The Economics of uberX (Part II)

N.B. I’ve updated this post to provide additional support for my assumption that (i) uberX passenger behavior mirrors yellow taxi passenger behavior, and (ii) surge pricing alone does not explain the high average fare figure.

If you’re feeling masochistic, you can read Part I here.

Uber released a blog post last week regarding the impact of recent uberX price cuts on overall driver performance. The post appears meant to refute recent criticisms that the price cuts have had a deleterious effect on driver incomes. Uber and its board members have been quite vocal that the net effect of the cuts has been positive: Uber’s customers, they contend, were relatively price elastic, meaning that any percentage drop in prices was more than offset by a greater percentage increase in quantity of trips demanded, leading to an overall increase in total revenue (i.e., gross fares). The data appear to lend credence to this claim, but I thought it would be worth examining the data in a bit more depth to see what else we can tease out.1

What else must be true?

Whenever a company releases data, there’s a strong tendency to fall into the trap of assuming that what you see is all there is. But we should always remember that corporations have every incentive to present the best version of themselves, and that taking their selectively released data at face value is an especially pernicious form of gullibility. For this reason, it’s always prudent to ask, “If these data are accurate, then what else must be true?” In other words, if we are to believe a given dataset, do we also believe the inferences that follow.

Let’s start by reviewing the data Uber released. Note that this data refers only to uberX drivers in New York City in each of the last three Septembers (2012 – 2014).

Uber_-_October_2014_-_Google_Sheets

Right off the bat, I see a problem with this data. It’s odd to me that in 2012 and 2013 the average uberX fare was (i) so large, and (ii) so much greater than gross fares per hour. Even 2014’s average fare seems strangely high. As I mentioned in my last post about uberX, the average NYC taxi fare is $13.40. Given that Uber explicitly benchmarked their price cut against taxi fares, it’s reasonable to assume that Uber views uberX and yellow cabs as near substitutes, which implies that an average ride on one service probably looks a lot like an average ride on the other. And if you don’t believe that line of reasoning, consider Uber’s own request heat map:

[Edit: Andrew Salzburg (Uber’s operations manager in NYC) has pointed out that these data reflect e-hails for uberTaxi, not uberX.  This is an important to distinction to make in light of Josh Mohrer’s (Uber’s GM in NYC) contention that yellow cabs are not a good proxy for uberX.]

heatmap

Here’s the same data visualized as a cartogram (again, this is from Uber’s own blog):

cartogram

Finally, if you’re still not sold, you can always look at the hour-by-hour video of pickup data that Uber released to YouTube last year. It bears out what’s said explicitly in the post: “If you look at all e-hail requests over a 2 week period, you see what you might expect: taxi drivers [who] use Uber get requests in all 5 boroughs, but demand is concentrated in the financial district and midtown Manhattan.”

Knowing this, it’s hard to believe that the average uberX fare is double the average taxi fare, since that would imply a completely different set of behaviors on the part of uberX customers relative to yellow cab customers.2 But for now, let’s call it a suspicion and move on. Maybe once we tease out some more inferences, this will all make sense.

Taxi drivers make money for two activities: (i) pulling the flag and (ii) driving a passenger (i.e., paid miles). While performing these activities, they endure the various incremental costs associated with driving, e.g., fuel and wear & tear (i.e., depreciation). These dynamics explain why the profit-maximizing driver’s dual objectives are maximizing the number of flag pulls while minimizing the total distance traveled. Knowing this, it’s worth asking if we can infer the two operational metrics that most affect taxi economics: (i) trips per hour and (ii) average fare distance. And as it turns out, we can do just that.

Let’s walk through the numbers for 2014. Trips per hour is a simple calculation. We just take the gross fares per hour ($36.16) and divide it by the average fare ($27.11), which implies an average of 1.33 trips per hour.

Fare distance is a bit harder. We know that, post-price cut, gross fare is calculated according to the formula:

Gross fare = base rate + (duration x per-minute rate) + (distance x per-mile rate)

Pre-price cut, though, the formula was slightly different:

Gross fare = base rate + (miles over 11mph x per-mile rate) + (minutes under 11mph x per-minute rate)

We don’t have any information about the amount of on-trip time an uberX spends above or below 11mph, but we do have a good proxy available. Using data from the 2014 NYC Taxi Factbook, we can deduce that taxis spend about 53% of on-trip time under 6mph. Since we’ve decided that yellow cab activity is a reasonably proxy for uberX activity, it seems fair to assume that uberX drivers spend a similar proportion of their on-trip time below 11mph.

Moving on, we know the rates in these equation because they’re posted on UberNYC’s home page. What we don’t know is duration or distance, and we need one to find the other. This would be quite a pickle, except Uber has kindly given us all the information we need to suss out duration. If we divide the minutes per hour “on trip” (29) by the number of trips per hour (1.33), we’re left with an average trip duration of 21.7 minutes. From there, calculating distance is a matter of subtracting time- and flag-related income from gross fares, and dividing the remainder by the per-mile rate, which yields an average fare distance of 7.2 miles. Finally, we can further calculate that average “on trip” speed is 19.8 MPH (distance / duration in hours) and that average utilization is 48% (minutes per hour “on trip” / 60).

If you’re more of a tables person, here’s everything I just said in pretty alignment:

Does this make sense?

As a reminder, we already know these metrics for yellow cabs, which, again, are ostensibly uberX’s nearest substitute. Per the 2014 Factbook (and logical inference):

In addition, the Factbook also tells us that utilization for the average taxi hovers near 50%.

Now that we know what must be true if Uber’s numbers are to be believed, I ask you: do these numbers make sense? They don’t to me. An average speed of almost 20mph? Even if the majority of uberX rides involve the outer boroughs (and they don’t), that seems crazy in a city where taxis barely average 10mph in the most common pickup/drop-off area.3 And besides, are taxi trips in the outer boroughs really so different from trips in Manhattan? According to data gathered for the 2014 Boro Taxi Market Study, the distribution of trip lengths in the boroughs is a little flatter than that of trip lengths in Manhattan, but not that much flatter:

Boro_Taxi_Market_Study_pdf__page_41_of_70_

And what about that average fare distance of more than 7 miles? Do we really believe that the average uberX ride originates in Lincoln Square and ends in Williamsburg? Or that the average trip is an airport run? Or do we believe that when people use uberX, they go on average almost three times the distance as they do when they take a cab, and that the trip takes them only twice as long? And despite all that, they still manage to maintain the same utilization as taxis? What about surge pricing? As Justin Overdorff notes, that’s one a plausible way to explain the discrepancy without assuming widely divergent behaviors. Luckily, that’s a hypothesis we can check using data collected by whatsthefare.com, and it turns out that surge pricing amounts to no more than around a 12.5% fare premium. Not nothing, but not explanatory either.

[Edit: Salzburg and Mohrer each point out plausible explanations for the disparity between uberX and yellow cab trip data: (i) most uberX trips occur in the outer boroughs and NJ, and (ii) uberX has an $8 minimum fare, which taxis do not.]

As it happens, I believe that the declared average uberX fare — a figure that’s more than double its benchmark — is wrong. And not just a little wrong, but horribly, weirdly wrong. So wrong that I have trouble ascribing motive to it.4 But that doesn’t change the fact that this data looks really wrong; in fact, it looks doubly wrong. So I’m going to perform a transformation on it based on nothing more than my intuition: I’m going to cut the average fare figures in half and see what effect that has on the metrics.

WARNING: It should be obvious that if you do buy the implications of Uber’s numbers, then you’ll give no credence to the analysis that follows. That’s your choice, and you are welcome to it. I’ve been wrong a lot in my life; I’m sure one more time won’t hurt.

 

And now we’re really in a pickle, because despite my transformation being based on little more than a hunch, it’s resulted in numbers that actually do make sense. Eleven-minute rides that go for 2.9 miles? Average fares of $13.56? An average speed a few miles per hour over a taxi? Realistic, likely, and plausible (because, outer boroughs).

uberX vs. yellow cabs

Ultimately, the only reason to release these numbers is to prove to drivers that the uberX program offers meaningful rewards. Uber’s made no bones of the fact that they are aggressively recruiting drivers. They’ve even gone so far as to offer income guarantees: for the last three months of the year, every uberX driver who spends 200 hours active in and around Manhattan and accepts 90% of their trip requests is guaranteed a monthly payout equal to $5,000 in net fares. Just how good of a deal is that? Thanks to the first footnote in the Three Septembers post, we can actually figure out the answer.

$25.10 per hour x 200 hours = $5,020, which means that, for the next three months at least, Uber is indemnifying full-time drivers against slow business.5 Good luck competing with that, companies-that-don’t-have-a-$1.5-billion-war-chest.6

Returning to the question at hand, is driving for uberX the “fantastic opportunity” that UberNYC’s GM says it is? It doesn’t seem like a bad opportunity. Of course, $25.10 isn’t really your take-home. There’s gas, wear & tear, car payments, interest, repairs, and insurance to account for, which could easily knock $5-10 per hour or more off your net. All told, a full-time driver probably averages around $15 per hour in take-home, which is right in the ballpark of what was inferable from Uber’s previous data release. To put it bluntly, Uber’s post provides further evidence that the economics of driving an uberX are exactly what we should expect them to be, that is, the same as the economics of driving a yellow cab. Or, at least, the marginal economics are the same. While taxi drivers pay to lease medallioned vehicles owned by others, uberX drivers enjoy the privilege of backstopping all the costs, liabilities, and risks that come with owning your own vehicle, which doesn’t sound so bad until you remember that cars break down sometimes, especially when you put 40,000 Manhattan miles a year on them. Not only do owner-drivers risk having to shell out for repair bills, they’ll have to pay those bills while their only revenue-producing asset sits idle.7 That situation is what the experts call, “operating without a net.”

A comment on utilization

Now, Uber bulls and board members have long argued that the key to this whole thing is network effects, which would presumably push utilization up in the range of 70-80%. Or, to use real numbers, they think that network effects plus algorithms will allow the average driver to produce 42-48 revenue-generating minutes per hour. This is an oft-repeated idea, that the scale advantages of Uber’s network will lead to a virtuous cycle that magically increases both availability and utilization.8 Respectfully, I call bullshit on that belief.

Early last year, the Earth Institute at Columbia University released a white paper entitled, Transforming Personal Mobility. The purpose of that paper was to assess the characteristics of centrally controlled, driverless car networks. Even under the absurd conditions of the paper (9,000 centrally controlled driverless taxis unimpeded by human-driven vehicles and pedestrians, and potholes), average utilization only reached 70%, and that was during peak periods.9 And that actually makes sense, because the behavior that drives utilization — customer demand — has to increase exponentially for utilization to increase linearly, because it’s not about aggregate demand, it’s about proximate demand. If you’re dropping someone off in Harlem at 12pm, it doesn’t much help your utilization if there’s high demand on the Lower East Side. There’s only so much fare chaining to be done, and when you factor in randomness (e.g., stop lights, traffic jams, other drivers, pedestrians, roadwork, potty breaks, fill-ups, blown tires, etc.) and waiting (Does everyone who calls a cab to their home really make it downstairs “in a minute”?), it’s nigh impossible to imagine a realistic scenario where uberX gets its utilization up over 60%, let alone 70-80%. And even if it did get up that high, consider what that would mean for availability. A car, after all, can’t be both utilized and available at the same time.10 If utilization goes up, the supply of cars will have to increase to keep up availability, which will exert downward pressure on availability. In other words, these aren’t metrics that can be independently pumped, they’re metrics that have to be simultaneously balanced. You are of course free to believe that the immediate future New York streets will be infinitely more orderly, predictable, and driverless than the present and past, but I’ll be taking the other side of that bet.

Conclusion

If someone truly wanted to “disrupt” the taxi market, they wouldn’t continue the decades-old trend of pushing risk onto drivers. Instead, they’d do something completely crazy, like buy a fleet of vehicles, hire a raft of high-quality drivers, pay those drivers a fair wage, invest in oodles of technology, and foster a culture that prioritizes efficiency, teamwork, and service. But I doubt such a company would get much interest from investors. Why invest in assets when you can invest in the platform that relies on them? So it goes.

By now it should be obvious that uberX, at least in New York City, is nothing more or less than a lightly regulated, heavily subsidized substitute for yellow cabs. Uber’s own data releases, when examined closely, bear this out. There is no technological magic here, there’s only the magic of subsidized prices, guaranteed wages, and bountiful promotion. Uber did a great and innovative thing by decentralizing the dispatch function and pushing it to mobile, thereby enabling them to build the first truly global taxi dispatch brand. They also did some very smart things with mobile payments, which helped them collect their rake with only a modicum of effort. Combine these innovations with exceptional execution, and it’s easy to see why they get so many plaudits (and so much investment). But at the end of the day, their business model boils down to taking a 20% rake of a large commodity services business while pushing assets, costs, and liabilities off the corporate financial statements and onto the shoulders of their not-quite-employee-partners. They own no cars, employ no drivers, and their vaunted network is built not on real assets, but monetary incentives. If the rest of their metros are anything like New York, they need global expansion to mask intra-city margin compression as their brand becomes less luxury and more commodity. That’s a tenuous position to be in, billion-dollar bankroll or not.

  1. All calculations in this post can be found in this spreadsheet. Feel free to copy and critique at your leisure. []
  2. Maybe it actually is the case that uberX passengers and taxi passengers use each service differently, but that would call into question the rationale behind Uber’s price cut. Mutually unique behaviors would suggest the two services aren’t substitutes at all, which would mean they’re not really competitors, which would obviate the need to “beat” the other’s price. []
  3. See pgs. 14-17 of the 2012 NYC-DOT Sustainable Streets Index. []
  4. When a company releases data that’s not just wrong, but beneficially wrong, it’s easy to ascribe motive. But sometimes, data just gets screwed up. []
  5. This is a good time to point out that, at $25.10 per hour, a driver would have to work 300 hours per month to net $90,766, which is what Uber has claimed to be the “median uberX small business income” in New York. I’m on record as doubting that claim, as are others. []
  6. I’ll now point out that this behavior is very much a form of predatory pricing, in that Uber is using its monumental cash reserves to sell its dispatch service to drivers at a loss with the express goal forcing less-well-funded competitors from the arena. The point of antitrust is to make companies compete on service, not balance sheets. If you’re sympathetic to the Chicago School, you’re unlikely to see that as a problem, because consumer welfare or something. Personally, I do think it’s a problem, but I’m more of a Herbert Simon kind of guy. So it is. []
  7. If you think insurance ever really makes a driver whole in that situation, well, that’s adorable. []
  8. This idea is especially magical when you remember that availability and utilization are countervailing concepts: at some point, increasing one can only come at the expense of the other. []
  9. Transforming Personal Mobility, pg. 25. []
  10. Yes, I’m aware of ride-sharing. No, I don’t think it changes the calculus, because I don’t believe that people in the US will use it in the volumes required to make the math work. I could be wrong. []