I’m breaking from my normal type of article here at Seeking Alpha that usually concerns either a company with limited research posted, or a general Macro / Investment Strategy piece. I’m also cognizant of the highly emotional points of view regarding Tesla, Inc.’s (NASDAQ:TSLA) stock, and its high-profile founder and CEO Elon Musk. Hence, I want to be clear from the start that there will be no recommendations at all regarding the company, the stock, or the product under discussion. I just want to provide what I think is an interesting case study about Tesla’s solar tile product to the public.
The system that I analyzed is not mine but was purchased by someone I know who has been gracious enough to allow me to investigate and utilize their data in terms of production and usage. I will provide a lot of data including Tesla’s original estimate of production, actual figures obtained through Tesla’s application that tracks and helps control the system, as well as cross verify the data with the Net Meter readings from the local electric utility. Due to my concern over the emotional responses by advocates for and against this stock, I’m going to keep the exact location and owner anonymous. I will provide pictures though to give readers an idea of what a solar roof tile system looks like in operation.
While I have not directly expended the financial and emotional effort to build a house myself, I did have tangential experience during my younger years. My parents built three houses in their lifetimes and helped refurbish another. During my teenage years, my brother and I even helped as the onsite liaisons between the builders and my parents for the construction of one home. These experiences have provided some basic understanding of the process for building residential properties, and one rule of thumb is to assume that things will always be more complicated than planned. More complicated also means you should assume more expensive than estimated.
Source: Tesla Solar Roof Installation, picture from home owner’s manager.
My parents were keen on pushing the envelope in terms of technology built into their homes. For example, their house built in the mid-1980s had 8 large solar panels for heating water on the roof. It was connected to a tank buried in the ground in the basement that supplied all of the house’s hot water needs. That home also had a pool with the first saltwater system that I knew of anywhere. Another of their homes built in 2001-’02 had a full geothermal system installed for both heating and cooling air conditioning, as well as an early version of solar tiles on a portion of the roof. Thus, I’ve grown up experiencing the pros and cons of installing cutting-edge technology into homes.
The downside of brand new technology is that it usually struggles to work consistently. The saltwater pool in the 80s was constantly having problems with algae blooms after hard rainfalls. The pumps it ran on also tended to break down, and we developed a pattern of resorting to using chlorine at times when the conditions warranted. The house built in the 00s also had a saltwater pool, but by then the technology was refined and significantly more dependable. We had the same experience with the geothermal system in the 00s, but my guess is the current iterations are probably improving materially. Finding trained people to maintain new technological systems has been an issue with these homes in the past.
I mentioned that the house built in ’01-’02 had an early version of solar tiles on part of the roof. These were indeed tiles not panels, so my past experience with that house is partly what lead me to reach out to study this new Tesla system. One of the primary questions we struggled with back then was a simple one: How much actual energy is the solar system producing? It sounds almost silly, but there wasn’t any way of telling exactly what the energy production was. Since the house was also completely new, we couldn’t even do a pre / post type of analysis using the electric bills. At one point our quest lead us to someone who installed a flash drive into the inverter that stored the data. We then would have to climb into the attic, disconnect the drive, and download the data into a computer. However, the data itself was frankly useless to a non-electrical engineer such as myself, and we gave up. This is one issue that current owners of Tesla Roof systems do not have to manage as I’ll go into more detail later, but the simplicity and ease of the Tesla App monitoring and control system for the phone is a significant step-up from my previous experiences.
Therefore, the purpose of this article will be to answer the following questions: 1.) How Much Energy is the Tesla Solar Roof System generating? 2.) Does the energy generation match what the company estimated in terms of production? Finally, I’ll do an analysis that will attempt to answer the question most of you are probably interested in debating: 3.) Is buying the solar tile system worth the cost? I will be showing the full cost for the owner who purchased it. As I said before, building anything with homes almost always winds up with complications that increase costs, and this example confirms that basic rule. Let’s get on with it!
Thanks to my previous experiences, determining the amount of electricity actually produced by the solar tile system is my primary goal for this analysis. Tesla’s solar tile and storage system is a significant step-up from the older panel-based systems in terms of aesthetics. While issues around profitability and its justification for the SolarCity acquisition abound in other articles, I’m only interested in studying the actual application of this system. This recent Bloomberg article details upheaval in the management of the Tesla Solar Roof system, but I’m going to solely focus here on trying to answer that first question above as thoroughly as possible. As you are about to see, confirming the produced electricity was not as easy as I had hoped for initially.
The first difficulty came from the inability again to do a pre / post installation analysis with the utility bills. The house was purchased but the solar roof installation happened relatively quickly in late 2019 before the new owner had really moved into the home. The house was also newly built on spec by another person. Hence, there was no one living in it to give me a base rate of electricity consumption over any period of time before installation. The roof was also installed at that time to take advantage of a tax credit that was expiring at that point. This credit was later extended, but in 2019 that possibility was not clear. Thus, there was some incentive to get the job done before the end of that year.
The owner of this system has told me that the initial customer experience with Tesla was very good. This was the first solar roof system installed in the area, and the Tesla group was motivated to get it done. The house is actually located in an historic district within its township, which posed an extra hurdle to receiving the necessary permits. Specifically, this town did not allow solar panels to be installed on top of roofs within the historic district. Tesla though sent company personnel to meet with the town’s board and answer questions about the tile system at no extra cost to the customer. This included a Tesla representative that came from the other side of the country with tile samples to present to the board. This is significant as generally you don’t want to do these board meetings on your own as it can get contentious. In the end, the company’s efforts were successful, and I am told that the town now is genuinely promotional of this tiled system as a solution for new homes in the area.
Source: Original Tesla Solar Roof Proposal
Installation began in the fall of 2019, and while severe storms did slow the progress of completion to a degree, the new roof was in place in time to claim the tax credit before year end. There was a bit more time till the roof began to produce electricity around testing the system, and getting approval from the utility to connect to the grid. Finally in December of 2019, the switches were thrown for good and the system was up and running.
Source: Original Tesla Solar Roof Proposal
The system itself comprises the roof tiles which cover 1,618 square feet of the roof, and are separated into two inverters that cover each side of the house. There are also two power wall storage systems that are designed to electrify the whole home during blackouts from the utility. While not directly factored into my analysis at the end of this study, it should be noted here that there is added value from this as the home is located in an area where loss of power is common. The electrical lines are still mostly above ground, and severe storms have knocked out power to residents for prolonged periods of up to a week or more in certain circumstances. Below is a picture I took of the installed inverters and power walls in the basement. The decision to install two Powerwalls came from Tesla’s recommendation based upon estimated energy demands, and the design for the system to be able to power the house for a number of days without grid support. To date, the owner has had a number of periods where the grid was offline during a blackout, and the system has transitioned automatically to maintain constant energy usage within the home. The longest such period has been for five hours to this point.
The key bit of data I’d like to focus on in this section from the samples I’ve shown above is the estimated annual production figure of 10,483 kWh. This was estimated by Tesla using satellite images of the home, geographical location and surrounding tree coverage, as well as the available square footage of the roof. This is where the story really begins. Now that there was just over a year’s worth of production data, did the real-world application match with Tesla’s estimate? Obviously, it wouldn’t be exact as there will be variations in cloud cover from year to year, but is the system producing close to estimated rates?
My primary goal in this study was to determine the actual rate of electricity produced by the solar roof system from as many independent sources as possible. In this case, that really came down to confirming production levels via both Tesla’s own app that tracks and controls the system, and data from the Electric Utility’s Net Meter. It sounds simple enough, but this is where things went a bit sideways.
Source: Original Tesla Solar Roof Proposal
As shown above, Tesla provides an application that runs on a phone that monitors and controls the system to a degree. Interestingly, this App is the only method available to gather data. There is no method for using a desktop program. Hence, I had to access the home owner’s phone application and figure out where I could access the data. It turns out I was able to email myself a spreadsheet. It took a little poking around to figure out how to access longer term periods than just the recent past. In the end, I accessed daily data that started in early December 2019 while they were testing the system, through to the end of June of this year 2021. The Tesla data provided figures for Home Usage kWh, Solar Energy Produced kWh, a figure for energy taken from the Powerwall storage system kWh, a figure for energy taken from the electric utility Grid kWh, energy sent to the Grid kWh, and the overall Net Grid kWh reading. Lots of data which is great, and it looks like this on a spreadsheet below:
Source: Tesla Application data and my spreadsheet.
Next, I went to the electric utility to see what data I could gather. The utility used a traditional website interface and thankfully did provide the ability to access data for longer term periods beyond just using statements. However, this data was segmented only by weeks and provided just Net Grid meter readings on a per kWh basis. My first problem here was a mismatch in period data. That could be solved relatively easily by converting the Tesla daily data into weeks, but the next issue was the time at which the effective meter readings were taken. Tesla segmented the data effectively at midnight, while the utility used a time that vacillated a bit in the afternoon. Hence, it became clear I wasn’t going to be able to produce a perfect fit to the decimal point to confirm the readings. I also couldn’t get a figure from the utility to show electricity production alone, just the Net Meter reading. How can I determine if the Tesla data for production and usage is accurate? I can compare the Net Meter readings from the Utility and Tesla to see if they match, but suddenly this looked pointless when I began to turn the Tesla daily data into weekly and chart it.
Source: Tesla Application 7 day simple moving average Electricity Production 2/1/20 – 6/30/21
First, I started with February 2020, because the production from the roof was intermittent in the first two months during testing, and I wanted clean data to estimate annual production. The first year looks like what one would expect in chart form, with production rising in the spring and summer period, and then falling into the fall and winter period when the amount of daylight and angle of the sun declines in the northern hemisphere. But what in the world happened towards the end of February 2021? According to the Tesla data the production from the roof more than doubles! Major problem and to make it more difficult, this occurred during a period when no one had been living in the home for almost two months between mid-February and early April of 2021.
I next checked the Tesla Net Grid data since I could compare it eventually with the utility to see if it confirmed a massive spike in production. Unfortunately, it did not, and in fact, the figures look accurate to what I was expecting. Note the negative Net Grid readings when the house was empty means that the roof production was exceeding house demand, and sending electricity to the grid itself. The system is designed to maintain the Powerwall storage first, but after that, any excess gets sent to the grid. The net readings year over year were also more in line with the alterations in weather patterns from sun exposure to total average temperature.
Source: Tesla Application 7 day simple moving average Net Grid readings 2/1/20 – 6/30/21
This started a series of communications with Tesla support to try and find a resolution to the problem now that the data was questionable. At first, this was a bit frustrating. I needed to get help from the owner to get Tesla to communicate with me. The initial response back to be honest felt like a stiff arm and I was a little disappointed. Eventually, persistence did lead to more emails with the appropriate response team, but the resolution was that I was not going to be able to find some hidden database that contained the missing data on the production. The best answer I can provide is the following: In late February while the home owner was away, the application sent a warning saying Tesla wasn’t receiving data from the solar system. A person that helped oversee the building of the roof system for the homeowner, went to investigate during that time period. What essentially was done was a reboot of the wireless internet system that the roof communicated with. My best guess is that this reboot caused the Tesla system to start fully communicating data properly. Recall that I mentioned there are two inverters for each side of the house. I believe that for whatever reason, one of those inverters wasn’t properly sending its data to Tesla. The angle of the house does favor one side of the roof more than the other, so my belief is that the inverter for the favorable side wasn’t communicating properly.
Eventually, an email came to the owner from Tesla explaining the problem as the following:
Hello, Thank you for your patience. The Tier II team reviewed there was an issue with the data that you had observed.
The problem was within our software and this has been corrected so that production monitors as normal. The firmware for the Powerwall has also been updated to the latest version and that should straighten that out.
While I now have confirmed the data was inaccurate up to 3/21, I need to do something to try and get comfortable that any of the data I have is accurate. Since the utility provides Net Grid meter readings as my primary source of historical data, I needed to try and match up the two data series on this metric to gain some comfort. This is where the timing became an issue. At first, I tried to look at percentage changes, and while the periods showed similarities, they weren’t close enough for my liking. In the end, I realized a simple overlay of the charts might do the trick.
Sources: TSLA App and Utility Data
While not perfect, you can see a fairly tight fit of the two series once I transposed the daily Tesla data into weekly. The Tesla data is also slightly post the Utility peaks and troughs consistently, which I view as displaying the timing meter reading issue. Overall though, this is the first point where I began to feel like I might be able to do something with the data I have. Now that we believe the Utility and Tesla data match at least in regards to the net Meter, I decided to use the data on production since 3/21 to at least estimate the annual production overall. It’s not what I wanted. I’d waited to do this until after a year had passed so that I had a clean annual figure, but it’s better than nothing so here goes.
I do have a year’s worth of data from Tesla that we now know is not counting all of the production. I decided to use this as a base and make a simple adjustment to see if it was close to the estimated production. The question is what should the adjustment be? When I started to look at the year over year change after the software fix, the rate of change of course was varied. The spring weather in 2020 where the house is located was particularly cool and cloudy relative to 2021. I was getting rates of change ranging from just over double, to as much as almost 1.5x the previous year. In the end, I decided to use the rate of change for the month of May ’21 – ’20. This was due to the closer approximation of average temperature readings supplied by the utility. As it turned out that was also the lowest year over year rate of change difference of only 91.4%. I then applied it to the base readings of the previous year from Tesla, and here’s how it looks on an estimated annual production basis:
Source: Tesla Application
Obviously, I wish I didn’t have to use a base rate of annual data, and then estimate what the total production would be using an adjustment. However, I am comforted that using the lowest rate of change for one month year over year is giving an answer in the correct ballpark for what Tesla was estimating. Hopefully, I can come back in another year’s time and show a full annual data set that matches my adjusted estimate. Either way, I am more confident now that the actual production of the roof is likely at least matching what Tesla originally estimated in their first proposal. This is key as potential purchasers can at least begin to assume the estimates are fairly accurate, and from there make their own assumptions around usage and costs of installation to determine whether or not it makes economic sense. Let’s move on to try and answer the third question.
Cost and Return Analysis:
There are estimates, and then there is reality. In the world of home building, the two are rarely the same, and the latter is almost always more than the former. This case was no different from that norm, as unforeseen issues cropped up to add to the total cost of installation. First, let’s take a look at the original estimate from Tesla. Below was the gross estimate, meaning excluding the potential tax credit the owner could receive from the government.
Source: Original Tesla Solar Roof Proposal
The owner decided to pay a third party to oversee and manage the installation process. Thus, it was known from the start that the total cost would be more than the Tesla estimate. However, at a later time, it was discovered that there was a problem with the original gutters installed on the house. The gutter system had copper elements which would become a risk to the house if in too close proximity to the roof’s power system. Thus, a new non-conducting gutter system would need to be installed increasing the total cost.
Source: Picture of new gutter system taken by author.
To cut to the chase, the total cost including the tax credit was just over $98,000. Tesla’s invoice was just about $6,000 more than the original estimate which isn’t too bad all considering. The cost of the new gutter system was included in the Owner’s Manager’s invoices. That also includes other repairs as well like drywall repair after installing conduits etc… The tax credit could only be applied to materials for the solar roof itself. Hence, it came in below TSLA’s original estimate by about $5,000, and none of the replaced gutter system costs could be offset.
Source: Owner’s Invoices and Accountant.
One last important part concerning the production level of the roof that we haven’t discussed is the path of depreciation over time. Tesla guarantees that after 5 years the roof’s production will be no less than 95% of the original annual estimate, and will decline no more than .5% annually for the next 25 years. The roof’s warrantied life is for 30 years in total and will be at least 82.5% of rated peak power at the end of the 30th year. This is important when estimating our likely return on the roof over its 30-year warranty.
Source: Picture of fully installed TSLA solar roof taken by author.
Now that we have confidence in the production figure and estimated rate of decline as well as the total cost of installation of the system as a whole, we need to look into the value of the produced electricity over the 30-year warranty of the roof. In the original estimate provided by Tesla, we find a statement that the company used a 2% expected rate of inflation for the value of electricity.
Source: Original Tesla Solar Roof Proposal
I wanted to check to see if this was a legitimate assumption to use for the home owner’s area. From a nationwide cost per kWh basis, the home owner’s value of electricity is among the highest in the nation, but what about the rate of inflation? According to the US Bureau of Labor Statistics, the compound annual growth rate of the region’s cost of electricity over the last 30 years was 1.82%. That’s close to Tesla’s 2% estimate, but I did use the 1.82% figure in my return analysis that follows. However, as you can see on the chart below, that inflation rate had a large range to it. Based off of the recent fairly static levels of cost in the immediately preceding decade, I’d assume the home owner is likely to see higher rates of inflation than the 1.82% estimate I’m using in the near future.
The next phase of this analysis was more complicated than I expected. I wanted to use a real-life analysis, and I found that using a regional kWh figure from a government agency didn’t capture all of the added fees we actually pay in our utility bills. The home owner allowed me to analyze all of the statements from the utility received to date. What I realized is that the solar roof production is effectively offsetting the Billed kWh from the utility, but they don’t provide us with the key thresholds where Delivery & System Charges are fixed for example. In the analysis below, you can see how the different parts of the statement are affected by the monthly Billed kWh. Remember that these Billed kWh are net of the roof’s production. Hence, you see the figure go to zero in March and April of 2021 due to the house’s vacant status.
Source: Utility Statements
As you can see, even when Billed kWh went to zero, there still were some monthly charges. In the end, I decided to use the median average of .2330 per Billed kWh as the starting point for determining the real-world total value to the home owner of the roof’s electricity production. Using the median versus the mean removes the extreme readings influencing the starting point. Now let’s put it all together.
Sources: Utility Statements, Tesla Original Estimate, BLS Regional Electricity Inflation Rate.
That $86,961 is awfully close to Tesla’s original gross cost of installation which would have left the profit produced to whatever degree of tax credit the home owner could claim. Of course, there were more costs, and the home owner chose to add on to them by having an external manager oversee the installation. Over the 30 year period for this home owner, I’m comfortable saying they’re likely to receive about 3-3.5% ROIC on their investment in the roof in terms of electricity produced. If it were based on Tesla’s gross estimate, then the ROIC would have been about 3.5-4%, and if Tesla’s original net of tax credit estimate were used, then the ROIC would have been more like 4.5-5%. That’s a big delta between what’s likely to be, and what could have been if the gutters didn’t have to be changed. If you read the article I linked to at the very beginning, then you’ll realize this case study is a prime example of what the company has struggled with since launching this product.
Is a 3-3.5% ROIC return worth it? If you asked me that twenty years ago it would have been an easy answer, but at a time when 30-year treasury rates are less than 2% with high-yield indexes at just over 4%? I leave that up to the consumer to decide. My primary interest was in determining if the production matched the estimate provided by Tesla. That answer I am comfortable with in the affirmative, although I went through significant pains to determine the accuracy. My simple return analysis also doesn’t include intangible benefits of the system’s power storage, as the home is in an area where power loss from the grid does occur frequently. Installing gas-powered generators in this area as a backup source is also forbidden by the town now. Hence, it certainly can be argued that the reserve energy system provides worthwhile value that I have ignored in this analysis.
I’m also conservative when dealing with estimates if you haven’t noticed, and a tweak here and there changes the return analysis to the upside. For example, the actual compound rate of inflation for the utility cost only needs to be 2.65% over the next 30 years to cover the home owner’s actual net cost of $98,000. I’d easily take the over on the next 30 years’ inflation rates versus the last 30. Also, if I used the mean average of .2423 kWh cost instead of the lower median as my starting point, then the total value shoots up another $5,000. All this is to say we can’t escape from using assumptions in return analysis, and it doesn’t take much to move the needle at these lower bounds.
The return analysis also assumes the system vanishes in 30 years which is likely not the case. In fact, I am told that the roof structure would last the lifetime of the house. In fire prone areas, the lack of flammable material on the roof could reduce the risk of catastrophic loss from wildfire, as most houses catch fire from burning embers landing on their roofs. One downside of the sturdy glass coating of the tiles though, is that Tesla has not provided a method for managing wet snow on the roof. The home owner explained that it tends to slide off the roof in large clumps, so one needs to be careful walking close to the sides of the house after heavy snowfalls. After 30 years the roof will be producing less energy, but the value of that production is likely to be higher due to inflation. On the other side of the coin, my analysis hasn’t addressed the issue of the Powerwall storage system, and the need for future replacement due to battery fatigue. I.e., as you can see one can improve or detract quite easily from the return analysis figure depending upon how and what they calculate. Essentially, the total return is tight, because the cost of installing the system varies from house to house. This is going to be a hard nut to crack for Tesla. The system itself though is working as designed in terms of production. Perhaps only installing on new construction would reduce the cost of installation enough to increase the ROIC of the system materially.
If you actually read this article to its end, then I hope you found it at least informative and interesting. Please keep the comments below civil. I will try my best to answer any question regarding the case study specifically, but I don’t wish to venture into any greater discourse among the faithful of the two opposing sides. All the best.