Stephen Lacey: We're revisiting a pretty important topic that has been batted around in solar. Is the technology only for the rich? Shayle and our guest, PowerScout CEO Attila Toth, use their analytical prowess to try and answer that question. And we're going to clarify the class-based perceptions of solar in this episode. Before we do that it's time to dig into our reading list. Shayle and I are reading the exact same thing today. A juicy piece of legalese from the newly-bankrupt solar manufacturer Suniva. What caught your attention about this document Shayle?
Shayle Kann: Well the headline most people picked up on is that Suniva's filing for bankruptcy. Which is news in and of itself, but that is not what is a big deal about this from a broader perspective. The reason this is a big deal is that in its bankruptcy filing, Suniva basically stated that it got some financing to take it through bankruptcy and that financing is contingent on Suniva filing what's called Section 201 petition. Now this is a little wonky but stay with me because it's going to be really important.
Section 201 is a section of the trade act of 1974. It's a mechanism for the US to impose remedies when a domestic industry has suffered what's called serious injury. And basically what it means is that Suniva intends to, but has not yet filed, a petition that if it went successfully through the International Trade Commission and to the President's desk, and we can come back to what that means, it could impose any number of different kinds of remedies from new important tariffs, to volume maximums, to minimum prices. And it could do so not just on China or Taiwan, which is what happened in the last solar trade case that was filed by SolarWorld. But this theoretically could be applied to the entire world. So this is going to be an early test of trade policy and the Trump administration. It will be a big deal for the cost of solar in the US. So if Suniva ultimately files this thing, it's going to be huge news.
Stephen Lacey: Right. So there are a lot of really important pieces to this. One is that the process is a lot faster then the tariff creation process we saw at the Commerce Department from 2011 through 2014.
Shayle Kann: Yes.
Stephen Lacey: That was a long 4-year process. We're looking at 120 days here.
Shayle Kann: Sort of. So the first process took a little bit longer. Took more like 9 months to a year to get sort of preliminary tariffs imposed. This is somewhat of an accelerated process. So the way that it works is if Suniva files, the International Trade Commission immediately looks at it. Within 120 days, so within 4 months, they have to decide whether the US industry has suffered serious injury. And sometimes they can extend that out to 5 months if they need to.
Stephen Lacey: And serious injury seems like a pretty high bar.
Shayle Kann: It is. In fact, it is a higher bar than what was set in the other type of petition. The type that SolarWorld filed.
Stephen Lacey: Which was what?
Shayle Kann: Which was called material injury. So this is serious injury. It's a higher bar, which in 4 months and potentially 5 months, if they extend it, the ITC would have to give a thumbs up or thumbs down on injury. Then by 180 days, so 6 months from the initial filing of the petition, if the ITC finds serious injury they have to issue recommendations about what remedies could be imposed. And so that's where they could explore a bunch of different options that I mentioned before. Those recommendations go directly to the desk of the President. And then the President has, as I understand it, a lot of leeway in determining exactly what he ends up doing about that. He could do nothing. He could accept all the remedies as proposed by the ITC or he could change them entirely.
Stephen Lacey: They go directly to the desk of a President who as a presidential candidate talked a lot about China as a currency manipulator. He's been very skeptical of global free trade agreements, and has especially cited Section 201 as a possible way of enforcing trade restrictions.
Shayle Kann: Yes. To be fair, Donald Trump himself has never probably mentioned Section 201. I'd be surprised if he ever heard of it in the first place. But it is true that in the trade policy document that came out in early March from the Trump administration, there were a bunch of trade remedies that were noted, specifically one of them being Section 201. I would just note, the last time this Section 201 was invoked was in 2002 in the steel industry. So it's pretty rare but I guess not a big surprise in the broader landscape that it would be introduced, or at least that it would be invoked, in the Trump Administration.
Stephen Lacey: The scope of this is really large as you said. So in the bankruptcy declaration document, that was filed this week in the Delaware court, the chief restructuring officer basically says that this could apply to South Korea, to Germany, to all sorts of European countries. This is a potentially global remedy. Not just regional specific.
Shayle Kann: Right. And you know, stepping back, the reason why that's what Suniva wants to do requires just a brief history lesson for anybody who hasn't been paying attention to trade and solar. Which is that SolarWorld filed this other kind of trade petition and anti-dumping countervailing duty petition in 2011 that did result in import tariffs on Chinese solar cells in 2012. So then a lot of Chinese manufacturers turned out to be shipping through Taiwan. So they would make the cells in Taiwan and assemble them into modules in China. So then SolarWorld filed a follow up petition to expand the scope to Taiwan that was largely successful. So then what ended up happening was most of the imports that are coming into the US now are not coming from either China or Taiwan. Many of them are coming from other countries in Asia. A lot of countries in Southeast Asia like Malaysia and Vietnam.
So I think what Suniva's trying to do here by filing Section 201, which allows them to encompass the entire world, is just say look, imports are the problem. It's not just imports from China. And we'll see whether they succeed.
Stephen Lacey: And so Suniva's basically saying to the President, let's test your policy on trade here. You know this President has never really mentioned solar other than to disparage the industry or to make offhanded remarks about how it doesn't work. But now all of a sudden solar could be integral to the President's execution of his stated trade policy.
Shayle Kann: It'll be interesting to see. I could imagine a couple of different things happening. One would be this stays quiet for everybody who's not in the solar industry as it goes through the ITC process. So for the next 6 months or so, you know we know about it and pay a lot of attention to it, but you never hear about it in mainstream media or from the President. And then if it gets to the President's desk, if that actually happens, then it becomes a big story. Or because this could be the first major trade action of the Trump administration you could imagine it gets elevated sooner. I guess we're just going to have to wait and find out.
Stephen Lacey: Well, let's get to what we came here to talk about this week. We're going to discuss another piece of analysis from you and Attila Toth.
So last year the Wall Street Journal issued an op-ed from its editorial board. And they called rooftop solar regressive, political, income distribution. The less fancy way of saying that is rich people are installing solar and poor people are paying for it. So you wanted to test that assumption. What did you and Attila actually do to test it?
Shayle Kann: This has always been a bit of a frustration for me, somebody who works with data for a living. There's never really been great publicly available data on the income distribution of solar customers. Residential solar customers. And so despite the fact that that question has come up a fair bit in political and regulatory debates. The Wall Street Journal wrote that editorial you mentioned while Nevada was going through its net metering dispute. It came up in some of the other net metering disputes in other states as well. Where one side would just sort of assert that solar tends to be for wealthy people. And poor people are paying for it.
The solar industry would disagree with that and nobody was really bringing a whole lot of data to the table. So it's always been a gap in my mind. And there have been some other pretty good attempts to try to answer that question. One from the Center for American Progress. One from Kevala Analytics. And they've been pretty good. The limitation on those has been that the data that is publicly available basically only goes down to the zip code level. So you don't actually know in those data sets, who's installing solar. You just know how much solar there is in a given zip code. And the issue with that is that there's pretty wide distribution of incomes within a given zip code. So that's always been frustrating to me and I ran across a different piece of analysis that Attila had done using PowerScout's data and methodology, which we talked about with him, that looked at the political leanings of solar customers and realized that if he could look at political leanings that he could also probably look at income distribution.
So we've been working together for the past few months, putting together this analysis that basically uses a pretty big sample of actual solar customers, actual solar households, representing a little over 500,000 individual solar households, at about 41 percent of the national solar market, and it looks at their income distribution to try to figure out how wealthy solar customers actually do tend to be.
Stephen Lacey: It's certainly a super important question so let's get into it and hear our conversation with Attila Toth the CEO of Power Scout.
Shayle Kann: Let's start by talking about what we did in this study that we put together, Attila. Specifically the data that we were trying to gather. As I mentioned in the intro, it's always been challenging for people to try to figure out the solar income distribution because there wasn't great data on who the actual households were that had solar. So you have a sort of solution to that, which is by using satellite imagery and what you've deemed your proprietary convolutional neural network machine learning algorithm, which sounds like something from a pitch deck in Silicon Valley. So can you explain exactly what that algorithm does, and how it allowed us to identify solar households?
Stephen Lacey: It actually sounds like something from Terminator, like Skynet.
Attila Toth: It may be out of a pitch deck, but it does work.
Shayle Kann: So how does it work?
Attila Toth: Convolutional neural networks is a fancy name for a machine learning algorithm for computer vision. This is the exact same algorithm that Tesla's using for self-driving cars or Facebook is using for the auto-tag feature, when they do facial recognition.
So what we have done is we used this technology to determine on satellite imagery which of the roofs have solar and which don't have solar. So first we had to train the models to understand what a roof is. How do you tell apart a room from a basketball court? And then we have to tag images. Hundreds of thousands of images with solar on the roof and without solar. So we had to teach what is the difference between a solar panel and a sun roof. What is the difference between a solar panel and solar thermal.
Once we have done this, then we unleash this algorithm on hundreds of thousands of satellite images. Once we have all these images, then we matched that to income data that we have procured from consumer marketing companies. At the intersection of those two databases, so we have the solar roofs and then we have the income data, we have identified about 520,000 homes that we know have solar and we have income data for. Then at the end we used GTM Research's data on total market size for the four markets that we have done this analysis to make sure that our sample size is statistically significant.
Shayle Kann: Right, so here's where we landed in this analysis. Having run through all of those satellite images through the convolutional network model and attaching that to income data, we ultimately identified a little over 520,000 individual residential solar households. We were only looking at four states just to limit the number of satellite images that we had to address. So that's California, New Jersey, Massachusetts and New York. Those states of course being big residential solar states, so in aggregate they represent a big portion of the overall market.
So we specifically identified about 62 percent of all the solar installations in those states. Or about 41 percent of all installations nationwide. So we think that's a pretty representative sample at least of those states. That allows us to come to some broader conclusions about the income demographics of solar customers.
So let's talk a little bit about the findings. And then I think we also want to address some of the limitations of this analysis and we want to talk about what we would have liked to have been able to do if we could. And what we might want to do in the future. But first order question is, is it true what the solar industry tends to say, which is that most solar customers are middle income households?
Attila Toth: Yes, so what we have found from the study is that middle and high income households are over represented in the solar sample. So about 87 percent of the solar sample that we have identified, the 520,000 households, have fallen in the middle and high income categories. Contrast that to 75 percent in the general population. So they are over represented. Low income households seem to be underrepresented. We found about 13 percent of households falling in those income brackets in our solar sample versus 25 percent in the general population.
Shayle Kann: So the findings here are a little bit nuanced, right? Because there's a couple of different ways to look at what we ended up discovering. One is the way that you just described it, which is that we found pretty universally that solar households skew somewhat higher income than the general population. As you said, if you're looking at low income and in this case defining low income as having household income below $45,000 a year, there's a smaller proportion of those than there are in the general, so a smaller proportion of solar households in that demographic than there is in the general population.
On the other hand, it appears also to be true that the vast majority of solar customers are what you might call middle income. So we're defining that being between $45,000 and $150,000 a year in annual household income.
Attila Toth: That's true, about 70 percent.
Shayle Kann: About 70 percent in the solar household, 65 percent in the general population.
So it skews a little bit higher income but not to say that it is only rich people. I think that would also be a false statement based on the data that we gathered here.
The other thing that I thought was interesting in the findings here was the demographic that seems to have gone solar the most, which is households with income between $100,000 and $150,000. So you might call those upper middle income households. That's where you see the biggest difference. Where it's like 25 percent of solar households but only something like 17 percent of all households. And I wonder Attila, as you guys are doing this consumer marketing for solar companies, are they targeting that demographic specifically? The sort of upper middle income type? Or does it just turn out to be that's who goes solar?
Attila Toth: When we are targeting and homeowners go solar, income is one factor that we look at. Above and beyond income, there are multiple other factors such as education level. Such as an estimation of their electric bill. Such as roof orientation where again convolutional neural networks come into play because we understand the potential solar that you can harvest from that roof. So what we do is we build a propensity model where income is one variable. But yes, I agree with you that those customers tend to be in the $100,000 to $150,000 range predominately.
Stephen Lacey: Another thing that I thought was interesting, one of the more surprising findings in this data, as I was looking through it was specifically about low income customers. So it is true that we found low income customers are somewhat underrepresented or low income households are somewhat underrepresented. Fewer of them have gone solar then exist. But it was still a pretty big number just to throw some data out there.
So we were estimating that of the sample that we found in those four states, about 13 percent of all the solar customers were what you might call low income households. Again, annual household income below $45,000 a year. That 13 percent would equate to, assuming that's a representative sample just across those states, that would equate to about 530 megawatts of low income solar. Again in California, Massachusetts, New Jersey and New York. So just to kind of place that in the market for you, the Obama administration right toward the end of Obama's years as president, put out what was supposed to be an ambitious goal to get 1 gigawatt of low income solar installed in the US, the entire US by 2020. And based on our data, we're suggesting that you know, there might already be half of that in place just in these 4 states.
And so while low income was somewhat underrepresented, to me this was surprisingly high. That you would have something like over a hundred thousand low income solar households in those 4 states. There's some great organizations like Grid Alternatives that have been installing for low income customers for a long time. But I'm just curious, were either of you guys surprised at either how high or how low that number was?
Attila Toth: Honestly I wasn't surprised. And we got a caveat to this. There are estimates in that number. But I wasn't surprised because the value proposition based on which solar has been sold in this country was save 15 percent or 10 percent on your energy bill. And that is the customer segment where the marginal savings represent the highest value. So I'm not surprised that they are responding to this type of value proposition.
Shayle Kann: I guess we should at this point take a step back and talk about why it's important to be trying to figure out income levels of solar customers. You know, there are a few different reasons for it, but I think that PowerScout's reasons are probably different than GTM's reasons. For you Attila, why is this question important to answer? Are you really just thinking about it from a customer acquisition perspective? Who should we be targeting and who is getting left out of the solar sales equation?
Attila Toth: Look, there is a customer acquisition perspective to this, but there is also a policy perspective to this. As I said, we have a lot of data. We have hundreds of data points on each one of the customers. Who has solar and what their income is is only two out of the multiple hundreds that we have accumulated. So the reason why we've decided to make this available to policymakers, installers, investors, is so we can inform them about the demographics of solar with the objective of making solar mainstream. This is not what our business is built on. Our business is built on leveraging all this data that we have about the consumers into propensity models. So who are the most likely to convert for solar, for electric vehicle charging, for battery storage systems or even an energy saving roof. A smart roof. And how can we reach those people cost efficiently? So that's the reason why PowerScout exists.
Shayle Kann: I don't know if this is too proprietary for you to answer, but if it's not I'm interested. In your designing of those propensity models, let's just say for solar specifically, like who is the perfect customer?
Attila Toth: It depends, right? It depends by geography. It depends by neighborhood. So the core of this is us being able to tag the customers who have already gone solar, right? And having hundreds of data points on them. Then we can extrapolate in their utility service, in their zip code, what are the common characteristics amongst them, right? So you could say, if you live in Menlo Park, California, and you have at least a 2,500 square foot home, your utility bill is at least $175, your income level is at least $125,000 per year and your education level is as least college, you are seven times more likely to convert. So that's where you spend your marketing dollars. As opposed to spending your marketing dollars on a very large population in a mass marketing campaign.
The data by itself is not valuable. What's valuable is how you connect what you know about the physical profile of that home. What you know about the owners of that home and how you connect that to an online profile so you can do cost efficient micro targeted digital marketing to those people.
Stephen Lacey: How do you think this will be interpreted by the solar industry? GTM is obviously not a policy advocacy shop. We're just interested in answering this question and figuring out what the market looks like and what consumer adoption looks like. But inevitably, as you alluded to Attila, this will become a piece of the policy discussion. So inevitably people interpret it in different ways, and I'm just curious if either of you have thought about the policy implications of these findings.
Shayle Kann: Look, I think that there's something in this data for everyone. You know, if you want to cherry pick from our findings you could make an argument in either direction. So if you want to say solar's only for wealthy people, there's data in here suggesting that solar skews higher income than the general population. So you could take that on it's own and suggest that solar is only for wealthy people. Though that's not entirely what the data shows. If on the other hand you want to show that we've gotten really good at installing low income solar, there's also data in there that suggests we've installed over 100,000 quite low income solar installations just in those four states, let alone in the entire country. So my hope is that this is a pretty nuanced view of the evolving demographics of solar and that it's not fodder for either side of this conversation.
Stephen Lacey: Oh how naive that you expect nuance.
Attila Toth: But guys the data point Shayle mentioned at the beginning of this conversation -- 70 percent of the solar homeowners that we have identified fall into the middle income category. I think that's also a very interesting data point that shows that solar is becoming democratized over time. And I stress how important it is to rerun this analysis a year from now. We've seen exponential growth in the past year. We're expecting to see healthy growth this year as well. We should be rerunning and trends are going to tell the real story.
Shayle Kann: Yes, I think that's a good point. And I would also just note that finding you mentioned, that the vast majority of solar households are middle income households. That is consistent with the other research that has been done. The stuff that was only at the zip code level but nonetheless found something very similar -- that the vast majority of the zip codes that have a lot of solar are middle income zip codes. So I think there's a pretty good preponderance of evidence to suggest that portion.
Attila Toth: One more point I would like to make, and maybe this is to the question of policy, which is there are differences. I believe that neighborhood effects have also had an impact in magnifying certain demographics in this data. Let me explain what that means.
Stephen Lacey: So you're talking about word of mouth?
Attila Toth: Not just word of mouth. You're driving around in your neighborhood and you're seeing solar in your neighborhood. People who are of similar background, similar income, live in the same neighborhood. So those neighborhoods that had early adopters benefited from a neighborhood effect that has magnified the adoption of solar by similar demographics. And NREL has put out a very good study about this.
Shayle Kann: There's also some research on that from UT-Austin from Dr. Varun Rai, that's really good. There's ample evidence of neighborhood effects and solar adoption. So that is a good point too.
Attila Toth: And we are seeing that in our data, right? We are seeing that solar adoption tends to be very concentrated by neighborhood.
Shayle Kann: I just want to throw one thing out there. Something is wrong with rich people in New Jersey. So of our four states, three of the four had at least an equal or higher proportion of solar households among the wealthiest families. This is households with income of over $250,000 per year. So in California, that's 10 percent of solar houses as opposed to 5 percent of the general population. Massachusetts 4 percent in each case. New York 6 percent of solar households and 4 percent of the overall population. But for some reason, in New Jersey there's a lower proportion of solar households that make over $250,000 a year than there is in the overall population, which means that New Jersey wealthy people are disproportionately not going solar.
So for our rich New Jerseyan listeners, get your act together.
Stephen Lacey: We are getting closer to figuring out who is buying solar. And it's not just for the rich. We can answer that question now. Shayle, how can people get this report?
Shayle Kann: Go to GTM. It's available. There's an article up. You can get the link there. You can also go to the GTM Research website, gtmresearch.com, there's a download link there.
Stephen Lacey: Attila Toth is the CEO of PowerScout and we really appreciate you coming on The Interchange and co-writing this report which is free on the GTM website. It's called "How wealthy are residential solar customers, household income and solar adoption in the US."
Attila, we appreciate your time.
Attila Toth: Thank you very much for your partnership guys.
Stephen Lacey: Shayle Kann is my co-host. He is our Senior Vice President here at GTM.
I am Stephen Lacey, GTM's Editor in Chief. You are listening to the Interchange. Our weekly conversation on the global energy transformation. We'll catch you next time. Thanks for being with us.
