Near-future quarterly financial projections

[mongo]

It's time to quantify the average realized gain Tesla might have were they to auction 2019 Model 3 LR coming off a 36 month lease. Obviously the clean ones they'll resell for higher prices than I now will show.
2019 Base Price $45,700
2022 current Mannheim Atlanta average 36,000 miles, clean $48,500
Tesla 2019 Model 3 LR typical residual value: $23,307
for these terms there will be no addition consumer charges because they did not exceed mileage allowance, no damage, no options.
It is almost absurd to realize they now can just auction those cars and take around $25,000 gain on each lease return. This probably cannot last too long but for 2022 if Tesla want to have some easy quick margin helpers all they need to do is sell off lease returns.
My calculations built from Atlanta because my Mannheim access is there.
This is such a large unit gain that I think we should quickly try to establish what lease return volumes are expected the rest of 2022.

I have not attempted to do that yet. Perhaps someone has a better approach than I would use, which is just extracting numbers from securitization pools.

Per P&D report, Q3 2019 had 8% of 79,600 Model 3s leased, so around 6,000 3 on lease.
Q4 2019: around 8,000 leased.

 

[unk45]

Just when second half can use the boost to lift off! An extra few hundred million.

 

[Doggydogworld]

Per P&D report, Q3 2019 had 8% of 79,600 Model 3s leased, so around 6,000 3 on lease.
Q4 2019: around 8,000 leased.

I believe Tesla's "subject to lease accounting" category includes a lot of 3rd party leases in Europe. Tesla doesn't own those cars and is probably not in a position to gain from their resale. Changing the terms of their direct leases has worked out incredibly well, though, even though the Robotaxi fleet is still...... not quite operational.

 

[Lance Boyle]

I'm pretty confident that won't happen until 2023. Single stack needs to be in beta for awhile.

True.

 

[unk45]

I believe Tesla's "subject to lease accounting" category includes a lot of 3rd party leases in Europe. Tesla doesn't own those cars and is probably not in a position to gain from their resale. Changing the terms of their direct leases has worked out incredibly well, though, even though the Robotaxi fleet is still...... not quite operational.

The German, UK, Netherlands rules are each quite distinct. I am not familiar with the Norwegian rules which probably are relevant. Depending on several unknown variables, I suspect most of the UK leases are outright sales to lessors, but tesla may have made arrangements to buy back the cars. Without knowing the precise agreements Tesla has with others it is not possible to know if they'd be eligible for lease accounting or not. There are also many fleet sales that began to happen around 2019-2020. Those too might be subject to lease accounting, but the specific deals will be determinative. Although not immediately relevant for 2022 statements, Hertz has no resale value obligations nor financing role with Hertz. Such agreements are common with other OEMs, but Tesla has resisted that.

 

[EVNow]

For some reason I was only focused on Q2 when I wrote that response. Yes, this could add up to a good amount of money this year.
There is upside with FSD as well. If "Autosteer on City Streets" gets wide released this year, there is a large amount of deferred revenue to recognize which I do not have in my forecast.

I'm pretty confident that won't happen until 2023. Single stack needs to be in beta for awhile.

Single stack is a distraction.

Autosteer, when it goes wide beta I.e. not early access as it is now will sati requirements to recognize revenue. This iswhat Tesla did after releasing traffic light / stop sign recognition.

But the quality (in terms of miles per disengagement) is still very poor for FSD. IMO needs 10x improvement for wide release and 100x or more for human level.

Elon has claimed both will happen this year - are they tied to each other ? Is the bar on wide release lower ? Needless to say 100x improvement isn’t happening this year … or for that matter 10x. Also note that frequency of releases has dramatically slowed this year - from 2 weeks to 2 months.

 

[kengchang]

Elon has claimed both will happen this year - are they tied to each other ? Is the bar on wide release lower ? Needless to say 100x improvement isn’t happening this year … or for that matter 10x. Also note that frequency of releases has dramatically slowed this year - from 2 weeks to 2 months.

Wouldn't surprise me that the release cadence slow down is due to NHTSA approval

 

[mongo]

Wouldn't surprise me that the release cadence slow down is due to NHTSA approval

NHTSA doesn't sign off of FSD software updates.
Even FMVSS is based on self-certification.

 

[Gigapress]

NHTSA doesn't sign off of FSD software updates.
Even FMVSS is based on self-certification.

Moreover, I spent hours researching this a couple months ago and found out that the Federal Motor Vehicle Safety Standards don’t even have any regulations on level 2 software (automated driver assistance systems) nor even level 3-5 (automated driving systems). In the US it’s all regulated at the state and local level for now.

 

[mongo]

IMO, Gary's $180M restructuring charge is very high for the 10% salaried headcount reduction.

Yeah, it occured one month before end of quarter, so the max hit based on layoff law (60 day notice or pay thereof) would be an additional 1 month. But maybe the severance is more generous.
Will have sone impact on stock compensation items.

 

[The Accountant]

Thanks! Most of the time the mode and the median are near each other and I guess the mean is slightly to the right of them. But just as a fun hypothetical, lets assume for some strange reason we have this strange near uniform distribution:
EPS=[0 50 50 50 ... 50] ($)
probability=[1.001 1 1 1 … 1 0.999] (%)

Ie the mean and median is ~$50 and the mode $0. Would you still give your estimate of the mode at $0, knowing that there is 99% chance that you are too low? Or would you make an exception and give the mean/median?

Another question, if TSLA has 10 events that each has a probability of 40% of happening with the same outcome, would analysts model 4 of them happening at 100%? Or zero of them happening?

This is based on my experience. I worked in the consumer products industry and the pharma industry where the approach is different.

With consumer products, the mode is used for both short term and long term forecasts. Things are binary. You include the event 100% or you exclude it 100%. This is because there is not much uncertainty with events in this industry. You will launch the new raspberry seltzer flavor or you won't. You will expand to Latin America or you won't.

In the Pharma industry, they often use the binary (in or out) method when doing a current quarter, current year forecast.
But when analysts complete the long range forecast to value a company (10 years out), they probablize the outcomes of drugs in the pipeline.
Drug A has a 70% probability of launch in 2024, drug B 40% in in 2025, etc.
In this case, the analyst takes 70% of the revenue and profits of Drug A and 40% of Drug B into the forecast. They do this for the entire pipeline which could be up to 20 to 25 drugs. The outcome is eventually binary. It will be either 100% or 0% (if the drug fails in clinical trials).
With your example, an analyst will take the 10 events and forecast the revenues/profits at 40% for each. If 4 of those events occur and 6 fail, they hit their forecast (assuming revenue and profit potential is the same for each drug).

 

[heltok]

This is based on my experience. I worked in the consumer products industry and the pharma industry where the approach is different.

With consumer products, the mode is used for both short term and long term forecasts. Things are binary. You include the event 100% or you exclude it 100%. This is because there is not much uncertainty with events in this industry. You will launch the new raspberry seltzer flavor or you won't. You will expand to Latin America or you won't.

In the Pharma industry, they often use the binary (in or out) method when doing a current quarter, current year forecast.
But when analysts complete the long range forecast to value a company (10 years out), they probablize the outcomes of drugs in the pipeline.
Drug A has a 70% probability of launch in 2024, drug B 40% in in 2025, etc.
In this case, the analyst takes 70% of the revenue and profits of Drug A and 40% of Drug B into the forecast. They do this for the entire pipeline which could be up to 20 to 25 drugs. The outcome is eventually binary. It will be either 100% or 0% (if the drug fails in clinical trials).
With your example, an analyst will take the 10 events and forecast the revenues/profits at 40% for each. If 4 of those events occur and 6 fail, they hit their forecast (assuming revenue and profit potential is the same for each drug).

Thanks! It just seems a bit strange that they call something an estimate, without saying specifically what they are estimating. If it’s the mean, the median, the mode etc.

Using different methodologies to estimate the EPS for next quarter and for next year seems a bit strange, ie to discard 49% events in the short term, but include them in the long term estimates.

I guess the question is what is the goal of the “estimate”. Is to be used to model the value of the stock? Is it to win a guessing competition? Is it to be the least wrong among a group of peers over a time period? Is it to change the stock price in a specific direction? And I get the feeling that they will first decide on an estimate and then try to rationalize it by picking things to include and not.

I liked ARK’s Monte Carlo model, but I guess it’s hard to print it in a headline and most people just a final number. Still a lot of relevant data is lost in this and a bias is intruduced where the short term estimates tend to often be about right, but sometimes be lower because they didn’t include low probability events.

 

[Chunky Jr.]

@heltok said...

Thanks! Most of the time the mode and the median are near each other and I guess the mean is slightly to the right of them. But just as a fun hypothetical, lets assume for some strange reason we have this strange near uniform distribution:
EPS=[0 50 50 50 ... 50] ($)
probability=[1.001 1 1 1 … 1 0.999] (%)

Ie the mean and median is ~$50 and the mode $0. Would you still give your estimate of the mode at $0, knowing that there is 99% chance that you are too low? Or would you make an exception and give the mean/median?

My statistics is a little rusty, but isn't mode defined as the most common value in a data set, so in the above example the mode would be 50, not 0. Or am I missing something?

 

[heltok]

@heltok said...


My statistics is a little rusty, but isn't mode defined as the most common value in a data set, so in the above example the mode would be 50, not 0. Or am I missing something?

I made an edit in the post under, then the post was moved but not the edit. Meant to have a uniform distribution [0 1 2 3 .. 99]

 

[petit_bateau]

Firstly many thanks to @Troy for passing me some historical production data splits by factory and by type. That doesn't drive my financial modelling, but it is of considerable interest to me. Ordinarily I can glean it for myself but if I am travelling then I miss things. Once again thanks to @Troy.

I'm having a proper update of my forecasts which I last did 12-months ago and apart from TSLA being a screaming buy at the prices it has been of late, what I am observing is that my previous cautious forecast of the future tax take may become significant. My quarterly model drives the 2022 and 2023 forecast, whereas the 2024+ are simply annual numbers. For the quarterly model I simply roll forwards the latest quarter's effective tax rate, i.e. approx 12%. For the subsequent years I have up until now applied a 25% rate as being a cautious OECD average. This has the effect of notching into (forecast) shareprice growth as you can see (I try to figure out a shareprice that is a sensible progression for PEG, PE, PR and both fwds and trailing). So ......... does anyone want to give me their view on what might be the average tax takes in the years to come ?

Needless to say the below is DRAFT whilst I carry on playing around with this and other aspects.

 

[The Accountant]

Firstly many thanks to @Troy for passing me some historical production data splits by factory and by type. That doesn't drive my financial modelling, but it is of considerable interest to me. Ordinarily I can glean it for myself but if I am travelling then I miss things. Once again thanks to @Troy.

I'm having a proper update of my forecasts which I last did 12-months ago and apart from TSLA being a screaming buy at the prices it has been of late, what I am observing is that my previous cautious forecast of the future tax take may become significant. My quarterly model drives the 2022 and 2023 forecast, whereas the 2024+ are simply annual numbers. For the quarterly model I simply roll forwards the latest quarter's effective tax rate, i.e. approx 12%. For the subsequent years I have up until now applied a 25% rate as being a cautious OECD average. This has the effect of notching into (forecast) shareprice growth as you can see (I try to figure out a shareprice that is a sensible progression for PEG, PE, PR and both fwds and trailing). So ......... does anyone want to give me their view on what might be the average tax takes in the years to come ?

Needless to say the below is DRAFT whilst I carry on playing around with this and other aspects.

View attachment 831446

For my model I used:
20% for 2023
22% for 2024
24% for 2025
25% for 2026 and thereafter.

However, I will likely take these numbers down in my next update.

Piper Sandler uses 20% for all years in their valuation. I will send you a copy of their Jan 2021 valuation model in a private message.
Tesla will benefit from higher tax deductions (vs book) for stock compensation for several years and they also have US NOLs to utilize.

Here are Amazon's effective tax rates for the past several years:
20.2% 2017
10.6% 2018
17.0% 2019
11.8% 2020
12.6% 2021

 

[mongo]

For my model I used:
20% for 2023
22% for 2024
24% for 2025
25% for 2026 and thereafter.

However, I will likely take these numbers down in my next update.

Piper Sandler uses 20% for all years in their valuation. I will send you a copy of their Jan 2021 valuation model in a private message.
Tesla will benefit from higher tax deductions (vs book) for stock compensation for several years and they also have US NOLs to utilize.

Here is Amazon's effective tax rate for the past several years:
20.2% 2017
10.6% 2018
17.0% 2019
11.8% 2020
12.6% 2021

Humm, I think linear is the wrong appoach since China and Berlin are taxed, but US is not until they burn through the loss carryforward. At which point we get the tax asset bump. (Or are those the same thing? With the asset being the tax on the net loss?)
Either way, big discontinuity at some point.

 

[petit_bateau]

Thanks re tax stuff. My ex was an intl tax lawyer so I know how absurdly low this can go, but I try to be conservative. Let me ponder and reissue.

When I reissue I will eliminate another error I picked up of 4x in the fwds PEG, just saying ....

Once again thanks all for your PMs.

 

[petit_bateau]

Once again thanks for the various PMs. Ultimately I went with a ramp in tax, making the tentative assumption that there are multiple countries involved + continuing investments that attract tax benefits + many tax treaties, and that they would combine to avoid a pure step-function cliff-edge. The annual numbers I assumed in the end were:

For quarterlies here is my latest forecast:

 

[The Accountant]

I have estimated that the Berlin/Austin Ramp Impact on COGS was about $905m in Q2 (when compared to Q1). That's about $0.75 per share (after tax).
This margin drag will continue for several more quarters (at decreasing amounts) until I believe Q2 2023 when it will likely not have a material impact on COGS. Perhaps the impact will be for example $750m in Q3, 500m in Q4 and down to 200m in Q1 next year.
The good news is that the expected efficiencies coming from Shanghai over the next few quarters will partially offset the Austin/Berlin Ramp impact.