The Upward Pricing Pressure Test and Sensitivity of the Diversion - - PowerPoint PPT Presentation

The Upward Pricing Pressure Test and Sensitivity of the Diversion Ratio

Lydia Cheung

Auckland University of Technology Presented at the 2nd ATE Symposium

16 December, 2014

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 1 / 21

Introduction

Contents

1

Introduction

2

Sensitivity of Elasticities and Diversion Ratios

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 2 / 21

Introduction

Merger Policy “in a Hurry”

Antitrust agencies receive many merger proposals (e.g. FTC & DoJ in the U.S. receive 1000+ merger proposals each year) Need to block mergers with large anti-competitive harm, e.g. big price increases (“unilateral effects”) 2 opposing price effects in even the simplest, static merger: loss in competition (P↑) vs. cost savings (P↓) Ideal: Specify model of industry / market, estimate it on computer, then simulate the merger Reality: Data and time constrained (30 days); may need shortcuts

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 3 / 21

Introduction

Old Shortcut: Market Share Analysis

1

Define market boundary

2

Find all firms that are “in”, and their market shares

3

Calculate increase in market concentration due to merger Problem: Market boundary definition is subjective, since most products are differentiated E.g. Merger of Whole Foods & Wild Oats (2007) Is the market boundary “premium organic supermarkets” or “all supermarkets”?

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 4 / 21

Introduction

Debate in court (2008): Are these substitutable? Should I include Walmart in the relevant merger market? vs. The UPP doesn’t require market definition anymore!

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 5 / 21

Introduction

New Shortcut: Upward Pricing Pressure (UPP) Test

UPP finds how much a firm’s pricing incentive changes with merger by approximating (via first order condition): Price increase with profits diverted to merging partner Price decrease with cost reduction Why is this new shortcut better? No need to define market boundary Allows product differentiation to affect merger outcome Big influence: U.S. and U.K. revised Horizontal Merger Guidelines in 2010, both incorporating UPP; Australia and E.U. are following suit

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 6 / 21

Introduction

What is the UPP?

Single product firms 1, 2 are merging. UPP for firm 1: UPP1 = D12(p2 − c2) − ec1 Demand-side input: diversion ratio D12 Cost-side input: ec1 = cost saving Analogous for firm 2: UPP2 = D21(p1 − c1) − ec2 Natural extension to multi-product firms Appeals: Small data requirement; easy to calculate

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 7 / 21

Introduction

Where to Get Diversion Ratio D12?

Farrell & Shapiro suggested “company internal documents” or “customer surveys” Some practitioners say this is picking a number out of the blue More data-driven way to get D12: simple demand system estimation

1

D12 =

• ∂q2

∂q1

• = ∂q2

∂p1 ·

• ∂q1

∂p1

• −1

= ε21 |ε1| · q2 q1

2

D12 =

• ∂q2

∂q1

• · q1

q2 = ∂q2 ∂p1 p1 q2

• ∂q1

∂p1

• p1

q1 −1 = ε21 |ε1| But demand estimation requires market definition! I show that this “catch-22” is not a big deal empirically

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 8 / 21

Sensitivity of Elasticities and Diversion Ratios

Contents

1

Introduction

2

Sensitivity of Elasticities and Diversion Ratios

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 9 / 21

Sensitivity of Elasticities and Diversion Ratios

Experimenting with Market Definitions

Supermarket consumer goods are ideal for experiments: product proliferation blurs market boundaries Smallest product category has 25+ items; big categories have 150+ Today I use a smaller category: sugar substitutes, in 2006 data

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 10 / 21

Sensitivity of Elasticities and Diversion Ratios

Summary of Sugar Substitutes Category

1433 supermarkets with 10+ unique products (UPC’s) 150 unique products Average store carries 20 products (min 4; max 37) Mean price per product: US$3.27 Mean price per oz.: US$1.01 Ingredient Obs. Saccharin & Dextrose 20% Nutra sweet 18% Aspartame 18% Saccharin 16% Sucralose 15% Fructose 6% Glucose 3%

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 11 / 21

Sensitivity of Elasticities and Diversion Ratios

Demand Model: Simple Nested Logit

Random utility of consumer i from product j: uij = β0 − αpj + ξj + [ζiN + (1 − τ)ǫij]

• combined error term

ξj: product j fixed effect (Mimicking practitioner’s situation, I do not use rich product characteristic variables) Equation to estimate: ln(sj) − ln(s0) = β0 − αpj + ξj + τ ln(sj|N) “BLP instruments” for endogenous prices Demand system gives explicit functional forms for elasticities and diversion ratios

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 12 / 21

Sensitivity of Elasticities and Diversion Ratios

Changing Market Definition

I experiment on market definitions by: Dropping an entire sugar type, in all markets Dropping 1-8 largest competing items, one by one, in each market When set of products changes (while fixing M), red terms below will also change in the data: ln(sj) − ln(s0) = β0 − αpj + ξj + τ ln(sj|N) The set of observations in the dataset also changes with J Thus, the “BLP” instruments will change too So estimated elasticities and diversion ratios will change

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 13 / 21

Sensitivity of Elasticities and Diversion Ratios

Diversion Ratios Unconstrained by Model

Simple nested logit model gives explicit functional forms for elasticities, thus the diversion ratios have explicit forms too D12 = ε21

|ε1| = (1−τ)s1|G+τs1 1−[(1−τ)s1|G+τs1], which takes the form x 1−x:

So the simple model does not inherently constrain values of the diversion ratio

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 14 / 21

Sensitivity of Elasticities and Diversion Ratios

Keeping track of 2 biggest goods

Two most frequently occurring items: Sweet ‘n’ Low 4.5oz; packets in box (3.86% obs.) Equal 3.5oz; packets in box (3.75% obs.) Item Variable Mean Min Max Sweet ‘n’ Low 4.5oz. elasticity −1.27 −1.89 −0.91 diversion ratio 0.017 0.00028 0.058 Equal 3.5oz. elasticity −2.96 −4.14 −1.48 diversion ratio 0.0087 0.000039 0.035 I keep track of their elasticities and diversion ratios as market definition changes

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 15 / 21

Sensitivity of Elasticities and Diversion Ratios Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 16 / 21

Sensitivity of Elasticities and Diversion Ratios Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 17 / 21

Sensitivity of Elasticities and Diversion Ratios Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 18 / 21

Sensitivity of Elasticities and Diversion Ratios Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 19 / 21

Sensitivity of Elasticities and Diversion Ratios

Takeaway from Experiments

As market definition changes, elasticities of the two goods often move in same direction Thus the diversion ratio does not change much in magnitude As more items are dropped from the market, diversion ratios have a tendency to increase (though by very small magnitude) Although my experiments see a ∼50% increase in diversion ratio, it is still a small magnitude (10−2) relative to price (∼ $5)

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 20 / 21

Sensitivity of Elasticities and Diversion Ratios

Alternative Approach

Some economists have estimated the diversion ratio using exogenous changes in product set Conlon & Mortimer (2013), “An Experimental Approach to Merger Evaluation”: Exogenously removing snack foods from vending machines and see how customers switch Another example: when a hospital exogenously shuts down, researcher can track where patients switch to in the data

Lydia Cheung (AUT) Sensitivity of Diversion Ratio in UPP 21 / 21