YT YouTube2Text

YouTube2Text → Transcripts → Lecture 6: Recent applications and advance

Lecture 6: Recent applications and advance — Transcript

by CEPR & VideoVox Economics · 17,401 words · 2,823 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome back.
0:02Um to this uh lecture six of uh STeG's
0:05virtual course
0:07on macroeconomic development. STeG is a
0:09research program run by CEPR on
0:11structural transformation and economic
0:13growth with a focus on the macroeconomic
0:15development of uh low-income countries.
0:18And as part of that, uh we've uh
0:21organized this course.
0:22Um STeG and this course are funded as
0:26part of the UK aid effort by the
0:28Foreign, Commonwealth and Development
0:30Office. We're thankful for their uh
0:32financial support to this.
0:34Um we're also thankful for uh Pete
0:37Klenow we have today uh here giving a
0:39lecture on misallocation.
0:41And um for Mandy Chan and um
0:46Kirsten McNeil and Lauren Wearing uh for
0:49helping organize this in the CEPR
0:51offices.
0:52Um
0:55I uh just remind you of the ground
0:58rules.
0:59Um Pete's going to lecture for about an
1:01hour and 15 minutes.
1:03And um
1:06uh he has a co-author, Keun Rhee,
1:09uh who will uh answer Q&A. So, Doug and
1:12I will not have to do that for those who
1:13are in the webinar uh portion. Uh for
1:16those who are panelists in the front row
1:18of students, um we ask you to mute, but
1:21you can feel free to unmute if you have
1:23a question. Um mostly clarifying
1:25questions during the talk and then uh
1:28broader questions uh people will have 15
1:30minutes afterwards for.
1:32Um so uh with that, oh, I should say the
1:36other thing is that the slides are
1:38available again on the course website.
1:40All the materials including the Q&A
1:43um and the recording will also be
1:45available uh
1:48So, with that, let me hand things over
1:50uh to Pete.
1:51All right. Thanks very much, Joe.
1:54Um let me share screen. So, I know I'm
1:56following Richard Rogerson who spoke
1:59last week and that's quite fitting
2:01because paper by Restuccia and Rogerson
2:04very much inspired Chang-Tai Hsieh and I
2:06to start working on this
2:08this topic. And I just remember being
2:11really wowed that that something I
2:13thought was kind of a small Harberger
2:15triangle could be this big distortion
2:18that could be holding back development.
2:20So, it really opened my eyes to a lot of
2:23um channels by which policies may be
2:24doing much more damage or market
2:26failures may be doing much more damage
2:28than I I previously imagined. I
2:29previously focused more on technology
2:31and human capital and this opened up a
2:33whole new whole new area.
2:36So, the outline I'm going to, you know,
2:38do a very simple model. Richard already
2:39laid out models in more detail. And then
2:42I'm going to talk about examples of
2:43misallocation including things that
2:45aren't misallocation. Um some of which
2:47he touched on and some of which he
2:48didn't. And then I'm going to try to
2:50recap and this most important part I'm
2:53trying to recap patterns that I see in
2:54the last 5 to 10 years of the
2:56literature. Um patterns I see that
2:58people have documented about
3:00misallocation that I think
3:01[clears throat] are quite useful going
3:02forward. And then briefly talk about
3:04some data sets that are publicly
3:06available in firms and plants and then
3:07some other ones that that are have been
3:09used but not so easily available.
3:12And then all the remaining time I'll
3:13spend on talking about, you know, some
3:15of my favorite papers in the literature
3:16and and what the essence of what they do
3:18and and, you know, what I think about
3:20them. So, um this is kind of an elastic
3:23portion because I'll just, you know,
3:25cover as many as I get through. Um okay.
3:28So, uh my first thing to start out with
3:30and I credit students, you know, at
3:32Stanford for forcing me to do this said,
3:35"You know, you use the term
3:36misallocation and then you give us, you
3:37know, a model where it where it
3:38features. What's the definition of it?"
3:40So, to me
3:41that was helpful for just forcing me to
3:43say,
3:44um you know, it's not only that that uh
3:48that that capital and labor or or people
3:51across occupations um that they could be
3:54reallocated but that the government
3:56could do something about that. So, not
3:58only could they be reallocated in a way
3:59that would increase output now or in the
4:01future uh in an efficient way. And so,
4:04it also forced me to clarify who would
4:07benefit from this and realizing that a
4:08lot of it is from a representative agent
4:10perspective. So, I kind of want to think
4:12separately that there's distributional
4:14questions that are very important.
4:16And then this half is like saying
4:18you know, if you if you wanted to get to
4:20some sort of Pareto frontier, um you
4:22might want to make sure that you
4:24allocate resources
4:25efficiently to maximize the the size of
4:28the pie. But, this definition is also
4:30helpful for clarifying things like I
4:32wouldn't call
4:33um
4:34you know, the fact that there trade
4:35costs that are shipping costs or there
4:37are transportation costs, things that
4:39are technological that mean that the
4:41rate of return to a resource you know,
4:44might be higher in one location than the
4:46other, I wouldn't call that
4:46misallocation. The planner, social
4:49planner, couldn't possibly improve on
4:50that. They can't repeal gravity. So,
4:52they can't make that any better. So, I
4:54wouldn't call that misallocation.
4:56And and for example, if it's if there's
4:58some sort of information problem, it's
5:00also if the government faces that same
5:03information problem as the private
5:04sector does, it's not clear the
5:06government can improve on it. So, it may
5:08be that the market isn't perfect, but it
5:10it can't be I can't be call it
5:12misallocation that the government could
5:14improve on if it can't observe uh
5:16anything more than the private sector
5:18can.
5:19Okay, so I just want to get that
5:20definition out there. And then the
5:22simple model which is, you know, nested
5:24by a lot of the stuff Richard talked
5:25about last week, I'm just a to fix
5:28ideas, picture aggregate output being a
5:30CES combination of different goods and
5:33elasticity of substitution sigma, and
5:36then this linear production, so output
5:39is linear in labor. So, for simplicity
5:41here, just one factor, but I'll
5:43elaborate on that in a bit. And then
5:44this is like process efficiency. But in
5:46this CES setup, I could have put A here.
5:50So, A turns out would look identical
5:52whether it's like quality, some sort of
5:54demand shifter, or a process efficiency
5:56down here. And I'll return to that in a
5:58few slides.
6:00So, picture a firm, a monopolistic
6:02competitor. They're the monopolistic in
6:03the sense that they set their own price,
6:06but they're a competitor in the sense
6:07that they take the wage, aggregate
6:09output, aggregate price, everybody
6:11else's prices as given. They view
6:12themselves as atomistic. So, it's like
6:15one firm per variety for I, per good I,
6:18intermediate good I.
6:20Um and then picture for simplicity just
6:22a fixed total supply of of labor to be
6:24allocated. So, the only allocation
6:25decision in this economy is how to
6:27allocate the production labor to these
6:29different I goods to be produced,
6:31intermediate goods to be produced. And
6:33let's for simplicity fix L and fix the
6:35number of of varieties or goods in this
6:37economy.
6:38And then what you get out of this this
6:39model is price is a a markup as a
6:42function of the elasticity of demand and
6:44marginal cost, but the the revenue tax,
6:47which I should have dwelled on. So,
6:48picture for the moment just like a
6:49firm-specific, product-specific,
6:52uh revenue tax rate tau Y.
6:55And I'm going to dwell on that at
6:57length,
6:57uh all the ways in which you might have
6:59wedges that look like taxes, but aren't
7:01actually government taxes and subsidies
7:03that might do as much damage as the
7:05taxes and subsidies that I'm describing
7:06here.
7:08Okay, so the firm now will price in that
7:11uh revenue tax rate. Um it's as if it
7:14raised its marginal cost. So, right
7:16there you get that it would be nice to
7:18isomorphic to put a tax on their inputs.
7:20In this model, just a single input,
7:22labor input, if you tax their labor
7:23input, that's the same as taxing their
7:25revenue in terms of it's like it's
7:27effectively in their first-order
7:28condition like increasing their marginal
7:30cost, and it'll get flowed through into
7:32their price. So, that will affect the
7:33allocation of labor because if they have
7:35a higher price, there'll be less demand.
7:37If I go, you know, and look about the
7:39upstream demand curve, you're going to
7:40demand's going to be inversely related
7:42to this output.
7:43I should have written down inverse
7:44demand curve, but I think I'll have it
7:46in a later slide in a bit.
7:48Um so, the higher the tau facing some I,
7:51the less labor will be devoted to it.
7:53Okay. So, one key thing here is that
7:56notice that TFPR, which was revenue
7:58divided by inputs, is not uh or is is
8:01proportional to the the tax rate here
8:03and is not equal to the underlying
8:06efficiency, which sometimes called TFPQ.
8:09So, TFPR for revenue productivity or
8:11revenue relative to inputs,
8:13and that in this simple example, there's
8:16no information about in revenue
8:18productivity about underlying
8:20productivity because PY over LI will
8:23only pick up this term.
8:25Um it won't show AI at all and that's
8:26because in this special case, there's
8:28perfect pass-through uh
8:31process efficiency into price. 1% higher
8:34AI means 1% lower price. That's AI shows
8:37up inversely in the price here. So, when
8:39you multiply by PI times Y over LI,
8:42which has AI in it, the AIs cancel. So,
8:45you get something that only involves
8:46this idiosyncratic tau. And the reason
8:48I'm dwelling on this is is cuz it has
8:50far-reaching implications. Even though
8:52this is a special model where TFPR says
8:55nothing about efficiency, even more
8:58generally,
8:59um it's not going to be the same as as
9:01as underlying efficiency. So, you really
9:03want to separately identify some
9:04underlying efficiency variable from some
9:06revenue productivity variable. And here
9:09it's showing that the more you tax
9:11firms, the more they'll look productive
9:13in this simple economy, and that's that
9:15doesn't necessarily reflect on their A's
9:17at all.
9:18It just reflects they'll have higher
9:19revenue per per worker. That might be a
9:22sign of a distortion in the economy
9:23rather than a sign that the firm is uh
9:25is really efficient.
9:27Um of course, in richer models, there
9:29could be a correlation between TFPR and
9:32AI because of things like AI not flowing
9:36inversely into price. So, things like
9:37variable markups. And I'll talk about
9:39those. But, still the the broader point
9:41is really worth stressing here, this
9:42difference. And the first people who who
9:44emphasized this terminology were Foster,
9:47Haltiwanger, and Syverson, who were
9:49really distinguishing revenue
9:50productivity from um uh kind of quantity
9:53productivity, they called it TFPQ.
9:56So, I'm borrowing and building on their
9:58terminology here, which is very
9:59clarifying.
10:00Okay, so aggregate TFP in this economy
10:02is like this generalized mean of the uh
10:05price of the efficiency the
10:07productivity variables, and then these
10:08wedges. And then you can't see it from
10:10just staring at this, but if you
10:12maximized uh
10:14you know, tau is some weighted average
10:15of the of the tau eyes, if if planner
10:18just wanted to maximize this, they would
10:19set all the tau eyes equal to tau. They
10:21would get This is just like sand in the
10:23gears. It's just distorting the
10:24allocation. This is a special economy
10:27where everything's efficient without
10:28these taus, so the taus just do damage.
10:31Um so, the the best thing to do would be
10:33to not have them. But, one thing that
10:35you can get from this simple example,
10:36and I'll add something on top of it, if
10:38you do log normal,
10:40log normally joint distributed A's and
10:42taus,
10:43and these are the means and the standard
10:45deviations of these A's and taus, and
10:47here's their correlation.
10:49Um then in that special case, you get
10:52you know, aggregate TFP in the economy,
10:54the log of aggregate TFP is a function
10:55of the mean A,
10:57and it's, you know, also a function of
10:59the uh variance of A,
11:02because you get um variance is good when
11:04sigma is greater than one. You can
11:05allocate more labor to the high
11:08efficiency firms. Um so, that's a
11:10positive. But, this whole term here,
11:12that's just coming from here. That's
11:13just coming from the generalized mean of
11:15A. That's what these two terms are. This
11:17is the unweighted mean of A, and this is
11:18like the weighted mean that you're going
11:19to get because you try to allocate more
11:22labor to the more efficient firms. And
11:24the the key misallocation term is this,
11:26which in this log normal case, in the
11:28CES economy, comes out really nicely,
11:30where it's the variance of these towels
11:33that does damage and how much damage is
11:35increasing in sigma. So the the closer
11:37substitutes these goods are in this
11:39economy, the more damage idiosyncratic
11:42wedges do idiosyncratic distortions do
11:44because labor will respond more heavily
11:47to it. So labor will be reallocated more
11:50uh strongly in response to a certain
11:52wedge the higher is the
11:53substitutability.
11:55Um so that's one key thing. And then if
11:57you if you wanted to put numbers on this
11:59and say you know, in the say in the US
12:01economy sigma squared uh tau is was 0.5,
12:05then you know, and sigma was four, you
12:08know, you say sigma squared sorry, sigma
12:10sigma let's say sigma tau is 0.5 so
12:12sigma squared is 0.25. Let's say sigma
12:14is four, that's the substitutability
12:16between varieties say. So you multiply
12:18that two, you get a 50% drop if you saw
12:22a 0.5 standard deviation in in uh
12:25revenue per worker cuz remember TFPR and
12:28tau is proportional to revenue per
12:30worker in the simple economy.
12:32Um if this was varying the standard
12:34deviation of this across firms or across
12:36plants was something like 0.5 which is
12:39not not far from what what people get in
12:42various studies,
12:43then we're saying 4 over 2 times sigma
12:45squared is 0.25 so you get 0.5. So you
12:49get we produce half as much output or
12:52you know, minus 50% as much output as we
12:55could if we had perfectly efficiently
12:57allocated labor to the point where we
12:59could equate uh labor um
13:02labor per worker.
13:04And of course one of the things that
13:04Richard brought up that I'll I'll talk
13:06about too is that um maybe this isn't
13:08misallocation cuz some of that
13:10dispersion is unavoidable. Some of that
13:11dispersion is from transportation costs
13:13or adjustment costs so the planner
13:15really can't reallocate from some of
13:17those low revenue productivity low TFPR
13:19to high TFPR. So that becomes a really
13:21important issue when thinking about
13:23whether these gains are achievable.
13:25Whether gains from from limiting this is
13:28is even achievable or feasible.
13:30And then there's issues like, you know,
13:31is there something wrong with this
13:33specification, this simple model?
13:35And what will altering the some of the
13:36some simplifying assumptions, what will
13:38they do to how we infer how much damage
13:41being done from a certain amount of
13:42dispersion in TFPR.
13:44Okay.
13:46All right. So, I've actually talked
13:47through some of these lessons already as
13:49I was going.
13:50Um that
13:52you know, the wedges show up in TFPR
13:54TFPR isn't TFPQ. So, really be careful
13:57about using this as a proxy for this.
14:00And there's really a whole generation
14:03of productivity research that did that.
14:05And so, every question that people did
14:07in that generation of productivity
14:09research that was using this as a proxy
14:11for TFPQ should be revisited. An example
14:14is a you know, paper by like Nico
14:16Voigtländer and
14:18and a collaborator which basically says,
14:20let's think about whether exporting
14:22allows firms to learn how to produce
14:24better. Let's examine it not in terms of
14:26TFPR, but in terms of TFPQ. And they
14:29found much stronger evidence for that in
14:30TFPQ than people had found previously
14:32when they were looking at TFPR. If TFPR
14:35was like a tax or a markup wedge, that
14:38might be stable as you enter export
14:40markets, but you get a lot bigger. And
14:42when you got a lot bigger, that's a sign
14:44that you're might have might have
14:45improved your efficiency. There might
14:46have been learning by exporting. And
14:48they that was uncovered when they
14:50focused on TFPQ, which was the
14:52underlying efficiency rather than this
14:54this TFPR.
14:56Okay. So, um other things I stressed is
14:58TFP is decreasing in the dispersion of
15:00these wedges. And as I I mentioned, more
15:02so the higher is is sigma. Um aggregate
15:06TFP in this static model is unrelated to
15:08the mean wedge, but in other models,
15:09like if there's free entry or there's
15:11investments in in quality, uh you know,
15:14ongoing R&D, then the mean wedge will
15:16start to have dynamic implications.
15:18So, that's just a simplification here
15:20that it didn't that the mean what didn't
15:22do anything at all.
15:24Um and then I just mentioning that
15:26dispersion is increasing in sorry that
15:28the aggregate TFP is increasing in the
15:29dispersion of TFPQ. That's kind of an
15:31interesting question related to
15:32misallocation because if the TFPQs
15:34themselves, their distribution is
15:36endogenous to the misallocation, then
15:39misallocation could be affecting the
15:40economy through its effect on the
15:42distribution of this and therefore the
15:43gains like if you had you know,
15:46borrowing from some work by um
15:48Bento and Restuccia, if you had
15:50a bunch of firms with really small
15:52productivity that enter because of lots
15:55of uh of barriers to to becoming bigger,
15:57say in India, then you might see um
16:00less dispersion of TFPQ or less in the
16:02right tail of TFPQ or maybe more
16:04dispersion cuz maybe few big firms that
16:06are efficient and you have a large
16:07number of small, inefficient firms. Um
16:11so you might have more or less
16:12dispersion of TFPQ as a result of
16:14policies there.
16:16So that's important for taking a full
16:17picture of the effect of misallocation.
16:21All right. Um
16:23so I wanted to kind of have an aside on
16:25on uh
16:27how to think about [clears throat]
16:27getting data on TFPQ.
16:30So some you know, if you if you the
16:33output here, um you can infer that the
16:35sort of step by on that. You can infer
16:36from inverting the demand curve. So the
16:38restriction is you need to know the
16:39demand curve. But if you know,
16:42you know, a a CES demand curve or if you
16:44knew the firm elasticity, you know, if
16:46there was a um
16:47a variable elasticity world, you could
16:49infer the firm's effective output from
16:52its revenue, uh something proportional
16:54to that by just inverting it using this
16:56the elasticity in this demand curve. And
16:58again, there could be a sigma I on this.
17:01Um and one thing I really want to
17:02emphasize
17:03is remember when I set it up, I said the
17:04AI I could have put it in quality.
17:06What you'd back out here would be
17:08quality as well as uh the quantity of
17:10what they're able to produce cuz what
17:12generates revenue for a firm is not just
17:14like how many cars they produce, but how
17:16good those cars are.
17:18Like, nobody really cares if people
17:19produced a million Model Ts from 1900
17:22now in the US market that have really
17:24high quantity productivity, but the
17:26quality would be so low that there
17:28wouldn't be much demand for it.
17:29This Y should capture what we really
17:31care about, which is this composite of
17:33quality and quantity.
17:35And so then when you think about
17:37dividing this Y that you've backed out
17:39in this way from some measure of inputs,
17:41and again, you can generalize inputs
17:42beyond just one here, keep in mind
17:44you'll get at not only process
17:46efficiency, but quality. Okay, and the
17:48reason I stress this is because
17:50sometimes people say, "Well, if you
17:51don't have data on unit prices, then
17:54you're you're stuck." And I'm saying,
17:56"No, actually, you're stuck if you're
17:57trying to get at process efficiency that
17:59way because you really need units of
18:00output, quantities of output divided by
18:02inputs to get at process efficiency."
18:05Um, but if you wanted to get at a
18:06composite process efficiency and
18:08quality, which is is broader and
18:10therefore in some ways more valuable,
18:12um, you can get that just by inverting
18:14the demand curve. I say just, that can
18:16be hard to figure out how to work the
18:17demand curve, but if you have a method
18:19for doing that, that will get you a
18:21composite you do really care about. And
18:23if you do on top of it in a subset of
18:25data sets, there is data on on physical
18:27quantities, so you can get average unit
18:29prices from that, too, but if you can
18:31use that physical quantities, let me say
18:32I call those QI, and I'm distinguishing
18:34that QI from this YI, which would have
18:37quality in it. So, this is just physical
18:38quantity. Then you can infer process
18:40efficiency from that, and that might be
18:42separately interesting to figure out if
18:44our firms high quality, and you infer
18:46the quality from the difference between
18:48the TFPQ and uh
18:50I should have said on the process
18:51efficiency here, that the TFPQ versus QI
18:54over LI here.
18:55So, the difference between those might
18:57reflect quality. Okay, and again,
19:00my prior is a lot of firms compete on
19:02quality and not on pure just like prop
19:04price of and we see tons of dispersion
19:07in price, but we also often see
19:09uh
19:10massive differences in sales for similar
19:12prices, and also some of the high price
19:14products and firms have really high
19:16sales. So that would be inconsistent
19:18with any quality differences
19:20if you if they had higher price and yet
19:23higher sales at the same time. So I
19:25think there's plenty of scope the data
19:28telling us there's plenty of scope for
19:29quality dispersion not just process
19:31efficiency dispersion.
19:34All right, so this was a simple model
19:37and and Richard tested some of this
19:39stuff. It's easy to generalize the
19:40physical capital.
19:42It's easy to generalize to intermediate
19:44inputs. I think that came up in a
19:45question last week.
19:47My my recent paper with Mark Bils and
19:49Kian Rahmani is here as well.
19:53He we we explicitly brought in
19:55intermediates and I'll talk about that
19:57in a couple slides.
19:58You can put wedges not just on revenue
20:00but on all inputs if you want.
20:03As Richard mentioned, you can't
20:04distinguish like a common wedge on labor
20:08capital materials versus revenue. Those
20:10will look the same in terms of firm
20:11behavior.
20:13So it's kind of like you can identify a
20:14common wedge that affects all inputs or
20:16revenue and then relative wedges that
20:18affect like capital relative to labor,
20:21labor relative to materials and so on.
20:23And then another thing that's easy to
20:25bring into this model is overhead costs.
20:27Bartelsman, Autor, and Scarpetta are
20:29examples of doing that. There's always
20:32the challenge of like how big are
20:33overhead costs, but if you had some
20:35information or way to infer them, you
20:37can easily adjust for them. In which
20:39case you back out these tiles not from
20:41TFPR but from a kind of overhead
20:43adjusted TFPR. You'd subtract overhead
20:46inputs from the inputs that you had in
20:47the denominator and then you'd get a
20:49measure of TFPR that was closer to the
20:52wedge that we're interested in.
20:54But they're they're certainly right in
20:55pointing out that if there are these
20:57overhead costs that what I define as
20:59TFPR as revenue over inputs would not
21:02totally reflect the wedge. It could also
21:03reflect reflect these overhead costs.
21:06You can also think about relaxing the
21:08assumption that that entry costs are uh,
21:11you know, that entry's fixed. And Dinger
21:13and Morrow is an example of a nice
21:14recent paper that I'll touch on later,
21:15too.
21:17Um, okay. So, harder to do in the sense
21:19of like you have to start modeling
21:21dynamic decisions. Um, these
21:23generalizations can all be done with
21:25static decisions. Like if you assume
21:27that the fixed amount of capital was
21:28being allocated at a point in time. Um,
21:30all of these things stay within a static
21:32framework and very tractable. Here
21:34you're starting to deal with dynamic uh,
21:36decisions or things that aren't
21:37closed-form solutions anymore. Um, so
21:40when these are the dynamic ones,
21:42financial frictions, that hasn't stopped
21:43a lot of people from doing that, you
21:45know, Joe. Uh,
21:47Caballero for example, uh, one of the
21:49leaders in bringing this in and and
21:50finding significant amounts of
21:51misallocation as a result of financial
21:54frictions that made some firms with
21:56really good productivity but not the
21:58internal financing to be able to to
22:00combine it with enough physical capital.
22:02So, if external financing has
22:05constraints, that would that would put
22:07sand in the gears of allocating capital
22:09to the the firms you really want to have
22:10them off.
22:12Okay. So,
22:13um, people have done adjustment costs.
22:14I'll talk about this paper a little bit.
22:16Uh, and also David and Venky, which is a
22:18more recent one. Um,
22:20that CES, you know,
22:22um, that's special.
22:24Uh, you can imagine doing variable
22:25elasticity of substitution. I wish we
22:27had like a consensus on what was a good
22:30uh, description of demand, like whether
22:32it was some particular VES with some
22:34particular functional form or parameters
22:36was was considered a a good
22:38approximation, but I think this the
22:39literature is still exploring
22:41alternatives and seeing what they do.
22:42So, I'll talk about those a little bit.
22:45Um, you can also say, you know, what
22:47about the single inputs that feed into
22:50output downstream? What about
22:52you know, the car industry uses steel
22:54and and all these other inputs and um,
22:58and then cars that can be used as inputs
23:00into making steel, etc. So, you know,
23:02general input-output structure.
23:04Fortunately, there's this beautiful
23:05paper by Bakkai and Farhi that that
23:08treats a very general environment in
23:09terms of input-output and in terms of
23:11things like constant returns to scale,
23:12which I was which I was imposing.
23:15And then there's all kinds of
23:16interesting issues. This is more of a
23:18dynamic thing, like I was mentioning,
23:20where you endogenize where the TFPQ is
23:21coming from. So, instead of taking the
23:23TFPQ distribution as fixed, and the taus
23:26kind of impact the allocation of of
23:30resources, capital, etc., to the
23:32different firms with different TFPQ,
23:34instead firms' investments, as I was
23:36mentioning, and TFPQ are a function of
23:38that. And then two other studies are
23:39some of the the the leading papers on
23:41that.
23:43Okay, so
23:45that was my kind of introduction and
23:46simple model and giving out some simple
23:48ideas. Now, I want to talk about, you
23:50know, I've already touched on some
23:51examples of misallocation, but I want to
23:53be explicit about some more, and then I
23:54want to
23:55talk about the the five fact patterns
23:57after that.
23:58Um
23:59okay. So, first, what is not? Before I
24:01give you examples, what what are things
24:03that are not
24:04misallocation? Uh even things that would
24:06generate TFPR dispersion that aren't
24:08misallocation by my definition. Well,
24:10anything that's just measurement error.
24:11Anything that's just like they made an
24:13error in recording revenue versus
24:15inputs. Maybe the true economy perfectly
24:17equates TFPR and the value of of of the
24:22marginal product of
24:23labor or capital across firms in the
24:25economy, but then when we measure it, it
24:27it doesn't, but that's just cuz of
24:28errors in the way things are recorded.
24:31Um that clearly wouldn't be
24:32misallocation, so we clearly have to
24:34worry about that. My paper with Mark
24:36Bils and Kian Ron is precisely about
24:38that.
24:39Um so, another one is adjustment costs.
24:41I'll mention that, um the the couple
24:44papers in that literature, as I as I
24:45promised.
24:46Um another thing I haven't brought up
24:47though is, you know, differential risk.
24:49Maybe the rate of return on capital is
24:51systematically higher at some firms or
24:53in some industries than others. That
24:55might be misallocation, or it could be
24:58um that's just riskier, and so perfectly
25:00functioning market is requiring a higher
25:02average rate of return to compensate for
25:04the systematic risk that's higher in
25:06those firms or the the as sectors. So,
25:08that in general people need to worry
25:10about. Um could be compensating
25:12differentials. If you looked at the
25:13value of the marginal product of labor,
25:16you could even do this for the same
25:17worker. You know, that my colleague um
25:20Isaac Sorkin has studied that. Um which
25:22is you know, workers may voluntarily
25:24move from a low-wage firm or high-wage
25:27firm to a low-wage firm. Um and you
25:29could say, "Wow, the same worker had
25:30different um marginal products and lit
25:32and wages." So, if the same worker has
25:34different wages across different firms,
25:36then you might say that's evidence of
25:38differences in their marginal product,
25:40and therefore misallocation. But, you
25:42have to be careful on that one. There
25:43could be amenity differ- Workers may be
25:45indifferent between these firms. Their
25:46marginal product may differ, but the
25:48amenities or disamenities may differ
25:49across those firms. So, I think this is
25:51a ripe area for people to study to what
25:54extent is is labor misallocated versus
25:57there's just amenities and disamenities
25:58that have to be incorporated.
26:00And then I think this came up in
26:01Richard's thing, too, which is
26:03um imagine that it looks like you know,
26:06the capital-labor ratio differs across
26:08firms, even within a narrow industry.
26:10You could say, "Oh, that's a sign
26:11there's misallocation because if they
26:14had the same technology and same face
26:15the same factor prices, um same capital
26:18intensity of their technology, um then
26:20they should have the same capital-labor
26:21ratios. Um so, maybe they have different
26:24costs of capital, for example. So, an
26:26alternative is that you know, the one
26:27that's has a lot of high capital ratio
26:29labor ratio just has, you know, more
26:31capital intensity in their production.
26:33The production, you know, picture
26:34Cobb-Douglas, and they have a bigger
26:35share on capital. Um so, then you have
26:38to worry about that that that would
26:39generate lots of dispersion in these
26:42naive TFPR measures that aren't really
26:44reflecting dispersion in the value of of
26:46marginal products of across these.
26:48That's the key idea of misallocation is
26:50if there's dispersion in the value of
26:52marginal products, then there's some
26:54gain that can be had by reallocating
26:56inputs from the their low value of
26:58marginal product uses to their higher
27:00value of marginal product uses. And
27:02these are all saying
27:04be careful, this would be it's it's
27:05imaginary, it's illusory, those gaps
27:07aren't really there. This is saying you
27:08can't realize those gaps cuz there's
27:10costs to move those inputs. This is
27:12saying there's risk that that those gaps
27:14are compensating for. This is saying
27:16there's amenities to workers that those
27:18those gaps are are reflecting. And this
27:20is saying you're not measuring the gaps
27:22and marginal products properly because
27:24they're not on the right production
27:25function.
27:27Pete?
27:28May I May I jump in here? So, when
27:30you're when you're
27:32when you're ruling out transportation
27:33cost as a misallocation, you're
27:36implicitly assuming that transport
27:39infrastructure is perfectly provided
27:40everywhere, right?
27:42No, just the way I wrote it may may made
27:45you think that though. Um so, I have a
27:47later slide that says there can be
27:48misallocation of investment in
27:50infrastructure, absolutely.
27:52I'm saying think about like conditional
27:55on efficient allocation of
27:56transportation, just the existence of
27:58physical transportation costs would mean
28:00that there'd be some wedges in the
28:02marginal product fight, but I completely
28:04agree with your point that that there
28:06could be misallocation and of
28:08infrastructure investments.
28:11Thanks.
28:13Yeah, okay.
28:15Um
28:16so, I'll mention that again in
28:18discussing the Asker-Coller Wexler uh
28:21the Locker paper.
28:23Okay, so what can generate, you know,
28:25misallocation? Examples would be, you
28:27know, literally tax rate differences
28:28across firms. In the United States,
28:29there's plenty of special tax rates for
28:31big firms versus small firms, but also
28:33whole sectors that get special tax
28:35subsidies. Economists love a carbon tax,
28:37but in the real world we have a carbon
28:39subsidy, uh at least in the US, which
28:41just seems kind of crazy.
28:42So,
28:43lots of size-dependent regulations, in
28:45fact, long before
28:47uh Restuccia and Rogerson, there were
28:48papers like Hopenhayn and Rogerson that
28:50talked about firing restrictions and how
28:52they would misallocate labor, you know,
28:53in a dynamic model where workers were
28:55moving across firms subject to these
28:57these dynamic costs. So, there's been a
28:59long history people like Restuccia and
29:01other people studying this is what
29:03Richard was calling the direct approach,
29:05which is you take an observable policy
29:08and then model what it's doing and
29:09provide evidence on what it's doing and
29:11figure out how much misallocation that
29:13might be generating. Whereas you'll
29:15notice a lot of what I'm doing is very
29:17much from the what he called the
29:19indirect approach, which is let's look
29:20for all these gaps in in in values of of
29:23marginal products and that's evidence of
29:26all types of things that might be
29:27contributing to misallocation. So, our
29:29approach was like, how much room is
29:31there for misallocation? Is this big?
29:34Does this add up to a big number?
29:36Restuccia and Rogerson pointed out that
29:37it could be big you know, in their in
29:39their model economy and we found there's
29:41evidence consistent with it can very
29:43well be big. So, if the gaps in the data
29:45were really small, we'd be saying
29:47theoretically that can contribute to big
29:49TFP differences, but we don't see any
29:50evidence consistent with that. Instead,
29:52we found plenty of evidence opening up
29:54the possibility of a lot large amounts
29:56of misallocation.
29:59Um but this direct approach that started
30:01to look at things like size of kind of
30:02regulations um preceded at least kind of
30:05indirect approach and it continues. So,
30:07there's like the the stuff by
30:09[clears throat]
30:09Luis Garicano, Claire Lelarge, John Van
30:12Reenen about
30:13size regulations in in France seeming to
30:15have a big effect as an example.
30:18Okay, so one category I really want to
30:19emphasize because I think it's it was a
30:21confusion for some reason is that price
30:23markup differences across products,
30:25that's generating misallocation.
30:27Okay, that this may seem obvious
30:29because, you know, we always think about
30:31the monopolist charge too high a price
30:33above marginal cost, the competitor
30:34charges price equal to marginal cost,
30:36the planner wants wants the monopolist
30:39to produce more. They don't want them to
30:41they don't want consumers to substitute
30:43away from something. Imagine two firms
30:45with the exact same marginal cost in
30:46different industries. One has a big
30:48markup, the other doesn't. We have too
30:50little of the markup good. From the
30:52planner's point of view, we want to
30:53equate the consumer decisions to the
30:56marginal cost. We want the you know the
30:58marginal rate of substitution on the
30:59consumption side of the downstream
31:01demand side. We want that to be equated
31:02to the marginal rate of transformation.
31:04So we want the marginal utilities or
31:06downstream marginal products to be
31:08equated to the relative marginal cost.
31:10And and markups are like a wedge in that
31:12process. It distorts it. It means that
31:14if Apple's markup is bigger than
31:16Samsung, we don't have enough Apple
31:17smartphones and we have too many Samsung
31:19smartphones.
31:20So there's some gain to reallocating it
31:22till those values of the marginal
31:24product um are equated.
31:27And I I stress this because I think some
31:29people reacted that you know, government
31:32imposed distortions, well that's
31:33inefficient, but if it's profit
31:34maximizing behavior, that's efficient.
31:36Well, no, of course, this is a market
31:37imperfection. This is just cuz it's
31:39profit maximizing doesn't mean it's it's
31:41not distorted. And the same thing goes
31:43for markdown differences and there's
31:45growing interest in that. There's a
31:47paper by Berger and um Herkenhoff and
31:49Mungiu that that emphasizes markdown
31:51dispersion and how much misallocation it
31:53can cause, for example.
31:56Okay. Um
31:58There's other things.
32:00Sorry.
32:00Can I ask you a question in terms of the
32:01price markups? So you said that that
32:03this can generate the
32:05misallocation, but in terms of for
32:06instance, your
32:072009 paper with
32:10with the share,
32:11um the
32:12you you using the framework of CES
32:14aggregator, so the the markup is the
32:16constant under this framework. How are
32:18you able to generate
32:20markup differences related misallocation
32:23in your framework?
32:25Yeah, that that CES one would not
32:27generate. If you if you could either
32:28deviate from CES, so variable elasticity
32:31substitution, or we had monopolistic
32:33competition, you could keep CES and put
32:35in, you know, Cournot, for example,
32:38rather than monopolistic competition.
32:39You know, small number of players within
32:41every subsector, and then their market
32:43shares would feed into their markups.
32:44The bigger firms would have higher
32:45markups cuz they'd have a lower
32:47elasticity they'd face, and the smaller
32:49firms would have a have a a lower
32:51markup. So,
32:53some combination of changing the market
32:55structure away from monopolistic or
32:56changing the preference structure away
32:58from CES. And so, people have done that
33:01um
33:01quite a bit.
33:03Yeah.
33:04Thank you.
33:05But, I I'll emphasize that it's not
33:06isomorphic to what we did.
33:08The the uh
33:10that saying one could change some of the
33:11assumptions and get lots of uh
33:13misallocation due to market dispersion
33:15isn't the same as saying that what we
33:18measured as wedges, you know, they might
33:20be market they might be uh fine under
33:23those alternative models, but that that
33:24the misallocation we got, the deadweight
33:26loss, the Harberger triangles we
33:28aggregate up to might be different
33:30because the economy is uh
33:32characterized by variable elasticity of
33:34substitution. So, I'll touch on that a
33:35little bit later.
33:37That's an important important issue I
33:39wanted to convey.
33:41Um okay. So,
33:43you know, there's there's also like not
33:45just thinking about a generic worker
33:47allocated across firms or how many
33:49workers are at this firm versus how many
33:51workers are at this, but which people.
33:53And you can imagine discrimination any
33:55number of places. And the reason I
33:56emphasize this is cuz I feel like
33:58there's an enormous literature on
34:00whether wage gaps reflect discrimination
34:02and in looking for discrimination in
34:04things like housing uh you know,
34:06mortgage loans and other loans by banks.
34:09And I think
34:10what, you know, I understand that that
34:12we're obsessed with this because we
34:14we want to know how much, you know, how
34:16much one group how much discrimination
34:17exists, and it's it's unfair, right?
34:20It's it's unjust. But, what surprises me
34:23from you know, I'm not outside of this
34:25literature, what surprises me is there's
34:27less emphasis than I thought there would
34:28be on the
34:30on the efficiency aspect of it. That um,
34:33that it means you're misallocating labor
34:36if the people with the best talent
34:37aren't the ones admitted to the graduate
34:39school or aren't the ones admitted to
34:40med school, aren't the ones hired for
34:42certain jobs. Um, or don't get loans for
34:45housing. So, if there's discrimination
34:47against the black population in the US
34:49in loans, you know, or business loans,
34:51that's going to mean there's
34:52misallocation of who's entrepreneurs and
34:54who's getting who's in houses. Um, and
34:56that's an efficiency loss. So,
34:59the misallocation literature comes
35:00naturally to that of saying not only is
35:01there a distributional not only is there
35:03a distributional implication, but
35:05there's a size of the pie implication as
35:07well. It's an efficiency
35:09uh, loss associated with that. So, it's
35:11unjust, increases inequality, and on top
35:14of that shrinks the pie.
35:16Okay. So, I'll give you an example of
35:17that in a bit. Um, you know, we already
35:20mentioned some financial frictions, but
35:21you can also think of state-owned banks
35:23or, you know, government wanting to
35:25favor certain firms that are politically
35:27connected as a way of maintaining power.
35:29That seems like it's true in the United
35:31States. We, you know, uh, it's a I think
35:34it's bound to be true everywhere and
35:35every time just to varying degrees.
35:37Um, efficiency wages haven't got as much
35:39attention. There's a lot more attention
35:40on on firm monopsony power recently, but
35:44efficiency wages generate misallocation,
35:46too, right? If some industries have to
35:48pay higher wages to try to induce people
35:50to to put forth effort in in the
35:52presence of a asymmetric information
35:55problem, then that means there's too
35:56much too little labor in those firms and
35:58in those industries. Now, this might run
36:00into what I was alluding to earlier,
36:01which is can the planner really improve
36:03on it? And it turns out they can. If the
36:06planner knows this set of firms or this
36:07set of industries is plagued more by
36:10asymmetric information than the than
36:12another one,
36:13it it if it subsidizes the the high
36:15efficiency wage firms and industries,
36:17that'll actually improve allocations.
36:19Even though it it's kind of
36:20counterintuitive when I first started
36:21seeing people do this, but but uh, but
36:24it's true in in in it's some of those
36:26models.
36:27Okay, so lots of government policies
36:29like licensing, entry, land use
36:31restrictions. Uh this is what I was
36:33alluding to earlier to the question.
36:34There could be under or over investment
36:36of of of public spending on things like
36:38R&D and infrastructure that itself could
36:40lead to misallocation. And then any
36:43externalities, positive or negative, or
36:45you know, something that lead to
36:46misallocation. Once I start listing
36:48this, you see what I said at the outset,
36:50which is misallocation is just what
36:52economists do is we study the efficiency
36:54of the allocation of resources. So
36:56that's why basically any decision that's
36:57made, I'm saying that could be
36:59misallocated. And we have to think about
37:01why it might be and quantify the effects
37:03of that when we want to think about how
37:05we could improve uh welfare, improve
37:08policy.
37:09So I have a question uh related with
37:12this um under over investment of R&D.
37:15Previously you mentioned that
37:16misallocation is not the result of being
37:19successfully I mean
37:21to develop successfully technology or
37:23not. But as I see, if you look at the
37:25thing at the moment of as a firm at a
37:27moment of time,
37:29you will tend to I expect that you will
37:32tend to value as more productive firms
37:35that were successful
37:37into the
37:39into developing some new technology or
37:40whatever. So if you look at the picture
37:42like at the at the picture of at the
37:44moment of time, you'll say, "Okay, so
37:47the idea will be to shift some some um
37:51some resources from this firm to this
37:52other firm, but what you're doing is
37:54observing just a successful a successful
37:58uh entrepreneur, let's say.
38:01That you cannot know uh ex ante.
38:04You can't know ex ante. So if it
38:07if it there's adjustment costs that that
38:09that like the market like didn't know to
38:11allocate more resources to Elon Musk or
38:14cuz they didn't know how how successful
38:17we still don't know. There's still a big
38:18big bet there on how how successful
38:20Tesla will be, but you're saying we
38:23don't know ex ante ex post some of these
38:25outcomes are, you know, much better than
38:27others. Those that I have to that is a
38:29version of the adjustment cost story of
38:30like we couldn't have known ahead of
38:31time. It's real information realized
38:33after we had to make these loans and
38:35these this allocation of resources. Some
38:37of them will pay off higher than others.
38:39But can
38:41can can be adjusted for that? I mean, is
38:43is it been done like to look, I don't
38:45know, a sequence of firms doing
38:47investment and to check whether
38:50some of them is allocation in in T plus
38:5210 is just story of being unlucky.
38:56Um I would say yes in the sense that,
38:58you know, when people have said
39:00adjustment costs should have very
39:01transitory differences in the marginal
39:03products of capital, whereas things like
39:06fine in financing frictions, too, for
39:07that matter, whereas things like markup
39:09differences might be more persistent or
39:11could be more persistent. So, one of the
39:12facts I was going to touch on is that
39:14there's lots of evidence, going to be
39:15one of my five facts, is there's lots of
39:16evidence that most of the TFP R
39:18dispersion is persistent. So, that would
39:21say
39:21Okay.
39:22Yeah, then then, you know, we figured
39:24out this was a good entrepreneur, we
39:26should have allocated them more
39:27resources by now. This is, you know,
39:29five, six, seven years worth of seeing
39:30the firm continually. But it but that is
39:32really important what you said is to
39:33separate out to the best we can the
39:36transitory stuff that that we couldn't
39:38have allocated resources better ex ante
39:41versus the, you know, ex post versus
39:43kind of stuff versus the stuff where
39:45it's like, well, it looks like it's
39:46persistent, so maybe the market should
39:47be allocated more.
39:49Yep.
39:50So, some data I'm checking it's actually
39:53goes with the line what you say. Usually
39:56I see like a lot of persistence in being
39:58successful doing R&D. So, yeah, I guess.
40:01Thank you.
40:02Yeah, yeah.
40:04Um okay, I just wanted to tell you about
40:06this one study with um King, Tashay,
40:08Eric Hurst, and Chad Jones. Um we were
40:11just looking, you know, facts that are
40:13qualitatively very familiar to people,
40:15which is that you look to 1960 and and
40:17we had these high-skilled occupations
40:19which are defined either by education or
40:21wages there they look very similar. We
40:23excluded social scientists that seems
40:25too self-referential and claiming that
40:28they were high-skilled. We are
40:29high-skilled. And then in the United
40:31States in 1960,
40:33you might wonder where where are
40:34Hispanic in this? Their number they're
40:37not they're in the black. There's black
40:38Hispanic that are in black and there's
40:40white Hispanic that are that are in in
40:42the white as well as the non-Hispanic.
40:44Because it's not estimated in a
40:46consistent way across time in the United
40:48States. We couldn't break that up
40:50continuously. So it's just it's just
40:52included in the white and black
40:53respectively. And then other groups like
40:56Asian and Native Americans um
40:58Asian Americans and Native Americans,
41:00those are too small to kind of go back
41:01to 1960 and look at these fractions.
41:03It's the especially Asian Americans have
41:05grown a lot, but that's why the study
41:07focused on on white versus black. Okay,
41:09so 1960
41:11you can see white men were about 40%
41:14well So white men were were over 90% of
41:17these occupations even though they're
41:18only 40% of the the population. So you
41:21can see of white men almost 20% were in
41:23these high-skilled occupations and just
41:25far fewer white men and black men and
41:27black women even as a share of full-time
41:29workers among these
41:30[snorts]
41:31so even this doesn't even take into
41:32account fact that that women were much
41:34less likely to be in the formal labor
41:36force. Even the ones who were full-time
41:38so ignoring part-time, ignoring out of
41:39the labor force, they were just much
41:41less likely to be in these high-skilled
41:43occupations. That's changed
41:44dramatically.
41:46This is even more recent than we had in
41:47the paper, but in the paper we stopped
41:49at 2010. It was a 50-year period, but
41:51most of that occurred in that 50-year
41:52period. Just dramatic equalization. It's
41:55not perfectly equal. There's plenty of
41:57room for further equalization. But our
41:59interpretation of this is the efficient
42:01thing would be for all these numbers to
42:02be the same. They were very
42:04inefficiently different in 1960
42:06suggesting differential opportunities
42:08either to acquire skills or to once they
42:11hit the labor market to get those jobs
42:13or in the 19 in by 2010 or 2018 there
42:17might have still be discrimination but
42:18much less.
42:20And so
42:21the thing that we're adding relative to
42:22an enormous literature that was
42:24documenting these differences in
42:25occupations and wages within and across
42:27occupations is asking what's the what's
42:30the the output impact of this? What's
42:32the efficiency effect of this? So we we
42:35put these wedges in that hit um women
42:37and black Americans um more than than
42:40white and then we have them fade over
42:41time and they're occupation specific so
42:43they're in you know big barriers to
42:45becoming a lawyer or a doctor and lower
42:48barriers to becoming like a nurse or a
42:50teacher or say for women.
42:52Um
42:53and [clears throat] because of
42:54comparative advantage that's a critical
42:55ingredient in our model it really
42:57matters who's the doctor and who's the
42:58nurse, who's the teacher and who's the
43:00the lawyer because people have different
43:02skills for that.
43:03And you might [clears throat] ask how do
43:04we know that?
43:05Well, we use the fact that there's a lot
43:07of dispersion in wages within
43:09occupations. And you might ask further,
43:11well maybe that's absolute advantage.
43:13Maybe the person who's an awesome doctor
43:15would have been an awesome nurse as well
43:18so it really doesn't matter whether that
43:20person became a doctor or a nurse, you
43:21could switch them around and it wouldn't
43:23matter. But one of the things that labor
43:24economists say is that people who
43:26transition between occupations um are
43:29are disproportionately the lower wage
43:31people within every occupation. That's
43:32suggesting the people with high wages in
43:34that occupation have a comparative
43:35advantage in that occupation.
43:38So then it matters um
43:41who who is the doctor and who is the who
43:42is the nurse and so because of that we
43:44get enormous gains from this
43:46equalization of opportunities. And a
43:48bunch of that when we look at GDP per
43:50worker it's smaller and you might say
43:52how's it so much bigger here because a
43:54bunch of it is due to rising labor force
43:56participation rates for women. We're
43:58saying women face less discrimination in
43:59the labor force so that caused them to
44:01participate more in the labor force and
44:03that generated some of this rising
44:05workers per capita that fed some of the
44:07US growth. And but even GDP per worker
44:09we're saying 1/4 of it over a 50-year
44:12period. Um you could ask you know, is
44:14this repeatable? Um on our metric we
44:17only say only another 10% increase in
44:19GDP per worker can be achievable, but
44:21we're missing lots of things like in our
44:23model you know how how good you are.
44:26And so you know um whether you should be
44:30a doctor and and the best ones the best
44:32women and black Americans in terms of
44:35their talent to be doctors still become
44:37doctors even in 1960, whereas in 2010 we
44:40get more of the marginal ones in that
44:42are better than the white men who are
44:43crowded out. But there's an efficiency
44:45gain to that. Um and they accumulate
44:47more human capital cuz there's less
44:48barriers to that along the way.
44:50But that framework kind of ignores
44:52differences in just opportunities early
44:54on. You know, are we missing people
44:56who could have been a doctor and
44:58you know, Jennifer Doudna um um one of
45:01the inventors of CRISPR. You know, are
45:03there Jennifer Doudnas we've missed
45:04because they just never were encouraged
45:06anywhere along the way or given
45:07opportunities to become to go down the
45:09research path. So our model could be
45:12underestimating both the gains we've
45:14experienced historically and the future
45:16gains, especially if you look at
45:17research cuz that's that's uh looks more
45:20like 1960 uh representation by race and
45:24gender uh rather than 2018
45:27distribution by race and gender. So it
45:29looks like there's you know, if you
45:30count inventors or you count professors
45:32inventors like Pat Enderle looks like
45:34there's tremendous opportunities for
45:36equalization um there that could that
45:39should make for more more research and
45:41more high high quality research output.
45:44Okay.
45:45So um so that was a bit of a a
45:47digression into hey, let's think about
45:49instead of allocating inputs across
45:51firms or plants, let's think about which
45:53workers do what. Um but there's lots of
45:56areas that people have already started
45:57to do to to to in this direction. Um so
46:01you know,
46:02discrimination by gender not only in the
46:03US but other places and of course
46:05discrimination on things like like caste
46:07or or other observables.
46:09Okay. Um,
46:11let me skip over this. This is I guess
46:13this is just talking about under what
46:14circumstances
46:16TFPR reflects the average products and,
46:18you know, pointing out we need some
46:20strong assumptions to go from average
46:21products to marginal products. We need
46:23Cobb-Douglas and then go from marginal
46:24products to marginal revenue products
46:26you'd need equal markups.
46:28Um, okay. Let me move on from that.
46:31So, one aside is just to say, and I
46:33think this came up in Richard's talk,
46:35which is what about gross output versus
46:37value added? So, one thing is you get
46:39smaller TFPR dispersion if you do it
46:41relative to gross output than value
46:43added, but distortions are amplified.
46:45Chad Jones has a couple nice papers.
46:47Ernest Liu has a nice recent paper.
46:50Um, Bakkain Farhi already mentioned
46:51their very nice paper. And that's a
46:53bunch of what my paper with with uh
46:55Kean and Mark Bills is about.
46:58So, part of it is that intermediate
47:00there are multiplier like Chad said and
47:01part of it is this double
47:02marginalization. That's in Ernest Liu's
47:04paper. He emphasizes distorting upstream
47:07firms and sectors is really damaging
47:09relative to distorting like final goods
47:12sectors.
47:13Um, so and that's an ingredient in
47:15Bakkain Farhi, too, that I'll return to.
47:18Okay. So, now I can do
47:20um, in the remaining parts of the the
47:21main lecture part I want to make sure I
47:22talk about um, some patterns I see in
47:25this mess up in the recent papers on
47:27misallocation. And then if I have time
47:29I'll talk about some data sets, but
47:30they'll be in the slides as a resource
47:32and also if I have any more time I'll
47:34talk about some some more recent papers
47:36in deep more detail.
47:37Okay, but I wanted to get out these five
47:39five key facts. So, one,
47:42which I've already kind of implicitly
47:43said, but I want to be explicit about,
47:45is that people are consistently finding
47:46more dispersion in this TFPR in
47:49developing economies than in advanced
47:51economies and in the richer economies.
47:53Um, and that's consistent with
47:54misallocation not only being large and
47:56important, but being a bigger
47:58uh problem in developing countries than
48:01in advanced economies. So, it's a
48:02problem everywhere, but it seems a
48:03bigger problem in developing countries.
48:05So, that's what makes it
48:07a very much a macro development
48:08question. But, it helps us understand
48:10Yep. I have a question. Sorry for
48:12interrupting. Um what about um
48:15uh this misallocation um by size? Like
48:19um
48:20if you split like very big firms from
48:22small firms and you compare countries,
48:24I'm curious about it.
48:27Um I mean, I'm going to touch on that
48:29cuz I'm going to say uh TFPR is strongly
48:31increasing in TFPQ and I'm going to
48:33argue that's going to make firms much
48:35smaller and plants much smaller in
48:37developing countries. Um
48:39that the fact that this is steeper in
48:41developing countries than in rich
48:43countries might be contributing to why
48:44average firm size is so different. Is
48:47that what you're alluding to?
48:48Uh
48:49this The thing is that that I really was
48:51discussing with a professor something
48:52that happens in Spain, that when you
48:54look at firms that are above 50 workers,
48:57uh things in productivity really look
48:59like other European countries. But, when
49:01you look at the below 50 workers, you
49:03know, it's where things get really
49:04messy. So, I was wondering if um
49:08if um
49:09you have some idea of how uh developing
49:11countries and uh developed countries
49:13look like in terms of the relative
49:16I mean, possibly relative misallocation
49:18between big firms or small firms.
49:21Yeah.
49:23Yeah, I think this is going to be
49:23related to the the fact too. So, let me
49:25let me come back to that shortly. And
49:27then, I I mean, I think discussion about
49:28why that might might generate some of
49:31the facts that you're talking about.
49:32Okay, thank you.
49:33Okay. And then, most dispersion is in
49:35this fixed effect. So, as much as we
49:38care about, you know, misallocation of
49:40capital relative to labor or labor
49:41relative to intermediates, etc., it
49:44looks like there's a common component is
49:45quite important. And in this we touched
49:48on already, much of the dispersion, the
49:50cross-firm dispersion, is very
49:51persistent. That it's not like if the
49:52firms are IID in terms of which ones are
49:54the high and low TFPR, that the majority
49:57of the variance seems to come from a
49:58component that's like a plant or firm
50:00fixed effect. And then, the last fact is
50:02that much of the dispersion in the
50:04average product of capital seems to be
50:06occurring even within firms. You know,
50:08like especially in the United This
50:09matters by country, but in the United
50:10States where you have a lot of
50:11multi-plant firms in manufacturing, then
50:15you can you can analyze this and say,
50:17you know, it's it's most of it, you
50:18know, if we had a single-plant firm
50:19everywhere, then it this wouldn't be
50:20meaningful. But, um in the United
50:23States, there's a lot of multi-plant
50:25firms, and we see a lot of dispersion in
50:26their average mar- average product of
50:28capital. So, then that becomes
50:29interesting. What can we learn from
50:30that?
50:32Okay, I think
50:33uh
50:34Richard already showed you this plot,
50:35which is just TFPR dispersion in the US
50:37versus China versus India, so that it's
50:40um
50:41significantly smaller. And so, if you do
50:43like He also showed you a table of like
50:45the the 90-10 ratios. So, you want to
50:48think about 90-10 ratios that might be
50:50like a, you know, a factor of two or
50:52more. So, if say if the 10% marginal
50:53product of capital at the Let's say
50:56let's do the 25th versus 75th
50:57percentile. So, the 10% marginal product
50:59of capital at the 25th percentile, there
51:01might be a a 23% marginal product of
51:04capital at the 75th percentile firm. So,
51:06we don't need these enormous differences
51:08in marginal products of capital or
51:10marginal products of labor similarly. If
51:11like thing same worker had a marginal
51:13product of labor worth $10 an hour at
51:16one firm, but worth $23 an hour at
51:18another firm. And that's like the inter-
51:20quartile range of of firms that we see.
51:22Um we don't need an enormous amount to
51:24generate a lot of misallocation. Of
51:26course, that that mat- what matters also
51:28is that how substitutable these firms'
51:29outputs are, but um as I mentioned
51:31earlier.
51:33Okay, so
51:34so, when people add it up, there's
51:35Bakkai and Farhi, I was quoting a 40%
51:37number for the US. They get a 20%
51:39number, maybe 25% depending on the
51:41version you look at. They were looking
51:43at these very big firms. So, the 40%
51:46number I was quoting was for like
51:47manufacturing which had a bunch of
51:48smaller firms in it and was only
51:50manufacturing. So Compustat has a lot of
51:52non-manufacturing, but it's also the the
51:54publicly listed firms. So, I'll dwell on
51:57this a bit, you know, why it's so big in
51:59a bit, but we found something bigger for
52:01manufacturing, but then
52:03um even bigger misallocation in China
52:06and India relative to the US. And I
52:07think people touched on it, these things
52:09changes over time.
52:10As as Richard said, we found a lot of
52:12reduction in misallocation in China
52:15that we found that it seemed to account
52:17for about a third of their growth, their
52:18rapid growth in manufacturing over the
52:20period we were looking at. And about 40%
52:23of that was due to um state-owned
52:26enterprises that were very inefficient,
52:28had very low
52:29uh TFPR either exiting or being
52:31privatized and shedding a bunch of
52:33inputs and their TFPR coming back up
52:35toward the toward the mean.
52:37So, China's kind of the best case
52:39because we can see it actually change
52:42over time, so it doesn't look like a
52:43pure measurement problem, and we can see
52:45it change. We see the sources of it seem
52:47kind of sensible, which is these
52:48state-owned enterprises. And when the
52:50state-owned enterprise faded, it looks
52:51like the misallocation, you know, it
52:54would diminish as well. And and they
52:56experienced rapid growth over that time
52:57period, too. So, it all seems to hold
52:59together nicely in China. India less so
53:02because it looks like there's massive
53:03misallocation. They've grown pretty
53:05well, but it doesn't look like the
53:06misallocation as we measure it has
53:08actually fallen in in manufacturing,
53:10anyway. So,
53:12the Inter-American Development Bank has
53:13a nice
53:14um a nice set of chapters in one of its
53:17one of its books
53:18uh that looks at the same types of
53:20calculations in in a number of Latin
53:21American countries. What they
53:22consistently find is a lot of
53:24misallocation compared to what people
53:26find in the US.
53:27Michael Peters has a nice paper on
53:28Indonesian manufacturing where he
53:30focuses on market dispersion as the
53:32source of this um and has an end theory
53:34of endogenous market dispersion. It's
53:36quite nice, as well as endogenous
53:37growth.
53:38Um and then Diego Restuccia with
53:40co-authors has has many papers, but just
53:43a couple of them I'll mention here on on
53:45agriculture in China and Malawi that
53:47find even larger losses in agriculture
53:50than in manufacturing. So, we have a
53:51little bit of this, you know, the the
53:53drunk looks under the light light post
53:55because that's the only place they can
53:56see, they look for their keys there. We
53:58have a little bit of that problem in
53:59this literature that
54:01most readily available data is
54:02manufacturing, so the vast majority of
54:04studies are about manufacturing. But, it
54:06could be there's more misallocation
54:08outside manufacturing. So, that's what
54:10Diego is finding with his collaborators
54:12in agriculture. Chang and I did
54:14something for Mexico that found
54:15something similar in the in the service
54:17and retail sectors, we found more
54:18misallocation
54:19um in Mexico than in the Mexican
54:22manufacturing sector. So, it's it's kind
54:24of an uh
54:26a frontier we've only only started to
54:28explore is what are things look out like
54:30outside manufacturing. And why might it
54:31be different? It's interesting, too.
54:34Okay, so an aside on on on Harberger
54:37triangles. So, like when I I mentioned
54:38that when I first saw
54:40Restuccia and Rogerson's big you know,
54:42like you tax productive firms and you
54:44subsidize unproductive firms, you get 30
54:47to 40% loss in GDP. I was like, "Whoa,
54:49how how is that possible?"
54:51Um I remember the Harberger triangle,
54:53which was like monopoly distortions
54:55aren't that bad. You know, they reduce
54:57GDP some fraction of 1%, maybe well
54:59beneath 1%.
55:02So, one of the things that's beautiful
55:03about this Buera and Farhi paper, it's
55:05got a lot of nice things in it. Like I
55:06said, it's got general input-output
55:08structures and nice decomposition
55:10decomposition into technological change
55:12versus changes in allocative efficiency.
55:14It's got a lot of great things about it.
55:16On top of that, it says, "What about
55:18this Harberger number? How can we get
55:1920% more or loss of GDP from
55:23misallocation and markup dispersion and
55:26and Harberger got far less than 1%?" So,
55:29they decompose it for us and they say,
55:31"Look,
55:32wedges or these towels within industries
55:35are much bigger across firms than they
55:37are when you average them and look
55:38across industries. So, you average out a
55:40lot of this dispersion when you look
55:41across industries.
55:43And when you look across industries,
55:45industries just aren't that
55:46substitutable, right? Compared to like,
55:48you know, Samsung versus Apple
55:50smartphones, those are highly
55:51substitutable and that and the wedge
55:53differences between them are
55:54substantial. Whereas if you say, "What's
55:55the average wedge difference between
55:58steel and cars?" It might not be that
55:59big and they're just not that
56:00substitutable. So, you get much smaller
56:03losses from those smaller wedges between
56:06less substitutable
56:07industries than you get, you know, big
56:09wedges within industries and lots of
56:11substitutability within industries. And
56:13then they show propagation from the
56:15input-output matrix matrix, which which
56:17Harberger didn't have.
56:19So, this is a very nice kind of saying
56:22why the recent literature is getting
56:24such big numbers, whereas Harberger got
56:27small numbers.
56:28Okay, and then there's also I don't know
56:30don't know where to put this because
56:31this also fits into dynamic effects of
56:34misallocation, but Jim Schmitz has been
56:36writing a a sequence of papers that are
56:37really fascinating about about
56:39misallocation due to monopoly in the
56:41United States. And it's really
56:43fascinating because it's it's not like
56:46this, you know, "Hey, it's like a tax on
56:48the firm and the firm just responds
56:49optimally to it." You know, it's it's
56:51markup is its own tax, so you can think
56:53of not literally a tax, but like a firm
56:55deliberately just reducing its output
56:57and raising its price. He's saying, "No,
57:00they do things like put in work rules
57:02that are inefficient as a way to
57:04redistribute that they're distribute
57:06some of the rents that are within these
57:07monopolies that end up destroying
57:09value."
57:10So, he's got lots and lots of examples
57:12of this happening uh partly due to
57:15government regulation, partly due to,
57:17you know, um industry market power. Um
57:20so, partly due to sometimes to, you
57:21know, union power and union decisions.
57:24But he describes it as a byproduct of
57:26monopoly power. So, he's basically
57:28saying monopoly power can do much more
57:30damage if you want if you carefully
57:32understand what's happening in these
57:33industries than our simple models
57:35suggest. So, I highly recommend tracking
57:38down all this stuff on this.
57:40Okay. So, I'm
57:42There's more TFPQ dispersion in India
57:44than China and the US. I won't talk
57:46about that much more. I want to talk
57:47about a couple of other things.
57:49Um okay. So, fact two is TFPR is
57:52increasing in TFPQ and more so in
57:55developing countries. So, like in India,
57:57we found a really steep relationship.
57:59This is TFPQ of a firm relative to the
58:01or plant relative to the mean. Here's
58:02its TFPR of the plant relative to the
58:04mean. You can see just a really positive
58:07slope. For every log point here, you've
58:09got like half a log point here. So, the
58:11elasticity is is about 0.5. The
58:13elasticity is still positive in the US.
58:15It's about 0.1, but that's dramatically
58:17lower.
58:18So, then the question is why? Why is
58:19TFPR so steeply increasing in TFPQ in
58:23India versus not the United States? Is
58:25it because if you try to get big in
58:27India, there's bigger transportation
58:29costs or there you know, the the
58:31transportation system doesn't work as
58:32well to try to scale up the size of your
58:34firm?
58:35Is it Nick Bloom's stuff of you you
58:37know, you don't trust
58:38people outside your family to to run
58:40plants or businesses that are at other
58:42locations from where you are? This is
58:44also in some work by Akshidan
58:46um
58:47Akshit Alp and and Peters.
58:50Um but people have built these into
58:52models like Bento and Restuccia did and
58:53said, "Look,
58:55if firms know that exposed, like say
58:57imagine this is tax rates on the formal
58:59sector and low tax rates on the informal
59:01sector, just to be to have one explicit
59:03example,
59:04then that will make you invest less in
59:06high quality that would make you big and
59:09make you a target of those formal sector
59:10taxes. So, that's just one force that
59:13would push you toward
59:14being informal. Another could be if
59:16there's big barriers to being informal
59:18to being formal and and big that and
59:21those firms are intensive in skill,
59:22maybe demand for skill isn't very high,
59:24so maybe supply for skill isn't very
59:26high. So, that's
59:27and so then you have a lot of low
59:28skilled workers on a lot of informal
59:29firms. So there were really nice paper
59:31for showing how you could have a lot of
59:33firms, you know, and Roberto Ferdal Jeff
59:36has a more recent AJ macro with that
59:38theme as well. There's on the big formal
59:40firms or the more successful firms wind
59:42up getting more firms of lower quality.
59:45So you can understand not only why firms
59:46don't get big, but why there's so many
59:48small ones endogenously as an outcome of
59:50that.
59:51And that's in some of Akeson Bursting's
59:53early work, too.
59:55Um but basically you can also argue this
59:57is pre-entry. Um this is entry. This is
1:00:00you know, this is quality of entrants.
1:00:02This is quantity of entrants. There's
1:00:03also what you do post-entry. You know,
1:00:05do you invest in getting better and
1:00:07expanding?
1:00:08You know, like in the United States
1:00:09there's a survey of like when you start
1:00:11a business, why did you start a
1:00:12business? An overwhelming majority is
1:00:14like, I just want to be my own boss. I
1:00:17just want to control. It's not like I
1:00:18have an idea and I want to scale up,
1:00:20right? But I suspect if you did the same
1:00:23survey in many developing countries, the
1:00:25fraction who are like who just say,
1:00:26well, this is the only opportunity I
1:00:28have and this is the best, like the
1:00:29subsistence entrepreneurs, this you
1:00:31know, I I can't get a formal sector job,
1:00:33I don't have the skills or whatever. I
1:00:34think that fraction might be
1:00:35dramatically higher. So I'd be
1:00:36interested in seeing those same surveys
1:00:38done
1:00:39elsewhere. Um but these papers are
1:00:41trying to go in that direction of trying
1:00:43to figure out whether this, what looks
1:00:45like a static distortion, has lots of
1:00:47dynamic implications for human capital
1:00:49investment, for firms' expansion
1:00:51post-entry, for the types of firms that
1:00:53are created and to begin with and how
1:00:55many firms are are created to begin
1:00:57with. So these these papers are they're
1:00:59going down that path, but there's plenty
1:01:00of plenty of road ahead, I think, to
1:01:02keep pursuing that and and refining our
1:01:04understanding of these
1:01:06these patterns.
1:01:08Okay. Um this is one thing I was
1:01:09alluding to is that post-entry, US is
1:01:12kind of an extreme. Post-entry in the
1:01:14US, they really expand. And half of this
1:01:16is like incumbents growing and half of
1:01:18it is
1:01:19is uh um firms exiting.
1:01:22Um and Mexico is intermediate, India's
1:01:26uh extreme in having almost no
1:01:27post-entry growth.
1:01:29If you start filling in with more
1:01:31countries, the US is an outlier on the
1:01:32upside and India's an outlier on the on
1:01:34the bottom side. So, a lot of Western
1:01:36European countries are in here, for
1:01:37example. Um so, the US looks unusual in
1:01:40how how big the firms get, and India's
1:01:42unusual in how small the firms stay.
1:01:45But, it's interesting that there's
1:01:46variation in this post-entry behavior
1:01:48across economies we need to understand.
1:01:50How much of this reflect, you know,
1:01:52different misallocation in these
1:01:54economies?
1:01:55Okay. So,
1:01:57um I was mentioning the fixed effect
1:01:59component. I'll call that the scale
1:02:01distortion. And then, distortions in
1:02:04capital relative to labor, I'll call
1:02:06those mixed distortions. I think this is
1:02:07terminology that Joel David and and uh
1:02:11Venky Venkatachalam um
1:02:14uh
1:02:15introduced scale versus mixed
1:02:16distortions. Um they find it's mostly
1:02:19scale distortions. And my paper with
1:02:20with um Kean and Mark found the same
1:02:22thing. And then, we that that's helpful
1:02:25because it says financial frictions may
1:02:26be important, but maybe that's the not
1:02:28the dominant source of TFPR variation.
1:02:30Maybe you need something common like
1:02:32markups, measurement error, um but also
1:02:34like tax and subsidy-type behavior. So,
1:02:36I think that's useful in guiding future
1:02:38theorizing and future empirical work.
1:02:41So, the persistence I already touched
1:02:43on. Um David and Venky found that was
1:02:45the biggest component in China and
1:02:47India. I'm sorry, China and the US. We
1:02:49found it in our study of the US and uh
1:02:52India and the US manufacturing. And
1:02:54again, to me that's that says maybe not
1:02:56financial frictions, which would be
1:02:58transitory because you can save your way
1:03:00out of financial frictions,
1:03:02um or adjustment costs, which should be
1:03:03transitory as well. So, those might be
1:03:06important, but they might not be as
1:03:08important as things like markup
1:03:09differences or or tax and subsidy
1:03:12differences across across these firms
1:03:13and plants.
1:03:16Okay. And then, lots of stuff is within
1:03:19firm and one reaction to that, my first
1:03:21reaction to that was, well, it can't be
1:03:23financial frictions because if the
1:03:25variation in the in the average product
1:03:27of capital is across
1:03:29plants within firms, then firms should
1:03:31should have an efficient capital market
1:03:33within the firm. They should equate the
1:03:35marginal product of capital across their
1:03:37firms.
1:03:38So, their explanation in this Caraglio
1:03:40Vincent paper, which is quite a nice
1:03:41paper,
1:03:42is actually it is financial frictions,
1:03:44but it's interacted with adjustment
1:03:46costs. So, the firm is saying, we can't
1:03:48raise external capital all right away,
1:03:51so what we're going to do is we're going
1:03:52to spread out like
1:03:55you know, upgrades of factor our
1:03:56factories. If we have multiple
1:03:57factories, we can't upgrade them all at
1:03:59once, you know, even if that's the most
1:04:00profitable thing to do with no financial
1:04:02frictions, we're going to space it out.
1:04:04So, spacing it out ends up generating
1:04:06more market dispersion, but they're
1:04:08saying, hey, it's it's kind of good
1:04:09dispersion because at least it
1:04:10economizes on some of those external
1:04:12financing costs and adjustment costs as
1:04:15well.
1:04:16Um
1:04:17So, but still there's other
1:04:18possibilities for this pattern. There
1:04:20could be differences in production
1:04:21technology that are persistent, you
1:04:23know, there's some capital intensive
1:04:24plants and less capital intensive
1:04:25plants.
1:04:26Um and that would be consistent with why
1:04:28a lot of it's persistent, whereas it
1:04:30wouldn't be very persistent if it was
1:04:31due to adjustment costs. Um and there
1:04:33also could be markup differences. A
1:04:34maximizing firm could just choose
1:04:37to, you know, for one one firm that's
1:04:39got more market power, one plant within
1:04:41the firm produce the product with more
1:04:42market power than another, they're going
1:04:43to persistently have higher value of
1:04:45their average product of capital.
1:04:48Um and of course, there's the
1:04:49possibility there's just misallocation.
1:04:51Like corporate finance literature is
1:04:52rife with the idea that managers don't
1:04:55perfectly allocate resources across
1:04:57their activities. And so, it could be
1:04:59multi-plant firms have, you know,
1:05:01misallocation just like the economy does
1:05:03across different firms.
1:05:05Okay, so I think I'll
1:05:08I'll skip some of the discussion of um
1:05:10some of the data sets, but that that's
1:05:12in the slides as a resource for you to
1:05:14to look at um and just start talking
1:05:16about some more of some of my favorite
1:05:18papers in the literature. I've already
1:05:20touched on some of them. Um some of my
1:05:21favorite recent papers, but I wanted to
1:05:24I think I have like 8 minutes left for
1:05:25for the presentation and then we open up
1:05:27for Q&A. So, let me do a little bit of
1:05:29that um
1:05:31of some of that literature. So,
1:05:34[clears throat]
1:05:35Okay.
1:05:36The one paper that I think is quite
1:05:38important. I'd like to see it updated
1:05:40and published in a nice place, which is
1:05:42saying, "Let's dive in and figure out
1:05:44how this data is created." And this
1:05:46paper says, "Look, the US Census Bureau
1:05:49cleans the data a lot." So, maybe that
1:05:51lack of dispersion in the US was just a
1:05:53lot of data cleaning before we
1:05:54researchers ever got it. And maybe in
1:05:56India they don't do that. That's the big
1:05:58question mark. Imagine India didn't do
1:06:00any cleaning and the US did a lot of
1:06:02cleaning. What they did, which is really
1:06:04cool, is said, "Let's look at the raw US
1:06:06data." And the raw US data has has like
1:06:08crazy dispersion, so it's like it would
1:06:11look like more dispersion there. And
1:06:12they're like, "Well, let's throw out
1:06:13some of the obvious errors in there and
1:06:15assume the Indian data at least does
1:06:16that." But they they get, you know, not
1:06:19as they they don't get as much of a
1:06:20contrast between the US and India
1:06:22anymore. They still get a contrast, but
1:06:24it basically widens US dispersion.
1:06:27Now, there's another force to be careful
1:06:29of though, which is that the Indian data
1:06:31that they're looking at and that I
1:06:32looked at with Shay was firms that are,
1:06:35you know, 10, 20 employees or more. Um
1:06:39and it turns out the majority of of
1:06:41workers in manufacturing in India are
1:06:43informal plants that are actually
1:06:45smaller than that. So, this is a nice
1:06:47paper for illustrating let's really
1:06:49think about differences in these data
1:06:51sets and the in the sampling of these
1:06:53data sets across countries. I remember
1:06:55Haltiwanger when he first saw Chang and
1:06:57I do China and India and the US was
1:06:58like, "You can't do this. These are
1:07:00different countries with different data
1:07:02sets. You know, it's a it's firm level
1:07:03data in China and then private firms
1:07:05above a certain threshold. It's a census
1:07:07of all plants in the US. And it's
1:07:10you know,
1:07:11a sampling of large, you know, medium to
1:07:14larger firms in India. And the
1:07:16underlying size distributions are
1:07:18different, so the sampling coverage is
1:07:19different everywhere.
1:07:21Um we went ahead and made the
1:07:22comparison, but his point is well taken,
1:07:24and this paper
1:07:25uh embodies that. Let's think carefully
1:07:27about what's in this what's in these
1:07:28data sets and how we can compare them.
1:07:30So, definitely keep that in mind.
1:07:33Um so, my paper with Mark and Keen
1:07:36So, we're, you know, worried about
1:07:38measurement error. And so, we came up
1:07:40with
1:07:41what we think is an interesting way to
1:07:43um to test for it, which is basically,
1:07:46you can think of it as as in the
1:07:47cross-section,
1:07:49you know, some firms have a lot of
1:07:50revenue relative to inputs,
1:07:52[clears throat]
1:07:52but what do the first differences look
1:07:53like? If you have panel data, does the
1:07:55change in revenue divided by the change
1:07:57in inputs that look a lot like the, you
1:07:59know, the high firms with high revenue
1:08:01relative to inputs, do they have a high
1:08:03change in relative growth change in
1:08:05revenue relative to inputs?
1:08:07Or if it was purely measurement error,
1:08:08there might be no connection between
1:08:09those two.
1:08:10And we embody that with a regression of
1:08:13of revenue growth on input growth, which
1:08:15don't think this is supposed to be
1:08:16structural. It's supposed to cuz the
1:08:18input growth is endogenous across firms,
1:08:19but it turns out, you know, in the in
1:08:22the presence of a lot of measurement
1:08:23error, um then this coefficient should
1:08:25be low when TFPR is is artificially high
1:08:28due to measurement error.
1:08:30And we found that relationship to be
1:08:31negative in the data, suggesting there's
1:08:33measurement error. And over time in the
1:08:35US, it's increasingly so.
1:08:37So, we argue that there's growing
1:08:38mismeasurement uh in the US, and maybe
1:08:41it's due to things like,
1:08:43you know, compliance with the survey is
1:08:45going down. You know, firms and the same
1:08:47thing that makes fewer firms respond may
1:08:49make them put less effort into the
1:08:51response. So, there may be less careful
1:08:53responding in that survey over time, so
1:08:55growing measurement error in the US.
1:08:57Um so,
1:08:59that's actually amazing because we
1:09:01if naively we it looked like allocative
1:09:02efficiency was plummeting in the US and
1:09:04we get a modest decline, still
1:09:06economically interesting but a modest
1:09:08decline.
1:09:09And um yeah, so that's
1:09:12that's uh our attempt to go straight at
1:09:14the measurement error explanation which
1:09:16would be the most undermining and I
1:09:18think of a
1:09:19I think it's the most undermining of the
1:09:21possibility of misallocation because it
1:09:22says it's just not there at all.
1:09:24Adjustment costs are different as I as I
1:09:25hope I'll have time to say that in a in
1:09:27a couple slides. Yeah, here.
1:09:29So
1:09:30the
1:09:31This is alluded to multiple times by by
1:09:33Richard and earlier today. So imagine
1:09:36there's just adjustment costs for
1:09:37capital and there's idiosyncratic
1:09:38shocks. That's going to generate
1:09:40dispersion in the in the average product
1:09:42of capital and in marginal products of
1:09:43capital and there's nothing the planner
1:09:45can or should do about that necessarily.
1:09:48And so they provide some evidence for
1:09:50that that there's bigger idiosyncratic
1:09:51firm shocks in developing countries and
1:09:53that's leading to even with the same
1:09:55adjustment technology, that's leading to
1:09:57more dispersion.
1:09:59So I want to read so so I have two kind
1:10:01of reactions to that. One is is that it
1:10:04still contri- several reactions. One is
1:10:06this can still contribute to TFP
1:10:07differences. If they're right, then one
1:10:10explanation for the TFP low TFP in
1:10:12developing countries is they have bigger
1:10:14idiosyncratic shocks that interact with
1:10:17um adjustment costs. Again, we can't do
1:10:19anything about it but now we understand
1:10:20some of that difference and where it's
1:10:21coming from. So we still learned
1:10:23something I think it's valuable.
1:10:25The second thing is you can ask This is
1:10:27kind of similar to I think it was
1:10:28Mateus's comment about infrastructure
1:10:31might not be efficient. Where are these
1:10:33firms tech- I shocks coming from? Why
1:10:35are there more bigger idiosyncratic firm
1:10:37shocks in developing countries than rich
1:10:39countries? Is it because they're small?
1:10:42In which case this might be a byproduct
1:10:43of whatever keeps firms small, firms can
1:10:45be bigger, we might have less misalloc-
1:10:47or less dispersion in in in average
1:10:49products and maybe less less or maybe
1:10:51we'll be able to produce more output
1:10:52because of it. Um the last thing is
1:10:54caution about data. What you know, I
1:10:57mentioned single countries because we
1:10:59don't have
1:11:00we don't have country, you know, samples
1:11:02for 50, 100 countries that are really
1:11:04comparable and really high quality. So,
1:11:06they did the best they could using these
1:11:07World Enterprise Survey data so they
1:11:09could get many countries at once, but
1:11:11there's there's a lot of missing data in
1:11:12that this study. So, this is a question
1:11:15I think it's still needs further study.
1:11:18And David and Venky, maybe this is one
1:11:19of the last ones I'll mention in more
1:11:20detail.
1:11:22So, I mentioned that they found a big
1:11:23fixed effect, but they also kind of
1:11:26poured some cold water on the adjustment
1:11:28cost explanation.
1:11:30Um, and Richard alluded to this, which
1:11:32is they said, "Look, if if you think
1:11:34it's adjustment cost, what you should
1:11:35see is
1:11:36the marginal that the average product
1:11:38capital differences should be
1:11:39transitory, and the investment rate
1:11:41differences should be persistent. So,
1:11:42it'd be like,
1:11:44you know, I get a positive shock that
1:11:45pushes up my my marginal product of
1:11:47capital, it's going to take me a while
1:11:48to build the capital stock. So, you
1:11:50should see capital growing continuously
1:11:52in the average product coming down,
1:11:54marginal product coming down
1:11:55continuously. So, you should see the
1:11:57transitory behavior of the average
1:11:59product, but the investment rate should
1:12:00be serially correlated. They don't see
1:12:02much serial correlation of the
1:12:03investment rate. That wasn't a moment
1:12:05that Asker, Collard, and Westland that
1:12:07Asker looked at. When they do look at
1:12:09that moment, they don't have much role
1:12:11for convex adjustment costs.
1:12:13Then you could say, "What about
1:12:14non-convex adjustment costs, like the
1:12:15kind Carrigan and Vincent
1:12:17were were analyzing?" They said, "Well,
1:12:18then you should see a lot of inaction
1:12:20and a lot of volatility in investment.
1:12:21We don't see that either." So, in the
1:12:23end they they end up saying, "Yes,
1:12:25there's adjustment costs. Yes,
1:12:27adjustment costs contribute to some of
1:12:29the dispersion in these average products
1:12:31in the US and China, but it's not it's
1:12:34not much of it. It's like a sixth of it,
1:12:35rather than the dominant role that Asker
1:12:37et al. were giving it." So, they find
1:12:39some sort of other big fixed effect in
1:12:40this average product, and things like
1:12:42markup dispersion might be important for
1:12:44that.
1:12:45Okay, so, I already mentioned the the
1:12:47Carrigan and Vincent, I think.
1:12:49Um,
1:12:50and this is this is a paper by Amit Seru
1:12:53that's that's talking about whether this
1:12:54dispersion in TFPR, not just an average
1:12:56product of capital, but in TFPR more
1:12:58generally, across plants within the US,
1:13:01whether that may be due to
1:13:02misallocation. Maybe that that may be
1:13:03due to um problems with management
1:13:06multiple locations and divisions. I
1:13:08think I'll stop there. And uh let me one
1:13:11one other thing before I open up for
1:13:12questions, cuz this is I think is a nice
1:13:14example. Um
1:13:16the this paper says by Haltiwanger,
1:13:18Kulka, and Kuceviks and says let's relax
1:13:21what was in my paper with Shea, which
1:13:22was CES plus Cobb-Douglas. Um or CES
1:13:26plus constant returns to scale, really.
1:13:28They said, let's relax both.
1:13:30They end up estimating roughly constant
1:13:32returns to scale,
1:13:33but um
1:13:35they they relax the
1:13:37the CES, and so they get markup
1:13:39dispersion. So, one reaction to this is
1:13:41this is still misallocation. Just
1:13:43because it's coming from markup
1:13:44dispersion and not some tax wedge tau,
1:13:47some mysterious wedge, it's still
1:13:49misallocation.
1:13:50But one thing this is really valuable
1:13:52for, I think the the paper that I would
1:13:54have discussed too was by Edmund
1:13:55McGrattan and Rui asking how costly
1:13:57markups are.
1:13:58Once you endogenize it in a particular
1:14:00way,
1:14:01then think about these gaps in in
1:14:03marginal products. There's a high markup
1:14:04firm and a low markup firm. I mentioned
1:14:06the sigma was really important. And one
1:14:08way to think about sigma is if they're
1:14:10really substitutable, then it's going to
1:14:12take a lot of misallocation to to pull
1:14:14the the relative marginal products of
1:14:16inputs back in line. Whereas if sigma is
1:14:19really low, it's like a lot of
1:14:20diminishing returns, and so the gap as
1:14:22we reallocate from the the low marginal
1:14:24product to high marginal product, the
1:14:25gap closes quickly.
1:14:27So, with variable elasticity of
1:14:28substitution, the gap closes more
1:14:30quickly. For a given wedges you observe,
1:14:32given what looks like dispersion in
1:14:34TFPR,
1:14:35um the variable elasticity makes the gap
1:14:38close more quickly. So, there turns out
1:14:40the same wedges yield less misallocation
1:14:43in the aggregate. So,
1:14:45they were kind of looking at this as the
1:14:47as you know CES was a special case, but
1:14:49I what I see in this paper and also the
1:14:51effort made you're going to do is
1:14:53endogenous market dispersion
1:14:55and it matters if it's coming from
1:14:58something like variable elasticity of
1:15:00substitution because not only does it
1:15:01rationalize the market dispersion but it
1:15:03it says it has particularly smaller
1:15:06aggregate implications still large but
1:15:08smaller than you'd get out of just
1:15:10assuming taxes in a CES environment.
1:15:12Okay, I'll stop and open up for
1:15:14for questions.
1:15:16In general discussion.
1:15:18Um
1:15:20Can I ask a question?
1:15:21Mhm.
1:15:22Okay, so um this more of a broad
1:15:24question about the whole like
1:15:27literature on misallocation. So I see
1:15:29that there's a big emphasis on
1:15:32understanding aggregate aggregate TFP,
1:15:34right? And like my understanding is that
1:15:37aggregate TFP only matters in so far as
1:15:39it contributes to some measure of
1:15:41welfare like utility of consumption or
1:15:44something like that, right? And so when
1:15:46I look at some of these frictions, I
1:15:48think that maybe some people are
1:15:49benefiting from this, right? I mean if
1:15:51you look at financial frictions and some
1:15:53firms are able to default in context
1:15:56where
1:15:57they don't get punished because there's
1:15:58not enough like legal institutions or
1:16:01things like that. So is there a way to
1:16:03kind of try and tease out the the effect
1:16:05that these frictions have on you know
1:16:09the welfare of agents
1:16:10in a way, right?
1:16:13Yeah, no no papers are coming to mind
1:16:15immediately on that but that's it's I
1:16:17think that's an area to work on is to
1:16:19say, you know, there's there's some
1:16:21things that like the government may be
1:16:23favoring some firms versus others and we
1:16:25might not like cuz that might actually
1:16:26increase inequality like the the large
1:16:29wealthy individuals large firms wealthy
1:16:31individuals might be politically
1:16:32favored. That doesn't sound like it'll
1:16:33be good on redistribution grounds or
1:16:35efficiency grounds, but there could be
1:16:37other cases like you're illustrating
1:16:39where maybe tolerating inequality
1:16:41actually reduce it. Sorry, tolerating
1:16:43misallocation reduces inequality. So, I
1:16:46think about like Walmart coming in to
1:16:48Mexico, the gains were, according to
1:16:51David Atkin and collaborators, the gains
1:16:52were for everybody, even though the the
1:16:54the people whose jobs were wiped out in
1:16:56the informal sector, but the gains were
1:16:58smaller for people in the informal
1:17:00sector who lost their jobs, don't shop
1:17:02as much at Walmart and these other big
1:17:03box retailers. So, I think it's really
1:17:05interesting that could be like part of
1:17:06the reason these governments are doing
1:17:08that or saying economically they're
1:17:09saying let's go and wipe If I let
1:17:10Walmart in, it's going to wipe out all
1:17:11these small retailers. That doesn't
1:17:12sound good for their welfare. So, I I
1:17:15completely am sympathetic with we
1:17:18that something I started with was that
1:17:19the representative agents' welfare,
1:17:20that's missing half the picture, which
1:17:22is what's the distribution of welfare.
1:17:23And is there some interesting trade-off
1:17:25in policy in some types of policies we
1:17:27might tolerate misallocation because of
1:17:29their distributional implications.
1:17:32You know, what I mean?
1:17:32Yeah, thanks. Thanks so much.
1:17:34may may I ask one question here? So, in
1:17:37in the language of of Richard's like
1:17:39like he
1:17:40gave last last week, so there's this
1:17:44apparent
1:17:46puzzle that is the in- indirect evidence
1:17:48on the size of misallocation is quite
1:17:51large, but then when this sort of direct
1:17:54evidence come come into play, the the
1:17:57effect they find is much smaller.
1:18:00So, one one thing that they hinted at
1:18:03and also Banerjee and Duflo hinted at at
1:18:05some point is that maybe this frictions,
1:18:08these distortions interact at some at
1:18:10some level and that might actually add
1:18:12up to the large indirect wedges that we
1:18:16find. Do you like there's there's an
1:18:19evidence that you showed today, they
1:18:21Cariga and Vincent, that about the sort
1:18:24of this interaction. Do you see any
1:18:26other
1:18:28like Is there any other literature
1:18:31advancing a lot on that?
1:18:35I mean, I would think the Bento and
1:18:36Restuccia and actually Peters are along
1:18:40those lines are saying that there's
1:18:41dynamic implications of some of these
1:18:43static distortions. Um but I I feel like
1:18:46there's more room for what you
1:18:48described. Like it reminds me of Chad
1:18:49Jones' thing about how, you know, if if
1:18:52sectors are complementary and some are
1:18:54distorted and others are not, you could
1:18:56get that
1:18:57you know, it's kind of like the O-ring,
1:18:59Michael Kremer's thing, you know, if you
1:19:00need to produce semiconductors in the
1:19:02Philippines, you need skilled workers,
1:19:04you need uh reliable roads and
1:19:06electricity. And if any of those are a
1:19:08problem, they're not going to locate
1:19:10fabrication plant there. So, you could
1:19:12improve,
1:19:14five of the 10 things you need and one
1:19:16still missing and they might not they
1:19:17might not develop anywhere near as much.
1:19:19So, then
1:19:20I'm really interested in that. It's
1:19:22harder to test because of the whole
1:19:23complementarity, but I'm really
1:19:24interested in whether that's like played
1:19:25a big role in in like China's growth
1:19:27vis-à-vis India, that China did a lot of
1:19:30things right and that's why, you know,
1:19:32they've really taken off and India did a
1:19:33bunch of things that were a bit helpful,
1:19:35but didn't haven't done enough to really
1:19:37like maybe the transportation system
1:19:39really needs reform or something.
1:19:41Um something else that's not
1:19:43uh that's still holding back
1:19:44development.
1:19:47Can I ask that question?
1:19:49Can I ask a question?
1:19:51Yep.
1:19:51Sorry to see you going back to to
1:19:53release the basic model, but I'm not
1:19:55sure I completely understand why the
1:19:57cost of misallocation is higher when
1:19:59goods are very substitutable because if
1:20:01you think for instance of two goods two
1:20:03firms and two goods that are pure
1:20:05complements and because of wedges a firm
1:20:07produces much less than optimal and the
1:20:10other much more, well, if they're
1:20:11complements, this is going to be really
1:20:13costly.
1:20:14So, I'm I'm not sure I understand why
1:20:16misallocation is not costly if goods are
1:20:19not very substitutable.
1:20:21Yeah, so I'm I'm picturing like, you
1:20:23know, they're already at least
1:20:24Cobb-Douglas, they're already like above
1:20:26one, so they're they're more than that
1:20:28and then we're moving them farther above
1:20:29one. So, you're right that that um
1:20:33that the whole story I was talking about
1:20:34complementarity was like here's
1:20:35different activities that that the more
1:20:37they're complementary, the more you move
1:20:38sigma below one, the more damage done to
1:20:41the other sector of having some costs in
1:20:43one sector. So, that's So, I was within
1:20:45this class of sigma is greater than one
1:20:47and we're making it higher than one. So,
1:20:49I think you're raising some interesting
1:20:50questions about complementarity going
1:20:52the other way.
1:20:53There's There's also this
1:20:55maybe you were heading hinting that a
1:20:56little bit, too.
1:20:58If they were perfectly substitutable,
1:21:00you could like substitute around
1:21:01whatever one's distorted and then
1:21:03there's no distortion. So, I think
1:21:04there's some discontinuity at at
1:21:07infinite substitutability, but
1:21:09the log normal structure is what I was
1:21:11using and saying it's it comes out
1:21:13analytically there.
1:21:15But also when I've done it in the using
1:21:17like Indian data, US data, or Chinese
1:21:18data, we just start playing around with
1:21:20the substitutability we we assumed and
1:21:22it's very clear higher that
1:21:24substitutability in the range that we
1:21:26were looking at in the data that we were
1:21:28looking at that that much bigger effect
1:21:30like with sigma equal four than sigma
1:21:31equal three. But I shouldn't state it as
1:21:33either as such a categorical thing
1:21:36either in theory or empirically.
1:21:39Thanks Matt. That's very useful. Thank
1:21:41you.
1:21:42Thanks for the question.
1:21:46Um
1:21:47is it possible to
1:21:49also say that one of the possible
1:21:52sources of misallocation
1:21:54is the fact that there are weak
1:21:56institutions in the state so that the
1:22:00you're not
1:22:01relating directly
1:22:03um
1:22:04the nature of the institutions to what
1:22:06firms specifically do, but it's just the
1:22:09case that the environment does not allow
1:22:12firms
1:22:13to work optimally. For example, if you
1:22:15have a weak legislative arrangement, for
1:22:17example.
1:22:18Is it a way to say that I mean, is it
1:22:20possible to say that that could be a
1:22:22source of um misallocation or it has to
1:22:25be specific to affecting
1:22:29what possibly firms do.
1:22:32What did you say? Did you say weak
1:22:33legislature? Is that what you said or
1:22:34did you say something else?
1:22:35Yeah, I'm saying that like weak
1:22:36institutions, for example, could it be a
1:22:38source of misallocation?
1:22:41And I cited as an example, if you have a
1:22:43weak legislative arrangement like the
1:22:45legislature, for example, is weak. So,
1:22:48the legal regimes
1:22:50are not very predictable, something like
1:22:52that.
1:22:54Yeah, I could see them having big
1:22:55dynamic effects. So, for example, you
1:22:57know, if you're making some investment,
1:23:00you know, like Intel was in was a
1:23:03non-trivial component of of GDP, not
1:23:05necessarily GNP, but GDP in Costa Rica
1:23:07after they located a plant there. Um I
1:23:10think they were 20% of their exports for
1:23:12while they were operating. Um
1:23:14and a big part of that was probably
1:23:15Costa Rica you didn't have the cheapest
1:23:17labor, they had some
1:23:19fairly skilled workers, certainly, but
1:23:21um was the stability of of the promises
1:23:24that the government could make. They
1:23:25could say, you know, "We're going to
1:23:26keep the tax rate low on you or here's
1:23:28the export processing zone we've created
1:23:30for you." And the credibility with which
1:23:32they could commit to that was probably
1:23:33pretty important. So, I agree with you
1:23:35that there could be lots of
1:23:36misallocation if if people are uncertain
1:23:38about the rules changing. Like if
1:23:40there's a lot of investments up front,
1:23:42even like you know, how much capital do
1:23:44you use in your factory versus how much
1:23:45skilled labor versus you know,
1:23:47materials. How many things do you do
1:23:49in-house? And that that relates to the
1:23:51the work that um Johannes Boehm and Ezra
1:23:54Oberfield are doing about you know, the
1:23:55border of what's done within firms
1:23:57versus out firms and how much do firms
1:23:58specialize. So, all of that could
1:24:00easily, you can imagine, feedback. They
1:24:03explicitly talk about how enforcement of
1:24:06of legal rights affects how much you do
1:24:08in-house versus have transactions with
1:24:10other firms um in that paper, but
1:24:12they're also just documenting it really
1:24:14interesting differences in the degree of
1:24:15specialization in India versus versus
1:24:18say the United States um that could very
1:24:20well be a byproduct of the legal system
1:24:23or or, you know, how how often the the
1:24:25tax rates change or how how they're
1:24:27enforced differently.
1:24:30Thank you.
1:24:38Indeed, I see a couple questions in the
1:24:40Q&A.
1:24:41Okay.
1:24:41Um,
1:24:42Stephen Barlow asks, "Companies move up
1:24:45the market chain as they produce more
1:24:46sophisticated, technically complex,
1:24:49higher-value, specialist products.
1:24:52Is there evidence that this progression
1:24:54correlates to a reduction in
1:24:55misallocation, or is the opposite
1:24:58found?"
1:25:00Yeah, so that makes me think across
1:25:02sectors, which I don't Nothing comes to
1:25:04mind right away, but it does make me
1:25:06think of one thing that was kind of
1:25:08surprising, which is
1:25:09when we first wrote the China and India
1:25:10versus the US paper, a common reaction
1:25:13was
1:25:14"Look, China and India are in the middle
1:25:15of reforms. So, the they changed a bunch
1:25:18of regulations and policies, there's
1:25:20going to be a bunch of reallocation of
1:25:21inputs is going to happen that So, add
1:25:24adjustment costs, and there's going to
1:25:25be a lot of a lot of gaps." One thing we
1:25:27found was that inputs just weren't
1:25:29moving around across firms in China and
1:25:32India as much as in the United States.
1:25:34So, the idea that oh, there's a bunch of
1:25:36things jumbling as part of the reform,
1:25:38so there's going to be some
1:25:39misallocation temporarily until things
1:25:41settle down and get reallocated. It just
1:25:43looks like the US looks much more
1:25:44dynamic, like inputs are flying around
1:25:47firms in the US, so the gaps are
1:25:48smaller, but inputs are flying around.
1:25:51So, it doesn't directly address your
1:25:52question, which is like high-tech more
1:25:54advanced sectors, are they Do they have
1:25:56less misallocation? But at least, you
1:25:58know, the US if we think about the US
1:26:00having these more advanced sectors
1:26:02disproportionately in China's like
1:26:03textiles and
1:26:05um, you know, although increasingly
1:26:06higher-tech stuff. It's an interesting
1:26:08question, I think it should be analyzed,
1:26:10uh, investigated whether that whether
1:26:11there's a pattern like that, um, like
1:26:13high-skilled versus low-skilled sectors
1:26:15or capital intensive or technology
1:26:17intensive by some measure, whether they
1:26:19have more dispersion or less dispersion.
1:26:20That's a good question.
1:26:24There's one more by Phillips
1:26:26Oluwagbenga
1:26:28Aboluwade. I probably mangled that. I
1:26:31apologize, Phillips.
1:26:33Um he says do government intervention
1:26:35programs if it's a big picture question,
1:26:37do government intervention programs
1:26:38affect misallocation? How can they be
1:26:40evaluated?
1:26:44Um well, there's that really nice paper
1:26:47I was I know there's a third co-author
1:26:48by Banerjee and Duflo. I'm trying to
1:26:50remember what what year it was published
1:26:52or it came out. They were evaluating a
1:26:53policy which was like earmarking some
1:26:56loans from state-owned banks to certain
1:26:58mid-size firms. So, it had a you know,
1:27:01random element to it that allowed them
1:27:03to gauge whether there was a high return
1:27:05on average to that policy and whether
1:27:07there was dispersion in the marginal
1:27:09product of capital. And they found
1:27:11really high on average but tremendous
1:27:12dispersion. So,
1:27:14I think you know, I get I've been a fan
1:27:16of just using it all the time. Using
1:27:18anytime we can, any policy where like
1:27:20this is a good policy, let's pursue
1:27:21this, just build in some random
1:27:23variation so that we can evaluate, you
1:27:26know, what whether there's misallocation
1:27:28or whether the policy reduces
1:27:29misallocation. Absolutely. So, I I guess
1:27:31I view that as more of something that's
1:27:33been done a few times, you know,
1:27:36and and there's plenty of room for it to
1:27:37be done more.
1:27:38Any other questions?
1:27:42There's a question from YouTube.
1:27:45Can you explain the
1:27:47the mechanism through which capital or
1:27:49output distortions
1:27:51affects TFP?
1:27:53Well, that's kind of probably too broad
1:27:55of a I mean, that's kind of what
1:27:57much of the lecture was on.
1:28:01Um
1:28:02I mean, I can try to give the the
1:28:04the thumbnail intuition though.
1:28:05Okay, go ahead.
1:28:06It's just like if marginal product is
1:28:08say
1:28:09$10 at one firm and $5 in the other, and
1:28:12you just reallocate your unit of capital
1:28:14from the $5 to the $10, you get $5 more
1:28:17of output with the same capital. Same
1:28:19total capital in the economy. So, if you
1:28:21say output in the economy divided by
1:28:24capital and labor, the denominator
1:28:25hasn't changed, but now output just went
1:28:26up $5. So, that's the fundamental
1:28:28intuition by which these gaps marginal
1:28:31products, if they exist, create the
1:28:33opportunity for increasing TFP by
1:28:35allocating things more effectively. And
1:28:37the bigger these gaps are, the the more
1:28:39you're surpassing uh output relative to
1:28:41like your total use of inputs. That's my
1:28:43That's my thumbnail version, Jeff.
1:28:45Okay. All right. Thank you, Pete. Um
1:28:49thank you very much. Um next week, we
1:28:52will continue.
1:28:53There is a uh TA session for Leonard
1:28:56Wantchekon's uh talk on um political
1:28:59institutions.
1:29:00Uh
1:29:01that's on Tuesday.
1:29:03And then um
1:29:05next week, things are a little bit
1:29:06reversed. The the mod We start a new
1:29:09module on agricultural productivity. Um
1:29:12but David Lagakos will give that lecture
1:29:14on Thursday.
1:29:16And then on Friday, Kristin Eady will
1:29:18give a lecture on um technology
1:29:21diffusion.
1:29:23Uh thank you again, Pete. Wonderful job.
1:29:26And uh thank you all the students for
1:29:27showing up. Uh let's unmike and applaud.
1:29:31You got to unmute your mic or you can
1:29:33you can't hear There we go.
1:29:37And we'll thank uh Mandy Chan again.
1:29:40Okay, take care.
1:29:41Bye-bye.
1:29:43Thank you.
1:29:44Thanks, Pete. That was great. Just
1:29:46fantastic.
1:29:48Yeah, thanks for all the comments and
1:29:49participation. I think Mandy said I'll
1:29:51get a copy of the the questions in the
1:29:53Q&A log.
1:29:55I think that's right. And and then we
1:29:58try to clean it up and post it on the
1:29:59web.
1:30:00Okay, perfect.
1:30:01You can monitor how well he ended.
1:30:04I was going to say, yeah, I think you
1:30:05can you can whether uh my answers were
1:30:08appropriate.
1:30:09Probably better than what I could have
1:30:11given, so.
1:30:12It's it's hard to do under time
1:30:14pressure.
1:30:16It is.
1:30:17We actually we we always talk about
1:30:18whether we should go back and correct
1:30:20all the mistakes that we made answering
1:30:22things quickly.
1:30:24So.
1:30:25Yeah.
1:30:25Anyway.
1:30:26Well,
1:30:27do you have a last question Noemi that
1:30:29you were asking in the chat I saw?
1:30:31Yeah, but but I think I got I got an
1:30:32answer. My my my question was but but
1:30:35maybe if if someone else has an answer
1:30:37whether you had nice examples of papers
1:30:39that use the direct approach
1:30:42that was mentioned like basically really
1:30:44trying to having a shock that gives
1:30:46wedges and and you can really observe
1:30:49full shocks and then only then doing the
1:30:52the welfare loss through misallocation a
1:30:54computation.
1:30:57I have a wonderful example.
1:30:59A new paper of mine.
1:31:03We look at directly at interest rate
1:31:05variation using a
1:31:07credit registry.
1:31:08Okay.
1:31:09So, this isn't exactly cost of capital
1:31:11varying by firms
1:31:13and building it into an aggregate model.
1:31:15Wait, what?
1:31:16posted on the website if you want to
1:31:17take it. It's not an experimental thing,
1:31:19but it's observing direct cost of
1:31:21capital.
1:31:22Okay, cool. Well, that's that's exactly
1:31:24what I'm looking for. Thank you.
1:31:26Thanks.
1:31:26Yeah, send me that too. I want to see
1:31:28that.
1:31:29Oh, okay. Great.
1:31:31Take care.
1:31:32All right. Thanks.
1:31:33Thank you.
1:31:34Thanks a ton, Keith.
1:31:35What?
1:31:36All right.

About this transcript

This page contains the full transcript of Lecture 6: Recent applications and advance by CEPR & VideoVox Economics, generated from the public captions YouTube serves with the video. The transcript has 17,401 words across 2,823 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.