1. Data
| Series | Frequency | Date Range | Use in estimation | Source |
|---|
| Stony Point and Ramapo scheduled volumes, available volumes, and reported maximum volume (Dth/d) | Daily | 2011-01-06 – 2026-09-05 (Stony Point from 2014-02-14) | Capacity Kₜ and scheduled flow Qₜ in (1) | S&P CapIQ |
| Algonquin Citygate and Henry Hub spot indices ($/MMBtu) | Daily, trade days | 2005-01-03 – 2026-07-24 | Dependent variable and upstream control in (1) | S&P CapIQ |
| Zonal and MA-average day-ahead and real-time LMPs, on / off / all hours ($/MWh) | Weekly averages | 2019-06-03 – 2026-08-24 | Dependent variable in (2), MA_AVG_RT_AllHours | ISO-NE |
| MA net generation by fuel — all fuels, natural gas, petroleum liquids (thousand MWh) | Monthly | 2001-01 – 2026-04 | Dependent variable in (3) | EIA |
| MA retail electricity sales by sector — residential, commercial, industrial, transportation, other (million kWh) | Monthly | 2001-01 – 2026-06 | Load L in (2)–(3) | EIA |
| Natural gas deliveries to MA electric power companies (MMcf) | Monthly | 2001-01 – 2026-06 | Generator gas volumes in the wholesale distribution | EIA |
| MA natural gas deliveries by end use (residential, commercial, industrial, vehicle, electric power), (MMcf) | Monthly | 1989-01 – 2026-06 | LDC volumes (5) | EIA |
| MA wholesale heating oil price ($/gal) | Weekly, heating season (Oct–Mar) only | 2013-10-07 – 2026-03-30 (343 obs.) | Cross-fuel substitution control in eq. (3), converted to $/MMBtu and averaged to monthly | EIA |
| TMAX / TMIN for a four-station New England composite | Daily | 1936-01-01 – 2026-08-27 | (1) - (3) | NOAA |
2. Econometric Framework
We develop three models that build on previous estimates to incorporate the effects of natural gas, electricity, and fuel oil. We estimate each model across the full data set and then in the winter (November-March) subsample to isolate Project Beacon's potential impact on Massachusetts.
2.1 Capacity elasticity of the citygate price
(1) estimates the elasticity of the local citygate price relative to the Stony Point compressor station capacity:
log PAGTt = α + θ log PHHt + β log Kt + δ log Qt + γ log(1 + HDDt) + XtΦ + ε1,t (1)
where PAGTt is the Algonquin Citygate spot price, PHHt the Henry Hub national benchmark, Kt the reported maximum Stony Point capacity, and Qt the scheduled flow (both in Dth/d), and Xt a matrix of month and year fixed effects (seasonal controls). The parameter of interest is β ≡ ∂ log PAGT / ∂ log K.
Estimation is in log levels rather than first differences because Beacon is a permanent level shift in capacity, whereas a differenced specification identifies the response to a transitory change. We estimate these regressions with HAC standard errors to account for this.
Capacity and scheduled flow enter separately because Project Beacon changes capacity, and capacity variation is driven largely by maintenance and outage scheduling and is more plausibly exogenous to the local price than demand-driven variation in flow.
2.2 Pass-through to wholesale power
Next (2) predicts the price of the average real-time power price across Massachusetts, where the goal is to estimate the change in power prices relative to the change in pipeline capacity:
log PRTw = α + φ log PAGTw + λ log Lw + γ log(1 + HDDw) + XwΦ + ε2,w (2)
(2) is estimated on weekly averages due to data availability, where PRTw is the Massachusetts average real-time all-hours locational marginal price, and Lw is Massachusetts retail electricity sales. Similar to (1), Xw is a matrix of temporal and seasonal controls. The elasticity of power prices with respect to pipeline capacity is ∂ log PRT / ∂ log K = φβ.
2.3 Oil displacement
Lastly, (3) estimates monthly fuel oil generation. Fuel oil is the fuel of last resort in New England winters. The response of oil-fired dispatch to the citygate gas price is estimated controlling for the price of wholesale heating oil:
log(Goilm) = α + ψ log PAGTm + ξ log Poilm + λ log Lm + γ log(1 + HDDm) + MmΦ + τt + ε3,m (3)
Mm carries monthly fixed effects and τt is a linear trend that absorbs the structural retirement of the New England oil fleet–a common specification used in time series econometrics.
Merit-order switching is a claim about relative prices, indicating that if oil-fired units simply burn whichever fuel is cheaper per MMBtu, then only the gas-oil ratio should drive substitution, and a rise in the gas price should move oil generation by exactly the same amount as an equivalent fall in the oil price. That symmetry is testable, and the data reject it (p = 0.0013). A 10% increase in the citygate price is associated with roughly 21% more petroleum burn; a 10% decline in the oil price buys about 10% more. Oil generation responds to gas scarcity at twice the rate it responds to oil prices—suggesting that variation in oil generation is more sensitive to natural gas prices rather than to oil prices.
2.4 Counterfactual construction and ratepayer incidence
To estimate a counterfactual change in natural gas prices relative to capacity, we multiply the observed price by the estimated capacity effect:
PAGT,cft = PAGTt · exp ( β ln [ (Kt + ΔK) / Kt ] ) (4)
Gas ratepayers pay the citygate price through the supply component of an LDC bill and electric ratepayers pay the wholesale power price through the supply component of an electricity bill:
ΔSgas = Σm ∈ W ( PAGTm − PAGT,cfm ) · VLDCm · η (5)
ΔSelec = Σm ∈ W PRTm [ 1 − exp ( φβ ln ( (Km + ΔK) / Km ) ) ] · Lm · η (6)
where W is the set of winter months in a season, VLDCm is measured gas delivered to residential and commercial customers, Lm is measured Massachusetts retail electricity sales, and η ∈ [0, 1] is the long-run pass-through of wholesale prices to retail tariffs.
3. Estimation Samples
| Model | Frequency | Sample | N | R² | Note |
|---|
| (1) | Daily | 2014-02-14 – 2026-07-24 | 3,095 | 0.701 | Stony Point data begin Feb 2014; the gas price series ends 24 Jul 2026 |
| (1) winter | Daily | Nov–Mar within the same span | 1,239 | 0.743 | As above |
| (2) | Weekly | 2019-06-03 – 2026-08-24 | 366 | 0.918 | The ISO-NE weekly price series begins in June 2019 |
| (2) winter | Weekly | Nov–Mar within the same span | 149 | 0.941 | As above |
| (3) | Monthly | 2014-02 – 2026-04 | 144 | 0.594 | Generation data end Apr 2026 |
| (3) winter | Monthly | Nov–Mar, with the heating oil control | 61 | 0.717 | The heating oil survey covers Oct–Mar only — present in 62 of 62 winter months, against 82 of 147 year-round |
| Incidence panel | Monthly, winter | 62 winter months, 2014-02 – 2026-03 | — | — | Both channels complete in 7 winters, 2019/20 through 2025/26; earlier winters are gas-only |
4. Estimated Parameters
Each reported coefficient is the elasticity of the dependent variable with respect to the driver. Standard errors are HAC with an Andrews (1991) data-selected Bartlett bandwidth.
| Model | Sample | Parameter | Estimate | HAC s.e. | t | p | HAC lags |
|---|
| (1) | Full | β (log K) | −0.705 | 0.194 | −3.64 | 0.0003 | 34 |
| (1) | Winter | β (log K) | −1.156 | 0.485 | −2.39 | 0.0171 | 18 |
| (2) | Full | φ (log P AGT) | +0.718 | 0.030 | +23.66 | <0.0001 | 6 |
| (2) | Winter | φ (log P AGT) | +0.737 | 0.046 | +15.98 | <0.0001 | 5 |
| (3) | Full | ψ (log P AGT) | +1.091 | 0.264 | +4.13 | <0.0001 | 5 |
| (3) | Winter | ψ (log P AGT) | +2.137 | 0.287 | +7.45 | <0.0001 | 4 |
Specification tests
| Restriction | Sample | Statistic | p | Result |
|---|
| θ = 1 — does the model behave as a log-basis spread? | Full | 1.820 | 0.177 | Not rejected |
| θ = 1 | Winter | 0.147 | 0.702 | Not rejected |
| β(log K) = −β(log Q) — is utilization a valid restriction? | Full | 0.277 | 0.599 | Not rejected |
| β(log K) = −β(log Q) | Winter | 1.359 | 0.244 | Not rejected |
| ψ = −ξ — is oil dispatch symmetric? | Winter | 10.331 | 0.0013 | REJECTED |
5. Elasticities
Applying the winter parameters at winter mean prices to an average capacity of 1,609,537 Dth/d:
| ΔK | Citygate, winter mean price | Citygate, winter p90 price | MA real-time power | MA oil burn |
|---|
| +100 MMcf/d | −$0.46 / MMBtu | −$1.00 / MMBtu | −$2.90 / MWh | −13.9% |
| +300 MMcf/d (Beacon) | −$1.23 / MMBtu | −$2.66 / MMBtu | −$7.94 / MWh | −34.5% |
Composite winter elasticities with respect to capacity: citygate −1.156, wholesale power −0.852, oil burn −2.471. The 95% confidence interval for the +300 MMcf/d citygate effect at winter-mean prices ranges from −$2.08 to −$0.24 per MMBtu. The year-round figures, at a sample mean citygate price of $4.53, are −$0.19 and −$0.52 per MMBtu.
Sensitivity of savings and pass-through assumptions
The per-winter total ranges from $181.6M in 2023/24 to $628.9M in 2025/26; the largest in the sample is the most recent, at $517.1M in 2021/22 and $452.8M in 2024/25. Six earlier winters (2013/14 through 2018/19) have only a gas component because the weekly power price series begins in June 2019.
| Pass-through η | Gas customers $M | Electric customers $M | Ratepayer total $M |
|---|
| 0.60 | 120.4 | 99.5 | 219.9 |
| 0.70 | 140.5 | 116.1 | 256.5 |
| 0.85 | 170.6 | 141.0 | 311.5 |
| 1.00 (reported) | 200.6 | 165.8 | 366.5 |