Causal Evaluation of Building Retrofit Effects on Rent: A Research Protocol Using Retrofit Design Evidence and Reproducible Building-Operation Data

1. Introduction

The question addressed here is narrower than the broad literature on sustainable buildings: how should a study estimate the causal effect of building retrofit on rent when the available background evidence concerns retrofit design, engineering performance, envelope materials, and operational data rather than rental outcomes? A credible answer requires careful definition of treatment before any econometric specification is chosen. Building retrofit is not a single intervention. It may involve insulation, glazing replacement, façade renewal, reflective roofing, shading systems, HVAC replacement, lighting modernization, or combined packages. If these distinct actions are collapsed into one indicator, any estimated rent effect mixes interventions that differ in technical content, timing, visibility to tenants, likely disruption during implementation, and plausible valuation channels. [1] [2]

The engineering literature gives reasons to avoid that simplification. Retrofit research distinguishes passive and active strategies and frequently evaluates them by several performance criteria rather than by energy use alone. It also reports that retrofit suitability and performance vary across climates and building contexts. Material- and façade-focused studies add another source of heterogeneity: systems that appear similar at the building-market level may differ in thermal behavior, durability, weather resistance, or life-cycle burdens. Those studies do not show how rents respond. They do show that any rent design needs documented intervention categories and comparators that reflect more than age and location. [1] [2] [3]

The cited studies are used only to inform exposure definition, outcome construction, and design choices. This article does not treat them as evidence on rent. Its contribution is prospective: a future quasi-experimental design linking documented retrofit events to rent panels, augmenting conventional controls with reproducible pre-treatment operational signatures, and estimating dynamic effects under explicit assumptions and diagnostics. No empirical rent analysis is reported.

2. Evidence Scope and Use in This Protocol

The evidence review is intentionally targeted. The cited studies are used only for three preparatory tasks: defining retrofit exposure with engineering specificity, identifying performance domains relevant to heterogeneity and mechanism analysis, and structuring reproducible pre-treatment operational controls. Because none of the studies estimates rent effects, the protocol does not infer tenant willingness to pay, capitalization into lease rates, or landlord pricing responses from them.

One group of studies concerns retrofit strategy and performance evaluation. The optimization review distinguishes passive and active retrofits, notes that combined strategies are common, and identifies energy efficiency, thermal comfort, cost-effectiveness, and environmental impact as major metrics used in retrofit assessment. It also emphasizes that many results are case-specific, which limits transferability across climates and building types. The industrial retrofit study operationalizes envelope-oriented decision making around energy use intensity, thermal discomfort hours, and life-cycle carbon emissions, and uses sensitivity analysis to identify influential parameters. Together, these studies justify a rent protocol that preserves technical distinctions among retrofits and records multiple post-treatment mechanism variables instead of presuming one dominant channel. [1] [4]

A second group of studies concerns envelope technologies and environmental assessment. The review of plasma-modified textile polymer materials reports that the reviewed studies associate plasma treatment with altered surface morphology and chemistry, changes in wettability and adhesion, enhanced hydrophobicity and reflective performance, and in some cases lower cooling demand or roof temperature for roofing or shading applications. The same review also stresses limited long-term building-scale validation, inconsistent performance reporting, plasma aging, scalability constraints, and incomplete techno-economic evidence. The façade life-cycle assessment study shows that façade alternatives do not dominate uniformly across impact categories and that different components act as environmental hot spots in different systems. For rent analysis, these findings matter because they caution against coding all envelope work as one exposure with one expected market effect. [2] [3]

A third group of studies is methodological. The HVAC-operation dataset study presents a reproducible pipeline for multi-building, multi-climate simulations with standardized variables such as indoor temperature, outdoor weather, setpoints, and HVAC power, and releases code and datasets for reproducibility. The heating-load prediction study describes a model-building workflow using historical building and weather data, lagged features, training and testing procedures, and residual-based evaluation. Neither study addresses rent. Their relevance here is prospective: they suggest how a later rent study could construct auditable pre-treatment operational features rather than relying only on coarse building descriptors. [5] [6]

3. Research Question, Target Estimand, and Causal Structure

The primary research question is: what is the causal effect of a documented retrofit event on rent per unit area over time relative to the rent path that the same building would have followed without that retrofit? The main estimand is a dynamic average treatment effect on treated buildings by event time, estimated separately for treatment cohorts defined by implementation timing and then aggregated over supported event times. Secondary estimands concern heterogeneity by retrofit class, visibility, and intensity.

The causal structure separates four variable types. First, treatment is the completed retrofit package, coded by class and implementation wave. Second, pre-treatment confounders include fixed building characteristics, submarket, climate context, pre-treatment rent trajectory, occupancy history, and baseline operational signatures observed before treatment. Third, post-treatment mediators include realized changes in operational or environmental performance, such as energy use intensity, discomfort indicators, or carbon-related measures. Fourth, the outcome is rent per unit area. This separation is substantive and methodological: baseline operational conditions may help explain why some buildings were selected for retrofit, whereas post-treatment operational changes may lie on the causal pathway from retrofit to rent and therefore belong in mechanism analysis rather than in the total-effect specification.

The protocol does not assume that superior technical performance yields a rent premium. Utility pass-through, lease duration, local vacancy, tenant search behavior, and the visibility of the intervention may all interrupt that link. The engineering literature justifies collecting structured retrofit and building-performance variables. Whether those variables predict rent responses remains a question for later empirical testing.

4. Retrofit Exposure Taxonomy and Event Coding

The proposed taxonomy uses four mutually recorded categories, while recognizing that this classification is a protocol design choice rather than a taxonomy taken directly from prior studies. Each retrofit wave is coded for the presence or absence of: passive envelope retrofit, active system retrofit, control retrofit, and combined package status. Passive envelope retrofit includes insulation, glazing changes, façade systems, reflective roofs, and permanent external shading. Active system retrofit includes major HVAC or lighting replacement. Control retrofit covers supervisory control changes, setpoint-program logic revisions, or building-management software modifications without major equipment replacement. A retrofit wave is marked as a combined package when at least two of the first three categories occur within the same implementation episode. [1] [2] [3]

Treatment timing is defined by an operational two-date rule that separates substantial completion from stabilized operation and then maps the daily stabilization date to the rent panel without ambiguity. For building i and retrofit wave r, let c_ir be the substantial-completion date, defined as the first calendar date on which commissioned scope share reaches at least 0.80. Commissioned scope share is the sum of completed scope weights divided by the sum of planned scope weights within that wave. The primary weighting scheme assigns each documented retrofit measure a weight equal to its affected area share if area is meaningful and documented; otherwise equal weights are assigned within class, and this fallback must be declared before coding begins. Let s_ir be the stabilized-operation date, defined as the first date of the earliest 14-consecutive-day window beginning on or after c_ir during which ordinary operation is observed on every day. Ordinary operation means no commissioning override, no temporary shutdown of the retrofitted system, and occupancy at or above 50% of the building’s typical pre-treatment occupied-area level, unless the asset type has a different pre-registered occupancy metric. The treatment date d_ir is defined as s_ir, not c_ir. To align the daily date with rent outcomes observed monthly or quarterly, the treatment panel period T_ir is assigned by a default majority-exposure rule: T_ir is the first panel period p for which at least one half of the days in p occur on or after d_ir. If d_ir falls exactly at the midpoint, assignment is to the next period. Periods that contain d_ir but fail the majority-exposure rule are classified as disruption and commissioning periods for that wave. Thus c_ir marks substantial completion, s_ir marks verified stabilized operation, d_ir = s_ir is the operational treatment date, and T_ir is the first rent panel period treated in estimation. If records identify c_ir but do not permit verification of a full 14-day ordinary-operation window, the wave is not eligible for the primary analysis and may enter only a sensitivity design with a separately declared timing rule. These thresholds and the panel-assignment rule are proposed protocol rules, not findings from the cited literature.

Visibility is coded to distinguish interventions that are externally or experientially salient from those that are largely hidden to tenants. Define Visible_ir = 1 if, by the operational completion date d_ir, documentation shows that the wave changes the exterior-facing façade, roof, glazing, or permanent external shading, or materially alters a tenant-accessible common area or leased area. For the main analysis, qualifying evidence must come from engineering records, permits, site logs, or leasing and marketing documents dated no later than the end of the panel period containing T_ir. Documents issued after that point may be used only to corroborate an already assigned classification, not to create it. Generic statements such as 'property improvements' do not qualify. If records conflict, the coding priority is: commissioned engineering scope, then permit or inspection records, then contemporaneous site logs, then leasing or marketing materials. This visibility determination is made after wave consolidation has been resolved under the timing rule for repeated interventions, so that the coded object is always the consolidated wave used in estimation. If the highest-priority records remain contradictory or do not identify the completed feature, visibility is coded indeterminate and excluded from visibility heterogeneity analysis unless resolved by independent review under the same priority rule.

Treatment intensity is measured with one primary scalar and one fallback. The primary metric is the Scope-Weighted Retrofit Share, SWRS_ir, defined as the commissioned scope share at completion and therefore bounded between 0 and 1 within a documented wave. Because SWRS depends on within-wave scope accounting, it is comparable across buildings only when the coding manual uses the same weighting hierarchy. The fallback metric is Class Count_ir, equal to the number of retrofit classes present among envelope, active system, and control retrofit. Heterogeneity analysis must identify which metric is primary. The two measures should not be pooled or interpreted as equivalent.

Repeated interventions are handled at the building-wave level with mutually exclusive episode windows, an overlap rule for stabilization conflicts, and an explicit adjudication rule for incomplete records. If two documented waves begin less than six months apart, they are consolidated into one wave r unless the record supports separate episodes for both waves. Separate episodes require evidence for each wave of at least one wave-specific scope document and either: (a) a wave-specific substantial-completion date c_ir and stabilized-operation date s_ir under the primary timing rule, or (b) a wave-specific substantial-completion date plus at least one dated document showing return to ordinary operation before the next wave begins. If a later wave begins before the earlier wave reaches its stabilized-operation date s_ir, the two waves are consolidated automatically in the primary analysis, because the earlier wave never attains a distinct post-stabilization period. If these conditions are not met for both waves, consolidation is mandatory in the primary analysis. When records are contradictory, the coding priority is: commissioned engineering scope and commissioning records, then permit or inspection records, then contractor or site logs, then owner or leasing records. If the highest-priority records still do not distinguish the waves, the analyst must consolidate them rather than exercise case-by-case discretion. If two waves begin at least six months apart, each wave may define a separate treated episode for the same building, subject to the same dating and support requirements. Let b index the building-wave episode (i,r). Episode b enters risk only at its own T_ir and remains in the estimation sample until the earliest of: the start date of a later documented wave in the same building, exit from the rent panel, or administrative study end. Outcomes observed after T_ir and before the first panel period containing the later wave’s start date are attributed to the earlier episode and retain that episode’s event-time index. The panel period containing the later wave’s start date is not attributed to the earlier episode in the primary design unless the later wave is later consolidated back into the earlier one; instead, censoring occurs at the start of that period to avoid contamination by anticipatory work, disruption, or partially overlapping scope. After censoring, a new risk set may be created for the later episode if it satisfies the same entry rules. A building with any earlier completed wave is never used thereafter as an untreated control for another building’s cohort, because its untreated potential outcome is no longer observed. Re-entry is allowed only as a new treated episode with covariates remeasured using information observed strictly before that later episode’s d_ir. The same building may therefore contribute multiple non-overlapping treated episodes across calendar time, but it cannot contribute overlapping treated risk time or serve as both treated and control in the same period. The protocol must report how many buildings contribute multiple waves, how much follow-up is lost to later-wave censoring, how many periods fall into the between-wave gap retained for the earlier episode, and how many later episodes fail support or data requirements.

5. Outcomes and Panel Construction

The primary outcome is rent per unit area at a regular panel interval, preferably effective contracted rent if concessions are observed. If only asking rent or face contract rent is available, the outcome definition must be stated separately and not treated as interchangeable with effective rent. The preferred scale is currency per square meter per month or equivalent local area-time unit. A monthly panel is ideal when lease records permit; otherwise a quarterly panel may be used if treatment timing and lease updates can be aligned without excessive aggregation.

Construction disruption can create temporary rent discounts, vacancy spikes, or missing observations. The protocol therefore requires analysts to distinguish at least three periods where data permit: pre-treatment, disruption and commissioning, and post-stabilization. Under the default panel-assignment rule, any panel period that contains d_ir but has fewer than one half of its days on or after d_ir is classified as disruption and commissioning rather than as treated post-stabilization time. The primary total-effect specification may begin at T_ir if the question concerns the effect of completed retrofit inclusive of initial disruption occurring before that first treated period. A secondary stabilized-service specification may instead exclude all disruption and commissioning periods from both treated episodes and matched controls by the same calendar-relative rule, but that exclusion must be pre-registered and applied symmetrically across cohorts. If exclusion rules differ across buildings because of inconsistent records, the design should treat that sample as exploratory rather than primary.

Standard building-level covariates include age, gross floor area, asset type, submarket, climate zone, occupancy history, and other documented capital improvements not part of the retrofit wave. The protocol does not treat these variables as sufficient. Retrofit adoption may depend on baseline inefficiency, unstable controls, thermal discomfort, or atypical load sensitivity to weather. Those features can affect both selection into retrofit and later rent trajectories. This is a hypothesis motivating additional pre-treatment controls, not an empirically demonstrated fact about all datasets.

6. Reproducible Pre-Treatment Operational Controls

Where operational data are available, the protocol proposes harmonizing them to a common schema modeled on reproducible building-operation datasets: timestamp, outdoor weather variables, indoor air temperature, heating and cooling setpoints, and whole-building HVAC power, all retained in physical units. In the simulation dataset study, released variables are stored without normalization, aggregation, or rescaling. A future rent study cannot assume that real-building source systems already satisfy that property. Instead, the study should publish a preprocessing script that maps raw records into the common schema and documents any aggregation or unit conversion applied before feature generation. [5]

Pre-treatment operational signatures are computed over the 12 months ending one full month before d_ir, provided that at least 75% of scheduled intervals are observed. If that requirement fails, a 6-month fallback window is allowed only when two conditions hold: observed coverage within that 6-month window is at least 75%, and the window contains at least 60 days from each operating season that is relevant to the retrofit classes analyzed in the primary specification. A heating-relevant season is present when the window includes at least 60 days with daily mean outdoor temperature below 12 °C or positive heating degree days relative to 18 °C. A cooling-relevant season is present when the window includes at least 60 days with daily mean outdoor temperature above 22 °C or positive cooling degree days relative to 18 °C. If the primary analysis is restricted to heating-oriented waves, only heating-season coverage is required; if restricted to cooling-oriented waves, only cooling-season coverage is required. If mixed classes are pooled in one primary analysis, both seasons are required unless the building is documented to lack the corresponding system or service in that season, in which case the pooled design is not primary and the building must move to a class-specific or exploratory sample. Features are generated separately for occupied and unoccupied periods when occupancy schedules exist; otherwise they are generated on the full interval series with a missing-schedule flag.

The default feature set includes: mean HVAC power within prespecified outdoor-temperature bins; slope of HVAC power on outdoor temperature estimated separately for heating-relevant and cooling-relevant conditions; mean absolute deviation of indoor temperature from the active setpoint; proportion of intervals outside a pre-registered comfort band around setpoint; variance of indoor temperature during occupied hours; and share of intervals with setpoint changes exceeding a pre-registered threshold within one hour, used as a control-instability proxy. Comfort-band and control-change thresholds are protocol parameters and must be set before analysis. For illustration only, a study might define a comfort band of plus or minus 1 degree Celsius around the active setpoint, but any chosen threshold remains hypothetical until pre-registered in a specific dataset. [5] [4]

A compact residualized-load feature is also proposed. The default predictor is a regularized linear model estimated on pre-treatment data only: Load_it = alpha + w'X_it + c'Cal_it + s'SP_it + l'L_it + q'Z_it + error_it. Here X_it is the weather vector, containing outdoor temperature and, when available under the same preprocessing rule for all buildings retained in the analysis, solar irradiation and relative humidity; Cal_it contains hour-of-day and day-of-week indicators; SP_it contains active heating and cooling setpoints; L_it contains pre-registered load lags; and Z_it contains interaction terms formed as outdoor temperature multiplied by season indicators. Coefficients are estimated separately for each component of each vector within one joint regression, not by collapsing weather inputs into a single scalar coefficient. The primary residual features are the mean signed residual over the final full heating season and the final full cooling season before treatment, stored separately as PreResidualHeat_i and PreResidualCool_i when both seasons exist. The comparison set for sensitivity analysis includes one tree-based model with tuning fixed by pre-registered cross-validation rules. Model choice for the primary analysis is not left open after inspection. [6]

These operational features are proposed confounding controls, not validated causal adjusters. Their role is to summarize baseline operating behavior that may correlate with retrofit selection and subsequent rent dynamics. Any empirical study using them should report whether they improve balance and whether results are sensitive to excluding them. High predictive accuracy in a load model is not enough to justify causal adjustment if the resulting residual feature is unstable across buildings or dominated by missing-data artifacts. [6]

7. Identification Strategy and Estimation Procedure

The preferred design is a staggered-adoption difference-in-differences event study with cohort-specific estimation rather than conventional two-way fixed effects. The analysis unit is the building-wave episode b = (i,r), not the building alone. The primary estimator is doubly robust group-time DiD applied after support restriction. For each treated cohort g and event time k, eligible controls are episodes belonging to buildings not yet treated at calendar time t = g + k plus never-treated buildings when available, excluding any building already exposed in an earlier wave. The analysis estimates a propensity score for cohort membership using pre-treatment variables only, trims observations outside common support, and fits an untreated-outcome model using the same pre-treatment variables and building and calendar fixed effects. ATT(g,k) is then estimated with an augmented inverse-probability weighted score. This is the default estimator because consistency requires correct specification of either the propensity model or the untreated-outcome model, conditional on the maintained assumptions.

Comparator construction proceeds in four fixed stages. Stage 1 restricts candidate controls to the same submarket or a pre-registered neighboring submarket and to comparable climate context. Climate comparability is enforced with three pre-treatment metrics measured for the building location or nearest weather station: annual heating degree days relative to 18 °C, annual cooling degree days relative to 18 °C, and long-run mean outdoor temperature. A candidate control is retained only if each metric lies within 10% of the treated episode's value or within an absolute tolerance of 250 degree-days for heating, 250 degree-days for cooling, and 1.5 °C for mean temperature, whichever tolerance is less restrictive for that metric; if these rules remove all candidates in a submarket, the cohort is unsupported for the primary analysis. Stage 2 estimates pre-treatment similarity using only variables observed before treatment: building characteristics, lagged rent summaries, occupancy, non-retrofit capital events, and operational signatures. Stage 3 enforces common support by computing propensity scores within the Stage-1 climate-eligible candidate set and trimming observations with scores outside the overlap interval defined by the 1st and 99th percentiles of the opposite group within that same set, unless a tighter pre-registered bound is chosen. Stage 4 estimates cohort-specific event-time effects in the supported sample. The identifying assumption is conditional parallel trends: absent retrofit, treated episodes and selected controls with similar pre-treatment observables would have followed similar rent paths over the relevant horizon.

The target estimand is defined at the building-wave level with event time indexed in the same panel units as the rent outcome. Let G_b denote the first treated panel period T_ir for episode b = (i,r), measured in calendar periods after applying the majority-exposure mapping from d_ir to the panel, and let k = t - G_b be event time. For supported cohorts g and event times k, ATT(g,k) = E[Y_bt(1) - Y_bt(0) | G_b = g, t = g + k, b in S(g,k)], where S(g,k) is the subset of cohort-g episodes with observed outcomes at event time k before any later-wave censoring or panel exit and with admissible support under the primary comparator rule. If a later wave starts during calendar period p, the earlier episode contributes outcomes only through the last period strictly before p unless the two waves are consolidated under the overlap rule; there is no re-indexing of the earlier episode after censoring. Aggregate dynamic effects are formed by a pre-registered default rule: ATT(k) = sum_g n_gk ATT(g,k) / sum_g n_gk, where n_gk is the number of treated episodes in S(g,k). Thus the primary aggregation is size-weighted over treated episodes with support at event time k. A sensitivity analysis will replace n_gk with inverse-variance weights computed from the cohort-specific standard errors, after excluding cohorts whose variance estimate is undefined or unstable under the pre-registered threshold. Both the set of contributing cohorts and the weights used at each k must be reported.

The estimation sequence is fixed. First, code treatment waves, classes, visibility, and treatment dates from engineering and leasing records. Second, assemble the building-period panel for rent and pre-treatment covariates. Third, generate operational features using only pre-treatment intervals. Fourth, estimate the matching or weighting model and report balance before outcome estimation. Fifth, estimate cohort-specific event-time effects. Sixth, aggregate them across cohorts for each event time. Seventh, estimate pre-registered heterogeneity by class, visibility, and intensity. Eighth, conduct placebo, pre-trend, and support diagnostics. Any redesign after a failed diagnostic requires full re-estimation and renewed validation under the revised protocol; post hoc adjustment alone does not restore causal credibility.

8. Prospective Validation and Diagnostic Rules

Because no empirical application is reported, validation is prospective. The first diagnostic layer concerns event coding. A coding audit should compare assigned treatment class, visibility status, consolidation of waves, and treatment month against engineering documents, commissioning logs, and leasing records. If disagreement on treatment month exceeds a pre-registered tolerance, such as one panel period, the building is either recoded through independent review or excluded from the primary analysis. The purpose is to reduce bias from announcement dates, partial completion, or prolonged commissioning.

The second layer concerns feature-generation reproducibility. Versioned code should transform raw operational records into the common schema and then into pre-treatment summaries using fixed rules. If whole-building HVAC power is unavailable for a large subset, that subset cannot simply receive imputed features and remain in the same primary analysis without qualification. Instead, the protocol requires either a reduced-feature specification estimated as a separate design or restriction of the primary sample to buildings with comparable operational variables. [5]

The third layer concerns residualized-load stability. The default linear load model and the pre-registered comparison model are both trained only on pre-treatment data. Evaluation is performed on held-out pre-treatment blocks, with attention to residual plots, seasonal structure, and instability across buildings rather than to a single fit statistic. If residuals remain strongly patterned by season or building identity, residualized-load features move to sensitivity analysis and are not used in the primary balancing stage. [6]

Pre-trend diagnostics are decision rules rather than proofs. If estimated lead coefficients show material departures from zero under pre-registered tests, the primary causal interpretation is suspended. Possible revisions include narrowing the comparator pool, changing the support rule, or enriching the set of pre-treatment controls, but any such revision constitutes a new analysis stage that must be re-estimated and re-diagnosed. A violated identifying assumption cannot be repaired by interpretive discussion alone.

9. Placebos, Negative Controls, and Sensitivity Analyses

The protocol uses placebo strategies more readily than negative-control outcomes because many administrative variables are not substantively meaningful outcomes. The preferred placebo is a pseudo-treatment date assigned within the pre-treatment period of treated buildings, aligned by calendar structure to the actual treatment date. A second placebo uses matched untreated buildings assigned pseudo-event dates corresponding to their treated counterparts. If these placebos generate systematic post-placebo effects, the comparator design is treated as inadequate.

Auxiliary placebo variables may be used when a data source contains time-varying administrative fields that should not plausibly respond to retrofit during the study window, but these variables are not central and must be justified case by case. Categorical identifiers that are effectively fixed, such as utility account numbers or insurer zone codes, are generally poor candidates and should not be presented as primary negative-control outcomes. Planned-but-cancelled retrofit projects, if documented before implementation and linked to the same buildings or close comparators, may instead serve as a more credible negative-control exposure.

Sensitivity analysis covers six departures from the primary design: alternative treatment-date thresholds within a pre-registered range; inclusion versus exclusion of the disruption and commissioning window; designs with and without operational signatures; alternative common-support trimming bounds; inverse-variance aggregation instead of the default size-weighted rule; and separate analyses for visible and non-visible interventions. In repeated-treatment settings, an alternative multiple-treatment estimator may be added only if enough buildings contribute more than one well-separated wave; otherwise the wave-specific design with later-wave censoring remains primary.

10. Relation to Prior Methods and Boundaries of Generalization

The protocol differs in purpose from retrofit optimization studies. Those studies evaluate retrofit options under engineering objectives such as energy use, comfort, carbon, and cost. This article does not import their optimization algorithms to estimate rent. It uses them only to justify explicit intervention coding, multiple performance domains, and climate-sensitive heterogeneity planning in a later observational rent design. [1] [4]

Generalization should remain narrow. Prior retrofit work emphasizes case specificity across climates and building types. The plasma-material review reports promising surface and thermal properties but also limited long-term building-scale validation, inconsistent reporting, and scalability constraints. The façade life-cycle study shows that environmental rankings vary by impact category and component. Any later rent estimate should therefore be interpreted as specific to the documented retrofit classes, leasing institutions, building stock, and climate contexts actually studied. [1] [2] [3]

11. Conclusion

This article has proposed a prospective protocol for causal evaluation of retrofit effects on rent. The proposal defines treatment by retrofit class, implementation wave, visibility, and explicit completion rules; constructs pre-treatment comparators using both conventional building descriptors and operational signatures; and estimates dynamic effects with a cohort-specific staggered-adoption event-study design. It also sets out coding audits, reproducible feature-generation rules, placebo tests, and failure responses for common identification problems.

What has been established is conceptual. The cited studies support finer retrofit classification, multiple engineering performance domains, envelope heterogeneity, environmental trade-offs, and reproducible operational data structure. They do not establish rent effects, and they do not validate the proposed causal design. Whether the protocol yields credible rent estimates remains for future empirical study using linked rent records, documented retrofit events, and, where available, pre-treatment operating data. [1] [4] [2] [3] [5] [6]

References

1. Reza-E-Rabbi, Sk.; Bhuiyan, Muhammed A.; Zhang, Guomin; Dodampegama, Shanuka; Atapattu, Kanishka (2026). Simulation- and Metamodel-Based Multi-Objective Optimization for Sustainable Building Retrofit Across Climatic Conditions. Materials. 19(8), 1649. doi:10.3390/ma19081649.
2. Azeem, Musaddaq; Amor, Nesrine; Kashif, Muhammad; Noman, Muhammad Tayyab (2026). Advanced Plasma-Modified Textile Polymer Materials for Building Energy Retrofit Technologies. Polymers. 18(11), 1395. doi:10.3390/polym18111395.
3. Burchart, Dorota; Schabowicz, Krzysztof (2026). Environmental Life Cycle Assessment of Selected Materials—Building Façades in Poland. Materials. 19(4), 807. doi:10.3390/ma19040807.
4. Duan, Zhongcheng; Li, Binhao; Zi, Yilun; Yao, Gang (2025). Building retrofit multiobjective optimization using neural networks and genetic algorithm three for energy carbon and comfort. Scientific Reports. 15(1), 38076. doi:10.1038/s41598-025-21871-0.
5. Domingo, Ferran Aran; Alonso, Pablo Fraile; Torrentó, Josep Rius; Batalla, Oriol Agost; Vilardell, Ignasi Barri; Mayoral, Jordi Vilaplana; Fornés, Jordi Mateo (2026). A reproducible method to generate multi-building, multi-climate HVAC operation datasets with a stochastic exploratory controller. MethodsX. 16, 103866. doi:10.1016/j.mex.2026.103866.
6. Emir, Ahmet; Topal, Halil Ibrahim; Ozturk, Suat (2026). Novel hybrid machine learning-based prediction of building space heating load: a comprehensive study. Scientific Reports. 16(1), 24924. doi:10.1038/s41598-026-55342-x.