Diving Into ARK Invest’s Bear Case On Tesla [TSLA]
I got a few requests from my last post to do a deeper dive into ARK’s Tesla bear case after providing my thoughts on its bull case.
Breakdown of the Model’s Assumptions
One of our readers, Pitounet, pointed out that ARK has hosted its model on Git. This has a lot more information on ARK’s Monte Carlo analysis than the blog post. You can find the model here, and download it for yourself as an Excel file. Thanks, Pitounet!
There are four main tabs to change: Tesla Valuation, Valuation ASP Tables, Monte Carlo Inputs, and Monte Carlo ASP Tables. The one tab I found interesting was the Monte Carlo Inputs tab. This tab limits the minimum and maximum for the simulation, and shows us what ARK set as the defaults for each value. I’ve selected a few rows where the bear and bull case are far from the possible minimum and maximum values. I find it interesting that so many key values are far from the minimum and maximum values, and how others are tightly constrained (such as the factory utilization factor being between 95% and 99%, out of a range of 70% to 100%). For ride-hailing service, the miles per year averages to 30,000 miles per car. Another point of interest is miles traveled per robotaxi, with the average set to 110,000 miles per year. I plan to reset the values to the minimum and maximum to see what changes, as an experiment. The bear case for max annual production increase is 50%, which I find odd and implicitly bullish.
Let’s take a look at what the model shows us when I reset the bounds.
The revised model gave me the following results for the stock price. Yes, it’s not a joke that the max price was more than $542 thousand dollars a share, with a minimum of -$1,962 per share.
I thought that was silly, so I capped everything below 0 to 0 and everything above 20,000 to 20,000. This gave me the following. The average and standard deviation of the bounded data makes more sense. The standard deviation is high as a proportion of the average, which means the data are widely spread from the average and not closely clustered around it.
The distribution of the stock prices looked like the following. You can see there is a significant number above $20,000 (28 simulations in fact).
Cars Sold and ASP
Next, I looked at 2025 cars sold and ASP. The average estimate of cars sold in 2025 was 8.6 million, with an average ASP of $45,345. This gives us a 2025 electric vehicle revenue of $390 billion! When I looked at the data closer, I found 453 records with more than 20 million vehicles sold by 2025. This was Tesla’s stretch target for 2030, and I removed all of those records. After removing those records, that gave me an average of 5.6 million vehicles sold, at an ASP of $47,169, and an estimated $266B in 2025 electric vehicle revenue. ARK had a target of $234B by 2025, with 5 million vehicles sold at an ASP of $45,000.
Let’s break down ARK’s bear case further. For Tesla to reach 5 million vehicles by the end of 2025, it needs to grow vehicle sales by a compound growth of 58.4% a year. The market is expecting vehicle growth of 750,000 to 900,000 this year. Of the three major revenue numbers, 5 million vehicle sales looks like it can be achieved, with revenue dependent on the cumulative vehicle production to determine ASP at the time.
Once Tesla finishes Austin and Berlin this year, it could start three factories next year, with a capacity of 500,000 each, and finish in 2023. Ramp up time looks to be 2 years, judging from the growth at Shanghai. Starting with close to one million this year, another three factories early in 2024 would get Tesla there. This is certainly a stretch, but I think Tesla will be close.
From recent research covered by Johnna Crider, researchers estimate pouch and cylindrical cells are going down in cost by 20% to 24% per doubling of capacity. Automotive revenue for Tesla last year was $27.236 billion, and with 500,000 vehicles sold, that gives us an Average Selling Price of $54,472. If Tesla meets Wright’s Law for batteries twice over the next 5 years and decreases the ASP of $54,472 two times by 20%, we get a 2025 ASP of $34,862. Multiplied by 5 million vehicles, that gives us $174 billion in revenue. This is an approximation, since the batteries are not the whole car, but a major part of it. I am excluding everyone else in the industry, but it’s possible we see Wright’s Law triggered a third time in 5 years with industry growth. My estimate is far below ARK’s bear case example of $234 billion in revenue. It also presumes Tesla can grow 58% a year, which as I said, would be phenomenal.
When I ran my simulation, I came out with an average of $11.282 billion for insurance revenue. This is far below ARK’s estimate of $23 billion.
I checked my auto policy at State Farm. State Farm is charging me $94.57 a month, for an annualized rate of $1,134.84. If we assume everyone is paying something similar and switches to Tesla Insurance, for 14 million vehicles, that comes out to $16.1 billion. This amount should be reduced as Tesla safety improves and ASP comes down. The extra $7 billion are charges due to ride-hail insurance coverage. The charges for ride-hail seem excessive to me, since commercial auto or ride-hail policies have been around for many years, due to the popularity of Uber, Lyft, and other driving services.
- Tesla Insurance may not be available in all states and countries
- Not everyone will switch to Tesla Insurance
- Tesla seems to have a profit and loss sharing agreement with independent insurance companies
- With more Teslas on the road and better safety data, insurance costs should come down due to lower accident rates and being easier to fix, pressuring insurance revenue potential
- Tesla wants to make it more convenient and cheaper to get insurance, not necessarily to make money off of it (right away)
- A loss ratio of 80% to start with would be amazing. New insurance programs have a combined loss ratio well above 100%, from my experience.
Human-Driven Ride-Hail Revenue
In my Monte Carlo simulation, my results gave me an average of $6.277 billion for human-driven ride-hail revenue. ARK estimated $42 billion. ARK does a good job of changing assumptions on per mile charge, total vehicles, and % in the fleet.
Let’s take a look at revenues for Uber, Lyft, and Doordash over the last year. This will give us an idea of the potential for Tesla human-driven ride-hail revenue.
To make things easy, everyone else doesn’t lose revenue, and Tesla gains $42 billion in ride-hailing and delivery revenue by 2025. That grows the total market to $62 billion, for a compound growth of 25.8%. Compared to delivery numbers, this seems very reasonable, especially as the world economy grows coming out of the pandemic. But, and this is a big but, if three of the established leaders only pull in $20 billion in ride-hailing and delivery revenue today, is it reasonable for Tesla, with no network advantages, no app at the moment, no ride-hailing service, to ride in and get $42 billion in revenue in 5 years from scratch?
$42 billion would be bigger than selling 500,000 electric vehicles per year. Think about that, and the hard work it took Tesla over a decade to get to that level. Tesla has done amazing things before, but this seems aggressive for ARK’s bear case. If you are a ride-hailing driver, would you prefer Uber, Lyft, or Tesla? Uber and Lyft will pay you more and have better chances of getting a fare. If it was me, I would pick Uber or Lyft all day, unless Tesla sweetened the pot and gave me a higher cut of the total. That cuts into Tesla’s profits, since revenue will be lower based on the lower per mile charge. I can’t judge how much ride-hail revenue there might be. The range of scenarios is too wide, including when Tesla starts a service, how many Tesla vehicles use it, how much Tesla charges for it, and compensation. It’s less than $42 billion, though.
What’s missing in the model?
- Battery supply is a big variable. It drives everything else.
- Energy storage is a big miss. I think 100% growth per year for the next decade is reasonable, provided battery cells are available.
- Tesla Semi and Roadster are missing — they are niche products, but they soak up cells and have six-digit average selling prices.
- For Volkswagen and Toyota to achieve 10 million in sales, it requires many brands, 4–8 volume models, with each model having 2–4 different body versions (such as sedan, hatchback, station wagon, coupe, etc.). Each body version has different powertrains, trim levels, and other options. ARK’s model needs to address these different markets and the resources necessary to create the models and versions. These are real-world limitations. Thanks to Maarten for breaking it down for me.
- Potential number of factories, ramp up time, and maximum factory capacity. This would be a better proxy, by year, on Tesla’s possible vehicle growth. It’s probably more accurate to think this way than Tesla magically growing vehicle sales because it is in a spreadsheet.
Thoughts on a Robotaxi
Thanos: I am inevitable.
Robotaxi maintenance: Hold my beer.
Suppose a robotaxi travels 100,000 miles a year. I am in the market for new tires. I asked our local tire shop Discount Tires how long the Michelins last on the Model 3. The person assisting me said Model 3 tires can wear out as fast as 8,000 to 9,000 miles to 20,000 miles due to the vehicle weight and torque. My set at 16,000 miles was one of the better ones he had heard about.
For 100,000 miles, tires will need to be changed 5 to 12 times per year. Good tires cost around $800 per four tires. That’s a cost of $4,000 to $9,600. Above this, we need to include exterior clearing, interior sanitation, air pressure checks, tire rotations, and other maintenance for our robotaxi. Electricity at 10 cents a kWh is $2,500 per year, assuming 4 miles per kWh. At 20 cents a kWh, the cost is $5,000 a year. Insurance, licensing, car payments, and depreciation are extra.
The $1 per robotaxi mile I mentioned in the last piece is too high. At that level, Tesla will steal share only from Uber and Lyft. For those who can afford a $1 per mile vehicle, many are likely not wanting to use it as a robotaxi. By having cheaper fares, it will lead to market expansion, but at an unknown level. Suppose Tesla is able to charge 50 cents per mile, and 25 cents per minute of traffic. My commute is 20 miles with 40 minutes of waiting. My commute would cost $20 one way, plus $4 for tolls. Round trip is $48, or $9,600 per year for 8,000 miles a year. That is a cost of $1.20 a mile. Once my Model 3 is paid off in a few years, my estimated cost is $2,400 per year, over 12,000 miles, and skip the tolls since I will drive locally. That is an all-in cost of 20 cents a mile. It’s a very individual case but true for many people living in the suburbs and commuting into city centers. The pandemic may reduce the number of trips we need to make to the office, since we may work from home a few days a week, further reducing future robotaxi demand.
The biggest competition for robotaxis are used EVs and gasmobiles. I think $0.20 to $0.50 a mile is feasible for mass market expansion. At the lowest rate, revenue is $20,000 a year. Variable costs are estimated at $6,500 to $14,600 a year (tires and electricity). Add in the cleaning costs, other maintenance, vehicle payments, depreciation, and ride-hailing insurance, pay taxes on the profits, and potentially not much profit is left.
The cheapest robotaxis will emphasize comfort and efficiency. Get rid of the steering wheel, brake pedal, accelerator pedal, some of the seats; lower the weight of the wheels; increase durability of the tires; use narrower tires; with smaller width tires, reduce the vehicle size and weight of the battery. Competition will increase the number of robotaxis and lower the amount each one finds work over time. High profit levels will increase competition for buying an autonomous vehicle, reducing profit margin. There’s a good chance that, like airlines, autonomous vehicles may charge you for your luggage or give you an allowance.
Here’s a summary of the three approaches. If I could create a Monte Carlo simulation template for any EV manufacturer with the things missing above, I might do it and publish it on CleanTechnica. The financial parts, I am not as interested in, since those are SWAGs anyway. ARK’s model has promise and a lot of engineering went into it. Some fundamental things are missing, though, such as model diversity, battery supply, actual factory start dates, factory years to scale, and factory max production capacities. An NPV analysis of free cash flow tied back to vehicle and energy storage growth would be key for financial simulations. The number of share price scenarios above $20,000 and the high number of scenarios for vehicle growth above 20 million in five years leave me concerned on the validity of the model. There seems to be some correlation between some of the variables that could be eliminated or reduced. ASPs seem to be related to vehicle growth and which vehicles are produced, as an example. Battery supplies drive how many vehicles can be produced and the mix.
Pitounet had a great comment from the last piece I wanted to share with all of you. This commenter has great knowledge of Monte Carlo models from their work in finance. “I think the bear case is even more attackable than the bull case. What you will see is 95% utilisation of Tesla factories as a bear case and the strict minimum is 70% and 100% is the maximum. I don’t know for you, but I find this to be very skewed. I’m sure most people who could call 70% the bear case, 95% is a bull case and 100%, well that is quite high…
“Another example: Max annual production increase: Bear case: 50%. Tesla communicates on 40%, that means that Ark’s bear case is above Tesla’s central case?
“Finally, I might have missed something but their model supposes the market is infinite. The only limitations of their model is Tesla’s execution to produce cars efficiently, there is no competition, no market resistance. As a result in the 5000 example paths given by Ark, there are 7 above the expected world car market for 2021, the 20th percentile is 4 millions cars a year, the 80th percentile is 12 millions cars a year, the maximum is 130 million cars a year, which I think you will agree will not happen, but he minimum is 1.2millions a year, this case has a very low probability but could happen in case of catastrophic events.
“So for me the model is extremely skewed.”
Thanks for reading!
Disclosure: I own 3 shares of Tesla as of today’s analysis. I have no plan to add or buy Tesla shares for the next 72 hours.