2t

(2)

P_3t = β_3t + β_3F F_3t + β_3C C_t + β_3D D_3t + β_3R R_3t + β_3N N_3t + β_3O O_3t + θ₃₂ P_2t + ε_3t

(3)

Equations (1)–(3) capture the price effects of inter-RTO dependence. Specifically, θ₁₂ > 0 in Equation (1) indicates that ISO-NE’s energy price moves with NYISO’s energy price. Similarly, θ₂₁ > 0 and θ₂₃ > 0 in Equation (2) indicate that NYISO’s energy price moves with ISO-NE’s and PJM’s energy prices. Finally, θ₃₂ > 0 in Equation (3) indicates that PJM’s energy price moves with NYISO’s energy price.

Whether Equations (1)–(3) make sense is an empirical issue best settled by the regression results presented in Section 5 below.

2.2. Estimation Strategy

We use two-stage least squares (2SLS), three-stage least squares (3SLS), and full-information maximum likelihood (FIML) to estimate the three equations, since the (endogenous) wholesale prices in neighboring markets are used as explanatory variables.

Moreover, despite our efforts to minimize other endogeneity concerns through the use of the forecasted demand, forecasts of generation outages, and natural gas prices set in day-ahead markets, some endogeneity may nonetheless persist due to our need to apply hourly data, rather than 5 min data, in our estimation of relationships dictated by data availability. This introduces the possibility of some feedback among the variables within an hour, further justifying the use of a system-wide estimator or an estimator that can address endogeneity concerns. We are unable to quantify the degree of bias, if any, that may remain in our coefficient estimates due to the use of hourly data after applying 2SLS and 3SLS, since we do not have more granular data for many of the PJM variables and thus cannot explore this topic.

We present the results from three different estimation methods to enable us to explore the sensitivity of the results to the estimation method selected. In particular, 2SLS is commonly used in empirical work. While more efficient, 3SLS and FIML may be more sensitive to model misspecification and distributional assumptions [33] (p. 413); [34] (Chapter 4); [35] (Chapter 8). In light of these sensitivities, we will focus on the 2SLS estimation results when providing some example applications of our results.

In our 2SLS and 3SLS estimations, we used the default set of instruments in the SAS software (Ver. 9.4), which, in this case, is the set of all exogenous variables. Including variables that are poorly correlated with the dependent variables may lead to bias in the coefficient estimates [36,37]. In light of this possibility, we also estimated the model with smaller sets of instruments by removing exogenous variables with weak correlations with the dependent variables. Coefficient estimates were sensitive to the set of instruments used in the estimation. Using the full set of exogenous variables as instruments generally provided more plausible coefficient estimates. A reader uncomfortable with our selection of instruments may opt to rely upon the FIML results.

3. Results

3.1. Data Description

Our large data sample comprises 74,283 hourly observations for the period of 1 January 2016 to 31 December 2024. All data are obtained through NRGStream (https://www.arcuspower.com/), aside from the daily RGGI carbon prices, which come from the Regional Greenhouse Gas Initiative.

Table 1. Descriptive statistics of the hourly variables for ISO-NE, NYISO, and PJM; sample period = 1 January 2016 to 31 December 2024; sample size = 74,283.

Variable: Definition	Mean	Standard Deviation	Minimum	Maximum	Price Correlation
					P1t	P2t	P3
P1t: ISO-NE’s real-time energy price ($/MWh)	40.61	41.39	−156.05	2455.00	1.00	0.45	0.39
P2t: NYISO’s real-time energy price ($/MWh)	27.08	39.23	−2558.47	2442.13	0.45	1.00	0.37
P3t: PJM’s real-time energy price ($/MWh)	33.60	41.98	−77.94	3463.00	0.39	0.37	1.00
Ct: RGGI carbon price ($/ton of CO2 emissions)	8.78	5.43	2.53	25.75	0.17	0.18	0.11
O2t: Generation Outages in NYISO (MW)	5799	3137.47	932	23.86	−0.08	−0.07	−0.04
O3t: Generation Outages in PJM (MW)	33,159.11	16,659.07	7369.53	73,541.00	−0.12	−0.10	−0.02
F1t: ISO-NE Adjusted Fuel Index	30.69	31.69	2.94	756.00	0.51	0.27	0.28
F2t: NYISO Adjusted Fuel Index	23.87	26.09	1.44	1009.80	0.32	0.19	0.22
F3t: PJM Adjusted Fuel Index	22.797	22.26	2.07	673.78	0.42	0.24	0.31
Q1t: Imports to ISO-NE from Canada (MWh)	1540.60	616.44	−1405.00	2869.00	0.00	−0.01	0.03
Q2t: Imports to NYISO from Canada (MWh)	740.91	383.77	0	2000.00	0.00	−0.3	−0.04
D1t: ISO-NE’s forecast system demand (MWh)	13,317.54	2712.51	6370.00	26,500.00	0.37	0.27	0.23
D2t: NYISO’s forecast system demand (MWh)	17,145.11	3193.65	9033.00	31,401.00	0.27	0.24	0.23
D3t: PJM’s forecast system demand (MWh)	90,387.46	16,189.18	55,512.00	155,655.00	0.34	0.28	0.29
S1t: ISO-NE’s solar + wind generation (MWh)	444.62	291.80	1.00	1893.75	0.03	0.01	0.03
S2t: NYISO wind + other renewable generation (MWh)	783.46	433.89	170.17	2802.67	0.01	−0.8	−0.01
S3t: PJM’s solar + wind generation (MWh)	3526.65	2417.40	8.80	16,010.33	0.09	0.02	0.02
N1t: ISO-NE’s nuclear generation (MWh)	3235	719.01	857.50	4044	−0.01	0.00	−0.01
N2t: NYISO’s nuclear generation (MWh)	4090	966.41	1159	5450	−0.14	−0.19	−0.13
N3t: PJM’s nuclear generation (MWh)	31,700	2153	15,324	36,954	0.15	0.10	0.02

presents the descriptive statistics of the variables used in our regression analysis.

The wholesale prices are positively correlated among the three markets in the Northeast for four reasons. First, these hourly energy prices move with the wholesale natural gas prices, as natural gas is often the marginal generation fuel, although the correlation among movements in wholesale natural gas prices at various trading points in the Northeast tends to be limited. Second, the cross-border price effects of fundamental drivers of wholesale prices, such as the system load, VRE generation, nuclear generation, and hydro generation, exist in interconnected markets [25]. Third, a multistate event like a summer heat wave, winter ice storm, or transmission failure will impact the energy prices in each of the geographically close, interconnected wholesale electricity markets. Finally, the import of hydropower from Canadian provinces (Ontario, Quebec, and New Brunswick) may reduce prices throughout the region.

The New England Internal Hub price serves to represent the wholesale price of electricity in ISO-NE. For NYISO, the New York Real Time Reference Load Balancing Market Price (LBMP) is used. PJM did not calculate a market-wide indicative price until 15 June 2018, leading us to calculate our own indicative price by taking a simple average of the hub prices reported for PJM’s Eastern, Chicago, Ohio, and Western energy trading hubs. These wholesale prices include losses and congestion, although such components may be much smaller than the losses and congestion costs typically found in some specific geographical areas within these markets.

The hourly energy prices are noisy, as Figure 2 demonstrates. While they have sample means of USD 33.6/MWh to USD 40.61/MWh, their standard deviations are as large as the means. The hourly energy prices are positively correlated. The correlation in prices between NYISO and ISO-NE is far greater than the price correlation between either market and PJM. This makes sense because PJM has a larger geographical footprint such that parts of the PJM market are less influenced by factors outside of the Northeast.

The data for the two variables representing imports from Canada are volatile and do not seem to correlate with the wholesale energy prices. Nevertheless, our regression results show plausible relationships between hydroelectric imports and wholesale prices in the Northeast after controlling for the effects of all other determinants of wholesale prices.

The fuel index variables are positively correlated with wholesale prices and range from 0.19 to 0.51. Ideally, a correlation closer to 1.0 should be found. Alternative natural gas price series (e.g., average of prices at relevant natural gas trading hubs) were tested, along with alternative formulas for determining when natural gas plants versus coal or other types of power plants might be on the margin. In the end, a better means of constructing the fuel price indices could not be found. This suggests that—particularly in NYISO—there are many factors affecting wholesale prices that are not associated with the production cost of the type of generation on the margin. These may include the structure of contracts to import hydroelectric power from Ontario or the pricing of in-state hydroelectric generation—variables for which we have no information.

Table 2. Correlations between various fuel prices and wholesale prices.

Variable: Definition	Price Correlation
	P1t	P2t	P3	F1t	F2t	F3t
Henry Hub natural gas price	0.40	0.26	0.35	0.40	0.40	0.51
Algonquin day ahead natural gas price	0.52	0.26	0.28	0.96	0.46	0.78
Transco NY Spot Price for natural gas	0.19	0.20	0.23	0.48	0.96	0.49
TETCO day ahead price for natural gas	0.46	0.26	0.33	0.83	0.50	0.96
North Appalachian price of coal	0.30	0.20	0.29	0.36	0.40	0.47

provides a “deeper dive” into the correlation between wholesale prices in the three markets and various natural gas and coal prices. While all correlations among electricity and fuel prices are positive, the correlations are lower than one might expect. In NYISO, in particular, our wholesale electricity price series is not strongly correlated with the natural gas prices at trading hubs.

The daily RGGI prices average USD 8.78/ton. They are widely dispersed and weakly correlated (r ≤ 0.18) with the energy prices.

The remaining variables are forecast system demands, renewable generation, generation outages in two of the markets, and nuclear generation, which have volatile data. While the system forecast demand data have price correlations of 0.23 to 0.37, the generation data generally have weaker price correlations.

Numerous variables were tested but eventually not included in the estimation because their impacts were insignificant or their inclusion led to counterintuitive results. A good case in point is hydroelectric generation in the three markets. The actual demand and actual hourly imports of energy from Canada were tested, but they were dropped in favor of forecast or scheduled values to reduce endogeneity concerns.

3.2. Regression Results

The regression estimation results using 2SLS, 3SLS, and FIML are provided in

Table 3. Price regression results; sample period = 1 January 2016 to 31 December 2024; sample size = 74,283; all coefficient estimates significant at the 1% level, except for those in red italics. A significant coefficient with a counterintuitive sign is indicated by red bold.

	2SLS			3SLS			FIML
	P1t	P2t	P3	P1t	P2t	P3	P1t	P2t	P3
Adjusted R2	0.24	0.24	0.18	0.22	0.26	0.12	0.34	0.26	0.23
RMSE	36.59	34.60	38.60	37.04	34.09	40.12	34.04	34.07	37.32
Intercept	−76.206	12.54	−76.05	−77.68	3.33	−94.12	−39.54	3.51	−71.76
P1t: ISO-NE’s real-time energy price ($/MWh)		0.66			0.24			0.66
P2t: NYISO’s real-time energy price $/MWh)	0.67		0.47	0.69		0.50	0.21		0.30
P3t: PJM’s real-time energy price ($/MWh)		0.32			0.29			0.29
Fjt: Fuel index for local region	0.36	0.02	0.33	0.35	0.07	0.30	0.39	0.09	0.38
Ct: RGGI carbon price ($/tonne)	0.47	0.34	0.20	0.46	0.21	0.33	0.35	0.25	0.24
Ojt: Local generation outages (MW)		0.0005	0.0005		0	0.0005		0	0.0005
Qjt: Local import from Canada (MWh)	−0.003	0.002		−0.003	−0.002		−0.0015	0.001984
Djt: Local forecast system demand (MWh)	0.007	0.0005	0.0008	0.0069	0.0015	0.0009	0.00471	0.001556	0.00075
Rjt: Local renewable energy generation (MWh)	−0.00086	−0.00764	−0.00035	−0.00044	−0.00393	−0.0007	−0.0022	−0.00474	−0.0005
Njt: Local nuclear generation (MWh)	−0.00426	−0.00211	−0.00035	−0.00329	−0.00361	0	−0.0031	−0.00389	0.000073

. As noted earlier, we generally favor the 2SLS estimates, given that this approach is less sensitive to model misspecification, which is often a concern in situations where public data are limited. The adjusted R² values are 0.34 or lower, reflecting the relatively poor goodness of fit, due chiefly to the noisy energy price data and our inability to model the transmission congestion and line loss portions of the wholesale prices using a regression approach. The coefficient estimates for the fuel price index are around 0.35 for ISO-NE and PJM but below 0.1 for NYISO, reflecting the challenge of determining the type of generation on the margin in these markets and the weak correlation between wholesale electricity prices and fuel prices, particularly in NYISO.

The impact of the import of hydroelectric energy from Ontario into NYISO is sensitive to the estimation method. Specifically, 3SLS provides a coefficient estimate that is negative and significant, consistent with our expectations. However, 2SLS provides an estimate that is insignificantly different from zero. Finally, FIML suggests a positive estimate that is counterintuitive.

The import of electricity into ISO-NE from Quebec and New Brunswick significantly suppresses the wholesale price in New England. The impact is similar to local nuclear generation’s effect on the wholesale price in that market.

The coefficients on the variables representing solar plus wind generation are consistently negative, supporting the hypothesized existence of economic merit-order effects from renewables in all three markets. Yet the coefficient estimate on wind plus solar energy generation for ISO-NE was insignificantly different from zero when 3SLS was used. Over the entire estimation period, wind and solar generation met less than 4% of the demand overall. Thus, its impact may remain too small to enable precise estimates of its impact on wholesale prices. Perhaps the presence of energy companies involved in both renewable energy and traditional electricity generation is affecting price formation and somewhat reducing the merit-order effect, as hypothesized in [24].

An increase in (baseload) nuclear energy generation generally lowers the wholesale market prices, particularly using our favored 2SLS estimates. However, it is noteworthy that the coefficient estimates for the impact of nuclear generation on the PJM wholesale price was insignificantly different from zero in both the FIML and 3SLS model estimations.

A change in scheduled generation capacity outages has little impact on wholesale prices in NYISO when 3SLS or FIML is used as the estimation method. This might reflect the successful scheduling of maintenance outages during periods when the generation capacity has a lower value.

Inter-RTO dependence exists based on the coefficient estimates for the variables representing prices in a neighboring market. Changes in wholesale market prices in one market affect prices in the neighboring market (s). As a consequence, rising electricity imports from Canada at least into New England will reduce ISO-NE’s energy prices, which will in turn place downward pressure on prices in the other two markets.

Regardless of the estimation method, an increase in demand has a higher impact on wholesale prices in New England than in the other two regions.

The coefficient estimates for C_t range from 0.2 to 0.47 across markets and across estimation methods, thus indicating P_jt’s passthrough of the marginal carbon emissions cost. Using the 3SLS estimates as an example, the coefficient estimates for C_t are 0.46 for ISO-NE, 0.21 for NYISO and 0.33 for PJM. As 5.35 × 0.053 = 0.28, the passthrough of the marginal carbon emissions cost by NYISO’s energy prices is 0.21/0.28 = ~75%. In contrast, the extent of passthrough is 0.46/(8.14 × 0.053) = 107% for ISO-NE and 0.33/(6.87 × 0.053) = 91% for PJM.

Using the 3SLS estimates as an example, the coefficient estimates for D_jt show that a 1 MWh increase in hourly demand causes energy price increases of USD 0.0069/MWh for the relatively small ISO-NE market, USD 0.0015/MWh for NYISO, and USD 0.0009/MWh in the larger PJM market.

Because [38] measured wind generation in PJM as the percentage of the generation mix, it is difficult to compare their estimate of the merit-order effects of wind generation to ours, since we use the total MWh of renewable energy generation as our variable. The analysis of wind generation and prices in ISO-NE conducted in [39] is also difficult to benchmark against our results, since that study focused on how prices are impacted by errors in forecasting wind generation, whereas forecasting errors are not considered in our models.

4. Discussion

Using our regression results, we explore two topics: the likely impacts on prices in the three markets of various additional proposed renewable energy projects and the impacts of recent nuclear power plant retirements on prices.

4.1. The Short-Term Impact of Proposed Large-Scale Renewable Energy Projects on Prices

There are off-shore wind renewable energy projects proposed for the Northeast, which have been put on hold for various reasons. They include the following: (1) NYISO—1034 MW Sunrise Wind, located approximately 30 miles off Long Island; (2) ISO-NE—800 MW Vineyard Wind, located 15 miles south of Martha’s Vineyard; and (3) PJM—2800 MW Atlantic Shores Offshore Wind South, located off the New Jersey coast.

Consider a “generic” renewable energy project with a capacity of 1000 MW and a capacity factor of 0.5. For illustration, we will adopt the 2SLS estimation results. A simple calculation is shown in

Table 4. Impact of new 1000 MW renewable project on prices.

	ISO-NE	NYISO	PJM
Unit size	1000	1000	1000
Capacity factor	0.5	0.5	0.5
Average output per hour (MWh)	500	500	500
Change in USD/MWh price associated with a 1 MWh decline in nuclear generation	−0.00086	−0.00764	−0.00035
Hourly average change in price	−0.43	−3.82	−0.175
Average wholesale price	40.61	27.08	33.6
Percentage change	−1.1%	−14.1%	−0.5%

, estimating that such a project would decrease system prices by 1% in ISO-NE and 14% in NYISO and have a much smaller effect in the larger PJM market. A refined calculation would require assumptions regarding the siting of the project and the temporal profile of its generation output.

4.2. The Short-Term Impact of Nuclear Plant Retirements on Prices

In recent years, many aging nuclear power plants in the Northeast have been retired, including the following:

• NYISO: 2060 MW Indian Point Energy Center with Unit 2 in April 2020 and Unit 3 in April 2021;
• ISO-NE: 680 MW Pilgrim Nuclear Power Station in May 2019;
• PJM: 652 MW Oyster Creek Nuclear Generating Station in September 2018; 908 MW Davis-Besse Nuclear Power Station in May 2020; 1268 MW Perry Nuclear Power Plant in May 2021; and 1872 MW Beaver Valley Power Station May 2021.

To derive a rough estimate of the increase in wholesale electricity prices resulting from the retirement of a nuclear power plant, we assume the retirement of a “generic” 1000 MW nuclear plant with a 0.9 capacity factor. We employ the 2SLS estimates of the relationships between nuclear generation and prices in the New England and New York markets.

Table 5. Impact of retirement of a 1000 MW nuclear plant in ISO-NE and NYISO.

	ISO-NE	NYISO	PJM
Unit size	1000	1000	1000
Capacity factor	0.9	0.9	0.9
Average output per hour (MWh)	900	900	900
Change in USD/MWh price associated with a 1 MWh decline in nuclear generation	0.00426	0.00211	0.00035
Hourly average change in price	3.834	1.899	0.315
Average wholesale price	40.61	27.08	27.08
Percentage change	9.4%	7.0%	1.2%

shows that the retirement of the 1000 MW nuclear plant would have the direct effect increasing prices by about 9% in ISO-NE, 7% in NYISO, and slightly over 1% in the larger PJM market. There would be spillover effects in neighboring markets as well.

5. Conclusions

Our regression analysis affirms that energy policies to promote the use of renewable energy resources in one state in the Northeast can affect prices throughout the region.

Wholesale prices in the Northeast are influenced by prices in neighboring markets, although the correlation between prices between the New England and New York markets is far higher than between either of these two markets and the larger PJM market.

We see evidence that increases in renewable and nuclear generation, as well as hydroelectric imports from Canada (at least those into New England), lead to a short-run reduction in wholesale electricity prices in the three wholesale markets serving the Northeast, although certain estimation methods may yield weak or counterintuitive results, thus reflecting that precise estimates are difficult to obtain when applying regression modeling to the publicly available noisy data from these markets.

As the Northeast U.S. continues to increase its reliance upon price-reducing intermittent renewables, the “missing money” problem may continue. This problem arises when revenues through the energy market are not sufficiently high to support new investment in energy generation or storage. Capacity markets have previously been implemented in the PJM, NYISO, and ISO-NE markets to remedy the missing money problem [40]. Although these capacity markets have proven expensive, some means of ensuring long-term resource adequacy is likely to continue to be necessary.

We would be remiss had we failed to acknowledge the caveats of our paper. The first caveat is that data availability imposed some limits on the scope and precision of our analysis. Although prices in these markets are set on a 5 min basis, hourly data were used in our estimation, due to the absence of public 5 min data on generation by fuel type in the PJM market.

The second caveat is that a single market-wide indicative price was used to represent real-time prices in each of the three markets, although unique prices may be set for up to 17,000 nodes or locations in PJM, about 11,000 nodes in NYISO, and roughly 8000 in ISO-NE. Using a single price for each market, we were unable to model the full distribution of prices in different areas of each market and the impact of transmission congestion. However, modeling the full set of 36,000 locational prices in this region would have been extremely difficult, requiring detailed information on transmission congestion, power plant availability, generation offers by power plant, the scheduling of hydroelectric imports, and the network topology. Given our research objectives, any additional insights that would have been obtained did not appear to justify the additional efforts involved in such modeling.

The third caveat is that the weak correlation between wholesale hub prices and fuel costs introduces modeling challenges, particularly for NYISO. The fact that local renewable energy development remains fairly limited in the Northeast also led to challenges, preventing us from separately modeling solar and wind generation. It is also likely responsible for an insignificant 3SLS coefficient estimation on renewable energy generation in the ISO-NE market.

The final caveat is that we did not study the impacts of specific state-level energy policies. We instead opted to examine the evolution of the three interconnected markets, without regard to the specific policies and trends in resource cost that led to their energy transition. To meaningfully address this caveat, we would require a comprehensive integrated resource planning model that can assess the detailed impacts of a combination of state policies. However, such an undertaking was well beyond the intent and scope of our paper.

There are limitless opportunities to expand and improve upon the analysis presented here. National changes in energy policy will impact the provision of electricity in the Northeast, affecting the relationships estimated here and the need for updated analysis. If renewable energy becomes a larger share of the generation mix in the Northeast, the impacts of such resources may lead to more precise estimates of merit-order effects in the future. The impacts of specific power plant additions, power plant closures, load additions due to new data centers in the region, and changes in state energy policies could be studied for their impacts on changes in wholesale prices in neighboring, as well as local electricity markets. If more granular data and additional variables become available from these system operators in the future, their use could enable modeling that may provide new insights. Modeling price formation within zones inside each of the markets could provide further insights into regions where prices may be most sensitive to changes in the supply mix in the three markets.