How to predict the future of shopping - Ulrich Kerzel @ PAPIs Connect
ย
trial lecture_aas
1. Consumer search and
price dispersion
Trial lecture during Ph.D. defense of
โEssays in Industrial Organization and Search Theory โ
by
รyvind N. Aas
2. Outline of talk
1. Motivation and questions
2. Theory
3. Empirics
4. Policy
Feb 5, 2016 Trial lecture - รyvind N. Aas 2
5. 43 โSpecific brandsโ
Trial lecture - รyvind N. Aas 5
Mean
(product)
Min
(product)
Max
(product)
Standard deviation 2,19 0,456 6,09 NOK
Coefficient of
variation 0,09 0,03 0,3199
Max - min 5,26 1 13,5 NOK
Histogram of savings by moving
from max to min
0,319
0,03 = 10,6
6. Motivating questions
1. How can identical products have different
prices?
2. What determines the prices and why are
the different dispersion across products?
3. Policy implications
โข M&A regulation?
โข Platform design?
Feb 5, 2016 Trial lecture - รyvind N. Aas 6
1. Inefficient
2. Inequality
๐๐๐๐๐
๐๐ข๐๐๐ก๐๐ก๐ฆ
Fig. Price 0.5l Coca Cola
๐
๐ท
7. Setup
โข How agents meet: random or direct matching?
โข How are contracts determined: bargaining or
price posting?
โข Sequential search (e.g., buying a house) vs.
nonsequential (e.g. job applications)
Feb 5, 2016 Trial lecture - รyvind N. Aas 7
$100,-
8. Search again<๐ ๐
Stop searchingโฅ๐ ๐
Basic setup
โข Ex. consumer searching for the
best quality of a product
โข No centralized market
โข Optimal stopping rule
Feb 5, 2016 Trial lecture - รyvind N. Aas 8
๐ข๐ฝ ๐ ๐ข๐ฝโฒ ๐ > 0
๐ข๐ฝ ๐
1 โ ๐ฝ
= ๐ข๐ 0 + ๐ฝ
0
๐ ๐ข๐ฝ ๐
1 โ ๐ฝ
๐๐น ๐
+๐ฝ
๐
โ ๐ข๐ฝ ๐
1 โ ๐ฝ
๐๐น(๐)
๐ข๐ฝ ๐ โ ๐ข ๐ 0 =
๐ฝ
1 โ ๐ฝ ๐
โ
๐ข๐ฝ ๐ โ ๐ข๐ฝ ๐ ๐๐น(๐)
Marginal cost
(less interesting conversation)
Marginal benefit
(more interesting conversation)
Michelle in a bar
Different men in the economy
9. Theory
โข Suppose centralized market for
lemon, with cost $1. Value = $10.
โข Suppose ๐ > 0 cost of finding the
price in a store
โข Diamond paradox: ๐ > 0 drastically
changes the equilibrium
Feb 5, 2016 Trial lecture - รyvind N. Aas 9
Equilibrium price = $1
Price = $1 Price = $2
Price = $1+ ๐ Price = ($1+ ๐) + ๐
Equilibrium price = $10
Price = $1 Price = $1
10. Theory
โข Diamond (1971, JET), search costs lead to uniform price
โข How can price dispersion come about as an equilibrium
result?
โข Reinganum (1979,JPE), Varian (1980,AER), Carlson and McAfee
(1983,JPE), Stahl (1989,AER), Stahl (1996,IJIO)
โข Burdett and Judd (1983,ECTA) What are crucial conditions
to generate price dispersion in an economy with identical
consumers and firms?
โข Answer: Ex post heterogeneity in consumer information
Feb 5, 2016 Trial lecture - รyvind N. Aas 10
11. Burdett and Judd (1983,ECTA), noisy sequential search
11
(MC = MB)
๐ = min ๐ ๐
, ๐ง
๐ =
0
๐ง
๐ง โ ๐ ๐๐น๐
1. Competitive price? Not an
equilibrium, incentive to deviate
2. Any price above competitive? Not an
equilibrium, since incentives to
undercut and sell to consumers with
two price quotes
3. Discontinuity in demand schedule, so
a pure strategy equilibrium cannot
exist
๐ โ ๐
๐=1
โ
๐๐ ๐[1 โ ๐น๐ ๐ ] ๐โ1
= ๐1 ๐ โ ๐
๐, ๐๐น(๐)
If ๐1 = 1, all observe only one price (Diamond result)
If ๐1 =0, all observe at least two prices (competitive result)
If ๐1 โ 0,1 , probability observe at least two prices
Pay c>0 to receive k
prices with prob. ๐ ๐
Ex post heterogeneity in consumer information
Prediction (Janssen & Moraga-Gonzalez,2004):
More companies in the economy will lead to higher prices
$10
$4
12. Empirics:
Estimating search costs from prices
โข Search costs can alter equilibrium predictions and policy
implications. So how empirically significant are they?
โข Given prices, can we say something about the search cost
among consumers?
โข Hong and Shum (2006) methodology to estimate search
costs
โข Main takeaway: prices are sometimes sufficient to estimate search
costs, because of equilibrium restrictions (optimality conditions
from consumers and firms) which rationalize observed prices.
13. Main idea
Probability of minimum
price of
No. of
prices
sampled
$2 $3 Expected
min price
Marginal
Expected
gains ฮ
ฮ โ ๐
(0,25)
2 pepl.
ฮ โ ๐
(0,06)
4 pepl.
ฮ โ ๐
(0,03)
4 pepl.
1 0,5 0,5 2,5
2 0,75 (=1-0,25) 0,25 2,25 0,25 0,0000 0,1875 0,2422
3 0,8750 0,1250 2,1250 0,1250 -0,1250 0,0625 0,1172
4 0,9375 0,0625 2,0625 0,0625 -0,1875 0,0025 0,0047
5 0,9688 0,0313 2,0313 0,0313 -0,2188 -0,0313 0,0013
Feb 5, 2016 Trial lecture - รyvind N. Aas 13
๐๐ ๐ ๐๐๐ = 3 = 1 โ 1
2
2
= 0,25
Constant search costs: c, first sample free,
Choose obs, so that EMB = MC
0%
20%
40%
60%
80%
100%
0 0,05 0,1 0,15 0,2 0,25
Estimatedsearch-costcdf
Seach costs
Search-cost distribution
14. Methodology, Hung and Shum (2006)
โข First, from the empirical price distribution, compute the cutoffs ฮ๐
โข Second, define
โข ๐1 = 1 โ ๐น๐(ฮ1) : proportion of consumers with one price quote
โข ๐2 = ๐น๐ ฮ1 โ ๐น๐(ฮ2) : proportions consumers with two price quotes
โข ๐3 = ๐น๐ ฮ2 โ ๐น๐(ฮ3)
โข ๐ ๐พ = 1 โ ๐=1
๐พโ1
๐ ๐ : proportions of consumers with K price quotes
โข Use indifference conditions from Burdett and Judd (1983), to estimate
๐1, ๐2, โฆ, ๐ ๐พโ1
ฮ ๐ = ๐ โ ๐
๐=1
โ
๐๐ ๐[1 โ ๐น๐ ๐ ] ๐โ1 = ๐1 ๐ โ ๐ = ฮ (๐)
โข Sort prices, ๐ = ๐1 โค ๐2 โค ๐ ๐โ1 โค ๐ ๐ = ๐
โข Let K โค ๐ โ 1 denote max number of prices sampled
Feb 5, 2016 Trial lecture - รyvind N. Aas 14
17. Discussion
โข Caveats:
โข Obfuscation (not all prices are ยซtrueยป price)
โข Search technology
โข Production costs
โข ยซDifferentiated productsยป across firms (e.g., return, shipping, handling,
etc.).
โข Hortacsu and Syverson (2004) importance of search frictions and product differentiation (e.g., mutual index funds + hedge fund)
โข Distribution of search costs have shifted over time, average search costs have gone down, upper percentile tends to increase, more novice
investors entering
โข Moraga-Gonzalez and Wildenbeest (2008), oligopoly with ML, Monte Carlo for robustness and counterfactual with tax increase
โข 15% sales tax reduces search intensity so consumers carry more than propotional increase, firms profit rise
โข Policy affects both firms prices and consumers search intensity
โข Wildenbeest (2011), product differentiation (61%) and search costs (41%) among retail chains in the UK.
Feb 5, 2016 Trial lecture - รyvind N. Aas 17
18. Matvareportalen
โข Competition Authority: the portal facilitates price cooperation (collusion)
(bad for consumers)
โข Consumer Protection Agency (CPA): the portal promotes competition
(good for consumers)
Feb 5, 2016 Trial lecture - รyvind N. Aas 18
โMatvareportalen kan bli forbudtโ
โ Bergens Tidene Oct, 27 2015
Vs.
19. โข โThe portal facilitates collusionโ (illegal price cooperation)
๐ก=0
โ
๐ฟ ๐กฮ 1 ๐1๐ก, ๐2๐ก
โข ๐ฟ < 1 discount factor, close to 1 more patience
โข Two firms, choose prices simultaneously each period
โข ๐ = ๐ is one equilibrium
โข ๐ ๐ as the monopoly price, symmetric trigger strategies, FE,PO
Matvareportalen โ Competition authority
Feb 5, 2016 Trial lecture - รyvind N. Aas 19
๐๐๐ก =
๐ ๐ ๐๐ ๐๐๐กโ1 = ๐ ๐
๐ ๐๐กโ๐๐๐ค๐๐ ๐
๐ = ๐ ๐ in t=0
Collusion alternative
20. Matvareportalen โ Competition authority
โข Information lags, it takes two periods before deviation is detected
โข Information lags softens punishment, strenghtens incentives to deviate.
โข More transparency, easier to sustain collusion, higher prices
โข Worse for consumers, better for firms.
Feb 5, 2016 Trial lecture - รyvind N. Aas 20
ฮ ๐
2
1 + ๐ฟ + ๐ฟ2 + โฏ โฅ ฮ ๐
ฮ ๐
2
1
1 โ ๐ฟ
โฅ ฮ ๐
โ ๐ฟโ โฅ
1
2
ฮ ๐
2
1 + ๐ฟ + ๐ฟ2 + โฏ โฅ ฮ ๐ 1 + ๐ฟ
ฮ ๐
2
1
1 โ ๐ฟ
โฅ ฮ ๐
1 + ๐ฟ
โ ๐ฟโ โฅ
1
2
>
1
2
(ColludeโฅDeviate)
21. Matvareportalen โ CPA
โข โThe portal promotes competitionโ
โข Suppose ๐ fraction of informed consumers
โข More informed consumers, decreases the lower bound, ๐ =
1โ๐
1+๐
โข More competition for informed consumers, expected price falls
โข More transparency, lower prices
โข Better for consumers, worse for firms.
Feb 5, 2016 Trial lecture - รyvind N. Aas 21
๐
1 โ ๐
2
+ ๐ 1 โ ๐น(๐) =
1 โ ๐
2
22. โข Petrikaite (2016) Collusion with costly consumer search
โข Q: how does stability of collusion relate to market transparency from the
viewpoint of the consumers
โข Sequential search, Stahl (1989) homogeneous prod., ๐ข = ๐ โ ๐
โข No price>๐ ๐, so indifference condition for competitive eq.
โข Lower bound, and CDF
Matvareportalen โ collusion and search
Feb 5, 2016 Trial lecture - รyvind N. Aas 22
๐
๐ ๐
๐ ๐ โ ๐ ๐๐บ ๐ = ๐ (MB = MC)
๐ ๐
โ ๐ , ๐
๐
โ(๐)
๐
As search cost increase, consumers
becomes less picky
๐
1 โ ๐
2
+ ๐ 1 โ ๐น(๐) =
๐๐ 1 โ ๐
2
= ฮ โ
๐ =
๐ ๐ 1 โ ๐
1 + ๐
๐น ๐ = 1 โ
๐ ๐ โ ๐ 1 โ ๐
2๐๐
(Competitive profits)
24. โข Suppose transparency is related to fraction of shoppers
(i.e., ๐)
โข More transparent, higher ๐
โข More shoppers implies less demand from uninformed
โข More shoppers implies lower bound goes down, so reservation
price goes down
โข Competitive profit goes down, but rate is lower with more
shoppers
Matvareportalen โ collusion and search
Feb 5, 2016 Trial lecture - รyvind N. Aas 24
โ โ โ
For more shoppers, inc. deviating profit > red. comp. profit ๐นโ
โ with ๐ More transparent, less
sustainable collusion
๐นโ โ with ๐ More transparent, more
sustainable collusion
For fewer shoppers, inc. deviating profit < red. comp. profit
1
1
๐นโ โ with ๐
2
0๐นโ
โ with ๐
๐ฃ
2
1
1 โ ๐ฟ
โฅ
๐ฃ
2
1 + ๐ +๐ฟ
๐ ๐ 1 โ ๐
2
1
1 โ ๐ฟ