Variance screens for detecting collusion · an application to two cartel cases in Italy Variance screens for detecting collusion: an application to two cartel cases in Italy Fabio

1

Variance screens for detecting collusion:an application to two cartel cases in ItalyVariance screens for detecting collusion:an application to two cartel cases in Italy

Fabio Massimo Esposito, Massimo FerreroAutorità Garante della Concorrenza e del Mercato / Italian Competition Authority

2nd ACLE Workshop on Forensic Economics in Competition Law Enforcement, Amsterdam, The Netherlands, march 17, 2006

2

Review of the literatureReview of the literatureÜ Theoretical:

Athey, Bagwell and Sanchirico (2004)

Harrington and Chen (2004)

Ü Empirical:

Connor (2004)

Abrantes-Metz, Froeb, Geweke and Taylor (2005)

Bolotova, Connor and Miller (2005)

3

Case I: Italian motor fuel market cartelCase I: Italian motor fuel market cartelÜ2000: Italian Competition Authority (“AGCM”) found that

all major oil companies in Italy created a cartel, implemented through so-called "brand agreements" between the companies and their distribution networksÜRetail price is set by service stations, but the companies

recommend a retail price to their network members. ÜThe companies set up a mechanism to determine the

purchasing price for service stations based on decreasing discounts as the quantity sold increased,

strong disincentive for the stations to diverge from the recommended price levels

4

The dataThe data

ÜWeekly prices (net of taxes) for gasoline and diesel fuel, for the EU-15 member states, from 1998 to 2005, collected by the European CommissionÜPrice series not stationary price levels turned into

differences of logs (≈ percentage changes), whose series are stationaryÜThis allows a straight comparison of the variances,

without having to weight them by the mean of the series

5

ResultsResultsÜ Italy’s standard deviations are the lowest both for

gasoline and diesel fuel markets (both in the whole sample 1998-2005, and in separated subsamples 1998-99 and 2000-05)ÜStandard deviations computed separately for each year

from 1998 to 2005 show that prices volatility in Italy was the lowest, the second-last or the third-last in every year for both markets, except in 2005.Ü Italy average prices for 2002-04 were the second-

highest among Euro-area markets: only The Netherlands for gasoline and Ireland for diesel fuel have average prices higher than Italian ones.

6

CommentsCommentsÜ According to the analysis Italian motor fuel markets could

have been the scene of a cartel among major oil companies.Ü AGCM investigation confirmed it for the years until 1999; the

conspiracy scenario seems nevertheless likely also for the period after the investigation

Ü Are there alternative reasons for these price patterns?

higher share of retail price in Italy attributable to tax elements

higher distribution network costs in Italy

7

Case II: personal care and baby food products in pharmaciesCase II: personal care and baby food products in pharmaciesÜ 2002: Italian Competition Authority fines national, regional

and provincial professional associations of pharmacies and pharmacists

Ü Documents proving attempts to fix prices of personal care and baby food products sold both in pharmacies and supermarkets

Ü Creation of price lists, advises to apply producers’ list pricesÜ Stronger collusive behaviour in Northwestern Italy (Piedmont

and Liguria), Lombardy and Emilia Romagna

8

DataDataÜ AGCM bought from Nielsen time series of volume and value

sales of the two most representative brands of selected personal care and baby food products sold both in pharmacies and in supermarkets, in 8 Italian macroregions (Northwest, Lombardy, Northeast, Emilia Romagna, Centre, Lazio Centre South, South)

Ü Sample: 45 product items, 25 months (march ’99 -march 01), 8 macroregions, 2 channels

Ü 1 antiseptic, 6 deodorants, 4 toothpastes, 5 toothbrushes, 16 feminine sanitary pads, 1 personal hygiene soap, 7 baby cereals, 5 baby food product items

Ü Average national price for products considered were higher in pharmacies, notwithstanding decreasing market shares

9

Sample of typical pharmacies' series (Emilia Romagna, thousands of italian liras)

3,5000

4,5000

5,5000

6,5000

7,5000

8,5000

mar-99

apr-99

mag-99

giu-99

lug-99

ago-99

set-99

ott-99

nov-99

dic-99

gen-00

feb-00

mar-00

apr-00

mag-00

giu-00

lug-00

ago-00

set-00

ott-00

nov-00

dic-00

gen-01

feb-01

mar-01

feminine pads 1

antisepticpersonal hygiene soap

feminine pads 4toothpaste 3

toothpaste 4

feminine pads 16baby food 4

feminine pads 6toothbrush 1

toothbrush 2toothbrush 3

10

Analysis setupAnalysis setupÜ Many series showed one almost flat segment, some were

“staircase-like” (almost flat segments separated by jumps)computing variances over the entire sample

would not allow to catch possibly collusive behaviourcompute variances over subsamples

Ü 6-month long subsamplesÜ Non-stationarity work on differences of logs,

making series (almost) stationary and directly comparable (prices are in different units)

Ü analysis carried out over standard deviations computed on subsamples of six subsequent differences of logs of prices.

11

Supermarkets as a benchmarkSupermarkets as a benchmarkÜUse supermarkets as a benchmark: lower average

prices, different institutional arragement make them more competitive

ÜAverage standard deviation is (slightly) greater for supermarkets is there a problem ? NO

Üpharmacies’ standard deviations (i) have a lower minimum (ii) 2,7% of them are below minimum supermarkets’ st.dev. (iii) about 30% of them are below the first quartile of supermarkets’ st.dev.

ÜThe benchmark has discriminatory power

12

Benchmark 1: lowest standard deviation in supermarkets subsamplesBenchmark 1: lowest standard deviation in supermarkets subsamplesÜ For each product item and each region, the minimum

standard deviation of subsamples of differences of logs of pharmacies prices was computed.

Ü If such a minimum was lower than the benchmark, the series was classified as “collusive”

Ü 47 “collusive” series, mostly belonging to Emilia Romagnaand Lombardy (23,4% each) , none to Northeast

Ü broadly consistent with AGCM findings, but stronger evidence for Northwest expected.

Ü However, visual inspection revealed that several series showing typically “collusive” patterns were not selected.

13

Noth East - series not caught by absmin threshold (price levels, thousands of liras)

3,0000

4,0000

5,0000

6,0000

7,0000

8,0000

9,0000

10,0000

mar-99

apr-9

9

mag-99

giu-99

lug-99

ago-9

9se

t-99ott

-99

nov-9

9dic

-99

gen-0

0feb

-00

mar-00

apr-0

0

mag-00

giu-00

lug-00

ago-0

0se

t-00ott

-00no

v-00dic

-00

gen-0

1feb

-01

mar-01

0,0000

2,0000

4,0000

6,0000

8,0000

10,0000

12,0000

14,0000

16,0000

feminine pads 4

toothpaste 2toothpaste 4antisepticpersonal hygiene soaptoothbrush 4toothbrush 5

deodorant 6

14

Benchmark 2: minimum productstandard deviationBenchmark 2: minimum productstandard deviationÜ 45 benchmarks: the minimum standard deviation of

subsamples of differences of logs of supermarket prices for each product

Ü 117 “collusive” series, mostly belonging to Emilia Romagna(>50% of series dubbed “collusive”), Lombardy (>50%) and Northwest (42%)

Ü Greater consistency with AGCM findingsÜ In addition, Centre and South are signalled as macroregions

deserving deeper investigationÜ A few “collusive” series maybe not caught, very few “non

collusive” caught

15

Methodological remarks /1Methodological remarks /1Ü significant standard deviations may occur even if there

is collusion, in particular when a series is characterized by strong stability over two periods separated by a jump

Ü small standard deviations may be due not to collusion, but to menu costs or other elements: antiseptic and personal hygiene soap product items display strongly collusive patterns, but they account for 2% of category sales in pharmacies; listed prices could have worked for them as a transaction costs reducing device, instead of a focal point for collusion

In the motor fuel case, high share of taxes may have reduced price volatility compared to other countries

16

Methodological remarks /2Methodological remarks /2Ü what kind of collusion is screened ?Ø low variance of aggregate price implies that neither price

nor quantities are changing very much collusion to stabilize market shares (and relative prices)

MOTOR FUELØ flat sections in aggregate data imply that prices are not only

constant, but also equal (as only in this case quantity variations do not affect average price) collusion is over the same price for everybody PHARMACIES

Ü aggregation matters ? aggregation of a large number ofseries may reduce variance of the aggregate series, biasingthe analysis