Lecture 12: Search and Price Dispersion

Lecture 12: Search and price dispersion

EC 105. Industrial Organization.

Matt Shum HSS, California Institute of Technology

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 1 / 31 Information costs

In many markets, consumers or ﬁrms may lack important information about each other Two types of information costs:

1 Search costs: Consumers are not fully aware of ﬁrms’ prices in the market gasoline, furniture 2 Veriﬁcation costs: Consumers are not fully aware of product quality: how reliable the product is automobiles, apartments, electronics Firms are not fully aware of consumers’ characteristics, actions insurance: are people driving safely, do they have unreported health risks? We focus on search costs in this lecture.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 2 / 31 mss # Hong & Shum; art. # 2; RAND Journal of Economics vol. 37(2) Why are prices for the same item so diﬀerent across stores? HONG AND SHUM / 259

FIGURE 1 RAW HISTOGRAMS OF ONLINE PRICES

A puzzle considering basic economic theory: review this. homogeneous products, so that only search frictions (arising from consumers’ imperfect informa- Considertion about stores’ the prices) benchmark and heterogeneity in search of perfect costs in the consumer competition population generate price dispersion in this market. EC 105. IndustrialAs a motivating Organization. example, (Matt we consider Shum HSS, 20 online California pricesLecture forInstitute the classic 12: of SearchTechnology) mathematical and price statistics dispersion 3 / 31 textbook Probability and Measure (Billingsley, 1992), retrieved by the MySimon (www. mysimon.com) and Pricescan (www.pricescan.com) search engines, on February 5, 2002. A histogram of these prices is given in the third panel of Figure 1.5 The long left tail of the histogram suggests that consumers may have an incentive to search, because the potential cost savings can be over $15 (the lowest price, $85.58, is over $15 less than the highest price of $100.87). On the other hand, the large spike in the histogram around $100 suggests that despite the low prices, consumers may not be searching very much, because otherwise firms would not find it optimal to “pile up” at a relatively high price. These arguments illustrate that the search models may imply conflicting interpretations of observed prices, if only the consumer or firm side is considered in isolation. For this reason, in this article we impose the optimality conditions for both consumer search and firm pricing in obtaining our estimates of search costs, thus rationalizing the observed prices as equilibrium outcomes for a given theoretical model.

5 These prices include shipping costs. © RAND 2006. Leaving the PC world

One important implicit assumption of PC paradigm is that consumers are aware of prices at all stores. This implies an infinitely elastic demand curve facing firms. (ie. if one firm raises prices slightly, he will lose all demand). Obviously, this assumption is not realistic. Here we consider what happens, if we relax just this assumption, but maintain other assumptions of PC paradigm: large #firms, perfect substitutes, etc.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 4 / 31 Search model

Each consumer demands one unit; Starts out at one store, incurs cost c > 0 to search at any other store. Consumer only knows prices at stores that she has been to, and buys from the canvassed store with the lowest price. “free recall” Utility u from purchasing product: demand function is

purchase if p u (1) don’t purchase≤ otherwise

What is equilibrium in this market?

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 5 / 31 Diamond paradox

Claim: a nonzero search cost c > 0 leads to equilibrium price equal to u (“monopoly price”) Assume that marginal cost=0, so that under PC, p = 0 n firms, with n large. Consumers equally distributed initially among all firms. Start out with all firms at PC outcome. What happens if one firm deviates, and charges some p1 such that 0 < p1 < c? Consumers at this store? Consumers at other stores? How will other stores respond? By iterating this reasoning ....

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 6 / 31 Now start at “monopoly outcome”, where all ﬁrms are charging u. What are consumers’ purchase rules? Do ﬁrms want to undercut? Given consumer behavior, what do they gain? Role for advertising? P. Diamond (1971), “A Theory of Price Adjustment”, Journal of Economic Theory

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 7 / 31 Remarks

Diamond result quite astounding, since it suggests PC result is “knife-edge” case. But still doesn’t explain price dispersion Assume consumers diﬀer in search costs Two types of consumers: “natives” are perfectly informed about prices, but “tourists” are not.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 8 / 31 Tourist-natives model

Tourists and natives, in proportions 1 α and α. L total consumers: αL natives, and (1 α)L tourists. − − Tourists buy one unit as long as p u, but natives always shop at the cheapest store. ≤ Each of n identical firms has U-shaped AC curve (1−α)L Each firm gets equal number of tourists n ; natives always go to cheapest store. c Consider world in which all firms start by setting p = minq AC(q). Note that deviant store always wants to price higher. Demand curve for a deviant firm is kinked (graph). Deviant firm sells exclusively to tourists.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 9 / 31 Deviant firm will always charge u. Only tourists shop at this store. If charge above u, no demand. If below u, then profits increase by charing u. First case: many informed consumers (α large) Number qu (1−α)L of tourists at each store so small that u < AC(qu). ≡ n In free-entry equilibrium, then, all firms charge pc , and produce the same quantity L/n. If enough informed consumers, competitive equilibrium can obtain (not surprising)

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 10 / 31 Second case: few informed consumers (α small) Assume enough tourists so that u > AC(qu). But now: hi-price firms making positive profits, while lo-price firms making (at most) zero profits. Not stable. In order to have equilibrium: ensure that given a set of high-price firms (charing u) and low-price firms (charging pc ), no individual firm wants to deviate. Free entry ensures this. Let β denote proportion of lo-price firms. Each high-price firm charges u and sells an amount

(1 α)L(1 β) (1 α)L qu = − − = − (2) n(1 β) n − Each low-price ﬁrm charges pc and sells

αL + (1 α)Lβ qc = − (3) nβ

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 11 / 31 In equilibrium, enough firms of each type enter such that each firm makes zero profits. Define quantities qa, qc such that (graph):

AC(qa) = u; AC(qc ) = pc .

(Quantities at which both hi- and lo-price ﬁrms make zero proﬁts.) With free entry, n and β must satisfy

(1 α)L αL + (1 α)Lβ qa = − ; qc = − (4) n nβ Solving the two equations for n and β yields

(1 α)L αqa n = − ; β = (5) qa (1 α)(qc qa) − − N.B: arbitrary which ﬁrms become high or low price. Doesn’t specify process whereby price dispersion develops. As α 0, then β 0 (Diamond result) → →

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 12 / 31 Nonsequential vs. sequential search Two search models:

Consider two search models:

1 Nonsequential search model: consumer commits to searching n stores before buying (from lowest-cost store). “Batch” search strategy. 2 Sequential search model: consumer decides after each search whether to buy at current store, or continue searching.

Question: from observed prices (as above), can we infer what consumers’ search costs in a market are? Consider data for the four textbooks.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 13 / 31 Nonsequential search model

Main assumptions: Inﬁnite number (“continuum”) of ﬁrms and consumers

Observed price distribution Fp is equilibrium mixed strategy on the part of ﬁrms, with bounds p, p¯. r: constant per-unit cost (wholesale cost), identical across ﬁrms Firms sell homogeneous products Each consumer buys one unit of the good

Consumer i incurs cost ci to search one store; drawn independently from search cost distribution Fc First store is “free”

q˜k : probability that consumer searches k stores before buying

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 14 / 31 Nonsequential search model Consumers in nonsequential model

Consumer with search cost c who searches n stores incurs total cost

c (n 1) + E[min(p1,..., pn)] ∗ − Z p¯ n−1 (6) =c (n 1) + p n(1 Fp(p)) fp(p)dp. ∗ − p · −

This is decreasing in c. Search strategies characterized by cutoﬀ-points, where consumer indiﬀerent between n and n + 1 must have cost

∆n = E[min(p1,..., pn)] E[min(p1,..., pn+1)]. −

and ∆1 > ∆2 > ∆3 > . ··· Similarly, deﬁneq ˜n = Fc (∆n−1) Fc (∆n) (fraction of consumers searching n stores). Graph. −

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 15 / 31 mss # Hong & Shum; art. # 2; RAND Journal of Economics vol. 37(2)

HONG AND SHUM / 261

FIGURE 2 IDENTIFICATION SCHEME FOR SEARCH-COST DISTRIBUTION IN Nonsequential search model NONSEQUENTIAL-SEARCH MODEL

!4 are the searchEC 105. costs Industrial Organization.of the indifferent(Matt Shum HSS, California consumers,Lecture Institute 12: of SearchTechnology) and where price dispersion a consumer with search costs16 / 31 !k is indifferent between obtaining k and k + 1 price quotes. Let Fˆ p denote the empirical distribution of the observed prices. First, we note that we can obtain estimates of these indifference points from the empirical price distribution Fˆ p, via the relation (2). Second, deﬁne q˜ 1 F (! ) : the proportion of consumers with one price quote; 1 ≡ − c 1 q˜ F (! ) F (! ) : the proportion of consumers with two price quotes; 2 ≡ c 1 − c 2 q˜ F (! ) F (! ) : the proportion of consumers with three price quotes. (3) 3 ≡ c 2 − c 3

We can estimate q˜1, q˜2,...by exploiting the firms’ equilibrium pricing conditions. To see this, note that a firm’s profits from following the mixed pricing strategy F ( ) are (see Burdett and p · Judd, 1983)

∞ k 1 "(p)=(p r) q˜ k(1 F (p)) − − ! k − p # "k=1 for all p [p, p]. The characterization of the equilibrium price distribution starts with the mixed- ∈ strategy condition that ﬁrms be indifferent between charging the monopoly price p (and selling only to people who never search but receive an initial free draw equal to p) and any other price p in the equilibrium support [p, p]:

∞ k 1 (p r)q˜ =(p r) q˜ k(1 F (p)) − . (4) − 1 − ! k − p # "k=1

The optimality equation (4) allows us to recover a nonparametric estimate of the search- cost distribution Fc from Fˆ p alone, as we now show. Let pˆ and pˆ denote the lowest and highest observed prices, respectively. For convenience, we index the n observed prices in ascending order, so that

pˆ = p1 p2 pn 1 pn = pˆ . ≤ ≤···≤ − ≤ Let K ( n 1) denote the maximum number of ﬁrms from which a consumer obtains price ≤ − quotes in this market. Given this condition, the indifference condition (equation (4)) for each of © RAND 2006. Nonsequential search model Firms in nonsequential model

Firm’s proﬁt from charging p is:

" ∞ # X k−1 Π(p) = (p r) q˜k k (1 Fp(p)) , p [p, p¯] − · · − ∀ ∈ k=1 Since ﬁrms use mixed streategy, they must be indiﬀerent btw all p:

" ∞ # X k−1 (¯p r)˜q1 = (p r) q˜k k (1 Fp(p)) , p [p, p¯) (7) − − · · − ∀ ∈ k=1

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 17 / 31 Nonsequential search model

Reconsider thismss data # Hong & Shum; art. # 2; RAND Journal of Economics vol. 37(2)

HONG AND SHUM / 259

FIGURE 1 RAW HISTOGRAMS OF ONLINE PRICES

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 18 / 31 homogeneous products, so that only search frictions (arising from consumers’ imperfect information about stores’ prices) and heterogeneity in search costs in the consumer population generate price dispersion in this market. As a motivating example, we consider 20 online prices for the classic mathematical statistics textbook Probability and Measure (Billingsley, 1992), retrieved by the MySimon (www. mysimon.com) and Pricescan (www.pricescan.com) search engines, on February 5, 2002. A histogram of these prices is given in the third panel of Figure 1.5 The long left tail of the histogram suggests that consumers may have an incentive to search, because the potential cost savings can be over $15 (the lowest price, $85.58, is over $15 less than the highest price of $100.87). On the other hand, the large spike in the histogram around $100 suggests that despite the low prices, consumers may not be searching very much, because otherwise firms would not find it optimal to “pile up” at a relatively high price. These arguments illustrate that the search models may imply conflicting interpretations of observed prices, if only the consumer or firm side is considered in isolation. For this reason, in this article we impose the optimality conditions for both consumer search and firm pricing in obtaining our estimates of search costs, thus rationalizing the observed prices as equilibrium outcomes for a given theoretical model.

5 These prices include shipping costs. © RAND 2006. Nonsequential search model Estimating search costs corresponding to graphs

We observe data Pn (p1,... pn). Sorted in increasing order. ≡ Take p = p1 andp ¯ = pn Given picture above, we want to estimate consumer cutpoints ∆i , i = 1, 2, 3, ... and also theq ˜k , probabilities that consumers search k stores. Consumer cutpoints ∆1, ∆2,... can be estimated by computing

∆n = E[min(p1,..., pn)] E[min(p1,..., pn+1)]. − using the observed prices Pn. Assume that consumers search at most K (< N 1) stores. Then can solve − forq ˜1,..., q˜K from

"K−1 # X k−1 (¯p r)˜q1 = (pi r) q˜k k (1 Fˆp(pi )) , pi , i = 1,..., n 1. − − · · − ∀ − k=1 ˆ 1 P where Fp = Freq(p p˜) = n i 1(pi p˜) is the cumulative distribution function of the observed≤ prices. ≤ n − 1 equations with K unknowns.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 19 / 31 mss # Hong & Shum; art. # 2; RAND Journal of Economics vol. 37(2)

HONG AND SHUM / 267

TABLE 1 Summary Statistics on Prices for Different Products

Standard Product n List Mean Deviation Median p p

Stokey-Lucas 19 60.50 66.60 5.64 64.98 59.75 86.80 Lazear 17 31.95 34.73 2.48 35.27 29.51 37.70 Billingsley 20 99.95 95.48 5.87 98.90 83.58 100.87 Dufﬁe 15 65.00 62.71 4.91 63.48 50.58 69.95

Note: Including shipping and handling costs. Price data for all products downloaded from Pricescan.com and MySimon.com: February 5, 2002. Summary price including S&H costs may not exceed the corresponding summary price without S&H costs, since we could not determine the shipping and handling charges from some of the websites.

the standard deviation of the prices with shipping costs was roughly the same magnitude as the prices without shipping costs. Moreover, the Spearman rank correlation statistics between the two sets of prices was around .90, indicating strong correlation. This evidence might cast some doubt on the possibility that online retailers are engaging in bait-and-switch strategies with regard to shipping costs. Table 2 contains estimates for the nonsequential-search model. The estimates of the q˜’s indicate that in most markets, about half of the consumers never search (more precisely, they shop at the store where they received their initial “free” price). For the Stokey-Lucas text, 52%(= 100 (1 .480)) of purchasers have search costs exceeding ! = $2.32, while over 60% of ∗ − 1 the Lazear book purchasers have search costs exceeding !1 = $1.31. As was the case with the example considered above, the substantial proportion of people who don’t search implies that we cannot identify the shape of theNonsequential distribution search for these model people. For example, we know nothing about the shape of the search-cost distribution above the 52nd quantile for the Stokey-Lucas book, and above the 65th quantile for the Lazear book. NonsequentialTable 3 presents the maximum model: likelihood results estimates for the sequential-search model. Com- paring these results to those obtained from the nonsequential-search models, we see that the sequential-search model predicts higher magnitudes for search costs: as an example, the median

TABLE 2 Search-Cost Distribution Estimates for Nonsequential-Search Model

Selling MEL a b c Product K M q˜1 q˜2 q˜3 Cost r Value

Parameter estimates and standard errors: nonsequential-search model Stokey-Lucas 3 5 .480 (.170) .288 (.433) 49.52 (12.45) 102.62 Lazear 4 5 .364 (.926) .351 (.660) .135 (.692) 27.76 (8.50) 84.70 Billingsley 3 5 .633 (.944) .309 (.310) 69.73 (68.12) 199.70 Dufﬁe 3 5 .627 (1.248) .314 (.195) 35.48 (96.30) 109.13

Search-cost distribution estimates

!1 Fc(!1) !2 Fc(!2) !3 Fc(!3) Stokey-Lucas 2.32 .520 .68 .232 Lazear 1.31 .636 .83 .285 .57 .150 Billingsley 2.90 .367 2.00 .058 Dufﬁe 2.41 .373 1.42 .059

a Number of quantiles of search cost Fc that are estimated (see equation (5)). In practice, we set K and M to the largest possible values for which the parameter estimates converge. All combinations of larger K and/or larger M resulted in estimates that either did not converge or did not move from their starting values (suggesting that the parameters were badly identiﬁed). b Number of moment conditions used in the empirical likelihood estimation procedure (see equation (17)). K 1 c − For each product, only estimates for q˜1,...,q˜K 1 are reported; q˜K =1 q˜k . − − k=1 d Indifferent points ! computed as Ep Ep (the expected price difference from having k versus k + 1 price quotes), k (1:k) − (1:k+1) ! using the empirical price distribution. Including shipping and handling charges.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 20 / 31 Sequential search model Sequential model

Consumer decides after each search whether to accept lowest price to date, or continue searching. Optimal “reservation price” policy: accept ﬁrst price which falls below some optimally chosen reservation price. NB: optimal strategy features no recall.

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 21 / 31 Sequential search model Consumers in sequential model

Heterogeneity in search costs leads to heterogeneity in reservation prices ∗ For consumer with search cost ci , let z (ci ) denote price z which satisﬁes the following indiﬀerence condition Z z Z z ci = (z p)f (p)dp = F (p)dp. 0 − 0 Now, for consumer i, her reservation price is:

∗ ∗ pi = min(z (ci ), p¯).

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 22 / 31 Sequential search model Results: sequential search model mss # Hong & Shum; art. # 2; RAND Journal of Economics vol. 37(2)

268 / THE RAND JOURNAL OF ECONOMICS

TABLE 3 Estimates of Sequential-Search Model

Mediana Selling Log-L 1 Product δ δ Search Cost Cost r αb F− (1 α; θ) Value 1 2 c −

Stokey-Lucas .46 (.02) 1.55 (.03) 29.40 (1.45) 22.90 (1.31) .58 19.19 31.13 Lazear .40 (.01) 1.15 (.01) 16.37 (1.00) 11.31 (.79) .69 4.56 34.35 Billingsley .25 (.01) 2.01 (.04) 9.22 (.94) 65.37 (.83) .51 8.43 23.73 Dufﬁe .21 (.02) 4.57 (.29) 10.57 (2.01) 28.24 (1.63) .54 7.00 18.93

Note: Including shipping and handling charges. Standard errors in parentheses. δ1 and δ2 are parameters of the gamma distribution; see equation (13). a As implied by estimates of the parameters of the gamma search-cost distribution. b Proportion of consumers with reservation price equal to p, implied by estimate of r (see equation (11)).

searchGenerally, cost for the Stokey-Lucas nonsequential text consistent (batch) with thesearch nonsequential-search model generates model (as smallerreported (more in the bottom panel of Table 2) is roughly $2.32, whereas the corresponding number for the sequential-searchreasonable) model search (as reported cost in estimates Table 3) is $29.40, over 10 times higher. Moreover, the estimates ofBatchr, the selling search costs, is are feature uniformly of lower online across search the four engines? books than the corresponding estimates in the nonsequential-search model.19 At first glance, the larger search-cost magnitudes for the sequential-search model might lead one to support the nonsequential-search model as a better descriptor of search behavior in the markets we consider. As pointed out by Morgan and Manning (1985), nonsequential- search strategies may be optimal when consumers face nonzero fixed costs of initiating a search, regardless of how many prices one obtains. Such a situation may describe the online market, since EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 23 / 31 there may be nonzero search costs associated with, say, seeking out a computer, logging on, and so forth. Moreover, due to search engines, price quotes might tend to be obtained in groups, which may fit better with the nonsequential-search assumption. However, as we remarked earlier, the large median search-cost estimates for the sequential model are due in part to parametric extrapolation. In the fourth column of Table 3, we report the implied estimates of α, the proportion of consumers with reservation price equal to p and, hence, the proportion of consumers who never search.20 α exceeds .50 for all four products, suggesting that the high median search costs estimated in this specification are due in part to extrapolation based on the gamma functional form for the search-cost distribution. To see if this is true, we also 1 computed, in the fifth column, implied estimates of F − (1 α), which are the search costs for c − the consumer who has a reservation price just equal to p. We see that for three out of the four 1 cases (excepting the Lucas-Stokey book), the value of F − (1 α) is comparable in magnitude c − to the search costs from the nonsequential model (and reported in the bottom panel of Table 2).21

! Speciﬁcation checks for the sequential-search model. On the other hand, we might worry that the sequential-search model may be biased toward large search-cost estimates, due to the necessary condition (remarked above) that the slope of the search-cost density f ( ) be negative c ··· in order for the price density (9) to constitute a valid equilibrium price density. For the gamma density function in (13), we found that when δ 1 (which was required for the density to be 1 ≤ strictly decreasing), the hazard rate parameter δ2 must be increased so that the tail of the density does not die off too quickly, which we need in order to ﬁt the price data (which do not have thin

19 As remarked above, we conﬁrmed that at the reported estimates, the implied search-cost distribution had a strictly decreasing density. This was true for all four books. Furthermore, to verify the robustness of these estimates, we also reestimated the sequential-search models using a variety of starting values, but the reported results were very robust. 20 α in the sequential model corresponds to q˜1 in the nonsequential model. 21 For the Lucas-Stokey text, the high search costs implied by the sequential-search model appear to be driven by the outlier price of $86.80 (more than $20 above the list price) charged by opengroup.com. We have conﬁrmed that this price was not a temporary oversight: on February 7, 2004, the price on this site was still $83.70 (without shipping costs). However, we note again that our modelling approach assumes that each price is “real,” in the sense that it generates positive sales in equilibrium. This assumption may not apply to this website. © RAND 2006. Sequential search model Test: sequential or non-sequential?

Look for evidence from online book markets. Do consumer recall? (Return to earlier searched stores) Nonseq search: recall is possible. Seq search: optimal behhavior has no recall Hortacsu, de los Santos, Wildenbeest paper

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 24 / 31 Sequential search model

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 25 / 31 Sequential search model

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 26 / 31 Sequential search model

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 27 / 31 Sequential search model When do people search? Lewis/Marvel paper

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 28 / 31 Sequential search model When do people search? Lewis/Marvel paper

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 29 / 31 Sequential search model When do people search? Lewis/Marvel paper

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 30 / 31 Sequential search model When do people search?

Pretty convincing evidence that people search when prices have been rising. Which contributes to greater price dispersion. (Also to “asymmetric price response”?)

EC 105. Industrial Organization. (Matt Shum HSS, CaliforniaLecture Institute 12: of SearchTechnology) and price dispersion 31 / 31