YT YouTube2Text

YouTube2Text → Transcripts → Lecture 7: Markets and Competition with Lauren Bergquist

Lecture 7: Markets and Competition with Lauren Bergquist — Transcript

by International Growth Centre · 18,132 words · 2,787 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00thank you
1:07foreign
1:32welcome thank you for joining us today
1:35uh my name is Dave Donaldson from MIT
1:37and along with David Atkin from MIT
1:40we've um been asked to put together on
1:44behalf of igc and Brad a sequence of PhD
1:47level lectures on the topic of firms and
1:50development and uh we're delighted you
1:52could join us here today I just want to
1:55advertise before we go any further that
1:56we um you know this is lecture number
1:59seven of eight and uh we have uh uh
2:02recordings slides reading lists for this
2:06lecture as well as all previous lectures
2:07and then the subsequent lecture taking
2:09place tomorrow all on the igc the
2:13website and there's a link to that in
2:15the chat below so today our lecture is
2:19uh going to be given by Lauren Bergquist
2:21from Yale uh Lauren needs no
2:24introduction she's written seminal
2:25papers on on the topic of today's
2:27lecture which is going to be sort of
2:29broadly markets and competition uh this
2:33um we think this relates to the lectures
2:36you've hopefully already seen but can
2:38again see on the website if you've
2:39missed any uh you know that sort of
2:41paint various pictures of um of
2:44misallocation of economic activity in
2:47developing countries uh in particular
2:50across productive activities and hence
2:52firms and how that could be a source of
2:55underdevelopment and today the theme is
2:57sort of uh you know uh competition and
3:00and lack thereof as a as one possible
3:03source of such misallocation
3:06um so uh we'll we'll start very shortly
3:09I just want to point out that um
3:10panelists uh welcome to um raise your
3:14hand and ask questions Lauren and ask
3:16those out loud when called upon Lauren
3:18will also pause try to do her best to
3:20pause occasionally to invite such
3:21questions and uh non-panelists that is
3:24everybody else uh feel free to just
3:26write your questions in the Q a box and
3:28um we'll uh do our best to try to find
3:32moments where we can intervene and ask
3:34those questions out loud to Lauren
3:36um anyway Lauren that's all I had to say
3:38please take it away and thank you so
3:40much for joining us Lauren great thanks
3:42so much for having me and thanks for
3:43that kind introduction Dave um it's
3:45really a pleasure to be a part of this
3:46course it looks like it's been a really
3:48fantastic one so far so thanks for
3:50thanks for including this
3:52um so let me give you sort of a high
3:54level agenda of where we'll be headed
3:56today
3:57um so we'll start by talking about uh
3:59just sort of understanding the landscape
4:01so you've heard a lot I think both in
4:02Dave and David's lectures about sort of
4:05um uh the overall uh landscape of
4:09um you know what firms involving
4:11countries looks like and you've heard
4:12sort of in your in your courses since
4:13then so I want to talk just overall
4:15about kind of the factors that might
4:17hinder and might support competition in
4:19low and middle income countries
4:22then we'll move kind of from you know
4:23the theoretical factors to the
4:25measurement side so how do we actually
4:26measure competition
4:28um I'll talk about some of the
4:29traditional approaches sort of in in the
4:32non-development literature and how that
4:34might be potentially
4:36um uh there might be potential
4:37challenges but also opportunities in
4:39applying that to the low and middle
4:40income country context uh and then we'll
4:43get into our experimental approaches
4:44including some work of my own with
4:46Michael dinnerstein uh in Kenya
4:49the third section will focus on
4:50implications of competition on a variety
4:53of outcomes including firm selection
4:54meaning sort of entry and exit by firms
4:58um prices quality and Contracting I'll
5:01spend the most time here in this Jensen
5:02and Miller piece uh and we'll if we have
5:04time we'll go over sort of briefly this
5:05other pieces I know that Rocco I think
5:07covered a portion of this this last
5:09piece on Contracting in the most recent
5:11lecture
5:12and then of course I'll conclude by
5:13talking about sort of opportunities for
5:15future research uh in this space
5:18so I want to start by kind of again
5:19understanding sort of the light of the
5:21land and talk about factors that might
5:23either limit or support competition uh
5:26in low and middle income countries let
5:28me start with a question to the
5:30panelists um and you're welcome in the Q
5:32a to also add responses if you'd like
5:34you know thinking back to again that
5:36kind of lecture one description of what
5:38it looks like for the market environment
5:40for firms uh in low and middle income
5:42countries what are some of the factors
5:44that you think might uh potentially
5:46hinder limit competition uh in this
5:48environment
5:58should we just unmute ourselves and talk
6:00I actually I actually can't see everyone
6:01at once so please do feel free to just
6:03uh just jump in if you can okay so sorry
6:06if someone raised their hands uh anyways
6:09I don't know I would think that maybe
6:10these markets are kind of characterized
6:12by a few very large firms potentially
6:16kind of state-owned Enterprises
6:19this kind of thing crowding out let's
6:21say smaller firms
6:25yeah so you could see there being very
6:26very few firms and then there's a second
6:28question of why are there very few firms
6:30right so maybe they're sort of
6:31regulatory barriers by the states in
6:33terms of which firms can enter
6:34um that might crowd out uh other
6:36potential entrants and fewer firms might
6:38need less competition absolutely
6:40how about some other other reasons
6:42access to capital for startups
6:45yeah access to Capital right so there's
6:47potential entrance you might want to
6:48enter into the into the industry but
6:51can't because of all the kind of typical
6:52you know credit constraints that we
6:54think about in developing countries I
6:55might limit the number of entrants into
6:56a market and thereby limit competition
6:58great
7:00yeah perhaps on the other end of the
7:02spectrum of what Peter was saying uh
7:04small firms but that are kind of parsing
7:07up the market into very specialized uh
7:10Goods maybe we at first look at them as
7:12they're all fruit vendors but they're
7:14all kind of specializing and I've got
7:15the tomatoes cornered and you've got the
7:18apples and
7:19stuff like that totally yeah Define the
7:21Market's going to be a really critical
7:22component of this so it might look like
7:25a big Market but when you think about
7:26the actual relevant markets
7:28um either by product or by geography it
7:31might actually be much smaller there
7:32might be much less competition in that
7:35area
7:36any other ideas
7:39information available to to customers on
7:42finding Alternatives yep absolutely so
7:45search costs information frictions can
7:46limit again sort of the size of the
7:48market right so you might know about
7:49your local supplier but you might not
7:51know about others and that will come up
7:52a lot in this Jensen and Miller piece so
7:54these are some great explanations let me
7:56kind of go through my own list which I
7:57think covers a lot of what you all
7:58discussed and is by no means it's
8:00comprehensive but you know the first is
8:02this idea of sort of a smaller Market
8:04size right so this has to do with the
8:06variable profits that are available to
8:07support firms right so you know if
8:09you're in an environment with a small
8:11Market size it's harder to support a
8:13large number of firms in the presence of
8:15their being fixed cost and here you know
8:17I think you already got it some of these
8:18smaller Market size can mean a couple
8:20things right of course it's going to
8:21mean just population and number of
8:23consumers it also can mean you know the
8:25ability of consumers to purchase so so
8:27you know poorer consumers
8:30um and it can mean to sort of the the
8:31segmented kind of markets again dude we
8:33talked about kind of information
8:34frictions also physical trade costs in
8:37the line right so this all sort of means
8:39again lower lower variable profits which
8:41means fewer firms can be supported in
8:43the presence of fixed cost which can
8:45mean lower competition
8:46there also could be greater barriers to
8:48entry so you mentioned some of these as
8:49well
8:50um you know the first of course is kind
8:52of legal or regulatory barriers to entry
8:54right so maybe maybe you know the state
8:56has regulation over who can open a
8:58business in certain industries
9:00um you know the World Bank has a great
9:02sort of doing business indicators where
9:03you can look and see buy country the
9:05particular barriers to entry
9:07there also could be sort of you know
9:09outside of the legal realm there could
9:11be essentially illicit
9:12um barriers to entry so think about
9:14corruption and bribe so you know maybe
9:15it's not just that you have to fill out
9:17a very extensive paperwork uh to become
9:20registered but you also have to on top
9:22of that pay a bribe to the official
9:24right you could think of that like
9:26um a sort of second barrier to entering
9:28I will say there's a little bit of sort
9:30of
9:31um uh I think sort of open question as
9:32to whether competition is an additional
9:35barrier to entry on top of the red tape
9:37or whether it actually is acting a
9:39little bit like a substitute to actually
9:41say well if you can pay a bribe instead
9:43of filling out the paperwork that might
9:44actually lower kind of the the barriers
9:47in practice
9:48so some debate there
9:50um but then of course you know I think
9:51this was uh mentioned in the in the
9:53answers here which I think is a really
9:54big one it's just these kind of de facto
9:56barriers to entry so credit constraints
9:58right there might be entrepreneurs who
9:59have a great idea could open a very
10:01profitable business to compete with
10:03incumbents but don't have access to
10:05working capital okay so those who are
10:06able to enter Then are the folks who
10:08already have access
10:10um maybe those who are wealthier to
10:12begin with and that's a smaller pool so
10:13yeah if you were if you were entrants
10:17okay so these are sort of the factors
10:18that determine you know the number of
10:20firms
10:21um smaller variable profits higher
10:23barriers to entry there's also a matter
10:25of how those firms interact right so I
10:27think you talked last week about
10:29um about contract and relational
10:30Contracting you know in the presence of
10:34other market failures that we think are
10:35sort of specific to or are common in uh
10:38low and middle income countries like for
10:40example lack of kind of formal contract
10:43enforcement
10:44relational Contracting as Rocco talked
10:46about can be sort of a substitute for
10:48formal contract enforcement but through
10:51the flip side of that as it means you're
10:52in sort of a captive relationship right
10:54between a firm and its consumers or in
10:56the case of monopsony power but you know
10:57firm and and producers
11:00um and that can limit competition your
11:01ability to kind of go outside that
11:03relationship to find kind of competing
11:05offers okay so these things can be
11:07deeply sort of um uh entwined okay so
11:10these are all factors that might make
11:11you think the degree of competition
11:12among firms could be quite limited
11:15um in the loan little income country
11:16context on the flip side there's reasons
11:19to think competition could be quite High
11:21um so what are some of the features of
11:23that environment that you think might
11:24support a competition or make you think
11:26actually competition could be quite high
11:28in this environment
11:37the presence of many many small firms
11:41so there's many many small firms and so
11:42so you know why would that be
11:48lack of differentiation and return to
11:51scale convexities yeah so there could be
11:54just kind of structural things to do
11:56with the industry sort of specialization
11:58that you see common in the moment in
11:59country so as you said there could be
12:01industries that sort of don't exhibit
12:03those scale economies
12:05um uh or they're kind of limited product
12:06differentiation and in which case you
12:08know you're sort of all competing over
12:10Barry homogeneous Goods absolutely
12:13okay
12:14other thoughts to closest subsistence I
12:17guess the outside option is pretty low
12:19so you're willing to go quite low in
12:21your price as is everybody else
12:24yeah so maybe sort of desperation and
12:25subsistence kind of leads to kind of
12:27Greater competition so it could be
12:28harder to maintain collusion for example
12:30if you're really desperate for that
12:32additional dollar as the business owner
12:34right
12:36other thoughts
12:47okay let me use some of my um list here
12:49again which somebody's recovered so yeah
12:51so specialization in industries that
12:52maybe don't feature strong economies of
12:54scales I think about sort of you know
12:56cottage industries
12:57um agriculture is potentially a big one
12:59so you know the most common firm in
13:01developing countries is a farm
13:03um I will say there's a debate here as
13:05to whether there are scale economies and
13:06agriculture uh the most recent paper in
13:08here that's pretty fantastic is by by
13:10Andrew Foster and Mark rosensweg
13:12um but you may imagine these are in
13:14general you know Industries where there
13:16aren't strong economies of scale okay
13:17and just to play that out what that
13:18means is you're not going to have kind
13:19of these natural monopolies that might
13:21emerge in an environment that you might
13:24have with you know with strong economies
13:25of scale
13:26some of this was mentioned you know
13:28maybe lower degree of product
13:29differentiation okay and that can come
13:30both kind of on the supply side so
13:32specializing in Commodities are very
13:34homogeneous goods and which we might
13:35naturally think competition is quite
13:37High
13:38um as well as in the demands I've read
13:39so poor household that might be less
13:41willing to pay for quality and so you
13:42just have left less differentiation
13:44based on quality measures
13:47um we discussed you know barriers to
13:49entry uh and the fact that they might
13:51legally be higher in some developing
13:53countries but of course there also might
13:54be less enforcement of those barriers so
13:57um you know I think it's quite clear
13:58when you look at say the large and
14:00formal sector right that maybe there's
14:01greater red tape to opening a business
14:03but you don't necessarily need to
14:05register your business to operate
14:07um and so you can have an environment of
14:09many small uh informal firms that are
14:12increasing the competition in the in the
14:14market
14:15there's also depending on the sector you
14:17know sort of relevancy here about
14:18thinking about limited sort of Ip
14:20protection right that might otherwise
14:22prevent sort of CopyCat or generic uh
14:24production
14:25and then finally this is sort of a
14:27nuanced point but I think is maybe a
14:28fruitful area for future researches you
14:31know these de facto barriers like credit
14:32constraints that we talked about being a
14:34barrier to entry they're also a barrier
14:36to firm growth right and so it's
14:39possible if they're constraining growth
14:40more than entry they might actually
14:42serve to maintain a market of many small
14:44firms right so
14:46um you know and I think the micro credit
14:48literature is sort of supportive of this
14:49point in that it seems like when you
14:51alleviate you know these these barriers
14:53these credit constraint frictions it
14:55doesn't do a whole lot to help people
14:56open new business it really seems to be
14:58most effective at kind of the right tale
15:00of very profit already profitable
15:01businesses so you might find if you were
15:03able to alleviate this credit constraint
15:05that we think sort of proliferates in
15:07low and middle income countries you
15:09wouldn't see a ton of Entry but you
15:10actually might see the existing firms
15:12the very profitable ones grow quite
15:13large that might actually limit the
15:15degree of competition could still be
15:17very positive for productivity on the
15:18whole okay I'm not saying it's not a
15:20good thing but thinking about its
15:21interactional competition I think is
15:23fairly nuanced and it really depends on
15:25sort of way wear along the distribution
15:27of latent productivity that constraint
15:29is most binding so I think this is an
15:30area of potential future research
15:34um
15:35questions on that before we go into kind
15:36of measurements
15:41okay so let's turn to measurement I want
15:44to talk first about kind of some of the
15:45standard approaches to measuring
15:48competition and measuring markups and I
15:50want to talk and highlight about how
15:51they these approaches can be potentially
15:54you know particularly challenging or
15:56particularly particularly fruitful in
15:59this development context
16:01so we'll go from kind of easiest to to
16:04most complex so the first thing you
16:05could do if you want to sort of measure
16:07our firms competing colluding is you
16:09could ask them
16:11um that might sound crazy but actually
16:13prior to sort of the Anti-Trust
16:15legislation in the U.S you know U.S
16:17firms used to fairly openly discuss for
16:20example price setting and price
16:21agreements and there's sort of this long
16:23history of the i o literature that
16:24actually uses minutes from board
16:26meetings to identify periods of
16:29collusion and periods where collusion
16:30broke down to study sort of pricing
16:32decisions under these two regimes so if
16:34you can imagine you know in the
16:36development context we run surveys we
16:37talked to firms quite a bit you could
16:40just ask are you competing are you
16:42colluding to set prices how are you
16:43setting prices
16:44so when you think about this approach
16:47in the context of you know working doing
16:50this in developing countries what do you
16:51see as sort of the the pros and cons
16:53overall and then in particular with with
16:55development
17:07everyone thinks it's a fabulous idea we
17:09should just stop here
17:13what are the pros and cons
17:24all right I guess I'll try to answer
17:28um but yeah so so the cons obviously of
17:31asking firms are you colluding
17:33I mean I don't know much about this I've
17:35never run a survey but I just I don't
17:37think they would be very honest but
17:40I don't know that seems pretty obvious
17:42to me
17:43um kind of more difficult to think of
17:44Pros but
17:47yeah so you know I think the the pro or
17:50the relative Pro I will say is that you
17:52know in many uh low and middle income
17:54countries that are sort of weaker there
17:56can be a weaker antitrust regulation or
17:58at least enforcement and so you might
17:59find firms at least more willing to
18:00openly discuss but of course as you say
18:03you know there's still going to be these
18:04strong social norms um against uh
18:06against collusion and that's going to
18:08clearly Drive self-reporting bias
18:10um this is actually a quote from the
18:12paper that that um that I'll describe in
18:14just a moment that I that I worked on
18:15with Michael dinnerstein and you know we
18:17directly ask trainers actually so this
18:19is working with agricultural traders in
18:21Kenya I'll tell you much more about the
18:22context in a moment we directly asked
18:25you know do you have a price agreement
18:26with other traders in this market and we
18:29found is that you know only 30 reports
18:31uh engaging in explicit price agreement
18:33but
18:35um 72 percent of Traders work in a
18:37market where at least someone else told
18:39us in that market that there was a price
18:40agreement and typically they say it's
18:41with everyone else in the market so you
18:44know I like this because it highlights I
18:45think two things the first is that you
18:46actually can learn quite a bit a lot
18:48about the environment through talking to
18:50firms okay so um I just want to
18:53emphasize there's a lot of qualitative
18:54work that goes on behind the scenes
18:56Behind these papers that doesn't show up
18:57necessarily in the paper and I would
18:59really encourage you if you're doing a
19:01specially sort of field work and
19:02development to talk to firms and ask
19:04them if of course it feels both safe and
19:06appropriate given the context but this
19:09also of course highlights the issue you
19:10mentioned that of course there's going
19:11to be very strong reporting bias and I
19:13don't think any anyone would believe
19:15this to be kind of a credible way of
19:16measuring competition but I do want to
19:18encourage you all to do kind of the
19:19qualitative work again with the caveat
19:21of being both safe and appropriate for
19:23the context but to actually talk to
19:25firms um and learn what they say about
19:26sort of how they're interacting with
19:28each other
19:29okay but you know with the with
19:31understanding this is not going to be a
19:32convincing method
19:34um what are some more quantitative or
19:35more convincing methods of measuring
19:37competition
19:38so the next sort of most straightforward
19:40thing you could do is you could use
19:41accounting data to estimate markups
19:43right so you know in theory prices are
19:45typically observable you can get
19:47accounting data to estimate marginal
19:49cost and then you can look at the
19:50difference and estimate markups
19:52so what do you think of the pros and
19:54cons of this approach
19:58um it's going to be hard to get good
19:59data
20:00yeah
20:05yeah so that's a big one so I would say
20:07you know in general even even putting
20:09aside kind of the low and Mill income
20:11country context you know this is
20:13generally not a super popular approach
20:16um uh uh because you know it's hard to
20:19estimate first of all
20:21um uh the value of capital in particular
20:23so many inputs but in particular capital
20:25is very hard to Value
20:27um it's hard for multi-product firms
20:29which a lot of firms are to assign costs
20:31across products
20:33um and then you mentioned I think this
20:34is this is certainly true this approach
20:36is probably even harder in the
20:38development context right so you know
20:39many firms don't keep records in fact we
20:41you know there's this whole literature
20:42and how do we convince firms to keep
20:44records
20:45um and so I think you know for for the
20:48development context it's not just the
20:50value of capital it's even the value of
20:51you know materials and labor that can
20:53also be can also be hard to measure
20:56the other sort of limitation here is you
20:58know imagine there's perfect measurement
21:00you might find large markups and that
21:01might be suggested in perfect
21:03competition but without knowing more
21:05about the future of the market in
21:06particular without knowing more about
21:07demand um you can't run counter factuals
21:09so you can know what markups are
21:11currently but it's difficult to know you
21:12know what would happen to markups if you
21:14had say greater entry or greater
21:15competition
21:17um so this is you know I think this is
21:18more appropriate as a sort of outcome
21:20measure if you already have an
21:22experimental variation or you've got
21:24sort of a quasi-experiment through a
21:25policy shock to use it as a measurements
21:27of an outcome of markups with all the
21:30caveats that I mentioned on measurement
21:31but it's not great for example for
21:33running counter factuals like the kind
21:35that will run for example in the piece
21:37that I'll discuss in just a bit
21:40um okay so in recognition of some of
21:42these measurement challenges
21:44um an approach that sort of uh grown in
21:47um an attraction is this idea of cost
21:49minimization or basically kind of a
21:51production function approach and the
21:53basic idea here is that if firms are
21:55minimizing cost
21:57then for any one given flexible input J
22:01markups are going to be equal to the
22:03output elasticity with respect to that
22:05input okay so what I mean here is you
22:07know imagine that input is labor right
22:09so as you change labor how does your
22:10quantity of output change
22:12over the input expenditures as a share
22:15of Revenue right so imagine this is your
22:17wage bill here over your total revenues
22:20so what are the kind of pros and cons of
22:22this approach relative especially to
22:23kind of the previous accounting measures
22:44as a comment in the Q a saying you have
22:46to estimate a production function by
22:48Nathan Gardner had history
22:50yeah so you have to that that can be
22:52potentially a con right so you have to
22:54actually measure this output elasticity
22:55which can be challenging absolutely
22:58any pros
23:08so you know the pro here in general is
23:11that you really only need decent records
23:12for one variable input right so in
23:14particular the issue about valuing
23:16Capital which will be very challenging
23:18you know we can basically set aside
23:20Capital here and say okay instead we're
23:21just going to focus on a different
23:22variable input labor is a very common
23:24one right and the idea here is it might
23:26be easier you know this denominator here
23:28is really kind of the win for this
23:30approach right so it might be easier to
23:32estimate the wage bill just this this
23:34part here rather than kind of the total
23:36cost and valuing capital okay so that
23:39might potentially be easier now the
23:41downside of course and I should say
23:42actually just you know uh quickly on
23:45that point I think that's especially
23:46true in wealthier countries with good
23:49um accounting data at least on on wage
23:52bills um that might still even be a
23:54challenge right think about smes in low
23:56and middle income countries right they
23:58have no employees or just a small number
24:00of employees own labor is going to be a
24:03really big part of that that wage bill
24:04and we know sort of it's notoriously
24:06difficult to Value uh own labor so this
24:08is not going to be quite the win that it
24:10might be if the only issue is valuing uh
24:12valuing capital
24:14um okay but that's the overall sort of
24:16benefit is the approaches you don't have
24:17to properly measure every input you can
24:19just sort of measure the input
24:20expenditure for for one
24:22um the downside is this point here about
24:24the output elasticity right so there's
24:26lots of identification challenges here
24:28so I think you uh heard a lecture by
24:29Eric bear hugin earlier in this in this
24:32series you know he notes he's got papers
24:34noting that basically if you don't
24:35account for Quality differences you
24:37might get biased estimates here so you
24:39know imagine that there's a firm that's
24:41using lots of inputs
24:43um and they're not actually producing a
24:44larger quantity of outputs but they're
24:46higher quality outputs right then you're
24:48going to underestimate if you don't
24:49adjust for quality or underestimate sort
24:51of this output elasticity
24:57um do you think what this exercise get
24:59easier or harder in the face of integer
25:02constraints so if if let's say my input
25:05is not actually smooth then I can scale
25:06up and down but actually there's only I
25:08can only have one worker or two I can
25:10only have one machine or two but I can't
25:13really scale these up I'm I'm trying to
25:15build an intuition if this is making my
25:17life harder or easier here
25:19yeah I mean I think there's there's
25:21there's uh there's two issues there so
25:23one is just kind of like the the
25:25mathematical issue of are you actually
25:27getting sort of a derivative here I
25:29think the other thing you mentioned is
25:30like are there constraints overall are
25:32you really getting the production
25:34function are you getting some reflection
25:35as constraints here
25:37um uh so I think you know I think that
25:39kind of all falls into kind of this
25:40issue around identification challenges
25:41absolutely I mean this is a place where
25:43I think I think development literature
25:45can do a little better actually
25:48um so you know we have for example
25:51um interventions that have tried to
25:53increase for example like labor drops
25:55right I mean we've we've seen
25:56interventions like that we've seen
25:57Capital drops um and so I think you can
26:00imagine kind of getting at this
26:01experimentally rather than looking kind
26:03of cross uh you know uh within a firm
26:05and kind of thinking about those those
26:06constraints
26:07um the other thing is just on the
26:08measurement side for Quality
26:10measurements so if you are you know the
26:11downside is you don't have accounting
26:13data but if you're going out and
26:14actually talking to firms and you could
26:16measure quality you could potentially
26:17you know try to actually are measuring
26:19and you can try to also measure quality
26:21as well so I think there are ways to
26:23kind of try to deal with them but as you
26:25said there's lots of issues there I
26:26think that complicate this um this
26:28estimation
26:30um and the final thing is of course this
26:31is still descriptive right so again I
26:33think very useful for if you're thinking
26:34on markups as an outcome of a policy
26:37shock uh this is a very useful approach
26:39it's not as useful if you want to think
26:40about
26:41um uh you know what happens if you
26:43increase the degree of competition or
26:44you know if you're trying to estimate
26:46the impact of policies that you're not
26:47observing in practice
26:50okay
26:51um and then sort of you know the last
26:52last kind of very large bucket of
26:54approaches that I'll talk about here is
26:55just kind of the full structural
26:56modeling um this is kind of the
26:58traditional i o approach and I'm going
27:00to go over in a really high level detail
27:01but you can think about for those who
27:02are familiar like the very classic blp
27:04approach
27:06and this approach you basically assume a
27:07particular utility or demand function
27:09you try to get supply shifters to
27:12estimate demand elasticity and then you
27:14assume particular Market structure and
27:15firm behavior and those two things
27:17together imply particular uh markups or
27:20marginal costs
27:21now of course the benefit here is you
27:23know everything so you can do lots of
27:24counter factuals
27:26um the downside is there's lots of
27:27assumptions so you're assuming you know
27:29the shape of demand
27:31um you're even assuming Market structure
27:33right so if the goal is to identify the
27:35market structure this is not a fruitful
27:36approach I will say In fairness you know
27:39not all these papers do this so this
27:40very famous paper by Aviv Nevo on um on
27:43ready to use cereals actually goes
27:45through exercise for three possible
27:47Market structures and then compares the
27:49kind of implied markups
27:51um to actual accounting estimates right
27:54so you know you can and that would be
27:55actually a somewhat similar approach to
27:57what we'll do in my paper with Michael
27:58dinnerstein
28:00um so they don't all assume this but a
28:01lot of them do and then it's sort of
28:02circular for not useful for identifying
28:05Market structure
28:06the final thing that I think is a little
28:08specific uh to uh development uh from
28:12this approach
28:13um but also maybe through a little bit
28:15in in the general literature but
28:17especially in development is I think
28:18it's pretty hard to find good
28:19instruments to estimate this demand
28:21elasticity right and so in particular in
28:23the developing country context you know
28:25I work for example a lot in agricultural
28:27markets so it's not so hard to think
28:30about
28:31um uh you know Finding shifters of
28:33Supply so think for example about
28:35um you know weather shocks shifting
28:36supply of production different areas
28:38but in a setting which you've got very
28:40small markets and your producers are
28:42also your local consumers of course as
28:44there's weather shocks you're changing
28:46household income which is going to
28:47change demand in a bunch of ways so
28:49actually identifying clean shocks I
28:51think in these smaller markets where you
28:52have kind of this overlap between who's
28:54the producer and who's a consumer
28:56um can be quite challenging
28:58that said I also view this as a as a
29:00place really as an opportunity for
29:01development right so we do a lot you
29:03know one of the things that development
29:05economists are good at though not
29:06exclusively as running experiments so I
29:08think there's actually lots of ways to
29:09think about experimentally estimating
29:12some of these uh key parameters that
29:14aren't going to then rely on these
29:16um uh instruments or identification
29:17assumptions
29:21okay so let's turn now to uh one example
29:23of that which is
29:25um work that I uh with Michael
29:26dinnerstein uh in Kenya and it's
29:28basically an experimental approach
29:31um to measuring the degree of
29:33competition in these markets let me give
29:35you a very kind of high level uh
29:37motivation behind this project
29:39um and just hold up the setting so uh
29:42we're working agricultural markets in
29:43Kenya it's pretty common to see pretty
29:46large variation in prices across these
29:48markets and it suggests these markets
29:49are sort of imperfectly integrated
29:52and this is a big interest of policy
29:54makers you know they're concerned kind
29:55of on two fronts one is that you know
29:57there seem to be low prices for farmers
30:00props are their main source of income
30:01and the flip side can have high prices
30:03for consumers and that raises issues of
30:05food insecurities there's a question
30:06that's kind of you know where is this
30:08this Gap coming from
30:10uh of course just High transaction costs
30:12play a large role so there's High
30:14transport cost search costs risk Etc
30:17but many policy makers are concerned
30:19that part of that Gap in prices between
30:21say farm gate price and Consumer Price
30:23Is markups by intermediaries so traders
30:26who are buying from Farmers
30:28um and selling to Consumers
30:30and the idea here in this context is
30:32there actually this is the context where
30:33there are not large barriers legal
30:35barriers to entry at least not ones that
30:37are enforced
30:39um but there are large de facto barriers
30:41so you have to have a good deal of
30:42working capital also a network of
30:44contacts to enter into the industry and
30:47the idea is these barriers might allow
30:49these intermediaries who are then you
30:51know the few who are able to overcome
30:53these barriers might be able to exert
30:54Market power
30:55and potentially it include
30:59so that's the back story so uh in this
31:01paper we're going to provide
31:02experimental and model based evidence on
31:05three questions so first what is the
31:07market structure and level of
31:08competition among Traders two how does
31:11this Market structure affect consumers
31:13and three what role does entry have in
31:15facilitating competition
31:17so I want to give you kind of I'll tell
31:19you more detail about the experiments in
31:21a bit but before I get into the theory I
31:23want to give you a very high level of
31:24understanding of the experiments because
31:26they map to the theory pretty closely so
31:29we uh work in 60 markets in Kenya we
31:31want run three field experiments and
31:34each of these experiments estimates a
31:36key parameter or moment in the model of
31:39competition demand and Market entry okay
31:41so the first experiment again I'll give
31:43you more details in a moment the first
31:44experiment is designed to estimate
31:46pass-through so in this experiment we
31:49offer to all traders who are located in
31:52randomly selected markets we offer them
31:55a subsidy per bag of maize that they
31:58sell to Consumers and we model this like
32:00a reduction in their marginal cost
32:03and we measure how much that reduction
32:05in marginal cost they pass through to
32:07consumers in the form of lower prices so
32:09we're going to get experimental
32:10estimates of pass through
32:13in the second experiment we estimate
32:14demand
32:15so we offer randomized reductions in the
32:18price to Consumers and we measure the
32:20corresponding quantities they buy at
32:22those randomly reduced prices and that's
32:24allow us to estimate demand and in
32:25particular I'll show you in a moment
32:27from the theory that the curvature of
32:28demand is going to be the really key
32:29feature here
32:31these two things together and tell us a
32:33lot about the underlying
32:35um model of competition that these
32:37Traders are operating under
32:40and then third experiment measures the
32:42impact of entry and competition so in
32:45this experiment we randomly offer
32:47um an entry subsidy to bring new Traders
32:50into the market and we measure what
32:52happens to prices and what happens there
32:54for the competition when new entrants
32:55come in
32:58we're going to then take these
33:00experimentally estimated parameters and
33:02embed them in a structural model and
33:04that's going to help us to identify the
33:05level of competition and the effects
33:08eventually
33:11okay so a little more detail here so
33:13we're working here in Western Kenya in
33:15these 60 markets you can see here on
33:16this around this area called bungoma and
33:18kakamega
33:21um the way these markets work they're
33:23sort of open air weekly markets so for
33:24example you might have your local market
33:26might be a Thursday market so every
33:27Thursday there's a market
33:29Traders show up bringing Maize to these
33:31markets uh they'll bring them in these
33:33trucks and they'll sell out of the back
33:34of their trucks
33:35this is sort of the most common thing
33:37you'll see where there are there are the
33:39model the modal uh Market has three
33:41Traders and they'll sell basically to
33:42Consumers um who line up right here out
33:45of the back of their truck
33:48um
33:49you know I think I'll skip most of this
33:51let me say I think the basic thing to
33:53take away here is this is
33:55um you know pretty much a spot Market of
33:56pretty homogeneous Goods so there's very
33:58little sold on credits
34:00um there's not much evidence of price
34:01discrimination or product
34:02differentiation or Price stickiness so
34:04you might think this is actually a
34:05pretty competitive market
34:07um on the other hand you can see these
34:09guys are sitting right next to each
34:10other it's not so hard for them to
34:12observe each other's prices and talk
34:18okay so let me now give you sort of the
34:20um the simplified theoretical framework
34:22that underlies uh our experimental
34:25approach
34:26um I should say and what I'm going to
34:27show you here I'm going to make a couple
34:28simplifying assumptions for example I'm
34:30going to assume constant marginal cost
34:32symmetry across Traders there's a full
34:35structural model in the paper that
34:37relaxes all those assumptions
34:39um it doesn't make a big difference for
34:41the results uh we still identify all of
34:44our our estimates experimentally when we
34:46relax uh those assumptions but I like
34:48this simplified approach
34:50um uh to show first because it really
34:52has a kind of clear mapping one to one
34:54between the parameters we're estimating
34:56in each experiment and the theory
34:59let me kind of walk you through this
35:00kind of simplified framework
35:03so uh Trader J is going to choose
35:05quantities right to Max maximize first
35:09his profits right so just price Minus
35:12cost times quantity again very simple
35:15plus the weighted sum and we'll come
35:18back these weights in a moment
35:19of the profits of all of his rivals
35:24and so how do we interpret this this
35:26profit weight here this Omega so this is
35:29going to basically embed two
35:30well-defined
35:32um uh models of competition okay so when
35:34this is zero right this when Omega zero
35:36this whole term goes away and we're just
35:38back to Cornell competition
35:42conversely The Other Extreme when Omega
35:43is one right that's as if the trader is
35:47putting equal weight on his own profits
35:49and his Rivals profits
35:51they're basically maximizing as if
35:53they're a single firm that's perfect
35:54collusion and there's ways to even embed
35:56kind of Birch Run competition in here as
35:58well
36:02so we can take the first derivative with
36:04respect to quantity to get the optimal
36:06condition for price so price is going to
36:09equal Cost Plus some markup term here
36:10right
36:12and this markup term is going to depend
36:13of course in the shape of demand which
36:15will estimate and come back to in a
36:16moment
36:17but it's also going to depend on two
36:19features that are determining
36:20competition okay first is n the number
36:22of traders in the market
36:24and then second this is Omega term right
36:26how do those Traders interact
36:32so we're going to uh summarize those two
36:34features of competition the number of
36:37traders in the market and the way those
36:39Traders interact into a single
36:41competitiveness index parameter right
36:44and this is going to be useful in just a
36:45moment
36:46and again we can think what this is when
36:48when we're no competitive environment or
36:50a collusive environment okay and again
36:52also nespertrand
36:54okay so I'll come back to why we've kind
36:56of gone through the algebra here on
36:57Sigma in just a moment
36:59let's go back to the first order
37:00condition here again remember prices
37:02equal Cost Plus a markup term here
37:05so we can take the derivative respect to
37:07cost okay and that's going to get an
37:08equation for pass through so what is
37:10passed through passthrough is How do
37:11prices change when costs change
37:14and what you can show is this is this
37:16past is going to be a function basically
37:17of two features of the market the first
37:19is that competitiveness parameter okay
37:21the sigma
37:23and the second is this e term okay this
37:25e term is reflected on demand and
37:27specifically e is equal to the
37:29elasticity of the slope of inverse
37:31demand so that's more complicated than
37:33just the elasticity of demand this is an
37:35elasticity of the slope how is the slope
37:37changing with quantity
37:39and I'll give you some graphical
37:41intuition on that in just a bit but you
37:42want to think of basically the curvature
37:43of Demand Being particularly important
37:45here
37:46right so again the takeaway here is that
37:48Pastor is reflective of the degree of
37:51competition in the market and demand
37:54so let me now map that to the
37:56experiments we're going to run in this
37:58paper
38:00so we're going to estimate rho or pass
38:02through from the pass through experiment
38:05we're going to estimate e this term on
38:07demand from the demand experiment and
38:09those two things together are going to
38:10allow us to back out the implied
38:12competitiveness parameter right so when
38:14we combine these two features of the
38:16data we're going to be able to say you
38:17know this is the underlying sort of
38:19model competition that best fits what
38:22we're observing in the marketplace
38:26let me pause there and ask their
38:27questions
38:31um Lauren there's one from the chat
38:34um how confident do you feel about the
38:36market boundaries is that Arbitrage
38:38going on between subsidized
38:39non-subsidized markets although you can
38:41you may want to answer that in a few
38:43minutes yeah no it's a great question so
38:45we um and that matters I should say uh
38:48for
38:49um for for uh for measurements in a
38:51couple ways right so
38:53um uh for example if our control markets
38:55are too close to our uh treated markets
38:58then we might be changing prices and
38:59control markets right and then we're not
39:01going to get the right estimate of
39:02passing so there's a really basic sort
39:04of measurement issue here so we
39:05intentionally spread out our markets
39:07quite a bit so we did a census of
39:08markets and then basically kind of Drew
39:10parameters or you know or or radii
39:13around each of the markets in our sample
39:15and so we intentionally spread them out
39:16and we could you know test for what
39:18happens to having a market sort of
39:20somewhat nearby that's treated it
39:21doesn't affect your your quantities
39:24um so we don't think there's an effect
39:25on our control group I will say it does
39:28seem like there might be some slight
39:29sort of business stealing from other
39:31closer markets that are not in our
39:33sample
39:34um and we have sort of an appendix sort
39:35of looking at that so I think this is um
39:37uh and that's real I should say that's
39:39real additional business for the Traders
39:41so I think it's sort of okay to include
39:42in our estimates here
39:44um but that is a slightly nuanced point
39:45and I think it's really critical when
39:47you think about in general doing Market
39:49level experimentation to think
39:51critically about these boundaries um in
39:53terms of how they affect one year
39:55measurement of the treatment effect and
39:57then two kind of your interpretation of
39:58what that means I'm happy to come back
40:00to that sort of once we go through the
40:01experiment or even at the very end when
40:03I discuss sort of at the end of this uh
40:05lecture I'll discuss more broadly kind
40:07of Market level experiments
40:09that's a great question
40:11there's one more quick one which is uh
40:14you mentioned uh that um it's often
40:18common that people are both producers
40:20and consumers
40:22um how do you deal with those issues for
40:24example pass through maybe diff
40:25different firms where the producers are
40:28for a consumer yeah that's a good point
40:30so in this context was we're working
40:32with large Traders
40:33um you know they are possible consumers
40:35they're not going to be in our demand
40:36experiment um there's a very small
40:38number of them I worry about that much
40:40more when you think about for example
40:43um you know working with individual
40:44farming households
40:46um as as being kind of the firm in this
40:48case The Firm here is going to be the
40:49trader and so they're going to be kind
40:50of the drop in the bucket in terms of uh
40:52in terms of demand
40:56um and for the panelists I should say it
40:58might help if you have questions if you
40:59click the raise your hand button I think
41:01you will then rise to the top of my
41:02screen which could be helpful so just I
41:03don't want to ignore any of you
41:06um okay so uh in experiment one let me
41:08tell you more details about each of the
41:10experiments so in experiment one
41:12um where we're targeting this this
41:13pass-through parameter so how did this
41:16work so we randomly selected markets and
41:19then those randomly selected markets we
41:20offered uh to all traders in the market
41:23a subsidy per Bank sold uh and the goal
41:26here again is to sort of mimic a common
41:28shock to their marginal cost
41:31so we offered to all traders in a given
41:33Market the same subsidy level but there
41:36was randomized variation across markets
41:38so in three quarters of the markets we
41:40offered
41:41um a low subsidy now low is just
41:43relative to high it's still pretty large
41:45and absolute value so it's eight percent
41:46of total price
41:48um and then according to the market we
41:49offered a high subsidy of 15 of total
41:52price
41:53the reason we have this variation become
41:55clear in a moment we're going to use
41:56this to test a particular functional
41:58form we're going to place uh on demand
42:02the subsea is offered for four weekly
42:04Market days in a row so if you're a
42:05Thursday Market gets offered for four
42:07Thursdays in a row covering a month in
42:09total
42:10um this is actually a big chunk of the
42:11selling season the really active selling
42:13Seasons about three or four months so
42:15this is you know about a third or a
42:16quarter of the selling season
42:18and then I won't have time to describe
42:20this much in detail but we have really
42:22detailed transaction level data we had a
42:23numerator so we're sitting there
42:24monitoring sort of each transaction
42:27um uh and that's going to allow us to
42:29estimate how much of the subsidy is
42:31passed through to Consumers yeah is
42:33there a question
42:34yeah what's the rationale for not
42:36varying within the same Market over time
42:39these rates wouldn't you get interesting
42:40variation there yeah I should say
42:42actually
42:44silly so
42:45I'm not sure I've been full but
42:46essentially basically markets are kind
42:48of rotating month by month through the
42:49different treatment statuses so we are
42:51able to basically include kind of Market
42:52fixed Effects by seeing what happens
42:54when a particular Market's in control
42:56yeah
42:57and that's actually pretty critical that
42:58reduces our standard errors quite a bit
43:01um
43:02it also allows us to see kind of if
43:04there's kind of any stickiness does it
43:05affect crisis Next Period answer is no
43:07so that's useful in a couple ways
43:10a good question
43:12um the second experiment is the demand
43:14experiment okay so here we offered
43:16randomized discounts and the prices that
43:18are paid by individual uh consumers in
43:20the market so here's how it works so
43:22consumers would agree to a price and
43:24quantity with a Trader
43:26then after that we would have them draw
43:28one of ten randomized discounts
43:30we then allowed customers to choose the
43:32quantity they wanted to purchase at that
43:34reduction in price and what that's going
43:36to give us basically is an exogenous
43:38shift in price and the resulting change
43:40in quantities and that's going to allow
43:42us to parameterize a pretty flexible
43:44functional form for demand and estimate
43:46that curvature term that e-turn and I'll
43:49give you more details on that that exact
43:50estimation in a bit
43:55the third experiment is estimating the
43:57effect of Entry
43:58so for each market
44:00um we that was randomly selected for
44:02entry we identify three sort of Entry
44:05Traders these are potential entrants
44:06folks we who have never worked in that
44:08market before and we offer a grant to
44:11enter the market on particular days and
44:13attempt to sell
44:15now we offer these three uh grants
44:17partially just to make sure we got sort
44:19of a strong first stage for entry
44:21benefit of these three different entry
44:23offers is we're able to randomize the
44:25size of those grants into low medium
44:27high
44:28um that's basically like variation in
44:31subsidizing their fixed costs to entry
44:33or it's a fixed Grant and that's going
44:34to allow us also to back up fixed costs
44:36which is going to be quite important for
44:38once we get variable profits
44:39understanding sort of what total profits
44:41look like
44:43again these offers are given for four
44:45weeks in a row the goal is here to do
44:47sort of long-ish run entry
44:49um and then we uh collect transaction
44:51level data to measure the effects of
44:53increasing the number of Traders on
44:55prices and we can back out then the
44:57model what happens to the underlying
44:59model of competition
45:01when entry happens
45:05questions there on the on the design
45:10yes
45:13um I was wondering can you say anything
45:14about how this um sort of encouraged or
45:18incentivized entry different from uh
45:22your the the entry of in the sort of
45:25like a natural condition where they just
45:28like do the do it themselves instead of
45:31being in descendants and advise to do
45:33that yeah how different is it gonna be
45:36yeah so so
45:38um so you know I think you might worry
45:40that this is sort of artificial is this
45:42kind of like you know natural entry
45:45um and so we can't say for sure I mean I
45:47will say we um you know we try to do it
45:49for kind of long run there were actual
45:50profits on the table for these folks
45:52they actually went in there and actually
45:53sold
45:54um so people took it pretty seriously
45:56there is a question of is this what you
45:57would see sort of in reality of natural
45:59entry and the way we can sort of answer
46:01that we can't answer that for sure you
46:02know that's an external validity
46:03question what I can tell you is given
46:05variation in what we do see you know how
46:08would you then apply that to kind of
46:09natural entry so for example
46:12um just to preview something I'll share
46:13in a moment we found that it was easier
46:15to get uh entrance so no one had worked
46:18there before but it was easier to get
46:20entrance in who knew the incumbent so a
46:23lot of these guys kind of know each
46:24other
46:26um so some people kind of knew some of
46:27that incumbent some people didn't we
46:30basically
46:31um for lower values of that entry Grant
46:34people were more willing to go if they
46:36knew the incumbents and so you might
46:37think that's then meaning that in
46:39reality if you weren't subsidizing the
46:41entrance I'm more likely to kind of know
46:42the incumbents then the second question
46:44you should ask is how does that then
46:46change the effects of Entry right
46:49um and we find basically and I'm you
46:51know totally spoiling the ending here
46:52but what we find basically is that
46:53entrance by you know entry by uh enters
46:57who know the incumbents doesn't doesn't
46:58increase competition at all
47:00um so it's going to look collusive
47:02before it has to look Lucid after
47:04um entry by new traders who don't know
47:06the incumbents does actually seem to do
47:08a little bit for increasing the degree
47:09of competition we do see more price
47:11drops with those with those uh that
47:13level entry but those are the guys who
47:15are hard to get in so if you think about
47:16kind of the marginal entrant in reality
47:18it's not going to be the totally new
47:20guys right
47:22um and so that's sort of what we can say
47:23given the variation that we see in the
47:24data
47:27yeah Timothy
47:31uh just a very operational question
47:33about like managing this in the field
47:35yeah the enumerators can see all of the
47:39like they're standing one place and can
47:41see all of the trades happening at once
47:43how how much can the Traders see of each
47:45other how do you hire the right number
47:48of people to make sure that you're
47:49observing enough that you're actually
47:51yeah yeah this was like a whole you know
47:53the whole field operation so
47:55um so uh we had a variable number of
47:58enumerators so there are there are many
47:59more than one there's sometimes
48:01one-to-one mapping of enumerators to
48:02Traders it really depended on the
48:04overall sort of
48:06um uh uh intensity of trade sort of in
48:09that part of the season so um for very
48:11intense part of the Season we had a
48:12one-to-one mapping of an enumerator to a
48:14to a Trader uh and less intense ones
48:17where there were fewer transactions you
48:18know we had a you know one enumerator
48:19maybe for every two Traders they can go
48:21back and forth but they're literally
48:22observing every uh every transaction
48:26um so it was very healthy this is not a
48:27cheap experiment it was very labor
48:29intensive and the traders in all of
48:31these markets can always see what the
48:33other Traders are doing yeah because
48:35um uh and there is some variability in
48:37this but if you look back to kind of the
48:39the layout so you know can these guys
48:41see what the others are doing maybe not
48:44but they're they're generally very
48:45clustered in the in the marketplace now
48:47there's reasons for that totally besides
48:48collusion which is like this reduces
48:50search costs for consumers right to know
48:52you kind of go to this part of the
48:53market and you see so I think this is
48:54not like prima facie evidence that's
48:55collusion but yes they're very close
48:57together and they're kind of sitting
48:58there chit chatting out of the back of
48:59their trucks
49:02um uh I will say we do in that
49:04measurement of transaction cost
49:06um uh we tried to also or sorry of
49:08prices we also measure are there sort of
49:10less obvious changes to prices like for
49:12example there are these little things
49:13called top-ups which means you might
49:15give a few extra kgs of maize that we're
49:18not included in the price you might even
49:19give a Coke or money for a motorcycle
49:21tax to take and those things uh when we
49:24include them that actually does increase
49:25the degree of pass-through observed by
49:27about one to two percent so that might
49:29be a slight way that there's kind of you
49:30know those are less observable ways of
49:32cutting the price but it's pretty
49:33Limited
49:36other questions
49:40okay so let me show you I've contribute
49:41already the results a little bit so we
49:42basically very limited pass through
49:44um so we're able to estimate a pretty
49:46precise uh 22.4 pass the rate
49:50um that doesn't change by uh first of
49:52all the low high distinction okay so
49:54it's almost an identical pass-through
49:56rate
49:57um even when you double the size of the
49:59the shock and we'll come back to that in
50:01a moment
50:02there's also no obvious heterogeneity by
50:03for example it's not the case that in
50:05larger markets you see greater faster
50:07than smaller that's already some
50:08evidence against uh corn no competition
50:10but we'll come back to that in a moment
50:12okay so overall kind of 22 pass through
50:14rate
50:15now you'll want to ask how do we
50:17interpret that pass through rate right
50:18what underlying model of competition has
50:21given a rise to this empirically
50:22observed 22 pass-through rate and I
50:25wanted to give you like very simple econ
50:27one graphical intuition on why we don't
50:29know unless we understand what demand
50:31looks like right and so what I want to
50:33point out here is you could have Justice
50:34equally been in a collusive environment
50:36or a Chrono competitive environment that
50:38gave rise that 22 passed through rate so
50:40why is that so we've basically induced
50:43this experimental reduction in marginal
50:44cost we've seen you know this reduction
50:47in price the ratio here is kind of
50:49roughly imagine it's 22 percent
50:52um uh this could have come from this
50:54model of collusive behavior but it could
50:56have also does this easily be Observer
50:59observational equivalent of a core no
51:01competitive environments uh uh with uh
51:04with this shape demand okay so the
51:05really critical thing to know are we in
51:07this world or this world is
51:09understanding the curvature of demand
51:11okay so that's that's why we have to go
51:13to the second experiment to estimate
51:15demand to back out from this
51:16pass-through rate what is the underlying
51:18model of competition
51:20so that's the demand experiment so this
51:23is just showing you kind of the raw data
51:24from the demand experiment so we offered
51:26these randomized reductions uh in price
51:29we measure the corresponding change in
51:31quantities okay so you know downward
51:33sloping demand that's that's a good
51:35first start
51:36okay so this is the raw data
51:38we're then basically going to use this
51:40experimental variation and price to
51:42estimate a general sort of below fighter
51:43class of demand functions now this may
51:46look a little bit funky or unusual but I
51:48want to just highlight a couple things
51:49you know the first is that this actually
51:50nests a bunch of demand functions that
51:52are probably much more familiar to you
51:54this is going to demand Nest linear
51:56quadratic isoelastic
51:59the reason we're using this particular
52:01class of demand function is because the
52:03the kind of defining feature of this
52:04class of demand functions
52:06um is it's going to capture curvature
52:08with a very small number of parameters
52:10okay the coverage we really care about
52:11and here under this class of demand
52:13functions this Delta here is going to be
52:15the really critical piece okay so here
52:17that very complicated e term is going to
52:19simplify the Delta minus one okay it's
52:21going to be pretty parsimonious and much
52:22more tractable
52:24now you might say okay it's parsimonious
52:26intractable but is it totally crazy does
52:27it actually fit the setting and that's
52:29where those two different size subsidies
52:31come into play right so one prediction
52:33out of this class of demand functions is
52:34that for you know a given Market
52:36structure which it was randomized the
52:38market structures should be the same and
52:39low high subsidies you should see a
52:41constant pass-through rate okay and
52:43that's what we find so I think there's
52:44some empirical support as well for this
52:48um so you know we're gonna take log
52:49first differences the previous equation
52:51and estimate this using GMM the really
52:53key thing here is we're using the
52:54randomized subsidy amounts as an
52:56instrument for price
52:58so these are the results
53:00um Delta here as I mentioned is kind of
53:01the the key term so harassment Delta's
53:04around four
53:05I want to highlight something very
53:06transparently which is we're getting
53:08really large confidence intervals here
53:10in Delta okay and that should not be a
53:11big surprise to you Delta is the second
53:13order sort of curvature term okay so you
53:16know we can just barely rule out kind of
53:18linear linear demand which would be a
53:19Delta of one
53:21we also cannot rule out you know a very
53:24curved Delta of of a very curved demand
53:26of adult of sex okay
53:28okay so we're not getting kind of
53:30precision this really critical term
53:31but what I'm going to show you in the
53:33next couple of slides is that first if
53:34you take Delta at its Point estimate
53:36okay the right model of competition is
53:38going to sort of jump out at you it's
53:39gonna be very obvious kind of what fits
53:41with this particular estimate
53:43the other thing I'm going to show you
53:44the second thing is that even if you
53:46take into account all the uncertainty in
53:47Delta we still have enough Precision to
53:50rule out some of the other common models
53:52of competition
53:53so let me show that to you now
53:55so remember sort of for the simplest
53:57model price can be a function of that
53:59competitiveness parameter and demand
54:01okay and now here because we have this
54:03particular functional form that e
54:04distrib is basically just simplifying to
54:06Delta
54:07okay so let's put all the kind of pieces
54:08together
54:09so now we've estimated Delta from the
54:12demand experiment
54:13we have
54:15um you know benchmarks for Sigma under
54:17various Market structures right we can
54:19test collusive or no competitive or sort
54:21of Patron competition
54:23and so we can say putting these two
54:25pieces together you know what pass
54:27through eight what row here should we
54:29have expected if markets are operating
54:31under each of these Market structures
54:34so let me show that Benchmark to you
54:36here okay so our predictions given our
54:39estimate of demand curvature is that if
54:41markets are are Bertrand right we should
54:44see 100 pass-through
54:46if they're acting core no competitive
54:48given kind of distribution of the number
54:49of Traders we should see on average of
54:5146 percent pass through and if they're
54:54collusive we should see almost a 20 pass
54:56through right okay so those are the
54:58benchmarks we want to now compare this
55:00to empirically what we actually observe
55:01as the past the range
55:04and so here's what we found so you
55:05already saw we have a 22 pass-through
55:07and this is the bootstrapped um
55:09confidence interval of 95 confidence
55:10interval around uh around around the
55:13estimate right so we can rule out here
55:14for no competitive embertron
55:17so it's very clear sort of given at
55:19least the point estimate Delta what the
55:20right model is
55:22now you should say well yes but there's
55:23that huge confidence interval around
55:24Delta are we really confident right
55:26and so we can do is basically just
55:28rearrange those terms right we're just
55:29going to solve here for a sigma and now
55:31we can take into account all the
55:33variation of Delta all the variation in
55:35in row right taking the bootstrap
55:36distribution of both of those right and
55:38get the the point estimate and
55:39distribution on Sigma
55:41I'm going to slightly adjust this to
55:43solve it sort of as the profit weight I
55:44think that's a little bit cleaner
55:45mapping back to the theory so remember
55:47one is collusive and zero for no
55:49competitive okay that's the weight that
55:50you're putting on your Rivals profits
55:54and we see as our Point estimate right
55:56is an Omega of 0.78 so almost an
55:59estimate of one basically and even given
56:02all that variation right even given our
56:03imprecision and our estimate it's also
56:05we can still rule out with 95 confidence
56:08an Omega estimate that's consistent with
56:10Cornell so we say is given the point
56:13estimate we're you know it looks really
56:14close to exactly expect given collusive
56:17competition uh collusion and we can rule
56:19out Cornell competition
56:24we do a bunch of other stuff in the
56:26paper so I mentioned we kind of moved
56:27beyond that simple model we account for
56:29a potential non-uh constant marginal
56:31cost again using additional experimental
56:34variation we bring in Trader hygenity
56:36demand heterogeneity
56:38it doesn't change much so our estimate
56:39of Omega given all that is is 1.07 okay
56:42but very you know again still very much
56:44centered around that kind of one that's
56:45the collusive benchmark
56:48we then I won't have time today to
56:50discuss kind of the the full entry
56:51result but I've already previewed them
56:52uh for you right so first the take-up of
56:54that subsidy amount is going to allow us
56:56to estimate the distribution of fixed
56:58cost
56:59then we see what is the impact of
57:01entrance okay and we find basically the
57:02entry by one additional Trader is
57:04reducing prices by about one percent so
57:06it's significant it's pretty precisely
57:07estimated but it's a pretty small
57:09reduction in prices we see this
57:11heterogeneity by this feature I
57:13mentioned before do Traders have
57:15contacts in the market okay so when they
57:16already have contacts in the market you
57:19can see it's really not changing Omega
57:20metric going from you know 1.07 to 0.95
57:23so you know you you can reject uh
57:26Cornell competition and you clearly
57:27cannot reject kind of collusion at one
57:30on the other hand entry by completely
57:32new traders who don't have any contacts
57:34with incumbents or don't know the
57:35incumbents at all actually does seem to
57:36decrease competition so here we actually
57:38can rule out a collusive behavior okay
57:41and we can't rule out uh Cornell so it
57:43does look like this uh entry by new
57:45Traders does seem to increase the real
57:47competition but if we go back to kind of
57:50those take up results right who's most
57:51likely to take up we find it's harder to
57:54get Traders without context to enter you
57:55literally need to pay them more to get
57:57them to go it suggests right that having
57:59contacts that sort of facilitates your
58:00willingness to enter
58:03okay so you know given all the structure
58:05we can do a lot to kind of quantify what
58:07do markups look like what do variable
58:08profits look like
58:10um I'll say you know markups are pretty
58:11large they're kind of in the 40 to 50
58:13range
58:14but fixed costs are also pretty large
58:16and so that's going to dissipate 71 to
58:19the variable profits okay so actually
58:20for the median Trader they're not making
58:22huge profits in this context however
58:25there's lots of variability across
58:27Traders okay so the largest traders who
58:29are serving the largest part of the
58:30market they are making really large
58:32profits and so you know the take home
58:34here is that even though the median
58:37Trader is not making huge profits the
58:39typical consumer is being served by a
58:41Trader who is so when you think about
58:42kind of total Surplus in the market
58:44Traders are capturing you know over 80
58:46percent of the Surplus in the market
58:50and then finally we can do you know
58:51counter factuals again because we've
58:52estimated sort of all the things we need
58:54uh to describe the market we can
58:56estimate counterfactuals what happens if
58:58we go from kind of this collusive
59:00Benchmark which what we estimate is sort
59:02of the right model of the current degree
59:04of competition and what if we're able to
59:06kind of lower that Omega down to core no
59:08competitive right to increase the degree
59:10of competition going to the left
59:12and so as you expect you know consumer
59:14surplus kind of goes way up
59:16um that's partially at the cost of
59:19Trader Surplus going down
59:21um but it's also partially coming from a
59:22reduction in debate loss right uh as
59:25you're decreasing sort of the market
59:26power in this environment
59:30um let me actually pause there and ask
59:32their questions on on that paper
59:36yeah please
59:39um so I was wondering if you could say
59:41um this probably goes beyond the scope
59:43of the study but I was wondering if you
59:45could say something about the structure
59:46of the market
59:48um like in a sense that because I feel
59:49like this is the the results shows that
59:52there is a stable sort of collusion
59:54going on so what is supporting this
59:56what's the mechanism behind this um
59:58joint profit maximizing and we also see
1:00:02there are like
1:00:03um some Giants basically working there
1:00:06uh earning most of the profit is it
1:00:09because they are they're so some of them
1:00:11they are including with each other and
1:00:13then other followers they Benchmark
1:00:15their price
1:00:17um based on those like the market
1:00:19leaders something yeah it's a good
1:00:21question I don't have actually at the
1:00:23last Point although it's a really
1:00:24interesting question as to kind of how
1:00:25the exact price setting agreements um uh
1:00:28happen so that that I don't know
1:00:30um in terms of kind of what's
1:00:31maintaining it I actually think the
1:00:32physical layout of the market becomes
1:00:34really important here so you can imagine
1:00:36kind of the period of benefits from
1:00:38defection might be quite low if you're
1:00:39kind of immediately detected
1:00:42um and then right away sort of the
1:00:43collusive agreement breaks down
1:00:45um so you can imagine this an
1:00:46environment where sort of you know
1:00:47immediate detection basically of of kind
1:00:49of that breakdown in the price agreement
1:00:51could lead to kind of a real loss in
1:00:53profits
1:00:54um so I think that's sort of the the the
1:00:55ability to kind of maintain I should say
1:00:58you know I'm not I'm not convinced that
1:00:59it's always even like an explicit
1:01:01agreement that you saw actually a large
1:01:03percentage of Traders do sort of
1:01:05largemouth markets have at least one
1:01:07Trader who's admitted there is an
1:01:08explicit agreement
1:01:09um but even sort of it could be
1:01:10potentially you know tacit but they're
1:01:12sitting there kind of watching each
1:01:13other as well
1:01:14yeah
1:01:16other questions
1:01:19are there hierarchies between the
1:01:21Traders could there be something like
1:01:23there's kind of the big guy who's always
1:01:26been there who's who's goes first and
1:01:28sets and the other ones just watch and
1:01:30see what they do right yeah I think it's
1:01:32kind of Science question I don't really
1:01:33know the answer that's a great question
1:01:34like is there kind of a a leader who
1:01:36said in the price and others kind of
1:01:38follow so I don't know that's a good
1:01:39question
1:01:40yeah
1:01:42um that may be relevant for thinking
1:01:43about kind of how do you if you wanted
1:01:45to Target policy to potentially break
1:01:47some of this down sort of who would you
1:01:48target you know absolutely
1:01:50okay so um in our last half an hour
1:01:53let's turn to sort of um we talked about
1:01:54kind of measuring competition let's turn
1:01:56now to kind of implications of
1:01:57competition on Market outcomes right so
1:02:00um we'll talk first about uh impacts on
1:02:02on firm selection sort of who survives
1:02:05who exits when when competitions
1:02:07increased
1:02:08um and then I'll talk just very briefly
1:02:10about sort of three other really nice
1:02:12papers looking at the impacts on prices
1:02:14quality uh and Contracting
1:02:17okay so we'll start with this Jensen and
1:02:19Miller piece um so the overall kind of
1:02:21motivating question behind this paper
1:02:22which is you know a really huge one in
1:02:24the space is why do good firms in this
1:02:27context why do you know why do good
1:02:29firms whether low cost High productivity
1:02:31high quality why do those good firms
1:02:33fail to grow and the flip side of that
1:02:36coin is why do these bad firms fail to
1:02:38exit right why do they remain in the
1:02:40market
1:02:41so one possible explanation is basically
1:02:44a lack of competition
1:02:46and in this context they're going to
1:02:47look at a lack of competition due to uh
1:02:50Market isolation Okay or limited
1:02:52integration so we kind of raise this
1:02:54point before how do you define that
1:02:56um this this paper essentially changing
1:02:58the boundaries of the markets
1:03:00okay so you know what are these barriers
1:03:02to Market integration lots of barriers
1:03:05uh that are possible High transport cost
1:03:06tariffs Etc here this paper in
1:03:09particular is going to focus on
1:03:10information frictions or search costs
1:03:12and the idea behind these information
1:03:14frictions is that it's difficult for
1:03:16consumers to learn about either the
1:03:18existence of firms and different markets
1:03:21and other markets than their own local
1:03:23market
1:03:24um or you might know about the existence
1:03:25of the firms but you might not have a
1:03:27good kind of credible signal on their
1:03:28quality you might not know the quality
1:03:30of those firms right so in some way you
1:03:32have better information about kind of
1:03:33your very local firms and you know less
1:03:35about firms that are further away that's
1:03:36going to limit kind of the size of the
1:03:38market and therefore how much firms
1:03:40across markets are competing
1:03:42so you know as a result of these
1:03:44information frictions consumers are
1:03:46gonna
1:03:47um often buy almost exclusively from
1:03:49their local producer conversely right
1:03:51producers are going to sell mostly local
1:03:52consumers and that's going to have two
1:03:54effects the first is it's going to limit
1:03:56the consumer base that can um I'm not
1:03:59going to inhibit good firms from growing
1:04:01okay so the first is kind of a tax you
1:04:03know a negative impact on good firms
1:04:06right so they can't expand to other
1:04:08markets that means they can't grow they
1:04:10can't split economies at scale that's
1:04:12going to limit kind of the the first
1:04:13part why do good firms you know fail at
1:04:15the growth
1:04:16the second point is that it's gonna have
1:04:18limited competition from across markets
1:04:19right so the bad firms are not going to
1:04:21be competed out of the market right
1:04:23they're going to instead survive and
1:04:25that might lead to kind of lower
1:04:26aggregate productivity in the entire in
1:04:29the entire area and also lead to kind of
1:04:31cross cross market across firm
1:04:33productivity dispersion
1:04:36okay so this Jensen and Miller piece
1:04:38basically says you know when a
1:04:39technology comes in that facilitates
1:04:42search right that that essentially
1:04:44reduces some of these information
1:04:45frictions do we see better Market
1:04:47integration and does that increase
1:04:49Market integration basically lead to
1:04:51competition disciplining the market
1:04:53right so do we see a reallocation of
1:04:56profits and market shares towards the
1:04:58good firms and away from the bad and
1:05:00this is sort of you know a micro
1:05:01empirical example of a very classic sort
1:05:03of you know mallets model of the pro
1:05:05competitive effects of you know trade or
1:05:07increased Market integration
1:05:11so this paper is going to use the same
1:05:13shock that a previous paper robs use
1:05:15uses which is the rollout of cell phones
1:05:17across markets in India so let me just
1:05:20give you two slides and kind of that
1:05:21initial paper just so you can see kind
1:05:22of the Baseline setup so this is looking
1:05:25at the state of Kerala in India
1:05:27um along the coastline and one of the
1:05:29big Industries in this area is is
1:05:31fishing right
1:05:33um and so this first paper the 2007
1:05:35paper is using the role of cell phone
1:05:37towers along the shoreline uh in the
1:05:41state of Kerala as a shock to base the
1:05:43information friction so what's happening
1:05:45that first paper is he studying
1:05:46fishermen fishermen tend to kind of go
1:05:49out into you know to to hunt for fish
1:05:51that day they come back in a baseline
1:05:53without cell phones they tend to just
1:05:54sell to their local market now you know
1:05:57some fishermen are kind of Lucky and
1:05:58have good catches that day some have bad
1:06:00and so what happens at Baseline is some
1:06:02of these local markets have very high
1:06:03supplies some have very low and you see
1:06:06lots of variability in prices across
1:06:08markets
1:06:09once these cell phone towers are allowed
1:06:11it becomes much easier once you're out
1:06:12here to say well maybe I won't go to my
1:06:14home Market that day I'll go to the next
1:06:16one over if they've got low Supply that
1:06:17day so the reduction of this sort of
1:06:20information costs through cell phones
1:06:22allows fishermen to better Arbitrage
1:06:24they end up going to different markets
1:06:26and you see much smoother Supply
1:06:29as a result you see much smoother prices
1:06:31across Market so this is like one of my
1:06:32favorite all-time graphs of all of all
1:06:34time we should all be so lucky to find
1:06:36uh this sort of results but this is
1:06:38showing you basically what happens when
1:06:39the introduction of cell phone comes in
1:06:41so you see really clearly right so
1:06:43region one prior to cell phones tons of
1:06:45price variability cell phones come in
1:06:47boom law of One Price basically
1:06:50same thing for region two lots of
1:06:52variability cell phones come in boom
1:06:54we'll have one price same thing for read
1:06:55and free okay so just super clear in in
1:06:57the figure is what's going on here a
1:06:59really big change
1:07:01um and reduction in prices version but
1:07:03that's kind of the backdrop for this
1:07:04paper yeah so uh that was the first
1:07:07paper this paper is gonna it's gonna
1:07:09turn things around a little bit and
1:07:10think about now the fishermen as
1:07:12consumers
1:07:13and what are they consuming they're
1:07:15consuming boats
1:07:17so this paper is going to look at the
1:07:18introduction of cell phones and say what
1:07:20was the effect on a boat builder quality
1:07:22and firm selection
1:07:25so let me tell you what's happening at
1:07:26Baseline and then we'll talk about what
1:07:27happens when cell phones come in so at
1:07:30Baseline these fishermen are buying
1:07:31boats almost exclusively from their
1:07:33local producer okay 97 are buying from
1:07:36the the boat builder in their home
1:07:38Village and there's roughly one book
1:07:40Builder per Builder so along the
1:07:41shoreline you see each Village has their
1:07:43own build Builder and all the fishermen
1:07:45from that Village are buying from their
1:07:46local build Builder
1:07:49um now some of those boat Builders are
1:07:50really good and some are not so good and
1:07:52so there's lots of variation and quality
1:07:54across the boat Builders and they do a
1:07:55really fantastic job of doing four
1:07:57different measures of the quality et
1:07:59cetera okay they've got independent
1:08:00independent Auditors they've got
1:08:02self-reported Etc
1:08:04but no matter how you cut it lots of
1:08:05variability and quality
1:08:08now what happens how they're going to
1:08:09use a slightly different uh cell phone
1:08:11rollout so I want to just be clear about
1:08:12about that so they're going to use these
1:08:14again three regions
1:08:16region one is going to get cell phones
1:08:18in 1999. uh region two is going to get
1:08:21it in uh 2000 okay and Region Three
1:08:24doesn't have cell phones for the
1:08:25entirety of their study period so Region
1:08:27Three is like a peer control
1:08:29okay so what do they predict is going to
1:08:31happen right kind of all of this Mellows
1:08:33model so the idea here the causal chain
1:08:35and I'm telling you this now because the
1:08:37graphs are going to really closely kind
1:08:38of mirror each of these steps
1:08:40in terms of the results
1:08:42so first cell phones will cause
1:08:44fishermen to travel to different markets
1:08:46to sell their fish okay that was already
1:08:47shown in the first Jensen piece but
1:08:49we'll see it replicated here okay so
1:08:51fishermen are going to New Markets to
1:08:52sell their fish
1:08:53in doing so they're interacting with
1:08:56other fishermen right
1:08:58um and they're chatting and they're
1:08:59chatting about for example where did you
1:09:01buy your boat how you know how good
1:09:02quality is that boat builder they're
1:09:04getting information about non-local
1:09:06Builders so they'll learn about again
1:09:08first the existence of non-local
1:09:11Builders they didn't know about before
1:09:12um but they'll also potentially form a
1:09:14more accurate signal about the quality
1:09:16right so there's some natural
1:09:17variability inside the quality of the
1:09:19boats the main measure they'll have is
1:09:21kind of the longevity or how long the
1:09:23boats last there's some natural
1:09:25variability there you might need
1:09:26multiple sort of
1:09:28um uh people's assessments to get an
1:09:31accurate signal of the quality okay so
1:09:32they can Source
1:09:34um uh multiple estimates from lots of
1:09:36non-local fishermen and get a more
1:09:38accurate sense of the quality of these
1:09:40non-local builders
1:09:42okay
1:09:43once they kind of know who the best
1:09:45Builders are including non-local
1:09:46Builders fishermen will start buying
1:09:48boats non-locally and this will lead to
1:09:50Greater competition among the boat
1:09:52Builders and this is kind of the
1:09:53anticipated effect
1:09:55as there's more competition what should
1:09:57happen is that you should see the good
1:09:59Builders expand okay and the bad
1:10:01Builders contract or exit
1:10:03and then finally you know this might
1:10:05improve improve aggregate productivity
1:10:06in the industry and perhaps might lead
1:10:09to improved outcomes for consumers okay
1:10:11so this kind of the predicted causal
1:10:12chain let's now see sort of um uh what
1:10:15actually happens in practice and I'm
1:10:17going to actually just show you the
1:10:18graphs here because I think they really
1:10:19sort of speak for themselves
1:10:21so how do you let me just tell you sort
1:10:23of how you you follow these figures
1:10:24since we're going to see a couple of
1:10:26these so each column here is a different
1:10:28result so this is looking at for example
1:10:29the percent of farmers who sell fish in
1:10:32their own village
1:10:33this top row is showing you for region
1:10:35one which got cell phones here this blue
1:10:36line in 1999
1:10:38region two got cell phones here in 2000
1:10:40and Region Three is a pure control they
1:10:42didn't get cell phones during this
1:10:43entire period
1:10:45so
1:10:46um let me actually talk about first what
1:10:47happens in Region Three which is based
1:10:49on the control group right so you see
1:10:50basically a hundred percent of fishermen
1:10:52are buying uh or selling fish uh in
1:10:55their own village okay
1:10:58uh when cell phones come in what happens
1:11:00so uh fishermen start selling fish you
1:11:03know outside their home Village right
1:11:05um similar in region two again kind of
1:11:07no pre-trend but right when cell phones
1:11:08come in they start selling fish outside
1:11:10their own village so they are going to
1:11:11new villages to sell their fish the idea
1:11:13here is they then start talking to each
1:11:15other and getting a more accurate signal
1:11:17on the quality of the boat Builders so
1:11:20panel B is looking at that error and
1:11:22estimating the quality of boats in
1:11:24particular the boat lifespan in years so
1:11:26they had kind of these objective
1:11:27measures of boat quality they also ask
1:11:30fishermen to estimate and they take the
1:11:31difference as kind of the absolute value
1:11:33of the difference as they error
1:11:34so again let's go to Region Three first
1:11:36which is basically the control we see on
1:11:39average right the estimates for the
1:11:41non-local builders the errors are much
1:11:43higher so people have worse information
1:11:44about non-local Builders than they do
1:11:47kind of their local guy and again that
1:11:49kind of persists throughout this entire
1:11:50time period in the control group
1:11:52you see a similar thing in region one
1:11:54but what happens when cell phones go in
1:11:56well it looks like sort of slowly over
1:11:57time
1:11:59uh fishermen start getting better
1:12:00information about the non-local Builder
1:12:02and sort of by the end of the period it
1:12:04basically converges so they're sort of
1:12:05equally well informed the error rate is
1:12:07the same between local you know
1:12:09non-local and local builders you see a
1:12:11very similar thing in region two right
1:12:13again no no pre-trend but then as soon
1:12:14as cell phones come in again sort of
1:12:16this convergence of knowledge
1:12:18okay so now fishermen are better
1:12:19informed about the quality
1:12:22what does that do to where they buy
1:12:23their boats so again basically a hundred
1:12:26percent in the control group are always
1:12:28buying locally buying in their own
1:12:29village what happens when cell phones
1:12:32come in this drops and drops pretty
1:12:33rapidly right so you go for basically
1:12:35100 buying in their local Village to you
1:12:37know about 20 to 30 percent
1:12:40same thing here for the instruction of
1:12:42cell phones in region two so this is
1:12:44kind of the first part of the causal
1:12:45chain you're seeing basically getting
1:12:46better information and buying
1:12:47non-locally
1:12:50now what happens to firm's election
1:12:52so here we're looking at the total
1:12:54number of firms so a reduction in the
1:12:56number of firms implies firm exit
1:12:58and you see you know in region two
1:13:01there's basically a pretty steady number
1:13:02of firms up until mobile phones are
1:13:04introduced and then you see a pretty
1:13:06rapid decline in the number of firms
1:13:08so you know this is like almost a 60
1:13:10decline
1:13:12same thing for region region two
1:13:14basically sort of a steady number firms
1:13:16up until cell phones are introduced and
1:13:17then a lot of firms exit when cell
1:13:19phones come in and no change in Region
1:13:21Three
1:13:23now the the particular prediction is
1:13:26that this is going to basically lead to
1:13:27Greater exits among the low quality
1:13:30firms right as they get competed out by
1:13:32the high quality so they can break this
1:13:34up into above below median quality as
1:13:37measured by these Baseline
1:13:39um uh lifespan of the years of the boat
1:13:41and what you see is there's some
1:13:43declines among both but they're really
1:13:44concentrated in the below median quality
1:13:47firms okay so it's really kind of these
1:13:49below median quality and quality firms
1:13:50who are seeing the greatest exit
1:13:53that's when the first step of the causal
1:13:54chain you introduce cell phones
1:13:55fishermen start talking they get better
1:13:58information they start buying
1:13:59non-locally and as they do this is going
1:14:01to lead to the exit especially by the
1:14:03lower quality firms
1:14:06they also ask do we see any aggregate
1:14:09productivity gains right as a result of
1:14:11this reallocation so they're going to
1:14:12here look at aggregate production this
1:14:14is production by all firms aggregated at
1:14:16the regional level
1:14:17the look first is here there are changes
1:14:19in the number of boats sold they don't
1:14:20find any differences there so overall
1:14:22quantity of both sold isn't changing
1:14:25um but the lifespan of these boats is
1:14:27changing okay so if you look at Baseline
1:14:29versus n-line and region one you see
1:14:31contents rightward shifts or all along
1:14:33the distribution improvements in
1:14:35um uh in in the years of the boat same
1:14:38for region two these big increases not
1:14:40much of a difference in region in Region
1:14:42Three where there's no cell phones
1:14:45uh they do a bunch of other pretty
1:14:47clever things to kind of uh isolate the
1:14:49exact called the mechanism what's going
1:14:51on here that I really like and I'll just
1:14:52mention them briefly so you know you
1:14:54could ask is this increase in average
1:14:56quality is that coming from a
1:14:57reallocation of uh to firms of higher
1:15:00Baseline quality and sort of the story
1:15:01I've been telling so far you can also
1:15:03Imagine That firms internally are
1:15:05upgrading their quality and response to
1:15:07competition
1:15:08um they estimate this super simply by
1:15:10basically throwing firm fixed effects
1:15:11into the regression and that kind of
1:15:13isolates the part that's within firm
1:15:15quality upgrading in response to cell
1:15:17phones and they find basically nothing
1:15:18there so their takeaway here is this is
1:15:20mostly happening from reallocation to
1:15:22kind of Baseline higher quality firms
1:15:24rather than firms kind of upgrading
1:15:26their quality within their their firm
1:15:30they also find a reduction in in cost
1:15:32okay especially uh labor costs in
1:15:34particular
1:15:36um and they can ask a similar question
1:15:37is this due to a reallocation to firms
1:15:39with lower cost at Baseline or due to
1:15:41kind of reductions in costs within firms
1:15:43again they use those firm fixed effects
1:15:45and here they see kind of 50 50. so this
1:15:48is partially reallocation to lower cost
1:15:50firms so more productive firms at
1:15:51Baseline it's also due to a reduction in
1:15:54cost and firms and this reduction cost
1:15:56is coming from a pretty interesting
1:15:57place they find basically evidence of
1:15:59economies of scale so basically as these
1:16:02firms are getting existing firms are
1:16:04getting larger right the ones who stay
1:16:05in business are getting larger they're
1:16:07able to have workers specialize so
1:16:10rather than having a single employee do
1:16:12all the parts of the boat building they
1:16:14can have two employees one who does the
1:16:16first part one knows the second and they
1:16:17specialize so they see a fallen task for
1:16:19workers
1:16:20um and that suggests there's some
1:16:21economies of scale here
1:16:25the final thing is looking at consumers
1:16:27so what happens to sort of prices and
1:16:28quality over on the market they do see a
1:16:31slight increase in the price of votes
1:16:33about a four percent increase but they
1:16:35see a much bigger increase in the
1:16:37average quality in terms of the boat
1:16:39lifespan okay so they can basically do
1:16:41sort of um uh a price adjusted by the
1:16:45quality so a price per boat year of Life
1:16:47uh and this Falls quite a bit now they
1:16:50know this is sort of the short-term
1:16:51results but it does look like sort of
1:16:53overall consumers are getting sort of
1:16:54higher quality at you know um at lower
1:16:57prices per per kind of year of the boat
1:17:01okay so this is sort of I mean this is a
1:17:03really I think nice paper because it's
1:17:04super reduced form but it's basically
1:17:05kind of showing you all those causal
1:17:07steps of you see a reduction some of
1:17:09these frictions that might be inhibiting
1:17:11competition right meaning you're kind of
1:17:12in these very sort of individual
1:17:14fiefdoms of these markets you reduce
1:17:16these frictions you increase competition
1:17:18across markets between these farms and
1:17:20you see kind of the classic Force you'd
1:17:22expect you see basically the bad firms
1:17:24are exiting the good firms are growing
1:17:26and they're able to specialize and take
1:17:28advantage of economies of scale
1:17:30let me pause you before going to the the
1:17:32next set of papers are there questions
1:17:34on this paper
1:17:41hmm
1:17:43okay so we only got another oh please
1:17:45yeah I mean sorry I there is a question
1:17:48from the Q a that I think relates to
1:17:51several of the papers you're talking
1:17:52about so it's from
1:17:55um mes uh how do you see the role of
1:17:58intermediaries in competition and how
1:18:00may it alter the implications of
1:18:01competition that you're talking about I
1:18:03guess in general so far
1:18:06yeah so um I should say that in the
1:18:09Jensen piece they go through a lot of
1:18:11work to kind of describe how like there
1:18:12aren't intermediaries in their context
1:18:15um
1:18:16I mean I guess I think of
1:18:18um you know the kind of link between
1:18:19these two papers intermediate areas is
1:18:21you know if if there were intermediaries
1:18:23between these markets
1:18:25um you might imagine that you know one
1:18:26of the frictions that that Jensen is
1:18:28thinking about reducing is search cost
1:18:30but you can imagine another friction is
1:18:31you know in a more General sense not in
1:18:33this particular context but could be
1:18:34intermediary markups
1:18:37um and so if you're able to reduce
1:18:38intermediary markups right that is
1:18:40another way of basically integrating the
1:18:42markets more directly so it'd be
1:18:44interesting to see you know again I
1:18:46don't think that's the the equivalent
1:18:48producers in the intermediary you know
1:18:50uh equivalent of the Kenya the Kenya
1:18:53agricultural stuff is Farmers and I
1:18:54don't I suspect Farmers don't have a lot
1:18:56of Market power but imagine you had
1:18:58intermediaries in a market where the
1:19:00original producers had Market power you
1:19:02might find if they're a way to kind of
1:19:03reduce markups among intermediaries that
1:19:05would be like better kind of integrating
1:19:07these markets you might also find kind
1:19:09of Upstream effects of increased
1:19:10competition among uh you know among the
1:19:13Upstream firms so I think neither one of
1:19:15these examples is kind of perfect sector
1:19:16for that but I think you can imagine
1:19:18other ones where there are kind of these
1:19:19like Upstream uh impacts of reducing
1:19:22intermediary markups
1:19:27was there another question
1:19:35okay so let me just very quickly go
1:19:37through I'm just going to give you one
1:19:38slide in each of these last three papers
1:19:39I just think they're really nice
1:19:40examples of
1:19:42um you know experimental or
1:19:43quasi-experimental results looking at
1:19:45the impact of competition empirically uh
1:19:48in in low and middle income country
1:19:50contexts so um this first paper is a
1:19:53paper by Busan galliana and they're
1:19:55basically randomizing entry into
1:19:58um uh the grocery store Market in the
1:20:00Dominican Republic one thing that I
1:20:01think is particularly clever here is you
1:20:03might think how do you run an RCT where
1:20:05you randomize Grocery Stores
1:20:07um you know both how do you convince
1:20:09grocery stores to enter how do you
1:20:10literally physically build grocery
1:20:11stores well they're not um they're doing
1:20:14something very clever here which is they
1:20:15basically have uh um sort of these
1:20:18administrative rules around a
1:20:20conditional cash transfer program so let
1:20:22me try to tell you the back story so
1:20:23this is a conditional cash transfer
1:20:25program in the Dominican Republic It's
1:20:27providing transfers to poor households
1:20:29the critical feature for their paper is
1:20:32that these households are getting their
1:20:33cash transfer on a debit card can only
1:20:35be used at Affiliated grocery stores and
1:20:38so when they say they're randomizing
1:20:39entry what they're really doing is
1:20:40randomizing
1:20:42um entry of existing incumbent grocery
1:20:44stores into this program this Affiliated
1:20:46program so it's a really clever way I
1:20:48think of getting kind of randomized
1:20:50variation that otherwise might be quite
1:20:51difficult to get
1:20:53um so they induce the entry of of you
1:20:55know 61 small retail shops into 72
1:20:57different markets they um are even able
1:21:00to get variation in sort of the the
1:21:01number of new stores here on the
1:21:03Intensive margin
1:21:04uh and that you know overall find kind
1:21:07of a reduction in prices two to six
1:21:09percent as you expect more in areas more
1:21:11entrants
1:21:12um and they see some improvements in
1:21:14self-reported service quality level okay
1:21:16so I just mentioned to say you know
1:21:17there's lots of ways to try to if you
1:21:18want to think about experimental
1:21:19approaches to get randomized entry that
1:21:21don't literally have to be kind of you
1:21:23know building building new stores and I
1:21:24thought this was pretty clever
1:21:27um uh I want to talk quickly about sort
1:21:30of uh this is going to be sort of
1:21:31impacts on on quality
1:21:32um so this uh Berkeley night kiss at all
1:21:35paper is looking at um entry into uh the
1:21:38pharmacy industry in Uganda
1:21:40so the back story for this paper is
1:21:42there's
1:21:43um you know counterfeit medications
1:21:45actually a really big problem so these
1:21:46are don't think here of kind of you know
1:21:48the CVS uh or Walgreens that are you
1:21:50know regulated by say you know
1:21:52everything's regulated by the FDA this
1:21:54is a situation where pharmacies in
1:21:55Uganda are these very kind of small
1:21:57mom-and-pop shops that are sort of
1:21:58unregulated
1:22:00um and and counterfeit medication is
1:22:02really a big problem okay so in
1:22:04particular they're looking
1:22:05anti-malarials that are medications to
1:22:07treat malaria
1:22:09and so what they're showing this paper
1:22:10is um entry can matter and in particular
1:22:13the identity of the entrance whether
1:22:15they're high quality or not can really
1:22:16matter so they randomly assign the entry
1:22:19of sort of a high quality provider of
1:22:22anti-malarial drugs okay in this case
1:22:23they're working with an NGO
1:22:26um that is a retailer providing
1:22:28basically authentic drugs that are
1:22:30priced at below market prices
1:22:32and so what they find is that first
1:22:34incumbent prices drop by quite a bit uh
1:22:37when the entrant comes in okay this is
1:22:38suggested that that at least some of the
1:22:40incumbents were initially pricing you
1:22:42know above cost
1:22:43um and sort of I think most
1:22:44interestingly find based on the share of
1:22:46incumbent firms that are selling fake
1:22:48drugs drops by 50 okay so bringing in
1:22:50sort of a high quality firm
1:22:53um can mean that other firms are going
1:22:54to start to match and compete in quality
1:22:56this also suggests by the way that part
1:22:59of the issue you might have assumed okay
1:23:00there's lots of there's lots of
1:23:02counterfeit medication but maybe these
1:23:03small mom and pop stores don't know or
1:23:05have no control over their quality and
1:23:08this is a problem coming Upstream this
1:23:09is suggested at least part of the
1:23:10problem can be ameliorated by the
1:23:12downstream uh you know retailer Behavior
1:23:16and they find sort of these welfare
1:23:17effects you spoke to expect to see which
1:23:19is the overall use of anti-malarial
1:23:21drugs increased right people kind of
1:23:22learned that the quality is better
1:23:25okay so who the entrance is can really
1:23:27can really matter
1:23:28the final thing I'll mention is a
1:23:30Contracting piece which I think Rocco
1:23:31already covered so I won't you know
1:23:33cover in great detail but I just want to
1:23:34mention it because I just want to say
1:23:36that you know competition is not always
1:23:37a good thing
1:23:38um there can be much more nuanced
1:23:40effects depending on the broader context
1:23:42in which it's operating right so the the
1:23:44overall punch line of this paper is sort
1:23:46of you know in a second best world
1:23:47competition might not always be good for
1:23:50consumers um or in their case they're
1:23:51looking at monopsony power so it might
1:23:53not always be good for producers let me
1:23:56just give you the quick summary here you
1:23:57know this is an environment they're
1:23:59looking at Coffee farmers who sell to
1:24:02these Mills
1:24:03um in Rwanda and there's lots of
1:24:05relational Contracting going on here so
1:24:07there's actually it's actually a pretty
1:24:08complicated contract but the most basic
1:24:10part is that before Harvest these
1:24:12farmers are super poor and credit
1:24:14constrained Mills are providing inputs
1:24:17extension services which means basically
1:24:18just training and cash loans
1:24:21and the expectation the kind of
1:24:22relational part is they're going to
1:24:23provide that before the Harvest and then
1:24:26post Harvest farmers are going to sell
1:24:27back to the mill okay and that's how the
1:24:29mill is going to recoup their costs and
1:24:30also basically capture some local Supply
1:24:34um now greater competition among Mills
1:24:36that there's multiple Mills that farmers
1:24:38can sell to that's going to increase the
1:24:40urge to or the temptation to side fill
1:24:42that's kind of the the AG from it's
1:24:44specific to this but it's basically
1:24:45renegging on the contract right so
1:24:46they're going to have better outside
1:24:47options
1:24:49um it's more easy after after the
1:24:51Harvest to sell to another another firm
1:24:53that's going to make the Mills less
1:24:55willing to give inputs up front it's
1:24:57going to make this relational contract
1:24:58harder to sustain
1:25:00so another very clever way to get
1:25:02exogenous variation in competition
1:25:05um so they don't have randomized
1:25:06variation but they're going to use an IV
1:25:08approach they they sort of luck into
1:25:10finding this um old engineering model of
1:25:13kind of what areas were suitable for
1:25:15Mills that wasn't actually followed in
1:25:17practice
1:25:18um but I can tell them kind of How
1:25:19likely it would be to kind of place a
1:25:21mill in this area whether it's kind of
1:25:22appropriate given kind of the number of
1:25:25trees and and Road access and all that
1:25:27now of course the suitability is going
1:25:29to affect kind of local Supply so
1:25:31they're in a condition on that on sort
1:25:33of your own suitability for your own
1:25:35placement and then kind of in a donut IV
1:25:37they're going to look at kind of a
1:25:39radius around the mill and say you know
1:25:41were you surrounded by more suitable
1:25:43areas with the idea you're controlling
1:25:45for all the factors that might affect
1:25:46your local Supply
1:25:48um but you're going to get exogenous
1:25:49variation kind of how exposed you are to
1:25:51other Hills
1:25:52and so what I find they find Mills
1:25:54facing more competition you know first
1:25:55they have worse performance
1:25:57um some of these things like higher unit
1:25:59costs and lower volumes could just be
1:26:01scale effects right so as firms face
1:26:03more competition they have lower
1:26:05variable product uh profits if there are
1:26:07some sort of scale effects that might
1:26:09have higher unit costs
1:26:12you know what can't be explained by just
1:26:14a scale effects that's pretty critical
1:26:15is they find lower aggregate quantity
1:26:17supply to the Mills so all Mills
1:26:19basically in the area when there's more
1:26:21when there are more of them
1:26:24um they also kind of critically find you
1:26:26know fewer relational contracts right so
1:26:27those do seem to break down
1:26:29um and farmers can be worse off as well
1:26:31okay so they're seeing basically sort of
1:26:32they're selling less the Mills and
1:26:34they're seeing lower overall revenues so
1:26:36just say the effects of you know of
1:26:38competition can be quite quite nuanced
1:26:41let me conclude by just talking very
1:26:43brief about kind of opportunities for
1:26:44future research so first specifically on
1:26:46this question on competition uh in in
1:26:49the low and middle income country
1:26:50context so we've talked about lots of
1:26:52those challenges right so there's
1:26:53limited record keeping especially for
1:26:54these smaller firms
1:26:56um kind of the the the joint nature of
1:26:59kind of household-run Enterprises uh
1:27:01make it very difficult to separate
1:27:03supply and demand shocks in a lot of
1:27:04contexts there's also you know kind of
1:27:07this history I think of distinct
1:27:09methodological approaches from the
1:27:10development literature and from say i o
1:27:12for example
1:27:14um but that also means there's a lot of
1:27:15opportunities okay so I should say just
1:27:17this distinct methodology your ears
1:27:19should kind of perk up if you're a PhD
1:27:20student you know I think this is an area
1:27:22where I think lots more work can be done
1:27:25and I think in particular I think
1:27:26especially in development there's lots
1:27:28of unique opportunities okay so first of
1:27:30all we saw an example uh in that paper I
1:27:31have Michael you know you could
1:27:32experimentally estimate a lot of these
1:27:34key features from the standard i o
1:27:36approaches right so for example
1:27:37experimentally estimating demand that's
1:27:39going to allow you to you know first of
1:27:41all overcome this issue around kind of
1:27:43getting separate separating supply and
1:27:44demand chunks but even relative um which
1:27:47I think is critical in the developing
1:27:48context but even relative to kind of the
1:27:50general i o literature
1:27:52um you know you're going to get around a
1:27:53lot of those typical I think fairly
1:27:55strong and then identification
1:27:56assumptions um so there's lots of
1:27:58opportunity here for kind of
1:27:59experimental work in this space
1:28:01and then my kind of final broader pitch
1:28:03I want to make is that you know you can
1:28:05I think partially because many of these
1:28:08markets are so isolated and small
1:28:10um and certainly in the hall but in many
1:28:12kind of developing country context it
1:28:14means that if you want to do
1:28:15experimental work
1:28:17um you can actually get exogenous
1:28:18sometimes even randomized variation at a
1:28:21market level
1:28:22um in ways that I think are can be more
1:28:24difficult in uh wealthier and often more
1:28:26integrated markets right so you can get
1:28:28variation and entry and competition at a
1:28:30market level and that can be really
1:28:31useful
1:28:32um and so I want to just conclude by
1:28:34kind of that broader pitch of like
1:28:35Market level experimentation and I'm
1:28:37going to talk about rcts but I think you
1:28:39can also Imagine quasi-experiments as
1:28:41well
1:28:42um this Market level experimentation is
1:28:44useful not just for measuring effects on
1:28:46competition but really any outcome
1:28:48that's going to have a market level
1:28:49effect so trade cost migration
1:28:52um you know relevant to this course is
1:28:54really any intervention I think that
1:28:55you've talked about from the policy
1:28:56perspective trying to alleviate you know
1:28:59uh late Market constraints Capital
1:29:01markets drains you know all those all
1:29:04those interventions their impact is
1:29:05going to depend
1:29:07um on how firms in the market are are
1:29:10interacting right so even the place for
1:29:12example the classic case of microfinance
1:29:14you might see if you give a loan to a
1:29:16firm that they expand their revenues but
1:29:17is that is that sort of truly new
1:29:19productivity and production or is it
1:29:21business stealing from from other firms
1:29:23that's a really kind of classic question
1:29:24and so you know I want to just emphasize
1:29:27there's sort of been this recent push to
1:29:28do kind of this Market level
1:29:29experimentation again I'm talking mainly
1:29:31about experiments but you can imagine
1:29:33um uh thinking about natural experiments
1:29:35as well just to highlight a few you know
1:29:37randomized saturation migration
1:29:39incentives uh I've got some work random
1:29:41randomizing uh a reduction search cost I
1:29:44mentioned before kind of this this micro
1:29:46credit thing there's this really nice
1:29:47paper by Jing Kai and Adam Seidel
1:29:49basically randomizing the saturation of
1:29:51loans across firms and they do actually
1:29:53find a good deal of kind of this this
1:29:55business stealing happening
1:29:57um randomized saturation on cash
1:29:59transfer business training Etc so this
1:30:01is kind of if you note the dates and all
1:30:03these papers they're all pretty recent
1:30:04papers
1:30:05um I think this is a really exciting but
1:30:07new place that the literature is going
1:30:09um and so I would love to see your paper
1:30:10here next so um I got one more minute
1:30:13let me answer uh ask a question by
1:30:15Timothy
1:30:17Laura I'm just interested in your
1:30:18perspective on the most interesting
1:30:20thing to learn next whether why more
1:30:23firms don't enter or why more firms
1:30:25don't leave
1:30:29you know I don't know whether I'd say
1:30:31one is more interesting than other I do
1:30:32feel like we've focused a lot on the
1:30:34first question about why more firms
1:30:36don't enter
1:30:37um and not much about why firms don't
1:30:38leave
1:30:39um
1:30:40and that's also true I think from kind
1:30:42of the the micro friction I think a lot
1:30:44of that comes from that statement is
1:30:46really specific to the development
1:30:47literature and a lot of the development
1:30:49literature has been very focused on I
1:30:50think questions that feel more easy to
1:30:52tackle like you know questions where you
1:30:55can randomize loans to individual firms
1:30:57to see if that helps them to to enter
1:30:59for example
1:31:00um or potential potential firms
1:31:02um I think now more and more with kind
1:31:05of this Market level experimentation we
1:31:07can we can try to get variation of
1:31:09forces that might also discipline the
1:31:11market
1:31:12um and lead to kind of proper reading I
1:31:13should say there's a whole trade
1:31:14literature you know looking at this
1:31:15question that's done it kind of uh from
1:31:17a trade perspective and looking at kind
1:31:19of Market level outcomes but I think on
1:31:21kind of the micro frictions we've
1:31:23focus a lot on the question of entry and
1:31:25I think that's really been uh driven a
1:31:27lot by kind of what we can get a handle
1:31:28on methodologically and as these sort of
1:31:30new approaches kind of open up I think
1:31:32we'll be able to say much more about the
1:31:33latter
1:31:36thanks
1:31:38um I should probably stop here I know
1:31:39we're out of time I was I was requested
1:31:41to give you one more reminder
1:31:43um that the last lecture is uh is
1:31:45tomorrow and it is
1:31:47day you'll have to remind me the lecture
1:31:49for tomorrow is uh wrecking you Hass is
1:31:51going to be talking about industrial
1:31:53policy a nice round out to the whole
1:31:55course
1:31:57um thank you so much Lauren that was
1:31:58fantastic and we're so excited uh so
1:32:01many of you could join us and we're
1:32:03excited to uh finish the course tomorrow
1:32:05uh with another Blockbuster lecture so
1:32:07thanks again and see you all tomorrow
1:32:09thanks everyone
1:32:12thanks Lauren

About this transcript

This page contains the full transcript of Lecture 7: Markets and Competition with Lauren Bergquist by International Growth Centre, generated from the public captions YouTube serves with the video. The transcript has 18,132 words across 2,787 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.