Among the arguments for Uber's current valuation is its option potential as the global platform for dispatching fleet-owned replacements for personally owned vehicles (POVs). In this vision of the future, these fleet vehicles may or may not be owned by Uber.
This view is inconsistent with my understanding of operating economics. Conventional wisdom makes several errors that artificially inflate the breakeven mileage of POV retention:
- Mis-classifying variable expenses as fixed, particularly insurance and depreciation
- Not penalizing fleet vehicles for higher variable costs
- Insurance is 2-3x higher due to higher liability limits and cancellation requirements
- Empty mileage penalty while repositioning of 10-90%
- Comparing apples/oranges car types (size, amenities, autonomy, and/or regulatory environment)
As a result, the limits of shared cost savings for Honda Civic-type fleet vehicles are much smaller than widely assumed - the POV breakeven today is less than 100 miles/month with minimally paid drivers, and about 500 miles/month with autonomous vehicles. Even if autonomy eliminates insurance requirements, the breakeven for an autonomous Civic is about 1,000 miles/month - less if Uber retains any platform margin.
Near-term POV replacement by driven fleet vehicles is likely to be caused by external one-time variable cost avoidance such as parking fees and hassle factors rather than increased capital utilization of the vehicles. This distinction makes a difference - for Uber, for car owners, for city planners.
An autonomous vehicle could likely park itself outside most high-cost CBDs with a lower VMT cruising penalty than it would receive in fleet use. This suggests that so long as autonomous vehicle acquisition costs are on par with current vehicle costs, the future may look a lot like today - only moreso.
While Uber could coordinate autonomous google-cars-as-a-service, those cars are likely inexpensive enough that the use case would not be cost savings or even convenience as much as regulatory - especially as the low-cost use eats into mass transit and results in overcrowding of roads.
Background
Economies of scale often vanish on closer inspection, or at least turn out to exist in forms different than that which was assumed - and the details matter immensely if planning an executable strategy.
A well-written May 2015 story in the City Paper by Emily Guendelsberger, who drove ~100 trips as an Uber driver earlier this year, quotes the Uber CEO referring to this potential:
Travis Kalanick, the CEO and founder of Uber, said at a conference last year that he'd replace human Uber drivers with a fleet of self-driving cars in a second. "You're not just paying for the car — you're paying for the other dude in the car," he said. "When there's no other dude in the car, the cost of taking an Uber anywhere becomes cheaper than owning a vehicle." That, he said, will "bring the cost below the cost of ownership for everybody, and then car ownership goes away."This view has been repeated often, sometimes citing (as Guendelsberger did) Columbia's Earth Institute's 2013 paper titled "Transforming Personal Mobility." This view is flawed*.
Bottom line for our discussion: The costs of an Uber-owned autonomous vehicle will not be lower than a privately owned autonomous vehicle above about 500 miles/month of usage.
Taxis in dense cities will likely be much cheaper and much more abundant, though.
Today's post has three sections, and a coda:
- For "Civic" level Uber cars in Pittsburgh, current costs favor POV use above about 120 miles/month - less if it's lots of small trips
- External variable costs such as parking fees can change the equation but only with drivers. Autonomous cars reduce the incentive to use Uber on any given trip
- There's no reason to believe autonomous vehicles would not continue to be privately owned outside small fleets of core urban taxis, or perhaps provided "as a service" by luxury brands.
1. For "Civic" level Uber cars in Pittsburgh, current costs favor POV use above about 120 miles/month - less if it's lots of small trips
Autonomous cars won't change this much past 500 miles/month; even if insurance requirements vanish it won't increase beyond about 1,000 miles/month.
For starters, let's compare apples to apples: an Uber/fleet vehicle that has the same actual marginal cost per VMT as a POV. It doesn't matter if we're talking about a large Chevy Suburban, a mid-sized Honda Civic, or a tiny Google car. For now, let's assume a Honda Civic.
Note that higher capital cost to operating cost ratios favor fleet use - a $150k luxury SUV has capital costs to operating cost ratio much higher than a Civic. At the extreme, consider an airplane. However, low operating margins also limit the ability to profitably re-position an airplane using an empty leg.
Today's first question is whether shared usage / increased asset utilization makes financial sense looking only at vehicle ownership/operating costs. In section 2 we'll examine the impact of other externalities such as parking costs.
Many flawed analyses begin to go wrong right here: often they compare a typical type of privately owned gasoline-powered car with some idealized tiny, electric powered vehicle - and then incorrectly ascribe the operating savings of the tiny electric vehicle to the fact that it was in a shared fleet.
In reality, the private owner could just as easily own the tiny electric-powered car if it makes sense for him. Fixed costs of vehicle ownership do matter, and they'll be spread farther on a fleet vehicle - but most normal cars have only a few hundred bucks of fixed costs to spread each month, and a specialized tiny google car is anticipated to have even less.
1a. Fleet vehicle variable costs are higher than an equivalent POV due to empty (unpaid) miles between fares
Consider a 2015 Honda Civic LX variable costs during the third year of ownership**:
- Fuel - $.097/mile ($1,449 for 15k miles)
- Maintenance/repairs - $.03/mile ($368+89 for 15k miles)
- Mileage-based depreciation - $.03/mile***
This is $.16/mile - but only for the POV.
Fleet drivers must cruise to find fares - which takes up 35-60% of their time. In NYC, street-hail taxicabs cannot accept central dispatch, and dispatched livery cabs cannot accept street hails under most circumstances. In other cities, cabs do both. Some of this cruising is a result of inefficient dispatching, the rest results from imbalances in the near-term distribution of start/end points.
In NYC, about 40% of taxi mileage is cruising - for every mile driven with a paying fare, two thirds of a mile are driven empty. This is consistent with my recent limited experience in Uber/Pittsburgh. So realized variable costs per mile of our fleet Civic are closer to $.16 + 2/3*.16 = $.27/mile
As long as we're comparing apples to apples, however you want to define variable costs (vehicle type, mileage, gas price, maintenance, etc) the POV will have a variable cost advantage due to the empty VMT of a fleet vehicle. This highlights the second main flaw of studies which project a fleet-based autonomous future: they assume an unbiased demand pattern.
1.a.1 Intuition building on biased demand patterns
Closer examination of Uber's ride-level data will likely highlight that ride demand is not only stochastic (can be analyzed though not perfectly predicted), but also biased: rides are not evenly distributed throughout the city, but rather follow time-based flow patterns.
In an oversimplified case, imagine that n people want to go from A to B in the morning, and B to A in the afternoon. There are three options. First, carpool/bus/subway in the same vehicle that carries n people. Second, take n number of vehicles more or less simultaneously. Third, take one vehicle, n times, traveling 2x the distance. There are pros and cons to each.
The first solution, broadly called bussing, has coordination costs that have implications for both pricing and costs. Under certain conditions this works, but vehicle size/cost/path must be optimized. Operators often end up using a bus as a taxi; the busses will often be oversized or undersized when ridership is measured instantanously.
The second solution, broadly called commuting in your own car, is often individually efficient but imposes costs on others due to traffic and parking. Sometimes those costs are also born by the driver instead of society. The number of vehicles is essentially equal to the number of commuters.
The third solution, broadly called taxi, removes duplicate vehicles at the expense of duplicate miles. In the case of NYC, the markup is about 67% more miles and likely varies by time and neighborhood, let alone city. This is why taxis often don't want to take you to neighborhoods where there isn't much demand. And if you call a cab from one, they're reluctant to come get you lest you cancel the request and leave them stranded without a fare back.
Higher numbers of fleet vehicles could reduce the number of empty miles - if the induced higher demand is at counterbalancing patterns.
Otherwise more fleet vehicles simply drops individual vehicle utilization - at the extreme it just becomes everyone in his own car, like the second solution. Like people Uber'ing home when they're done at the bar - you need about as many cars as destinations. With fewer cars, drivers with multiple trips will drive double-mileage.
1b. Fixed costs of car ownership spread quickly. In Pittsburgh today, owning a Honda Civic still beats using Uber at 10 round trips of 3 miles each month.
As noted in Part 4b, today's Pittsburgh rates are likely below cost for Uber drivers. Drivers are subsidizing ridership, perhaps because they don't understand their costs. Anecdotal information suggests a downward trend on vehicle quality, and driver turnover may also be quite high - as new drivers realize the actual economics, it's likely they will leave.
1.b.1 Financing costs are definitely fixed, though should be calculated based on the point-in-time financing cost of the vehicle, since the choice of whether or not to finance out of pocket is separate.
So if we're talking about the same 3-year old POV and Uber vehicle, the value of the car is the same for both owners at any point in time.
If that three-year-old car is worth $12,000
- Private financing at 5% is $50/month in financing costs for a collateralized asset
- Perhaps Uber can borrow for collateralized assets at 3%, or $30/month in financing costs
1.b.2 Age-based depreciation is $50/month for either party
1.b.3 Fixed-cost total is $100/month for POV, $80/month for a fleet-owned vehicle
A fleet-owned vehicle might drive 60,000 miles a year or more; those fixed costs get amortized to less than $.01.
A POV driven 937 miles/month has amortized fixed costs of $.107/mile and, with an additional $.16 of truly variable costs is about equivalent to the variable costs of a taxi with 40% cruising VMT.
This means a fleet-owned civic only has an absolute cost advantage over a POV civic driven fewer than 11,244 miles in a year. Beyond this, even at variable cost pricing, the POV owner saves because he's not paying for the empty miles of the cab.
Of course, this ignores insurance.
1.b.4 Insurance is a hybrid of fixed/variable costs, and also more expensive for fleets. In the long run, fleet-owned vehicles will continue to have higher insurance costs due to higher liability limits and stricter cancellation requirements
Note that liability exposure is mileage-based:
- Legacy personal line auto policy structure has flat rates which (among other things) subsidize high-mileage drivers by overcharging low mileage drivers. This will (slowly) change.
- Metromile recently launched in Pittsburgh; two policies that cost me ~67/month each with USAA would cost me <$30/month + $.06/mile with Metromile
- Commercial policies are 2-3x as expensive due to higher limits of liability and cancellation requirements. This will likely shift to mileage-based much more quickly if it isn't already, but fleet insurance will likely remain more costly than individual
Civics are among the highest dollar amount of loss, possibly due to its low cost being most attractive to younger, less responsible drivers. And low vehicle weight contributes to higher injuries during a crash.
I priced a 2015 Civic with USAA and received a quote $1.24/month higher than my current auto insurance (1998 4Runner, 2003 Accord) of about $67/month apiece. So I'll assume the Metromile quote is valid for a Civic.
Insurance varies heavily by region and risk profile; a mature driver with a reasonable record should easily be able to get Civic insurance for less than $100/month here in Pittsburgh. And the metromile quote of $30+ $.06/mile seems a more accurate breakdown than assuming a straight $67/month fixed cost.
1.b.5 Assuming reasonable insurance costs, Uber platform fees and driver costs lower current POV breakeven to ~80 miles/month for short trips (~130 miles/month for a single large trip)
A future autonomous POV Civic breakeven is ~500-1,000 miles/month, depending on insurance costs
Using my Metromile insurance rate of $30 + $.06/mile, I'll estimate commercial insurance costs of 2x, or $60+$.12/mile. Just like 1.b.3 above, the fixed cost of insurance basically goes away for fleet vehicles due to high usage and we're left with the variable piece.
Said differently, the POV owner of the civic pays $100+30=$130/month for the option to drive at $.16 + $.06 = $.22/mile.
The customer of the fleet taxi pays $0/month for the option to pay a cost of $.16 + $.12/mile and a 67% mileage penalty, or $.47/mile. Excluding a driver fee and platform/vehicle profit, those curves cross at 520 miles/month from just variable operating cost alone.
In our autonomous future, insurance rates are likely lower, though fleet operated vehicles will likely remain more expensive. This would increase the breakeven mileage point - but even if the insurance cost went to zero the POV would be advantaged above 937 miles/month (as per 1.b.3).
Today, at real Uber rates, POV breakeven is between ~80 and ~130 miles/month
Today in Pittsburgh, Uber costs $1.20 per mile, and for reasons explained in 3b these prices are likely already below a combined minimum wage and car expense. Including insurance (apples/apples!), the POV Civic driver pays $130/month for the option of $.22/mile. Those curves appear to cross at 132 miles/month, but it's actually closer to 80 miles if they're short trips.
When the "start fee" of $1.50, "safe driver fee" of $1.00, and $.20/min "wait time" are considered, the line crosses lower - the minimum trip cost is $5, which is basically a $2.50 startup fee, a $2.40 mileage charge for a two mile trip, and 30 seconds of wait time. Twenty-six $5 trips/month is $130 (and 52 miles). Adding a mile to each of those trips is $31, for a total of 78 miles and $161 for Uber.
78 miles in the POV at .22/mile is only $17.60 + $130 = 147.60 for POV.
That's one 3-mile round trip every three days, if you're paying current Pittsburgh Uber driver wages.
It should be noted that short trips benefit Uber; in the example above Uber's platform fee is $2.50+ 20%*2.50 = $3. On a $5 fare that's a 60% platform margin. The City Paper article does a nice job illustrating that Uber's platform revenue is often 30-40% because of the prevalence of these smaller trips, not the "advertised" rate of 20%.
Of course, some portion of this higher platform margin is paid back to drivers as part of the guaranteed hourly wage instituted in early 2015. I'll work out how much when I get to the future posts on recommendations for Uber and its competitors.
1c. Electric vehicles don't change this balance much. Small, inexpensive cars will save the least from amortizing fixed costs; large, expensive cars will save the most
Let's look first at electric vehicles - how will this change the equation? As we looked at in Part 4b, the "fuel" costs of electricity will be much lower - in that example, electric cars saved 65% on fuel (from $.1126/mile to $.0389/mile). Maintenance is likely lower as well.
What that calculation leaves out, though, are State and Federal taxes that support roadway construction and maintenance. Here are current rates from the EIA; I believe there is a wide amount of discussion already on the negative impact that improved fuel efficiency has on maintenance budgets. Electric vehicles will likely be assessed for road usage at a rate comparable to gas powered vehicles
In PA, the total tax is currently $0.70/gal on about $2.50/gal of gas, or about 28%. New York is also high. For illustrative purposes in this conversation, I'll assume that variances in gas taxes due to local conditions (weather impact on road maintenance in the North) drives most of the differences in the price of gas nationwide.
In this case, 28% of the $.1126/mile gas cost (from Part 4b) would be road maintenance tax, or $.0315/mile. So the electric "fuel" cost comparison in high gas-tax states such as PA and NY is more like $.0704 (half would be road maintenance fees!), and the savings from electric fueled vehicles is about 37%.
How does this change our equation above? If fuel is 37% lower, and maintenance and mileage-based depreciation remain the same, our total variable costs per VMT for the electric Civic is $.061/mile (fuel), $.03/mile (maintenance), $.03/mile (mileage-based depreciation), and $.06 for insurance for a total of $.181/mile.
Our fleet vehicle, with $.12/mile insurance and 67% more miles, costs $.402/mile. At $130/month in fixed costs for the POV, the curves now cross at 588 miles per month - up 13% from 520. Because both the fleet and POV benefits from reduced variable costs, with no increase in fixed costs, there isn't much change in the equilibrium.
Because Uber pricing is essentially marginal cost pricing, an interesting question is what happens when some drivers use electric cars but the fleet as a whole doesn't. The answer is those drivers pocket the difference, until enough of the fleet is electrified, then those savings pass to the consumer.
If electrification doesn't justify an Uber-fied Civic, what would?
- External variable costs such as destination parking (section 2 below)
- Very high fixed costs to amortize (section 3 below)
- Higher capital cost vehicles - either luxury or autonomy at a very high price
- Monthly ownership fees such as parking garages
2. External variable costs such as parking fees can change the equation but only with drivers. Autonomous cars reduce the incentive to use Uber on any given trip
From a trip-based perspective, assume for a moment that I already have a car. On each trip, my fixed monthly costs are sunk - so I will take an Uber if it makes more sense for that trip.
There are two separate questions:
- For the given trip, do I choose Uber or my POV?
- Over time, am I using my POV on enough remaining trips to justify its fixed costs?
On any given trip, the operating cost of the Uber car is higher than my POV due to higher insurance and empty cruising miles - so I will choose it every single time, if choosing by variable operating cost alone. Today, the Pittsburgh "Uber penalty" is $2.50 + $1/mile.
- On the 4-mile trip from my house to downtown, this is about a $13 penalty round-trip
- Parking costs are $5 on nights and weekends, $8/hr max $18 during weekday
- In NYC it appears parking runs about $15-60/day
- Consider hassle factors of waiting (for Uber) or parking (for me)
- Perhaps I'd Uber for a weekday business meeting, provided I was 100% sure I didn't want to go somewhere else before heading home
- Unless I parked downtown frequently enough to justify a $300/month parking lease
- In that case I would almost never Uber downtown at least in Pittsburgh
- Beyond 7 miles each way, it's always more expensive to Uber in Pittsburgh - even with the parking cost savings
Perhaps this is obvious, but in the short-term Uber replaces trips for people who already don't have cars, or have high single-trip costs for their POV (parking costs, anticipated drunkenness, etc)
In the long term, an autonomous car could lower that "Uber penalty" - but if I had an autonomous car, why couldn't it go park itself somewhere that wasn't high cost? Uber's costs will go down, but so will the cost of the alternative to using Uber on any given trip.
The breakeven threshold for owning a car will increase, but the use case for Uber despite owning a car will disappear. I suspect this dynamic is responsible for much of Uber's growth in cities like San Francisco - use of an Uber instead of a POV on a one-off trip because a taxi costs more than parking but Uber doesn't. With autonomous cars, these rides would go away, offset to some degree by increased leakage from public transportation.
With autonomous cars, the POV electric Civic breakeven for ownership is the 500-600 miles/month described in 1c above, reduced by any margin Uber takes. This is a higher bar than the 100 miles with a driver, but still less than 20 miles a day before accounting for any Uber margin.
Unless the fleet penalty for insurance costs or VMT drops significantly, or autonomous technology is outrageously (capital) expensive, most people will buy the nicest autonomous car they can afford - like today, only moreso.
3. There's no reason to believe autonomous vehicles would not continue to be privately owned outside small fleets of core urban taxis, or perhaps provided "as a service" by luxury brands.
Someone has probably already said, "The future will look a lot like today, only moreso." A mentor sent me this link last week on predicting the future, much along the same lines.
Higher adoption of shared vehicle fleets is likely with higher parking costs and other externalities such as congestion pricing. Higher vehicle costs, either through costs of autonomous technology, or decreased economies of manufacturing scale through lower volumes, could also drive adoption.
Autonomous vehicles, if affordable, will negate any parking cost advantage from a fleet vehicle. And the remaining economics don't overwhelmingly favor fleet use of autonomous vehicles.
3.a Cars don't have the same benefits from high utilization as airplanes or warehouses. I think this is where a lot of intuition misfires with a "just so" story
- At high ratios of capital cost to operating costs, high utilization is required when margins are thin. I'll use monthly (capital) to hourly (operating) to illustrate because it's easy to show the impact on usage cycles
- A $100m 737-900ER has monthly capital costs of $417k, and a $7,794 hourly operating cost (includes fuel, maintenance, labor, engine replacement. Excludes fixed costs such as crew salaries, hangars, etc)
- At 10% variable cost margins ($8,660 rev/hr) or $866 it would take 481 hours just to pay the capital costs. That's 30 days of 16 hours/day
- Of course, there's (limited) age-based depreciation and insurance, too - hopefully the airline gets more like 15% variable cost margins on average
- It's clear why turns is more important than earns in a low-margin, high capital cost environment - and why commercial airliners can't afford to fly around empty or even at low load factors. This is (partly) why lessors like ILFC and ALC make the money in this industry
- Our POV Honda Civic above had $50/month in capital costs, and $130 in total fixed costs. Variable operating costs are $.22/mile
- In city driving averaging 12mph, hourly costs are $2.64; at 10% margins ($2.90/hr in revenue), it would take about 45 hours to cover all fixed costs
- In highway driving averaging 60mph, hourly costs are $13.20; at 10% margins ($14.52/hr), it would take less than 9 hours to cover all fixed costs
High capital costs relative to operating costs are most useful as a barrier when they're not fungible - Amazon built out a series of large, capital intensive distribution centers in which the marginal cost of storing or fulfilling one more item is negligible.
Competitors who build a DC in one physical location can't pick it up and move it (or even modify it's structure substantially) without incurring large costs. So they won't build one until they're sure they have enough volume to make it work - or they get investors to build ahead of expected demand.
Fixed capital intensive industries (railroad, fiber networks, etc) tend to have boom-bust-boom cycles where the long-run success is built upon the failure of investments that built out capacity, which is then brought back online for a fraction of its development cost.
Competitors who build a DC in one physical location can't pick it up and move it (or even modify it's structure substantially) without incurring large costs. So they won't build one until they're sure they have enough volume to make it work - or they get investors to build ahead of expected demand.
Fixed capital intensive industries (railroad, fiber networks, etc) tend to have boom-bust-boom cycles where the long-run success is built upon the failure of investments that built out capacity, which is then brought back online for a fraction of its development cost.
High capital costs relative to operating costs are not really a barrier to entry for airlines, since the mobility of the capital asset means it can be redeployed at a moments' notice - ensuring the industry structure evolves to keep the high-cost asset in service through leasing a large portion of the fleet.
After deregulation, airline industry margins have accrued to lessors, not operators, over the long run. Manufacturers of the largest planes have been kept in a duopoly largely through state intervention preventing consolidation into a monopoly. These dynamics have been somewhat softened recently with extras like baggage fees, etc charged by the airlines.
Profits from these ancillary fees are essentially the entirety of airline profits. And in the long run, "industry discipline" will be required to avoid using those profits to subsidize even lower ticket costs to drive market share.
After deregulation, airline industry margins have accrued to lessors, not operators, over the long run. Manufacturers of the largest planes have been kept in a duopoly largely through state intervention preventing consolidation into a monopoly. These dynamics have been somewhat softened recently with extras like baggage fees, etc charged by the airlines.
Profits from these ancillary fees are essentially the entirety of airline profits. And in the long run, "industry discipline" will be required to avoid using those profits to subsidize even lower ticket costs to drive market share.
3.b Competitive dynamics limit platform margins and impact breakeven levels of usage
This is the essential problem with Uber as a business - if there's money to be made in using a car as a taxi for the next hour or two, anyone not using their car will immediately put it to use as long as the variable costs will be covered.
The vehicle density required to provide adequate ride coverage in a city for a competing ride network is small - a couple dozen vehicles in a city the size of Pittsburgh, and maybe a couple hundred in NYC. From there, the number of vehicles can scale with the number of users.
The low cost/mile of UberGoogle when compared to public transit is the real kicker - it's likely to draw more users out of public transit than replace existing private cars. Most public services operate under a subsidy model; that is, since they aren't allowed to profit, they "reinvest" their profitable operations into unprofitable ones.
When competition cherry-picks the profitable routes (such as FedEx/UPS did to the USPS), the public service may collapse unless externally subsidized. Most transit systems globally operate at a loss in this manner already, with the amount of subsidy a local political decision.
I'll cover these dynamics in a later post, but this dynamic is likely to accelerate the adoption of congestion pricing in the urban cities most suitable to Uber. These taxes on roads will be used to subsidize public transit, essentially making it free (or even paid if the abuse incentives can be managed) in order to keep street traffic to a manageable level.
The vehicle density required to provide adequate ride coverage in a city for a competing ride network is small - a couple dozen vehicles in a city the size of Pittsburgh, and maybe a couple hundred in NYC. From there, the number of vehicles can scale with the number of users.
It doesn't take much contribution margin to pay for a dedicated vehicle - as soon as it's clear that the revenue is high enough to cover the additional insurance and empty VMT, unused cars will be pressed into service unless regulations or other cost barriers intervene.
The impact to the owner may be little more than their car's useful life is consumed faster - perhaps meaning they get a new car every 3 years instead of every 5 or 10.
It's challenging to believe Uber will be able to extract enough margins from just the rider transport side of the equation. Competitive dynamics will keep the core business margin low, forcing Uber to find profit in ancillary services.
The impact to the owner may be little more than their car's useful life is consumed faster - perhaps meaning they get a new car every 3 years instead of every 5 or 10.
It's challenging to believe Uber will be able to extract enough margins from just the rider transport side of the equation. Competitive dynamics will keep the core business margin low, forcing Uber to find profit in ancillary services.
3.c Uber/POV breakeven in an autonomous world will depend heavily on vehicle type/cost, insurance costs, empty VMT, and platform margins. Let's look at best-case taxi fleet economics
The lowest operating cost solution is purpose-built, small electric cars like the Google car. Using the above framework, a motivated reader could construct a best-case cost curve that illustrates the underlying principles:
- POVGoogle might have a variable cost of $.08/mile
- Electric cost $.03/mile
- Road maintenance tax of $.03/mile
- Mileage-based depreciation of $.01/mile in electric vehicle with no maintenance
- Basic insurance rate of $.01/mile
- Fleet-based UberGoogle might have a variable cost of $.11/mile
- Base rate variable cost $.08/mile
- Insurance "penalty" for fleet use of $.01/mile for higher limits and cancellation reqs
- VMT "cruising penalty" average of 20%
- No higher depreciation/costs from non-owned use - riders treat the car as if they own it
- Using UberGoogle is a (minimal) penalty of $.03/mile plus .011/mile/10% operating margin UberGoogle charges
- For every $12k of average capital value there is monthly cost of
- $80/month for POV
- $50/month (capital)
- $30 (age-based depreciation)
- $60/month for UberGoogle
- $30/month (capital, lower borrowing cost)
- $30 (age-based depreciation)
- At 10% variable cost margins
- UberGoogle penalty is $.041/mile to POV, its margin is $.011/mile
- POV breaks even at 1,951 miles/month for each $12k in average capital value
- UberGoogle breaks even at 5,454 revenue miles/month/vehicle
- At 30% variable cost margins
- UberGoogle penalty is $.062/mile to POV, its margin is $.033/mile
- POV breaks even at 1,290 miles/month for each $12k in average capital value
- UberGoogle breaks even at 1,818 revenue miles/month for each $12k in average capital value
- At 50% variable cost margins
- UberGoogle penalty is $.085/mile to POV, its margin is $.055/mile
- POV breaks even at 941 miles/month for each $12k in average capital value
- UberGoogle breaks even at 1,090 revenue miles/month for each $12k in average capital value
- At 100% variable cost margins
- UberGoogle penalty is $.14/mile to POV, its own margin is $.11/mile
- POV breaks even at 571 miles/month for each $12k in average capital value
- UberGoogle breaks even at 545 miles/month for each $12k in average capital value
Note the tension between margin and POV breakeven - as UberGoogle extracts more margin from its platform, it reduces the breakeven point of owning the POV (or launching an alternative taxi service).
Because of the capital cost of this model, it would be interesting to explore whether an Uber or a vehicle manufacturer/bank would be the most natural owner of the vehicles. Rental car companies will be significantly impacted.
Because of the capital cost of this model, it would be interesting to explore whether an Uber or a vehicle manufacturer/bank would be the most natural owner of the vehicles. Rental car companies will be significantly impacted.
3.d These tradeoffs make an UberGoogle most profitable when:
- "Cruising" miles are kept to a minimum to reduce VMT penalty (dense urban core in a fleet sized for balancing peak hours with variable pricing)
- Passenger total VMT are kept to a minimum to minimize breakeven of self-ownership alternative (intra-urban trips)
- Vehicles have very high capital cost relative to operating cost (think a luxury SUV that costs 10x to purchase but only 2x to operate)
- Margins are kept low enough to dissuade competitive entry
This is a pretty dynamic problem - in a high-density area cruising is minimized and total miles driven are lowest, but profit-taking (margins) will be challenging because of minimum costs of entry of a competing platform - or the low cost of owning your own dedicated commuting car.
In a low-density area, cruising miles are much higher and so are total passenger miles. So the UberGoogle variable cost penalty increases, with a population which naturally incurs more miles over which it can amortize self-ownership. This group ends up paying a low premium to be able to personalize their car and keep stuff in it between rides.
Total vehicle miles driven doesn't reduce (if anything, it increases unless something like dynamic carpooling increases).
3.e Low-cost public transit replacement is where things get really interesting
The low cost/mile of UberGoogle when compared to public transit is the real kicker - it's likely to draw more users out of public transit than replace existing private cars. Most public services operate under a subsidy model; that is, since they aren't allowed to profit, they "reinvest" their profitable operations into unprofitable ones.
When competition cherry-picks the profitable routes (such as FedEx/UPS did to the USPS), the public service may collapse unless externally subsidized. Most transit systems globally operate at a loss in this manner already, with the amount of subsidy a local political decision.
I'll cover these dynamics in a later post, but this dynamic is likely to accelerate the adoption of congestion pricing in the urban cities most suitable to Uber. These taxes on roads will be used to subsidize public transit, essentially making it free (or even paid if the abuse incentives can be managed) in order to keep street traffic to a manageable level.
There are some interesting distinctions as to whether those taxes would be implemented in the form of a mileage tax that varies by boundary, a "toll" to cross boundaries (including an infinite toll such as a maximum number of operating licenses within a zone), or some combination thereof.
The Earth Institute paper seems flawed in four ways. As a result, the scenarios ascribe benefits to shared autonomous vehicles that are actually artifacts of the scenario construction.
First, the scenarios conflate (artificially high) fixed costs with variable costs. On page 13 it references a 2012 AAA study showing that two thirds of the $0.59/mile POV ownership cost (15 15k miles/year) of a small sedan is fixed costs. For comparison, Edmunds estimates lower fixed costs on a Civic and costs at $.50/mile at 15,000 miles/year
While $.16/mile seems reasonable for gas & maintenance (assuming $3.50/gal gas), AAA also incorrectly assumes that insurance and depreciation are almost entirely fixed. The paper also doesn't recognize the higher commercial insurance cost for fleet vehicles. AAA also tags vehicles with a $440/year "license, registration fee, and taxes" which seems quite high.
The net effect is that it sets a much higher fixed cost bar for the POV sedan to cross, and doesn't recognize that fleet vehicles are inherently more expensive in direct operating cost per paid passenger mile than an identical POV.
In Section 1 above, I outlined what I believe is a more accurate approach. It's hard to imagine that fleet vehicles have lower variable operating costs than POV due to higher insurance costs; the question is how much of the "fixed" costs does the POV need to amortize, and how much platform margin is the fleet operator collecting.
Second, the scenarios assume away empty VMT. When you make this assumption, it's easy to pretend cruising disappears with minimal wait times and high utilization percentage of taxis. In reality, most short rides are commutes to/from work and lunch. Even if ride start/end points were evenly distributed, the vehicles needed for commuting would be idle during the rest of the day regardless of who owned them. A close look at the "ground truth" in Fig 1 here illustrates the VMT challenge in NYC - a city with much higher population density than Ann Arbor.
Third, the scenarios assume a capped fleet rather than dynamic interchange of vehicles depending on the going platform cost. It (more or less reasonably ignoring the two issues above) recognizes that amortization of all fixed costs disappear when scaling above about 1,000 rides/hour for a fleet in an Ann Arbor-sized town. This approach doesn't seem to recognize that fleet operators will need to earn a margin, and the charging of this margin will also change the calculus of whether "unused" vehicles "join" the platform during peak periods.
The question here is capital efficiency - what type of return the platform is able to charge will impact its ability to finance the vehicles that appear on it. A platform rate high enough to finance its own vehicles will paradoxically encourage substitution of owned vehicles at the margins. It seems like the equilibrium here is some minimum number of dedicated "platform" vehicles optimized for intra-city commuting, which sounds a lot like an autonomous version of today's taxi fleet.
Because fleets of higher-cost "upscale" vehicles would only come online when pricing hit their marginal cost, they would remain mostly unused - either privately owned, or in nicer "date night" or "executive" fleets which could routinely charge for the excess VMT. Because of the low cost of platform entry, it seems like you could plausibly imagine a "BMW" fleet, etc. Perhaps these vehicles would loan themselves to the Uber fleet during peak hours - occasionally you'd get an unexpectedly nicer ride. Instead, I would expect separate ride networks - if coordinated by Uber it would be for a minimum margin.
Outside of the densest urban cores, higher VMT means most people would be using a personally owned autonomous vehicle, and the hassle factor of not storing stuff in your own car / additional wear and tear will prevent them from part-time contribution to any shared vehicle fleet.
Fourth, the scenarios articulate an opportunity cost of driving time as additional "upside". This is a problem because (a) it ascribes average wage earnings to marginal cost of time. Under this logic, sleeping at night is so costly no one should do it. (b) It also assumes a passenger will be just as productive as when he's not in the car. (c) It also seems to account for the time spent searching for parking, which disappears if autonomous vehicle ownership is as low-cost as the author assumes
Taken together, the proposed scenarios are strawmen where the benefit ascribed to the autonomous fleet is mostly an artifact of the scenario construction.
The question is not whether a centrally controlled fleet of autonomous electric vehicles is cheaper than gas-powered medallioned Yellow Cabs with drivers and legally mandated fee structures and coordination prevention (hint: it is)
The question is rather what dynamics will result from open competition - and whether sufficient fees will accrue to Uber to justify the cost of expansion (and capital cost of owning own vehicles).
A related question is compensation to medallion owners for the loss of value incurred by side-stepping the regulations; as noted in last week's Executable Strategy post, the amount of medallion value lost in NYC alone appears equivalent to Uber's global investment to date.
TL;DR With driver costs currently outweighing high parking fees, Uber appeals to a wider market today - and enjoys higher margins - than it likely will once autonomous vehicles come online. Cannibalizing from public transit may help, but will likely also trigger a significant regulatory response
Thanks for reading,
Greg
*While I have considerable respect for Jeff Sachs, he appears to have had little to do with that paper.
It may have been written in large part by a "William C Jordan" of "Jordan Analytics LLC."
Jordan Analytics LLC and "William C Jordan" may refer to an R&D analyst at General Motors, per INFORMS class of 2007. I made no attempt to contact him, or the Earth institute, or anyone else, prior to publishing this post. If I do, I'll update (including their response) here
**Note, I used the 2015 model as a base because Edmund's True Cost to Own (TCO) tool does not appear to allow selection of past years; I used the historic data of a 2010 model to illustrate the breakdown of the age and mileage components of depreciation. This data is illustrative, not analytic
***Depreciation is often seen as fixed. After driving a brand-new car off the lot, where it immediately loses 10-15% of its value, depreciation is as much mileage-based as time-based. This is complicated and varies by model and year due to shifts in production volumes, gas prices, buyers' preferences, etc.
For an example, consider a 2010 Honda Civic DX 4-door automatic; if sold at dealer invoice of $14,434, when it left the lot it was only worth about $13,000
Today, that car in outstanding condition has a base value of $7,713 for a private party transaction in Pittsburgh, PA - it lost about $5,300 in value
Coda: There are four main flaws in the Earth Institute's paper
The Earth Institute paper seems flawed in four ways. As a result, the scenarios ascribe benefits to shared autonomous vehicles that are actually artifacts of the scenario construction.
First, the scenarios conflate (artificially high) fixed costs with variable costs. On page 13 it references a 2012 AAA study showing that two thirds of the $0.59/mile POV ownership cost (15 15k miles/year) of a small sedan is fixed costs. For comparison, Edmunds estimates lower fixed costs on a Civic and costs at $.50/mile at 15,000 miles/year
While $.16/mile seems reasonable for gas & maintenance (assuming $3.50/gal gas), AAA also incorrectly assumes that insurance and depreciation are almost entirely fixed. The paper also doesn't recognize the higher commercial insurance cost for fleet vehicles. AAA also tags vehicles with a $440/year "license, registration fee, and taxes" which seems quite high.
The net effect is that it sets a much higher fixed cost bar for the POV sedan to cross, and doesn't recognize that fleet vehicles are inherently more expensive in direct operating cost per paid passenger mile than an identical POV.
In Section 1 above, I outlined what I believe is a more accurate approach. It's hard to imagine that fleet vehicles have lower variable operating costs than POV due to higher insurance costs; the question is how much of the "fixed" costs does the POV need to amortize, and how much platform margin is the fleet operator collecting.
Second, the scenarios assume away empty VMT. When you make this assumption, it's easy to pretend cruising disappears with minimal wait times and high utilization percentage of taxis. In reality, most short rides are commutes to/from work and lunch. Even if ride start/end points were evenly distributed, the vehicles needed for commuting would be idle during the rest of the day regardless of who owned them. A close look at the "ground truth" in Fig 1 here illustrates the VMT challenge in NYC - a city with much higher population density than Ann Arbor.
Third, the scenarios assume a capped fleet rather than dynamic interchange of vehicles depending on the going platform cost. It (more or less reasonably ignoring the two issues above) recognizes that amortization of all fixed costs disappear when scaling above about 1,000 rides/hour for a fleet in an Ann Arbor-sized town. This approach doesn't seem to recognize that fleet operators will need to earn a margin, and the charging of this margin will also change the calculus of whether "unused" vehicles "join" the platform during peak periods.
The question here is capital efficiency - what type of return the platform is able to charge will impact its ability to finance the vehicles that appear on it. A platform rate high enough to finance its own vehicles will paradoxically encourage substitution of owned vehicles at the margins. It seems like the equilibrium here is some minimum number of dedicated "platform" vehicles optimized for intra-city commuting, which sounds a lot like an autonomous version of today's taxi fleet.
Because fleets of higher-cost "upscale" vehicles would only come online when pricing hit their marginal cost, they would remain mostly unused - either privately owned, or in nicer "date night" or "executive" fleets which could routinely charge for the excess VMT. Because of the low cost of platform entry, it seems like you could plausibly imagine a "BMW" fleet, etc. Perhaps these vehicles would loan themselves to the Uber fleet during peak hours - occasionally you'd get an unexpectedly nicer ride. Instead, I would expect separate ride networks - if coordinated by Uber it would be for a minimum margin.
Outside of the densest urban cores, higher VMT means most people would be using a personally owned autonomous vehicle, and the hassle factor of not storing stuff in your own car / additional wear and tear will prevent them from part-time contribution to any shared vehicle fleet.
Fourth, the scenarios articulate an opportunity cost of driving time as additional "upside". This is a problem because (a) it ascribes average wage earnings to marginal cost of time. Under this logic, sleeping at night is so costly no one should do it. (b) It also assumes a passenger will be just as productive as when he's not in the car. (c) It also seems to account for the time spent searching for parking, which disappears if autonomous vehicle ownership is as low-cost as the author assumes
Taken together, the proposed scenarios are strawmen where the benefit ascribed to the autonomous fleet is mostly an artifact of the scenario construction.
- Scenario 1 (Ann Arbor): The chart on p.15 incorrectly ascribes fixed cost savings to shared, driverless vehicles that are actually variable costs. It doesn't account for platform margins or a realistic VMT penalty. Finally, the savings described by autonomous purpose-built vehicles are mostly an artifact of the lower cost vehicle and could be earned by a private owner for much less than the gap implies
- Scenario 2 (Babcock Ranch, FL): The charts on p.20 and 21 correctly describe the costs of centrally providing a shared ride service within Babcock Ranch using autonomous, conventionally powered vehicles. It is an ideal environment - defined as a closed community with infrequent, short, internal trips. While a small population could certainly benefit from a shared fleet, the scenario indicates (on p 21) that the driverless fleet could "complement the use of personally owned vehicles." This scenario makes the most sense if personally owned vehicles weren't autonomous, or parking costs were extremely high. Otherwise, why wouldn't someone use their autonomous POV instead of the shared service?
- Scenario 3 (Manhattan, NYC): The charts on p.24 and p.25 comparing UberGoogle to Yellow cabs make several apples/orange errors resulting in an artificially lower operating cost and higher utilization. It's trite to assume violation of the laws yellow cabs are required to follow (fee structure, vehicle use, no centralized coordination) and on top of it assume an empty VMT rate that's a fraction of what existing fleets experience (and research suggests is not achievable).
The question is not whether a centrally controlled fleet of autonomous electric vehicles is cheaper than gas-powered medallioned Yellow Cabs with drivers and legally mandated fee structures and coordination prevention (hint: it is)
The question is rather what dynamics will result from open competition - and whether sufficient fees will accrue to Uber to justify the cost of expansion (and capital cost of owning own vehicles).
A related question is compensation to medallion owners for the loss of value incurred by side-stepping the regulations; as noted in last week's Executable Strategy post, the amount of medallion value lost in NYC alone appears equivalent to Uber's global investment to date.
TL;DR With driver costs currently outweighing high parking fees, Uber appeals to a wider market today - and enjoys higher margins - than it likely will once autonomous vehicles come online. Cannibalizing from public transit may help, but will likely also trigger a significant regulatory response
Thanks for reading,
Greg
*While I have considerable respect for Jeff Sachs, he appears to have had little to do with that paper.
It may have been written in large part by a "William C Jordan" of "Jordan Analytics LLC."
Jordan Analytics LLC and "William C Jordan" may refer to an R&D analyst at General Motors, per INFORMS class of 2007. I made no attempt to contact him, or the Earth institute, or anyone else, prior to publishing this post. If I do, I'll update (including their response) here
**Note, I used the 2015 model as a base because Edmund's True Cost to Own (TCO) tool does not appear to allow selection of past years; I used the historic data of a 2010 model to illustrate the breakdown of the age and mileage components of depreciation. This data is illustrative, not analytic
***Depreciation is often seen as fixed. After driving a brand-new car off the lot, where it immediately loses 10-15% of its value, depreciation is as much mileage-based as time-based. This is complicated and varies by model and year due to shifts in production volumes, gas prices, buyers' preferences, etc.
For an example, consider a 2010 Honda Civic DX 4-door automatic; if sold at dealer invoice of $14,434, when it left the lot it was only worth about $13,000
Today, that car in outstanding condition has a base value of $7,713 for a private party transaction in Pittsburgh, PA - it lost about $5,300 in value
- With 0 miles it's worth $2,235 more, or $9,948
- A $3,000 loss in five years, or about $50/month due to age
- At 12,000 miles per year, or 1,000 miles/month, that's $.05/mile
- With only 50,000 miles it's worth $917 more, or $8,630
- That's $.026/mile for the first 50,000 miles
- With 100,000 miles it's worth $1,266 less than base value, or $6,647
- That's $.04/mile for the second 50,000 miles
No comments:
Post a Comment
Focused, topical critiques are welcome though my responses may be limited. Off-topic comments, trolling, spam, or anything not pulling its weight will be deleted.