Skip to main content

Does locational marginal pricing impact generation investment location decisions? An analysis of Texas’s wholesale electricity market

Abstract

Using data from Texas’s wholesale electricity market, we investigate the relationship between nodal prices and investment location decisions of utility-scale generation. We find some evidence that new investment arises in areas with recently elevated nodal prices. However, we find no evidence that new generation resources receive a nodal price premium post-entry as projected by the expectation of higher nodal prices. Further, a logit regression analysis suggests that the probability of natural-gas-fired generation investments tends to increase with expected nodal prices in peak hours. However, the estimated relationship is statistically and economically weak and sensitive to model specification. These findings suggest other factors are more important drivers than nodal prices of location decisions for utility-scale generation investments in Texas.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8

Notes

  1. 1.

    For a summary of the studies performed in several U.S. markets, see Neuhoff and Boyd (2011). For a broader summary of the trade-offs associated with nodal pricing, see Weibelzahl (2017).

  2. 2.

    Synapse Energy Economics (2006) compares generation investment in PJM to lagged LMPs and finds no evidence that generation investment occurred in regions with higher lagged LMPs. Brown and O’Sullivan (2019) document the spatial and temporal variation in the value of solar power across the U.S. using nodal price data. However, unlike our analysis which focuses on observed investment, the authors consider simulated solar PV investment.

  3. 3.

    We utilize 15-min LMPs in our analysis because of computational tractability and 5-min prices are nearly identical within a 15-min interval.

  4. 4.

    See PUCT Subst. R. 25.195(c): http://puc.texas.gov/agency/rulesnlaws/subrules/electric/25.195/25.195.pdf.

  5. 5.

    Potential market changes to implement the recognition of marginal losses in dispatch decisions were debated and ultimately rejected in PUCT Project No. 47,199: Project to Assess Price-Formation Rules in ERCOT’s Energy-Only Market, https://interchange.puc.texas.gov/Search/Filings?ControlNumber=47199.

  6. 6.

    In addition, resource nodes that were created as the result of entry during our sample do not have historical LMP data that we can utilize to understand how lagged LMPs relate to entry decisions near a given node.

  7. 7.

    Accounting for a plant’s random availability vastly complicates the subsequent discussion without the benefit of additional insights, see Woo et al. (2019) for the effect of a plant’s availability on a generation plant’s profitability.

  8. 8.

    In addition, it is possible that \( E\left( {\pi_{t} } \right) \) increases with the expected nodal price variance because rising electricity price volatility implies a higher likelihood of high price hours (Woo et al. 2016). However, the high correlation of expected LMP levels and variance makes it difficult to include both covariates. We discuss the robustness of our results to the inclusion of expected nodal price variance measures below.

  9. 9.

    Our empirical results throughout the paper are robust to decomposing the net demand variable into the individual demand and observed generation by technology variables.

  10. 10.

    We utilize observed values on net demand and Henry Hub prices to estimate LMP levels out-of-sample. This implicitly assumes that firms have perfect foresight on these variables. Constructing forecasts of these variables will introduce additional variability in the expected LMP measures. This would likely elevate the variability in our LMP estimator leading us to be less likely to conclude that LMPs drive investment location decisions. In the Conclusion, we stress the importance of future research that relaxes the perfect foresight assumption.

  11. 11.

    Because our analysis is reduced-form in nature, we are unable to explicitly model the impact of generation capacity additions on expected LMPs post-entry. In the conclusion, we stress the importance of future research that establishes a structural model that permits such counterfactual simulations.

  12. 12.

    We also estimate a forward-looking average LMP in super peak hours where super peak reflects weekday hours between 12:00 PM and 8:00 PM. Our on-peak coefficient results are robust to this alternative specification.

  13. 13.

    In Sect. 4.1, we noted that a firm’s expected operating profits will depend on expected fuel input prices. However, in our annual discrete entry regressions, we are identifying off of the variation across nodes within a weather-zone and year. Consequently, we do not include Henry Hub gas prices as a regressor as there is limited-to-no variation in expected natural gas prices faced by generators across nodes within a weather-zone and year. ERCOT’s weather zones segment Texas into 8 regions, for details see http://www.ercot.com/news/mediakit/maps.

  14. 14.

    We are unable to include node fixed effects because this would limit the variation we are identifying off of to the node-by-year level. This often results in multi-collinearity as there is limited variation in entry of a specific technology within a node-year.

  15. 15.

    We also tested if the results were impacted by including existing coal generation as a separate regressor. This coefficient was statistically insignificant in all specifications and the other coefficients were largely unaffected.

  16. 16.

    We observe a weak negative correlation between existing nearby capacity and LMPs ranging from − 0.003 to − 0.08.

  17. 17.

    The HIFLD (2020) data reflects the transmission network as of 2019. This creates limitations to the degree that the transmission network changed over our sample. To ensure our results are robust, we consider alternative definitions of nearby nodes based on geographical distance.

  18. 18.

    Figure 10 in the “Appendix” presents the Lorenz curve by year as an alternative representation of the dispersion of LMPs in our sample. The dispersion trends in Fig. 10 reflect those presented in Fig. 2.

  19. 19.

    The LMP contour plots in this figure were created using code in Elhabr (2018)

  20. 20.

    The other category includes batteries/flywheel storage (11), hydro (1), biomass (1), coal (2), and other natural gas (54). The coal units are additions to existing facilities. Other gas represents small assets (< 20 MWs) with technologies such as landfill gas, steam turbines, and one large facility with natural gas compressed air storage.

  21. 21.

    See Table 7 in the “Appendix” for a detailed summary of capacity additions by year and technology in MWs.

  22. 22.

    We also considered alternative lagged structures such as one and 2 years to categorize the pricing tiers and find that our results are robust.

  23. 23.

    Similar results arise if we focus only on investments between the period 2012–2019.

  24. 24.

    Recall, we utilize 2011–2012 data to forecast forward for the period 2013–2019, 2011–2013 data to forecast the period 2014–2019, and so on until we are utilizing 2011–2017 data to forecast RTM nodal prices for the period 2018 and 2019.

  25. 25.

    We undertake a log transformation on the existing capacity measures to account for the highly skewed nature of the data. We adjust existing capacity by adding a constant equal to 1 to deal with the zero observations. However, because this can bias the regression results under certain circumstances (Bellégo and Pape 2020), we consider other transformations such as square root, cubic root, and including a dummy equal to 1 when existing capacity is positive and zero otherwise. We find that our results are robust to these alternative transformations.

  26. 26.

    Liu et al. (2016) and Zarnikau et al. (2019a) find similar henry hub coefficients in the range of 7–9. While there are a handful of negative henry hub coefficients, they are not statistically significant and often arise at nodes in industrial regions where idiosyncratic local consumption behavior is more likely to drive nodal price patterns.

  27. 27.

    From Table 3, \( \sigma_{NetDemand} = 2497.65 \) such that a one standard deviation change in net demand gives \( 2497.65 \times 0.052 \approx 129.88 \).

  28. 28.

    It is possible that expected LMP variance can impact investment location decisions. However, high correlations between our average LMP level and variance measures raises concerns of multi-collinearity. As a robustness check, we include expected LMP variance measures in our logit regressions with average expected LMPs across all hours. These covariates are systematically statistically insignificant.

  29. 29.

    The variability in the number of observations across technologies arises from the fact that we have weather zone and year fixed effects to control for important regional and time-varying factors that drive investment decisions. Consequently, our analysis identifies off of variation in entry decisions within a year and weather zone. If there is a year or weather zone with no investment in a certain generation technology, then the nodes in this zone-year are dropped from our sample due to multi-collinearity.

  30. 30.

    The missing existing wind and solar capacity variables are arising because CT units have no existing capacity of these technologies near their nearest node.

  31. 31.

    More specifically, we match the new generation assets via the coordinates from the EIA’s Form 860 data (EIA 2019a) to the transmission infrastructure in the HIFLD (2020) database. We document that CCGTs and CTs are more likely to locate near 345 kV and 138 KV lines, respectively.

  32. 32.

    For a detailed analysis of the impacts of solar PV on wholesale prices, see Bushnell and Novan (2018).

  33. 33.

    More specifically, regressing the nodal solar irradiance measure on weather zone and year fixed effects yields an R-squared of 0.90.

  34. 34.

    To investigate these two channels further, we interact the peak and off-peak expected LMP covariates with a dummy that equals one only if there is existing wind capacity nearby. We find the positive peak LMP coefficient is stronger at nodes without existing nearby wind capacity providing some support to conjecture that new wind facilities target nodes with higher expected LMPs in peak hours. However, a structural model is needed to fully disentangle these two effects. We emphasize the importance of such an extension in the Conclusion.

  35. 35.

    More specifically, we evaluate all of the covariates at their observed value and consider a one standard deviation increase in the value of the covariate under question. The statistical significance provided indicates the significance of the coefficients under consideration. We also computed traditional AMEs of an incremental (one-unit) increase that computes standards errors via the delta method. This yields analogous intuition and statistical significance, although the magnitudes of the effects decrease.

  36. 36.

    These numbers are calculated by using the observed number of assets by technology in Table 1, divided by 160.

  37. 37.

    Utilities like Pacific Gas and Electric (PG&E) and Bonneville Power Administration (BPA) have been seeking non-wire solutions based on DR and DSM (Sreedharan et al. 2012; Woo et al. 2014; E4 The Future 2018).

  38. 38.

    For example, in 2014, ERCOT approved a transmission line to alleviate local congestion in Houston, a move that was opposed by several generation companies (Tiernan 2015).

References

  1. Andrade, J., & Baldick, R. (2017). Estimation of transmission costs of new generation. University of Texas Energy Institute. White Paper UTEI/2016-09-2.

  2. Baughman, M., Siddiqi, S., & Zarnikau, J. (1997). Advanced pricing in electrical systems, part 1: Theory. IEEE Transactions on Power Systems, 12(1), 489–495.

    Article  Google Scholar 

  3. Bellégo, C., & Pape, L.-D. (2020). Dealing with logs and zeros in regression models. Série des Documents de Travail. Retrieved from: https://ssrn.com/abstract=3444996.

  4. Brattle Group (2018). Estimation of the market equilibrium and economically optimal reserve margins for the ERCOT region.

  5. Brown, P., & O’Sullivan, F. (2019). Spatial and temporal variation in the value of solar power across united states electricity markets. MIT Center for Energy and Environmental Policy Research. Working Paper Series, CEEPR WP 2019 – 011.

  6. Bushnell, J., & Novan, K. (2018). Setting with the sun: The impacts of renewable energy on wholesale power prices. UC Davis Energy Economics Program Working Paper 020.

  7. CRA (2008). Update on the ERCOT Nodal Market Cost-Benefit Analysis. Charles Rivers Associates International. Prepared for: Public Utility Commission of Texas.

  8. E4 The Future (2018). Non-Wires Alternatives. Case Studies from Leading U.S. Projects.

  9. EIA (2019a). 2018 Early Release Data. Form EIA-860 Detailed Data with Previous Form Data (EIA-860A/860B). U.S. Energy Information Administration. Retrieved from July 9, 2019.

  10. EIA (2019b). Table 6.5. Planned U.S. Electric Generating Unit Additions. U.S. Energy Information Administration. Retrieved July 9, 2019.

  11. EIA (2019c). Texas Electricity Profile 2017. State Electricity Profiles. U.S. Energy Information Administration. Retrieved from December 9, 2019.

  12. Elhabr, T. (2018). Re-Creating a Voronoi-Style Map with R. Retrieved https://tonyelhabr.rbind.io/post/recreate-voronoi-map-r/. Accessed 25 Aug 2019.

  13. ERCOT (2019). Capacity Changes by Fuel Type Charts, October 2019. Retrieved November 29, 2019, from http://www.ercot.com/gridinfo/resource.

  14. FERC (2014). Price formation in organized wholesale electricity markets. Docket No. AD14-14-000. Federal Energy Regulatory Commission.

  15. Green, R. (2007). Nodal pricing of electricity: How much does it cost to get it wrong? Journal of Regulatory Economics, 31(2), 125–149.

    Article  Google Scholar 

  16. HIFLD (2020). Electric power transmission lines. Homeland Infrastructure Foundation-Level Data. Access: April 25, 2020.

  17. Liu, Y., Woo, C. K., & Zarnikau, J. (2016). Ex post payoffs of a tolling agreement for natural gas-fired generation in Texas. Journal of Energy Markets, 9(1), 21–45.

    Article  Google Scholar 

  18. Neuhoff, K., & Boyd, R. (2011). International experiences of nodal pricing implementation. Climate Policy Initiative.

  19. Northbridge Group. (2017). Economic equilibrium and reliability in ERCOT. Retrieved from http://www.northbridgegroup.com/publications/Economic_Equilibrium_and_Reliability_in_ERCOT.pdf. Accessed 10 Sep 2019.

  20. NREL (2020). Physical solar model V3. National renewable energy laboratory. Retrieved May 10, 2020, from https://developer.nrel.gov/docs/solar/nsrdb/psm3_data_download/.

  21. Potomac Economics, (2017). 2016 State of the Market Report for the ERCOT Electricity Markets. Independent Market Monitor for ERCOT.

  22. Potomac Economics, (2019). 2018 State of the Market Report for the ERCOT Electricity Markets. Independent Market Monitor for ERCOT.

  23. Schweppe, F., Caramanis, M., Tabors, R., & Bohn, R. (1988). Spot pricing of electricity. Norwell, MA: Kluwer.

    Book  Google Scholar 

  24. Sen, A. (1973). On economic inequality. Oxford: Clarendon Press.

    Book  Google Scholar 

  25. Spees, K., Newell, S., & Pfeifenberger, J. (2013). Capacity markets—Lessons learned from the first decade. Economics of Energy & Environmental Policy, 2(2), 1–26.

    Article  Google Scholar 

  26. Sreedharan, P., Miller, D., Price, S., & Woo, C. K. (2012). Avoided cost estimation and cost-effectiveness of permanent load shifting in California. Applied Energy, 96, 115–121.

    Article  Google Scholar 

  27. Stern, L. (1998). Merchant power plants: Competing for financing in a deregulated industry. The Journal of Structured Finance, 4(3), 47–55.

    Article  Google Scholar 

  28. Synapse Energy Economics (2006). LMP Electricity Markets: Market Operations, Market Power, and Value for Consumers. Prepared for: American Public Power Association.

  29. Tiernan, T. (2015). PUCT sees need for houston import project; Calpine, NRG Disagree. Power Grid International. Retrieved from http://www.power-grid.com/2015/09/30/puct-sees-need-for-houston-import-project-calpine-nrg-disagree/#gref. Accessed 10 Sep 2019.

  30. Tierney, S. and E. Kahn (2007). A cost-benefit analysis of the new york independent system operator: The initial years. ANALYSIS group report.

  31. Weibelzahl, M. (2017). Nodal, zonal, or uniform electricity pricing: How to deal with network congestion. Frontiers in Energy, 11, 210–232.

    Article  Google Scholar 

  32. Woo, C. K., Horowitz, I., Zarnikau, J., Moore, J., Schneiderman, B., Ho, T., et al. (2016). What moves the ex post variable profit of natural-gas-fired generation in California? The Energy Journal, 37(3), 29–57.

    Google Scholar 

  33. Woo, C. K., Milstein, I., Tishler, A., Zarnikau, J. (2019). A wholesale electricity market design sans missing money and price manipulation. Energy Policy, 134, Article 110988.

  34. Woo, C. K., Sreedharan, P., Hargreaves, J., Kahrl, F., Wang, J., & Horowitz, I. (2014). A review of electricity product differentiation. Applied Energy, 114, 262–272.

    Article  Google Scholar 

  35. Zarnikau, J., Woo, C. K., & Baldick, R. (2014). Did the Introduction of a Nodal Market Structure Impact Wholesale Electricity Prices in Texas (ERCOT) Market? Journal of Regulatory Economics, 45, 194–208.

    Article  Google Scholar 

  36. Zarnikau, J., Woo, C. K., Zhu, S., & Tsai, C. H. (2019a). Market price behavior of wholesale electricity products: Texas. Energy Policy, 125, 418–428.

    Article  Google Scholar 

  37. Zarnikau, J., Zhu, S., Woo, C. K., & Tsai, C. H. (2019b). Texas’s operating reserve demand curve’s generation investment incentive. Energy Policy, Article 111143.

Download references

Acknowledgements

We would like to thank Jenet Dooley, Yiang Guo, and Lucy Zhu for their excellent assistance, as well as numerous helpful comments and suggestions from anonymous referees. This project was supported by the Government of Canada’s Canada First Research Excellence Fund under the Future Energy Systems Research Initiative.

Author information

Affiliations

Authors

Corresponding author

Correspondence to David P. Brown.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix

Appendix

A: Supplementary figures and tables

See Figs. 9, 10, 11, 12 and Tables 7, 8 and 9.

Fig. 9
figure9

Nodes in sample (N = 160)

Fig. 10
figure10

Lorenz curve of LMP dispersion by year

Fig. 11
figure11

Capacity additions (MWs) by pricing tiers—nearest node by geographical distance (2012–2021)

Fig. 12
figure12

Capacity additions (MWs) by pricing tiers and year—nearest node by geographical distance

Table 7 New generation facilities by year and fuel source (MW)
Table 8 Average LMPs at nodes with and without recent entry—nearest node (1-year lag)
Table 9 Average LMPs at nodes with and without recent entry—nearest node by geographical distance (3-year lag)

B: Logit regression robustness

Tables 10 and 11 present the results of our discrete logit regression with the spatial distance measure that links entry that occurs to the nearest two nodes. We will highlight the statistically significant changes from those presented in Tables 4 and 5. In Table 10 columns (3) and (4), log existing wind capacity is now negative and statistically significant. This is consistent with CT units locating to at nodes away from existing wind assets that suppress expected LMPs. In columns (6) and (8), the expected average peak and off-peak LMP variables are no longer statistically significant. However, they maintain the same sign.

Table 10 Nodal entry logit regression by technology (CT, CCGT)—nearest two nodes
Table 11 Nodal entry logit regression by technology (solar, wind)—nearest two nodes

In Table 11 columns (1) and (2), the existing solar variable is no longer statistically significant. In addition, the average expected off-peak LMP variable is no longer statistically significant in column (2). However, these variables maintain the same sign. The average expected peak LMPs is now statistically significant and positive in columns (6) and (8). These results strengthen the positive relationship between expected peak LMPs and the likelihood of entry of a wind asset. Looking at columns (5)–(8), in addition to existing wind capacity being statistically significant, wind speed is now statistically significant.

Tables 12 and 13 present the results of our logit regression, removing the final year of our sample. In column (4), the coefficient on average expected peak LMPs is no longer statistically significant. As noted in the main text, this emphasizes that the relationship between CT entry location decisions and expected peak LMPs is sensitive to model specification. The CCGT results are identical to those included in the main analysis as no CCGT entry was observed in 2018.

Table 12 Nodal logit regression by technology (CT, CCGT)—nearest node, excluding 2018
Table 13 Nodal logit regression by technology (solar, wind)—nearest node, excluding 2018

Table 13 considers solar and wind assets. In columns (1)–(4), the existing solar capacity remains positive, but is no longer statistically significant. While the precise quantitative numbers have changed in Table 13, the sign and statistical significance is unchanged for the remainder of the covariates.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Brown, D.P., Zarnikau, J. & Woo, CK. Does locational marginal pricing impact generation investment location decisions? An analysis of Texas’s wholesale electricity market. J Regul Econ 58, 99–140 (2020). https://doi.org/10.1007/s11149-020-09413-0

Download citation

Keywords

  • Electricity
  • Regulation
  • Entry
  • Locational marginal pricing

JEL Classification

  • L11
  • L51
  • L94
  • Q41
  • Q48