Screens: some comments Univ.Prof.dr. Maarten C.W. Janssen CEG Europe.

Screens: some commentsUniv.Prof.dr. Maarten C.W. Janssen

CEG Europe

General Principles

Statistical inferenceDo conflicting hypotheses yields sufficiently different predictions that

on the basis of empirical evidence we can discriminate between

them?

Quality of Data

Cost-benefit analysis whether or not to engage in it Compare cost/benefits alternatives (e.g., leniency program for cartel

detection purposes)

How difficult is it to get data? how likely you get conclusive results?

Sometimes almost no alternative available, e.g., when calculating

damages

How well can we distinguish between collusion and competition?All firms charge same price

Could also be the outcome of Bertrand competition with identical cost

But if firms have different cost …For this we need to have cost data

Cost pass through (how much of cost changes is passed through in terms of prices)

No information on cost per se needed, only cost changes

Cost pass through smaller than 1 sign of market power, collusion

But monopolist with two part tariff also has a cost pass through of 1 in the price per unit

Mobile telephony in Netherlands: retail price per minute does not react (or negatively to termination rates)

Everyone believes this is a competitive market

An example from own research (joint with Riemer Faber)In NL large gasoline companies (Shell, BP, Esso, etc.) daily announce a suggested price to (their) retailers Do these suggested prices help retailers coordinate prices?

Our hypotheses: With suggested prices, price difference between any two pair of retailers much more stable than without them

Retailers advertise “I am x cents below suggested price”

We looked at so-called rank reversal test

For any pair of stations, we look at how often (% of days) the

station that is most often more expensive, is actually cheaper

First we consider Vienna

Wien, Euro 95

Amsterdam, Euro 95

0

200

400

600

800

1000

1200

1400

0.0 0.1 0.2 0.3 0.4

Series: REVERSAL2Sample 1 3782Observations 1865

Mean 0.057176Median 0.004193Maximum 0.469388Minimum 0.000000Std. Dev. 0.097575Skewness 1.986917Kurtosis 6.301016

Jarque-Bera 2073.884Probability 0.000000

Rotterdam, Euro 95

0

400

800

1200

1600

0.0 0.1 0.2 0.3 0.4

Series: REVERSAL2Sample 1 4830Observations 2393

Mean 0.057102Median 0.003623Maximum 0.457143Minimum 0.000000Std. Dev. 0.092841Skewness 1.895103Kurtosis 6.058314

Jarque-Bera 2364.977Probability 0.000000

What to conclude?Is this enough evidence to have some conjecture that “something is going on”? How to investigate this further?

Or, are there other explanations? Maybe Dutch market is more oriented towards the international spot market and suggested prices translate spot market price into retail price?

We did advanced econometric model where we studied whether suggested price contains more information than spot market price to predict retail prices

Answer is: yes, quite a bit?

Maybe retailer’s input price is linked to suggested price and retailers simply react to their cost price?

Don’t have independent evidence on cost prices.

Concluding

Screens can be extremely useful

But have to be applied with great care

Often issues are not as simple as they seem at first sight if one

wants to eliminate alternative explanations

Often: no conclusive proof

good for detection purposes

But only if data are relatively easily available and it is clear what

are different testable implications of alternative hypotheses

Sometimes few alternatives available (like when used to calculate damages)