solar | Béla Figge

Did California's Solar Mandate Increase Solar Adoption?

Wed, 29 Jul 2026 12:25:00 -0700

California solar mandate series

The Strange Economics of Rooftop Solar
How We Combined Engineering and Economics
The Solar Gap
Did California’s Solar Mandate Increase Solar Adoption? (current)

One question appears in almost every policy discussion.

Did the policy work?

It sounds simple.

In practice, it is surprisingly difficult to answer.

Suppose rooftop solar installations increase after California introduces a mandate.

Does that mean the mandate caused the increase?

Not necessarily.

Solar installation costs were changing. Electricity prices and utility tariffs were changing. Federal tax incentives remained available. California had been expanding rooftop solar for years before the statewide mandate took effect.

Simply comparing before and after cannot distinguish the effect of the mandate from all of these other changes.

Our goal was therefore slightly different.

Instead of asking whether solar adoption increased after 2020, we asked:

How much would adoption have increased even if California had never adopted the mandate?

Answering that question requires constructing a counterfactual.

Policy timeline. San Francisco’s 2017 Better Roofs ordinance preceded California’s statewide low-rise residential requirement, effective in 2020.

The problem of missing histories

Every policy evaluation faces the same challenge.

We observe what actually happened.

We never observe what would have happened otherwise.

California either adopted the mandate or it did not. History unfolds only once.

The challenge for empirical researchers is to estimate the missing alternative—the counterfactual.

Constructing a synthetic comparison

To estimate that counterfactual, we used synthetic control methods.

The basic idea is intuitive.

Rather than comparing the treated jurisdiction with any single untreated place, synthetic control constructs a weighted combination of other places that resembles it before the policy.

No individual city looks exactly like San Francisco, and no individual state looks exactly like California. But a weighted combination can sometimes closely reproduce pre-policy characteristics and outcome trends.

The resulting synthetic comparison provides an estimate of what adoption might have looked like without the mandate.

One advantage of the method is that readers can inspect the comparison directly. If the treated and synthetic series track one another before the policy and then diverge afterward, that pattern supports—but does not by itself prove—a causal interpretation.

The counterfactual problem. Synthetic control constructs a weighted comparison that tracks the treated geography before the policy, then uses the post-policy divergence to estimate the effect.

San Francisco: directly observing solar on new buildings

San Francisco’s Better Roofs ordinance took effect in January 2017. It applied to new residential, commercial, and municipal buildings, with exemptions and alternative compliance pathways including solar thermal systems and living roofs.

For this analysis, we observe individual new-construction permits and whether residential projects installed rooftop solar. We construct synthetic San Francisco from a donor pool of 18 other Bay Area municipalities that had not yet adopted their own mandates.

The standard synthetic control places most of its weight on Oakland and San Jose. Before the policy, synthetic San Francisco tracks the city’s solar-adoption share reasonably well despite quarterly volatility.

After implementation, the series diverge.

Our main synthetic-control estimate indicates that the ordinance increased the share of new residential buildings with rooftop solar by 33 percentage points, or about 120 percent relative to the pre-policy level. A complementary synthetic difference-in-differences estimator produces a similar effect of 31 percentage points, with a standard error of 9 percentage points.

Post-policy photovoltaic adoption was approximately 57 percent rather than 100 percent. That is consistent with the ordinance’s design: projects could qualify for exemptions or comply through solar thermal systems or living roofs.

When those pathways are considered, total verified compliance rises to roughly 80 to 85 percent by 2018–2019, and only a small remaining share cannot be verified as compliant or exempt.

California: using a proxy for new-construction adoption

The statewide mandate took effect in January 2020 for newly constructed low-rise residential buildings.

At the state level, however, we cannot directly observe the share of new homes that installed solar across California and all potential comparison states.

Instead, we construct a proxy outcome:

Total residential solar capacity additions per new residential building permit.

The numerator includes all residential solar capacity added in the state, including installations on existing homes. The denominator is an average of residential building permits issued several quarters earlier to account for the lag between permitting, construction, and solar interconnection.

This measure therefore should not be interpreted as the average system size installed on a newly constructed home.

It is a state-level proxy for whether residential solar capacity increased relative to the flow of new construction.

We validate this proxy using the Bay Area permit data. When applied to San Francisco’s citywide mandate, it produces a substantial post-policy increase consistent with the direct new-construction adoption measure.

What we found statewide

For the statewide analysis, synthetic control constructs a comparison primarily from New York and Hawaii in the monthly and quarterly specifications.

The estimates are consistent across time aggregations:

the monthly synthetic-control model estimates an increase of 5.39 kilowatts, or 38 percent, per new residential building permit;
the quarterly model estimates 6.38 kilowatts, or 45 percent; and
synthetic difference-in-differences estimates 8.46 kilowatts, or 60 percent.

These magnitudes are in the range of the capacity of a typical residential photovoltaic system, although the outcome itself is a proxy and cannot be interpreted as a system installed directly on each new permit.

The methods differ in their comparison weights and identifying assumptions, but all point to the same qualitative conclusion:

California’s statewide mandate substantially increased residential solar capacity relative to new construction.

Why robustness matters

Synthetic-control estimates are only as credible as the counterfactual they construct.

The paper therefore includes several supporting analyses:

pre-policy comparisons of the treated and synthetic series;
balance tables for predictor variables;
placebo tests that apply the same method to untreated cities and states;
synthetic difference-in-differences estimates; and
validation of the statewide proxy using the richer San Francisco permit data.

For both policies, the post-treatment increase stands out relative to placebo estimates, although the statewide outcome and comparison necessarily remain less direct than the building-level San Francisco analysis.

Increased adoption is not the end of the policy question

At this point, it would be tempting to conclude:

The mandate worked.

Our paper is more careful.

Increasing adoption answers one important question:

What changed because of the policy?

It does not answer every important question:

Was that change worth achieving through this particular policy?

A complete policy evaluation must also consider whether the additional installations generated private and social benefits sufficient to justify their costs, how much private value depended on tax credits and net-metering transfers, and how rooftop solar compares with alternatives such as utility-scale renewable generation.

Our engineering analysis finds generally positive private payoffs under the policies in place, but those returns depend substantially on transfers. Removing the federal Investment Tax Credit reduces the median estimated payoff by 41 percent; replacing net metering with a less valuable compensation policy reduces it by 58 percent; removing both reduces it by 75 percent.

The paper also notes that utility-scale solar can generally produce renewable electricity at lower capital cost, while rooftop solar may offer distinct distributed-generation or grid benefits.

The results therefore support a precise conclusion:

San Francisco’s and California’s mandates substantially increased rooftop solar adoption or capacity relative to their estimated no-policy counterfactuals.

Whether rooftop mandates are the most efficient way to increase renewable generation is a separate welfare question, and the paper finds limited support for preferring them to larger utility-scale projects on those grounds.

Measuring policy versus judging policy

One lesson from this project is the importance of distinguishing estimating a policy effect from evaluating policy desirability.

The first asks what changed because of the policy.

The second asks whether that change was worth its full social cost and whether another policy could have achieved the same goal more efficiently.

Those questions are related, but they are not identical.

Our evidence provides a clear answer to the first: the mandates substantially increased solar adoption.

It also provides several inputs into the second—without pretending that a treatment-effect estimate alone settles the broader policy debate.

This post is based on Solar Adoption by Mandates, joint work with Stefano Carattini, Wade Davis, and Anton Heimerdinger. The full working paper includes the engineering model, policy details, robustness checks, placebo analyses, and supplemental results.

Previous: The Solar Gap

Return to the Solar Adoption by Mandates publication page.

The Solar Gap

Wed, 29 Jul 2026 12:20:00 -0700

California solar mandate series

The Strange Economics of Rooftop Solar
How We Combined Engineering and Economics
The Solar Gap (current)
Did California’s Solar Mandate Increase Solar Adoption?

If you ask a simple question—

Should a profitable investment be adopted?

—the obvious answer is usually yes.

Not because people are assumed to be perfect, but because expected profitability should generally make adoption more likely, even when other considerations matter.

Our study asked whether that relationship holds for rooftop solar on newly constructed homes.

The answer was:

Only weakly.

That finding became one of the central results of the paper.

Measuring expected profitability

In the previous article, I described how we estimated the expected private payoff from rooftop solar for thousands of new-construction projects using NREL’s System Advisor Model.

Each building received an estimated net present value based on local weather, rooftop characteristics, electricity prices, expected electricity consumption, financing, incentives, installation costs, and other project characteristics.

The result was an estimate of what rooftop solar was expected to be worth financially for each individual project.

The next step was straightforward.

We compared those estimated payoffs with what builders actually did.

What would we expect to see?

Imagine ordering every building from least profitable to most profitable.

If expected profitability strongly determined adoption, we would expect a simple pattern.

The least profitable projects would rarely install solar.

The most profitable projects would install it much more frequently.

Reality looked different.

Estimated payoffs and observed adoption. Green distributions show projects that installed solar; blue distributions show those that did not. The lowest-payoff projects are less likely to adopt, but across most of the distribution adopter and non-adopter payoffs overlap substantially.

The projects with the very lowest estimated payoffs were indeed less likely to install solar.

Beyond that lower tail, however, builders chose not to install solar across the full range of positive estimated payoffs.

Some projects with relatively modest estimated returns adopted solar. Others with substantially higher estimated returns did not.

The relationship between expected profitability and adoption existed, but it was much weaker than a simple investment model would predict.

Quantifying the relationship

Visual patterns can be misleading, so we also estimated statistical models relating observed adoption to estimated net present value.

The relationship remained modest.

Across model specifications and alternative engineering parameterizations, a $1,000 increase in estimated net present value was associated with a relatively small increase in adoption probability. In the primary specification with city and year fixed effects, a one-standard-deviation increase in payoff within a city-year corresponded to an increase in adoption probability of only about 1.2 percentage points.

Estimated net present value by itself explained less than 8 percent of the variation in adoption decisions. Specifications adding location, time, and income controls explained more overall variation, but the payoff-adoption relationship remained limited.

That does not mean profitability is irrelevant.

It means that many other factors appear to influence adoption.

Estimated profitability explains little of adoption. Across the paper’s linear probability models, NPV explains less than eight percent of observed variation in solar adoption.

What is the solar gap?

We use the term solar gap for the evidence that builders sometimes forgo rooftop solar despite positive estimated private payoffs—and, in particular, for the weak relationship between the size of those estimated payoffs and observed adoption.

Notice what this definition does not say.

It does not claim that every builder who skipped rooftop solar made a mistake.

Nor does it imply that our engineering model perfectly measures the true profitability of every project.

Instead, it points to something more modest.

If estimated profitability were the dominant driver of adoption, we would expect projects with higher estimated payoffs to adopt much more frequently.

That is not what we observe.

Why might this happen?

The paper discusses several possible explanations.

1. The engineering model may miss some costs

Administrative effort, permitting, contractor coordination, grid interconnection, and design revisions can all create real costs.

These “hassle costs” are difficult to observe and quantify. The paper estimates that, across most parameterizations, unobserved costs or model misspecification would need to exceed roughly $4,000 per project to rationalize even half of the observed non-adoption decisions.

That is possible, but it is not trivial relative to the modeled payoffs.

2. Builders and future occupants may face different incentives

The builder typically pays for installing the solar system.

The future homeowner or tenant receives much of the electricity savings.

If builders cannot fully recover that value in the building’s sale price or rent, they may rationally install less solar than would be optimal for the eventual occupant. This type of split incentive is familiar from research on energy efficiency.

3. Builders may have different expectations

Our engineering model uses information that would have been available when builders made their decisions.

Nevertheless, builders may have formed different expectations about future electricity prices, net-metering rules, installation costs, equipment performance, or tax incentives.

Reasonable decision-makers can disagree about the future.

4. Solar payoffs may be relatively small and easy to overlook

This explanation is easy to miss when the payoff is reported in dollars.

The median estimated payoff in our primary analysis is roughly $7,900. That sounds substantial.

But the median construction cost among projects with cost data is approximately $500,000. For most of those projects, the estimated solar payoff is less than 2 percent of total project cost; for almost all projects, it is below 5 percent.

Builders manage financing, permits, inspections, subcontractors, and dozens of other decisions simultaneously.

In that context, a benefit equal to one or two percent of project cost may simply receive relatively little attention—especially when it is uncertain or primarily benefits a future occupant.

Possible mechanisms, not definitive conclusions. The analysis is consistent with several explanations for the weak payoff–adoption relationship.

What the paper does—and does not—show

The paper does not identify a single reason builders often declined rooftop solar.

Instead, it documents several related empirical facts:

estimated private payoffs were positive for most projects under a wide range of assumptions;
adoption remained low before mandates;
the very lowest-payoff projects were less likely to adopt; and
above that lower tail, the size of the estimated payoff only weakly predicted adoption.

These findings support the existence of a solar gap, while leaving room for unobserved costs, differing expectations, split incentives, and other rational explanations.

The extent of the gap is therefore bounded by the assumptions in the engineering model. The most optimistic net-present-value calculations provide an upper bound on what builders could be said to forgo.

Why this matters beyond rooftop solar

The broader lesson extends beyond California.

Climate policy increasingly depends on technologies that appear economically attractive: heat pumps, electric vehicles, home batteries, building retrofits, and distributed solar.

For each technology, policymakers often ask:

How much money does it save?

That is an important question.

But this research suggests a second question may be equally important:

How strongly does expected profitability actually predict adoption?

Those are not the same thing.

A technology can appear financially attractive while still spreading more slowly than expected. Understanding that difference is essential before deciding whether additional information, financing, incentives, standards, or mandates are warranted.

This post is based on Solar Adoption by Mandates, joint work with Stefano Carattini, Wade Davis, and Anton Heimerdinger.

Previous: How We Combined Engineering and Economics

Next: Did California’s Solar Mandate Increase Solar Adoption?

How We Combined Engineering and Economics

Wed, 29 Jul 2026 12:15:00 -0700

California solar mandate series

The Strange Economics of Rooftop Solar
How We Combined Engineering and Economics (current)
The Solar Gap
Did California’s Solar Mandate Increase Solar Adoption?

One of the hardest questions in economics is surprisingly simple:

What would have happened if someone had made a different decision?

Suppose a builder chooses not to install rooftop solar on a newly constructed home.

Would installing solar have saved money?

Or was avoiding solar actually the better financial decision?

The problem is that we observe only the choice that was made. We never observe the alternative.

To answer that question, we combined engineering and economics.

Economics can observe choices—but not necessarily opportunities

Economic data are very good at recording what people do.

We can observe whether a builder installed rooftop solar. We can observe where the building was constructed, when permits were filed, and characteristics of the building itself.

But none of those data tell us something equally important:

How profitable would rooftop solar have been for this specific project?

Without that information, it is difficult to know whether non-adoption reflects a project with poor financial prospects or barriers that prevented an otherwise attractive investment.

Estimating those expected payoffs became the first part of our project.

Using an engineering model

Rather than building our own engineering model, we used the System Advisor Model, or SAM, developed by the U.S. National Renewable Energy Laboratory.

SAM is used by engineers, utilities, consultants, and solar developers to evaluate photovoltaic systems. Given enough information about a building and proposed system, the model estimates electricity production, utility bills, installation and operating costs, financing, incentives, and ultimately the expected net present value of installing rooftop solar.

In other words, SAM estimates what a builder or homeowner might expect to gain financially from installing a system.

That made it the ideal starting point.

But we did not simply run SAM

Many previous engineering studies evaluated only a handful of representative buildings.

Our objective was different.

We wanted to estimate expected private payoffs for thousands of actual construction projects. That required assembling a much richer dataset than had previously been used in evaluations of rooftop solar mandates.

For each building project, we combined information from multiple sources:

municipal building-permit records identifying new residential projects across the San Francisco Bay Area;
rooftop shading estimates from Google Project Sunroof;
local weather and solar irradiance from NREL’s National Solar Radiation Database;
hourly electricity-consumption profiles from restricted-access PG&E metering data for nearly 63,000 addresses;
historical electricity tariffs and net-metering rules;
changing federal and state incentives; and
California interconnection records describing installation costs across cities and years.

Together, these datasets allowed us to parameterize the engineering model using characteristics of actual projects rather than hypothetical examples.

From observed buildings to estimated payoffs. We combine building permits with rooftop shading, local weather, electricity-use profiles, tariffs, incentives, and installation costs, then use NREL’s System Advisor Model to estimate the expected private payoff from rooftop solar for each project.

Every project receives its own estimate

One detail that often gets overlooked is that rooftop solar systems do not all have the same optimal size.

A building with higher electricity demand benefits from a different system than one with lower demand. Roof size matters. Electricity tariffs matter. Shading matters.

Rather than evaluating one fixed solar system for every project, we optimized system size separately for each building to maximize its expected private payoff, subject to the building’s characteristics and rooftop constraints.

That means every project in the dataset receives its own predicted net present value under its own circumstances.

The result is not a single estimate of “the” profitability of rooftop solar. It is a distribution of expected payoffs across thousands of projects.

Modeling expectations at the time of the decision

The adoption decision occurred when the builder was planning and permitting the project—not years later, when its eventual performance could be observed.

We therefore attempted to parameterize SAM using information that would have been available when each permit was first filed. Historical installation costs, tariffs, net-metering rules, incentives, financing assumptions, and technology characteristics vary by project year.

For electricity demand, our primary model uses city-level mean hourly consumption profiles for recently created PG&E premises. This approach is intended to approximate what a builder could reasonably expect about a future occupant’s electricity use. As a robustness check, we also use matched building-level interval data for the subset of projects where those data are available.

This distinction matters. An engineering model can generate a very precise estimate under a given set of inputs, but the decision-maker does not know the future with certainty. Our estimates represent plausible expected payoffs at the time of the decision, not guaranteed realized returns.

Engineering tells us what appears financially possible

After running the engineering model, we knew something economists rarely observe directly.

For every building, we had an estimate of:

What would rooftop solar likely have been worth?

Under our primary parameterization, expected private payoffs were positive for every project. For single-family homes, they ranged from approximately $2,700 to $17,200, with a median of about $7,900.

Alternative assumptions produced more variation. Higher installation costs, lower electricity consumption, less favorable compensation for exported electricity, or the absence of the federal tax credit reduced estimated payoffs. With project-matched electricity-consumption data, 12 percent of the relevant subsample had negative estimated payoffs, although some of those estimates likely reflected incomplete consumption records or buildings that were not yet occupied.

That heterogeneity is itself an important result.

Earlier engineering studies often concluded that rooftop solar was cost-effective across all representative new-construction cases. Our analysis suggests a more nuanced picture: expected private payoffs were generally positive, but their magnitude—and occasionally their sign—depended on project characteristics and assumptions about future conditions.

Two linked questions. Engineering estimates the expected payoff from a system; economics compares those incentives with observed adoption.

Economics asks a different question

Once we estimated expected payoffs, the project shifted from engineering to economics.

Instead of asking:

Would solar appear to pay?

we could ask:

Did builders actually install solar where the expected payoff was highest?

Those are different questions.

The engineering model estimates the private financial opportunity under specified assumptions. The adoption data reveal behavior.

Comparing the two is what led us to the idea we call the solar gap, which I discuss in the next article.

Why combine disciplines?

Many climate-policy questions require both engineering and economics.

Engineering tells us what technologies are capable of achieving and what they appear likely to cost or save.

Economics helps explain how people respond to those opportunities and whether policies change their behavior.

Neither perspective is sufficient on its own.

A technology can work extremely well and still fail to spread. Likewise, observing slow adoption does not necessarily mean the technology is ineffective or that non-adopters are making mistakes.

Understanding both requires combining the two perspectives—and being clear about what each can and cannot establish.

This post is based on Solar Adoption by Mandates, joint work with Stefano Carattini, Wade Davis, and Anton Heimerdinger.

Previous: The Strange Economics of Rooftop Solar

Next: The Solar Gap

The Strange Economics of Rooftop Solar

Wed, 29 Jul 2026 12:10:00 -0700

California solar mandate series

The Strange Economics of Rooftop Solar (current)
How We Combined Engineering and Economics
The Solar Gap
Did California’s Solar Mandate Increase Solar Adoption?

For years, engineering studies supporting California’s rooftop solar policies suggested that installing solar would be cost-effective on essentially all new homes.

If that were true, however, a puzzle remained.

Before solar became mandatory, most newly constructed homes in the San Francisco Bay Area did not install it.

Why would builders pass up an investment that appeared to pay for itself?

That question motivated our paper, Solar Adoption by Mandates. We examine the economics of rooftop solar requirements in California, beginning with San Francisco’s 2017 Better Roofs ordinance and then the statewide requirement that took effect in January 2020.

The paper asks three related questions:

How profitable was rooftop solar on actual new construction projects before the mandates?
Did builders adopt solar on the projects where its estimated payoff was highest?
Did the mandates increase adoption relative to what would otherwise have happened?

The answers are more nuanced than either “solar obviously pays” or “builders were irrational.”

Looking beyond the typical house

The engineering studies used to support early solar mandates generally evaluated a small number of hypothetical buildings with stylized characteristics.

We took a different approach.

We assembled detailed data on thousands of actual residential construction projects across the San Francisco Bay Area. We then used the National Renewable Energy Laboratory’s System Advisor Model to estimate the expected private payoff from installing rooftop solar on each project.

The calculations account for differences in local sunlight, rooftop shading, electricity consumption, utility tariffs, installation costs, incentives, financing, and system size. Rather than assuming one representative house, we estimated payoffs across the range of buildings that developers were actually constructing.

That produced much more variation than earlier policy studies suggested.

Under our primary assumptions, estimated net present values were positive for every project in the sample. For single-family homes, they ranged from approximately $2,700 to $17,200, with a median of about $7,900. Under alternative assumptions—including higher installation costs, less favorable electricity compensation, and different electricity-use profiles—a minority of projects had negative or very small estimated payoffs.

Estimated private payoffs from rooftop solar. The primary model produces positive but heterogeneous net present values. Alternative assumptions reduce the estimated payoff for some projects and reveal greater variation than studies based on a small number of representative buildings.

The first lesson, then, is not that rooftop solar was equally profitable everywhere.

It was that solar appeared financially attractive for most projects, but the size and even the sign of the payoff depended on project characteristics and assumptions about future costs, electricity use, incentives, and utility policy.

Profitable does not necessarily mean important

A median estimated payoff of roughly $8,000 sounds substantial on its own.

But the median construction cost among projects for which we observe costs was approximately $500,000. For most projects, the estimated solar payoff was less than 2 percent of total construction cost, and for almost all projects it was below 5 percent.

That relative scale matters.

A builder manages financing, architectural plans, contractors, inspections, permitting, and a long construction schedule. Even when rooftop solar has a positive estimated return, that return may be small relative to the rest of the project. Additional design work, administrative effort, interconnection delays, or other costs not captured in the engineering model could further reduce its importance.

The expected benefit may also accrue primarily to the future homeowner through lower electricity bills, while the builder bears the installation cost and implementation burden. Whether the builder can fully recover that value in the sale price is uncertain.

So the relevant question is not simply:

Does solar have a positive estimated net present value?

It is also:

Is that payoff large, certain, and salient enough to affect the builder’s decision?

Positive does not necessarily mean salient. For projects with observed costs, the median estimated solar payoff was about $7,900 compared with a median construction cost of roughly $500,000.

Profitability only weakly predicted adoption

Once we had estimated payoffs for individual projects, we compared them with actual solar adoption.

There was a modest positive relationship: projects with the very lowest estimated payoffs were less likely to install solar.

Beyond that lower tail, however, builders declined to install solar across the full range of estimated positive payoffs. Projects with comparatively high expected returns often remained without solar, while some lower-payoff projects adopted it.

Estimated net present value explained less than 8 percent of the variation in adoption decisions.

We refer to this weak connection between estimated private payoffs and observed adoption as the solar gap.

The finding does not prove that builders behaved irrationally. Our payoff estimates may omit project-specific constraints or costs that builders understood. Expectations about future tax credits, electricity rates, or net-metering policies may also have differed from the assumptions in the model.

Still, the results suggest that estimated profitability was not the dominant factor in many adoption decisions.

What could explain the gap?

Several mechanisms are plausible.

Builders may not capture the future electricity savings enjoyed by occupants. Financing or liquidity constraints may make upfront costs more important than long-run returns. Information may be incomplete. Developers may be uncertain about future electricity tariffs or incentive policies. Solar installation may also involve permitting, contractor coordination, interconnection, and other “hassle costs” not fully represented in an engineering model.

Another possibility is simple scale and attention. An expected solar payoff equal to one or two percent of total project cost may receive little attention amid the larger demands of constructing a building.

Our analysis does not isolate one definitive explanation. Instead, it shows that a simple model in which builders adopt whenever estimated net present value is positive does a poor job of describing the observed data.

What the mandates changed

California’s policies allow us to ask what happened when rooftop solar was no longer left entirely to voluntary adoption.

San Francisco’s 2017 Better Roofs ordinance applied to new residential, commercial, and municipal buildings, although it included exemptions and allowed solar thermal systems or living roofs as alternative compliance pathways.

Using synthetic-control methods, we estimate that the San Francisco policy increased the share of new residential buildings with solar photovoltaics by 33 percentage points—more than doubling the pre-policy adoption rate. A complementary synthetic difference-in-differences analysis produces a similar estimate of 31 percentage points.

California’s statewide mandate took effect in January 2020 for newly constructed low-rise residential buildings. At the state level, we cannot directly observe the solar share among new homes, so we use total residential solar capacity additions per new residential building permit as a proxy outcome.

Depending on the estimator and time aggregation, we find an increase of approximately 5.4 to 8.5 kilowatts of residential solar capacity per new building permit, equivalent to an increase of roughly 38 to 60 percent relative to the counterfactual.

These results indicate that both mandates substantially increased solar adoption.

That does not mean they produced universal rooftop solar. San Francisco’s ordinance contained exemptions and alternative compliance pathways, while California’s code also included exemptions and options such as community solar. Policy design and implementation therefore matter when interpreting post-mandate adoption.

Increasing adoption is not the same as proving a policy is efficient

It is tempting to move directly from “the mandate increased rooftop solar” to “the mandate was good policy.”

Our paper is more cautious.

The engineering model estimates private payoffs to adopters, not the full social costs and benefits of the mandate. A substantial share of those private payoffs came from the federal Investment Tax Credit and California’s net-metering policies. Removing the tax credit reduces the median estimated payoff by 41 percent; replacing net metering with a less valuable compensation policy reduces it by 58 percent; removing both reduces it by 75 percent.

Those policies transfer value to rooftop solar adopters, but transfers are not themselves net social benefits.

A complete policy assessment would also compare distributed rooftop solar with alternatives such as utility-scale renewable generation, which can often produce electricity at lower capital cost. Rooftop solar may provide other benefits, including distributed generation and potential grid-management value, but those benefits must be weighed against its higher cost.

Our results therefore support a narrower and more defensible conclusion:

Solar mandates substantially increased rooftop solar adoption, and many of the projects induced to adopt likely received positive private payoffs under the policies in place.

That is different from showing that rooftop mandates are the least-cost way to increase renewable electricity generation. The paper ultimately finds limited support for preferring mandates for rooftop solar over larger utility-scale projects on efficiency grounds.

The broader lesson

The most interesting part of this project is the gap between technological possibility and observed behavior.

Engineering models can estimate whether a technology is likely to generate electricity and produce a positive financial return. They are less well suited to predicting whether builders, households, or firms will actually adopt it.

That second question depends on incentives, information, uncertainty, financing, transaction costs, institutional design, and who bears the cost relative to who receives the benefit.

The same issue arises with heat pumps, insulation, electric vehicles, batteries, and climate-adaptation measures. Lower costs and better technology are essential, but they do not guarantee adoption.

Understanding how technologies perform is an engineering question.

Understanding whether they spread—and whether policy should compel them to spread—is also an economic and institutional one.

This post is based on Solar Adoption by Mandates, joint work with Stefano Carattini, Wade Davis, and Anton Heimerdinger. The full working paper contains the technical details, robustness analyses, and complete results.

Next: How We Combined Engineering and Economics

Solar Adoption by Mandates

Mon, 01 Sep 2025 00:00:00 +0000

Public-facing writing

This paper is explained in a four-part series under Writing: