Measuring consumer switching costs in the television industry

by Oleksandr Shcherbakov

October 4, 2016

O. Shcherbakov

Switching Costs

In many markets, consumers are reluctant to switch

Examples: Banking Telecommunication Electricity Paid-TV

O. Shcherbakov

Switching Costs

Consumers repeat purchases from the same provider

Some reasons: Provider consistently oers the best price-and-quality products Exit/termination fee (monetary costs) Installation fee (monetary costs) Too much hassle to switch (utility costs) Brand loyalty (utility costs) Switching costs: additional monetary and utility costs of switching to an

alternative provider

O. Shcherbakov

Switching Costs

Anecdotal evidence of switching costs

Survey results for paid-TV: Approx. 80% of satellite subscribers report high levels of satisfaction with their service (Nielsen Media Research survey, 1997). By contrast, a dramatically lower percentage (45%) of cable subscribers are satised. Yet, only 10% of cable TV consumers indicated that they are very likely to switch to satellite service (Chilton Research Services Survey, 1997).

O. Shcherbakov

Switching Costs

What are the eects of switching costs on competition?

Quantifying switching costs is important to answer the following questions:

1. Do markets become more competitive due to rms' incentives to invest in their customer base? 2. When do rms start extracting rents (harvesting) from locked-in customers?

O. Shcherbakov

Switching Costs

Approach

Step 1. Develop a

dynamic

empirical framework that identies and

estimates customer switching costs using

market-level data.

Identication requires separating the following eects from

aggregate statistics : i. Consumer heterogeneity in preferences ii. Switching costs

Step 2. Simulate counterfactual and compare dierences in the optimal policy of cable TV providers under static and dynamic monopoly and duopoly structures.

O. Shcherbakov

Switching Costs

Outline of Talk

1. Overview of Paid-TV Industry 2. Model (dynamic):

i. Flow Utility ii. Dynamic Programming iii. Purchase Probabilities (random logit) and Market Shares 3. Data 4. Identication 5. Estimation algorithm (nested loops) 6. Results

i. Switching Costs ii. Counterfactual simulations 7. Conclusion

O. Shcherbakov

Switching Costs

Paid-TV Industry

Before early-1990s, cable TV providers were

local monopolies.

Since then, a Direct Broadcast Satellite (DBS) service was launched and there is new entry into the market. Competition is unusual because cable TV providers set prices and quality locally, whereas DBS providers set them at the national level. Products are vertically dierentiated and more expensive bundles uniformly include all the channels from low quality bundles - used to construct scalar quality measure.

O. Shcherbakov

Switching Costs

Users face substantial switching costs

Customers face the following costs when switching providers: Up-front installation fees Equipment purchases Hassle costs (e.g. installation appointments, obtain landlord's permission to install satellite dish, etc.)

O. Shcherbakov

Switching Costs

Model

They propose a dynamic model of consumer behaviour in a market with switching costs. Denitions:

t

- time period (one year).

g ∈ {o, c, s}

- outside option (free over-the-air TV), cable and

satellite services

pgjt

- monthly subscription fee

qgjt

- quality of program content measured as a weighted average

total no. of channels (more weight given to more costly channels)

O. Shcherbakov

Switching Costs

Consumer's Flow Utility

Consumer's ow utility

Ui (ait , ait−1 ) =

from

service

g

is:

 ˜  q¯gt ) + ξgt +εigt −ηig · 1(ait−1 6= g ) + |δig (p¯gt , {z }

if ait = c, s,

 

otherwise

mean utility,δig

εiot

where ηig denote switching costs that are known to consumers ait ∈ {o, c, s} - consumer i 's choice of provider ξgt - unobserved quality shock to econometrician but observed by consumers

(similar to BLP)

εigt - i.i.d. random taste shocks for providers; follows EVT1 distr.

O. Shcherbakov

Switching Costs

Dynamic Programming

The Bellman equation for a consumer's dynamic maximisation problem is:

Vi (Ωt , ait−1 , εit ) =

max {Ui (ait , ait−1 ) + βE [Vi (Ωt+1 , ait , εit+1 )|Ωt , ait , εit ]}

ait ∈{o,c,s}

s.t. Ωt denote current service characteristics and any factors aecting future characteristics. It evolves according to a rst-order Markov process.

O. Shcherbakov

Switching Costs

Purchase Probabilities

In each year (one time period), each consumer type chooses either cable, satellite or the outside option. The conditional probability of choosing service

g

when they were previously with

Pr(ait = g , ait−1 = k) = Pr =



k

is:

−ηig · 1(g 6= k) + Vig + εigt ≥ −ηil · 1(l 6= k) + Vil + εilt , ∀l 6= g



exp(−ηig · 1(g 6= k) + Vig ) exp(Vi0 ) + exp(−ηic · 1(c 6= k) + Vic ) + exp(−ηis · 1(s 6= k) + Vis )

O. Shcherbakov

Switching Costs

Market Shares

Current period

predicted

market shares are given by:

sigt = Pr(ait = g , ait−1 = c) · sict−1 + Pr(ait = g , ait−1 = s) · sist−1 + Pr(ait = g , ait−1 = o) · (1 − sict−1 − sist−1 ) and

aggregate

market shares:

Z sgt =

sgt (p¯ct , q¯ct , ξct , p¯st , q¯st , ξst , sit−1 , ωi ) dGω,sict−1 ,sist−1 (ωi , sict−1 , sist−1 |θ) | {z } sigt

O. Shcherbakov

Switching Costs

Heterogeneity Moments

Predicted product-specic market shares for cable TV:

R

!

1 j = arg max {αip pgj 0 t + αiq qgj 0 t } sigt dGω (ωi , sit−1 |θ)

sj|g ,t =

O. Shcherbakov

j 0 ∈Jgt

R

sigt dGω (ωi , sit−1 |θ)

Switching Costs

Data and Instruments

Cable TV:

Exhaustive information (e.g. market size, no. of subscribers, prices, channel lineups) from 1992-2006. Product-specic shares available at market level (a market,

n

is

dened as the no. of homes passed by a cable provider) Satellite TV

Data collected from internet Only total share of satellite rms are available and at a coarser level (typically spans a few markets)

O. Shcherbakov

Switching Costs

Instruments for Price and Quality

E [pigt , qigt |ξgt ] 6= 0.

Therefore, need to instrument price and quality.

IV 1. Average prices and quality of other cable systems that belong to the same multiple-system operator (MSO) IV 2. Bargaining power of MSO, proxied by the no. of homes passed and no. of subscribers of the parent company IV 3. MSO average capacity level

O. Shcherbakov

Switching Costs

Identication Strategy

Separating the eects of consumer heterogeneity (measured by random coecients) and switching cost parameters from market level data is tough. Therefore, he proposes the following strategy:

Identies

exogeneous shifters

switching cost parameters from

of the

previous period decisions (i.e. if switching costs do not exist, then exog. changes in previous period should not aect current period's decisions) E.g. cost-reducing innovations.

Identies

consumer heterogeneity from variation in observable product

characteristics

across

markets. Product-specic market shares provide

important information on the distribution of preferences.

O. Shcherbakov

Switching Costs

Estimation Strategy

The vector of parameter to estimate is:

(α ¯c , α ¯s , α ¯p , α ¯ q , σαc , σαs , σαp , σαq )

Estimation Algorithm: Inner Loop: Solves dynamic programming problem for

each

consumer type

and calculates aggregate market shares. Middle Loop: Solves for

ξct

and

ξst

that match observed market shares to

the ones predicted by the model. Outer Loop: Searches over the parameter vector for values that minimize

GMM objective function

O. Shcherbakov

Switching Costs

Solving for

ξct

and

ξst

Match model predictions for aggregate provider-specic shares to the ones observed in the data, i.e. solve the following system of equations:

( sct = sct (p¯ct , q¯ct , ξct , p¯st , q¯st , ξst |θ) sst = sst (p¯ct , q¯ct , ξct , p¯st , q¯st , ξst |θ)

O. Shcherbakov

Switching Costs

Moment Conditions

1.

E [ξ˜cnt |zcnt ] = E [ξ˜snt |zsnt ] = 0

2.

E [ucjt |It ] = 0 where

ucjt

(mean independence assumption)

is the measurement and approximation error between the

data and observed predictions of

O. Shcherbakov

product-specic

market shares.

Switching Costs

Larger switching costs in satellite than cable services

O. Shcherbakov

Switching Costs

Switching costs have nontrivial eects on mkt eqlbm

DM vs DD - DM oers 51% lower mean utility than DD (competitive

eects due to satellite entry) DM vs SM - SM oers 89% lower utility (cable providers need to

lower prices to attract customers with high switching costs) SD vs DD - SD oers 40% lower utility (if switching costs do not

exist)

O. Shcherbakov

Switching Costs

When do rms start harvesting?

O. Shcherbakov

Switching Costs

Conclusions

1. Switching costs are signicant and constitute approx. half of the annual variable service costs for each service provider ($190 for cable; $240 for satellite). 2. Professional installation costs only amount to 20% of switching costs, with the remainder attributed to hassle costs. 3. From counterfactual simulations, they nd that due to switching costs, cable rms have incentives to invest in customer base, hence increasing utility.

O. Shcherbakov

Switching Costs