Subscribe to our blog :

Posted on 15 Jun 2017 by Steve Johnson

A Bayesian on Uber Suggests Less than 40% Chance of Failure

A Bayesian on Uber Suggests Less than 40% Chance of Failure

 

What are the odds Uber is alive in 10 years’ time? Just 1%, according to Hamish Douglass of Magellan.

My guess is he was making a point using hyperbole – 1% is a very low number. But it is an interesting question. And one that we can learn a lot from trying to answer.

On a recent holiday I finally got around to reading Superforecasting: The Art and Science of Prediction. It’s the best book I have read in a very long time (thanks Chris), and it clarified something I have been thinking about for many years.

Bayes Theorem, a slightly obscure corner of the world of probability theory, gets a chapter in the book. We have written a few Bayes Theorem blogs but I’ve never really been able to succinctly summarise why it is so important. Sure, the results can be counter-intuitive. But why does everyone seem to think it so important that it gets its own chapter in a book (Nate Silver’s excellent Signal and the Noise was half Bayes Theorem)? Superforecasters answered that question for me.

My summary is this: the importance of base line probabilities is often overlooked by investors. Our brain wants to jump straight to a conclusion based on the most recent information we have, and that can be a big mistake.

Here’s a reminder of the text book example. Assume one in 10,000 people in the population has a rare disease. A test is available from the local chemist that is 95% accurate. You take the test and test positive.

Time to freak out, right?

Still low chance

No, it’s not. Yes the test is fairly accurate. But the starting probability is so low that even those who test positive are still unlikely to have the disease. Imagine 10,000 people take the test. Only one of them actually has the disease (95% likely, but not certain, to be accurately diagnosed). Of the remaining 9,999, 5% of them are going to get an incorrect positive result. That is approximately 500 people testing positive for every one who actually has the disease (the exact probability has changed to 0.19% for those who want to look up the formula and use a calculator).

We don’t need to worry about the maths to use the principle effectively. Most real world probabilities are not exact numbers. The key here is to make sure you think about base line probabilities before you jump to conclusions.

So is Uber going to be bust in 10 years’ time? Ignore the temptation to think about driverless cars and how much of a twit CEO Travis Kalanick is. Let’s try and think about base line probabilities first.

Start with what percentage of companies go bust. A quick Google search tells me roughly half of all new businesses in the US fail in the first five years and one third survive 10 years or more. That means just 33% of the survivors fail between years five and 10. Uber is 8 years old so, assuming the failure rate continues to decline, I would estimate that roughly 25% of businesses that make it to 8 fail within the next decade. Now we have a useful starting point from which to incorporate Uber-specific information.

Leave Uber for last

The company apparently generated US$6.5bn in gross revenue last year on more than US$20bn of bookings (Uber keeps roughly 30% with the rest going to the driver). That’s a big business, which dramatically increases the chances of survival. I would probably adjust the 25% down to 10% thanks to Uber’s growth and size.

It is losing money, though. Apparently more than $2bn last year. So that increases the chances of failure if shareholders refuse to keep funding it. My personal opinion is that Uber could be extremely profitable tomorrow if they stopped trying to grow, but we’ll leave personal opinions out of it for now. Let’s say its current losses increase the failure likelihood back to 20%. Driverless cars? That’s a risk, but also a potential opportunity. Bump it up to 30%? The Kalanick factor? He is definitely a twit, but how much does that increase the failure rate by? A couple of percent perhaps?

All of this is completely subjective, of course, but it does help make a point. We aren’t getting anywhere near 99%. There are many factors to contemplate, but the base rate is uber important.

A Bayesian on Uber Suggests Less than 40% Chance of Failure
A Bayesian on Uber Suggests Less than 40% Chance of Failure  A Bayesian on Uber Suggests Less than 40% Chance of Failure  A Bayesian on Uber Suggests Less than 40% Chance of Failure  A Bayesian on Uber Suggests Less than 40% Chance of Failure

16 thoughts on “A Bayesian on Uber Suggests Less than 40% Chance of Failure

  1. What happens if you start with a different baseline – say, survival rate for companies with operating losses of $2B? Or companies that have lost money for 8 years straight? Is the outcome different, or does it all balance out when you add the other factors back in?

    • As long as you can get an appropriate sample size, the more data you can find about similar businesses the better. Tech companies, more than $1bn of revenue, growing at astronomical rates, substantial annual operating losses. If you can find the data, that would be a better base line. There is no point in me spending 2 weeks doing the work, but my guess is you will find a

        lower

      failure rate among that cohort than the population average.
      Salesforce and Xero are the two most obvious I can think of. Nothing but losses so far, but these are wonderful, wonderful businesses. The shares are expensive, because the market knows it, but I see the way Salesforce has become ingrained into our little business and I see a very long annuity stream with almost no marginal cost of us as a client.

  2. I suspect Hamish assumes private funding markets at ridiculous valuations will dry up very soon. In that case, Uber might not have enough time to stop growth& losing money before it runs out of cash – hence the ponzi analogy. Also it has surprisingly low barriers to entry and the 20% take-rate will diminish as competitors catch up. Very unlikely that the return on invested capital will be acceptable once stable (however that is a different question to default!), again ponzi analogy comes to mind where a $ invested returns less than a dollar in value.

  3. You might find that describing Travis Kalanick as a “Twit” rather than a “Twat” is less genitalially confrontational.

  4. Perhaps the question is not ‘what percentage chance is there that Uber goes bankrupt in the next 10 years’ but ‘how much does a certain fund manager benefit from attracting reader interest with his hyperbolic claims?’

    Listcorp told me that MFG was the most searched stock this week with 210% more interest than usual 😉

  5. Another key point is population studies are good at predicting the success, be it in medicine or anything else, of something at a population level but poor at an individual level. If a treatment is 90% successful it does not mean my odds are 90%, it just means 90% of a group will be fine. The actual odds of an individual are unknown.

  6. Interesting article. I ordered the book you mention (Superforecasting) after looking it up.

    Largely irrelevant for the purposes of the article but I had different maths of failure from reading the survivor rates. If 50% of businesses fail in the first 5 years and 33% survive after 10 then that would imply that the baseline is 17% (0.5-0.33 or 0.67-0.5), rather than 33%.

    I think the definition of failure is worthwhile considering in this scenario as most failures are determined based on the operating status of registered entities which is very empirical. Given the size of Uber’s revenues and the number of smartphones the app is installed on there is significant value in this alone but you also raise the critical point of losses and the need for shareholders to continue to fund the growth of the business. Has it failed if the availability of outside funding dries up (this has happened several times in the past few decades), the business collapses as a self-sustaining entity and then sold for a fraction of the current valuation?

    One of the curiosities of Venture Capitalism in the US, is that there would appear to be an incentive towards businesses that can’t fund their own growth as it allows VC firms to continue deploying more capital at high management fees into these ‘successful’ companies. This occurs at consistently increasing valuations – determined at least in part by the VCs that are already invested (a market price is what someone is willing to pay!). This is the ponzi characteristic I suspect Hamish (and others) see not just in Uber but in a number of these VC-backed companies.

    • Hi Alan, if Uber was a startup that maths would be correct. But you know that it has already survived to the 5 year mark.If the failure rate goes up from 50% to 67% between years 5 and 10, then 17% more have failed but that is out of the remaining 50%. So our 33% came from 17/50.

      On the sustainability front, my personal view is that Uber could be profitable tomorrow if it needed to be. It makes perfect sense to keep funding it while they are growing so quickly but if you turned off the taps I think it could survive within its $US6bn of revenue quite comfortably.

  7. Hi Steve,

    I agree that base rate exercises are useful – survival rates are interesting and so are security returns. It might be interesting to know what the base rate probability is for a stock to achieve an above-index return, and over different time periods, for example.

    But once you start looking at Uber (or any other company) at individual security level I think the qualitative factors outweigh the stats. Because at the most granular level you are reduced to a sample size of 1 and the specifics of the stock you’re looking at.

    For Uber, I think the sustainability of the first mover network effect advantage will be the biggest driver in the medium term. If all the passengers and drivers are on Uber, why go elsewhere? (says someone who doesn’t know much about Lyft and still calls for a taxi). I think driverless cars are a longer term issue.

    For passengers/drivers, replace with consumer/supplier equivalents for any network business – payments (PayPal, Visa etc), recruitment (Seek), retail (Amazon, eBay), advertising (Google) etc. Theres a pattern in there somewhere!

    Cheers

  8. I thought the same as you Steve until I listened to the below (and have seem Uber’s management implode over the last month). The driverless car thing is a long time away but it’s a real threat. Uber’s strength is in its user experience, and their is no doubt they dominate this, but they’re not a tech innovator, and they are going to throw billions at the driverless car thing. Google’s partnership with Lyft puts them in the prime position to be the tech/service that succeeds.

    http://exponent.fm/episode-115-business-matters/

Leave a Reply

Your email address will not be published. Required fields are marked *

Security Test *