Pricing in Designed Markets:

The Case of Ride-Sharing

Jonathan Hall

Uber Technologies

John Horton

MIT & NBER

Dan Knoepfle

Uber Technologies

The “sharing" or "gig" economy

2

Banking

Food

Hotels

Real Estate

Retailing

Healthcare

Transportation

Diversified Labor

Personal Services

Corporate Services

Rental Cars

To what extent do platform markets

"set" prices on their platforms?

The choice of price structure

​

• Platform takes a percentage of transactions; supply side takes the rest
• Some platforms set the price faced by buyers
• Seems sensible, but the comparative statics for a price change are puzzling

Suppose Uber cuts fares in a city by 20%.

Passengers demand more trips

​

But drivers now make less per trip, and presumably want to supply fewer

How is the "gap" closed?

This talk

• Show how the "gap" is closed
• Discuss the economic implications of the market adjustment process

Driver paystub from a week of working

Drivers gross hourly earnings rate

w = $11.30/hour gross earnings rate

Hours of work are transformed into hours of transportation services

(assuming US avg. trip duration)

~16 hours of transportation

Fraction of hours-worked that are "on trip" is "utilization" or technical labor productivity

"Utilization" = x =

Flow of money earned while on a trip

Driver hourly earnings rate decomposed

Flow of $ while on a trip

Fraction of hours- worked on trip (utilization)

*

w =

The supply of driver hours-worked

H(w)

Hours

Hours online

w

The transformation of hours-worked into hours of transportation services: hours scaled by utilization

x H(w)

Hours

H(w)

Hours of driving with passengers

w

Key point: Utilization

changes shift the

product market supply curve

Demand for hours of transportation services

Hours

D(p)

Price /

Hour Transportation

p

A market equilibrium: Hours demand = hours supplied

x H(w)

Hours

D(p)

Price /

Hour Transportation

Driver earnings rate / Hour Worked

Equilibrium quantity of hours of transportation

p

px

Suppose prices are lowered

Fare cut

x H(w)

Hours

Prices

D(p)

1. Uber decreases fare by dp.

p

w

Gap, before the market adjusts

x H(w)

Hours

Prices

D(p)

2. Hourly earnings rate falls by dp x

p

w

"gap"

Paths to a new equilibrium after a fare increase

• Both passengers and drivers are completely inelastic
• Drivers have a downward sloping labor supply curve
• Changes in surge pricing essentially "undo" a fare change
• Service quality deteriorates, shifting in the demand curve
• Increase in driver technical productivity i.e., increase in utilization

​

Paths to a new equilibrium after a fare increase

• Both passengers and drivers are completely inelastic
• Drivers have a downward sloping labor supply curve
• Changes in surge pricing essentially "undo" a fare change
• Yes, in part
• Service quality deteriorates, shifting in the demand curve
• Yes, in part
• Increase in driver technical productivity i.e., increase in utilization
• Yes, in part

​

Results preview

Empirics

N = 43 US Cities with UberX

T = Weeks from

June 2014 to Jan 2017

x_it

UberX price index in city i in week t

y_it

Average outcome in city i in week t

Price Index: Price of a 6 mile, 16 minute unsurged trip that week, in that city.

x_it

UberX price index in city i in week t

y_it

Average outcome in city i in week t

Price Index: Price of a 6 mile, 16 minute unsurged trip that week, in that city.

E.g.,

t = 3/6/2016 - 3/12/2016

i = New York City

x_it , Trip Price Index = $24.04

y_it , Hourly earnings rate = $34.00

Price variation

Fare goes up

Fare does down

(seemingly not touted)

Small between-city differences in the precise week of a change

Between-city variation in size of the change.

Distribution of fare changes

City-week grand means over time

Base trip price index

Long-run decline in fares

But little evidence of sustained higher surge

Surge

Large increases in utilization

Utilization

No persistent changes in the hourly earnings rate

Hourly earnings rate

Identification

Leaks about Uber decision-making on pricing

City-specific changes in the base fare

• Random?
• No - Uber has a team that works on pricing.
• But they seem to have been selecting on "observables"
• Little evidence of forecasting
• Double fixed effects panel (with unit specific time trend) can "handle" selection on observables
• 2021 Update: Who knows what TWFE with continuous treatment does anymore...
• No evidence results are sensitive to city-specific weather, unemployment rate, competitors, etc. - See paper
• The pattern of variation in fare changes looks promising
• E.g., all are "treated," small timing differences, etc.
• Highly granular data makes parallel trends assumption assessable

​

Regression approach

The "long-run" specification

Outcome of interest in city i at week t.

Base price index at city i at week t.

City i specific effect

Week t specific effect

City i specific time trend

Base fare index (log) in that particular city & week

10% fare increase reduces avg. surge multiplier by 2%

10% fare increase reduces utilization/tech. productivity by 7%

Positive but small effect on the gross hourly earnings rate.

Accounting for the adjustment process

​

Showing the dynamics

Main specification: 15 pre, 25 post (see paper for alt. leads/lags)

Surge multiplier

Surge multiplier

No evidence of pre-period pre-trends

Surge multiplier

Gradual decline in surge following a fare increase (~30% by end of the period)

Utilization

Utilization

8% reduction in utilization following a 10% fare increase by week 8; close to 10% near end.

Hourly earnings rate

Hourly earnings rate

Initial pass-through, then steady decline, turning negative by week 8.

Other outcomes related to

market clearing

"D"

"w"

"x"

Surge multiplier (changes in "p")

"H"

Wait-times (related to φ(x))

# trips and hours with passengers decline with higher prices

Driver compensation effects are sensitive to definition, but mostly positive

Clear decline in driver technical productivity with higher prices

Surge goes down, frac. of earnings from promotions goes down; no change in market share

Driver hours-worked increases with higher fares; effect seems to be on the intensive margin

Wait-times clearly increase; some evidence that trips get shorter

Back to the simple model

x S(w)

Hours

Prices

D(p)

p

w

Adjustment: Some increase in surge offsets dp

x S(w)

Hours

Prices

D(p)

p

w

Adjustment: Increase in wait-times (demand shift)

x S(w)

Hours

Prices

D(p)

p

w

D₂(p)

Adjustment: Increase in utilization

x S(w)

Hours

Prices

D(p)

p

w

D₂(p)

x₂ S(px₂)

Utilization works on supply curve two ways: "inside" (more hours) & "outside" (more productive)

Implications

What we know from the emprics

• There is clearly a price/utilization trade-off at the market level
• There is clearly a utilization/wait-time trade-off at the market level
• Some of the base price changes are "undone" by surge
• Welfare effects of fare increases are ambiguous for drivers but probably negative, at least on average
• Fare increases seem to lower hourly earnings rates, but the increase in hours-worked suggest they are better-off overall

​

Who gains and loses?

​

Market clearing condition: hours demanded is hours supplied

Disutility from Passenger wait-time

Driver net hourly earnings rate

This price can reflect platform surge changes

x-axis axis: Driver Utilization

x-axis axis: Driver Utilization

y-axis: Price

Drivers want really

high prices & high utilization

($$$)

y-axis: Price

x-axis axis: Driver Utilization

Driver Happy Place

Passengers want really

low prices & low utilization

(cheap & short wait-times)

y-axis: Price

x-axis axis: Driver Utilization

Driver Happy Place

Pax

Happy Place

But only some of these price/utilization combinations are equally equilibriums

y-axis: Price

x-axis axis: Driver Utilization

Driver Happy Place

Pax

Happy Place

Possible (p, x)

Equilibria

Which one do drivers like?

y-axis: Price

x-axis axis: Driver Utilization

Driver Happy Place

Pax

Happy Place

Drive indifference curve

Which one do passengers like?

y-axis: Price

x-axis axis: Driver Utilization

Driver Happy Place

Pax

Happy Place

Pax

indifference curve

Making both groups happy is a knife-edge condition

Driver-preferred equilibrium

Pax-preferred

equilibrium

But both sides can agree about what to do about prices in certain conditions. E.g., prices are too high:

Both sides want fare cuts

Or prices are too low:

Both sides want fare

increases

But in a range, both sides are implaccable

Drivers want fare increases; Passengers want fare cuts

Thank you

John Horton http://john-joseph-horton.com/

@johnjhorton

Backup slides

Driver labor

Hours-worked per driver

Driver churn

More hours-worked / less exit when hourly earnings rate is higher

Hours-worked per driver

Driver churn

But it persists once hourly wage effect has turned negative.

Why more hours-worked when hourly earnings are lower? Answer: Utilization is costly to drivers.

With lower utilization, less fuel expenditure; less wear and tear.

Not captured: greater effort; short breaks have a higher opportunity cost

Service quality (wait-times)

Median wait time from dispatch to pick-up

With a higher fare/lower utilization, nearest available care is close and so wait-times fall

Market quantities

Much less precise, but clear evidence of utilization trade-off: More hours-worked but fewer hours with customers

Thoughts on welfare

Effect of a fare cut

(ignoring wait-times)

Social planner wants to increase this area as much as possible

Gap between supply and demand versus utilization

Accounting for costs

Costs depend in part on mileage

At $0.54/mile and 6 mile avg. trip, expenses ~$185

Costs per hour of driving depends on the city and utilization

Speed when w/ passengers

Speed when not with passengers

Log hourly costs

net

gross

Log hourly earnings

Implications of endogenous

technical productivity

x, utilization

w, wage

Driver indifference curve

x, utilization

w, wage

More hours

Fewer hours

Satisfying a fixed

amount of demand, D(p)

x, utilization

w, wage

Fewer hours-worked,

high utilization

Many hours-worked, lower utilization

What if there is a lower

product market price?

x, utilization

w, wage

Ways to meet demand with a lower product market price

Higher utilization

More

hours

Possibilities with p = p_H

Possibilities with p = p_L

Which hours/utilization combo have drivers earning MRP (w = px)

(i.e., could be equilibria)?

x, utilization

w, wage

Low price equilibrium in which drivers get MRP

High price equilibrium in which drivers get MRP

Slope = p_L

Slope = p_H

x(p_L )

x(p_H )

Many possible equilibria with depending on the product market price set by the platform

x, utilization

w, wage

A = High price, low utilization

equilibrium

B = Low price, high utilization equilibrium

A

B

What equilibria should "we" prefer?

Consider small changes in utilization, dx

x, utilization

w, wage

Value of more output greater than the cost to worker, or:

p dx > w

Consider small changes in utilization

x, utilization

w, wage

Value of more output greater than the cost to worker, or:

p dx > w

Value of more output less than the cost to the worker, or:

p dx < w

x, utilization

w, wage

Value of more output = cost to worker

​

Should we expect the decentralized efficient price?

Long-history of taxi regulation---and relatively unsuccessful de-regulation attempts in 70s and 80s.

By centralizing pricing, are ride-sharing platforms solving the "Diamond (1971)" problem?

Thank you

Authors: Jonathan Hall, John Horton and Dan Knoepfle

Link: http://www.john-joseph-horton.com/papers/uber_price.pdf

Backup Slides

Promotional payments (e.g., earnings guarantees)

Hourly earnings without promotions

Hourly earnings with promotions

Greater pass through of fare changes without promotional payments included

Alternative within-city approach

• Basic idea: Compare UberX/UberBlack within the same city
• Advantages:
• Alleviates some of the concern about city-specific selection
• City-specific shocks (concerts, sports, very localized weather etc.) are netted out
• Disadvantages
• UberX and UberBlack compete with each other
• In many cities, UberBlack drivers can take UberX requests

Some implications of the market quantity results

• Reduction in trips means price effect outweighs the quality effect
• Change in hours-worked is an intensive margin rather than extensive margin effect
• Long-run hours-worked change despite no change in hourly earnings rate, consistent with highly elastic labor supply to Uber

Uber taking prices was not ex ante obvious

• Factors suggesting inelastic supply:
• Unique flexible work option
• Some past evidence of target earning by drivers
• Regulatory/capital barriers to entry are not zero
• Factors suggesting elastic supply:
• Drivers have complete flexibility of hours-worked
• Barriers to entry are not that high
• Existence of competitor platforms allow driver to easily switch

Related literature

• Spillover effects of the introduction of Uber on:
• Drunk driving (Greenwood & Wattal; Brazil & Kirk)
• Bankruptcy (Nian et al.)
• Entrepreneurship (Burtch, Carnahan & Greenwood)
• Preferences and welfare of riders and drivers
• Value of flexibility to drivers (Chen et al)
• Racial discrimination (Ge et al.)
• Price sensitivity of riders (Cohen et al.)
• Labor supply elasticity of drivers (Chen & Shelden)
• Descriptive
• Who Uber drivers are & what do they want (Hall & Krueger)
• Gig economy labor (Katz & Krueger)
• How does the market "work"? (this paper)

​

Conceptual framework assumptions

• Riders face a fare that has (1) fixed fee and (2) a variable component based on time and distance
• For simplicity, we are going to assume riders buy hours of transportation at a price, p.
• Drivers are paid a % of the gross receipts
• For now, I'm going to assume Uber's take is 0%.
• I'm going to ignore cost of gas, wear and tear, miscellaneous costs, and so on.
• Role of surge pricing:
• By assumption, it "works" to keep wait times more or less constant (hence keeping the demand curve fixed).
• It is applied as a multiplier, m to the base fare rate, b, and so p = bm.
• Demand curves slope down, supply curves slope up

​

Robustness check - can synthetic control approach work for previous outcomes?

Same pattern of results: no long-run change in hourly earnings because of changes to utilization & surge

# of trips

#

hours-worked

# drivers working

Fare increase leads to a reduction in the number of trips taken

Fare increase has seemingly no effect on the number of active drivers

Fare increase leads to a reduction hours