Does Merger Simulation Work? Lessons from a Case in the Swedish Analgesics Market
Jonas Björnerstedt Frank Verboven KKV and KU Leuven
November 2012
Topic of paper Ongoing debate in IO Are structural models useful to predict merger e¤ects? Or can we learn more from direct evidence on past mergers? See Angrist-Pischke versus Nevo-Whinston These questions extend well beyond mergers and beyond IO!
Start from a merger in the Swedish painkiller market During the investigation in March 2009, we developed a structural demand model to predict the price e¤ects. Two years later we updated the dataset to perform an ex post analysis of the merger.
Two purposes Ex post merger evaluation Evaluation of merger simulation as a tool in merger policy (including the performance of a new demand speci…cation).
Selected literature Merger simulation Hausman, Leonard, Zona (1994), Nevo (2000), Ivaldi and Verboven (2005)
–> We compare two nested logit speci…cations in integrated framework. Ex post merger evaluation Borenstein (1990), Hastings (2004), Ashenfelter and Hosken (2008)
Evaluation of merger simulation Peters (2006), Weinberg and Hosken (2009)
We compare the predictions as made during the investigation, with what happened afterwards. –> can test richer predictions because large size of the merger
The market and the merger
The market for OTC analgesics (painkillers) Brands are sold under three main active substances: paracetamol, ibuprofen and ASA Note: paracetamol = acetaminophen in U.S. Brands are sold under two main forms: tablets and …zzy tablets
The merger between GSK and AZT only companies selling paracetamol do not sell brands in the other segments
A central question is therefore whether the market is segmented according to the substance.
The market and the merger (2)
Market shares in 2008, by form and active substance Form Tablet Fizzy tablet Total
Paracetamol 36.1 6.0 42.1
Ibuprofen 29.0 29.0
ASA 2.6 26.3 28.9
Total 67.7 32.3 100
The market and the merger (3)
Market shares in 2008, by …rm/brand and active substance Firm AZT GSK McNeil Nycomed Meda (Ellem) Bayer Total
Brand Alvedon Reliv Panodil Ipren Treo Magnecyl Ibumetin Alindrin Bamyl Aspirin Alka-selzer Albyl
Paracet. 29.3 2.2 10.6
Ibupr.
ASA
Total 31.5 10.6 44.7
19.1 22.5 3.1 9.2 0.7
42.1
29.0
9.2 3.4 2.7 0.4 0.0 0.2 28.9
0.6 100
The merger investigation
Swedish competition authority gathered following evidence traditional review of the market, the …rms and planned reforms questionaire and merger simulation
Merger cleared in April 2009 consumers base decisions more on brands than on active substance deregulation of pharmacy monopoly will lead to more entry “merger simulation results imply the merger would not raise prices signi…cantly”
Predictions and ex post development Predicted price increase: +34.5% (no cost saving), +7.4% (25% marginal cost saving). Actual price evolution: +42% (in absolute terms), +35% (relative to competitors)
1.4
Price (in Krone) 1.6 1.8
2
Price evolution analgesics (April 2007-April 2011)
2007m1
2008m1
2009m1 Month Paracetamol Ibuprofen
Note: vertical line refers to the month of merger (April 2009)
2010m1 ASA
2011m1
The dataset
Panel consisting of monthly observations during 1995-2008 on average 67 products per month product=brand, form, packsize, dose, e.g. Alvedon tablet, 30 pices, 500 mg/piece.
Variables turnover, volumes and price marketing expenditures macro variables (GDP, #sick persons, ...)
The dataset
Summary statistics Variable revenue (rjt = pjt qjt ) price per tablet (pjt ) price per ddd price per normal dose number of tablets(qjt ) number of ddd number of normal doses marketing sickwomen sickmen GDPnom (in billions) popwomen (in thousands) popmen (in thousands)
Mean 1.24 1.06 6.02 1.61 1.11 .21 .77 564.1 822.9 524.5 621.6 4524.2 4437.4
St. Dev. 2.56 .46 2.21 .60 2.19 .43 1.57 1445.7 197.0 108.0 107.4 54.8 72.5
Min. .00 .27 1.74 .43 .00 .00 .00 0 391 254 443.2 4471.4 4366.1
Max. 22.95 2.55 15.50 3.88 16.61 3.07 11.08 13536 1204 763 859.7 4652.6 4603.7
Demand model
Two-level nested logit upper nest is form (tablet or …zzy tablet) lower nest is active substance (paracetamol, ASA, ibuprofen)
Two speci…cations unit demand model: linear price constant expenditure demand model: logarithmic price
Demand model (cont.)
Utility of consumer i for product j (in month t) uij = xj +
j
+ f (yi ; pj ) + "ij
where "ij follows “nested logit” extreme value distribution. Cond. individual demand under two speci…cations for f (yi ; pj ) Unit demand uij = xj +
j
+
(yi
pj ) + "ij ) dij = 1
Constant expenditures demand uij = xj +
j
+
( ln yi
ln pj ) + "ij ) dij =
yi pj
Demand model (cont.) Random utility maximization max fui 0 ; ui 1 ;
; uiJ g
yields “nested logit choice probability” for product j: sj =
exp(( j )=(1 exp(Ihg =(1
1 )) exp(Ihg =(1
2 )) exp(Ig )
1 ))
2 ))
exp(Ig =(1
exp(I )
Aggregate demand for product j is qj =
I X i =1
= sj dij
sj ( ; ) I
= sj ( ; ) pBj
(unit demand) (constant expenditures demand)
where I is potential number P of consumers, and B = Y is potential budget (Y = Ii =1 yi ).
Estimation Step 1. Invert choice probabilities as in Berry (1994) Unit demand ln(sj =s0 )
1
ln(sj jhg )
2
ln(shjg ) = xj
pj +
j
Constant expenditures demand ln(sj =s0 )
1
ln(sj jhg )
2
ln(shjg ) = xj
ln pj +
j
Step 2. Equate choice probabilities to observed market shares Unit demand qj sj = ; I
sj jhg = P
qj
j 2H hg
qj
Constant expenditures demand
;
P
j 2H
shjg = PHhg Phg h=1
qj
j 2H hg
qj
P pj qj pj qj j 2H pj qj sj = ; sj jhg = P ; shjg = PHhg Phg B j 2H hg pj qj j 2H hg pj qj h=1
Price elasticities Own-price elasticities are Unit demand
jj
=
1 1
1 1
1
1 1
1
2
2
sj jhg
1
2
sj jhg
1
sj jg
sj
pj ;
2
sj jg
sj
1:
2
Constant expenditures demand
jj
=
1
1 1
1
1
1 1
1
Similar expressions for cross-price elasticities
2
Oligopoly model
Multiproduct …rms set prices to maximize pro…ts X (pk ck ) qk (p) f (p) = k 2F f
First-order conditions yield p=c
F
(p)
1
where F is the ”ownership matrix“ and demand derivatives.
q(p): (p) is matrix of
The model serves two purposes Compute current marginal costs Compute post-merger equilibrium (due to change in
F
and c).
Extension of oligopoly model
F
Extend ownership matrix
to allow for partial collusion
Example: …rm 1 owns product 1 and 2, …rm 2 owns product 3 no collusion F
=
1 1 0
1 1 0
1 1
1 1
0 0 1
!
1
!
partial collusion ( 2 (0; 1)) F
=
Implementation
Both estimation and simulation carried out in Stata
Simulation: started with …xed point iteration
pt+1 = cpost
F post
1
(pt )
q(pt ):
Convergence not always obtained –> Newton method
Demand parameters and price elasticities Constant expenditures Unit demand Param St. Err Param St. Err price ( ) -.304 .097 -2.042 .147 .835 .019 .928 .012 subgroup ( 1 ) group ( 2 ) .667 .018 .792 .010 marketing 15.50 2.66 8.85 1.75 sickwomen .357 .123 -.699 .081 sickmen 1.145 .235 .809 .155 time trend .0013 .0005 .0007 .0002 yes yes yes yes monthly f.e. R2 0.983 0.972 Implied price elasticities (December 2008) Average Range Average Range "jj -2.68 -2.84; -1.91 -12.4 -24.1; -3.9 "jk , k 2 Hhg .16 .00; .93 1.5 .00; 8.3 "jk , k 2 Gg .04 .00; .29 .25 .00; 1.76 "jk , k 2 = Gg .01 .00; .06 .02 .00; .16 Preferred model is constant expenditures model.
Predicted price e¤ects
Results as obtained during the merger investigation
Bertrand partial collusion Bertrand partial collusion
price increase no cost # 25% cost # Constant expenditures +34.0% +4.7% +28.4% –0.1% Unit demand +12.9% +1.6% +16.1% +9.0%
Lerner .49 .76 .16 .54
Conclusions
Without e¢ ciencies, there is large predicted price increase, because GSK and AZT become monopoly in paracetamol segment.
This is especially so in the constant expenditures model and under multi-product Bertrand
Smaller e¤ects under 25% marginal cost savings.
Predicted price e¤ects: further results
Paracetamol Ibuprofen ASA AZT GSK Nycomed Meda (Ellem) McNeil Bayer
% Price e¤ects Market share e¤ects Bertrand Part. coord. Bertrand Part. coord. Predictions at the level of the active substance +34.1 +28.0 –7.1 –5.4 +0.7 +4.1 +3.7 +2.7 +0.8 +3.0 +3.3 +2.7 Predictions at the level of the …rm +21.3 +19.5 –3.4 -2.7 +59.8 +45.1 –3.7 -2.7 +0.6 +4.0 +1.3 +0.9 +0.1 +2.7 +0.6 +0.5 +1.7 +4.1 +5.1 +3.9 +0.1 +2.5 +0.1 +0.1
smaller …rm GSK raises price by more than larger …rm AZT market shares of all competitors increase
Ex post merger analysis Price evolution
1.4
Price (in Krone) 1.6 1.8
2
Price ev olution analgesics (April 2007-April 2011)
2007m1
2008m1
2009m1 Month Paracetamol Ibuprofen
Note: vertical line refers to the month of merger (April 2009)
2010m1 ASA
2011m1
Ex post merger analysis Sales evolution
20
Market share (in percent) 30 40
50
Market share ev olution analgesics (April 2007-April 2011)
2007m1
2008m1
2009m1 Month Paracetamol Ibuprofen
Note: vertical line refers to the month of merger (April 2009)
2010m1 ASA
2011m1
Ex post merger analysis: actual e¤ects Di¤erence-in-di¤erence regression at level of active substance Speci…cation e.g. Ashenfelter and Hosken ln pit =
Constant Ibuprofen ASA Paracetamol*merger Ibuprofen*merger ASA*merger R2
i
+
i PostMergert
Price Param. St. err. .303 .004 .171 .006 .208 .006 .351 .006 .001 .006 .100 .006 .969
+ "it ;
Market share Param. St. err. .468 .002 -.199 .003 -.204 .003 -.033 .003 .050 .003 -.016 .003 .986
price increase merging …rms segment: +42% (in absolute terms), +35% (relative to competitors)
Ex post merger analysis: actual e¤ects Di¤erence-in-di¤erence regression at level of …rm
Constant GSK Nycomed Meda McNeil Bayer AZT*merger GSK*merger Nycomed*merger Meda*merger McNeil*merger Bayer*merger R2
Price Param St. err. .304 .011 -.004 .016 .107 .016 -.121 .018 .229 .016 -.149 .016 .356 .016 .379 .016 .012 .016 .029 .018 .084 .016 .105 .016 .907
Market share Param St. err. .344 .003 -.221 .004 -.254 .004 -.316 .004 .052 .004 -.339 .004 -.056 .004 -.003 .004 .001 .004 .011 .004 -.027 .004 .005 .004 .990
Conclusions
Preferred model is constant expenditures nested logit This model predicts the average price increase of merging …rms well Results are more mixed on predicting price e¤ects of individual merging …rms: not equal for both …rms price e¤ects of outsiders: depends on partial collusion parameter market share e¤ects: no change in one segment