Approximating diversion ratios for retail chain mergersmergersPresentation to CRESSE European Conference on Competition & Regulation 5 July 2008on Competition & Regulation, 5 July 2008

Chris Walters*

Assistant Director, Mergers

*Views expressed are mine only*Views expressed are mine only

OverviewOverview

E t i th d l b d● Econometric methodology based on Somerfield/Morrisons (2005) for approximating diversion ratios to measure competitive effects ofdiversion ratios to measure competitive effects of local retail mergers

● Apply methodology to two recent (2006) local retail mergers examined by OFT and CC: Vue/Ster (cinemas) and Waterstone’s/Ottakar’sVue/Ster (cinemas) and Waterstone s/Ottakar s(book stores)

● Suggest some thresholds against which predicted diversion ratios can be judged

BackgroundBackground

L l t il t d t h b● Local retail mergers tend to have been examined by OFT/CC using isochrone/fascia count/market share rulescount/market share rules

● E.g. supermarkets, book stores, cinemas, pharmacies, bingo halls, off-licences, licensed betting offices, funeral homes, pubs

● This methodology is convenient but may be unrealistic. Alternative?

● Diversion ratios—closeness of competition

Diversion ratiosDiversion ratios

● Diversion ratio from A to B represents proportion● Diversion ratio from A to B represents proportion of revenue from A’s customers who would choose B as their second choice as opposed to C ppor D (cross-pB to A/own-pA)

● In an undifferentiated/equidistant market● In an undifferentiated/equidistant market, diversion ratios match market shares: if A, B and C each have 33% shares, the diversion ratio from A t B i 50% (33%/66%)A to B is 50% (33%/66%)

● In differentiated single-good Bertrand model, g gdiversion ratios may be combined with margins to ‘predict’ post-merger price increases

Somerfield/Morrisons (2005)Somerfield/Morrisons (2005)

C l t d i iti b S fi ld f 115● Completed acquisition by Somerfield of 115 geographically non-contiguous, mostly mid-range supermarkets from Morrisonsrange supermarkets from Morrisons

● Isochrone/fascia count/market share rules suggested acquisition of 56 of these supermarkets potentially problematic

● Survey of 5,400 shoppers at these stores asking for second choice supermarket

Survey resultsSurvey results

40%

50%

60%

20%

30%

40%

0%

10%

e s s n e m k h y y r n k n e n ll ll n m s et m y e d d g d e n s n e y a e o o k ll d y y w m n n r ll n e h n s p

John

ston

eP

eebl

esS

outh

Shi

elds

Pock

lingt

onFr

ome

Yar

mN

ewar

Mid

dles

brou

g hFi

ley

Pais

ley

Brid

ge o

f Wei

Whi

tbur

nPr

estw

icC

hape

l Alle

rton

Pool

eSt

ockt

onC

ambe

rwe

Cor

nhi

Littl

eham

pto n

Balh

amLo

ftus

Uni

on S

tree

Wic

kham

Birtl

eySh

andw

ick

Plac

eKn

ight

swoo

dO

akw

ood

Kilw

inni

ngM

uire

ndW

imbo

rne

Brec

hin

Pres

ton

Pans

Ryt

onC

arlu

keB

roug

hty

Ferr

yAl

exan

dria

Kilb

irnie

Ca

D'O

roKe

lso

Selk

irkW

esth

iPo

ntel

and

Thor

nbur

yD

erby

Ren

frew

Cot

tingh

amD

avid

son'

s M

ain

Bed

lingt

onD

onca

ste

Ferr

yhi

Sprin

gbur

nH

ale

Kils

yth

Kilb

urn

Saltc

oats

Sid

cup

Approximating diversion ratiosApproximating diversion ratios

Di i ti b l b i t● Diversion ratios can be laborious to measure, e.g. with customer surveys, especially at phase 1

● Can they be approximated using the traditional isochrone/fascia count/market share methodology?

● Econometric model relating diversion ratios to● Econometric model relating diversion ratios to such local market characteristics

Pseudo ML fractional logit modelPseudo-ML fractional logit model● Diversion ratios (d) are 80%

90%

100%

● Diversion ratios (d) are non-negative and cannot exceed 1, therefore usual to assume they are 40%

50%

60%

70%

80%

1/[1

+ex

p(-X

ygenerated by a model d=1/[1+exp(-xβ)]

● Logit transformation 0%

10%

20%

30%

40%

d=1

β)]

● Logit transformation gives ln[d/(1-d)]=xβ, which can be estimated by OLS

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10

Xβ

4

6

8

10

y

● However, 8 diversion ratios in sample are 0, so preserve these by using -4

-2

0

2

4

ln[d

/(1-d

)]

preserve these by using pseudo-ML estimator instead of OLS (Papke & Wooldridge, 1996)

-10

-8

-6

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10

Xβ

Pre merger independent variablesPre-merger independent variables

● Di i ti rvey

● Diversion ratios implied by market shares .5

.6ra

tio fr

om C

C s

u

● Number of proximity stores in isochrone .3

.4om

er d

iver

sion

r● Drive time to nearest

proximity store .1.2

-Som

erfie

ld c

usto

● Number of competing fascias

0M

orris

ons -

0 .1 .2 .3 .4 .5 .6Morrisons-Somerfield diversion ratio implied by pre-merger market shares

Results (marginal/partial effects)Results (marginal/partial effects)Independent variable dy/dx Std Err z P>z X*Independent variable dy/dx Std. Err. z P>z X

Number of proximity stores 0.027 0.008 3.25 0.001 1

U b d † 0 048 0 022 2 23 0 026 1Urban dummy† 0.048 0.022 2.23 0.026 1

Drive time to closest proximity store -0.000 0.003 -0.13 0.899 4.8

Urban dummy X proximity drive time -0 014 0 006 -2 37 0 018 3 0Urban dummy X proximity drive time 0.014 0.006 2.37 0.018 3.0

Diversion ratio from pre-merger market shares 0.183 0.067 2.73 0.006 0.1

Number of pre-merger competing fascias -0.030 0.012 -2.46 0.014 3Number of pre merger competing fascias 0.030 0.012 2.46 0.014 3

†For discrete change from 0 to 1.

*Value of independent variable at which marginal effect is calculated (generally mean or median of independent variable).

Predicted and actual diversion ratios

.4.3

atio

.2dive

rsio

n ra

.1P

redi

cted

0

0 .1 .2 .3 .4Surveyed diversion ratio

DiscussionDiscussion

M d l t di t l di i● Model appears to over-predict low diversion ratios (<20%) and under-predict high diversion ratios (>20%)ratios (>20%)

- Could add other independent variables but o ld competition a thorit ha e these ewould competition authority have these ex

ante?

F h i hi d d di i● Further, is this over- and under-prediction an issue from the perspective of the competition authorities?authorities?

ExtensionExtension

Si lt it b t di i ti d b● Simultaneity between diversion ratio and number of competitors?

- Customers come to supermarkets but supermarkets locate where customers are

● Instrument number of proximity stores and fascia count with population in isochrone, number of

ki d i i 2car parking spaces and store size, using 2-step procedure of Wooldridge (2005)

● Results essentially unaltered

Sensitivity analysisSensitivity analysis

A l d l t d t l t i bl● Apply model to data on explanatory variables from Vue/Ster (cinemas) and Waterstone’s/Ottakar’s (book stores)Waterstone s/Ottakar s (book stores)

● Both comparable to supermarkets in terms of multi-product nature, differences in store size, small number of large players

● But unitary demand for cinemas, and book stores are not destinations

Vue/Ster (2006)Vue/Ster (2006)● In April 2005 Vue acquired 6 Ster multiplex cinemas● In April 2005, Vue acquired 6 Ster multiplex cinemas

- No national concerns- Local overlaps in Basingstoke, Edinburgh, Leeds and p g , g ,Romford

● OFT examined fascia counts of multiplexes and market shares in 20-minute drivetime isochronesshares in 20 minute drivetime isochrones- Found possible problems in Basingstoke, Leeds and

Romford● CC examined 4 overlaps on a case-by-case basis

- Found problem only in Basingstoke and Vue divested the acquired cinemaacquired cinema

Results for Vue/SterResults for Vue/Ster

Market shares Implied Predicted

Local National diversion ratio Proximity diversion

Locality Ster Vue Ster Vue Local National Number Drivetime Fascias ratio

Basingstoke (urban) 47 53 14 2 100.0 2.6 1 5 2 65.2

Cardiff (urban) 25 0 14 2 0.0 2.6 0 30 5 2.6

Edinburgh (urban) 15 16 14 2 18.8 2.6 1 5 5 4.9

Leeds (urban) 13 9 14 2 10.3 2.6 1 6 5 3.9

Norwich (rural) 31 0 14 2 0.0 2.6 0 30 3 4.7

Romford (urban) 100 0 14 2 0.0 2.6 0 22 1 2.7

Waterstone’s/Ottakar’s (2005)Waterstone s/Ottakar s (2005)

● In September 2005 Waterstone’s announced it intended to● In September 2005, Waterstone’s announced it intended to acquire 141 Ottakar’s book stores

- No national concerns- 33 local overlaps (within 1 mile on high street)

● OFT analysis emphasized degree of direct competition between Waterstone’s and Ottakar’s on non-price factors

● CC cleared merger unconditionally given no difference in range or service quality between overlap and non overlaprange or service quality between overlap and non-overlap stores

- CC surveyed customers at 33 overlap locations and 40CC surveyed customers at 33 overlap locations and 40 comparable non-overlap locations

- Obtained diversion ratios at 33 overlap locations

Results for Waterstone’s/Ottakar’sResults for Waterstone s/Ottakar s1

7.8

.9tio

.5.6

.7di

vers

ion

rat

.3.4

Pred

icte

d d

.1.2

.1 .2 .3 .4 .5 .6 .7 .8 .9 1Surveyed diversion ratio

Thresholds for diversion ratios (1/3)● Three sources of possible thresholds● Three sources of possible thresholds

● First, from increment to HHI (Δ) per OFT Merger Guidelines

● For HHI of 1,000, Δ of 100 problematicFor HHI of 1,000, Δ of 100 problematic

- Implies 7.6% threshold (7.1%/92.9%)F HHI f 1 800 f 0 bl i● For HHI of 1,800, Δ of 50 problematic

- Implies 5.3% threshold (5%/95%)

Thresholds for diversion ratios (2/3)

S d f i l d f i t l● Second, from previously used fascia count rules

● OFT and CC previously have used 5-to-4 and 4-p yto-3

● S mmetric 5 to 4 merger implies 25% threshold● Symmetric 5-to-4 merger implies 25% threshold (20%/80%)

● Symmetric 4-to-3 merger implies 33% threshold (25%/75%)

Thresholds for diversion ratios (3/3)

Thi d f CC M G id li bi d● Third, from CC Merger Guidelines, combined market share of 25% potentially problematic

● Combined symmetric 25% market share implies 14.3% threshold (12.5%/87.5%)

● Not equivalent to 8-to-7 fascia count as nothing assumed about market shares non-merging firmsassumed about market shares non merging firms

Further extensionsFurther extensions

L t t l d l?● Latent class model?

.3.4

io.2

dict

ed d

iver

sion

rat

0.1

Pred

● Pool samples for Somerfield/Morrisons and

0 .1 .2 .3 .4Surveyed diversion ratio

● Pool samples for Somerfield/Morrisons and Waterstone’s/Ottakar’s?

ConclusionsConclusions

● Di i ti b i t d i t diti l l● Diversion ratios can be approximated using traditional rule-of-thumb isochrone/fascia count/market share rules

● M d l di t ibl di i ti i 25 t f● Model predicts sensible diversion ratios in 25 cases out of 37, when applied to Vue/Ster and Waterstone’s/Ottakar’s

● Se eral important ca eats and possible e tensions● Several important caveats and possible extensions, however