The Impact of Local Law 87 on Building Energy Efficiency: Evidence from NYC Multifamily Buildings
Yu Qiao and Qianru Zhou
Introduction
Buildings account for approximately 70% of New York City’s greenhouse gas emissions, making the sector the primary target for municipal climate policy. To address this, the city enacted Local Law 84 (LL84) and Local Law 87 (LL87), which together serve as the foundational pillars of New York City’s long-term climate transition. While LL84 focuses on annual energy benchmarking to promote transparency, LL87 requires periodic technical audits and operational “tune-ups” to drive direct performance improvements.
This study evaluates whether addressing informational frictions and operational inertia through these local laws yields a measurable reduction in energy consumption. We test the hypothesis that providing owners with technical data and requiring low-cost system optimizations under LL87 has a detectable impact on building energy performance. If newly audited buildings demonstrate significant reductions in site energy use intensity compared to those awaiting compliance, it would suggest that informational and operational barriers are primary hurdles to decarbonization. Conversely, a null result would imply that audits and retro-commissioning alone are insufficient to drive meaningful improvement in building energy performance, signaling a need for the stronger financial incentives or strict performance standards, such as the more recent Local Law 97.
Policy Background
The Greener, Greater Buildings Plan
In 2009, New York City enacted the Greener, Greater Buildings Plan (GGBP) to target energy use in the city’s largest properties. Under the plan, Local Law 84 (LL84) established a mandatory benchmarking and disclosure program. By requiring buildings over 50,000 square feet to report annual energy use, LL84 aimed to motivate voluntary improvements through transparency.
Along LL84, the city enacted Local Law 87 (LL87) to move owners toward more direct action. Unlike the purely informational nature of benchmarking, LL87 requires a two-part intervention for every 10 years. The first part is energy audit, which is a diagnostic requirement where professionals identify Energy Conservation Opportunities (ECOs). This removes the information barrier by providing owners with a clear “diagnosis” of their building’s inefficiencies. The second part is retro-commissioning (RCx).Unlike the audit, RCx mandates action. Owners must optimize existing mechanical systems to ensure they perform as originally intended (e.g., repairing leaky valves or calibrating sensors). This addresses operational neglect by forcing “no-cost/low-cost” repairs.
Identification Strategy: A Natural Experiment
The implementation of LL87 provides a natural experiment for evaluating the causal impact of mandatory audits and retro-commissioning. Rather than allowing for self-selection into treatment, the city established a staggered 10-year compliance cycle where a building’s deadline is determined exogenously by the final digit of its Borough-Block-Lot (BBL) number.
Because the last digit is uncorrelated with building characteristics, energy performance, or owner intent, the BBL assignment mechanism provides as-if random variation in the timing of the policy intervention. This allows us to exploit the staggered rollout using a difference-in-differences design, comparing the energy trajectories of “treated” buildings against a control group of properties yet to reach their assigned compliance year. This framework effectively isolates the policy’s impact from endogenous factors like voluntary green investments or management quality.
Data description
Data Sources and Analytic Sample
This study integrates two primary administrative datasets publicly available via the NYC Open Data portal to construct a comprehensive longitudinal panel. Annual energy performance metrics are derived from the Local Law 84 (LL84) Benchmarking database, which provides building-level, self-reported annual consumption data for electricity, fuel oil, and steam. Information regarding policy compliance and intervention timing is sourced from Local Law 87 (LL87) submissions, which record the specific calendar years in which buildings completed their mandated energy audits and retro-commissioning. Together, these sources cover the population of large buildings in New York City—typically those exceeding 50,000 square feet—including commercial, residential, and institutional properties.
The unit of observation is the building-year, with each record representing a property’s energy profile for a specific calendar year. For the final analytic sample, we construct a panel spanning 2011 through 2018, retaining buildings with valid observations over multiple years. This longitudinal structure facilitates a robust within-building comparison of energy performance before and after audit compliance, providing the necessary statistical framework to identify the impact of LL87 while controlling for time-invariant building characteristics.
Key Variables and Data Cleaning Protocol
The focal outcome measure for this analysis is Site Energy Use Intensity (Site EUI), calculated as the total annual energy consumed at the building site normalized by its gross floor area (kBtu/ft²). Site EUI serves as the primary dependent variable because it provides a direct reflection of operational energy efficiency, isolating the performance of the building envelope and internal systems from upstream utility distribution losses. This normalization ensures that energy performance can be meaningfully compared across properties of varying sizes, providing an accurate proxy for building-level operational quality.
To mitigate the inherent noise of self-reported LL84 benchmarking data, we implemented a data-cleaning protocol. First, observations with missing consumption or implausible Site EUI values, defined by the NYC DOB standard of below 5 or above 1,000 kBtu/sq ft, were excluded as likely data entry errors. Second, entries lacking valid or standardized Borough-Block-Lot (BBL) identifiers were removed. This standardization is critical for cross-dataset integration and the precise attribution of LL87 treatment timing across the longitudinal panel.
Descriptive statistics
Sub-Sample Selection
Property type is a critical dimension in energy-performance research, as operational energy demand and baseline Energy Use Intensity (EUI) vary substantially across building categories. Distinct sectors, such as commercial, institutional, and residential, exhibit fundamentally different operating schedules, fuel mixes, and technical retrofit opportunities. Consequently, a pooled analysis of the entire benchmarking population would introduce significant unobserved heterogeneity, risking the conflation of divergent energy behaviors and masking the true impact of the policy intervention. To ensure the internal validity of our causal estimates, we narrow our analytic focus to a single, homogeneous property class.
Dominance of the Multifamily Sector
Multifamily residential buildings represent the largest and most data-rich subset of the New York City building stock subject to Local Law 84. As shown in Table 1, this category overwhelmingly dominates the benchmarking universe with 18,553 unique Borough-Block-Lot (BBL) identifiers—a count that far exceeds any other property type.
By focusing exclusively on multifamily properties, we maximize the statistical power of our longitudinal panel. This high density of observations is essential for credible event-study analysis and difference-in-differences estimation, as it provides a robust number of “treated” and “control” units across every year of the staggered LL87 rollout.
| Property Type | Number of Buildings |
|---|---|
| Multifamily Housing | 18553 |
| Office | 2534 |
| K-12 School | 1564 |
| Other | 688 |
| Non-Refrigerated Warehouse | 598 |
| Hotel | 512 |
| Retail | 371 |
| Mixed Use Property | 309 |
| College/University | 291 |
| Manufacturing/Industrial Plant | 269 |
| Note: | |
| Building counts are based on distinct Borough-Block-Lot (BBL) identifiers for each property type. |
Trends in Energy Use Over Time
The temporal trajectory of average log Site EUI for the multifamily
sector exhibits a pattern of high inter-annual variability rather than a
linear decline. The series begins with a significant fluctuation,
characterized by a sharp drop from its initial level to a series trough
in the second year of the study. This decline was immediately offset by
a rapid and sustained increase, leading to a prominent peak in the
middle of the observation period.
Following this peak, the data reveals a multi-year period of consistent energy intensity reduction. During this interval, average log Site EUI trended downward, reaching a secondary low toward the end of the decade. However, this downward trend did not persist through the final year of the study, as the series concluded with a notable uptick. Overall, the descriptive evidence suggests that while the sector achieved periods of efficiency gains, these improvements were interspersed with cyclical increases, leaving the final energy intensity higher than its recent minimum but below the starting point in 2011.
Unbalanced Panel Structure in LL84 Data
A key feature of the LL84 benchmarking dataset is its highly unbalanced
panel structure. The number of multifamily BBLs reporting each year
increases substantially over time. Reporting grows from roughly
9,000–9,500 buildings in 2011–2015 to more than 14,000 in 2017 and over
16,000 in 2018. This pattern reflects expanding compliance, improvements
in data collection, and the entry of newly covered properties into the
reporting system. While this growth is expected, it also means that the
set of buildings represented in early years is not the same as in later
years, complicating year-to-year comparisons of average energy use.
The table summarizing years of available data per BBL underscores the extent of this imbalance. Only 5,726 multifamily buildings (30.9%) report in all eight years of the study window, while the majority appear only intermittently: there are about 45 percent of multifamily buildings have just one or two years of data from 2011 to 2018. Very few buildings have three to six years of observations. This distribution shows that most of the dataset consists of short, fragmented reporting histories, with buildings entering and exiting the sample frequently.
| Number of Years of Data | Number of BBLs | Percent of BBLs |
|---|---|---|
| 1 | 3224 | 17.4% |
| 2 | 5185 | 27.9% |
| 3 | 565 | 3.0% |
| 4 | 538 | 2.9% |
| 5 | 601 | 3.2% |
| 6 | 879 | 4.7% |
| 7 | 1835 | 9.9% |
| 8 | 5726 | 30.9% |
Such incomplete reporting raises important concerns for longitudinal analysis. Apparent changes in average energy use over time may partly reflect changes in the mix of buildings reporting, not genuine performance trends. Buildings missing in some years may differ systematically from those that report consistently, potentially biasing estimates if not addressed.
Recognizing this issue, we also constructed a balanced panel composed of the 5,726 buildings that reported in all eight years. This stable subset provides a consistent observational base and is used alongside the full unbalanced panel in later sections to strengthen our analysis.
Empirical Strategy
This study estimates whether mandatory energy audits under Local Law 87 (LL87) lead to improvements in building energy performance, as measured by Site Energy Use Intensity (EUI). Because audit deadlines are staggered across buildings based on an externally assigned BBL rotation schedule, not owner choice, the timing of audit exposure provides quasi-random variation.
Importantly, all multifamily buildings are eventually treated. Therefore, identification relies on comparing buildings before and after their own audit, while using buildings that have not yet reached their audit year to form the “counterfactual trend” in each year. This structure supports a staggered difference-in-differences (DiD) design.
To do this, we estimate the following two-way fixed effects (TWFE) specification: \[Y_{it} = \delta D_{it} + \sigma_i + \tau_t + \varepsilon_{it}\] Where: - \(Y_{it}\) = log(Site EUI) of BBL \(i\) in year \(t\) - \(D_{it}\) = 1 if building \(i\) has completed its LL87 audit by year \(t\), 0 otherwise; - \(\delta\) = treatment effect (≈ % change in energy use) - \(\sigma_i\) = BBL fixed effects - \(\tau_t\) = year fixed effects - \(\varepsilon_{it}\) = error term \(D_{it}\) equals 1 if building iii has completed its first LL87 audit by year t. The coefficient \(\delta\) captures the average percentage change in energy use following the audit. Building fixed effects \(\alpha_i\) absorb time-invariant characteristics such as construction type and equipment vintage, and year fixed effects \(\gamma_t\) control for citywide shocks such as weather, energy prices, and broader policy changes. Standard errors are clustered at the building level.
A key threat to causal interpretation is that buildings audited earlier may differ systematically from those audited later in ways correlated with energy performance. Our identification strategy mitigates this concern in two ways. First, building fixed effects remove all time-invariant structural and occupancy attributes (e.g., size, heating system type). Second, year fixed effects account for common external shocks that affect all buildings, such as cold winters or fuel market volatility.
Another threat arises from the possibility of differential pre-treatment trends or anticipation effects if owners begin improving efficiency before audits occur. The event-study model directly tests the parallel-trends assumption by examining whether treated buildings diverge from comparison buildings prior to their audit year. We further account for serial correlation by clustering standard errors at the building level, ensuring valid inference despite persistence in energy consumption patterns. Together, these steps substantially strengthen the credibility of the estimated effects.
Findings
1.Regression results: TWFE Estimate
Based on the TWFE DID regression, using 89,311 building-year
observations and fixed effects for 15,317 multifamily housing buildings
across 8 years, yields:
| Statistic | Estimate |
|---|---|
| Coefficient (δ) | -0.0148 |
| Standard Error | 0.0091 |
| t-statistic | -1.631 |
| p-value | 0.1028 |
| Approx. % Change (100*δ) | -1.48% |
| Exact % Change (exp(δ) - 1) * 100 | -1.47% |
The table above reports the estimated average effect of completing an LL87 audit on multifamily buildings’ log Site Energy Use Intensity (Site EUI). The coefficient on the audit indicator is –0.0148, which translates into an approximate 1.48 percent reduction in Site EUI following the audit. Although the sign of the coefficient aligns with LL87’s intent that identifying inefficiencies through audits may help reduce energy use, the estimate is not statistically significant at 5 percent levels (p = 0.1028). The standard error (0.0091) produces a t-statistic of –1.631, indicating considerable uncertainty around the true effect.
Because the dependent variable is expressed in logs, the interpretation naturally centers on percentage changes. Here, the estimated decline is small and imprecisely estimated: the standard error is almost as large as the coefficient, and the confidence interval easily includes zero. This means the model does not provide statistically reliable evidence that LL87 audits reduce log Site EUI. While the direction of the effect is consistent with modest energy savings, the data do not allow us to conclusively distinguish this from no change in consumption.
2.The TWFE Event Study
Recent econometric literature has demonstrated that traditional Two-Way
Fixed Effects (TWFE) estimators can produce biased results in research
settings characterized by staggered treatment timing and heterogeneous
effects. To address these complexities, we move beyond a static
“post-treatment” dummy variable and instead adopt a dynamic event study
framework. This approach decomposes the estimated impact into discrete,
annual intervals relative to the year of compliance, rather than
estimating a single, aggregate average treatment effect.
By utilizing this event study specification, we can map the temporal evolution of building energy performance both before and after the LL87 intervention. This granular decomposition serves two primary purposes: first, it allows for a rigorous test of the parallel trends assumption by examining the coefficients in the pre-treatment period; and second, it reveals whether the impact of the audit and retro-commissioning persists, fades, or intensifies in the years following compliance. This methodology ensures that our causal identification is robust to the unique challenges posed by the staggered rollout across NYC’s multifamily housing stock.
Utilizing a comprehensive panel of 89,311 observations across 15,317 multifamily buildings, the event study identifies a distinct temporal pattern in the policy’s impact. The analysis evaluates the evolution of site EUI relative to the audit year to distinguish between immediate operational shifts and long-term capital improvements.
## OLS estimation, Dep. Var.: log_site_eui
## Observations: 89,311
## Fixed-effects: BBL: 15,317, year: 8
## Standard-errors: Clustered (BBL)
## Estimate Std. Error t value Pr(>|t|)
## rel_year::-4:evertreated 0.015144 0.013618 1.112096 0.26611453
## rel_year::-3:evertreated -0.009094 0.010635 -0.855117 0.39249957
## rel_year::-2:evertreated -0.011131 0.008065 -1.380135 0.16756532
## rel_year::0:evertreated 0.006536 0.007775 0.840701 0.40052881
## rel_year::1:evertreated -0.000726 0.009897 -0.073404 0.94148545
## rel_year::2:evertreated -0.016141 0.011740 -1.374807 0.16921131
## rel_year::3:evertreated -0.047757 0.014279 -3.344494 0.00082631 ***
## rel_year::4:evertreated -0.041524 0.020095 -2.066389 0.03880853 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.34551 Adj. R2: 0.480913
## Within R2: 5.231e-4The pre-treatment coefficients (leads) are centered close to zero, providing no evidence of differential energy trends prior to the mandated LL87 audit. This statistical non-significance validates the parallel trends assumption, confirming that the site EUI of buildings in different compliance cohorts was moving in tandem before the intervention. Consequently, the subsequent reductions can be attributed to the policy rather than pre-existing trajectories or selection bias.
The results reveal that the impact of Local Law 87 on building energy performance is characterized by a significant implementation lag. Between the audit year (\(t = 0\)) and the second year post-compliance (\(t = 2\)), the estimated coefficients remain small and statistically insignificant, indicating that the mandate does not yield an immediate reduction in Site EUI. The first measurable impact appears three years after the audit, where the estimate reaches -0.048 (p < 0.001), representing an approximate 4.8% decline in energy use relative to the baseline. This effect persists into the fourth year post-audit, with a statistically significant estimate of -0.042 (p < 0.05).This temporal pattern suggests that the observed energy savings are not primarily driven by the immediate operational “tune-ups” associated with retro-commissioning. Instead, the delay likely reflects the time required for building owners to evaluate the diagnostic audit findings, secure necessary capital, and complete the implementation of more substantive, low-cost Energy Conservation Opportunities (ECOs) identified during the diagnostic phase. Ultimately, the findings indicate that while LL87 is effective, its benefits are realized through a gradual, phased response rather than an instantaneous shift in building performance.
Conclusion
Key Takeaways
Our analysis provides several critical insights into the effectiveness
of Local Law 87 (LL87) audits in reducing energy consumption within New
York City’s multifamily sector. Across both the two-way fixed effects
(TWFE) and dynamic event-study models, we find no evidence of immediate
reductions in Site EUI following audit completion. Specifically,
buildings undergoing LL87 audits show no statistically significant
change in energy consumption during the audit year (t=0) or the
subsequent two years (t=1, 2).However, the event-study results reveal a
modest but measurable decline in site EUI, approximately 5–7 percent,
emerging roughly three years after the audit deadline.
These results imply that LL87’s audit-and-disclose approach may influence building performance only after owners have had time to interpret audit findings, plan improvements, and implement recommended changes. At the same time, the overall magnitude of the effect is small, and the absence of short-run reductions suggests that audits alone may not be sufficient to drive rapid or large-scale improvements in energy efficiency across the city’s building stock.
Limitations
A central limitation of this analysis is the high degree of
unbalancedness in the LL84 benchmarking data. Only 30.9 percent of
multifamily buildings report complete data for all eight years in our
study window, meaning that most buildings enter or exit the sample at
some point. This raises the possibility of selection bias if buildings
with consistent reporting differ systematically from those with missing
years. To better understand this limitation, we conducted several
comparisons between the balanced panel and the rest of the dataset.
| Borough | Total Buildings | Balanced Panel Buildings | Incomplete Panel Buildings | Percent Retained | Percent Dropped |
|---|---|---|---|---|---|
| BRONX | 4125 | 1023 | 3102 | 24.8% | 75.2% |
| BROOKLYN | 4761 | 1178 | 3583 | 24.7% | 75.3% |
| MANHATTAN | 6744 | 2569 | 4175 | 38.1% | 61.9% |
| QUEENS | 2779 | 918 | 1861 | 33.0% | 67.0% |
| STATEN IS | 164 | 42 | 122 | 25.6% | 74.4% |
| NA | 7 | NA | 7 | NA% | 100.0% |
Beyond these variations in retention and potential selection patterns, the spatial data for the multifamily sector reveals a persistent geographic hierarchy in building energy performance. Throughout the observation period, multifamily buildings in the Bronx and Queens consistently exhibit the highest mean Site EUI, with energy intensities frequently reaching the upper end of the 75–110 kBtu/ft² scale. This trend is most prominent in 2013, when the multifamily stock in Queens emerged as a citywide outlier, reaching the absolute peak intensity observed in the dataset.
While all five boroughs display synchronous temporal volatility—evidenced by a uniform energy dip in 2012 and a collective spike in 2014—the multifamily stock in the Bronx and Queens remains elevated relative to Manhattan and Brooklyn across nearly every year. By 2018, the citywide multifamily profile shows a visible convergence toward lower energy intensity; however, the Bronx and Queens continue to represent the higher end of the energy-use spectrum. These findings suggest that while external shocks like weather drive performance shifts citywide, the multifamily housing stock in these two boroughs remains characteristically more energy-intensive than the rest of the city.
Secondly, the distribution of gross floor area shows clear size differences between the two samples: the balanced panel skews toward larger buildings, with its density curve shifted to the right and a heavier upper tail. In contrast, the incomplete panel is more concentrated among mid-sized buildings, with fewer very large properties represented. This indicates that larger multifamily buildings are substantially more likely to report consistently across all eight years, while smaller and mid-sized buildings drop out more frequently. Because building size is correlated with key determinants of energy use, such as system complexity, investment cycles, and occupancy patterns, this imbalance suggests that the analytical sample may over-represent larger properties relative to the citywide building stock.
Thirdly, the age comparison shows that the balanced panel disproportionately represents older and mid-century buildings, with larger density peaks before WWII and again in the 1950s–1960s. The unbalanced sample, by contrast, contains more modern buildings, including a noticeable concentration of properties built after 1980 and especially after 2010.
Because LL87 began implementation around 2013, many of the newest buildings were not yet due for their first audit or may have had fewer years to appear consistently in the LL84 dataset. Their absence in the balanced panel therefore reflects both true reporting attrition and policy timing, rather than purely data quality differences. As a result, the balanced panel provides a stable but older-skewed view of NYC’s multifamily stock, while the unbalanced sample captures a broader set of newer properties that are still entering the compliance cycle.
Next Steps & Policy Implications
Taken together, our findings demonstrate that Local Law 87 (LL87) audits
yield only gradual and modest energy reductions within New York City’s
multifamily sector, with no detectable impact in the immediate
short-term. The significant 4.8% reduction in Site EUI observed three
years post-audit suggests that for multifamily owners, audit findings
function as a long-term planning resource rather than a catalyst for
immediate operational change.
To bridge this “information-to-action” gap, policymakers should consider evolving LL87 from a diagnostic mandate into an actionable pathway for building improvements. This transition could include pairing audit requirements with targeted, low-cost financing for the specific energy conservation opportunities identified in the reports. Furthermore, requiring follow-up verification for low-cost operational “tune-ups” would ensure that the most accessible savings are realized immediately rather than being deferred during long capital planning cycles.
As the city enters the era of binding emissions limits under Local Law 97, the role of LL87 must be repositioned as a strategic roadmap for compliance. Future research should prioritize the integration of benchmarking data with Department of Buildings (DOB) permit records to determine exactly which retrofits drive the observed medium-run savings. Understanding this causal link is essential for assessing whether the multifamily sector can effectively leverage informational tools to meet the aggressive decarbonization targets necessary for New York City’s climate goals.
Finally, the persistent energy intensity observed in the Bronx and Queens introduces a critical energy justice dimension that mandates a shift toward direct investment. While New York City shows a general trend toward energy convergence, the multifamily buildings in these two boroughs consistently remain the most intensive users, even after audit cycles are completed. Because these boroughs contain a higher concentration of low-to-moderate-income communities, these persistent inefficiencies may point to a disproportionate economic and environmental burden. To address these disparities, the city must move beyond informational mandates and prioritize direct capital investment and technical assistance for these vulnerable communities, ensuring that the transition to a low-carbon future does not leave behind the neighborhoods facing the greatest structural and financial barriers.
References
Mayor’s Office of Long-Term Planning and Sustainability. (2013). New York City Local Law 84 Benchmarking Report. New York City Government. https://www.energystar.gov/sites/default/files/buildings/tools/The%20New%20York%20City%20Local%20Law%2084%20Benchmarking%20Report%2C%202013.pdf