YT YouTube2Text

YouTube2Text → Transcripts → Lecture 4: Misallocation with Pete Klenow

Lecture 4: Misallocation with Pete Klenow — Transcript

by International Growth Centre · 16,296 words · 2,582 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01foreign
0:32on my screen
0:34didn't switch to me that's weird
0:37normally it switched to whoever's
0:38speaking I wanted to see if it switched
0:40to you if you raise raise your hand
0:42um my mind says you speaking I know
0:48but I always I've just raised my hand
0:50and okay so what I'm speaking to so that
0:53may maybe let me but you popped up
0:55before you spoke so I think I think
0:57we're good
0:58okay
1:00so feel good about that
1:04good so I'll start then yeah we're all
1:06good yeah yeah so um my name is Dave
1:09Donaldson uh along with David Atkin
1:12we've been asked to convene a a PhD sort
1:15of level Zoom course via on the behalf
1:17of bread and igc about the topic of
1:19firms and development uh this is lecture
1:22number four previous lectures are online
1:25on the igc website uh recordings and
1:28slides and reading lists and uh
1:30encourage you to check those out if you
1:32haven't already but today we're
1:34delighted to have Pete CLE now he's a
1:36professor of Economics at Stanford
1:38University he's a giant in this field
1:39and he's very generously uh donated his
1:43time today to to contribute to the
1:44series uh he's going to be lecturing on
1:47the topic of misallocation which touches
1:49on themes we've already covered in the
1:52first lecture we um we talked about how
1:54at a broad level you can think about
1:57um individual firms becoming more
1:59productive learning new techniques uh
2:02learning new technologies getting access
2:04to those things and that's what the last
2:06couple of lectures have been about
2:07today we're pivoting in the course more
2:10towards the question of misallocation
2:12that is to say sort of firms getting
2:13access to inputs uh
2:16and since asking about the allocation of
2:18inputs to firms relative the relative to
2:21other uses like other firms and that's
2:24the broad theme that Pete's gonna sort
2:26of uh Usher us into starting today so
2:29with all that said Pete it's over to you
2:31and sorry just to remind you of protocol
2:32everybody um so panelists that is a
2:35randomly chosen people who've been
2:36invited to be panelists you're welcome
2:38to raise your hand uh at any time uh to
2:41the extent Pete um uh feels the time
2:44allows he'll he'll call on you to um to
2:47speak you'll be able to unmute yourself
2:48and ask your question out loud everybody
2:51else
2:52um feel free to write your questions in
2:54the Q a box uh to the extended time
2:56it'll allows we'll read out those
2:58questions for Pete to answer as well
2:59anyway take it away Pete thanks
3:02okay thanks very much for the
3:04introduction Dave and thanks for you and
3:06David for inviting me to participate
3:08this is um a great opportunity
3:10um okay so I want to talk about
3:11misallocation and specifically
3:14um my outline is I'm going to start with
3:16a definition and a simple model and then
3:19I'm going to give examples of things
3:21that I think are and are not starts to
3:23misallocation then I'm going to talk
3:25about five facts
3:27um that I think have already been
3:29documented in the literature and then
3:31three promising areas for future
3:32research I think people are working on
3:34right now and there's still plenty of
3:35room for more progress
3:37um and then at the end you know I have
3:39some stuff in the slides about data sets
3:42in case people are curious so probably
3:43we'll just skip over that
3:45um and then if there's time I'll talk
3:47about some select papers in the
3:48literature that I haven't talked about
3:49already that I particularly like and
3:51think are particularly insightful okay
3:54so let me dive in so what how would I
3:58Define misallocation this is my own
3:59personal definition I don't know if
4:01there's an accepted definition so I
4:04would describe it as
4:06um a misallocation of inputs of some
4:09form that could be people across
4:11occupations it could be Capital across
4:13firms or you know labor across plants
4:17within a firm there's any any number of
4:19um versions of that but one of the
4:22reasons I wrote this definition down was
4:24to try to clarify that it wouldn't be
4:27something like if you could repeal the
4:29laws of gravity and you could reallocate
4:31Goods across space more easily that that
4:33would be the fact that we don't do that
4:35wouldn't be a source of misallocation
4:37we're constrained optimal made perhaps
4:39in the allocation of goods across space
4:41given the physical constraints we have
4:43I'll elaborate a little bit more on that
4:45but so I I go even farther to say the
4:48planner must know something and see
4:51something so that they could tax and
4:53subsidize some activities
4:55um to reallocate inputs in a way that
4:57would be a Pareto Improvement but I
4:59guess I'd call this the potential Pareto
5:00Improvement
5:02um so one way to say this is deficiency
5:04of resource allocation is what all of
5:06Economics is about and that's that's
5:08true
5:09um although there's all it's also about
5:10maximizing social welfare at least in my
5:13view so misallocation is not so much
5:15about that that's like a separate
5:17question although I'll touch on that in
5:18terms of what I think are promising
5:19areas for future research
5:21okay so now move on to the simple model
5:26um so picture CES aggregation of
5:29different Goods I so y i is production
5:31of these different goods and there's a a
5:33fixed measure M of those or fixed number
5:35M of those and sigma is this elasticity
5:38of substitution and then you can define
5:40an ideal price index for this economy
5:42that's a generalized mean of of the
5:44individual prices and picture the
5:46production of individual Goods is just
5:49linear in labor just for Simplicity but
5:51you could add lots of different things
5:52to that Capital intermediate switch
5:54which are interesting but for the simple
5:56just to get fixed ideas just imagine
5:58linear production and imagine here's the
6:00mysterious part
6:02um imagine there's some firm specific or
6:04good specific I so these could be firms
6:06that produce these these Goods
6:09um so it could be firm specific uh
6:11Revenue tax rate that varies across
6:14firms the Y is for signifying its
6:16attacks on on output
6:19um and then picture the firm's
6:20monopolistic competitive meaning it
6:22takes input prices for its L is given
6:24takes the the price index capital P is
6:27given even the aggregate production
6:28which will affect its residual demand
6:30takes all those as given there's a fixed
6:32resource constraint and just for
6:34Simplicity not only is there a fixed L
6:36but there's a fixed number of firms
6:39okay so in this economy firms are are
6:42all due to charge the same markup if if
6:45the the towels which I'm now calling one
6:47over one minus the Revenue tax so it's
6:49like a grossed up version of this tax so
6:51a higher Revenue tax means a higher Tau
6:53they'd actually charge higher prices if
6:55they face higher Tau which makes sense
6:57um so it's as if they had higher markups
7:00but in this economy with cs demand and
7:02monopolistic competition the markups
7:04would be identical if if not for this
7:06Revenue tax if you want to call it that
7:08okay so one key thing that's emerged
7:10from this literature is this distinction
7:12it was first raised to my knowledge by
7:14Foster healthwire and Severson which is
7:16between revenue productivity and some
7:18online underlying measure of let's call
7:21it physical productivity or real
7:22productivity
7:24um and here Revenue productivity would
7:25just be Revenue per worker and in this
7:28CES setting this in the special
7:30assumptions of constant returns to scale
7:31which is production linear and labor and
7:34and CES you get that Revenue
7:36productivity does not reflect underlying
7:39uh efficiency this uh this a variable
7:43which we can call
7:44tfpq it doesn't reflect that at all
7:46there's like no relationship now in
7:48other models there'll be some
7:50correlation between tfpr and tfpq and of
7:53course if the Tau happens to be like
7:56like government's tax formal big firms
7:58more than they attack small firms then
8:00Tau would be positively correlated with
8:01tfpq but there's no
8:04um endogenous underlying reason in this
8:06simple setup why tfpr would be connected
8:09to tfpq and that's kind of a stunning
8:14a stunning uh feature or attribute of of
8:18this model that will generalize in the
8:21sense of even if you move away from some
8:23of these Stark assumptions you'll get
8:24the tfpr isn't tfpq and if you wanted to
8:27measure underlying efficiency you need
8:29something more than just Revenue per
8:31unit of input because that might reflect
8:33things more like wedges or distortions
8:35or markups which could be correlated
8:37with underlying efficiency but are not
8:39the same so that means if for example
8:41you were studying whether there's rapid
8:44learning by doing as a result of
8:45entering export markets you'd want to
8:48look at tfpq's Evolution rather than
8:50tfpr tfbr might be the the markups where
8:53they evolve as you start exporting and
8:55learn from exporting if there's no
8:56relationship between markups and that
8:58learning you might not see anything even
9:00though the underlying tfpq might be
9:02soaring as a result of learning by
9:04exporting okay so in this economy real
9:07aggregate tfp just output per unit of
9:10input is some generalized mean of of
9:13these underlying efficiencies but then
9:14there's this sand in the gears that the
9:16these wedges that are disrupting things
9:18and you can't see it by staring at this
9:21equation so I like to do the log normal
9:23case where it comes out really nicely
9:24and then aggregate tfp are the log of it
9:27is a function of the mean of these A's
9:29so a higher mean is good that makes
9:30sense in this case if Sigma is greater
9:32than one then dispersion in a is good
9:34and that's because implicit in this is
9:37that they're going to pile more L and
9:39the economy will pile more L on top of
9:41the high a Goods in production of IA
9:43Goods we have some substitutability
9:44Sigma the higher that substitute ability
9:47is if that's greater than one then the
9:49good better this variance is for the for
9:51aggregate welfare because you can pour
9:53inputs onto onto efficient Goods
9:57um and then if Sigma is less than one
9:58that's that's bad but we need Sigma
10:00greater meaning dispersion and
10:02productivity is bad because we can't
10:04substitute very well all away from the
10:06high productivity
10:08um I had the low low efficiency firms
10:09but a sigma is greater than one we can
10:11substitute really easy to the high
10:13efficiency firm that's where this term
10:14comes in which is not my focus my focus
10:16is more this last term which is saying
10:19the higher the variance of these taus
10:21these firm specific wedges or product
10:23specific wedges the more damage that
10:25does this negative means that's dragging
10:27down at tfp and so that is sand in the
10:31gears this equation is showing that so
10:33the more dispersion is bad in these
10:34wedges but then higher elasticity is bad
10:37too because the economy is going to
10:39respond more to these wedges so you can
10:40reallocate inputs more the higher a
10:43sigma so you'll get more Distortion out
10:45of it so this is like the deadweight
10:46loss triangle and it's bigger the bigger
10:48the the wedge and the bigger the
10:50response of the economy to the wedge in
10:51terms of these sigmas
10:53again this is a very special case
10:56um where things uh are very nice for
10:58example the correlation between Tau and
11:01underlying a doesn't show up here and
11:03that's a special property of a log
11:04normal that doesn't survive very
11:06generally
11:07um let me pause it looks like there
11:08might be some questions already
11:18by way Dave
11:20thanks just asking a written question
11:22from the Q a that has also intrigued me
11:26uh in the past too so how would you
11:29think about the word friction relative
11:32to something like wedge in this
11:33allocation if that's something you come
11:35to later obviously just feel free to
11:37shunt on it
11:39yeah okay so that's fair I think of
11:43distortion as my favorite term which I'm
11:45using wedge as a as a proxy for that and
11:48and I guess I don't care for friction as
11:51much because I'm worried that would
11:52confuse real frictions like say real
11:55adjustment costs of moving Capital
11:57across firms or real frictions and and
12:01reallocating workers they have to find a
12:03job etc those those are like gravity
12:06um that are constraints that are that
12:08that you know any planner would face
12:11they wouldn't want to subsidize the
12:14reallocation of capital ignoring those
12:16frictions those real frictions or
12:17subsidized reallocation so just
12:20semantically you know I personally favor
12:22Distortion over friction
12:24okay
12:26um all right so let me move on and say
12:28okay so what this very simple model what
12:30do we what do we get from it
12:32um well wedges can show up in tfpr not
12:35in tfpq tpr is not the same as tfpq
12:38which I've already emphasized aggregate
12:40tfp is discreasing decreasing and
12:41decreasing in the dispersion of these
12:43wedges aggregate tfp is unrelated to the
12:46mean wedge but just in the special model
12:47where there's say a fixed number of
12:49firms more generally the average wedge
12:51would affect things like entry and half
12:53efficiency effects and aggregate tfp is
12:55increasing the dispersion of tfpq
12:57because I mentioned that so one comment
12:58I wanted to make right here though is
13:01that this is a special model where cfp
13:03has and Tiff PR has nothing to do with
13:05tfpq as I've mentioned but
13:08um
13:08any model you write down say you wrote
13:11down uh non-constant elasticity of
13:14substitution so variable elasticity and
13:16or you had non-monopolistic competitions
13:18so you had endogenously varying markups
13:21you could do the same thing in that
13:22economy which is back out uh tethered
13:25being endogenous markup in that version
13:27so Sigma would have a sub I it's a
13:30function of these non-cs preferences or
13:31this non-monopolistic competition Market
13:34structure so you have a richer model
13:35with some Sigma eyes here those would be
13:38themselves distortions those are
13:40misallocating inputs for the standard
13:42monopolist produces too little arguments
13:44the ones with the high markups would
13:45produce too little but you could also
13:47back out a taui that would be the only
13:49Distortion you'd have an endogenous
13:51Distortion here you could back out the
13:53tower and you could back out the AI so
13:55if this setup seems really special it's
13:57actually not that special in the sense
13:59that you can generalize everything and
14:01still back out you know what the AIS and
14:04what the uh Tau eyes are the economy and
14:07there might be some correlation between
14:08them
14:09um between say the sigma I's the markups
14:12endogenous markups in the a so you still
14:14wanted to but you still want to
14:15distinguish the wedge that's distorting
14:17things from the underlying efficiency
14:19okay
14:21um so I think we get some insights from
14:22just this very simple setup
14:25okay so a little bit more on this
14:27measurement wise so the way I've done it
14:30the one question is well how do we get
14:31get information on this out real output
14:35um if you have an individual unit price
14:37which some data sets have I would
14:39actually call that
14:41um if you if you had a measure of
14:43physical output that's usually what they
14:45mean by
14:46um having having a unit price they might
14:49have an average unit price they might
14:50have some measures or quantities of
14:52production q and they can infer what I
14:54would call Process efficiency which is
14:56physical units of production divided by
14:57inputs
14:59um but this notion of of of the AI and
15:02that I wrote down that could be quality
15:04doesn't need I wrote it down as if it
15:06was process efficiency but it could
15:08easily be things like quality
15:10or if you're looking at a firm or a
15:11plant that produces multiple products it
15:13could also be the variety of products
15:14that produces and in the Cs framework
15:17they're isomorphic and other Frameworks
15:19they wouldn't be exactly the same but
15:20but conceptually this AI can reflect
15:23much more than just physical process
15:25efficiency
15:27um and then if you had if you knew
15:28something about the demand elasticities
15:30these could have Sigma eyes on them you
15:32can back out from a firm's Revenue it's
15:34um effective output including the
15:37quality and variety of what it's doing
15:38are plants effective output so I just
15:40want to distinguish my version of tfpq
15:42has things like process sufficiency and
15:45quality or even within within a variety
15:47in it and that's distinct from this
15:50other separate object which I think is
15:51also interesting which is this process
15:53efficiency this you can get from
15:55physical productivity data so if you
15:57know something about demand you can
15:58infer effective output so that can be
16:00really useful to get this broad measure
16:02of tfpq if you happen to have a data set
16:04where you have a quantity measure then
16:06you can get process efficiency which is
16:08also interesting but isn't the only
16:09thing we care about in terms of
16:10underlying a a technology which would
16:14have things like quality and Variety in
16:15it okay let me move on
16:18um this I've already mentioned ways to
16:20generalize this model in terms of the
16:22market structure and in the and the um
16:25and the demand and the elasticity the
16:28demand variable versus constant let me
16:30just talk about some easy
16:32generalizations to add fiscal Capital
16:33add intermediate Goods add wedges on all
16:37inputs you know so things like Financial
16:39frictions which might not be
16:40misalocation I guess by my my uh term
16:43that I used a moment ago they might show
16:46up
16:47um you could have those in as wedges on
16:49on all inputs
16:51um on any number of inputs you can add
16:54things like overhead costs and of course
16:55you could have things like entry costs
16:56that that make the M endogenous
16:58interesting
17:00okay so there are many other
17:01generalizations so Financial frictions
17:03I'm being inconsistent here because I'm
17:05using because I think I think that's
17:07there's some tension there in terms of
17:08financial frictions which is that
17:11um if you if you think about what the
17:13underlying source of the friction is
17:16some of those frictions might be
17:17information frictions uh or enforcement
17:20of contract frictions that get
17:22conditional on those the government
17:23can't do any better than the banking
17:25system other ones might be such that if
17:28the government just subsidized say firms
17:30that had high average product of capital
17:33or value of the marginal property
17:34Capital which is the more what we're
17:37really trying to get at and Dave and
17:39David were very helpful in their slides
17:41and stressing that distinction that what
17:43we're really caring about is the value
17:45of the marginal product of inputs and
17:47it's high at one firm and load another
17:48firm or one good another good the
17:51planner wants to reallocate so
17:54um depending on the exact source of the
17:55friction the government could improve
17:58upon
17:58um what the financial system does with
18:00certain targeted subsidies and and other
18:02frictions that can't these are all
18:04models where where
18:05um the particular underlying sources
18:07such that there is misallocation
18:10um not just you know imperfect markets
18:12okay so people have added adjustment
18:15costs um again those aren't necessarily
18:17things that um the market can improve on
18:19the planner would face those adjustment
18:21costs so these were more kind of like
18:23well maybe it's not misallocation maybe
18:25it's adjustment costs it's all talk more
18:26about those papers if I have time
18:29I already mentioned the variable markups
18:31monopolistic competition you can also
18:33add input output Matrix the work of baki
18:36bakai and fari is particularly uh
18:38clarifying in this regard and
18:39generalizing
18:41okay and I'll return to the um to their
18:43paper in a bit okay so that's the what I
18:46wanted to do with the outset
18:47um I know there's a raised hand so let
18:49me let me uh see what Brian wants wants
18:53to say
18:54yeah nice to meet you hey baby I was
18:58just wondering did you mention this
19:02there was kind of
19:05a role of the in the long term I mean I
19:08seem to remember that this the amount of
19:10misadocation or the amount of distortion
19:13depending on the Persistence of the
19:15functions in the persistence on the
19:17productivity so I'm wondering in this in
19:20this general setting these distortions
19:23that you are calling
19:25and somehow we endogenized if we can see
19:29if depending on the government actions
19:32they can increase or if they or if we
19:35can expect them to disappear in the long
19:37run
19:38okay so I think the point you were
19:41making about Maul's paper was that he
19:43was describing imagine two polar cases
19:46Imaging that firm specific shocks were
19:48IID and that there's a lag in adjusting
19:51Capital uh to the current productivity
19:53there's nothing that can be done
19:55if it's perfect I'm sorry uh permanent
19:59differences in productivity across firms
20:00then eventually even if you can't get
20:03external financing uh perfectly because
20:05of financial fractions or the nature of
20:08the banking system or the or the
20:10um incentive problems or repayment
20:12enforcement issues then firms would
20:15still self-finance an alarm so that's
20:17the point I just wanted to reiterate
20:19what you were saying that one of all's
20:20key points is if there was a ton of
20:23persistence in the economy in the long
20:25run would would reallocate capital on
20:27its own through internal savings so you
20:28wouldn't have misallocation so he was
20:30emphasizing you know it really depends
20:32on how much persistence there is so
20:34people have tended to find enough
20:35persistence that that there's still some
20:38Capital misallocation so it doesn't go
20:40away in the long run there's these
20:41ongoing shocks such that you at any
20:43point in time the financial system isn't
20:45reallocating enough you have this
20:46mismatch where some people have the
20:48wealth but not the good investment
20:50projects other people don't have the
20:51wealth but have good investment projects
20:53so that the presence of ongoing shock
20:55seems to be enough profit generate
20:56misallocation that persists
20:59having said that one of my facts that
21:01I'm going to talk about will come back
21:02to this about like how persistent these
21:04wedges are
21:06um they're a little too persistent to be
21:07adjustment costs or even maybe Financial
21:09friction it's a bunch of them
21:11okay thanks
21:13um let me see if there's any other
21:15questions before moving on to
21:17examples
21:19okay Ali
21:22hi I had a few questions but first of
21:26all the data used to investigate in this
21:28allocation are usually related to
21:30Industry section
21:32candidates data represent the entire
21:34economy
21:35uh second
21:37kind different source of installation
21:39and be prioritized for policy making is
21:43there a method to decompose
21:45misallocation such as measuring barriers
21:47to economic growth
21:50finally
21:51Tiffy has two factors efficiency and
21:54Technology apart from misallocation the
21:58literature about the efficiency Factor
22:00also includes other subjects such as
22:04unproductive activities and idle
22:07resources and so on and as far as I know
22:10no suitable no medical indicators has
22:14been introduced to measure them
22:16is it correct to attribute
22:22efficiency fluctuations to misallocation
22:24does efficiency emerge in a measurable
22:27way other than this allocation thank you
22:30okay so the first issue I think you
22:34raised that a lot of the data sets that
22:36we've used so far have been for
22:37manufacturing and I think that is a
22:39constraint that we have fewer data sets
22:42for things like the agricultural sector
22:44and especially outside agriculture
22:46manufacturing and those are quite big
22:48and often much bigger than manufacturing
22:49so I think that's the limitation so far
22:51of the literature it's also an
22:53opportunity for anybody any students
22:55listening to this that if you have
22:57access to a data set on the service
22:58sector in India or in Mexico I know
23:01there's a couple that I've I've seen
23:03some work on in those particular
23:04countries so it suggests misallocation
23:07if anything bigger in the service sector
23:09agriculture that's an ongoing debate how
23:12much misallocation there is in
23:14agriculture um but obviously really
23:15important for developing countries to
23:17study this in agricultural sector
23:18absolutely
23:20um policy I think I'm going to get to
23:22that in a little a little bit later
23:24you know the I think the modest humble
23:28thing I want to say is that that we
23:30don't have that we've nailed what these
23:32towers are and there's a lot of
23:34different things that can contribute to
23:35town I'm just about to give you some
23:36examples and without knowing exactly
23:39which policies we can't tell policy
23:41makers what to do so I think this is
23:42still mostly at the basic research stage
23:44that doesn't do justice to some papers
23:47out there for example
23:49um Atkin favor and Gonzales Navarro
23:51um studied what what um FDI and the
23:54retail sector in Mexico did to
23:57productivity and Welfare of different
23:58groups that has a pretty clear policy
24:00implication and you could say if there's
24:02policies that inhibit FDI those are you
24:06know generating some misallocation on a
24:08on in terms of
24:09um different types of firms so I think
24:12there's many exceptions to that but it's
24:13more like a hand-to-hand combat thing
24:15than individual studies talking about
24:17individual policies that can be reformed
24:19and improved I'm going to be taking more
24:21of this complementary approach Which is
24:22less policy less directly relevant for
24:25policy was to say how much room is there
24:27for misallocation how do we think about
24:29different categories of misallocation
24:31rather than talk about exactly the
24:33policy implications that that could
24:35follow
24:36um your third one about idle resources
24:39is really interesting maybe we can talk
24:41about that more offline I won't have
24:43much more much to say about that today
24:45okay so touching on a little bit on on
24:48what was just mentioned and I wanted to
24:50mention some things that aren't
24:51misallocation so I already mentioned
24:54um things like unavoidable adjustment
24:56costs or Transportation costs one hazard
24:59in this data is there's lots of
25:00measurement air so a lot of this
25:02variation if you back out the taus or
25:05even the tfpqs a bunch of it is coming
25:07from just mismeasurement of the revenue
25:08and inputs and that's
25:10um there's some ways to try to tackle
25:11that but that's that's like a major
25:13caveat I think
25:15there's also things like if you back out
25:17and looks like a firm has a high
25:20um value of marginal product
25:22um of say Capital
25:24um maybe that's coming from risk maybe
25:25that's the plan or wouldn't want to want
25:27more capital and labor allocated to that
25:29activity because it's risky relative to
25:31other ones and that's a perfectly
25:32functioning Market would do that
25:33similarly with labor there could be
25:35compensating differentials so that the
25:37value of the marginal product of labor
25:39lists low in one activity but that's
25:41because workers are getting lots of
25:43amenities to being there and some will
25:45have high values of marginal products
25:47and and uh that's because of
25:49disadvantages
25:50um so the market might be functioning
25:52fine you wouldn't gain you know welfare
25:54sense from reallocating
25:56inputs if they reflect compensating
25:58differentials and then kind of a general
26:00thing is you know
26:02any any any specification of a model in
26:06backing out of an A and A tau is going
26:08to be subject to if those if the model's
26:10been specified in terms of preferences
26:11Market structure Technologies and that
26:13includes the production function so one
26:15big category I think there's been less
26:17work on than there should is
26:18heterogeneity and production
26:19elasticities say across plants or across
26:22firms with within Industries there's
26:24bound to be a lot of that and we
26:26wouldn't want to mislabel
26:28um that as misallocation of capital or
26:30misallocation of Labor and that's really
26:32important of course distinguishing the
26:34misallocation from these technology
26:36differences
26:37um in terms of inferring gains that we
26:40had for misallocation potentially policy
26:42implications
26:43okay so you know what do I have in mind
26:45well I I literally talked about a
26:47Revenue tax those do differ in things
26:49like subsidies they differ across firms
26:51across Industries across um sizes of
26:54firms
26:55um so in subsidies as well
26:57um there's things like size dependent
26:58regulations that that are kind of an
27:00effective Shadow Tau that are hitting
27:02some firms more than others
27:04um the markup differences are so these
27:07might sound very Republican saying you
27:09know these wedges are are government
27:11induced and and but these are like
27:13market failures so there can be
27:15government policies that are that are
27:17generating the distortions and they
27:19could be Market imperfections that are
27:21generating these distortions so markup
27:23differences wage markdown differences
27:25and growing interest in that I think
27:27most of that work is is in developed
27:29countries so I'd like to see
27:31um a lot more work in in developing
27:32countries I'm not a micro development
27:35person so maybe more than I realized but
27:37from the macro development literature
27:38that I know a little bit better there
27:39seems to be a lot of opportunity to
27:41study markdown differences more foreign
27:43development countries
27:45okay
27:46um so there's many other potential
27:49sources I mentioned Financial frictions
27:50but you can also think of state-owned
27:52banks you can think of just cronyism
27:55um you can think of things like
27:57discrimination and I'm going to give an
27:58example of that and and that can be
28:00occur in many different forms
28:02um that could generate misallocation
28:04um things like Efficiency wages when I
28:07was first taught that I thought that was
28:09um one of those things the market
28:09couldn't improve on because if there's a
28:11monitoring problem and workers
28:13um need to be paid Efficiency wages in
28:16order to induce uh more effort then how
28:18can how can the planner improve upon
28:21that but it turns out if the planner
28:22knows that certain firms or sectors uh
28:26face more monitoring problems than
28:28others then they can actually subsidize
28:30those sectors and increase welfare so
28:32that's a category that I mean it's a bit
28:34of an outdated literature but I don't
28:36know why exactly it seems like it should
28:37be revived there's all kinds of things
28:40like land use restrictions these are
28:41more like policy related
28:43um and then you can think of this is a
28:45sense in which this is like all of
28:46Economics is there misallocation is
28:48there is there the right allocation of
28:49resources so you can think of
28:51misallocation of r d and infrastructure
28:53and that's one way in which
28:55you know if you think about
28:57um saying oh there's there's
28:59Transportation costs and therefore that
29:01you can't call that misallocation that's
29:03generating wedges across space but if
29:05there's a government Investments and
29:07infrastructure of sub-optimal then
29:09that's a form of misallocation that will
29:10reduce uh those wedges and and may
29:13generate welfare gains and of course
29:15there can be externalities of all kinds
29:16that as well known generate
29:18misallocation I think I'll see if
29:20there's another question now there's
29:22some popping up
29:32or not
29:33um feel free to raise a hand at any
29:35point
29:36um okay so one paper to 10 okay it looks
29:40like there's uh somebody raised their
29:43hand hi yeah there's one question that q
29:45a chat they got three upvotes uh okay on
29:49metallication and competition says is
29:52there a conflict between policies trying
29:54to reduce misallocation and increase
29:55competition if so how can we resolve
29:58this for example could we reduce
30:00misallocation by giving all resources to
30:03a few of the most productive firms or
30:05would that reduce competition with the
30:07usual negative effects
30:09I mean that that's really interesting
30:11um so yeah I mean certain models like
30:15variable markups like there's a nice
30:17paper by
30:18um Boren Michigan on this Karina Bora
30:21and virgila Midge again that talks about
30:22attention between you know you mentioned
30:25competition but also uh just inequality
30:28that if you if the most efficient
30:31entrepreneurs are most productive firms
30:33are owned and you don't have completely
30:34Diversified ownership then subsidizing
30:37those so if they have high markups
30:39they're not going to produce enough so
30:40if you solve that by subsidizing them
30:42you're going to subsidize the most
30:43profitable wealthy entrepreneurs owners
30:46Etc so but you mentioned competition but
30:49I think the idea is
30:50hey uh by just subsidizing them you get
30:53them to produce the right amount so
30:56um you're already pushing against the
30:57the market uh markups being higher at
31:01those big firms and so even though
31:03you're making competition even less the
31:05subsidy is trying to offset the
31:07Distortion from that so I think there's
31:09a trade-off there not only with with
31:11having to think about its endogenous
31:13impact and competition which you
31:14mentioned but also on inequality and
31:16I'll return to that if I have time
31:19I think that's a a highly relevant so
31:22there are other things that where
31:24misallocation might be increasing
31:25inequality and this is an example of
31:27that
31:28um this is a work I did with Chang Tai
31:30Shay a frequent collaborator uh Eric
31:32Hurst and Chad Jones and we were just
31:35pointing out something that is an
31:37voluminous literature on which is that
31:38there's been some occupational
31:40convergence across skill that high
31:42skilled occupations which you could
31:44measure as high wage you could measure
31:46as high education
31:47um or dominated by white men more in the
31:50past than reselling so still inequality
31:52and if you did this for economics it
31:54would be more Stark we just did white
31:55versus black because we could those
31:57categories were consistently measured
31:59back in time we wanted to go back or far
32:01away but there's many other categories
32:02and more recent data that where you
32:04could break things out
32:06and there was you know
32:08disproportionately white men in these
32:10high skilled occupations in 1960 still
32:12disproportionate but the ratios were
32:15were um far less us unequal and so we
32:19argued this was misallocation or was
32:20consistent with misallocation that
32:22there's talent to be a lawyer doctor
32:24Economist that was distributed the same
32:27on in men versus women white versus
32:29black and so these numbers should be in
32:31an efficient economy with no
32:32misallocation should be should be equal
32:35and they're more equal now so maybe
32:37there's less misallocation of talent now
32:39than there used to be and we've tried we
32:41tried to quantify that and
32:43so he said that these patterns
32:45consistent with falling discrimination
32:46can generate gains if people have
32:48comparative advantage and there's people
32:50who were born to be you know with the
32:51ability to be doctors more than better
32:53than to be nurses but if all men were
32:56the doctors and all women were the
32:57nurses that's going to misallocate a
32:59bunch of talent so so we said it may
33:01have accounted for this reduced
33:02discrimination may have recounted for a
33:05significant chunk of growth in GDP per
33:07capita Less in in GDP per worker because
33:10a bunch of the gains we think came can't
33:12took the form of increasing labor force
33:14participation of women so this gap
33:16between the 25 and 40 is it may have
33:18contributed to Rising GDP more than
33:21Rising GDP per worker because a bunch of
33:23it was increasing workers per capita
33:24there's a question of whether all those
33:27gains have been realized obviously not
33:28all of them but how big are the
33:30remaining gains
33:31um and the gains might be big you know
33:34an example is a student of mine John
33:36Felix who has a paper on on women
33:38inventors arguing that that uh that
33:41misallocation is still quite large and
33:43still large gains to be had from
33:46equalizing
33:48you know opportunities and access to
33:50becoming researchers by
33:52um a gender okay and that's in the U.S
33:55in that study arguably that's going to
33:57be even bigger in other countries
33:59okay
34:00um so anyway this is just a nice example
34:02of illustrating that even though a lot
34:04of my discussion will be about firms and
34:05plants
34:07um or for that matter manufacturing that
34:09these Concepts can be applied to
34:12um not just non-manufacturing firms and
34:14plants but other aspects like like
34:17allocation of people across occupations
34:20okay
34:21um now I want to say you know there's a
34:23literature on this what are some broad
34:24facts that have been uncovered
34:26um some patterns that seem to be common
34:28across a number of the studies and then
34:30certainly have I want to have to make
34:31sure I have time for
34:33um promising directions for future
34:34research
34:36okay so what are these five facts so
34:38this lists them but I'm going to talk
34:39about them in some details so let me
34:41just go straight to that next the first
34:44one is that tfpr this measure of Revenue
34:46per unit of inputs is quite dispersed
34:49across
34:50um plants within Industries so uh this
34:54is the U.S distribution
34:56um and the one hazard in this is that
34:59the samples aren't quite the same so in
35:01the US this is all Employers in
35:03manufacturing in China this is that
35:06initial plants and China these are
35:08actually firms
35:09um and they're firms above a certain
35:10size threshold picture about you know 25
35:13workers and India same thing it's above
35:16a size threshold that's somewhere
35:18between
35:19um
35:20you know you know somewhere around 25
35:22workers depending on the year so and it
35:25and they're in India they're closer to
35:27plans
35:28um so we have some apples and oranges to
35:29these comparisons but I would argue that
35:32you know to the extent that subsequent
35:33literature other than this picture has
35:36has brought in the informal sector of
35:39smaller firms smaller uh in India
35:41they've gotten even more dispersion so
35:43so and I would argue probably the same
35:46would be true of China so the ways in
35:48which these data sets are inconsistent
35:49probably understates the difference in
35:51dispersion between China India and the
35:53US here and you do see even despite
35:55these these uh discrepancies in the
35:57sample that the U.S distribution is much
35:59less dispersed there's a ton of
36:01dispersion in the US which is
36:03interesting so you could just look at
36:04that and say that's consistent with a
36:07lot of misallocation
36:08um because you know these are
36:09differences of a factor of four in
36:11Revenue per unit of input so you have to
36:13ask where those where those coming from
36:15are those are those real are there
36:17measurement error which case they're not
36:18real gains to be had
36:20um are they adjustment costs which case
36:23you can't necessarily improve upon it or
36:25they do things like markups which would
36:27mean our markdowns which would be
36:29massiveness allocation okay
36:31um so but the fact that people found
36:33very consistently is kind of the Richer
36:36countries tend to have narrower
36:37distributions of tfpr and poorer
36:39countries have wider distributions
36:41um so that's consistent with distortions
36:44being bigger and and developing
36:46countries and we're dragging down
36:48aggregate tfp in developing countries
36:49and maybe contributing to the to why
36:51they're underdeveloped relative to say
36:53oecd economies
36:55okay
36:57um so if you say how much you know if
36:59you try to translate this into this if
37:01you use some of the like the simple
37:03model there was a formula that came out
37:05in terms of how tfp is related to the
37:07variance of tfpr and you can generalize
37:09that to non-normal distribution to to
37:12include things like capital and labor
37:14separately
37:15you do that in China in 1998 compared to
37:18the US in 1980 98 India 1994 compared to
37:21the US in 1994. you get in manufacturing
37:24these gaps aren't as big as for the
37:26whole economy so maybe agriculture is
37:27playing a disproportionate role here so
37:29if you're if you're wondering why U.S
37:31has only two to three times higher tfp
37:34than China and India in those years
37:35that's because it's looking at
37:37manufacturing and the differences in
37:39allocative efficiency taking these tfbr
37:41differences naively is not reflecting
37:44measurement Air at all are are big so
37:46they're somewhere about a half of the
37:48story and and uh us and China in the log
37:51space and a third of the story in India
37:54so it looks like there's room for a
37:56bunch of misallocation that's actually
37:57happening here that might be
37:59contributing to these differences in tfp
38:01in a non-trivial way
38:03and that's interesting to see what
38:04happened over time and in India they
38:07didn't go down over time if anything
38:09over this sample we chose this
38:11particular sample in this study that I
38:12had with Chang because we had a
38:14comparable definition of Industries here
38:16and there were a bunch of reforms in the
38:18middle of the sample for India so we
38:20were surprised we thought that part of
38:22India's rapid growth in recent decades
38:24would have come from
38:26um reducing things like licensing
38:28restrictions that would cause uh
38:30reallocation that would mitigate
38:32misallocation and cause aggregate cop to
38:34grow so it's interesting that that we
38:36find it actually got worse over this
38:38this sub-period over longer periods with
38:41more recent data it's more flat
38:43um but again surprising that it hasn't
38:45improved
38:46and then in China it's important
38:48misallocation seems to have fallen about
38:50two percent a year over this the sample
38:52and if you look at actual tfp growth
38:54it's only about a third of actual tfp
38:56growth from China and this this
38:58interpretation in India looks it's kind
39:00of interesting because if you think
39:01about why did they have such low top
39:03growth maybe one of the reasons was was
39:05this this is surprising too because I
39:07would have thought tfp would have gross
39:09would be a big source of of India's
39:10rapid growth
39:12okay
39:13um and then you can ask you know where
39:15is this coming from this is probably the
39:16closest thing I'll say to a directing
39:18directly relating these misallocation
39:21measures to policy if you look at State
39:24on Enterprises which shrank is a share
39:26of the economy and their tfp gap between
39:28surviving state-owned Enterprises and
39:30private Enterprises that shrunk so they
39:32privatized the most inefficient ones or
39:35they exited most inefficient State on
39:37Enterprises with the lowest tfpr and
39:39then the ones that survived
39:41um they exposed them more to to Market
39:43pressure and so their tfp our Gap so
39:45they might say they reduce their
39:47subsidies and so the tfpr Gap shrank and
39:51that's of the say 15 increase in tfp
39:54that we estimate from reduced
39:55misallocation about 40 percent of that
39:57six percentage point to the 15 seem to
39:59come from state-owned Enterprises so
40:02again we all know that there was a shift
40:04away from state-owned Enterprises we can
40:06imagine that increased productivity
40:08what's useful about this is it's trying
40:10to quantify exactly that contribution
40:13um so it's interesting that it looks
40:14looks big but not the whole story of
40:16China's miraculous growth
40:18okay let me
40:20um okay let me say one more thing before
40:23opening up questions some more questions
40:24which is so
40:27Chang and I studied China and India
40:28versus the US people have studied other
40:31um other countries that in American
40:34Development Bank had a study of many
40:35Latin American countries Michael Peters
40:37studied in manufacturing uh Diego
40:39restitution collaborators have done a
40:41lot of studies for at the agricultural
40:43sector in places like Malawi and China
40:46and elsewhere that I'll touch on if I
40:48can later
40:49and it's pretty consistent this finding
40:51that misallocation is is large it seems
40:54to be larger in developing countries
40:55than in uh developed countries so that's
40:58why this is probably their most robust
40:59fact out there
41:01um these are like 37 countries some
41:03countries are studied repeatedly in
41:05different years here so it's not
41:07completely independent just to try to
41:09see what's the broad pattern and
41:12um if you estimate a line through this
41:14this is relative tfp compared to the US
41:15the US is one and and the US doesn't
41:18have perfect allocative efficiency so
41:19the US has had like something like 75
41:21percent this allocative efficiency is
41:23the inverse of the missile the gains
41:24that you could achieve for your
41:25misallocation so if your point if you're
41:2750 allocative efficiency you could
41:29double output by reallocating capital
41:31and labor potentially in this naive
41:34interpretation of the data with no
41:36measuring there
41:37so this is showing it's not perfectly
41:39correlated with with tfp across
41:40countries but it's somewhat correlated
41:42so if you run a regression you get an
41:44elasticity of 0.4 you know not a you
41:47know some noise in that so you one
41:50Temptation would be to say maybe 40 of
41:52the differences in tfp could come from
41:54differences in allocative efficiency
41:55what's if true would be huge
41:57so many caveats in that the T this
42:00measure misallocate allocative
42:01efficiency is for manufacturing except
42:03for a few studies of Agriculture and
42:06some of services there
42:08um and then uh the sample frames differ
42:12which I already mentioned that's that's
42:13a problem these use one heroic Sigma
42:16it's arguably too low which would mean
42:18the misallocation is being understated
42:20but there's probably a lot of
42:21heterogeneity in this across sectors and
42:23so you'd want to incorporate that in a
42:25more careful study
42:27um I'm taking studies off the shelf and
42:28they all they all happen to use the same
42:30Sigma measurement air could be worse
42:32than poorer countries and again too few
42:34studies outside outside Manufacturing
42:36okay this is a study across Mexican
42:39states that this has misallocation
42:42rather than allocative efficiency and
42:44it's saying the states with higher GDP
42:46per capita have less dispersion in tfpr
42:49across
42:50um firms and that's associated with less
42:52misallocation in the Richer richer state
42:55so again some broad suggestive evidence
42:57that this contributing in a in a big way
42:59not just these country about country
43:01studies calculating potential gains from
43:03this allocation but but observing that
43:05looks like misallocation might be bigger
43:07in developing countries and contributing
43:09in a sizeable way to the income gaps
43:11okay
43:13um maybe I'll stop now for pause now for
43:16a second for for questions
43:31all right uh Cincy I don't know if I'm
43:33pronouncing your name right but yeah so
43:36I have a question on the production
43:39function and electricities do you think
43:41it makes sense to borrow some like the
43:43new message from IO literature on
43:45production function estimation so sort
43:47of a more clearly uh look into this kind
43:50of measurements uh related in the
43:53medical development literature oh no
43:57ABS absolutely I think you're you're
43:58right there's this burgeoning literature
44:00on production function estimation
44:03um that's made a lot of progress
44:05um that it could be there's an Arbitrage
44:07opportunity which is to take that and
44:10say to what extent is that generating is
44:11there heterogeneity in these production
44:13elasticities
44:15um so that you know some of this
44:16dispersion in let's say the value of the
44:19Mars the marginal product of capital
44:21versus labor that that you would get
44:23under constantly less assumption is in
44:25fact just the heterogeneity and the
44:26other sister so it would narrow I would
44:28think would narrow the
44:29dispersion in the value of the marginal
44:32products and the question is how much
44:33and Country by country so that's that's
44:35an opportunity yeah I mean one one
44:37comment I wanted to say about that
44:39literature because I'm not a direct
44:40contributor I'm become more of a
44:41consumer uh I tend to use like just cost
44:44shares but you need constant returns to
44:46estimate production elasticities but you
44:48need two strong assumptions there at
44:50least
44:51um you need uh competitive input markets
44:53with maybe three it's wrong you need
44:55constant returns to scale and you you
44:57need some measure of the cost of capital
44:58so but that's the way I I go at it
45:02um and but I'm a consumer of watching
45:04what's happening in these production
45:05function elasticity papers and one thing
45:08I was going to say is I really like the
45:09papers that lay out the environment and
45:11also show the environment in which their
45:14estimator is consistent and also kind of
45:15demonstrate that it works you know like
45:17in other words if you generate data from
45:20that model that the estimator comes up
45:22with the the true production
45:24elasticities and the data generating
45:26mechanism so that's just some good
45:27practice that I've seen in some of the
45:29papers and really admire
45:31okay
45:32um one comment I wanted to get back to
45:34you know size of misallocation might be
45:38big this study is asked is making a
45:41really interesting point this is Baka
45:43and fari again which is you know what
45:45Chang and I were often asked this
45:47question we didn't have a good answer
45:48which was I thought from the classic
45:50work of Arnold Harvard or the deadway
45:53Lost triangles were small the cost of
45:54Monopoly were less than one percent of
45:56GDP the fact that some Industries or
45:58some firms are monopolists and other
45:59firms weren't that was just trivial that
46:01was just second order
46:03um so Buckingham farhi point out you
46:05know they get say 20 misallocation and
46:08uh in the U.S and that's not just
46:10manufacturing by the way they're looking
46:11at copy stat firms many majority of
46:13which aren't and they did a really nice
46:16thing which was decomposed how they're
46:18able to get a number that's 100 times
46:20bigger than Hardware's number and one of
46:22the reasons was if you look within
46:24Industries the wedges are bigger than if
46:26you look across the industry so if you
46:28look like electricity versus
46:31um manufacturing the average wedge
46:33differences aren't big but if you look
46:34within uh manufacturing you might see
46:37big differences in markups and same
46:39thing with
46:40within electricity provision you might
46:42see massive differences in markups so
46:43there's much bigger uh wedges within
46:47than across Industries and critically
46:48remember the losses due to misallocation
46:50are increasing in substitutability so if
46:53Harbor was focusing more on most on like
46:55electricity versus manufacturing
46:57factoring there's not that much
46:58substitutability between them but
47:00there's a lot of substitutability
47:02between different manufacturing firms or
47:04products and so any distortions can do a
47:07lot more damage so the wedges the
47:08distortions are bigger the damage to
47:10distortions are is bigger because more
47:12substitutability and then they get a
47:14bunch of propagation from the input
47:15output net Network as well which
47:17obviously relates to a lot of their work
47:19um Jim Schmitz has a bunch of work on
47:22this that's it's even other dimensions
47:25in which Monopoly can do damage you know
47:28by limiting entry by limiting the
47:31technology used by competitors and on
47:34and on so I think he's on to lots of
47:37channels by which monopolies might be
47:39doing during damage okay
47:42um
47:43I think this was mentioned by Dave and
47:45David and their introductory lecture
47:47there's more tfpq dispersion that's kind
47:50of a corollary in developing countries
47:52so I'll skip over that they also
47:54mentioned that the Distortion seem to be
47:56size dependent this is one version of
47:58that this is tfpr in a plant relative to
48:02the industry mean in India and this is
48:03the tfbq of that plant so you can see a
48:06very strong positive relationship you'd
48:08also get this with measurement error and
48:10revenue or input so that's that's a big
48:12caveat to a plot like this but one thing
48:14that's interesting is this is much
48:16steeper in rich countries than in much
48:18flatter and I'm sorry much deeper in
48:20developing countries and flatter uh but
48:22still positive and and rich countries so
48:25as much as people are talking about big
48:27firms have high markups in the U.S
48:28there's evidence consistent with even
48:30stronger relationship in places like
48:32India okay
48:35um one interesting side aspect of this
48:38is that if you think about distortions
48:40rising in
48:41um productivity that might affect
48:43incentives Dynamic incentives to invest
48:45in productivity so Bento and rustucia
48:47have a really nice demonstration of that
48:49in their AJ macro
48:51um also I'll actually Alpine Peters have
48:54a nice example of how this firms won't
48:56replicate themselves into more units if
48:58there's a they had in particular
49:01enforcement uh problem that meant that
49:04an entrepreneur couldn't delegate start
49:06opening up new
49:08um new establishments and so firms would
49:09be smaller so that would that might
49:11apply even though they're
49:13um the data they used was for
49:15manufacturing that could apply easily to
49:17non-manufacturing where
49:18multi-establishment firms are quite
49:19common things like retail and services
49:23um and then of course entry can affect
49:25entry too
49:26um there's nice papers on that
49:28okay
49:29um so this is a version of this which is
49:31what how does tfpq this is average
49:34employment so it's an input but if you
49:36the implicit thing that we looked at
49:39explicitly in the paper but this is just
49:40a more transparent way to show it is to
49:42say average employment if you normalize
49:44the youngest firms to one and then look
49:46at older firms they're much bigger in
49:48the US kind of they grow in Mexico and
49:50then India very little growth and if you
49:53construct a tfpq measure it looks like
49:55this as well so it looks like tfpq Rises
49:57more with age in the US than in Mexico
50:00or India
50:01people have filled in more countries
50:03since then one interesting thing is that
50:05a lot of countries are just here in the
50:07middle so the U.S looks like an outlier
50:09even relative to place like Germany and
50:11India looks like an outlier on the low
50:13side even you know like relative to
50:14Mexico and many other countries that are
50:16in here so but this illustrates that
50:20um misallocation can have these Dynamic
50:22implications that affect firms
50:23incentives to grow large as they they
50:25age and it could be to what extent
50:27they're opening multi
50:29um multi-establishment locations or
50:30improving uh their process efficiency or
50:33the quality or variety of their products
50:35as they age
50:37okay
50:39um all right so a third fact so the
50:41first fact was that
50:43the distortions appear bigger in
50:45developing countries
50:46and that contributes potentially sizably
50:49to the underdevelopment the second fact
50:52was that
50:53um
50:53that the size the Distortion seems size
50:56dependent and so that can do maybe
50:59direct damage in terms of misallocation
51:01but also have this indirect potentially
51:03powerful effect on tfpq and the
51:05distribution of tfpq and the average
51:07level of tfpq in the economy in effect
51:09you know the technology term that's one
51:11way to say it misallocation may have a
51:13direct reduction in output relative to
51:16inputs but it also may affect technology
51:19um and that could be new further damage
51:21so the third fact is that
51:24um is it skilled distortions appear to
51:26be more important than mixed Distortion
51:28this terminology I'm borrowing from
51:29Davey David Joel David uh and and venky
51:34at NYU which is
51:36um scale they called was a common
51:38component of here I should really have
51:40in the data we're measuring the value of
51:42the average products of capital labor
51:43and intermediates X is for intermediates
51:45here but is emphasized by David David
51:47what we really are trying to get out is
51:49the marginals so here that we're just
51:51I'm just uh reporting what they report
51:53which is about averages and they're
51:55hoping those are informative about
51:56marginals
51:57so they were pointing out and they
51:59looked um I looked in China and the U.S
52:02at firms and and they found that the
52:05scale this this person was bigger and
52:08more important than mixed dispersion and
52:10this is mixed with being ratios of
52:12things like the average product of
52:13capital to labor so what could what
52:15would show up in the scale be common to
52:17all of them well uh markups and the
52:20output Market
52:21um Revenue taxes and subsidies and of
52:23course measurement error that was common
52:25to all inputs or hit Revenue
52:27and but it does say that maybe the the
52:30that there's big distortions Beyond just
52:32Financial frictions that would show up
52:34you think in the capital labor ratio or
52:36risk that isn't a distortion but just uh
52:39compensating differential that would
52:40show up in the capital labor Ratio or
52:42things like wage markdowns
52:44is that where the if any one of these
52:46were the dominant source of tfpr
52:48dispersion you should see it in the mix
52:50and not in the scale
52:52um okay so this is an example of that
52:56um so you know even though we did a
52:59different calculation and it's nice for
53:01illustrating this so suppose what we did
53:03is we said here's allocative efficiency
53:05and say India and our sorry these are
53:09the gains from reallocation
53:11yeah if you reallocated inputs to the
53:14same extent that if you reduce
53:15misallocations that was the U.S level
53:17India would have a 59 higher tfp and
53:20China 31 percent higher in in these
53:22years that we looked at this is assuming
53:24no heterogeneity in production
53:25elasticities here we assumed all mixed
53:29distortions was really heterogeneity and
53:30production elasticities so that's kind
53:32of like saying there is no misallocation
53:34no mixed Distortion and you can see that
53:37reduced the the amount of misallocation
53:40but the majority of it was remaining
53:43so this is our studies in China and
53:45India which were more Frontier in Plants
53:47here
53:48um illustrate this and then but there
53:51there's David venky study of U.S and
53:54China found the same thing
53:55okay so another fact I wanted to
53:57emphasize is that these tfpr differences
54:01look pretty persistent and most of the
54:03variance in them is looks like a
54:04permanent component as opposed to a
54:06transitory component so so
54:09um David venky emphasized this uh at
54:12these U.S compustet firms and in Chinese
54:14firms as well
54:16um and my paper I had with
54:18um Mark bills that's the B and key and
54:20Ron we found the same thing for Indian
54:22plants and U.S plants and Manufacturing
54:25okay so I would say why would it be
54:28persistent well it's not really
54:30consistent with a dominant role for
54:31financial production so this ties into
54:33this question earlier about confirms
54:36save their way out of financial
54:37fractions well in the presence of
54:39ongoing shocks no there's always going
54:41to be this misallocation in the cross
54:42section
54:44um or or just dispersion in marginal
54:46products that are unavoidable if if you
54:48can't improve on these Financial
54:49versions policy can't improve on it
54:52um but uh at any loan firm it had like a
54:56big positive shock then capitals can
54:58adjust so in any one firm there should
55:00be mean reversion a lot of mean
55:01reversion in their average product or
55:03capital or hopefully their value of the
55:05marginal product capital and and
55:07therefore they're tfpr so the fact that
55:08we see very persistent differences isn't
55:11consistent with big Financial friction
55:13or for that matter adjustment cost some
55:14firm has a positive shock and then it's
55:16going to have a high average product to
55:18Capital value of the marginal cost of
55:20capital hopefully but it's it's the
55:22reason it's not investing more is
55:23because there's adjustment costs and you
55:25should see that slowly fade over time
55:27so the fact that these differences are
55:28so persistent says well maybe it's
55:30something that's not going away like a
55:32markup a price cost markup could be
55:34persistent a measurement error could be
55:36persistent too to have things like taxes
55:38or subsidies could be persistent so this
55:40isn't you know this misallocation that
55:42that this literature is finding is so
55:44massive potentially there's plenty of
55:46room for big distortions from Financial
55:48frictions um or Mark knockdowns and
55:51other sources that would show up in the
55:53mix and be and be relatively transitory
55:56for some of them like Financial
55:57fractions
55:58um but but it does say maybe the bulk of
56:00the actions in something like uh
56:02persistent markups
56:04okay
56:06um the last let me pause see if there's
56:08um questions or comments
56:31Ollie again
56:37hi hey I had a question about uh Chief
56:40you're on the one hand the chief your
56:44represent productivity on the other hand
56:46it represents Marco for monopolistic
56:49price making companies
56:51uh
56:54um what exactly does this index mean is
56:59it uh have a
57:01dual Behavior or something
57:06okay I guess I would say it's more
57:10closely related to it and then the
57:11simple model I wrote it would be only
57:13related to the distortion
57:14or measurement error
57:16um so but that might endociously be
57:18correlated with underlying tfbq so I
57:20think the first thing would be to say
57:22you know you really want to think of
57:24tfpr as a distortion and then estimate
57:28tfbq separately so if you really want to
57:30get a tfbq don't use tfpr as a perfect
57:33proxy for that that's that's the kind of
57:35main reaction that I have
57:37um to that and even though they're
57:38correlated
57:40um that they're not the same and so if
57:42you really want to get at tfpq then try
57:43to measure that that that distinctly
57:47um
57:47okay
57:49also opened another question this one
57:51pops up
57:53um okay so the last one this is less
57:55we'll die yep sorry just asking a
57:58written q a question from uh Victor it
58:01goes back a little while but I at some
58:04point I think it would be interesting
58:05for you to comment on it so um he writes
58:08uh I was wondering if differences in
58:10technology between countries uh for
58:12example agricultural Innovations for the
58:14US may be not be directly used in India
58:16and the cost of adapting them to a
58:19different context could be captured as
58:20misallocation
58:24okay
58:25that's a really good question I mean one
58:28angle I would have is that you know is
58:29there misallocation globally in
58:32Innovation because of differences in
58:34property rights you know like is there
58:36less Innovation into agricultural or
58:39health interventions that would be
58:40particularly helpful in developing
58:42countries and you can imagine this is
58:44something Michael Kramer has emphasized
58:46um Jacob moscono's recent work too
58:50um that that there's a potential that
58:53better intellectual property rights
58:55could improve the allocation of of
58:58research in a way that would be have a
59:00big effect on agricultural productivity
59:02or health in developing countries in
59:04particular and
59:06you know that that that's a possibility
59:08so that's one reaction I have to that
59:11um so one thing I skipped over it with
59:14the the fact that tfpq dispersions were
59:16different is is to emphasize that we
59:19weren't imposing that the tfpq
59:21distributions were the same we weren't
59:22assuming countries had the same
59:23technology and in fact I was kind of
59:26emphasizing that that Distortion
59:28misallocation and Market imperfections
59:30in government policy distortions could
59:33themselves affect the technology
59:34distribution I guess that's where I was
59:37going and responding to this question
59:39um but you know it does it you know I'm
59:43not an expert on technology adoption and
59:45agriculture in developing countries but
59:47it's fascinating for me as an observer
59:49to think about
59:50how our approach is like why isn't this
59:52being adopted and it seems like one of
59:54the lessons from that literature is
59:55people are behaving pretty optimally
59:57given the environment they're they're in
59:59so but that includes some things like
1:00:01information frictions of not knowing
1:00:03exactly what you know what fertilizers
1:00:06would use or what how to use the new
1:00:08hybrid seeds and what the payoff to be
1:00:10them to be Etc that's a tricky area too
1:00:12which is information frictions which
1:00:14because the government might be able to
1:00:15improve on those but it might also you
1:00:17know not be able to depending on how
1:00:19um anyway so I think that your question
1:00:21brings up a whole host of interesting
1:00:22interesting things so I'm glad glad you
1:00:25um glad I got to hear it
1:00:27okay
1:00:29um this last fact which I would say I
1:00:31would like to see more documentation of
1:00:32it I'd like to see it documented for
1:00:34labor like to see a document for other
1:00:35countries but Kerry and Vincent have
1:00:37this fascinating point where they say in
1:00:39U.S manufacturing across plants most of
1:00:42the dispersion in the average product of
1:00:44capital I'd say this because that's what
1:00:46they're measuring again we we're hoping
1:00:48this is reflecting the value of the
1:00:49marginal product to Capital but without
1:00:51production in less distance you can't
1:00:53quite
1:00:54um can't quite get there but we're using
1:00:57they're using this the proxy they say
1:00:58most of this dispersion is across plants
1:01:00within firms so one reaction to this was
1:01:03Well it can't be Financial friction to
1:01:05them because a firm is facing external
1:01:07Financial fractions but internally it
1:01:09shouldn't allocate its own Capital
1:01:10efficiently now that's that's a leap so
1:01:13so maybe that's not true in fact they
1:01:15they have a a subtle argument which is
1:01:18that the this Gap within firms is
1:01:20reflecting Financial professions
1:01:22interacted with lumpy adjustment costs
1:01:24so imagine there's this these non-convex
1:01:26adjustment costs for Capital
1:01:28and these big investment projects then
1:01:30that you're going to do if you face
1:01:32Financial frictions you're not going to
1:01:33do them all at once across the firms
1:01:35across the planets within firms you're
1:01:36going to stagger them so because of the
1:01:38financial fractions you're going to see
1:01:39more dispersion and that's optimal given
1:01:42the financial frictions and the
1:01:43adjustment cost but they're going to be
1:01:44gains to limiting the financial
1:01:45frictions that might show up in um in
1:01:49reducing this this uh dispersion
1:01:52um but I'd also like to emphasize you
1:01:54know there's things like plant specific
1:01:56markups they're you know different carb
1:01:58models being produced at different
1:02:00plants within a within a manufacturing
1:02:02firm they may have different markups and
1:02:04that's why they'll be very persistent
1:02:05differences in the average product to
1:02:07Capital across those plants within firms
1:02:09and of course which would not be
1:02:11misallocation the production
1:02:12elasticities could vary across plants
1:02:14within firms so I think there's a lot of
1:02:17opportunity for seeing to what extent
1:02:18this is true for labor out as well as
1:02:20capital what extent it's true in other
1:02:22countries and figuring out what the
1:02:23source of it is so I think this is a
1:02:25great paper for for pointing out this
1:02:27fact in a potential interpretation of it
1:02:29but if um I'm hoping it also leads to a
1:02:32bunch of follow-up work
1:02:34okay
1:02:35um now I'm glad I I have at least almost
1:02:38half an hour to devote to you know not
1:02:41just what people have documented already
1:02:42even though I've touched on some
1:02:43opportunities to do more things like
1:02:45outside Manufacturing in particular and
1:02:48but I want to emphasize you know where I
1:02:51think the most exciting work is most
1:02:52recently and and potentially in the near
1:02:54future
1:02:55okay so I put into three categories and
1:02:58one relates to something that was
1:03:00brought up earlier which is how does
1:03:01this relate to policies so if you time
1:03:04this allocation to explicit policies and
1:03:06distortions then the policy implications
1:03:08for that are more clear
1:03:11um another is like better identification
1:03:12so the approach that I've taken is like
1:03:15I said more like looking at at samples
1:03:18that of the whole population and just
1:03:19asking how much room is there for
1:03:21misallocation which is a very very
1:03:23direct in terms of policy implications
1:03:25but also you know you it's not very
1:03:27clear but the closest I got to a policy
1:03:30experiment would be that when I said
1:03:32that shrinking of state-owned
1:03:34Enterprises in China seem to contribute
1:03:36x amount 40 of their gains and
1:03:38allocative efficiency and contributed
1:03:40this much to tfp growth
1:03:43um and then a third category that I want
1:03:45to talk about is is how inequality
1:03:47trades off or doesn't with misallocation
1:03:51okay so explicit policies so remember
1:03:54misallocation is is allocated is it the
1:03:57flip side of allocative efficiency so
1:03:59it's all of Economics so not
1:04:00surprisingly there are many Papers
1:04:02written decades before Chang and I
1:04:04talked about this
1:04:06um you know they talk about you know uh
1:04:08misallocation so hopeenheim Rogerson
1:04:10talk about firing costs in Italy causing
1:04:13labor to be misallocated and trying to
1:04:15quantify that misallocation
1:04:17um there's many studies preceding us and
1:04:19and
1:04:20proceeding us that preceding proceeding
1:04:23us about size dependent regulations in
1:04:26developing countries or developed
1:04:27countries advanced economies and how
1:04:30that might misallocate uh labor garakano
1:04:33at all are looking particular in more
1:04:34recent work at Dynamic implications of
1:04:36this does it affect incentives of firms
1:04:38to increase their productivity and find
1:04:40sizable
1:04:41um distortions as a result of it there's
1:04:44the differences in tax rates across U.S
1:04:47states that lead to a non-trivial amount
1:04:49of of misallocation it looks like and
1:04:51then this is one of my favorite recent
1:04:53Studies by
1:04:54um cavalski Martinson Santos which is
1:04:58that a lot of the financial friction
1:05:00work on misallocation are just on
1:05:02financial frictions
1:05:04um has a model and then observes some
1:05:07data
1:05:08um and then often the implication is
1:05:10that there's like a shadow interest rate
1:05:12that might differ across firms some
1:05:13firms are rationed other firms aren't
1:05:15but there might not be a explicit
1:05:17interest rate difference being charged
1:05:19across firms so they document
1:05:21um using banking data in Brazil big
1:05:24differences in the interest rates
1:05:25charged across firms
1:05:28um across borrowers in Brazil and so you
1:05:31know One Direction to go with that is
1:05:32how much is that it's default risk and
1:05:34they argue it's not default risk
1:05:36um and there's also how much does it
1:05:38reflect risk more generally and they
1:05:39argue again it's not they argue what's
1:05:41happening is is actually monopsony power
1:05:43on the part of the banks the banks are
1:05:45charging High interest rates to more
1:05:46profitable firms and one thing that's
1:05:49really interesting about that is that
1:05:52that implies that there'll be less
1:05:54saving out of this Distortion over time
1:05:56so they they get they get rationing
1:05:59still occurs they get this explicit
1:06:01variation in the in the cost of capital
1:06:04and because the efficient firms say the
1:06:07small productive firm is trying to grow
1:06:09it's being charged High interest rates
1:06:10to to borrow more then that's going to
1:06:13reduce its profitability net of Interest
1:06:14costs and reduce its ability to save its
1:06:16way out of this financial friction so I
1:06:19think it's just a great paper in terms
1:06:21of the quality of the data and allowing
1:06:22us to see something we normally don't
1:06:23see and then potentially major
1:06:25implications for allocative efficiency
1:06:28or at least aggregate tfp that reminds
1:06:30me one thing I didn't say that I wanted
1:06:31to say that I really like to say is that
1:06:34suppose governments can't do anything
1:06:36about financial affections or adjustment
1:06:38costs
1:06:39but suppose those financial infections
1:06:41are bigger in some environment than
1:06:42another that's still going to affect
1:06:44aggregate tfp if adjustment costs are
1:06:46bigger in one economy than another or
1:06:48even the size of the adjustment costs
1:06:50the same but the shocks are bigger
1:06:51hitting firms so that that will generate
1:06:53more dispersion in tfpr and that even
1:06:56though it's not misallocation will lower
1:06:58aggregate tfp so things like adjustment
1:07:01costs and financial for instance those
1:07:02are real frictions that are affecting
1:07:04real aggregate tfp even if the
1:07:06government can't do anything about it
1:07:07even if they're the constrained optimal
1:07:09so we still need to understand how much
1:07:11of that is happening
1:07:13um if we want to understand uh economic
1:07:15activity even if the implication is okay
1:07:17that's a source of of reduced tfp that
1:07:21we that there's no direct policy
1:07:23implication for it okay
1:07:26um natural and policy experiments so
1:07:28increasingly people have looked at
1:07:29things like you know relaxing a change
1:07:31in capital requirements increasing
1:07:32Capital requirements on banks you know
1:07:34if you have great data you can see
1:07:36that's binding on some firms and not on
1:07:38others and so you can see what happens
1:07:39you cannot use that to estimate
1:07:41dispersion I should have this recent
1:07:43paper by
1:07:44um Dave Donaldson and collaborators and
1:07:47I think it's yeah I think it's Ecuador
1:07:50is that right
1:07:51um anyway that that finds
1:07:54um that uses a natural experiment and
1:07:56finds very small differences in the
1:07:59marginal product of I think it's labor
1:08:01across firms relative to the average
1:08:04product differences so that's an example
1:08:06of using a
1:08:07um a natural experiment where there was
1:08:09increases in in government contracts
1:08:12across some firms but not others try to
1:08:14estimate these gaps and finding small
1:08:16gaps so I have no stake you know I want
1:08:18to know the answer right if the answer
1:08:20is measurement errors is responsible for
1:08:22a lot of this dispersion that's really
1:08:23good to know because it's not real
1:08:24allocation if if it's misspecification
1:08:28or production elasticities maybe we can
1:08:30find that in a in a natural or policy
1:08:32experiment find that the gaps and True
1:08:34Values of marginal products aren't as
1:08:36big as the measured gaps in average
1:08:37products
1:08:38because you know we we really need to
1:08:40know what's real versus what's what's
1:08:41missed specification or mismeasurement
1:08:43in order to try to get it uh policy
1:08:45improvements and it might actually make
1:08:46things better
1:08:48um Natalie bow and Adrian betray have
1:08:50really nice paper coming out in
1:08:51econometrica about how FDI
1:08:54liberalization in India which we've kind
1:08:56of studied to death uh or not we but I
1:08:58mean I have one paper on it but a lot of
1:09:00people have studied they took a
1:09:02different angle on it which is to say
1:09:03the firms if this is a form of reducing
1:09:06a capital friction because the foreign
1:09:09firms come in and Provide Capital when
1:09:11they when they buy up a local Enterprise
1:09:13then
1:09:15um then you should see the firms that
1:09:16respond to most of the ones with
1:09:18initially High average products of
1:09:20capital and that's exactly what they
1:09:21find so that's a really interesting
1:09:23example of how you know you couldn't see
1:09:26it by just looking at industry
1:09:27productivity you might see an effect but
1:09:28you might not know where it's coming
1:09:29from they say it's it's in particular
1:09:32rapid Capital expansion on for for
1:09:35um firms or plans with initially High
1:09:37average products at Capital and then
1:09:39people have been looking at Branch uh
1:09:41Brand Bank expansion into poorer cities
1:09:44that were underbanked had less
1:09:46Banks per capita data and a good
1:09:49explicit government policy to open more
1:09:50branches and use that variation across
1:09:53space so as a as a natural policy
1:09:56experiment to see what it does to things
1:09:57like firm entry and growth and so I
1:09:59would just say that's a great
1:10:00opportunity to look at impacts on on
1:10:02efficiency in this application as well
1:10:06okay let me pause see if there's
1:10:08questions comments at this point
1:10:20it's a version of like I'm not going on
1:10:21in class until someone asks
1:10:24and ask a question or answer the
1:10:26question
1:10:27um I won't literally do that but
1:10:33okay feel free to interrupt if something
1:10:36comes up
1:10:37um Pete uh let me send to you so uh
1:10:40there's a question in the written q a
1:10:42there are there are several we don't
1:10:44time doesn't allow unfortunately every
1:10:45but um but one that I think might
1:10:48Intrigue people would be from Fang XI
1:10:51Wang uh asking um going back a bit sorry
1:10:54but you know uh the chance to pause
1:10:57could you please uh maybe say a bit more
1:11:00or possibly give examples of um
1:11:02environmental issues environmental
1:11:05externalities and their relationship
1:11:06with misallocation
1:11:10um that's really interesting I mean I
1:11:12think I know you're testing one but if
1:11:13you could say a bit more about uh be
1:11:15appreciated yeah um
1:11:18I think that seems like an area a
1:11:20promising area for people to start
1:11:21working on more I know of only a few
1:11:22things
1:11:24um one is I think there's some work
1:11:27saying that that um
1:11:30that uh some of the quantitative easing
1:11:34by central banks and advanced economies
1:11:37may have affected big firms that are
1:11:39that are actually disproportionate
1:11:42polluters
1:11:43um and so may have you know to try to
1:11:46reduce misallocation of capital they
1:11:48might have actually not only tilted an
1:11:50allocation of capital across big firms
1:11:51or small firms but
1:11:53um kind of worsened environmental well
1:11:56going the opposite direction of
1:11:57internalizing the environmental
1:11:59externality so that's one example of
1:12:00something I've seen
1:12:02um another example is done just looking
1:12:04in in Chilean data uh with Ernesto
1:12:07paston about to what extent there's
1:12:09differences in tfpr with with carbon
1:12:12usage so if you had a carbon tax with
1:12:14that amplify misallocation or would it
1:12:17mitigate it would it would it um would
1:12:20help solve some of this misallocation or
1:12:22would it
1:12:23um or or would it make it worse
1:12:26um so I guess another way to say that
1:12:28would be that the
1:12:30um that some of this some of the uh the
1:12:33misallocation that we're finding could
1:12:35be
1:12:37um could actually be efficient if some
1:12:39firm is producing too little but it's a
1:12:40polluter that would be the simplest way
1:12:42to say it imagine the firms with big
1:12:43markups or big polluters well the
1:12:45markups are actually doing a little bit
1:12:46of good in that maybe on that there we'd
1:12:49like to do them even more if we had a
1:12:50carbon tax maybe they'd shrink even more
1:12:53so I think that's just a really
1:12:54interesting topic
1:12:56um
1:12:57you know there's local pollution and
1:12:58Global pollution too so I so I'd like to
1:13:01see more on that I guess that's
1:13:02consistent with my starting to look at
1:13:03it in the Chilean day
1:13:05um so but there might be more out there
1:13:07that I'm not aware of
1:13:08um but I think that's that's really
1:13:10important because that's another way in
1:13:12which the this framework is limited
1:13:14which is if you're just looking at these
1:13:16private returns you're missing these
1:13:17social externalities environmental but
1:13:19but there's other other types as well
1:13:21knowledge externalities you know we
1:13:24naturally think the Market's
1:13:25misallocating when it comes to pollution
1:13:27externalities they're negative or
1:13:29knowledge externalities are positive so
1:13:31we'd like the government to to tilt
1:13:33things and maybe amplify some of the
1:13:36tfpr dispersion there in service of
1:13:39internalizing externalities
1:13:41so yeah
1:13:43um
1:13:43all right so I guess this relates a
1:13:45little bit to
1:13:47um I'll say misallocation versus
1:13:48inequality first to say where something
1:13:51that improved allocative efficiency
1:13:53might have worsened inequality so
1:13:56um so you could either suffer the
1:13:58misallocation and as a way to reduce
1:13:59inequality or you could reduce the
1:14:01misallocation and that's going to make
1:14:02an inequality worse so
1:14:05um I would view a lot of the trade
1:14:06literature as a version of saying hey
1:14:08trade utilization which is reducing
1:14:11misallocation
1:14:12um actually worsens inequality so you
1:14:14can imagine countries choosing to
1:14:18um to endure some of the misallocation
1:14:20I'm doing some work in progress saying
1:14:22you know hey maybe the majority of the
1:14:24people would suffer given there's a big
1:14:26size wage premium so there's Rich
1:14:28workers race workers high-wage workers
1:14:30at high tfpr firms if you subsidize them
1:14:33Not only would that have competition
1:14:35issues which were mentioned earlier but
1:14:37it would raise and and maybe increase
1:14:39inequality among owners which is
1:14:41emphasized by Bohr and midrkin it might
1:14:43increase worker wage inequality as well
1:14:45you could have a majority of workers
1:14:47that lose from that or at least less
1:14:48gain less and Atkin Faber and Gonzalez
1:14:51Navarro is a good example of they found
1:14:53everybody gain not everybody literally
1:14:55but in you know different groups
1:14:57um even you know say that I'm a
1:14:59tremendous the bottom decile or the
1:15:00bottom quartile or what it was but they
1:15:03even gained despite the fact that
1:15:04there's a lot of disruption of the
1:15:06informal sector
1:15:07um as a result of the FDI but the point
1:15:10is you could also have things like
1:15:11policy makers say I don't have an
1:15:13interest in allowing Walmart into India
1:15:16it's going to wipe out all these all
1:15:17these small firms that have low wages
1:15:21low consumption anyway so it's going to
1:15:22worsen inequality so it could lead from
1:15:24a political economy Point of View to a
1:15:26coalition to block it it could also be
1:15:28bad for a majority of people so if you
1:15:31have a social welfare function that puts
1:15:32enough weight on them or even puts equal
1:15:34weight on them that it could actually
1:15:36reduce the gains from reducing this
1:15:38allocation because you're worsening
1:15:39inequality and you might want to pull
1:15:41back on you might do
1:15:43reduce misallocation lesson as a way to
1:15:46to mitigate inequality I mean that's
1:15:48just another version of what we do when
1:15:50we say we tax and transfer we might
1:15:52think that causes distortions but we're
1:15:53okay with that because it's achieving
1:15:55our objective of of promoting social
1:15:57welfare by reducing inequality
1:15:59okay so this is I guess a little bit
1:16:02environmental There's issues with like
1:16:04this fascinating paper to me at least
1:16:07um about water rights in India that
1:16:09suggested there's a lot of misallocation
1:16:10of water across firms but that people
1:16:13also view it as Fair what's currently
1:16:15being done so that if you allowed
1:16:18um more efficient firms to access more
1:16:19water you'd increase inequality and
1:16:21people people didn't want that to happen
1:16:23so that's really interesting example of
1:16:25that
1:16:26okay um this is what I was referring to
1:16:30um big plans pay much higher wages if
1:16:32you think the size premium is big in the
1:16:33U.S it's gotten a lot of study it looks
1:16:35even more steep in in places like India
1:16:37and Mexico
1:16:38Okay so
1:16:40I wanted to emphasize too that there are
1:16:43ways in which reducing misallocation can
1:16:46reduce inequality so it's instead of
1:16:48misallocation versus inequality like you
1:16:51either suffer a misallocation or you
1:16:53suffer inequality here your
1:16:55misallocation is is uh is costing you
1:16:58equality so
1:17:00um managing the flow talk about a bunch
1:17:02of small and medium Enterprises with
1:17:04high average products of capital so they
1:17:06were below average in terms of the you
1:17:08know relative to the bigger firms
1:17:10um so reallocating more Capital to them
1:17:12with high returns
1:17:14um this was more like a a policy
1:17:16experiment or uh that was done there or
1:17:20natural policy experiment so maybe this
1:17:22is actually the the wrong phrase they
1:17:24were trying to estimate that marginal
1:17:26product value marginal products Capital
1:17:27there
1:17:28then my study I did with with Chang and
1:17:30Eric and Chad
1:17:32um reduced wage inequality across about
1:17:35gender and race we think we think this
1:17:37kind discrimination has reduced us a
1:17:39major source of wage inequality
1:17:42so that was one where reducing
1:17:43misallocation may have reduced
1:17:46um inequality so so these misallocations
1:17:50not be it uh reducing it might not be
1:17:52the expense of more inequality it could
1:17:53actually reduce high quality
1:17:56um so it's
1:17:57land landed and lottie's study of
1:18:00politically connected firms in in Italy
1:18:03um that could that favors big firms the
1:18:05politically connected ones make these
1:18:07fixed Investments and in political ties
1:18:09and so it arguably increases inequality
1:18:12that this crony capitalism exists and so
1:18:14attacking coronary capitalism might
1:18:16reduce inequality
1:18:19um Bento has some nice papers looking at
1:18:22um similar to what would what what I did
1:18:24in this paper but focusing on
1:18:26entrepreneurs in particular either by
1:18:27gender or by by race in the US and
1:18:30arguing that um there's barriers so so I
1:18:33mentioned this study in particular uh
1:18:35this recent study by morrisonian PSY
1:18:37which is is fascinating because it it um
1:18:40it touches on many things that we talked
1:18:42about for one it says it looks like the
1:18:45average product to Capital is high among
1:18:47female entrepreneurs in the U.S so it
1:18:49looks at newly started firms and looks
1:18:51at their return on Capital now you could
1:18:53say things like well maybe it's kind of
1:18:56Industries they're in so it survives
1:18:58controlling for industry in particular
1:18:59it fades with age
1:19:01so if you thought it was just industry
1:19:02then it shouldn't Fade with the age of
1:19:04the firm and the fading with age is
1:19:06consistent with they're able to save
1:19:07their way out of it and the other thing
1:19:09they found that was fascinating was that
1:19:11you might have said well maybe they have
1:19:12higher default rates higher risk their
1:19:13default rates were lower so it's like
1:19:15women were starved for Capital had less
1:19:18um external Capital the Reliant had to
1:19:20rely more on internal Capital had higher
1:19:22average products they seemed to save
1:19:24their way out of it and they'd pay their
1:19:25the the bank loans at a higher rate so
1:19:29um this is evidence again from the U.S
1:19:31consistent with maybe reducing that
1:19:34might be discrimination
1:19:36um it could just be Financial frictions
1:19:37though but reducing it might reduce
1:19:39inequality
1:19:41um and then there's a paper by Punk car
1:19:43and Goldberg that talks about how female
1:19:45entrepreneurs employ more women so
1:19:48reducing barriers to female
1:19:49entrepreneurs could potentially reduce
1:19:52gender inequality in wages and
1:19:54employment
1:19:55um and they get they get big effects of
1:19:57that's but that's a fascinating new
1:19:58angle that I hadn't seen before that
1:20:00that
1:20:01um that women versus men entrepreneurs
1:20:04employ women women just women
1:20:06entrepreneurs employ women
1:20:07disproportionately so there might be
1:20:09distortions in who they hire as well as
1:20:11distortions in their access to
1:20:12entrepreneurship to begin with
1:20:14so again these are these are more
1:20:16hopeful because they're all ways in
1:20:18which reducing misallocation wouldn't
1:20:20wouldn't give us a further headache in
1:20:21terms of inequality but instead could
1:20:24actually reduce inequality and promote
1:20:26equality so a more but a nice optimistic
1:20:29um area of potential research Brian
1:20:32yeah since they seem to be related with
1:20:35equality and inequality I'm wondering if
1:20:37it is some scope or if there is a
1:20:41relationship to conflict I mean
1:20:44political conflict
1:20:45using this allocation can reduce
1:20:48political conflict and probably if there
1:20:50is any relationship with other political
1:20:52economy
1:20:55so miss allocation and comfortable I
1:20:58I don't have a great answer to that off
1:21:00top of my head I mean I think
1:21:04and this reminds me of a of a paper I
1:21:06discussed Years Ago by uh dutella
1:21:10um which is like why isn't capitalism
1:21:11more popular in developing countries and
1:21:14he was saying basically it's associated
1:21:17with crony capitalism like it could be
1:21:19associated with selling off Monopoly
1:21:22rights to a foreign multinational that's
1:21:24like neocolonials that doesn't sound
1:21:26great at all or or giving Monopoly
1:21:28rights to politically connected local
1:21:30firms so
1:21:32um
1:21:33I think misallocation to the extent that
1:21:35it's tied to crony capitalism could be a
1:21:37source of political conflict you know
1:21:39like you could I don't know if it's the
1:21:40source or it's wrapped up within
1:21:43political conflict like if an Insider
1:21:44group is seizing resources or trying to
1:21:46seize a resources you know the expense
1:21:48of another group our groups are battling
1:21:50for access to natural resources or other
1:21:52so those
1:21:54they could be wrapped up in each other
1:21:55and I think I mentioned in the Indian
1:21:57example how I mean you could also have
1:22:00you know policies to reduce
1:22:01misallocation could be really unpopular
1:22:04and generate you know uh you know
1:22:06overthrowing of governments either
1:22:07democratically or or on Democratic I
1:22:10mean I guess you could even say in the
1:22:12United States you know
1:22:14um the decline of of of low-skilled job
1:22:17opportunities or medium skill job
1:22:18opportunities due to the
1:22:20de-industrialization due to exposure to
1:22:22trade with China probably not a leap to
1:22:24say that that has an impact on on U.S
1:22:28trade policies and you know Trump being
1:22:31elected and and maybe even you know um
1:22:33rejecting
1:22:35um Democratic norms and and uh
1:22:38respecting you know outcomes of
1:22:40Elections so that would be an example
1:22:41where probably not too much of a leap to
1:22:43say uh policies that reduce
1:22:46misallocation but that increased
1:22:47inequality may have furthered political
1:22:50conflict
1:22:51so I'd have to think about a lot more I
1:22:53think that's a really interesting thing
1:22:54to think about the interactions of those
1:22:56I mean anytime you see things like
1:22:58governments raising prices of subsidized
1:23:00products say food uh or Necessities or
1:23:04subsidized and it reduces those
1:23:05subsidies and then there's riots that's
1:23:07an example of political conflict
1:23:09interacting with attempts to reduce
1:23:11misallocation so I guess that's why one
1:23:14of the reasons why I think it's so
1:23:15interesting for us to try to find ways
1:23:17in which reducing misallocation can also
1:23:19reduce inequality rather than increase
1:23:21it there's potential for oh even then
1:23:25right there's some policies that benefit
1:23:27the majority but if a minority suffers
1:23:29that might generate political conflict
1:23:32um okay
1:23:33so I have some stuff in here on on firms
1:23:36and the data sets I'm just kind of skip
1:23:38over that
1:23:39um and uh I want to talk a little bit in
1:23:42my remaining
1:23:44um six minutes or so about some of my
1:23:46favorite papers in the literature that I
1:23:47haven't touched on already or maybe some
1:23:48of the ones I've touched on I want to
1:23:50say a little bit more about
1:23:52um so one I think you know I mentioned
1:23:54the measurement error as being a major
1:23:55issue so it's really nice paper by Julia
1:23:58Rosenberg and Kirk white that can see
1:24:01what cleaning is done in the U.S census
1:24:04data and their point is look there's
1:24:06tons of cleaning
1:24:07and so one of the reasons that looks
1:24:10like dispersion in tfpr is small in the
1:24:12US might be there's just a lot of
1:24:13cleaning done and maybe less cleaning is
1:24:15done in other countries and so it may be
1:24:18the cleaning process is making it look
1:24:20like there's more misallocation in India
1:24:22relative to the US than there really is
1:24:24that's that's
1:24:26um but I think this raises you know the
1:24:28issue of measurement which is really
1:24:29important you know we don't want to
1:24:31mislabel mismeasurement true
1:24:33misallocation with policy implications
1:24:35so I kind of want to emphasize this
1:24:37study uh up front
1:24:40um so and I I did some work that I
1:24:42already touched on that tried to get it
1:24:44Miss measurement that suggested in the
1:24:46US that it might be a growing problem
1:24:48over time
1:24:49um so adjustment costs I mentioned
1:24:52adjustment costs earlier there's this
1:24:54asker at all paper that's quite nice
1:24:56that that finds that in countries with
1:24:58basically bigger tfpq shocks but the
1:25:01same adjustment costs they'll display
1:25:03more dispersion in the average products
1:25:05to Capital and they're saying look
1:25:07that's that's not inefficient that's
1:25:08constrained optimal uh there's issues
1:25:10with the sample of this but the David
1:25:13venke when they looked at the sources of
1:25:15capital misallocation are a nice update
1:25:17of this and what they found was that
1:25:21they didn't just look at the dispersion
1:25:22in the average product of capital
1:25:24um but they looked at how persistent it
1:25:26was and if you had a convex adjustment
1:25:29costs you should see a lot of
1:25:31persistence to it and they don't find uh
1:25:33that much persistence to say for example
1:25:35investment rate differences investment
1:25:37rate differences if you're slowly trying
1:25:38to re you know build a new higher
1:25:41Capital stock in response to a high an
1:25:43improvement in your tfpq you should have
1:25:45a lot of Serial correlation when looking
1:25:47at this additional moment investment
1:25:48persistence they find a much smaller
1:25:50role for convex adjustment costs and you
1:25:52might think non-convex adjustment costs
1:25:54but then then there should be a lot of
1:25:56volatility so when they try to get at
1:25:58this in the U.S and Chinese data they
1:26:00find it's important but it's responsible
1:26:03for something like a seventh of the
1:26:04dispersion and the average product of
1:26:06capital so saying adjustment costs are
1:26:09an important source of this dispersion
1:26:10but it's not the the dominant source
1:26:13okay so
1:26:15um I already mentioned that the Carrick
1:26:16and Vincent paper
1:26:18um and this is a nice paper I'd like to
1:26:20see them it's been followed for a while
1:26:22I'd love to see them follow up on it
1:26:24they found the same thing there's a lot
1:26:25of dispersion across plants but they
1:26:27actually argue with misallocation within
1:26:29firms hence related to you know
1:26:31principal agent problems and management
1:26:32quality so that's I think just a really
1:26:35interesting potential area of research
1:26:38um to to go after
1:26:40um
1:26:41I kind of touched on this already but I
1:26:44wanted to revisit it which is halter
1:26:45cool a conceiverson have a nice paper
1:26:47where they say hey I think things that
1:26:50are being labeled misallocation are
1:26:52really variable markups or non-constant
1:26:55returns to scale now variable markups
1:26:57that's misallocation so it's it's
1:27:00endogenous misallocation as opposed to
1:27:02the this these these wedges that these
1:27:05mysterious wedges Tau that could be
1:27:06Revenue taxes
1:27:08um but but as I would call misallocation
1:27:11um returns to scale less than one so I
1:27:13guess what I wanted to emphasize though
1:27:15about this study
1:27:16is that deviations from CES and CRS they
1:27:21can reduce misallocation for given
1:27:22wedges so what I mean by that is that
1:27:25that
1:27:26um CES and and uh monopolistic
1:27:29competition and cost returns to scale
1:27:31are not just they're not there's nothing
1:27:34knife edge about that in the sense that
1:27:36you can relax any of those assumptions
1:27:37and you're still going to come back out
1:27:39some Taos or some distortions that are
1:27:41endogenous with markups you can back out
1:27:43distortions and you can back out tfpq
1:27:45nothing's going to be ultra sensitive to
1:27:48that but
1:27:50um the the Damage Done by those
1:27:53distortions can be different under
1:27:55things like diminishing returns to scale
1:27:57or are variable markups so for example
1:28:00if a firm has high markups and dodgesly
1:28:03as opposed to just being taxed a lot
1:28:05um then the elasticity of demanded faces
1:28:08is smaller so that's going to do less
1:28:09damage so that's what what um work by
1:28:12Edmund Michigan and Zhu for example find
1:28:14that when they have variable elasticity
1:28:16demand systems so you have endogenous
1:28:18markups that are bigger for bigger firms
1:28:20but that doesn't do as much the same
1:28:22markup doesn't do as much damage with
1:28:25basically a low stigma endogenously but
1:28:27it than it would if the stigma was
1:28:28exogenously high in the tower was was
1:28:32hitting them so I think the details of
1:28:34exactly what the market structure is the
1:28:36preferences are and the production
1:28:38technology are
1:28:40even to the extent that they don't get
1:28:41away come get get read completely of the
1:28:43wedges they're going to affect how much
1:28:45damage the wedge does in terms of
1:28:46aggregate tfp
1:28:48so that's one of the I think the
1:28:49enduring lessons of this paper
1:28:51have a really nice paper what you know
1:28:54there where they drive home that markup
1:28:55dispersion is generating misallocation
1:28:57but they also look at how how
1:29:00um how entry is distorted and they don't
1:29:02find find big effects there and Edward
1:29:05major do the same thing they find
1:29:08um this is they find small amounts of um
1:29:11misallocation from markups in the U.S
1:29:14and interestingly of that small amount
1:29:17only a little bit it comes from markup
1:29:18dispersion they've got a particular uh
1:29:21variable elasticity of substitution
1:29:23preferences that limits how much Market
1:29:24dispersion you can get so for example
1:29:26bake infaria assumed arbitrary Market
1:29:29dispersion so they got a ton more
1:29:31so this number might be might be too low
1:29:34potentially but it's interesting that
1:29:36they point out you know the average
1:29:37Market can distort this is the really
1:29:39the double marginalization point that
1:29:41that
1:29:42um Dave and David brought up which is it
1:29:45can distort intermediates relative to
1:29:46labor for example and of course things
1:29:48like labor Supply
1:29:50um so but so far these studies haven't
1:29:51found a big effect on on Entry you know
1:29:54in these quantitative Theory models
1:29:57um Peters has a really nice paper react
1:30:00interacting misallocation due to markups
1:30:02with growth incentives and also things
1:30:04like entry barriers and there's a
1:30:06follow-up paper that Michael has with
1:30:08Connor Walsh it's very nice
1:30:10um that looks at also again kind of
1:30:13interacting
1:30:14um endogenous markups with incentives to
1:30:16innovate and for both incumbent firms
1:30:18and entrants so I think this is a growth
1:30:20area that I kind of touched on when I
1:30:22said that there's um there's Dynamic
1:30:25implications of misallocation but this
1:30:27is these are explicit papers in the
1:30:28growth literature that are they're going
1:30:30at this so I'm I'm running out I'm
1:30:32running out of time I think so I'm going
1:30:34to stop now I left more slides than I
1:30:36knew I could cover because I figured
1:30:37people might just be able to look at
1:30:39them and see some comments I have on
1:30:40some of these other papers in the
1:30:41literature but I think I'll I think I'll
1:30:43start now
1:30:45thanks very much for having me
1:30:47thank you so much Pete that was a
1:30:49wonderful lecture and uh we'd like to
1:30:51thank Pete and hope everyone joins us
1:30:53next Thursday is our next lecture at uh
1:30:55uh 10 a.m uh where we'll be uh I think
1:30:58talking about Labor and and capital uh
1:31:01uh which are obviously an important part
1:31:03of this misallocation story
1:31:06yeah the Chris's we'll we'll cover some
1:31:08of that right so
1:31:10um thanks for this opportunity and also
1:31:11the unusual time that that accommodates
1:31:13my time zone nicely thank you
1:31:16well we thank you
1:31:18and thanks everyone for attending see
1:31:20you Thursday thank you
1:31:23very much

About this transcript

This page contains the full transcript of Lecture 4: Misallocation with Pete Klenow by International Growth Centre, generated from the public captions YouTube serves with the video. The transcript has 16,296 words across 2,582 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.