08 October, 2015

Uber (Part 6) - Transit network cross-subsidies make it easier to cherry-pick profitable routes

Summary: At a high level it appears at least plausible that in Pittsburgh today, the entire bus system could be profitably discontinued if the Port Authority operated an Uber-like service instead.

It's common for a business to offer a portfolio of products, some of which are more profitable than others.  This can either strengthen the business, or open it up to devastating competition, depending on the situation.

A product might even be sold at a loss (negative profit) as part of a marketing strategy.  Industries from banking to specialty chemicals, however, have slowly learned this can be a devastating strategy:  it invites new competition to cherry-pick profitable product volume, leaving a company saddled with both loss-making products and legacy overhead costs that are challenging to adjust in the short-term.

This is similarly true with customers who are cheap/expensive to serve/sell to due to idiosyncrasies in their structure or behavior.  Often times, these cross-subsidies aren't intentional - they evolve from organizations' cost structures bloating into the available revenue.

For example, the US postal service has a dual government mandate to provide both universal service and to break-even financially.  But years of having FedEx and UPS cherry-pick the most profitable routes, combined with an internet-driven shift away from "baseload" mailings, have left the postal service struggling to service its legacy costs.  In 2012, even the US government spent only $337m of its $4.8b in mailings with the USPS (7%).  The situation at USPS is going from bad to worse.

This concept of cross-subsidies is both an opportunity and a threat for Uber:

  1. Transit network subsidies are large, particularly in bus networks
  2. Profitable transit routes may fall to Uber-like competition, especially once autonomy is available, sending public transit systems into an accounting death spiral
  3. Meanwhile, Uber should be careful that its adoption of a legacy ride pricing structure doesn't leave it vulnerable to a similar disruption

1.  Transit network subsidies are large

Transit networks such as subways and buses are structured in a way such that some routes subsidize others.  Of course, in the United States the entire system often receives an external government subsidy - for capital and sometimes also operating expenses.

Farebox recovery ratios - the ratio of fares collected to operating costs - are dependent both on strategic investment priorities (favoring cars or transit) and whether variable pricing has been instituted.

Breaking farebox recovery ratios down even farther, individual routes often fail to cover their variable costs.  Here's Cap'n Transit with a breakdown of which NYC bus routes cover their variable costs in 2012.  Note that once indirect overhead costs are considered, none of the NYC bus routes were profitable.

In Alon Levy's 2011 post The Option of Profitable Transit, he cites a useful framework proposed by David Levinson to discuss profit distribution across routes:
Mass transit systems in the United States are collectively losing money hand over fist. Yet many individual routes (including bus routes) earn enough to pay their own operating (and even capital costs). But like bad mortgages contaminating the good, money-losing transit routes are bogging down the system.
We can divide individual systems into three sets of routes:
1. Those routes break-even or profit financially (at a given fare). This is the “core”.
2. Those lines which are necessary for the core routes to break-even, and collectively help the set of routes break-even. These are the “feeders”.
3. Those lines which lose money, and whose absence would not eliminate profitability on other routes. These money-losers are a welfare program. We might politely call them “equity” routes.
Levy continues to write that underused routes are a lesser problem in the US than high labor costs (overstaffing and sometimes salaries), poor design, artificially low fares, bad regulations, and auto-oriented public policy.

His blog - which I highly recommend - is mostly about how better management and operating practices could make public transit more efficient and hence widespread.

One of his recent posts is an interesting discussion on how much more labor intensive US train systems are than their European counterparts - a topic that US train advocates rarely seem to mention (instead focusing on desired capital investments).

2.  Profitable transit routes may fall to Uber-like services, once autonomy is available

In that 2011 post, Levy notes that "the best-performing routes do not form a trunk system, but are for the most part short-hop crosstown buses, with very high ridership per kilometer of route length."

With Uber's $5 minimum fare, it's easy to see that a customer might prefer waiting 5 minutes to take a private Uber directly to a destination within ~2 miles instead of waiting ten minutes to pay $3 to sit on a crowded bus (or pay $1 or even $0 if it's a downtown "circulator").  Particularly if the bus stops on either end require walking.

Customers in core downtown areas with best taxi economics (high demand, low empty VMT) seem like the same places that are the most profitable for bus systems - often because their fares greatly exceed the cost of service.  If bus systems lost these profitable riders, they would lose the economic surplus generated by these riders - which will further increase the subsidies they require to operate.

VOX wrote last summer about Lyft Line (carpooling) being "the beginning of the end of public transit."  In a large part, because "self-driving cars plus ride-sharing will make buses obsolete."

Last week, Levy surprisingly disagreed.  His analysis, though, painted with an overly broad brush.  He compared fully loaded POV vehicle costs of ~$.30/km ($.50/mile) to NYC subway costs of $.21/km ($.35/mile).  This is apples and oranges, both because of fixed/variable cost issues as well as cars vs subway.

On p. 13 of the .pdf that Levy links to (showing operating costs of top 50 transit authorities in US in 2011), the NY MTA shows "operating expense per passenger mile" of $1.35 for a bus and $.34 for heavy rail (subway).

These numbers clearly show that, without subsidy, bus is already perilously close to Uber fares ($2.15/mile in NYC).  On p.21 the Allegheny County Port Authority bus opex/passenger-mile of $1.47 is already greater than the local Uber fare of $1.20/mile.  Even our light rail cost ($1.31/passenger-mile) is greater than the local Uber fare.

Obviously these are route averages; some cost more, some cost less.  It also ignores "flag" fares and minimum fares of Uber.  At a high level it appears at least plausible that in Pittsburgh today, the entire bus system could be profitably discontinued if the Port Authority operated an Uber-like service instead.

And autonomy will only make this choice starker.  In Uber (Part 3c para 3c), we saw that an autonomous electric Civic used in an Uber-like service might have variable operating costs as low as $.11/mile (before insurance).  This is one third of the NYC subway operating costs, and even less if there are multiple people in the car.

Other countries such as China have begun testing autonomous busses; it's not clear if labor agreements in the US will allow for autonomous busses to replace drivers.  Or even if they do, if the service costs would be better spent on a fleet of smaller, more nimble and responsive vehicles like autonomous taxis.

In any case, disruption of the economics of intra-city transit systems, particularly busses, is a likely near-term effect of Uber, particularly once autonomy is in play.

Given the apparent economics, perhaps cities should give serious thought to launching their own Uber-like service.  We'll return to some of these dynamics in a later post, when we look at re-regulation possibilities that Uber may want to encourage.

As part of that re-regulation discussion, the coming urban autonomy apocalypse will likely require defensive congestion pricing of surface street traffic, preferably in a way that further subsidizes public transit systems.

3.  Uber should be careful that its legacy ride pricing structure doesn't leave it vulnerable

We started this post by looking at the risk of having profitable services/customers subsidize unprofitable ones: that organizations can be blind to how attractive "cherry picking" a profitable customer can be, and blind to the hangover induced from excess overhead costs spread among a base of remaining unprofitable customers.

When Uber and its competitors launched, they adopted the legacy taxi pricing structure - a "flag fee", a mileage fee, and a wait-time fee.  This structure makes sense for a decentralized taxi industry, because drivers need to be paid both for waiting as well as empty VMT... as was noted at the "original problem statement" of Uber (Part 2a).

Medallions and higher pricing was an imperfect solution - the pricing is arguably too high, as noted by the rising value of the medallions (discussed at length in my Coda to Executable Strategy: Sweat more in peace, bleed less in war)

Uber/Lyft/etc have not yet developed a pricing model that takes full advantage of technology to price rides in a way that eliminates some riders from being subsidized - either by other riders or by drivers.

If we examine the cost of differing rides, the biggest thing that stands out is empty VMT.  As we saw in Part 4c, empty VMT in NYC varies from 10%-90% depending on time and location, and averages about 40%.  This has two parts, time-based and location-based.  As argued in Part 4c, while perfect planning/coordination by Uber may reduce this, the issue is mostly structural based on ride patterns.

First, passenger fees should vary based on the amount of empty miles the car has to travel to make the pickup.  This should be a "theoretical" base (not an actual one which may be randomly high or low).  In low-demand times of day, there are likely fewer drivers on the road - resulting in higher empty VMT to pick up a passenger.  Per-mile fares should be time-varying to account for this.  This would be, in essence, a "penalty" to requesting a ride at an off-peak time.

Second, passenger fees should vary based on the ride density of both pick-up and drop-off points.  In other words, taking a car out of a high-demand area to a low-demand area is costly - it will require more empty VMT to get the next fare.  And vice versa.  These rides should be priced to reflect the full costs they impose on the network. (there's probably some tradeoffs with idling drivers vs empty VMT that should also be considered)

Of course, this may be hard to explain to consumers - who already have difficulty wrapping their heads around surge pricing.  My personal belief is that with proper demand analytics and scheduling of drivers, surge pricing could be eliminated in favor of this more transparent scheme.  Particularly if it also reflected more explicit framing of allocation of driver pay (one reason there is surge pricing when bars are closing on a weekend night is that fares at that hour are one-and-done).

If it worked, less transparent pricing would enable the platforms to price to get just the users they want (like personal auto insurance).  It also might ease some of the economic pressure of the race-to-the-bottom by forcing riders to price-shop.  The increase in search costs, by increasing the multi-homing costs to riders, may paradoxically give some breathing room from pure price competition.


Under the current system, riders who take Uber at low-demand times or to outlying areas are being subsidized - explicitly through Uber's minimum wage guarantee to drivers (paid out of platform fees), as well as by drivers who aren't reimbursed for empty VMT.

An enterprising competitor could find a way to take advantage of this

Thanks for reading,

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.