YT YouTube2Text

YouTube2Text → Transcripts → Supplemental lecture: Labour market frictions and development

Supplemental lecture: Labour market frictions and development — Transcript

by CEPR & VideoVox Economics · 16,170 words · 2,348 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome back. I hope you all enjoyed
0:03your virtual spring break. Um
0:07uh we are delighted to have Mark
0:10Rosenwine
0:12uh giving a supplemental lecture on
0:14labor markets in developing countries.
0:17Uh Mark's a professor at Yale. He's a
0:20microdevelopment economist, a labor
0:22economist. So he's a perfect person to
0:24give this lecture. Uh he'll be focusing
0:27most mostly on agriculture. I should say
0:30welcome uh and give you a little bit of
0:32the ground rules. Of course, this is the
0:34STEG uh course in macro development. Um
0:38STEG is a research cirrun research
0:41program funded as part of the UK's uh
0:45aid effort through the foreign
0:47commonwealth and development office. Um
0:50I want to remind uh the panelists, the
0:53front row students, you can interrupt
0:55using the microphone uh if you have a
0:57question. Um, but otherwise, if you
1:00don't have a question, please mute so
1:01that we can hear Mark. Um, the rest of
1:04the class that's in the webinar, uh, you
1:07can ask questions in the Q&A and Doug
1:10and I will do our best to try to answer
1:12what we can and if there's, uh,
1:14something that's, uh, you know, we can't
1:16answer that a lot of people have
1:18interested in, um, we will ask Mark at
1:20the end. Um, I also want to thank, um,
1:24well, thank Mark again. All these
1:27lectures are really doing this out of
1:28the goodness of their heart. I want to
1:30thank Kirsty and Mandy as well at CPR
1:33for organizing this for us. Um, go
1:36ahead, Mark.
1:37All right. Well, thank you.
1:39I'm going to now share the screen
1:44and
1:46I can get this thing out of the way.
1:48I'll be able to
1:50do full screen mode. Okay. So this I'm
1:54reporting on research that is carried
1:57out jointly with Andrew Foster where the
2:00the big question that we're asking is
2:02are there too many farms in the world
2:04which is the same as asking are the
2:06farms in general too small in low-income
2:09countries and it turns out that labor
2:11market frictions uh in agriculture
2:14appear to play an important role in
2:16understanding what's going on with
2:18respect to with respect to scale. So
2:21this lecture is is going to focus on
2:23labor market frictions in agriculture in
2:26low inome countries. Now that's
2:28obviously not the only labor market
2:30frictions that exist in the world in the
2:32industrial sector. However, a lot of
2:34those frictions are legislative. They're
2:37based on labor laws that there might be
2:39specifically legislated hiring costs,
2:41but the ones in agriculture are in a
2:44sense natural and they're they're not
2:46obviously something that that policy
2:49could directly affect. Now, now what why
2:51is that? Okay. So, we have to sort of
2:54understand what's different about
2:56agriculture compared to to industry.
2:58Agricultural production takes place in a
3:00series of sequential operations
3:03and the timing and length of those
3:05operations is stochastic. It's going to
3:06depend on weather. So, you can't know in
3:08advance when you're going to fertilize,
3:11when you're going to need to spray, when
3:14you're going to be weeding. It's all
3:15going to depend on stochcastic events
3:17that are going to be revealed over time.
3:20And the combination of the the episodic
3:23and stochastic nature of agricultural
3:26production combined with small scale
3:29because in principle, if the scaler was
3:30large enough, you could contract to to
3:32have a laborer around, you know, all the
3:35time. there's enough work to do but it's
3:37but at small scale you're just not going
3:39to see uh relational uh contracting
3:42going on and that that's what we observe
3:44in the data as I'll as I'll show you. So
3:46labor and agriculture is typically hired
3:49uh if it's hired at all and we'll we'll
3:51talk about the importance of that on a
3:53daily basis. uh and um there are
3:56therefore search costs that so when a
3:59when a farm needs to hire labor uh it
4:02needs to find that labor and and uh
4:04concominantly laborers need to find the
4:07farmers that are uh needing this this
4:10labor. So what we do in this research
4:14and I'll talk about is unlike in in a
4:17lot of macro work where where sometimes
4:19the transaction costs or these costs are
4:21backed out of of of a calibrated model.
4:24We're going to actually use direct data
4:27that's going to enable us to um to
4:29estimate these transaction costs and
4:30quantify them. But we're also going to
4:32use a calibrated model to to also
4:34extract them. And we're going to be able
4:36to compare the two. And in in
4:38particular, we're going to show who pays
4:40for these these contraction costs. Um so
4:43you if farm if individual laborers have
4:45to search they they incur their own
4:47their own cost of travel etc. uh the the
4:50question is in the labor market how much
4:51of that is compensated and we're going
4:53to then embed these frictions in a model
4:56of agricultural production and we're
4:59going to show that using these models uh
5:01what role these frictions play in
5:03lowering productivity through the
5:05misallocation of labor both across farms
5:09and also as I'll show an
5:11underutilization uh in a large fraction
5:14of the farms that exist in low-income
5:16countries
5:18and so sort On the micro side, we're
5:20going to test some of the implications
5:21of of this uh of this model that embeds
5:24these transaction costs, but we'll also
5:26calibrate the model uh to a number of
5:29moments in the in the data as I said and
5:32we're going to use the calibrated model
5:35which which whose basic elements seem to
5:37pass these these micro tests that we
5:39carry out to perform a major
5:41counterfactual. So, you know, an
5:43important aspect when we talk about
5:45development as a dynamic process is
5:47structural adjustment, right? the
5:48transformation of an economy from rural
5:50and agricultural to industrial. And what
5:53we're going to do in this counterfactual
5:54is we're going to force a structural
5:56adjustment. And by doing that, it
5:59corresponds in in some ways to policies
6:02that have been actually carried out,
6:04which is a major land reform, a
6:07consolidation of land holdings. And by
6:10doing that, we're able to see by how
6:12much specifically there are too many
6:15farms and how much output total from the
6:19from the same amount of land and how
6:21much output per labor is uh is costed is
6:24sacrificed by having this inefficient
6:27distribution of uh of land sizes. Okay.
6:31So
6:33you in in talking about low-income
6:35agriculture, we you just can't ignore
6:36the the role of scale or agriculture in
6:39generally in general and you know the
6:41key difference between low-income
6:42agriculture and agriculture in
6:44developing countries is is clearly scale
6:47and in low-income countries productivity
6:49whether per worker or income per worker
6:52or productivity of the land is is
6:54significantly lower and I'll show you
6:55what how much lower. Okay. So we need to
6:58look at this big picture as it were of
7:01of scale and uh and and look at the role
7:04of labor market frictions and and
7:05understanding why there are differences
7:07in productivity by by scale. Okay. And
7:10this has to include the role of
7:12mechanization. I mean in part because me
7:15machines can substitute for labor which
7:17means they in principle can save on
7:19these labor market transaction costs but
7:21they're also themselves a source of of
7:23scale economies. And we need to look at
7:25both the scale economies associated with
7:28these frictions in labor markets as well
7:30as the the the technological scale
7:33economies that are in some sense
7:35specific to to agriculture in many ways
7:37as I'll as I'll explain now the we found
7:42historically
7:44today's discussions of agriculture at
7:46least among microeconomists is mostly
7:48about you know marginal effects of
7:50getting farmers to use more fertilizer
7:54or credit subsidies or other sort of
7:56marginal impacts on agriculture. And
7:59there's just very little discussion in
8:01modern development economics about the
8:02scale issue. But when we go back to to
8:051948
8:07and look at the report of a commission
8:08that was set up in 1946, at the time
8:11that India became an independent nation,
8:14they had to determine what to do with
8:16these very large land holdings held by
8:17the the Zamandiri in the in the United
8:21Provinces. And and so what you see on
8:24the screen is is the the fundamental
8:27statement of this committee which is the
8:28fundamental problem in agriculture in
8:30India is farms are too small. And in
8:331948 the farms were considerably larger
8:36than they are today. In India the
8:37average farm size about three acres.
8:40Okay, which is 33 times less uh in size
8:43than the typical than the average farm
8:45in the US. Okay. And they were really
8:48quite bold because they said well you
8:49know so what should the scale be? It's
8:51obviously lower than it what it should
8:53be. And they came to a technological
8:55conclusion that the minimum Favre scale
8:57would be one that could sustain the use
8:59of what was then the the machine or the
9:03mode of power at the time, which was a
9:04bullet pair. Okay? And say they they
9:06came out and said minimum size should be
9:0810 acres. So we're going to look at how
9:10many farms are actually uh
9:14less than 10 acres uh in in today's
9:16economy in India as well as a whole
9:18bunch of other other countries. So they
9:20came out with an idea of 10 acres should
9:23be the minimum and they actually um in
9:26carried out a counterfactual of making
9:30all farms 10 at least 10 acres and how
9:33many farms were in excess based on that.
9:35So we're going to do what they did but
9:37we're going to use a much more
9:38sophisticated model and we're going to
9:39take in account that there's now modern
9:41machinery and these transaction costs.
9:44So we'll start with what may be familiar
9:47to you in a number of dimensions but not
9:49all. This uh looks like five global
9:53facts even though it says four. So um
9:57first as I said farming in low-income
9:59countries is small scale and I'll show
10:01you that and it's large scale in
10:03developed countries. Uh productivity
10:06based on just looking at output per land
10:09uh is much higher in in high-income
10:11countries. But also there's so then you
10:15would conclude if you looked across
10:16countries that you know well you know
10:18scale seems to be really important here.
10:20Large scale farming is clearly more
10:22productive if we run a regression across
10:24the across countries. But then when you
10:27look within low countries there's this
10:28stylized fact that that exists in every
10:31low-income country which is that the
10:33smallest farms are the most productive
10:34whether you measure profits GFP or just
10:37output per per acre. Okay. And a a
10:41fourth stylized fact, and I don't I
10:44don't know why we use stylized. These
10:45are facts that I'll show you. Uh is that
10:48a very large fraction of farms in
10:50low-inccome countries are operating
10:52outside of the labor market. By that I
10:55mean that there's no family member who's
10:57working in the labor market while
10:59they're farming and they're not hiring
11:01anybody either in the labor market. So
11:03for example in the data that we have
11:05from India in terms of land area 50% of
11:08the land area it tends to be farmed in
11:11this artic way that that that has no
11:13contact with the labor market. So any
11:15good model is going to have to explain
11:16the existence of these of these farms
11:18and these frictions are going to are
11:20going to play a role here and then
11:23obviously in high income countries we're
11:24going to see a lot of mechanization.
11:26Okay. And so the question is is there
11:28some barrier to mechanization? In low
11:30income countries does it have to do with
11:31scale? So let me just show you quickly
11:33these these facts. So this is the
11:36fraction percent of households
11:39uh that are
11:4210 acres or less in u
11:46across a lot of countries. So you can
11:48see that in the US you know less than
11:51about 10% of farms are are 10 acres or
11:53less in in Canada it's only about 6%.
11:57But in a whole array of countries around
11:59the world and Asia that are that are
12:00that are low income, we see that you
12:03know there's basically a minority very
12:05small minority in some cases of farms
12:08that are as at as large as as 10 acres,
12:10right? So this is a massive difference
12:12in in scale. We look at productivity per
12:16acre. You know, we all know the income
12:18per capita is much higher in high income
12:21countries than low countries. about the
12:23productivity measured as what one gets
12:25out of land. We look at a common crop
12:28that's grown across these countries
12:29which is soybeans and we can see that
12:31the output per acre is four times higher
12:33in the United States compared to
12:35Indonesia, India and the Philippines.
12:37China looks like a little bit of an
12:39outlier. You know, it's only two times
12:41but turns out that China is a big
12:44overuser of fertilizer which just tells
12:46us that we can't use output. we need to
12:48look at either TFP or or profits that
12:51take in account the the use of of inputs
12:53when we're making these these
12:54comparisons. Okay? And you know it's not
12:57just soybeans when we're just looking at
12:59India for example that's what we'll
13:00focus on here we see that it's true
13:03across a wide variety of crops that
13:05there's just a a very large uh
13:07difference in productivity. Now in in
13:10terms of for example rice I mean India
13:12is the major producer of rice in the
13:15world right and yet you know their
13:17productivity is considerably low. If we
13:19think about the issue of food
13:21availability right if if it's really
13:23true that we can we can up the u
13:26productivity of land in agriculture in
13:29India by a transformation of scale of
13:32farming. Okay, then you know that that's
13:34going to have a major impact not just on
13:35incomes but on the amount of food
13:38available in the world. Okay. All right.
13:40So here's this this uh fourth fact which
13:43is this inverse relationship. So here is
13:46four different countries. I don't know
13:47if you can see it well. Bangladesh,
13:50China, Indonesia and Nigeria, right? So
13:53representing different continents,
13:54right? And what you can see is in all
13:57those countries and I'll show you it's
13:58true for India too in terms of profits
14:02or yields per acre it's significantly
14:05higher at lower scales right there's
14:08this inverse relationship which goes
14:10against what we see worldwide okay and I
14:13think that this may in fact be
14:14responsible for why when people talk
14:16about land reforms in low-inccome
14:18countries they talk about moving from
14:21large farms to small farms and indeed in
14:24China they went from communally
14:27farmed land that was large to individual
14:30plots of land farmed by by families.
14:33Right? Now part of that was due to the
14:34issues of moral hazard associated with
14:36collective farming. Right? But one
14:38question is they sacrifice scale and the
14:41issue is whether by doing that they
14:43sacrifice productivity. Okay. Soic
14:47farming. So I've already defined it as
14:50farming that takes place or operations
14:52specifically that takes place without
14:54any member of the family working in the
14:56labor market or hiring any labor. A lot
14:59a lot of data sets provide
15:03information at the operational level. So
15:05the the Nigerian LSMS gives you two
15:08operations. One they call planting and
15:10the other they call harvesting. and we
15:12look at the planting operation, we find
15:14that
15:16more than a third of those operations
15:18are turkey. So it's a it's a big
15:20fraction. Uh the the China data enables
15:24us to see it only over the whole season
15:26and over the whole season 17% of the
15:28farms never hired any labor and none of
15:30the family members ever worked in the in
15:32the labor market. And in the India data
15:34that we'll concentrate on uh 30 over
15:37more than again more than a third of the
15:39operations and here we have 12 different
15:41operations uh were uh were artic okay
15:46and the relationship between size and
15:49utarchy is not uh not monotonic. It
15:52looks like this in uh in the three
15:55countries where we can actually compute
15:57uh autoarchy. The way to do that is you
15:59actually need information on family la
16:01family member labor supply as well as
16:04inputs by whether they're hired or not.
16:06And actually a lot of data sets just
16:08don't provide that kind of information.
16:10So these three do and what you can see
16:12they all show the same shape that the
16:15that the arthury is somewhere not at the
16:18bottom right somewhere not quite in the
16:20middle but uh well definitely not in the
16:23middle but u has this shape. Okay. And
16:26so again, a good model is going to be
16:29able to explain the inverse relationship
16:32between profits per acre and acreage and
16:35is also going to be able to explain the
16:37shape of this curves as well. And then
16:41of course the the final fifth fact is
16:44that the farms in the United States are
16:46big and they're getting bigger and uh
16:49much more mechanized. Okay. So when we
16:53talk about this issue of too many farms,
16:55okay, it's not a fantasy. You could have
16:57said the same thing in the United
16:59States. If we look in 19, so this is
17:01going to plot the number of farm. This
17:03plots the number of farms in the US from
17:0519 from 1850 to the current day. Okay.
17:09And as well as the total amount of land
17:11and farming and average land size. And
17:13what you can see if we were looking at
17:15the number of farms in 1935, 7 million.
17:18Okay. We could have asked the question,
17:20is that too many farms? Well, it turns
17:22out, yeah, because in
17:25now we have only two million farms.
17:27That's a reduction of 5 million farms
17:29with the same amount of land.
17:32And the blue line here tells you what
17:35happened at least since 1948 in output.
17:37Output actually grew. So you had a
17:40massive reduction in farms and a
17:42significant increase uh in output over
17:45time excess farms and of course at the
17:47same time a reduction in the amount of
17:50labor used on each of the farms. So if
17:53you know if this is the natural
17:57path of development of structural
18:00adjustment that occurs uh in in when
18:03there's true development, right? We
18:05should see this in these low- income
18:07countries as well, but we don't. They've
18:08been stuck in this low productivity
18:12small farm scale for a very long time
18:15with no policies uh in low countries
18:18that are discussing even the idea that's
18:20that's anymore that scale is a is an
18:23issue. Okay. So
18:26it seems like the mere fact that farms
18:30are of different size uh is a is a
18:33problem because the scale economies are
18:35not monotonic and and you know there's
18:37differences across scales. Okay. So and
18:42this notion that they're scaled this
18:44economy seems at odd you know with what
18:46we see in the world and so we need to to
18:48understand that. So I don't have to
18:50convince you that you know it's a it's a
18:52big question for development. There's
18:54lots of farms. It's about 95 million.
18:55Now, this is an old statistic. Okay. And
18:58just to give you a preview, when we do
19:00our counterfactual based on our
19:01calibrated model, which passes tests uh
19:04a number of of tests of not just fit,
19:07but of sort of what we see through
19:09comparative statics in the data, we show
19:11that you can reduce the the number of
19:13farms by 87%. That there's over 80
19:17million too many farms in India. By that
19:19I mean you can do that and have output
19:23actually increase and specifically you
19:26can have the output per labor increase
19:28by 68%. By just in reducing the number
19:32of farms keeping the same amount of land
19:34in uh in production and um labor force
19:39actually goes down a bit. We'll we'll
19:40talk about that. So that's going to be
19:42the bottom line here. Again you see the
19:44role that transaction costs play. Yeah.
19:46There's been a couple questions in the
19:47Q&A about the definition of farms,
19:50whether that's changed.
19:51I'll talk about that. I'll talk about
19:52that. This is just the introduction. So,
19:54let's hold off on questions because I'm
19:56going to define farms and plots and a
19:59lot of things. Okay? And and you'll see
20:01and if I don't, you know, then please
20:03please ask. Okay? So, um so we'll we'll
20:08we'll get to that. Okay? So now the the
20:11challenge to there are a number of
20:13challenges to identifying scale
20:14economies and the the first is that in
20:19most of the countries the data sets that
20:21are available are representative data
20:24sets of the population and farm scale is
20:26is very small. So you don't actually see
20:29any large farms in most of these
20:30low-inccome countries. So it's very hard
20:32to find what largecale productivity
20:35might be in in these contexts. Okay. uh
20:39and also of course you know the scale of
20:42farming may not be exogenous right it
20:44may be something that's induced and
20:45indogenous and correlated with a number
20:47of unobservables associated with land
20:51quality with farmer ability and other
20:53kinds of of things uh the land
20:55allocation process may may be p
20:57purposive or market oriented and other
20:59things so there's a number of of
21:01challenges um there's also some issues
21:03about measurement uh as well and of
21:06course the challenge to identify if
21:08there are real age scale economies. We
21:10surmount all those challenges of of data
21:12and and correlations and spuriousness,
21:15you know, of of why there are scale
21:16economies. And there we're going to need
21:18a model and we're going to need really
21:19good data as I'll show you. Okay? And
21:22the the data
21:24challenges are such that almost no other
21:27data sets in the world, but the one that
21:28we use is capable of actually building
21:31up and and empirically verifying the
21:33model that we're going to have. Okay. So
21:36I'm going to describe the data
21:38uh and I'm going to show the
21:40relationship in our data which could be
21:41from India the dis the relationship
21:43between land uh productivity and uh and
21:47acreage. Okay. Then we'll get rid of all
21:51the other sort of nonsense stuff about
21:53measurement issues and correlated with
21:54farmability in ways that I I think will
21:57convince you. Uh then we're going to
21:59quantify the importance of these
22:00transaction costs in the labor market
22:03and then we're going to build them into
22:04this to this model that will incorporate
22:06those transaction costs as well as
22:08economies of scale and farm equipment
22:11that we need to to explain what is
22:13ultimately going to be what we see in
22:15the world a U-shape in this productivity
22:18land area relationship and then we'll
22:21have then we'll look at some tests of
22:22transaction costs the implications of
22:24that we'll assess whether there are
22:26really economies of scale on equipment
22:28which which also has some very strict
22:30data requirements. Uh we'll estimate
22:33structural parameters about equipment.
22:35Um and then we're going to calibrate the
22:37model based on a number of moments in
22:39the data which is going to enable us to
22:41identify the optimal scale of farms
22:43given existing technology and by
22:46existing I mean the technology of the
22:48equipment that's now available in India
22:51right now and I'll show you in fact a a
22:54storefront of of equipment when we see
22:56that then we'll carry out this
22:57counterfactual of uh of land
23:00consolidation and then we'll talk about
23:02you know why isn't it happening
23:03naturally in the market. So that's where
23:05we're going. All right. So let's talk
23:06about the the key data set that we use
23:09which is the ikrat village studies in
23:12India. Icat stands for the international
23:14crop research institute in the semiarat
23:17tropics. Uh they actually started this
23:21long-term survey in 1975 and it
23:24continues today in in uh in six of the
23:26villages but it's expanded. We're going
23:27to look at the the the panel from 2009
23:30to 2014, which was the latest available
23:33as of as of a year ago. This covers 20
23:36villages in six states, 819 farmers,
23:392,000 plots. And here I'll get to this
23:42distinction between plots and and farms
23:44in a in a second. There are two features
23:47of this data which are a little bit
23:48different from a lot of uh data sets.
23:50One very different. The first is in
23:52addition to a sample of these farmers in
23:57each of these villages, there's a
23:58complete census of all the households in
24:01the villages. Right? So we can say
24:02something about the full picture of the
24:04land distribution in these areas. And
24:07the second key feature of the sampling
24:11frame is that it over samples large
24:14farms. The sampling frame is not meant
24:16to be representative, but rather it's
24:19divided up into four equal-sized groups.
24:23Landless labor households,
24:26households with small farms, medium
24:27farms, and large farms. And rather than
24:30talking about, you know, what the
24:30cutoffs are, I'm going to now show you
24:32what the distribution looks like in
24:34these um in these villages using the
24:36census data and what we get from the
24:40survey data which oversamples large
24:42farms. Okay. So this is the cumulative
24:45distribution of farm sizes in these uh
24:49villages in the six states 20 villages.
24:53So the x-axis is uh acreage. Okay. And
24:56the red line tells you the fraction of
25:00farms of a given acreage that are below
25:03that cutoff. So let's take our important
25:06cutoff of 10. That's what the Zaman
25:08theory commission talks about. A lot of
25:10the data that you see across country is
25:12a rate of what fraction or below 10 as
25:14we saw and what we see is in these
25:17villages 90% of the farms are less than
25:2010 acres okay below the minimum that the
25:23Zandere commission thinks is is
25:25important but when we look at the survey
25:27data that has this intensive information
25:30which is the the uh blue line we see
25:35that in fact
25:38the 10 40% of the farms that we're going
25:41to be able to observe are above 10
25:43acres. So we have just a large a much
25:45larger fraction of larger farms that
25:49were able to look at in the India data
25:51compared to only almost every other data
25:53set in the world. There are two
25:54exceptions that we know of. One in
25:56Africa that was collected just a few
25:57years ago. Now we're going to
26:00distinguish between plots and farms. The
26:02basic unit of operation is a plot which
26:04I'll get to. Farm is a collection of
26:06plots. Okay. So when we look at the the
26:08plot distribution, it looks it looks
26:10like this. Again, there's more plots
26:13above uh 10 acres than you get by
26:15looking at the farms above 10 acres.
26:17Okay. So this data also is very rich in
26:21having information on inputs that are
26:23used, the prices of inputs, multiple
26:25measures of plot quality, input price
26:27schedules which are going to be very
26:29important for measuring the transaction
26:30costs. uh there's information how the
26:33plots were acquired and basically these
26:35all of these plots were acquired
26:37essentially by inheritance there's just
26:39very thin market for sales as I'll
26:41quantify in a second there's inventories
26:43of assets and there's a little bit of
26:46information about the capacity of
26:47equipment unfortunately and this is a
26:50big deficit in almost every data set in
26:53the world you know at you know at the
26:55aggregate level people talk about
26:56capital which is just nothing right it's
26:58just an aggregate of the cost of whole
27:01bunch of different things for
27:03agricultural data sets. You might get
27:05that there's a tractor and a thresher,
27:07but that's not enough either. There are
27:09big fat tractors and small tractors and
27:11there's big sprayers and small sprayers
27:13and that the capacity of of machinery
27:16really is the issue here. That's where
27:18the scale economies are and you can't
27:20see them in most data sets, but you can
27:22here. All right. So, what about this
27:24inverse? I I promise I'll talk about
27:26plots and and farms in a second more
27:29precisely. So what happens when we look
27:31at the relationship between farm size
27:34acreage and output and we're going to
27:37use profits per acre. So we'll take into
27:39account all the inputs that are used.
27:41Okay. So this is what it looks like
27:43across 12 acres up through 12 acres
27:46which is 96% of the farms in these
27:49villages. Right? And that looks exactly
27:51the same as we saw for those four other
27:53countries Nigeria and Mexico Bangladesh.
27:56Right? So, you know, we're there's
27:58nothing peculiar about this environment,
28:01right? If we had a representative sample
28:03of farms, this is what we would see
28:06basically because there' be almost no
28:07observations above 12. But in the survey
28:11data, when we extend it out, this is
28:14what you see. It's U-shaped.
28:17So all of the studies that have been
28:20focused and there have been a lot of
28:21them in the old literature on the
28:23inverse relationship in low-income
28:24countries is asking the wrong question.
28:27They've got to have a model that
28:28explains the U because that's what
28:30really exists in these countries. You
28:32just can't see this part because you
28:34don't have enough farms in most of the
28:36of the most of the samples. Okay. So
28:39when you look at this, you can say,
28:40well, all right, you know, there's lots
28:42of hypothesis you might have about why
28:44larger farms are, you know, might be
28:46more productive than some of the
28:47smallest farms. You know, like these
28:49guys are richer, they're less credit
28:51constrained, maybe the land quality is
28:52different, maybe they're growing
28:54different crops, lot lots of different
28:56things. Okay, what we're trying to do is
28:58get at, you know, pure scale economies
29:00that are associated with the input
29:03factor costs and other things. Okay, so
29:05by the way, we can immediately rule out
29:06that it's different crops. So if you
29:08just take one crop that's commonly grown
29:10cotton, this is what it looks like.
29:12Okay? So it's not differences in crops.
29:15You could do it for you could do it you
29:17can't do it for soybeans because there's
29:18nobody growing soybeans on the small
29:21farms, but everybody's growing cotton
29:23almost. Okay? So
29:27these plots are measured directly by the
29:31survey researchers, right? So there's no
29:33this is not asking farmers how big's
29:35your farm, right? They're measured very
29:38well and they were measured twice
29:40actually in over time and so you can
29:43actually compare the different
29:44measurements and you know measurement
29:45error is just trivial. So that's
29:46irrelevant here. There's 24 measures of
29:49plot characteristics. There's you know a
29:51lot of aronomists are working in on on
29:53these data surveys and they they you
29:55know they they really know that plot
29:57quality matters and and so there's a lot
29:59of good measures of that. Okay. Now
30:03the key one that we worry about is that
30:06there's a correlation between the scale
30:10of operation, the wealth of the farmer
30:12and the ability of the farmer. And we
30:15can rule out that those are solely
30:18responsible for the U-shape because we
30:21can look at the scale of operation of
30:25the different plots owned and operated
30:28by the same farmer. So more than half of
30:32the farms are operated on multiple
30:34plots. Okay. So here's a an example of
30:38the quality of the soil. All right. So
30:41this is a plot a graph distinguish
30:45between plots and graphs here. Um the
30:48red line is what you saw. That's across
30:50farms in profits per acre and acreage.
30:54Okay. The green line is the relationship
30:58between profits per acre and land size
31:03controlling for the 24
31:07plot characteristics. So it controls
31:09immensely for for soil quality and you
31:12can see that does explain some of the
31:14upper tail of the on the left. Okay, but
31:17you still get a U-shape and it has
31:18nothing to do with the the this part of
31:21the the curve as well. And then finally
31:24controlling again for the 24 plot
31:26characteristics. The blue curve is the
31:29relationship between the scale of plots
31:32for the same farmer
31:34and profits per acre.
31:36These within farmer estimates of this
31:39relationship control perfectly for
31:41access to capital and they control
31:43perfectly for farmerability.
31:45So we can say that while farmer ability
31:48may be relevant and capital market
31:50constraints may be relevant the U-shape
31:54exists
31:56independent of that. Okay. So it's
31:58there. So this is just to show there's a
32:00real scale seems to be a real scale
32:02phenomenon there. Now is this a
32:06technological issue?
32:08So for example, if all of the plots that
32:11were farmed by the farmer were sort of
32:14consolidated in terms of operation,
32:16right? Then in a sense there's no
32:19difference in the plots in terms of
32:22individual operations and so it could
32:24just be pure scale economies. But it
32:26turns out that farmers
32:29cultivate plots differently with
32:32different timing and the data enables us
32:33to to show that. This is just if you do
32:35it with um output per acre. Okay. And at
32:38the same time, what we see in the data
32:40is that output per per labor just grows
32:43almost monotonically across farm size.
32:46Okay. So, if we move to larger farm
32:48size, we're going to get higher. It
32:50would appear, right? That's what's has
32:51to be tested. It would appear we're
32:53going to get higher output per acreage
32:55and we're going to get more output per
32:56worker as well. All right. So, plots and
32:59farms.
33:01So plots are the spa the basic spatial
33:04unit of production and they're not
33:06choice variables here. So in 2009 there
33:10was a
33:12a module that told us about the origin
33:14of these plots. So 85% of all the owned
33:17land in 2009 was inherited land. Okay.
33:20And this didn't differ across large and
33:23small holdings. Okay. across the six
33:26years of the panel less than
33:27threequarters of a percent of any of
33:30these plots were sold or bought. Okay?
33:32So it just stays the same. So plot size
33:36does not vary from year to year. Farmers
33:38are not choosing how much of a plot, how
33:40much acreage I should plant. They may
33:43plant different crops across years on
33:44the same plot. And the plots are really
33:47important here. So as this the unit of
33:49production, they even have names in the
33:51data set. So this is not just a panel of
33:54farmers. that are followed over time.
33:55It's actually a panel of plots that are
33:57followed over time. Okay, so I I hope
34:01that answers what that question was,
34:03right? What's a farm? What's a plot?
34:04Right? So a farm is a collection of
34:06these of these plots. Okay, and as I
34:09said, the the plot is the basic unit of
34:12of operation. the time we have in the
34:15iCaret data unique to to to data sets
34:18around the world the actual dates at
34:21which each operation was initiated by
34:24plot. Okay. And what we can see is
34:26across plots for the same farmer we
34:28almost never see the initiation of the
34:31same operation occurring at the same
34:33time across the plots. So here's a graph
34:36of that. This is the percent of
34:37operations starting on the same day
34:39across farmer plots with farmers that
34:41have two or more plots. Right? and it's
34:43just it just never happens. So we can we
34:45can in a sense treat the plot as as if
34:48it's a farm. Now the correlation between
34:50plot sizes and farm sizes is is over 7.
34:53So I'll sometimes casually say farm size
34:56and some I'll say plot size and it's not
34:58going to make a difference. Sometimes it
35:00it will and well I'll be very specific
35:01about when when it really matters. Okay.
35:04All right. So can we explain this
35:07pattern? We need to explain the U shape.
35:09We need to explain the existence of
35:11artic farms. We need to explain why the
35:13artic farms are not in the smallest
35:15scale farms and they're not in the
35:16largest scale farms for example of the
35:18things we've got to to fit in uh in the
35:21data. Okay. So the model as I said is
35:24going to have these transaction costs
35:27of hiring labor. Okay. And it's going to
35:31have economies of scale and equipment
35:32but we're going to measure them right.
35:34So we're not going to we're not going to
35:36back out either from a calibration.
35:38Okay, we're going to actually estimate
35:40the capacity scale economies and
35:42equipment and we're going to estimate
35:44directly from the data what the
35:45transaction costs are that that we
35:48observe and we'll back them out from a
35:51calibration.
35:53Okay, so
35:56labor as I said are hired on a daily
35:58basis. There's no long-term contrast.
36:00Um, and
36:03there's good reason to think that these
36:06transaction costs are not going to be
36:08tiny. Okay, there's travel costs. For
36:11example, the data provides us with the
36:13average distance between a plot that the
36:16farmer is cultivating and where he
36:18lives, which is on average 1 kilometer.
36:21Okay? And you know, you have to look at
36:23what the spatial organization of these
36:25villages is. And we have a satellite
36:28view from Google Maps. This is one of
36:32the 20 villages.
36:34This is Shirore. Okay, this is we're
36:36actually looking at one of the villages.
36:38And what what we see is which is typical
36:40in in India is that all the villagers
36:42live in a central compound. So this is
36:44what the village looks like. There's
36:46streets. Sometimes they're paved.
36:47There's electricity. There's there's
36:50lights on some of the streets sometimes
36:51or or not. Okay. So whether you're a
36:53landless laborer or you're a quote big
36:56farmer, this is where you live. And the
36:58rest of the area called the village is
37:00where the plots are. And you could you
37:02could sort of make out that these these
37:04the the individual sizes of the farms.
37:07So the farmer has to go to the to the
37:09farm and the laborers have to go to the
37:11farm and a laborer has to find the
37:13farmer that is um that needs him for a
37:16particular day. Of course, those
37:18arrangements might be made, you know,
37:19within within the village, but there is
37:22a lot of movement that is required for
37:25for laborers. The the the growth center
37:28at Yale carried out its own survey in in
37:31one state, Tomato, which is has detailed
37:34information about where laborers work
37:36and who they work for. And what it shows
37:38is that almost a quarter of the laborers
37:41report working for a farmer outside the
37:43village. So that picture of what a
37:45village looks like is not the full
37:46picture, right? A lot of labor goes
37:49outside of the village. Concominantly,
37:51the farmers report that they employ wage
37:54laborers from outside the village. Okay?
37:56And among the workers who work outside
37:59the village and who travel by foot on
38:01average they travel 2 kilometers and
38:04among those traveling by bus they
38:06average about 8 kilometers. So there's
38:08there's daily commuting among these
38:11these laborers. Okay. Now, another fact
38:15in the Icarat data that you don't see in
38:17almost any other data set in the world
38:21is
38:22hours worked by individual laborers. And
38:25the reason for this is in typically in
38:27agriculture as I've said labor is hired
38:31daily and wages are report daily. So
38:34what you see in data sets that report
38:36wages is wages per day, which is fine.
38:41And I know when I started out working on
38:43in Indian data a long time ago, I just
38:46assumed that all the workers worked an
38:49eight hour day, right? They're paid to
38:51to work for a day. Well, theat shows us
38:54data shows us that's not true. There's a
38:56large mass of workers who work less than
38:59eight hours a day. And this is going to
39:01be key to the issue of the importance of
39:03transaction costs, which is going to
39:05enable us to explain the the shapes that
39:08we see in the in the data. Right? So
39:10this is there are two sources of
39:12information on hours worked by laborers
39:14in the Icarat data. One is in 2009 and
39:182010 rounds some representative farmers
39:21were asked to describe the wage
39:24schedules. They were asked to say you
39:26paid by different types of labor and for
39:29specific operations for 12 different
39:31operations. What was the daily wage you
39:34paid and how many hours was the worker
39:36working? Okay, so that's one set of
39:38data. Another set of data is from the
39:40input transaction files which are
39:42collected every three weeks in the
39:44acroat data which tell us the number of
39:46hours worked by hired laborers input and
39:49the wages they were paid. Okay. So we
39:53can compute hourly wages from from this
39:55which you don't often see in in
39:58agricultural data sets. All right. So
40:00table one
40:02first column is a simple regression of
40:05log hourly wages which is hours divided
40:09into the daily wage. Okay. And on the
40:11right hand side we have whether the
40:15worker worked eight hours per day versus
40:18a worker who worked less than eight
40:20hours. And what you can see is from
40:23these wage schedules
40:25that there's a 33%
40:28difference in the hourly wages for
40:30full-time workers and part-time workers.
40:34Now we can look at the input part of the
40:36survey say for 2014.
40:39Take into account that the hours that
40:41are used are endogenous. Instrument them
40:44by land quality variables and the size
40:47of the plots. Okay, which is done in the
40:50second column. So this is IV.
40:53This is all fixed effect operation. So
40:55we control for what operation is. So
40:57none of these differences are due to
40:58being different operations. And you get
41:00the same answer. It's a 35% discount if
41:04the worker is full-time versus
41:06part-time. Okay? And they could do that
41:08for bullet pairs. And when we look at
41:10equipment, you don't see these fixed
41:12costs. So when we talk about these these
41:14transaction costs in the labor market,
41:16they're the ones that appear to be
41:18important, not know not so much for for
41:21equipment as well. Okay. So we have
41:24these discounts for part and full-time
41:27labor. We can use these combining with a
41:31distribution of hours worked, okay, and
41:34the wages that are paid to back out
41:37directly from the data what the fixed
41:39costs are. So um before we do that
41:42though, I just want to show you that
41:45indeed the discount is higher if the
41:47plot distance is higher. Okay, but that
41:51the trans what appears to be the
41:53transaction cost seem to be higher on
41:54the larger plots, which is you know not
41:56obvious why that's true, but just to
41:58show you that we're going to need to
41:59take that into account when we do the
42:01the calibration. Okay. So
42:05these discounts that we see by hours
42:07worked by an individual worker is
42:10consistent with the existence of fixed
42:12transaction costs. Okay. So we can
42:15assume that workers entering the labor
42:16market for offer work face some fixed
42:18transaction costs. Okay. And in
42:20equilibrium employers are going to have
42:22to compensate at least some part of
42:24those fixed costs when they hire workers
42:26in a competitive market. Okay. So we can
42:29hypothesize that this is what
42:31characterizes the wage schedule. So the
42:33hourly wage for a worker works eight
42:36hours is equal to some fixed cost plus
42:40the number of hours worked times the
42:41marginal hourly wage. That's W1. So W0
42:45is the fixed compensation. W1 is the
42:47hourly wage. And this applies. So the W0
42:50applies no matter how many hours are
42:52worked. Okay? W1 also applies but the
42:55total wage is going to depend on the the
42:57two things. Okay. So to calculate fixed
43:00cost we need the distribution of hours
43:01as I said so the average wage that's
43:05paid for part-time workers that is
43:08workers working less than eight hours.
43:11Okay we take that average weighted by
43:14the distribution of people in the
43:15different hour categories because the
43:17average hourly wage is is nonlinear
43:20right it depends on how many hours are
43:21worked. That's the definition of the
43:23hourly wage. We divide the the daily
43:26wage by the number of hours, right? And
43:28similarly for the full-time workers
43:31defined as those who work 8 to 12 hours.
43:33We don't see anyone working more than
43:3512. Okay, we get the hourly wage there.
43:37Okay. And the regressions told us that
43:40the hourly wage for full-time for
43:44part-time work is so 1.35 times the
43:49hourly wage for full-time work. Okay. So
43:52substituting into those equations with
43:54the marginal wage being 21 rupees that
43:56we can get from just observing workers
43:58work an additional hour over eight hours
44:00where the fixed cost is is is low. Okay.
44:03Yielded estimate of W0 of 178 rupees.
44:07Okay. What's that tell us? Means for a
44:09full-time worker 51% of the daily wage
44:13is fixed transaction costs. Okay. The
44:17average profits of a farm in this
44:20environment is about 4,000 rupees,
44:22right? So just paying upfront the 178
44:25rupees every time you hire a worker,
44:28okay, is not a trivial cost if if if
44:31this calculation is is correct. Okay, so
44:35it looks like transaction costs are
44:36really important. Okay, so now we're
44:39going to put them into um the the model.
44:42Okay. Um the relevant unit is the
44:45operation. As I said, farming is a
44:47sequence of operations, right? So for
44:50each operation, you have a demand for
44:52inputs to carry out that operation. So
44:54if the operation is weeding, you need to
44:58hire someone or not or use family
45:00workers to weed. If it's for spraying,
45:02you need to hire equipment for spraying
45:04and someone to to use the spray. Okay?
45:07So to to to fix ideas, we're going to
45:08have the simplest model, one operation.
45:11Okay? And then it can be extended to
45:13multiple operations. And I'm going to
45:15show you just with transaction costs as
45:18as modeled as you saw we're capable in
45:21this model to generate the U-shape of
45:25profits per acre by acreage. Okay.
45:28Without putting in artificial scale
45:31economies of any kind in the production
45:33technology. You know, it's not going to
45:35be sufficient to explain the increasing
45:38rise in profits per acre. you know, as
45:40we continue to increase scale without
45:43adding machinery, and I'll show you why
45:45machinery and machine uh scale economies
45:47are going to be ultimately important.
45:49Okay, so here's the the basic model.
45:53So,
45:54agriculture is is really
45:56straightforward. Okay, the the hard part
45:59is figuring out, you know, what to do
46:01about the vagaries of weather and using
46:03the latest technology, but the essential
46:06feature is that you're going to feed
46:09seeds that are planted in the ground.
46:11That's what you're doing. So, you feed
46:14them with water, you feed them with
46:15fertilizer, and you make sure that the
46:18competitors to the nutrition of your
46:20seeds are defended against. So, those
46:23could be pests or they could be weeds
46:26that compete. Okay. So output is
46:29basically a function of land and the
46:32nutrition that the seeds get. Okay. And
46:36we're going to hypothesize that that is
46:38constant returns to scale. If you double
46:40the seeds and you double the land and
46:41you double the nutrition, you're going
46:42to get double the output. Nothing
46:44magical, right? And no farmer would
46:46argue against that. Okay? It's all going
46:49to be how you do those things. All
46:51right? So that's that's the fundamental
46:53production function. And we're going to
46:55also assume it's it's not explicit
46:56there, but we'll estimate it in the the
46:58fully calibrated model that TFP, which
47:00be an outside parameter, is also scale
47:03invariant. Okay, that there's no magical
47:06TFP that is higher for large farms for
47:11no reason that anyone can think about.
47:13So we're not going to have any magic in
47:15this model. It's all going to come
47:16through the cost side and the input
47:18side. Okay. So we'll assume that the
47:20that E
47:23the the plant nutrition production
47:25function we'll start with the simplest
47:27one. It's only labor family labor and
47:29hired labor. Okay. So labor is used. It
47:32can be hired. It can be the family
47:34labor.
47:37So consistent with our our wage schedule
47:40for an individual worker, we now have to
47:42come up with a wage schedule that takes
47:44in account that we may hire more than
47:46one worker. Okay? And we'll assume for
47:49the moment, though we're going to end up
47:51estimating this, that you have to pay
47:53the fixed cost for every laborer that
47:55you hire. We'll we'll look at whether
47:57there's team hiring. So maybe you can
47:58save if you hire more than one at once,
48:01but that that doesn't happen very often,
48:03but I'm going to eventually find that
48:06there is no team hiring. Right? So this
48:08then when we take in account
48:11the number of hours you work that has to
48:13be divided up into laborers because each
48:15labor is not going to work more than
48:17some limit of hours say eight hours a
48:20day right so we get this ceiling
48:22function I don't know if you're familiar
48:24with ceiling functions but it basically
48:25says that the fraction that's in the
48:27parenthesis is always going to be at the
48:30upper digit of that fraction so for
48:33example if you only need four four hours
48:36of work and The limit that any worker
48:39will work is eight hours. So that
48:42parentheses number will be 1/ half,
48:44right? Then the ceiling is one, right?
48:46So you're going to pay the fixed cost
48:48once. So it's W 0 + 4 * W1 will be the
48:53cost for four hours of work. Now, if you
48:55need 11 hours of work to carry out this
48:58operation because no worker works more
49:00than 8 hours, you're going to have to
49:02hire two workers. So 11 over 8 is going
49:05to be greater than one. So, uh, you kick
49:08up to the the next highest digit, which
49:10is two. You're going to have to pay
49:11twice the fixed cost and then the 11
49:13hours times the the wage cost. So,
49:15that's that's what's built into the to
49:17the model. Okay.
49:19And you can also figure out what the
49:21opportunity cost for applying
49:25family labor uh on a farm where you the
49:29the worker could work also off the farm.
49:32Removing one laborer from the family
49:33labor force working on the farm or a
49:35marginal hour. Okay, the marginal hour
49:38is simple. If workers are working both
49:40on the farm and off the farm, that's W1.
49:43But this is the equation you get if you
49:45said let's move a worker from one to the
49:47other. So that's built into this model
49:49as well. That's a floor function. It's
49:50pretty straightforward. Okay. So then we
49:53get farm profits. The farmer's
49:56programming problem is to maximize
49:58income which takes an account that there
49:59can be offfarmm work. Okay. That's it.
50:02Really simple. Okay. And this model is
50:04capable of giving us the U-shape.
50:07Okay. So given the labor transaction
50:10costs, there's going to be three
50:11regimes. There are going to be three
50:12regimes characterizing use of family and
50:15hired labor with two land thresholds in
50:18the model. Okay. So the first the
50:20regimes are going to be defined by land
50:22by the size of operate the scale of
50:25operation. Right? So we're going to
50:26start with the first regime starting
50:28from the very smallest farm and
50:30increasing land size up to some
50:32threshold a star. Now what's that? At
50:34the very smallest farms
50:37family members will be working on the
50:39farm and off the farm. the scale is is
50:41not not sufficiently high that it can
50:45consume all of the hours that are
50:46available for family workers. So they'll
50:48work off the farm as well. Okay. So the
50:52critical upper bound is when all of the
50:54hours of all the family labors are
50:56exhausted and now it becomes potentially
51:00profitable to bring in higher more
51:03labor. Okay. So up to that AAR
51:05threshold, all of the labor is family,
51:08but the family members are working off
51:10the farm. And in and you can see that in
51:13the data, right? At the smallest farms,
51:15the the average days worked in the labor
51:17market is much higher than it is for the
51:19smallest farms, right? So a lot of the
51:21the hours of farmers who are farming
51:26small plots of land is used actually to
51:27work for other other farmers. Okay? So
51:31now at the upper bound a a star okay and
51:36throughout the span of land sizes up to
51:39a star the marginal hour the marginal
51:42product of an hour is equal to the
51:44marginal wage. Okay so if all farms were
51:47in this regime then labor would be
51:49allocated efficiently across farms.
51:53So fine, but then we go up to the AAR
51:57threshold. Land size continues to
52:00increase, right? But now
52:03the labor force from the family is fully
52:06exhausted at at say L. Okay. So we then
52:11have a regime of autoarchy.
52:14the the marginal increase in acreage
52:16above a star, okay, would induce more
52:21hours that would be productive. Okay,
52:24but because of this fixed transaction
52:26cost, the farmer if when he compares
52:29what profits are by hiring that
52:30additional hour of work from the labor
52:32market, paying that fixed cost with not
52:34hiring that labor, profits are clearly
52:37dominated by not hiring labor. And this
52:40continues as land size increases in this
52:42regime until we hit the threshold a
52:45double star where it finally becomes
52:47profitable to hire that extra labor and
52:50pay the the fixed cost. But in that
52:53interval which is non-trivial in the
52:56data given the high fixity of of costs
52:59it's it's understandable as land size
53:02increases in this interval
53:06profits per acre decline.
53:10And the marginal product of labor
53:13increases over this interval as acorage
53:16increases which tells us that in this
53:19regime the marginal product of labor is
53:22higher than in among the smallest farms.
53:25Okay. And is also unequal across farms
53:28because what the marginal product is
53:30going to be among farms are going to
53:31depend on the the land size of the farms
53:34and the size of the family labor force.
53:36So if all farms were a taric there would
53:39be a labor misallocation and as I said
53:42at least a third of the operations in
53:44these data take place in otterarchy. So
53:47there's a lot of misallocation of of
53:49labor but the key point is in this artic
53:52regime there's declining profits
53:55per acre as acreage increases because
53:58the farmers are not using higher labor.
54:03Okay, so that's what I said. Then we
54:07reach the A star double star regime and
54:10now the farms will start to hire labor,
54:11but they'll be incurring these these
54:13high fixed costs, right? They'll be
54:15using little fewer hours and then as
54:18farm scale expands, they'll be expanding
54:21the hours that each worker works. So the
54:23fraction of total labor costs that is
54:26fixed cost decline and the average
54:28hourly wage is declining. So the cost of
54:31labor in hours is lower the larger the
54:34scale. And this with a simple
54:37simulation, this is what you get
54:39basically, right? So for the smallest
54:41farms up to a star, you get that as you
54:44increase land size, it doesn't affect
54:46profits per acreage because they're
54:47always allocating labor such that the
54:49labor that's used on the farm is equal
54:51to the marginal margin the wage. Okay?
54:54So there's in a sense it's efficient
54:56across these scales, right? Then we
54:58reach a star and now because of the
55:01limits of the family labor force and the
55:03reluctance to hire labor there's this
55:05decline to reach a double star in which
55:08case they start to hire labor and
55:10profits per acre start to rise.
55:13At best they're never going to exceed
55:16the the that for the smallest farms with
55:18this model but you do get the decline.
55:21Okay. So for if most of the farms are
55:25concentrated in in this regime and in
55:27this little this simple uh simulation we
55:31get it about 12 acres which is 96% of
55:33the farms right you're just you're going
55:35to see basically this this inverse right
55:37but when you increase the scale you're
55:39going to see the upward move okay so we
55:43saw the we get the three regimes we get
55:45the U-shape just from transaction costs
55:47okay and then we can look at the
55:48marginal return to to to land as as
55:51Well, okay. So now, if we want though to
55:55have the ability of larger farms to e to
55:58be even more productive than the
56:00smallest farms, we've got to do more
56:02than eliminate transaction costs. Okay,
56:05we've got to we think right we
56:07hypothesize add machinery that differs
56:09is differs in in capacity. Okay,
56:14so we do that and here I'm not going to
56:17go over the details of this because this
56:19is a lecture that's concentrating on
56:20transaction costs, right? But just to
56:22point out that you can't avoid the issue
56:25of machinery for two reasons. One is it
56:28is going to be responsible for the scale
56:29economies in the upper part of the
56:31distribution and also because machinery
56:34and labor are substitutable
56:36and therefore what happens to
56:39productivity with regard to the fixed
56:42costs in the labor market is going to be
56:44affected by the ability to use
56:46machinery. Okay. So a key distinction in
56:49thinking about machinery is the
56:51distinction between the time you use a
56:53machine and the effective capacity of
56:55the machine. And by capacity we mean how
56:58much the machine can process per unit of
57:01time. So when we think of computers, we
57:02know that processors have different
57:04capacity, right? They can different
57:06meglops. They can render a 4K video in
57:08two minutes or six minutes, right? That
57:10you know a computer is not a computer.
57:12It's a processor with different
57:14capacities. Right? Now the key thing
57:18about agriculture which differentiates
57:19it from a lot of other industries is
57:23that capacity has a spatial component.
57:26It depends on the scale of operation in
57:29a very mechanical way. All right. So,
57:31two examples
57:33you can get. You can purchase an eight
57:35row harvester and a four row harvester.
57:40A four row harvester would be perfect if
57:42you have four rows to harvest, but an
57:44eight row harvester would be irrelevant
57:46if you only have four rows to harvest.
57:48Okay? So, four row farms are never going
57:52to be using eight row harvesters. Okay?
57:56If eight row harvesters are more
57:58efficient than four row harvesters, you
58:01can see where scale economies are coming
58:02from. But we have to also pay attention
58:04to what the pricing schedule is for that
58:06capacity. Another example, sprayers.
58:10Okay, so the capacity of a sprayer is
58:12measured by the radius of the spray. And
58:15clearly the radius of the spray is going
58:19is spatial. And so if the radius of the
58:21spray is 10 mters, but you've only got a
58:23farm of 5 mters, then you're wasting
58:26spray and you're not going to use the
58:28the higher capacity sprayer. Okay? So
58:30there's lots of examples where
58:33harvesters and sprayers and threshers
58:36have different capacities and the
58:39usefulness, the effectiveness of the
58:40capacity is going to depend on scale. So
58:42we can hypothesize there's a
58:46capacity function that depends on scale
58:48and has a positive first division uh
58:50derivative. That's the five function.
58:53All right. So and then what about the
58:56cost though of capacity? So we can
58:58hypothesize that the rental price per
59:00unit of time. So the machine is used m
59:03hours, right? So the rental price per
59:06hour is going to be equal to some fixed
59:09price uh P Q times the amount of
59:12capacity to the new. Okay. So we're
59:14going to allow for nonlinear pricing.
59:17And if new is less than one, then the
59:19cost per per hour of increasing capacity
59:23is going to exhibit in a sense economies
59:25of scale. So we're going we're going to
59:28end up estimating new and we're going to
59:30estimate the fi function from the iat
59:33data. Okay, we have to take into account
59:36that the capacity farmers choose is
59:38endogenous and therefore the price they
59:41pay is indogenous and we do that
59:43econometrically in the in the data but I
59:46don't have time to talk about the
59:47details of of that. Okay, but first it
59:51we now modify the model. So now the
59:53nutrition production function is going
59:55to have both labor L and machines and
59:58capacity. And now our production
1:00:00function has the ability to substitute
1:00:03between machine hours and labor. Okay,
1:00:06we have a delta function for that. We
1:00:08have our capacity function built in. We
1:00:10have the shares of of labor and uh and
1:00:12machinery in our production function.
1:00:15Okay.
1:00:17Now, the the total hourly cost of course
1:00:19of actually running a machine is going
1:00:22to equal to the rental cost of the
1:00:23machine plus of course the labor cost of
1:00:26running that machine. So, we assume that
1:00:28for each hour you run a machine, there's
1:00:31theta amount of labor. But it turns out
1:00:33not surprisingly, it's one to one. The
1:00:35theta is basically one. You can see it
1:00:36in the data which has the number of
1:00:38hours a sprayer is used for example and
1:00:40the number of hours in spraying labor is
1:00:44used. and they're essentially identical.
1:00:46Okay, so now we have the expanded profit
1:00:49function which has the the the modified
1:00:52production function. Okay, uh in the way
1:00:55we've specified this production
1:00:57function, you can actually determine
1:00:58optimal capacity independent
1:01:01of the rest of the production function
1:01:03parameters. Okay. And it satisfies this
1:01:06first order condition.
1:01:08And you can from that first order
1:01:11condition determine that capacity is
1:01:14going to increase with uh with scale and
1:01:17capacity is going to be higher if the
1:01:19labor that you use to run the the
1:01:22machine is is higher priced. Okay. Not
1:01:25not surprising. Okay. So
1:01:29you can get this ever rising thing by
1:01:31introducing machines. All right. So now
1:01:34I want to show you tests of the
1:01:37existence of scale economies and what we
1:01:39see in the in the data in terms of
1:01:42production and pricing and then uh I'll
1:01:45show you our estimates of new and of the
1:01:48five function and then we'll do the
1:01:49counterfactual in uh the remaining 10 or
1:01:52so minutes that I have. Okay. So first
1:01:55we see that if we look at the fraction
1:01:57of operations that use lowhour wage
1:02:02labor by low hour I mean less than eight
1:02:05eight hours. Okay, we see that at first
1:02:08rises, then it falls, which is really
1:02:10consistent with the data, right? So for
1:02:12the smallest farms, right, when they
1:02:14first start hiring labor, it's going to
1:02:16be low hours, right? But then as the
1:02:18farms get bigger, they're going to use
1:02:20more and more hours and there's less
1:02:21lowour labor the higher the scale. As a
1:02:24consequence, because low hour labor is
1:02:27more expensive labor, the real hourly
1:02:30wage has this shape. Okay, it rises with
1:02:33scale and then it ultimately starts to
1:02:35fall. Okay, because of the increasing
1:02:38use of small scale, small hour labor.
1:02:41All right, so if you run regressions of
1:02:46plot size on
1:02:48what fraction of the labor is low hour,
1:02:52you see that on average it's negative.
1:02:54The that the larger plots are using are
1:02:56less likely to use low hours labor. It's
1:02:58sort of dominated by the the upper tail
1:03:01here. Okay. And as a consequence, the
1:03:03average hourly wage as scale increases
1:03:06is lower. Okay.
1:03:08Now, we always worry about plot quality,
1:03:13right? We have the 24 measures, but you
1:03:15know, someone can always say, "Oh, you
1:03:17know, you're there's something else
1:03:18going on with these plots." Okay? But
1:03:20what we can do is plot fixed effects,
1:03:22and we can use the classic instrument of
1:03:24microdevelopment economics,
1:03:27rainfall. exogenous not determined by
1:03:30any decisions of farmers and rainfall in
1:03:32these icat villages importantly affects
1:03:35the demand for inputs. Okay. So what
1:03:37we're going to do is we're going to take
1:03:38the same plot and look over time at the
1:03:43variation in the amount of labor that
1:03:46the plot uses determined by the
1:03:50variation in rainfall. Okay. So in in
1:03:53principle to show this that when rain is
1:03:55more ample the demand for labor goes up.
1:03:57Right? So for a given plot, the number
1:04:00of hours is going to increase
1:04:04if rainfall is higher. And what our
1:04:07model implies is that the marginal
1:04:11increase in the demand for labor for
1:04:13lowhour labor is going to be different
1:04:15for small plots and for large plots. In
1:04:17particular, for small plots, there's
1:04:19going to be an increase in demand for
1:04:21lowhour labor when there's an increase
1:04:23in rainfall. And for the larger plots,
1:04:26there's going to be a decrease because
1:04:27they're they're already employing the
1:04:29low wage labor, low hour labor, and now
1:04:31they're expanding that labor force.
1:04:33Okay, so I'm going to show you what
1:04:36those coefficients look like by farm
1:04:38scale, right? These are the sort of
1:04:40linear quadratic equations, right?
1:04:42First,
1:04:43this is a plot of what the coefficient
1:04:46of rainfall is on profits, okay? With a
1:04:4995% confidence interval. The x- axis is
1:04:53the the scale of the plot. Okay, this is
1:04:55plot fixed effects
1:04:58and the the blue line is the
1:05:00coefficient. So what it says is what we
1:05:02know from these data. Higher rainfall
1:05:04increases profits and it's about the
1:05:06same for every plot size, right? This is
1:05:09profits for per acre. Okay. But now when
1:05:12we look at the effects of rainfall with
1:05:15plot fixed effects on the fraction of
1:05:17operations using low hour hired male
1:05:20labor, we can see at the low scales the
1:05:24coefficients positive. Meaning that an
1:05:26increase in rainfall increases the use
1:05:29of lowhour labor. But for the larger
1:05:32farms, it's the opposite. An increase in
1:05:35rainfall decreases lowhour labor.
1:05:39And concominantly we see when we look at
1:05:41average hourly wages that for the larger
1:05:44farms when rainfall is higher that
1:05:46actually reduces their hourly wage costs
1:05:48while it tends to be always positive for
1:05:51the for the larger scale plots for the
1:05:54smaller scale plots. Okay again these
1:05:57are this is a coefficient right the
1:05:59coefficient actually changes sign across
1:06:01the plots for both of these things in a
1:06:04way that the model uh explains. So we
1:06:06see one implication of these fixed costs
1:06:09seems to be true in these data that's
1:06:12that
1:06:14and therefore the model that's
1:06:15predicting the use shape seems okay.
1:06:17Okay. So I'll skip over all of the
1:06:20econometrics of of sprayers just to show
1:06:22you that the hours of sprayer use first
1:06:26increases and then decreases. That can
1:06:29only be consistent with hours per acre
1:06:32going down with an increase in capacity.
1:06:34And in fact we see that the log of
1:06:38capacity of sprayers increases with with
1:06:41scale as we would expect. The amount of
1:06:45weeding labor costs go down with scale
1:06:47because of the increasing use of higher
1:06:49capacity sprayers. Cost of spraying is
1:06:52relatively stable because you're both
1:06:54using less hours of spraying but
1:06:57increasing capacity and paying a little
1:06:58more for it. Sprayers. Now what about
1:07:01the cost of sprayers? So this is an
1:07:03actual
1:07:06price list for sprayers in India. Okay,
1:07:09these are the pictures of the sprayers.
1:07:11This is the liter per hour. That's the
1:07:13measure spray capacity, which is the one
1:07:15I used in in showing you those those
1:07:17graphs. And you can see that the higher
1:07:19powered sprayer is 13 times has 13 times
1:07:23more capacity than the lower one, but
1:07:26the price is
1:07:29nowhere near 13 times. Right? So it
1:07:31looks like the new is clearly less than
1:07:34one. Okay. And in fact, we do see that
1:07:37the farmers pay more per hour for
1:07:41spraying for the sprayer they rent that
1:07:44are large compared to small. They're
1:07:46clearly buying higher capacity sprayers.
1:07:48Okay. Now
1:07:51the
1:07:52you know what we see is waiting hours go
1:07:54down with acreage, sprayer hours go down
1:07:56with acreage, price goes up, which you
1:07:58saw on the graphs. Okay, we estimate new
1:08:02from our data using the the the moments
1:08:05of from the capacity functions and from
1:08:07the price schedule using his instruments
1:08:09the land quality and and scale. We
1:08:12parameterize the five function which
1:08:14relates the capacity to scale as a
1:08:17quadratic which means that we can
1:08:20actually if if this works um give
1:08:24provide the optimal scale the scale at
1:08:26which the capacity is uh no longer
1:08:30increases profitability. Okay. So this
1:08:33is our estimate of new. How does it
1:08:36compare? Our GMM estimate of new from
1:08:38the actual data of the use of sprayers
1:08:43in India is.3.
1:08:46The priceless schedule, if we have four
1:08:48observations, we could still estimate
1:08:50new is.5. We look at a US price schedule
1:08:53for sprayers, it's 0.1. So, we're in the
1:08:55ballpark. These are not really
1:08:56statistically significantly different.
1:08:58Maybe there's more scale economies for
1:09:00sprayers in the US than India, but it's
1:09:02not very important. So we think we have
1:09:04a good estimate of of um of new. Okay.
1:09:08What about our estimate of the
1:09:10parameters of the five function?
1:09:13They're precisely estimated and the
1:09:16optimal scale at which there's no more
1:09:20benefits from increasing capacity is 25
1:09:24acres.
1:09:25Now why is it 25 acres?
1:09:28It's obvious that there are much bigger
1:09:30farms in the world elsewhere. Uh so why
1:09:34would we get 25? We get 25 because there
1:09:38are no
1:09:42equipment available in India for farms
1:09:45larger than 25. 25 is the top
1:09:50of the farm size in most of India.
1:09:53Right? So no one is going to be selling
1:09:54equipment that's relevant for 100 acre
1:09:58farms in India. So we can only calculate
1:10:01the optimal farm scale for existing
1:10:04equipment in uh in India. Now we could
1:10:07do simulations where we change the new
1:10:09and the capacity function and get larger
1:10:11scales. But you know that that goes
1:10:13beyond what we think would would be
1:10:15reasonable right now. Okay. So, for one
1:10:17example,
1:10:19you know, there's you can spray with a
1:10:21drone and this drone can, you know, can
1:10:24spray uh a 4,000 to 6,000 square meter
1:10:28area in just 10 minutes, right? Well,
1:10:30none of these three acre farms are going
1:10:32to, you know, rent a sprayer. So,
1:10:34there's clearly better technology than
1:10:36what we saw from the Chrysen graph price
1:10:39list. Okay. So, we're going to calibrate
1:10:41the model.
1:10:46three aims. Does the calibrated model
1:10:50fit the data? Right? And we're going to
1:10:53fit it to the U-shape that we see. We're
1:10:55going to fit it to two moments only,
1:10:56right? Uh we're going to calculate the
1:11:00optimal size of farms because the the
1:11:02true optimal size of farm is going to
1:11:03depend on the price of labor and the
1:11:05other parameters of the production
1:11:07function. So it it will embed the five
1:11:09function in it but the actual optimal
1:11:11scale will depend the price of labor and
1:11:13other factors. Okay. And then we'll do
1:11:15the counterfactual land consolidation,
1:11:18land redistribution. Okay. The
1:11:21calibration is also going to enable us
1:11:22to identify things you don't see in the
1:11:24data. So, for example, we'll be able to
1:11:26distinguish between hiring costs paid by
1:11:29by farmers, which we could see in the
1:11:31schedules, but also the fixed cost of
1:11:34entry to the labor market, which you
1:11:35can't see in any schedule of of
1:11:37laborers. And it's going to turn out
1:11:38they're going to be different. We'll
1:11:40also be able from the calibration to
1:11:42estimate what the marginal product of
1:11:43labor is on Atari farms by farm scale,
1:11:46which again you can't see in the in the
1:11:48data. It's a structural estimate.
1:11:51Okay. So and then from that we'll be
1:11:53able to to estimate the true marginal
1:11:55cost of hiring a laborer across
1:11:57different scales of of farms which will
1:11:59allow. Okay. So I won't have time to
1:12:02talk about all the details of the
1:12:04parameterization. Um we have to take in
1:12:06account that there's multiple
1:12:07operations. We know the number of
1:12:09operations is actually 12 in the data.
1:12:12Okay. So we can also look at the data to
1:12:14see how many days each operation uh
1:12:18takes. It's going to depend on farm
1:12:20size. So we we know what that is. Okay.
1:12:23We assume a Cobb Douglas
1:12:25parameterization of the production
1:12:26function where we have a TFP parameter
1:12:28SI uh 21 days the total number of days
1:12:32worked uh on a in a particular
1:12:34operation. We have our production
1:12:36function. We have estimates of new that
1:12:38we bring to the calibration and of W1
1:12:40and we have the parameters of the FI
1:12:42function. Otherwise we we calibrate the
1:12:44other parameters there just from the the
1:12:48the two moments we fit. We assume that
1:12:51the number of workers a family has which
1:12:53is about what the average is is three
1:12:54workers and the maximum that a worker
1:12:57will work as eight hours which we have
1:12:58to know as well. We allow hiring cost to
1:13:01vary with acreage and we take into
1:13:03account that there can be economies of
1:13:05scale there. So we get calibrated
1:13:08parameters. Okay. and the implied
1:13:10elasticity of substitution between labor
1:13:12and um the machines is very high. We get
1:13:17that the the fixed cost just for the
1:13:20calibration we didn't build it into the
1:13:22model is 159 uh rupees that compares to
1:13:26what we estimated directly
1:13:27econometrically from the the the
1:13:30schedules of 177. So we fit that really
1:13:32well. uh we estimate that the fixed cost
1:13:36a worker incurs going into the market is
1:13:38actually 42 rupees which is less than
1:13:41what the farmers pay and we think that's
1:13:44because our workers live in the village.
1:13:47Okay, so their transaction costs are
1:13:49actually lower than the transaction
1:13:51costs incurred by workers 25% of whom
1:13:54come from outside the the village. Okay,
1:13:57the marginal product of an infram
1:13:59marginal hour on average is 43 rupees.
1:14:03That compares to 21 rupees for the
1:14:05marginal hour in the labor market. So
1:14:07it's more than twice as high. Okay. So
1:14:10labor is surely underutilized on the 30
1:14:14more than a third of farms that are
1:14:17operations that are target which make up
1:14:1950% of the the land size. Okay. We can
1:14:22also get the average marginal cost of
1:14:26hiring a laborer by farm size for the
1:14:29for the largest farm because of what
1:14:32appears to be higher hiring cost for
1:14:34large farms. It's 43 acres for hiring a
1:14:36worker. That's not the marginal hour.
1:14:38The marginal hour is still 21 acres.
1:14:40Sorry, 21 rupees for the smallest farms
1:14:43and the largest farms. It's just 44 for
1:14:47the Ataric farms. Okay, so we have that.
1:14:49How does it fit? So first it's spot on
1:14:52for predicting the fraction of farms in
1:14:54autoarchy 33.8 versus 34 in the data.
1:14:58Okay, this is the fit to um what we're
1:15:00trying to fit which is profits per acre
1:15:03by land size. So this is the data and
1:15:06this is our model. Okay, we don't have
1:15:08heterogeneity in the calibration. So our
1:15:11model predictions are going to be less
1:15:13smooth than the than the data. We also
1:15:15fit it to output per worker which rises
1:15:18with with farm scale. This is the data.
1:15:20This is our fit to the data. The green
1:15:24line shows you what would happen if you
1:15:25ignored machinery, right? You're just
1:15:27not going to get the upper tails here
1:15:28and you're going to get a flat
1:15:30relationship of output per worker as
1:15:32well. What about a relationship we don't
1:15:35fit to? Well, we can look at labor hours
1:15:38per acre. This is what the data have
1:15:40show us and this is what we fit. So, it
1:15:42does it does pretty well. Okay. So we
1:15:45have chronometric tests of the
1:15:47implications of the model that I showed
1:15:48you and we have this calibration that
1:15:50that seems to fit the major moments in
1:15:52the data pretty well. So let's let's at
1:15:55least pretend that we have the right
1:15:57parameters and the right model and let's
1:15:59carry out our counterfactual
1:16:02consolidation. Now there's going to be a
1:16:04fixed point issue here, right? because
1:16:06the the optimal scale of the farm is
1:16:08going to depend on the equilibrium wage
1:16:10rate and the equilibrium wage rate is
1:16:12going to depend on what the equilibrium
1:16:14number of laborers there are and what
1:16:16the demand for labor is. So we have to
1:16:19calibrate the model do the
1:16:22counterfactual using the parameters of
1:16:24the model which takes into account this
1:16:26fixed point. Right? So what we're going
1:16:27to get when we move the all farms to the
1:16:30optimal scale is an optimal scale that's
1:16:33consistent with the equilibrium wage
1:16:35that would exist given the demand for
1:16:37labor at that optimal scale farm. Now
1:16:40when we do this where what are the
1:16:42sources of the gains? One source of
1:16:44gains is that we eliminate farms. So
1:16:47therefore we eliminate the misallocation
1:16:49of labor across farms. Okay. We also
1:16:52reduce the the the amount of fixed cost
1:16:55in hiring labor. Okay. And we reduce low
1:16:58hour usage because all our farms are are
1:17:01bigger. And of course we take we exploit
1:17:04the scale economies associated with with
1:17:06machinery. Right? So those are all the
1:17:08sources of the of the gains. So what do
1:17:10we get? Okay. There's also details. We
1:17:13have to convert hours that we see in the
1:17:15data into laborers which we get from
1:17:17theat data. We use the number of farms
1:17:20in India from the 2011 Indian census and
1:17:23we uh have the size of the urban labor
1:17:26force. We're going to assume unlike our
1:17:29surplus labor development economists of
1:17:32the old days like Arthur Lewis and u and
1:17:35others that there is an actual wage
1:17:37elasticity and the best estimates of
1:17:40that for India and across a lot of
1:17:41countries is the urban wage elasticity
1:17:43is minus.4. So we're going to use that
1:17:45one in our equilibrium model. Okay,
1:17:49the Zamanderia Commission did the same
1:17:51exercise for 10 acres and they found
1:17:53that you could reduce the number of
1:17:54farmers by 66% if you did that. Okay, so
1:17:57here's our estimates. Let me summarize
1:17:59them. The mean farm size is going to go
1:18:02up from 3 acres to 24 acres. The number
1:18:05of farms is going to be reduced by 87%
1:18:08by 80 by 80 million farm over 80 million
1:18:11farms. The labor force size is going to
1:18:13be shrink is going to shrink by 16%.
1:18:17Total output is going to rise by 42%
1:18:19with these fewer farms and output per
1:18:22worker because there's fewer workers is
1:18:24going to rise by 68%. Now I don't know
1:18:27of any intervention policy intervention
1:18:30that microeconomists
1:18:34estimate from RCTs that comes close to
1:18:37increasing output per workers by 68%.
1:18:40But that's this is what this this will
1:18:42do. So there's 83 million surplus farms
1:18:44and there's 29 million surplus
1:18:46agricultural laborers. Now we were a
1:18:49little bit surprised by the fact that
1:18:51the size of the labor force didn't
1:18:53shrink that much. And the reason for
1:18:56that is that because of the existence of
1:18:59a large fraction of farming, okay, labor
1:19:02is actually underutilized
1:19:06right now. Okay. So when we eliminate
1:19:08our taric farms that actually increases
1:19:10the demand for labor which is offset of
1:19:12course by the the the
1:19:16substitution of machinery for farms.
1:19:18Okay.
1:19:20So machine capacity rises by 72%. So a
1:19:24lot of this comes from the use of
1:19:25machines. All farms now use machines as
1:19:28opposed to most of the small farms not
1:19:30using machines at all. Um the average
1:19:33machine hours per acre at seven acres is
1:19:367 at the optimal 24 acres it's it's
1:19:40almost four right so there's a massive
1:19:42increase in the use of machines uh and
1:19:44and substitutability so this is what I
1:19:47discussed okay how restricted is labor
1:19:50useic farms a oneacre farm uses about 10
1:19:53hours per acre per day but on the worst
1:19:56of the farms they use only you know less
1:20:00than three acres per Okay.
1:20:02All right. So,
1:20:05part of the conclusion is obvious. It
1:20:07would appear there are too many farms.
1:20:08It would appear that the trajection
1:20:10costs of labor are a real impediment
1:20:13given existing farms to the productivity
1:20:16overall of agricultural labor. Okay? and
1:20:19that there are real positive scale
1:20:21economies associated with saving on
1:20:24labor and the the the true features of
1:20:28the available technology and machinery
1:20:31that exists in India. So the last
1:20:34question is so why aren't farms
1:20:36increasing their scale? Why doesn't a
1:20:37farmer increase the scale? And the
1:20:39answer is that there's a real challenge.
1:20:44This is the challenges of a policy but
1:20:45not going to talk about this as a policy
1:20:47just as why we don't get this naturally
1:20:50in the market. Okay. Average farm size
1:20:52is three. You're not going to start to
1:20:54get scale economies till you're around
1:20:5612 acres. You got to So there's a big
1:21:00chasm in the U-shape between 3 acres and
1:21:0312 acres. Adding an acreage to a
1:21:06threeacre farm actually lowers profits
1:21:08per acre. Okay. So that means a farmer
1:21:11who wants to get another acreage from a
1:21:13fellow farmer because it needs to be
1:21:15contiguous. Okay, that fellow farmer has
1:21:17a higher return on that acreage than the
1:21:20farmer would get. And so the offer the
1:21:23asking price for that acre is going to
1:21:26be higher than the returns to the to the
1:21:28farmer at that increment. The farmer
1:21:30therefore has to leap over that chasm
1:21:33from three acres to 12 acres by
1:21:36simultaneously
1:21:37transacting with multiple farmers, maybe
1:21:41as many as eight farmers, all of whom
1:21:43have very small plots. And those eight
1:21:45farmers that are relevant are the
1:21:47farmers that are contiguous to his land.
1:21:50So there's holdup problems and other
1:21:52issues. And in the history of the world,
1:21:54it would appear that there's never been
1:21:57an endogenous land consolidation that
1:22:00has occurred.
1:22:02There have been exogenous ones, but I I
1:22:04have to stop to to to take more
1:22:06questions that I can talk about. There
1:22:08was one in world right after World War
1:22:09II in France. And just to show you what
1:22:12it looked, this is what France
1:22:13agriculture looked like after World War
1:22:15World War II. And this is this is from
1:22:18an airplane view. This is the same land
1:22:21after the reform. There's a tractor.
1:22:24Couldn't get a tractor here. All right.
1:22:26So, let me take questions.
1:22:27We have any question? Thank you, Mark.
1:22:29Um, do we have any questions from the
1:22:31front row? There are a few from the Q&A.
1:22:33Let's do that. Maybe it'll warm up the
1:22:35front row people.
1:22:37Okay. Um, Yonas uh Goten had a question.
1:22:43Let me see where it is here. Maybe I can
1:22:46just add him as a panelist. Jonas was
1:22:49saying these fixed these search costs
1:22:52uh that the let me see I want to find
1:22:55his exact word he said I didn't
1:22:56understand the argument for high search
1:22:58costs in these labor markets on slide 39
1:23:01right below information is given that
1:23:03laborers work on average for seven
1:23:05different farmers isn't this an argument
1:23:08against high search costs and why would
1:23:10labor contracts be daily if search costs
1:23:13are so high it seems a super inefficient
1:23:16contract in a setting with high search
1:23:17costs costs
1:23:18it well it it um the problem is the
1:23:22scale, right? You you would like to hire
1:23:25labor on a long-term contract to avoid
1:23:27these these costs, right? But the reason
1:23:30you can't is you don't know how much
1:23:32labor you're going to need in advance.
1:23:35And given the small scale, uh the type
1:23:37the number of hours you need is actually
1:23:39very small. So you you just you just
1:23:41can't do this these contracts. So that's
1:23:43why scale is really important part of
1:23:44the story here. You know, it's not that
1:23:46search costs don't exist in, you know,
1:23:48in all labor markets, right? We all
1:23:50incur search costs even for any of the
1:23:52jobs that we have now, right? There's
1:23:54there's clearly transaction costs all
1:23:55over. But here, it's it's just much
1:23:57worse that prevents long-term contracts.
1:24:00It's both the sarcastic nature of the of
1:24:04the work that you need to do, which you
1:24:06can't know in advance. Okay? Sometimes,
1:24:08by the way, it's zero, right? It that
1:24:10that there's no rain and there's there's
1:24:12literally nothing to do. there's just
1:24:14lots of non-trivial instances where the
1:24:17farmer has nothing. Okay. And also the
1:24:20the small scale that that prevents it.
1:24:23So it's a good question, right? But it
1:24:25it it just highlights how special
1:24:27agriculture is compared to other
1:24:29industries and why through this this
1:24:31scale issue.
1:24:32It seems that you have arguments for the
1:24:34search costs, but but a lot of the the
1:24:37ultimate conclusions just could be any
1:24:39kind of a fixed cost.
1:24:40Yeah. No, exactly. It doesn't dep right
1:24:42it doesn't depend on that we're we're
1:24:44making up stories about what they are
1:24:46right it I mean there's commuting costs
1:24:48too right we we talked about that right
1:24:50that you have to convince a laborer to
1:24:52come to spend you know an hour to come
1:24:54to your farm on a particular day he's
1:24:56not going to do that unless he feels
1:24:57he's fully compensated so that's another
1:24:59component but again we're making up
1:25:00stories for them right that you know we
1:25:03we can't know for sure we could ask
1:25:05people I suppose but but we for sure see
1:25:08that the pricing of hourly
1:25:12work is clearly exhibits having fixed
1:25:15costs. Do we have any questions from the
1:25:17front row?
1:25:18Yeah, Mark, can I ask a question?
1:25:20Sure.
1:25:21Um, I'm just uh the upper threshold
1:25:24stuff with a double star. It seemed to
1:25:27be where it makes sense to just hire a
1:25:29worker, but wouldn't there be like
1:25:32another threshold for hiring two
1:25:33workers?
1:25:34Yes, exactly. Yes. And in the That's
1:25:37exactly right. You're completely
1:25:38correct. And in the calibration we allow
1:25:41for that. It's just the simple model.
1:25:43You're it's absolutely right. Yes. We
1:25:45allow for for as you you know you go
1:25:48with one threshold where you you've
1:25:50hired a worker now he's worked his eight
1:25:51hours and now the increase land size you
1:25:53got to hire another worker. Right.
1:25:55That's why we have the ceiling function.
1:25:56Right? So there is a cost to jumping the
1:25:59by the eight hours each time and each
1:26:01time it has to be incurred. But the
1:26:03averages are still still going to be
1:26:05lower than the larger farms, right?
1:26:06because the you know the full you've got
1:26:08one full-time worker and then a
1:26:10part-time worker compared to just a you
1:26:12know part-time worker but that's yes
1:26:13it's exactly right and as I said it's in
1:26:15the calibration and that that gets to
1:26:17the issue by the way of whether there's
1:26:19team hiring right so if you needed two
1:26:21workers and you could somehow hire a
1:26:23team and incur one cost right then then
1:26:25there wouldn't be that that jumpiness
1:26:27but in fact we find the data you seem to
1:26:29have to pay it per worker even if you're
1:26:32using more than one so the ceiling
1:26:34function seems to be the right way to to
1:26:35model it Any other questions?
1:26:38Yeah. Uh I had an observation maybe this
1:26:42is more anecdical uh but it is more like
1:26:46uh even small farms like in villages uh
1:26:50there are this corporate societies which
1:26:52like where people like pull up and hire
1:26:55machinery and labor. So but when we are
1:26:58like doing this modeling we are like
1:27:01looking at individual farms and
1:27:03individual plots but can we like
1:27:07consider like a society of multiple
1:27:12farms as like one large farm or
1:27:15okay yeah I got I got it.
1:27:18Yeah.
1:27:18Yes. This comes up all the time right?
1:27:21you know, why don't the these guys get
1:27:23together and coordinate, right? And then
1:27:27they can rent the large tractor and, you
1:27:29know, farm these different plots, right,
1:27:31as one contiguous farm. And in fact, the
1:27:34Zamon Theory Commission talked about the
1:27:36alternative to, you know, a massive
1:27:39forced land redistribution as forming
1:27:41these cooperatives of of farmers, right?
1:27:44I don't know why they don't do it, but
1:27:46for sure they don't. And the and which
1:27:49we saw is even at the individual farmer
1:27:52scale he the farmer doesn't even
1:27:54coordinate his own plots right that they
1:27:57start the operation at at different
1:27:59sizes. Now we can think of we can think
1:28:01of of of stories about this you know
1:28:03that the farmers would have to cooperate
1:28:05at a lot of dimensions. For example they
1:28:07might to do this correctly they'd all
1:28:09have to grow the same crop right because
1:28:12you know the timing and the inputs would
1:28:14be would be the same. But that
1:28:16eliminates, you know, diversification of
1:28:18of crops, right? And so if they're in a
1:28:20world of risk, that might that might be
1:28:22dangerous. But yes, it's it's a very
1:28:24good question about, you know, this this
1:28:26issue of cooperation and it it just
1:28:29doesn't it just doesn't happen. Now,
1:28:30it's not about a cooperation of
1:28:32financing, you know, a tractor. You
1:28:34know, if you've got a oneacre farm, you
1:28:36can't even use the the tractor very
1:28:38much, right? It it is it is about
1:28:39cooperation and increasing the scale.
1:28:42So, and it's been tried even, you know,
1:28:44to to try to form these cooperatives,
1:28:46and it just doesn't it just doesn't
1:28:47work.
1:28:48Mark, there's a question from Rakkesh
1:28:50Banerjee in the Q&A, and it's related, I
1:28:52guess, to the search costs, but he said,
1:28:54"Is it possible to know from the data
1:28:55what precisely are these fixed
1:28:57transaction costs for labor? From a
1:28:59policy perspective, and that's why I'm
1:29:01reading this one to you, because you
1:29:02talked about the policy perspective on
1:29:03the land side, but not necessarily the
1:29:05labor market side. from a policy
1:29:07perspective if one were to be interested
1:29:09in reducing these transaction costs we
1:29:11need to shed light on the precise nature
1:29:12of the of the costs so I guess as an
1:29:15add-on is there anything that you think
1:29:17on the on the labor side is there any
1:29:19kind of policy u that that's an
1:29:22excellent question as well we haven't
1:29:24given that that much thought I I guess
1:29:28because we were sort of skeptical that
1:29:29much could be done about that right but
1:29:32you know there's new technology now and
1:29:34certainly the dissemination and
1:29:36information through cell phone
1:29:37technology and everything else has
1:29:38certainly advanced, you know, compared
1:29:40to 1948, right? Where in principle we
1:29:42might be able to set up something where
1:29:44a farmer who needs work, you know, is
1:29:46able to use the that technology to to
1:29:49find workers. So, so I think that while
1:29:52we haven't applied our brains to
1:29:54thinking about it very much, it would
1:29:55actually be fruitful uh to think in that
1:29:58direction because it is it is it is an
1:30:01information. You know, we're not going
1:30:02to deal with the commuting cost. you
1:30:03know the spatial nature of the the farms
1:30:05cannot be eliminated right except by
1:30:08consolidation. So the fact that a worker
1:30:10has to travel 8 kilometers to get that
1:30:12job, there's still going to be that
1:30:13fixed cost. But in terms of finding a
1:30:16job, yes, there might be some some
1:30:18harnessing of current technology that
1:30:20could could reduce those costs and
1:30:23therefore reduce the amount of autaric
1:30:26farming that's taking place, which is
1:30:28sort of the the the most um the biggest
1:30:30divergence from the the prop the
1:30:33efficient allocation of labor.
1:30:35We're we're running out of time, but
1:30:37there's a a couple of questions and if
1:30:39you could give quick hits on one is
1:30:41I have an observation.
1:30:44Oh, go ahead. Sorry.
1:30:45Yeah. Um from the model we could see
1:30:48that um the agricultural model is
1:30:51frictionless
1:30:52um with low cost of such such cost and
1:30:56um less
1:30:57sorry what's
1:30:59I'm sorry the model is clearly has
1:31:01frictions in it.
1:31:02Yeah, frictionless. the agricultural um
1:31:05no the model has frictions in it. It's
1:31:07not frictionless.
1:31:09Okay. Um then
1:31:10that's the whole point of this lecture.
1:31:13Okay. Then um what I observe is that um
1:31:15the model is quiet about um exogenous um
1:31:19transition shock from weather.
1:31:22No, we used we used weather. No. Okay.
1:31:25because I went a a lot of the
1:31:27econometrics has uh village year fixed
1:31:30effects which are going to capture fully
1:31:32weather a and if you look back at the
1:31:35tables you'll see that so this is a good
1:31:37point you're raising and secondly of
1:31:39course we use weather variation as one
1:31:41of our ways of estimating the role of
1:31:44fixed costs in okay shaping the cost
1:31:47function but yes that's always very
1:31:49important and it's always a good
1:31:50question to ask when someone's working
1:31:51on agriculture is what are you doing
1:31:53about the the variation in weather. And
1:31:55as I you know at the beginning, it's the
1:31:58fact that there's variation in weather
1:31:59that's one of the reasons for the
1:32:01inability to to have contractual labor
1:32:05beyond the day.
1:32:06So Mark, from the Q&A, uh do you think
1:32:08CAST has anything to do I'm going to
1:32:10give you three of them. CAST has
1:32:11anything to do with the labor market
1:32:13frictions and would there be something
1:32:15you could incorporate in the model?
1:32:16Good. and then uh other labor
1:32:18opportunities in the villages and and
1:32:20Rea in particular, how might this
1:32:22interact with the model?
1:32:24Okay, so um one Yep. So one there there
1:32:29can be more done because as we know
1:32:32there are barriers to employment along
1:32:34cast lines, right? So the existence of
1:32:36these cast boundaries, okay, are going
1:32:38to raise the costs of hiring labor,
1:32:41right? So the the a farmer may be of one
1:32:44cast and the only available laborer may
1:32:46be of another cast and he will not hire
1:32:48that labor. So that's going to raise
1:32:50that's going to increase in a sense the
1:32:52transaction cost of labor. So there is a
1:32:53role for looking at the of cast in uh in
1:32:57in the transaction cost. I think that's
1:32:59a good idea. um what was the second
1:33:03question the second part of that
1:33:05about how the model interacts with other
1:33:07labor market opportunities in the
1:33:08villages and rea
1:33:11so um I'm doing a lot of work that's
1:33:14emphasizing the issue that was raised in
1:33:16the previous question about you know the
1:33:17the the fluctuations in weather and risk
1:33:20and and other things and rea is uh let's
1:33:24think about it we we don't think about
1:33:25it here so much um the and rea certainly
1:33:29enforced the 2009 200 4 period and rea
1:33:33for those who don't know is a program in
1:33:35which the government sets a statutory
1:33:38minimum wage and guarantees 100 days of
1:33:40employment. Now what Enriga is going to
1:33:43do in equilibrium it's going to raise
1:33:44the wage. uh it and in in work that
1:33:48Chris Eugri and I are doing it has
1:33:50actually complicated effects on the
1:33:52demand for labor and the supply of labor
1:33:54in terms of migration because by setting
1:33:56a floor in bad times right it raises the
1:33:59over it makes the cost of bad weather
1:34:01for farmers even higher than it was
1:34:03before which makes them more
1:34:05conservative and they take less risk. So
1:34:07the the effect of ENREA is very
1:34:09complicated but in the context of this I
1:34:12think in what ENA is is mainly going to
1:34:14do is make labor more expensive and more
1:34:17scarce.
1:34:18Okay. Thank you um very much Mark. Uh
1:34:21the front row can give a round of
1:34:23applause to him.
1:34:26Never works very well in Zoom.
1:34:28Well well not when the microphones are
1:34:30off.
1:34:32Very well done. Thank you. I haven't
1:34:34heard the sound of one hand clapping
1:34:36right in a long time.
1:34:39We had that.
1:34:40Um,
1:34:42so tomorrow we have uh Richard Roger
1:34:45will be presenting on misallocations. We
1:34:47hope to see you all uh tomorrow and
1:34:50thank you very much for participating.
1:34:52Bye-bye.

About this transcript

This page contains the full transcript of Supplemental lecture: Labour market frictions and development by CEPR & VideoVox Economics, generated from the public captions YouTube serves with the video. The transcript has 16,170 words across 2,348 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.