GUY ARIE
OLEG BARANOV
BENN EIFERT
HECTOR PEREZ-SAIZ
BEN SKRAINKA
Bundling Software:An MPEC Approach to BLP
Extension of BLP to multi-product markets
Observation: a large share of word processors and spreadsheets are sold as part of a suite (or bundle).
Interpretation 1: word processors and spreadsheets are complementary products (in the usual sense).
Interpretation 2: people have positively correlated preferences for a variety of software applications.
The Problem
Goal: to estimate consumer preferences over observed and unobserved characteristics of products in a market.
Application: Gandal, Markovich and Riordan (2006), office software. Extend BLP (1995) to markets with bundling and product complementarities.
Idea: think of the product space as containing every possible combination of word processors and/or spreadsheets. Generates accounting problem.
Data: US market shares for Microsoft, Lotus and Novell spreadsheets, word processors and suites, 1992-1998.
The office software space in the 1990s
-three companies (Microsoft, Lotus/IBM, Novell/Corel)
-two types of individual products (spreadsheets, word processors) plus suites
-fifteen possible combinations a consumer could buy
-significant changes in prices and product availability over the 1990s
Structure of the model, I
Heterogeneous consumers with preferences over product
attributes
Heterogeneous consumers with preferences over product
attributes
Probabilistic demands for individual consumersProbabilistic demands
for individual consumers
Multidimensional quadrature formulas
Multidimensional quadrature formulas
Products and their characteristics
Products and their characteristics
“Market share” functions for all possible product combinations
“Market share” functions for all possible product combinations
Structure of the model, II
“Market share functions” for all possible product combinations “Market share functions” for all possible product combinations
Constraint: predicted shares = observed shares
Constraint: predicted shares = observed shares
Residuals (“unobserved product quality”)
Residuals (“unobserved product quality”)
InstrumentsInstruments
Aggregate market shares for individual products and bundles
Aggregate market shares for individual products and bundles
GMM objective functionGMM objective function
Our Approach
Main obstacles:
numerical instability, convergence problems, slow in MATLAB.
usual methods require inner loop, outer loop
Solutions:
Substitute multidimensional quadrature for Monte Carlo
MPEC/AMPL/KNITRO takes ~ five seconds.
Impose constraints instead of using nested loops.
Multi-starts to deal with tons of local minima (still a problem...)
The basics
exp( )ˆ , 1,...,
exp( )jt jt
jtkt ktk
ps d j k J
p
i
jt jt ii
kt kt iμ
X β Z μμ
X β Z μ
1,..., , 1,...,ijt jt jt ijtu p j J t T jt jt iX β Z μ
• Consumer i’s utility for each product j as a function of product characteristics and individual preferences:
• Aggregate market shares computed by integrating over distribution of preferences:
The basics
• For a given set of structural parameters, compute ξjt by implicit relation:
ˆˆ ( , ) 1,..., 1,...,jt jt jts s j J t T θ
• Using instruments Zjt , form GMM objective function:
ˆ ˆ ˆ ˆ ˆˆ ˆarg min ( ) ( )jt jtE E jt jtθ
θ Z θ Ω Z θ
Gaussian quadrature interlude…
Integration Technique
Integration technique…
Quadrature faster and more accurate…but still problem of many local minima
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x 10-20
0
0.5
1
1.5
2
2.5
3
3.5
4distribution of 50 best objective function values from 5000 starts
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x 10-19
0
5
10
15
20
25distribution of 100 best objective function values
Results plausible at best objective function value?
Factor Coefficient $ Equivalent
Price -0.034 -
Bundle 1.89 $90.01
Microsoft 5.00 $238.10
Lotus -1.84 -$87.62
Quality (7 to 10) -0.317 -$15.09
Rho -0.05 -
Sigma.WP 4.72 -
*Results from solution with lowest objective function value
…but some parameter estimates are unstable even among “good” solutions
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5Histogram for rho coeffic ient (1% of All Solutions )
rho coeffic ient
Fre
qu
en
cy
Price coefficients are stable among “good” solutions
-0.1 -0.09 -0.08 -0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.010
1
2
3
4
5
6
7
8
9Histogram for Price coeffic ient (1% of All Solutions )
Price coeffic ient
Fre
qu
en
cy
Trends in unobserved product quality
92 93 94 95 96 97 98-4
-3
-2
-1
0
1
2
3
4
5
6Unobserved means
Years
Va
lue
IBM SS
COREL WP
IBM S
COREL S
MS WP
MS SS
Summary
Solution much improved over MATLAB method in working paper.
Numerical stability is still a significant problem.
Model is probably not well-identified: need more diagnostics.
One thing is for sure: Microsoft fixed effect is huge!
Reaching out to a new demographic?