col-xs-12
col-sm-12
col-md-12
col-lg-12
col-xs-12
col-sm-12
col-md-12
col-lg-12
videoType
dynamic-media
videoDmUrl
https://delivery-p136806-e1377785.adobeaemcloud.com/adobe/assets/urn:aaid:aem:a4da26a0-24bf-4c68-8277-efcd10bc780d/play?assetname=ABCDE23_Session_2.mp4
keyFrameImage
videoDescription
This video features the following papers presented at the ABCDE 2023 Conference: Unequal Global Convergence, The Global Race for Talent: Brain Drain, Knowledge Transfer, and Growth, and The Long-Run Development Impacts of Agricultural Productivity Gains: Evidence from Irrigation Canals in India.
timestamp

00:02 OK,

00:02 um,

00:03 very,

00:04 uh,

00:04 warm welcome to this,

00:06 uh,

00:06 second session on,

00:08 uh,

00:08 growth and convergence.

00:11 Um,

00:12 we have,

00:12 uh,

00:13 3 papers and they're all terrific.

00:15 Uh,

00:16 I applaud both the authors and the,

00:18 uh,

00:18 organizers of this conference for selecting them.

00:21 They're really outstanding papers,

00:24 um,

00:24 and I very selfishly,

00:26 I teach a course on

00:28 economic growth at Georgetown and I'm.

00:30 Already looking at what I can cannibalize from these lovely papers,

00:33 so,

00:34 uh,

00:34 each,

00:35 uh,

00:35 presenter has,

00:36 uh,

00:37 20 minutes,

00:38 uh,

00:38 and I've been told to,

00:39 uh,

00:40 uh,

00:40 be rigorous in enforcement.

00:42 So we will start with,

00:44 uh,

00:44 Chamiro and,

00:45 uh,

00:46 and then,

00:46 uh,

00:47 move on from there.

00:48 Thank you.

00:51 Is there

00:56 So sorry.

01:07 All right,

01:07 uh,

01:08 uh,

01:08 thanks a lot to the organizers for including our paper in today's,

01:11 uh,

01:12 agenda.

01:12 Um,

01:13 so this is joint work with Elissa Genone who is at CRE in Barcelona and Can Cuno,

01:18 our PhD student at Penn State,

01:20 um.

01:21 Uh,

01:22 before I begin,

01:22 I must say that parts of this project are sort of still work in progress.

01:26 This has been a huge kind of data collection exercise,

01:28 and I'm going to point out to those pieces,

01:31 uh,

01:31 where we should be

01:32 slightly careful about interpreting the quantitative numbers.

01:35 But just to begin,

01:37 let's,

01:37 I mean,

01:38 what do we know about.

01:39 Economic growth in the last 30 years,

01:41 we know that there has been a lot of growth

01:44 amongst the poorer countries in the world and on an

01:46 average poorer countries have caught up with richer countries.

01:49 Um,

01:50 and then there's a little bit of evidence of what has happened within countries,

01:53 uh,

01:54 um,

01:54 mostly the US and,

01:56 and some in the UK.

01:58 What we really don't know is what has

02:00 happened to the evolution of spatial inequality,

02:03 uh,

02:03 uh,

02:04 within

02:05 countries over time in these last 30 years which has

02:08 seen a lot of economic growth in the world,

02:10 right?

02:11 And,

02:11 and with the evolution of spatial inequality in particular

02:16 what has been the role of um um structural.

02:19 Change

02:19 in shaping this,

02:21 this evolution.

02:22 So that's going to be the focus of our paper,

02:25 um,

02:25 uh,

02:25 a key contribution of this project is kind of building this new data set

02:30 on the GDP per capita at subnational units over time

02:35 so both kind of states and provinces of countries,

02:38 but also at the city level.

02:40 And then more importantly,

02:42 something that did not exist at all

02:44 is the labor force composition by sectors within

02:48 each region of various countries of the world,

02:51 right?

02:51 So using this data set,

02:52 we first kind of document three facts.

02:55 One is for the average country in the world in the last 30 years,

02:59 um,

03:00 convergence between regions within a country has been falling over time.

03:05 Um,

03:05 the second is that the,

03:06 the,

03:07 the,

03:08 this,

03:08 this stall in uh spatial convergence is related to growth of

03:13 services as the key driver of development in the last,

03:17 um,

03:17 um,

03:17 20 years or so.

03:19 Um,

03:19 and the,

03:20 and the last fact,

03:21 Is that

03:22 much more than happened during the manufacturing phase in the services phase,

03:27 um,

03:27 especially kind of professional and business services,

03:30 employment is getting very,

03:32 very concentrated in few regions of,

03:35 of,

03:35 of countries.

03:36 So,

03:38 Then we were,

03:38 so once we had the facts,

03:39 we were thinking like,

03:40 could,

03:41 could a sort of canonical structural change model with space built in it

03:45 explain these facts,

03:46 or do we need some other ingredients?

03:48 And in terms of the theory,

03:49 I'm going to show you a

03:50 very,

03:51 very simple kind of canonical model of structural

03:53 change with kind of space built into it,

03:56 um,

03:56 which is going to at least qualitatively match,

03:59 um,

03:59 um,

04:00 our facts,

04:01 um,

04:01 very well.

04:02 Um,

04:03 we're contributing sort of to two large literatures.

04:06 Uh,

04:06 uh,

04:06 uh,

04:07 one is just the literature on the evolution of spatial income disparities,

04:12 um,

04:12 and the second,

04:13 which is a more recent literature,

04:14 which takes kind of structural transformation and space,

04:17 uh,

04:18 uh,

04:18 more seriously,

04:19 um,

04:20 um.

04:20 Um,

04:20 on the first one,

04:22 we are just expanding the evidence to,

04:25 to many,

04:25 many more countries than existed before.

04:28 And in the second one,

04:29 we are kind of highlighting,

04:31 um,

04:31 the role of agglomeration and services,

04:34 uh,

04:34 uh,

04:34 which is going to be kind of a key feature of our,

04:37 of our theoretical framework.

04:39 So just to spend a little bit of time on,

04:41 on the data,

04:42 we are building on an existing data set by Giannoli,

04:45 LaPorta Schleifer,

04:47 uh,

04:47 uh,

04:47 but we have added many,

04:49 many more countries that,

04:50 that data set did not have.

04:52 We have also went on and corrected some of the data.

04:55 So for example,

04:56 the data for India in that data set

04:58 was missing for many,

04:59 many years.

05:00 Uh,

05:01 uh,

05:01 uh,

05:02 uh,

05:02 so we have updated it to the most recent year possible for our sample of countries.

05:07 Um,

05:08 from the Economist,

05:08 we also have Data on the GDP at the level of cities,

05:12 um,

05:12 in the world.

05:13 So sort of combining these two data sets,

05:15 we have like a balanced panel on GDP per capita by region within a country,

05:20 um,

05:20 for 34 countries between 1980 and 2017,

05:25 but we also have an unbalanced panel that starts much further back

05:28 and essentially the reason is that for the advanced countries you,

05:32 you,

05:32 you can go back further,

05:34 but for the developing countries you don't get data,

05:36 um,

05:36 going back far in time.

05:38 We have added to this.

05:39 This data on years of schooling by subnational region by year,

05:43 um,

05:43 in the world,

05:44 um,

05:44 and,

05:45 um,

05:45 the labor force composition by,

05:47 by kind of broad sectors.

05:49 We are,

05:49 we are,

05:50 so this summer we are making a,

05:51 a,

05:51 a new effort to kind of,

05:53 so right now we have the labor force composition by just broad sector agriculture,

05:57 manufacturing and services.

05:59 Now we are making an effort to break down services into further subsectors so

06:03 that we can get the tradable high skill part of the services in,

06:07 in much more,

06:08 um,

06:08 richer detail.

06:10 In terms of coverage,

06:11 well,

06:11 you can see that we're doing sort of a pretty good job in most continents,

06:16 um,

06:16 um,

06:16 except for Africa and since I'm standing at the World Bank,

06:20 um,

06:21 here is a call for help,

06:22 uh,

06:22 uh,

06:23 if any of you can help us add more countries in the African continent,

06:27 that would be terrific.

06:27 We have

06:28 tried a lot.

06:29 Like we write to like national statistical agencies in various countries.

06:33 We sometimes get replies and other times don't.

06:36 But,

06:36 but,

06:36 but Africa is one continent that is kind of,

06:39 so we have done some robustness with night lights,

06:41 etc.

06:42 but it would still be useful to have

06:45 actual employment and GDP figures at the subnational level.

06:49 So let me start with the facts and so

06:51 this is kind of your familiar convergence regression,

06:54 except now we are doing it at the sub,

06:55 uh,

06:56 uh,

06:56 uh,

06:56 within countries.

06:57 So you regress kind of

06:59 growth in GDP per capita,

07:01 uh,

07:01 uh,

07:02 for the next 10 years,

07:03 um,

07:04 for a given state or province

07:06 on the initial level of GDP per capita,

07:09 um,

07:09 at time T0.

07:11 So for a particular country.

07:13 Uh,

07:13 uh,

07:14 and as must be familiar to all of you,

07:15 if this beta is kind of less than 0,

07:18 then that points towards convergence,

07:19 and if it's positive,

07:20 then that means the

07:22 poorer regions are not catching up with the richer regions.

07:25 So we run this regression

07:27 for all the countries in our sample,

07:29 country by country.

07:30 So we have one beta C,

07:32 uh,

07:32 uh,

07:33 at a given point in time.

07:35 Uh,

07:35 uh,

07:36 uh,

07:36 for each country and then for the figure I'm going to show you,

07:39 we're going to average this across all countries,

07:42 right?

07:42 So,

07:42 what this,

07:43 so what this point is showing is

07:46 the average beta C for all the 34

07:49 countries in our sample and essentially showing that

07:52 in the 1980s,

07:54 there was a lot of within-country convergence on an average in the world.

07:58 And then over time there has been a sharp decline

08:01 in the within-country convergence rates and by the 2000s,

08:04 the within-country convergence process has basically come to a complete stall.

08:09 Um,

08:09 these are unweighted regressions.

08:10 If you do population weighted regressions,

08:13 the,

08:13 the,

08:13 the key fact that in the,

08:15 in the 2000s,

08:16 there is no within-country convergence.

08:18 It's still true except that the decline happened,

08:21 uh,

08:21 much faster.

08:23 Um,

08:23 this is true for like 56% of our countries,

08:26 but,

08:27 but this covers like for,

08:28 for most of the large countries by economic size or population,

08:33 um,

08:33 the within-country convergence rate is lower in,

08:36 in,

08:36 in the 2010s than it was in the 1980s.

08:41 Um,

08:41 um,

08:41 this is also true if you do the same convergence regressions,

08:44 not at the level of provinces or states,

08:47 but by cities.

08:48 Uh,

08:49 uh,

08:49 it's also true for conditional convergence,

08:52 not only,

08:52 um,

08:53 unconditional convergence.

08:55 It's true if you exclude India and China.

08:58 Um,

08:59 the

08:59 one sort of big challenge is accounting for regional price differences,

09:02 so we are collecting data we now have.

09:04 Have it for India,

09:05 China,

09:05 Britain,

09:06 uh,

09:06 and the US,

09:08 and for all these four countries,

09:09 our facts are quite robust.

09:12 We are trying to get it for more countries and hopefully we'll,

09:14 we'll be able to say more.

09:16 You can

09:17 cut this by

09:18 any dimension that you want,

09:19 but like the big fact that in the 2000s for most countries of the world,

09:24 there is

09:25 very little within country convergence kind of holds.

09:28 So,

09:29 for the main story,

09:30 let me just tell you,

09:31 tell this to you through the lens of like two countries.

09:34 So what I'm plotting here is GDP per capita of

09:36 India and China relative to the US over time,

09:39 right?

09:39 So what this graph is showing

09:41 is the enormous catch-up

09:43 that US,

09:44 uh,

09:44 India and China have had with the US.

09:47 But the key thing is that when you look

09:49 what is happening within these countries,

09:51 so now I'm plotting GDP per capita for

09:54 regions within these countries arranged by deciles,

09:57 right?

09:57 So

09:58 the orange line is the top decile regions in India and China,

10:01 and the,

10:02 the blue lines are the bottom decile,

10:04 uh,

10:05 regions within India and China.

10:06 They used to be very close to each other in the 1990s and over time,

10:10 um,

10:11 um,

10:11 the divergence has just increased,

10:13 right?

10:14 What we think is this is related to is,

10:16 is a growing concentration of tradable high-skilled services.

10:21 So,

10:21 if you just focus on the national employment

10:25 in sectors that is concentrated in the top regions

10:29 for manufacturing in India,

10:31 that has declined.

10:32 That's consistent with deindustrialization in India,

10:35 right?

10:35 The top regions are de-industrializing.

10:37 China has relatively been stagnant.

10:40 Um,

10:40 services overall,

10:42 India has been stagnant.

10:43 China has increased a little bit,

10:45 but if you look at professional and business services,

10:47 which is this high skill services,

10:49 uh,

10:49 India has gone from something like 39% to 47%.

10:53 So 47% of the national high skill services in India.

10:57 In terms of employment is just concentrated,

11:00 uh,

11:01 uh,

11:01 uh,

11:01 in these five states

11:03 and you see sort of similar patterns,

11:05 um,

11:05 um,

11:06 kind of in China.

11:07 So if I,

11:07 if I look,

11:08 if I take this to the entire sample

11:10 and if I regress,

11:12 uh,

11:12 the internal convergence rate on the services

11:15 employment shares of course with development,

11:17 the internal convergence rate is falling,

11:20 but what it is falling,

11:21 uh,

11:22 uh,

11:22 starkly with is the growth of services within these countries.

11:25 So ideally once we have the full.

11:27 Data set with the professional business services.

11:29 I want to replace the x axis

11:31 not by broad services but just by

11:34 uh professional and business services which we have done for 22

11:37 countries where we have the data and the thing holds,

11:40 uh,

11:40 uh,

11:40 but we want to sort of um do it for the full,

11:42 full sample.

11:44 Here's the same thing in regression so the Y variable is the

11:49 within country convergence rates.

11:51 All regressions have country fixed effects,

11:53 so we are tracking the evolution of within

11:56 country convergence rate over time within a country.

12:00 It's of course rela related to development.

12:02 So as development progresses,

12:04 as employment moves out of agriculture,

12:07 the within-country convergence rate falls,

12:10 but

12:11 it's not so much related to manufacturing than it is,

12:15 um,

12:15 um,

12:15 to the growth of services,

12:17 um,

12:17 within these,

12:18 within these countries.

12:19 Now,

12:19 one valid question to ask is maybe with development.

12:23 There is no regional,

12:25 there is no sizable regional inequality left and so

12:28 this might just be reflecting kind of the solo lesson

12:31 that once your regional inequality is down a threshold,

12:34 uh,

12:34 uh,

12:35 there is,

12:35 there are no gaps left to close,

12:39 uh,

12:39 uh,

12:39 that is not fully true either.

12:41 So now on the Y axis we.

12:43 Have just the coefficient of variation in GDP per capita between regions,

12:48 um,

12:48 within a country.

12:50 Of course,

12:50 with the development,

12:51 this

12:52 sort of fell sharply,

12:54 but then it kind of,

12:55 when services hit kind of the 30% mark,

12:58 uh,

12:58 uh,

12:59 this kind of stabilized at around 18% or 17%.

13:02 And since then,

13:04 um,

13:04 these gaps between GDP per capita between regions haven't,

13:08 haven't been closing.

13:10 Um,

13:12 So here is where we should be a little bit careful.

13:14 So it turns out that even if you look at the EU,

13:17 uh,

13:17 they don't have a,

13:19 they don't have

13:20 good data on,

13:21 uh,

13:21 labor force composition by sectors before 2000.

13:25 It depends on the country.

13:26 For example,

13:26 Italy,

13:27 you don't get this data,

13:28 which is why.

13:30 I don't want you to focus on anything before 2000s,

13:32 but

13:33 after 2000,

13:34 so what this,

13:35 uh,

13:35 chart is plotting is the,

13:37 um,

13:38 employment in,

13:39 in each sector in the top region versus the bottom region.

13:43 So lesson number one,

13:45 services is much more concentrated than manufacturing or agriculture,

13:50 right?

13:51 The second is

13:52 the growth in concentration in services,

13:55 which is the slope of this line.

13:57 Um,

13:57 is much more for services than manufacturing or agriculture.

14:01 So,

14:02 not only is it more agglomerated in,

14:04 in,

14:04 in services,

14:05 the rate of increase,

14:07 um,

14:07 in concentration,

14:09 um,

14:09 is,

14:09 is,

14:10 is also much higher.

14:12 Now,

14:12 we can't do,

14:12 we,

14:13 in terms of testing what other factors might be involved,

14:16 we can't do much because you don't get a lot of

14:18 sort of time varying controls by region within a country,

14:22 but we can look at cross-country relationships.

14:24 So if you put a time fixed effect

14:26 and compare

14:27 two different countries,

14:29 one which has a lot of internal convergence and the other that does not,

14:33 and see what factors would explain this,

14:35 and I'm only showing some factors here.

14:38 So this is like a kitchen sink regression.

14:40 It turns out that

14:42 it's services,

14:43 employment share and the growth of productivity in services

14:47 that seems to be the leading predictor of why certain countries have high

14:51 internal convergence,

14:53 um,

14:53 than others and in particular,

14:55 uh,

14:55 uh,

14:56 growth of services associated with less,

14:58 um,

14:58 internal,

14:59 internal convergence.

15:01 So,

15:03 I'm going to sort of

15:05 To see,

15:06 to see whether this,

15:06 all these facts can be rationalized by a simple sort of structural change model.

15:11 I'm gonna explain this in words.

15:12 The model is not complicated at all,

15:14 right?

15:14 So you,

15:14 you take the most canonical structural change model

15:17 where structural change happens because of non-hommothetic preferences,

15:21 right?

15:21 And you put it in a standard model of economic geography.

15:25 So there are multiple regions,

15:26 people choose which region they want to locate in.

15:29 And they produce stuff and they,

15:31 and,

15:31 and they consume.

15:32 The,

15:33 the only departure from standard models is going

15:36 to be so in standard structural change models,

15:38 you have,

15:38 so this is the sectoral

15:40 growth rate of TFP

15:42 in agriculture and manufacturing.

15:45 Uh,

15:45 uh,

15:45 uh,

15:45 it's all exogenous and,

15:47 and,

15:48 uh,

15:48 uh,

15:49 uh.

15:49 Uh,

15:49 uh,

15:49 as a standard in the structural change models,

15:52 what we have done for services is we have put in an additional kick

15:56 that comes from agglomeration and services.

15:58 So,

15:59 growth rate of services productivity is,

16:01 is due to an exogenous factor,

16:04 but also because there are more people located in a particular region,

16:09 um,

16:09 um,

16:10 in services.

16:11 So

16:12 the calibration is preliminary and,

16:13 and pretty standard.

16:14 So in the interest of time,

16:15 I'm not going to focus much,

16:17 but just show you this.

16:19 Um,

16:20 um,

16:20 the,

16:20 the red dots with the dashed lines is what

16:24 the,

16:24 the,

16:25 the,

16:25 the data says about the evolution of within-country convergence.

16:29 That very simple model just matched to the

16:32 initial concentration of employment shares in different regions

16:37 can pretty much predict

16:38 the decline in the uh,

16:40 the decline in the within country convergence rate over time,

16:44 um,

16:44 um,

16:44 at least qualitatively.

16:46 And we can,

16:47 of course,

16:47 use the model to see what role this agglomeration force is playing,

16:51 right?

16:51 Right.

16:52 So,

16:52 what role is,

16:53 is,

16:54 is,

16:54 is,

16:54 is this factor playing here?

16:57 So in the model,

16:58 if you do this decomposition where you choose like some agglomeration force

17:02 that is calibrated to the data versus setting it to zero,

17:06 it turns out that you still get some decline in beta convergence,

17:10 even if there was

17:11 no agglomeration.

17:13 But then the effect just multiplies,

17:16 uh,

17:16 uh,

17:16 if there is agglomeration in services.

17:18 So if I,

17:19 if I excluded

17:20 the agglomeration factor from the model,

17:22 there are two pieces of the data that I won't be able to match.

17:26 One is

17:27 the speed of this decline,

17:28 like it would be much slower without the agglomeration kick,

17:32 and the second is increase in concentration,

17:35 uh,

17:35 uh,

17:35 of services,

17:36 employment in,

17:37 in,

17:37 in,

17:38 in particular regions.

17:40 So,

17:41 kind of the big picture message of our,

17:43 of our,

17:43 of our research is this.

17:44 So,

17:44 one is,

17:45 of course,

17:45 like putting out the data set,

17:46 which,

17:46 of course,

17:47 I'm assuming would be useful for many of you.

17:50 Uh,

17:50 but also to say this,

17:51 I mean,

17:51 the World Bank had this fantastic report showing that

17:54 future growth is going to be driven by services.

17:57 And if that is true,

17:58 and,

17:58 and,

17:58 uh,

17:59 and,

17:59 and assuming that it's all sort of professional tradable high skill services,

18:04 then spatial inequality is going to be a free gift of it,

18:07 um,

18:07 um,

18:08 that we should be,

18:09 um,

18:09 kind of cognizant about.

18:11 So thank you.

18:13 Amazing.

18:17 Thank you so much,

18:18 Micro,

18:18 and you,

18:19 uh,

18:19 and you,

18:20 uh,

18:20 are great on time.

18:21 So,

18:21 uh,

18:22 let's move right on to,

18:23 uh,

18:24 Marta for,

18:25 uh,

18:25 the second paper.

18:32 Great.

18:32 So thank you so much for,

18:34 to the organizers for inviting me to present this project

18:37 that's called the Global Race for Talent,

18:39 Brain Drain,

18:40 Knowledge Transfer and Growth.

18:43 And in this paper,

18:44 I studied the effects

18:46 of migration on innovation and growth

18:49 and the starting point is that

18:51 uh inventors are very highly mobile individuals.

18:55 22

19:05 Great.

19:06 So to illustrate this,

19:07 I want to start with this example

19:09 where Jean Calvignac here is a prolific French inventor

19:13 that appears in my data from the European Patent Office.

19:17 And this is an example of a real patent that he filed in 1991

19:22 and from this we learned that at this time

19:24 Calvignac lived in the city of Lagos in France.

19:28 But also most of his collaborators at this time,

19:31 that is the other co-inventors that contributed to

19:34 this technology were also based in France.

19:38 Then in 1998,

19:39 Calvin Jaak moved to the US.

19:42 And we know this thanks to his later patents such as this one from 2004,

19:47 which shows that first at this time he lived in North Carolina.

19:51 Second,

19:52 most of his collaborators at this time were also based in the US.

19:56 But most interestingly,

19:57 even after moving to the US,

19:59 Kaviac continued to work with some inventors based in France,

20:04 and this is not an isolated example.

20:06 In fact,

20:07 there is a systematic flow of inventors across countries.

20:11 To give a sense of this using patent data between 2001 and 2010,

20:16 I observed that about 6%

20:18 of European national inventors live in the US in any given year on average,

20:23 but only about 0.4% of American inventors live

20:27 in Europe in any given year on average,

20:30 and the difference between these two numbers,

20:31 the net migration is what I refer to as the brain drain.

20:35 It has been a cause of concern

20:36 and policy interventions both on the European and the American side.

20:41 So in this project,

20:41 I first ask

20:43 how does inventors migration affect migrants' productivity,

20:46 their collaboration networks,

20:48 and their knowledge spillovers on local inventors.

20:51 And second,

20:52 I study the positive implications of some real world policies

20:55 that have been implemented to manage its migration flows.

20:58 So I asked what is the role of tax and migration policy

21:01 in shaping migration flows,

21:02 the innovative capacity of the economy,

21:05 productivity and output.

21:08 And in this paper,

21:09 I answered these questions in 3 steps.

21:11 The first step is to bring a theoretical framework to

21:14 a literature that had so far been mostly empirical.

21:18 So I built a two country dynamic general equilibrium innovation based growth model

21:22 that has two important new features.

21:25 One is that inventors in this model are allowed to move

21:28 across countries and to return to their country of origin.

21:31 And second,

21:32 inventors here accumulate human capital by learning from others,

21:36 but this learning happens inside endogenous interaction networks that

21:40 produce knowledge spillovers both within and across countries.

21:44 Now the key takeaway from this model

21:46 is that migration has ambiguous effects on innovation and

21:50 output in the short and in the long run,

21:52 and this is because it produces several forces that have opposite signs.

21:57 And the two main forces that come out of this model are on the one side,

22:00 the effect of the brain drain,

22:02 as it is partly offset on the other side

22:05 by the positive benefits for knowledge spillovers.

22:10 Now,

22:10 given this ambiguous theoretical result,

22:12 the next question is how can we quantify these different forces.

22:16 And this takes me to the 2nd part of the project where

22:19 I build a micro-level data set of migrant inventors between Europe

22:22 and the US using data from the European Patent Office.

22:26 Now here one of the challenges is that

22:29 there is very scarce availability of data

22:31 that consistently track individuals across countries,

22:34 which is why here I focus on Europe and the US

22:37 but more generally the theoretical insight from this model

22:40 apply more broadly not just to individuals that

22:42 invent technologies but also individuals that adopt technologies

22:46 and so this model could apply more generally not

22:48 just to advanced economies but also to developing countries.

22:52 Now,

22:53 the key findings from my data are that first,

22:56 after migration,

22:57 migrants,

22:57 both European and Americans

22:59 tend to increase their patent applications by 42% per year on average.

23:05 Next,

23:05 what I think is even more interesting in this data is to look at

23:09 what happens to the collaboration network of these migrants.

23:12 And here I find that after migration,

23:14 migrants actually continue to work with inventors in their country of origin,

23:19 but less frequently

23:20 as they work more often with investors and

23:22 desti inventors at destination than other migrants.

23:26 So then the last question is what happens to these inventors in the country of origin

23:31 after somebody that they work with moves away.

23:34 And here what I find in the last point is that the local inventors at Origin

23:38 also increase their patent applications by about 15% per year

23:42 after their collaborator moves away.

23:45 And this is what I will interpret through the lens of

23:47 my model as the effects of the knowledge spillovers channeled by migrants

23:51 on the productivity of the local inventors at origin.

23:55 Finally,

23:56 the last step in this paper is to bring together the

23:58 theory and the data by calibrating the parameters of this model

24:02 to match his empirical result

24:03 and finally to study counterfactual policy

24:06 exercises that resemble real world policies implemented

24:09 either by European countries or the US

24:11 to address these migration flows.

24:15 So let me start with a brief overview of this model.

24:20 On the human capital side,

24:22 here,

24:22 the population is exogenously split between

24:24 inventors that produce technologies or ideas

24:27 and workers that produce the final good.

24:30 On the production side,

24:32 these technologies or ideas are used to improve the quality of intermediate goods

24:36 that are then sold to the final good aggregator.

24:39 So that overall innovation here drives

24:42 aggregate productivity growth through the improvement in intermediate quality,

24:46 as is common in the innovation-based growth models.

24:50 Now,

24:50 the novelties that I bring to this framework are the following.

24:54 I consider a two country version of this economy with two countries,

24:57 A and B.

24:59 And the goal is to describe the endogenous evolution

25:01 of TFP ABC in each of the two economies.

25:05 At the heart of the process of growth are inventors.

25:09 And inventors here produce innovation,

25:12 but on top of that,

25:12 they are allowed to move across countries and to return to their country of origin.

25:17 And second,

25:18 they meet others in the economy and they learn from them.

25:21 And this interaction and learning process generates

25:23 knowledge spillovers both within and across countries.

25:28 Finally,

25:28 in this model,

25:29 I also allow for some additional exogenous technology diffusion across countries

25:34 to capture other sources of ideas flows such as foreign direct investment

25:38 or trade in intermediate goods.

25:40 And finally,

25:41 these countries also trade the final good

25:43 and share a common exogenous interest rate.

25:46 So let me give you some details on this process of migration for these inventors.

25:52 Inventors are born in a given country in this model,

25:56 and they have heterogeneity across two key dimensions.

25:59 The first,

26:00 as Ufuk mentioned in his presentation,

26:02 is that

26:02 inventors have heterogeneous talent that I call Z.

26:06 But in this model inventors also have

26:08 heterogeneous idiosyncratic opportunities in different countries,

26:12 and this is what I call their foreign productivity differential epsilon.

26:16 And together these two variables determine how productive an inventory is.

26:20 So an inventor in every period produces a certain bundle of ideas or technology

26:25 Q that is equal to their talent when they live in their home country.

26:29 But this is equal to the talency plus the

26:31 foreign productivity differential epsilon if the inventor moves abroad.

26:36 Now this epsilon here is meant to capture idiosyncratic reason why a

26:40 certain inventor could be more productive in one place versus another,

26:43 such as

26:44 their individual field expertise that could be a good match

26:47 with a firm in the other country in this period.

26:50 So the important point here is that you can see that

26:52 inventors will choose to move when they have a high draw

26:55 for these idiosyncratic opportunities in the other location.

26:59 But also these opportunities epsilon are allowed to move over time

27:03 and if they become low enough,

27:05 the inventor will choose to return to the country of origin.

27:08 So here migration decisions are endogenous

27:10 to the idiosyncratic

27:12 productivity opportunities of these inventors in different countries.

27:16 The second important piece here to the right of these

27:19 slides is that their return to innovation for these inventors

27:23 are also proportional to the aggregate TFP ABC in each location.

27:27 And this means that a 2nd reason for inventors to move here is also

27:31 that they may want to go to the country with the highest TFP,

27:33 the technology frontier,

27:35 where they earn more from their innovations.

27:40 Finally,

27:41 the third point is that inventor silent Z also

27:44 evolves over time thanks to this learning process.

27:48 And this works in a way such that

27:50 with some probability lambda an inventor meets another individual

27:54 in the economy and the result of this

27:56 meeting is that both individuals become more productive.

28:00 However,

28:01 a novelty here is that the probability of meeting

28:03 different individuals is different for locals and for immigrants,

28:07 and this is due to matter frictions.

28:09 And this liter frictions means that

28:11 a certain inventor that is born in a certain country C and lives in some

28:17 other country D has different probabilities of

28:19 meaning inventors that live in different places and

28:22 come from different countries.

28:24 The important point of this structure is that first of all,

28:27 individuals have a 3rd reason to move because

28:29 when they move they can change their interaction network

28:32 and look for better learning opportunities.

28:35 But also the other important point is that

28:37 when individuals move with some probability these migrants

28:40 can still interact with inventors in the country of origin

28:44 and with this meaning they transfer knowledge that

28:46 makes the locals of origin more productive.

28:48 So this is the way in which the model captures the idea that

28:52 the

28:53 knowledge spillovers channeled by migrants make the

28:55 local inventors at origin more productive.

28:59 So as a summary of the effects of migration in this model,

29:02 if we think about the migration flow from country A to country B,

29:05 which is the frontier.

29:08 What we find is that there are now more inventors in B

29:10 who are also more productive when they move because they're positively selected,

29:14 although some of these migrants may crowd out

29:17 some of the local inventors,

29:19 and on,

29:19 on the other side in Country A,

29:21 there are fewer inventors because of the brain drain,

29:24 but the migrants transfer knowledge to the locals at origin,

29:26 making them more productive.

29:28 And on top of that,

29:29 the Country A benefits from higher innovation through the

29:32 front at the frontier also through exogenous technology diffusion.

29:36 Which is why overall the net effect of migration in this economy is ambiguous.

29:41 And this motivates the need for an

29:43 empirical and quantitative analysis in this model.

29:47 So to do that I focus on an application of this model where location A is Europe

29:52 and location B is the US

29:53 and again this is due to data availability,

29:56 but what could think of a different application and calibration of this

29:59 model where one country is the US and the other country is,

30:01 for example,

30:02 a developing economy.

30:04 Now,

30:04 I use my patent data to calibrate the parameters of this model.

30:09 And to do that,

30:09 I first show this fact that represents the

30:12 collaboration network of local and migrant inventors.

30:16 So what I'm doing here is to

30:18 group the inventors in my data into 4 categories that are

30:21 locals and migrants of European and American origin.

30:24 And here I plot on the y axis the fraction of collaborators of these inventors

30:29 that belong to each of the four categories.

30:32 So for example,

30:33 the first bar on the left

30:35 here shows that when I look at European local inventors,

30:39 in orange,

30:39 this means that they mostly work with other local European inventors.

30:43 However,

30:44 the second bar looking at the European inventors that move into the US

30:48 shows in green that

30:50 these migrants are much more likely to work with American inventors

30:54 as well as other migrants,

30:56 and the similar pattern is true for American inventors.

31:00 Now,

31:00 this is helpful to think about the network frictions in my model,

31:03 and it says that inventors are more likely to meet

31:06 and interact with other inventors that either live in the same country

31:10 or also they have their same nationality.

31:16 Now the next step here is to illustrate the evolution of patenting activity

31:20 both for the migrant inventors as well as for their local collaborators.

31:25 So to do this,

31:25 the strategy here is to match every migrant in my sample with one placebo or

31:31 control inventor who is not a migrant and not a direct collaborator of a migrant,

31:35 but this control inventor comes from the same country of origin,

31:39 appears for the same first year in the sample,

31:41 and also has the same cumulative patent stock

31:44 by the time of migration as the migrant as a proxy of productivity.

31:49 And finally,

31:50 as the last step,

31:51 I collecting the data,

31:52 the full network of collaborators in the country of origin,

31:55 both for the real migrants

31:57 and for the control group.

32:00 And here is what I find.

32:02 In this first figure I'm plotting here on the X axis,

32:05 times 0 is the time when the real migrants move away.

32:08 And the Y axis is showing the mean of patent applications per year.

32:13 The blue line is the dependent applications of the real migrants,

32:16 and the red line is for the control group.

32:19 So what we see here is that the key point is that after migration,

32:22 migrants tend to increase their patent

32:25 applications relative to the control group.

32:28 The next question is what happens then to the local

32:31 collaborators of these migrants in the country of origin.

32:35 That's what I show in the second figure where the

32:38 X axis still shows the time 0,

32:40 the time when the real migrants move away,

32:42 but now in the blue line I'm tracking the patent applications of

32:46 the collaborators of the real migrants in the country of origin,

32:49 whereas the red line are the patent applications

32:51 of the collaborators of the control group.

32:55 And the key message here is that also

32:56 the local collaborators of the migrants at origin

33:00 see an increase in their patent applications after the migrant moves away

33:05 relative to the patent applications of the collaborators of the control group.

33:10 Now this,

33:11 this fact is very important to calibrate the parameters of my model.

33:14 Why is that the case?

33:16 That is because

33:17 I can replicate these two event studies

33:20 from a simulated sample of inventors in my model.

33:23 And I calibrate my parameter in such a way that the model simulated sample in red

33:28 produces the same magnitude as the data event studies

33:31 that we've just seen that here is in green.

33:34 And the key figure here is the one for the local inventors on the right

33:38 because in the model,

33:39 the reason why the local collaborators of the migrants at origin

33:43 see an increase in their patent applications

33:45 is thanks to the knowledge spillovers

33:47 that the migrants abroad channel

33:49 to these inventors back in the country of origin.

33:54 Now in the last few minutes,

33:55 I want to show some policy exercises.

33:58 Today I'm going to talk about a tax on inventors' profits

34:02 that is allowed to be heterogeneous across countries

34:05 and yet in my my calibration is initially higher in Europe and lower in the US.

34:10 And in the paper I talk more also about migration policy in the US.

34:15 Now the policy exercise that I do here is to study a

34:19 a tax cut in the European Union

34:22 for American immigrants and European returned migrants,

34:25 and this is motivated by the fact that several European countries

34:29 have implemented policies of this flavor starting from the Netherlands,

34:32 Denmarks,

34:33 more recently also Italy and other countries

34:36 with the goal of trying to revert this brain drain.

34:40 In my model this means solving the transitional dynamics from

34:43 the initial equilibrium where the tax rate is higher in Europe

34:47 to the new equilibrium where the tax rate in

34:49 Europe is lower for Americans and returned migrants.

34:53 Now in the next few figures,

34:55 I'm ploying the results of this policy,

34:57 so the X axis is always going to be

34:58 the number of years since the policy implementation.

35:02 This first figure here on the y axis is plotting

35:04 the stock of migrants as a fraction of domestic inventors.

35:08 So what we see here in terms of allocation of talent

35:11 is that

35:12 the green line shows that the stock of

35:14 American migrants in Europe increases and this is

35:17 because Americans know that now if they go to Europe they'll have a lower tax rate,

35:22 but at the same time the stock of European migrants in the US,

35:25 the red line

35:26 is decreasing,

35:27 and this is because European migrants know now that if they

35:31 return to Europe they will have this lower tax rate.

35:34 And the dashed line here is showing that

35:36 uh the magnitude of this policy is such that

35:39 the level of net migration that is the dash line

35:42 is going down to zero in the long run.

35:45 Now,

35:45 what is the,

35:46 the effect of this reallocation of talent on innovation?

35:50 That's what we see in this figure,

35:51 where now the y axis is an innovation plotted as a percentage of GDP.

35:56 And what we see is that this reallocation of inventors

36:00 towards Europe implies that there is an increasing innovation in Europe

36:03 and a decline in innovation in the US.

36:06 Now the last question is,

36:08 what is the impact of this change in innovation on output in the two economies?

36:14 So in this last figure I'm plotting outputs in Europe and the US

36:18 relative to the counterfactual equilibrium

36:20 where there is no policy change.

36:23 So what we see here in the green line is that output in the US is declining,

36:27 and this is because innovation in the US is declining and

36:31 the US is the innovation and technology frontier to start with.

36:35 But more interestingly,

36:36 here in the red line we see that output in Europe

36:39 increases for a few decades,

36:41 but eventually,

36:41 and this is thanks to higher innovation in Europe,

36:44 but eventually at some point output in Europe starts to decline.

36:48 And to understand why we need to look at the table of the right,

36:51 where I provided the composition of the different

36:54 forces that I discussed in the theoretical section,

36:57 quantify the two different time horizons.

36:59 And the one that I want to highlight here is given by

37:02 the knowledge spillovers that are at the center of this model.

37:06 And what's happening here is that this policy

37:08 by reducing net migration flows to zero,

37:11 it also reduces knowledge spillovers across countries

37:14 and eventually this has a sizable negative effect on output in Europe,

37:18 but this is also a force that takes a long time to become

37:21 effective because it operates on the distribution of human capital in the economy

37:25 that is a slow moving object.

37:29 So

37:30 to summarize,

37:31 this paper presents a framework to study migration decision,

37:35 knowledge networks and innovation

37:36 and some new empirical evidence to think about

37:39 the magnitude of these different forces.

37:43 More importantly to me,

37:44 the key takeaway from these policy exercises are the following.

37:48 First is that

37:49 um policies that try to change migration flows and reduce the brain drain.

37:55 Actually generate several forces on the economy that have opposite sides

38:00 and so they even

38:02 they may even end up in backfiring and reducing output in the long run.

38:06 On the other hand,

38:07 I also highlight that knowledge spillovers are very important.

38:10 They have a sizable quantitative effect on output.

38:14 So perhaps if anything rather than

38:16 focusing on policies that affect migration flows,

38:19 one could think about policies that

38:21 foster these knowledge spillovers across countries.

38:24 And finally,

38:24 I think we still have some open questions,

38:27 especially relevant for this audience.

38:29 Due to data availability,

38:31 this,

38:31 this paper focuses on Europe and the US

38:34 and we need to have more research to understand,

38:36 for example,

38:37 how sizable these knowledge spillovers are

38:39 for other countries,

38:40 especially the developing countries.

38:42 Thank you.

38:47 OK,

38:47 great.

38:48 Very,

38:48 very nicely done and,

38:49 and right on time,

38:50 uh,

38:51 as well.

38:52 Uh,

38:52 let's now have the 3rd,

38:54 uh,

38:54 paper.

38:55 Alison,

38:55 uh,

38:55 20 minutes.

38:59 Um,

38:59 great.

39:00 Thank you for having,

39:01 um,

39:01 our paper as part of the session.

39:03 Um,

39:04 this paper is with Sam Asher,

39:05 Paul Novisad,

39:06 and Doug Galen,

39:07 who you'll hear more from later

39:09 on the long-run development impacts of canal irrigation in India.

39:13 So the motivating question of this paper is

39:15 one of the oldest in development economics.

39:17 How do gains to agricultural productivity

39:20 affect broader growth and structural transformation?

39:23 This question naturally arises from the observation that um

39:26 Lewis put in 1954

39:28 that economies in which agriculture is stagnant

39:31 do not show uh industrial development.

39:33 So there's been a lot of theory,

39:35 um,

39:36 and exploration of exactly why and how,

39:38 uh,

39:38 gains in agricultural productivity are translated into,

39:42 uh,

39:42 into industrial development.

39:44 Um,

39:45 and more recently,

39:46 there's been an empirical literature trying to quantify this effect,

39:50 uh,

39:50 much of it looking at the green revolution

39:52 and the introduction of high yielding seed varieties,

39:55 um,

39:55 through the 20th century.

39:57 Um,

39:57 and in both the theory and in the empirics,

39:59 um,

40:00 the relationship is,

40:01 is complicated,

40:02 and we want to understand more about

40:04 exactly when,

40:05 um,

40:06 technical advancement in agriculture

40:08 translates into growth in the non-farm sector as opposed

40:11 to crowding out growth in the non-farm sector.

40:14 So in this paper,

40:14 we add what we think is a unique

40:16 angle by looking at the generally equilibrium effects

40:19 of,

40:20 uh,

40:20 a massive agricultural productivity shock at high spatial resolution.

40:25 So that allows us to look at exactly how and where,

40:28 um,

40:28 structural change might take place in the long run.

40:32 So we're looking at the canal network in India,

40:34 which is very large.

40:35 It runs for 300,000 kilometers and serves,

40:38 uh,

40:38 about 1/5 of all villages in India.

40:41 What's interesting about canals as a,

40:43 as a shock to agricultural productivity

40:45 is that unlike other technical advancements

40:48 like seeds or fertilizers or mechanization.

40:51 Um,

40:51 canals are fixed in space.

40:53 So when you build a canal,

40:54 you introduce a long-run differential,

40:56 um,

40:56 in productivity for places that get access to canal water versus those that don't.

41:01 Um,

41:01 they don't diffuse across space through time like many other technologies do.

41:05 Um,

41:06 these canals were mostly built,

41:07 uh,

41:08 through the 19th and 20th century,

41:09 and the majority of the canals in our data set were completed over 40 years ago.

41:13 So we're going to be looking at outcomes in,

41:16 uh,

41:16 the long run in the modern cross section from this,

41:18 uh,

41:18 canal network

41:20 that was built to deliver water primarily during the dry season,

41:23 the rubby growing season.

41:25 Um,

41:25 to allow farmers to have,

41:27 uh,

41:28 better,

41:28 more production in a year.

41:30 So you're getting more production per hectare per farmer,

41:32 and they're more resilient to the annual weather seasonality.

41:37 Um,

41:38 so as I mentioned,

41:38 the canals,

41:39 uh,

41:40 are an old technology.

41:41 They were invested heavily in by the British Raj and

41:43 then also by the early independent government in India.

41:46 They were actually the dominant source of irrigation before

41:48 the rise of groundwater in the last 40 years,

41:51 and they're still a very significant,

41:52 um,

41:53 source of irrigation for Indian farmers.

41:56 Just so you have a picture of what these canals look like.

41:58 These are not the Suez Canal or the Erie Canal.

42:00 They're not built for massive transport.

42:02 You can see here there's a boy standing in the middle of this canal,

42:05 so it's fairly shallow,

42:06 and the water level actually fluctuates such that sometimes these canals are

42:10 empty,

42:11 um,

42:11 and the water level will go up and down as,

42:13 as the,

42:14 as the season changes.

42:16 So don't picture big transport canals.

42:18 These are really built for,

42:19 uh,

42:19 irrigation purposes for agriculture.

42:22 So our goal in this paper is to estimate the impact of these,

42:26 uh,

42:26 getting access to irrigation from these canals in India.

42:29 And we're going to do that in 3 ways,

42:31 estimating 3 different local average treatment effects.

42:34 The first is the direct treatment.

42:35 So,

42:35 in villages that get access to canal water versus those that don't,

42:39 what are the agricultural and

42:41 non-agricultural outcomes we see?

42:43 Then we'll look at spillovers from those directly treated,

42:46 um,

42:46 areas into other villages that are proximal.

42:50 And lastly,

42:50 we'll look at the timing of canal construction

42:53 to understand regional urbanization.

42:55 So our gains from canal irrigation being invested in the region,

42:59 um,

42:59 into urban centers.

43:02 To give you an idea of where we're going,

43:03 uh,

43:04 to cut to the chase,

43:05 we're going to show that

43:06 this increased access to canal irrigation,

43:09 we can actually measure an increase

43:11 in canal irrigated agricultural land that

43:13 leads to increased agricultural productivity,

43:16 that's been sustained,

43:17 um,

43:17 into the modern day.

43:19 And that increase in agricultural productivity leads to an increase in migration.

43:24 So,

43:24 the population density in irrigated areas drastically increases,

43:28 but we don't see any long-run,

43:30 um,

43:31 uh,

43:31 gains in consumption.

43:33 So,

43:33 for the general population,

43:35 places that are irrigated versus not irrigated,

43:37 uh,

43:38 their consumption are about equal,

43:40 except for landowners.

43:41 So landowners,

43:42 um,

43:42 who get their land irrigated have sustained gains,

43:45 um,

43:45 in the modern data.

43:47 And we see no effects on rural industrialization,

43:50 but we do see regional urban growth.

43:52 So basically,

43:52 in the long run,

43:53 um,

43:54 the story we're going to try to show here is that productivity

43:56 gains introduced by these canals are equilibrated by labor flows across space,

44:01 and any structural change they,

44:03 uh,

44:03 create is concentrated as,

44:05 as urban growth.

44:07 So to walk you through our empirical strategy in more detail,

44:09 just consider two,

44:11 villages that are similar,

44:12 except for their elevation.

44:13 So we have one that's slightly higher,

44:14 maybe 10 m higher in elevation than the other.

44:17 When the canal is constructed through the area,

44:19 because canals require gravity to deliver water to their endpoint,

44:23 the canal will only be able to feed the village that lies below the canal.

44:27 So this is going to introduce a discontinuity,

44:29 um,

44:30 that allows us to do a regression discontinuity design based on

44:32 the elevation of the villages relative to the canal placement.

44:36 So in our first stage,

44:38 we're going to see that access to this canal water is going to give,

44:40 um,

44:41 these irrigated villages higher agricultural productivity,

44:44 and we'll test for any other non-farm outcomes.

44:48 We'll then look at spillovers into those,

44:50 uh,

44:50 villages that lie above the canal that weren't directly treated,

44:53 but are proximal to this boost in agricultural productivity

44:56 by comparing to a third set of distant villages that are much farther from the canal.

45:01 And then lastly,

45:02 we'll look at,

45:03 uh,

45:03 regional urbanization to see if these gains

45:05 from productivity in the treated villages,

45:07 uh,

45:08 result in growth of urban population.

45:11 The data we're going to use for this,

45:12 um,

45:13 I won't go into detail here,

45:14 but it's all administrative,

45:15 publicly available data from India between 2011 and 2013.

45:20 The one thing I'll highlight is that we don't have,

45:22 um,

45:22 a,

45:23 a direct measure of crop yields at the village level.

45:25 So we're going to use a satellite proxy,

45:27 which is the enhanced vegetation index.

45:30 So that essentially gives us,

45:31 uh,

45:32 a measure of the green up in every village during every season.

45:35 So that's what we're going to use for our productivity measure.

45:38 If you want to know any more about this data,

45:40 you can check out the Shrug open data product that's,

45:43 um,

45:43 made available through Development Data Lab.

45:45 There was just a second version of this

45:47 data platform released last week,

45:48 which was very exciting.

45:50 Um,

45:50 there's a lot of documentation there and happy to talk about it later.

45:54 All right,

45:54 so for our first,

45:55 um,

45:56 uh,

45:56 identification strategy,

45:57 we have a standard regression discontinuity,

45:59 uh,

45:59 specification where our running variable

46:02 is elevation of villages relative to the nearest canal.

46:06 So what I'm showing here on the,

46:07 in the map,

46:08 you'll see each polygon is a village,

46:10 and the villages colored purple lie,

46:13 um,

46:13 below the nearest canal and the villages

46:15 colored orange lie above the nearest canal.

46:18 The gray villages are far enough away from

46:20 a canal that they're excluded from our sample.

46:22 And just for this district,

46:23 I'm plotting in,

46:24 um,

46:25 on the Y axis of that plot,

46:27 you see,

46:27 it's the share of agricultural land that's irrigated by canals.

46:31 So this is just to,

46:32 to show you the discontinuity we were describing,

46:35 um,

46:35 the orange villages have much lower levels of irrigation than the,

46:39 uh,

46:39 purple villages.

46:40 So that's the discontinuity that we want to exploit in this,

46:43 um,

46:43 design.

46:44 So looking at a result with our full sample,

46:46 um,

46:46 here's this,

46:47 uh,

46:47 regression discontinuity,

46:49 again with canal,

46:50 um,

46:50 irrigation as the outcome variable.

46:53 And we see on the right-hand side,

46:54 um,

46:55 those are the villages that are below the canal that get access to irrigation,

46:58 and on the left-hand side,

46:59 those are above the canal that do not get access.

47:01 So we see a jump at this boundary,

47:04 um,

47:04 and I'm going to show you a bunch more results here

47:06 that just show you that jump at the boundary is on coefficient plots,

47:10 but that's what we're measuring,

47:11 um,

47:11 with the,

47:11 the,

47:12 uh,

47:13 design.

47:14 So here

47:15 is a coefficient plot showing irrigation outcomes.

47:17 So the second one you'll see there is

47:19 the same one we just looked at.

47:20 It's percent of your agricultural area that is irrigated by canals.

47:24 And we're plotting this as a normalized treatment effect.

47:27 So all of these,

47:28 uh,

47:28 coefficients are in terms of standard deviation of their outcome variable,

47:32 so that we can look,

47:33 compare them all together.

47:35 And we see that the villages that lie below the canals really do have,

47:38 um,

47:39 increased irrigation that's driven by canals.

47:41 And

47:42 we notably don't see,

47:43 um,

47:44 too well,

47:44 which is groundwater irrigation or any other source of irrigation.

47:47 There's no difference,

47:48 um,

47:49 at the boundary of in our

47:50 regression discontinuity.

47:52 And these irrigation outcomes lead to agricultural productivity gains.

47:56 So villages below the canal

47:58 have a larger share of,

47:59 uh,

47:59 their area that's cultivated.

48:02 And they also have higher productivity,

48:04 and that's going to be concentrated

48:06 in the rubi,

48:07 uh,

48:08 season,

48:08 the dry season,

48:09 versus the Kari harvest season.

48:11 So,

48:12 that's what we would expect because these canals really

48:14 were designed to deliver water during that dry season,

48:16 and that's where we're detecting,

48:17 uh,

48:17 productivity gains.

48:19 And these villages are also more likely to grow water-intensive crops.

48:22 So we're pretty confident that this illustrates a first stage of,

48:25 of,

48:25 uh,

48:26 what we're hypothesizing here is that

48:28 the,

48:28 um,

48:29 canals really are bringing additional irrigation,

48:31 which is creating long-run,

48:33 uh,

48:33 sustained differences in agricultural productivity.

48:35 So the big question is,

48:36 do these translate into any local non-farm economic outcomes?

48:41 And the answer is,

48:42 uh,

48:43 there is one major one which is population density.

48:45 So

48:46 areas that are irrigated by canals have much higher population densities,

48:50 but notably don't have,

48:52 um.

48:53 Any increase in non-farm employment.

48:56 So we don't see an increase in industrialization,

48:58 so no

48:59 increase in manufacturing or services or even agro-processing employment

49:03 directly in the villages that are serviced by the canals.

49:06 Um,

49:06 this increase in population density,

49:08 I won't show the details here,

49:09 but we have suggestive evidence that this really is driven by

49:11 migration as opposed to changes in fertility or mortality patterns.

49:17 Um,

49:17 just to dig in on one result we showed here is there's,

49:20 there's also no difference in consumption,

49:21 but if we desegregate that by landowners and landless,

49:24 so in the bottom graph here,

49:26 you'll see,

49:26 um,

49:27 landowning households on the top and landless households on the bottom.

49:30 So if you compare landowning households that have

49:32 access to irrigation versus those that don't,

49:34 they have,

49:35 uh,

49:35 higher consumption

49:36 in the modern day,

49:38 while landless households have about equal consumption across that,

49:41 that discontinuity.

49:42 And on the top,

49:42 we just break it out by,

49:44 um,

49:44 the percentile of,

49:46 uh,

49:46 the size of landholdings.

49:48 So the wealthiest households that own the largest land

49:50 get the best returns.

49:52 So the gains from this,

49:53 uh,

49:53 the increase in agricultural productivity are accruing to the fixed factor.

49:57 So landowners are benefiting while landless laborers,

49:59 um,

50:00 are the,

50:01 the gains from the productivity increase are,

50:03 are eaten up by the population density increase.

50:07 Um,

50:08 these results are,

50:09 are robust to several different specifications and checks.

50:12 I won't go into detail for the sake of time,

50:13 but happy to talk more about these later.

50:16 Um,

50:17 so the first question we have is if you have,

50:20 you know,

50:20 an increase in agricultural productivity in these

50:23 local villages,

50:24 there may be many reasons why this would spill over into other villages in the area.

50:29 Um,

50:29 the first could just be that as these villages are,

50:31 are taking water from the canals,

50:33 that allows the water table to run.

50:35 And that groundwater recharge might lead to positive agricultural effects in

50:39 nearby villages that don't get access to the canal water.

50:42 Also,

50:42 labor and goods could be flowing between these,

50:44 um,

50:45 villages and we might see,

50:46 uh,

50:47 a null result in the regression discontinuity.

50:49 That just means the entire area proximal to a canal

50:52 may be,

50:52 uh,

50:52 experiencing,

50:53 um,

50:54 economic gains.

50:56 So we're going to test this with a regression where we hold out the

51:00 above canal tree,

51:01 uh,

51:01 group.

51:02 So those are the villages that are near the canal,

51:04 but don't get treated by,

51:06 um,

51:06 getting access to the canal.

51:07 And we'll compare those

51:09 again to the villages that are below the canal and then to a third

51:12 group of distant villages that are more than 15 kilometers away from the canal,

51:16 and we're confident are not affected

51:18 by the introduction of the canal.

51:21 To create that,

51:22 um,

51:22 comparison group,

51:23 we use an entropy balance weighting so that we ensure that characteristics,

51:27 uh,

51:27 like these geographic fundamentals are matching.

51:29 So we try to match the,

51:30 the distant

51:32 village group to the villages that are treated by the canals.

51:36 Um,

51:36 now,

51:36 these are a little bit hard to,

51:37 hard to see,

51:38 but I just want you to focus on the right-hand side.

51:40 That's the coefficient that's estimating the spillover effect.

51:43 So it's directly comparing

51:44 the above,

51:45 uh,

51:46 villages that lie above the canal,

51:47 but proximal to the canal to distant canals.

51:50 And we mostly see,

51:51 No spillovers in agricultural and irrigation outcomes.

51:55 So there's no difference in agricultural productivity

51:58 or percent of the land that's irrigated.

52:00 There's a small spillover where villages near the canal are more likely to,

52:04 um,

52:05 grow water-intensive crops.

52:07 And moving on to non-agricultural outcomes,

52:10 again,

52:10 we see mostly no spillover.

52:12 So our,

52:12 our villages that are near the canal look very

52:15 similar to the villages that are far away,

52:17 except for population density.

52:18 So there's a small increase in population density,

52:21 um,

52:21 in the areas right near the canal,

52:22 which

52:23 mirrors the influx in population density we saw,

52:26 um,

52:26 from the regression discontinuity.

52:29 But the,

52:29 the major takeaway here is we don't see spillovers in rural manufacturing.

52:33 So there's no increase

52:34 in non-farm employment,

52:36 in service sector employment,

52:37 or manufacturing employment

52:38 in areas that are in,

52:40 that are proximal to the canal but not treated.

52:42 So that leaves us to conclude that we

52:44 don't have much evidence of rural industrialization.

52:48 So our final,

52:49 uh,

52:49 set of results is going to look at

52:52 regional urban growth.

52:53 So our,

52:53 our first two strategies really relied on the spatial distribution

52:57 of the canal network and the villages,

52:59 both the distance and the elevation and how those are all related.

53:02 But we know that the gains in agricultural productivity

53:05 might be

53:06 invested into urban growth in a way

53:08 that's totally orthogonal to that spatial relationship.

53:11 So we could see over time as canals are,

53:13 um,

53:14 constructed in the,

53:15 uh,

53:16 in the region,

53:17 that those gains are invested into town growth.

53:19 So what we do is we have a panel of towns.

53:21 So we have every town from the 2011 population census,

53:25 and we have the population going back every decade until 2001.

53:28 And so for each town,

53:30 we draw a catchment area of a 20 kilometer radius around the town,

53:33 and we look at the area.

53:34 So that's what's

53:35 in yellow,

53:36 and then in blue,

53:37 we have the area serviced by canals,

53:39 um,

53:39 over time as,

53:40 as those change.

53:41 So for every town,

53:42 we can define a share of the area that's treated by canals

53:46 and look at that change over time in a difference in difference analysis.

53:50 So in this,

53:51 uh,

53:51 first result,

53:52 we see that time 0 is when a town gets

53:54 treated by canals in the catchment area surrounding it,

53:58 and we see a significant increase in urban population.

54:01 Um,

54:02 for the periods after the town received access

54:05 to canal water in those outlying areas.

54:08 Um,

54:08 we can also look at an outcome that is town appearance by size.

54:12 So we can threshold town size and say,

54:14 as soon as this town hits 5000 or 10,000 or,

54:17 or 50,000,

54:18 so at what,

54:19 at what part of the size distribution

54:22 is town growth sensitive to the,

54:24 uh,

54:25 canal construction?

54:26 And what's important here is that we see it's concentrated in smaller towns.

54:30 So,

54:30 the smaller town sizes are more sensitive to,

54:33 um,

54:33 increases in canal irrigation in the areas surrounding it.

54:36 So this tells us that small regional towns that might

54:39 be very reliant on agro-processing or trade from agricultural goods

54:43 are sensitive to growth when

54:45 there's more productivity

54:47 around them,

54:48 while large cities uh

54:50 that might be more diversified aren't going to be as affected by,

54:53 by,

54:53 um,

54:53 the canal construction.

54:56 So,

54:57 kind of in summary,

54:58 what we've seen from both the,

55:00 uh,

55:00 the regression,

55:01 discontinuity and spillover analysis as well as this,

55:04 uh,

55:04 town panel of urban

55:06 population data,

55:07 is that there are large movements of people both

55:09 from into rural areas that are treated by canals

55:12 and into urban areas that are in the region.

55:14 So

55:15 the first question you might have is how big are these population movements.

55:18 So if we take these coefficients and do a back of the envelope calculation,

55:22 uh,

55:22 we see that

55:23 the effect is about.

55:23 5 million people living in urban areas and 29 million people living in rural areas,

55:28 um,

55:28 due to this canal network.

55:30 So just for an order of magnitude,

55:32 the most,

55:33 um,

55:34 significant migration event in Indian history is obviously the,

55:37 the partition,

55:37 which

55:38 displaced about 17 million people.

55:40 So we're talking about a large scale movement of people,

55:43 um,

55:43 obviously over a long period of time,

55:45 but it's quite economically significant,

55:47 um,

55:48 to think about the spatial,

55:49 uh,

55:49 mobility of people,

55:50 uh,

55:51 just from these rural infrastructure projects.

55:55 So to conclude,

55:56 um,

55:56 we'd see that these canals introduce

55:58 a long-run difference in agricultural productivity.

56:00 Again,

56:00 these were built,

56:01 um,

56:02 through the 19th and 20th century,

56:04 and we're now looking at the long-run effects in the modern cross section,

56:07 and we still see sustained differences in agricultural productivity.

56:11 Um,

56:12 but those differences

56:13 mostly were equilibrated by movements of labor,

56:16 um,

56:17 across space rather than between sectors in a highly localized way.

56:21 So,

56:21 the structural change that we can detect is not directly in rural areas,

56:24 but rather in,

56:25 in regional urban centers,

56:27 um,

56:28 and any sustained living standard changes

56:30 we see are concentrated among landowners,

56:32 uh,

56:33 as opposed to landless laborers.

56:36 So the implications for these findings are that,

56:38 um,

56:39 you know,

56:39 generally in the long run,

56:40 development entails substantial movement of people across space.

56:44 Um,

56:45 and as we rural infrastructure projects might not

56:48 lead to the kind of rural industrialization.

56:50 But rather relieving barriers to uh mobility and allowing people to move to cities

56:56 might be an effective way to bring about structural change.

56:59 Um,

57:00 and one other thing I'll add is that,

57:02 you know,

57:02 this paper we,

57:03 we deal with uh a shock to agricultural productivity that's driven by

57:07 agricultural technology,

57:08 but we also know agricultural productivity will change in the future,

57:11 especially with,

57:12 with climate change.

57:13 Changing the patterns of what places are most productive,

57:16 um,

57:17 and which are less productive.

57:18 So

57:18 we can expect other shocks to agricultural productivity

57:21 to inspire similar movements of people,

57:24 um,

57:24 across space.

57:25 So it's important,

57:26 uh,

57:26 not definitely not the first paper to point to,

57:28 to mobility due to climate change,

57:30 but,

57:30 um,

57:31 to just in line with those other,

57:32 other findings.

57:33 So,

57:33 um,

57:34 that's all,

57:34 and,

57:34 uh,

57:35 thank you very much.

57:40 OK,

57:40 that was,

57:41 that was awesome as well.

57:42 Thanks very much,

57:43 uh,

57:43 Allison.

57:43 And so Chris Papageorge's going to,

57:45 uh,

57:45 discuss the three papers.

57:54 Alright,

57:54 thanks everyone.

57:55 I'd like to thank the organizers for

57:58 um

57:59 the generous invitation to be part of this great event

58:03 and for for taking a look at these three fantastic papers.

58:07 I can really not do justice to,

58:10 you know,

58:10 the depth of these papers in 15 minutes,

58:13 so

58:14 I would be sending bilateral detailed comments to the authors

58:18 and what I'll be doing here is just,

58:20 just,

58:20 uh,

58:21 present to you what are my top highlights from my readings of these papers.

58:25 I'd like to spend a few minutes in the end of

58:29 zooming out and looking at country convergence and some of my own thoughts

58:34 about some of the emerging challenges uh to developing economies that I see.

58:39 Um,

58:40 so let's start with,

58:41 uh,

58:42 Sumiro's,

58:42 uh,

58:43 paper.

58:45 Which is a fantastic paper uh

58:48 on specifically on the data side

58:52 because it fills a gap,

58:53 a major gap in the literature we have done work on convergence

58:57 looking um

58:59 at country level where the unit of information is uh the country a lot of work there.

59:05 We have also some work that is at the individual level

59:08 but we have very little at the regional level and this

59:11 paper uh

59:12 exactly uh

59:14 fills that gap

59:15 so

59:16 the paper is very novel because it it constructs 600

59:21 regional income level data.

59:24 Um,

59:24 and it spans 34 countries

59:27 for the period 1980 to 2015,

59:30 so for those of you who have done

59:32 work with data,

59:33 this is an incredible achievement.

59:35 Um,

59:36 in the paper you can see how

59:38 the authors are homogenizing,

59:41 uh,

59:41 different regions,

59:42 and we are talking about Tanzania,

59:43 Peru,

59:43 and the US,

59:44 very difficult,

59:45 uh,

59:45 task.

59:46 Then,

59:47 as,

59:47 uh,

59:48 Schmiro have shown

59:49 they have

59:50 used a,

59:50 a very standard spatial

59:52 structural transformation model,

59:54 uh,

59:55 to explain some of the,

59:56 uh,

59:56 main findings.

59:57 Really the key finding of the paper is that there is a sharp slowdown,

1:00:02 uh,

1:00:02 within in within country regional convergence.

1:00:05 Uh,

1:00:06 and,

1:00:06 and then they are using their model to provide a mechanism

1:00:11 of why that's why they observe this,

1:00:13 this fact,

1:00:14 and the way they describe it then in the

1:00:16 model is that this is basically due to the transition

1:00:20 to services

1:00:22 and uh and the economic activity

1:00:25 being specially concentrated

1:00:27 which results in this decline in regional convergence.

1:00:31 Uh,

1:00:31 so I have a few comments.

1:00:33 Definitely the first one is on data,

1:00:36 um,

1:00:37 you know,

1:00:38 uh,

1:00:38 the devil is in the details on these constructions,

1:00:41 and,

1:00:41 uh,

1:00:41 I have seen that the authors

1:00:44 relied quite a bit on interpolations,

1:00:46 and I've never been

1:00:47 a fan of,

1:00:48 uh,

1:00:49 interpolating data.

1:00:50 There is a lot of work.

1:00:51 Uh,

1:00:52 for example,

1:00:53 from Angus Deaton and others that show that

1:00:55 when we finally figure out actual data and we replace the interpolated data,

1:01:01 many of the results

1:01:02 are overturned.

1:01:03 So I would

1:01:04 say in my bilateral comments,

1:01:05 uh,

1:01:06 Schmitra,

1:01:06 basically I guide you to some

1:01:08 new,

1:01:09 uh,

1:01:09 now casting machine learning,

1:01:11 uh,

1:01:12 uh,

1:01:13 empirics where basically they take into

1:01:16 account all of the information that you have out there.

1:01:19 To feel uh this data,

1:01:21 basically the signal to noise that you will be obtaining

1:01:24 from,

1:01:25 uh,

1:01:25 what I suggest I think is is higher than what you have right now.

1:01:29 Um,

1:01:29 the second issue that I wanted to bring up is,

1:01:32 um,

1:01:33 sort of,

1:01:33 you know,

1:01:34 I take a little bit of an issue of how quickly

1:01:36 the authors come to the conclusion

1:01:38 about country convergence and that

1:01:40 the evidence is that we have actually country convergence,

1:01:44 um,

1:01:44 I have done quite a bit of work.

1:01:46 I have a gel paper in 2020 with Paul Johnson.

1:01:49 And we had looked at,

1:01:50 we looked at the the evidence very carefully,

1:01:53 um,

1:01:54 and

1:01:54 what we came out with is that

1:01:56 the jury is still out so

1:01:58 basically the way we explain some recent papers which

1:02:02 focus this convergence because

1:02:03 they look at shorter periods we call these episodes of convergence if you'd like

1:02:08 I'll relate this comment actually to my final points about

1:02:12 kind of zooming out and looking at

1:02:14 uh cross country uh convergence.

1:02:16 It's a fantastic paper let's move on to the next paper

1:02:19 next paper

1:02:20 takes us somewhere else on growth the

1:02:23 one of the key

1:02:25 production functions,

1:02:26 the human capital

1:02:28 specifically on talent and uh

1:02:30 innovation migration basically the paper

1:02:33 does a fantastic job in offering a new model

1:02:36 of brain train

1:02:37 brain gain

1:02:38 it's a very complicated question.

1:02:40 Um,

1:02:42 the,

1:02:42 the paper,

1:02:44 um,

1:02:44 complements the model with

1:02:47 fantastic micro level data.

1:02:50 Um,

1:02:50 they look at basically patterns in the EU US migrant,

1:02:55 uh,

1:02:55 corridor.

1:02:57 And finally they have they don't stop with positive analysis they do

1:03:00 also

1:03:01 some policy experiments,

1:03:02 some key policy experiments one tax cuts

1:03:06 for uh the migrant innovators and the second

1:03:09 in increasing the cap of H-1B visas in the United States.

1:03:14 So

1:03:14 main findings,

1:03:15 and I'll stick only to the policy here

1:03:17 is not surprising the tax cuts

1:03:21 have positive effects in the short run,

1:03:23 negative effects in the long run.

1:03:25 The increase in the cap of H-1B visas is good for everyone,

1:03:29 the US,

1:03:31 the,

1:03:31 the European Union,

1:03:32 the global growth.

1:03:34 So two main comments I think,

1:03:36 uh,

1:03:36 Martha has already

1:03:38 picked up on the first one.

1:03:40 Um,

1:03:40 you know,

1:03:41 this is a fantastic tool,

1:03:42 a framework,

1:03:43 a template,

1:03:44 and we look forward to having the published version

1:03:48 and taking this and applying it to developing economies like

1:03:51 the corridor being

1:03:53 EU,

1:03:54 I don't know,

1:03:54 Africa or Asia.

1:03:56 Uh,

1:03:57 Europe,

1:03:57 uh,

1:03:58 so those would be very exciting

1:04:00 ways by which we can bring in developing economies in this.

1:04:03 I understand the

1:04:05 reason why they went to EU US for,

1:04:08 for,

1:04:08 for data availability,

1:04:09 but I think this is a very huge opportunity.

1:04:12 Have

1:04:12 uh to go in different corridors with developing economies

1:04:16 then of course we can go further we can look

1:04:19 at refugee uh innovators we can also you look at

1:04:22 more specialized innovators

1:04:24 uh with all the new reality of AI.

1:04:28 The second,

1:04:29 uh,

1:04:29 comment I have is for

1:04:31 to get some clarity from the authors,

1:04:33 uh,

1:04:33 to get them to think a little bit

1:04:35 further on

1:04:36 the assumptions made is a complicated model

1:04:40 and assumptions are very important.

1:04:42 One that I have a little bit difficulty understanding.

1:04:46 Is

1:04:47 whether that assumption made about the inventor's experience

1:04:51 a productivity boost when they are moving abroad

1:04:55 kind of

1:04:56 pinning down the results and and whether

1:04:58 uh I'm interpreting this correctly,

1:05:00 you know,

1:05:01 basically that's the conclusion of the model

1:05:03 and policy implications depend on that assumption,

1:05:06 something that we can discuss further

1:05:08 in the Q&A,

1:05:09 but again a fantastic uh

1:05:11 uh paper Martha

1:05:13 so finally we come to the last paper.

1:05:15 Uh,

1:05:15 Allison's paper,

1:05:17 uh,

1:05:17 on,

1:05:17 uh,

1:05:18 the process by which we grow the structural transformation,

1:05:21 especially early stages on agricultural productivity,

1:05:24 a fantastic paper,

1:05:26 you know,

1:05:27 very rich,

1:05:28 uh,

1:05:28 it took me forever to read,

1:05:29 it's like 78 pages,

1:05:31 so,

1:05:32 uh,

1:05:32 but,

1:05:32 uh,

1:05:33 but,

1:05:33 uh,

1:05:33 wonderful,

1:05:34 so

1:05:34 it careful,

1:05:36 carefully documents how investment in irrigation using the India example,

1:05:41 the canal example.

1:05:43 Um,

1:05:43 results in this positive return on productivity,

1:05:47 but to me the most exciting part of the paper bear is the disaggregation,

1:05:51 right?

1:05:52 So

1:05:53 you go from

1:05:54 looking at the results at impact,

1:05:56 which is the irrigated villages.

1:06:00 You know then you have the

1:06:02 connection

1:06:03 with rural areas and then you go to

1:06:06 regional urban economies and then you have possibly

1:06:10 the global economy.

1:06:10 This is a fantastic way

1:06:12 of mapping out

1:06:13 the impact

1:06:15 and the ripple.

1:06:15 Effects the total economy so I,

1:06:17 I

1:06:18 for me that was the most exciting part of the

1:06:21 of the paper and for for that to happen of course

1:06:24 the authors use different empirical methods,

1:06:26 uh,

1:06:27 a multi-sector location uh model by

1:06:31 you know I think motivated by boosters and and so forth

1:06:34 um

1:06:35 so

1:06:36 the findings are clear

1:06:38 for me the biggest finding is on population movement.

1:06:42 Uh,

1:06:42 being so central to

1:06:45 productivity,

1:06:46 uh,

1:06:46 gains and,

1:06:47 uh,

1:06:47 the key mechanisms for structural transformation,

1:06:50 a lot of papers have been talking about this,

1:06:53 but this paper documents,

1:06:55 uh,

1:06:55 quantitatively that impact.

1:06:57 Uh,

1:06:57 the numbers I had was 9 million,

1:06:59 uh,

1:07:00 drawn to cities and,

1:07:01 uh,

1:07:02 32 million in rural areas.

1:07:03 I think the numbers changed a little bit now

1:07:05 is 5 and 29,

1:07:07 but still,

1:07:08 uh,

1:07:08 massive,

1:07:09 uh,

1:07:09 impacts.

1:07:10 Um,

1:07:11 on my main comments is for future work mainly,

1:07:14 uh,

1:07:14 one,

1:07:15 I understand the limitation of the paper

1:07:17 to draw

1:07:19 implications,

1:07:20 um,

1:07:21 at the aggregate level,

1:07:22 and,

1:07:23 uh,

1:07:23 I understand that the limitation comes from measuring labor flows directly.

1:07:28 I was wondering there and this is what I'm sending in my bilateral,

1:07:31 uh,

1:07:31 comments whether.

1:07:33 You,

1:07:33 you could do surveys to fill those gaps and we have come a long way on this.

1:07:37 The World Bank is an expert on this

1:07:39 um the fund is catching up.

1:07:41 I think

1:07:42 actually this surveys can be done and can be done effectively and

1:07:46 go that extra distance to say something about aggregate

1:07:50 effects,

1:07:50 uh,

1:07:51 which now I think you

1:07:52 you are you come a little short.

1:07:54 Um,

1:07:55 the second I think is something that

1:07:57 Alison picked up at the very end of the presentation which is

1:08:01 exactly these shocks how do major shocks impact agricultural productivity,

1:08:07 both economic

1:08:08 but also climactic,

1:08:09 and uh I'll come to this

1:08:11 in my discussion in a minute.

1:08:14 OK,

1:08:14 so

1:08:15 I'll spent the last 5 minutes I have on kind of zooming out and talk

1:08:19 about um

1:08:20 country convergence and

1:08:22 before I get to my

1:08:24 uh to the challenges that I worry most about uh facing developing economies

1:08:28 I'll start with the current context.

1:08:31 Most of the papers that have been written

1:08:33 and trying to calculate country convergence stop at 2019,

1:08:38 2018,

1:08:39 even before that 2015.

1:08:41 Uh,

1:08:42 a lot has happened and not in the right direction for developing economies.

1:08:46 Uh,

1:08:47 some people can call it transitional.

1:08:49 There are all these shocks,

1:08:49 the pandemic shock,

1:08:51 Ukraine war,

1:08:52 um,

1:08:53 of

1:08:53 financing tightness in the,

1:08:55 in the global markets.

1:08:56 But we also know that,

1:08:59 you know,

1:08:59 uh,

1:09:00 impacts that seem transitory or are transitory for

1:09:03 emerging markets and advanced economies are permanent for

1:09:06 developing economies that,

1:09:07 that worries me.

1:09:09 But also this is in the back of

1:09:11 what we think is a shift in the global economy

1:09:14 for many decades we have been thinking about convergence

1:09:18 when the economy,

1:09:18 the global economy was doing fantastically maybe because of China

1:09:22 and other factors but now we are getting into an era where

1:09:25 the projections from the World Economic Outlook elsewhere show

1:09:30 a truly a real

1:09:33 global slowdown mainly because of China

1:09:35 so.

1:09:36 You know there is this the the so the initial condition for

1:09:39 many of the developing economies given the shocks and given the new phase

1:09:44 of,

1:09:44 uh,

1:09:45 global slowdown

1:09:46 in growth

1:09:48 is not very promising

1:09:50 but this is not really what worries me mostly for developing economies are these

1:09:54 emerging challenges and here I I'd like to talk about 3

1:09:59 and I have to move on here.

1:10:03 OK this is it

1:10:04 um climate AI and fragmentation you all know about climate,

1:10:09 uh we have seen it in the fund

1:10:11 uh

1:10:11 how many of the countries come to us

1:10:14 for financial assistance because of the shocks

1:10:17 this demand is increasing at an increasing rate,

1:10:20 um,

1:10:21 so

1:10:22 you know

1:10:23 you wonder for example Alison how

1:10:27 your work would change when you take into account

1:10:30 this climatic changes both

1:10:31 increasing,

1:10:32 you know.

1:10:33 Uh,

1:10:33 frequency,

1:10:34 intensity,

1:10:35 duration of all these,

1:10:36 uh,

1:10:37 major shocks we had

1:10:38 floods in India and so forth.

1:10:40 I think,

1:10:40 I think that that would be a very nice natural experiment for you guys.

1:10:44 Um,

1:10:45 second is AI,

1:10:46 um,

1:10:47 conceptual frameworks show that.

1:10:50 You know robots,

1:10:51 especially the low end robots,

1:10:53 substitute for low skilled workers,

1:10:56 and

1:10:57 many of you know worked in

1:10:59 developing economies there was this demographic dividend

1:11:02 dream in Africa,

1:11:03 right,

1:11:03 that in 20 years,

1:11:05 you know,

1:11:05 we'd have the youngest population.

1:11:07 You know this new robots that keep work

1:11:11 basically in China

1:11:13 uh where it should have been in Africa by real people

1:11:17 it's something that worries me quite a bit so.

1:11:20 Uh,

1:11:20 Martha,

1:11:20 I think it is very interesting to think about

1:11:22 your work with

1:11:24 AI migrants,

1:11:26 of course,

1:11:26 for

1:11:27 your,

1:11:28 uh,

1:11:28 work is EU and US,

1:11:30 but I mean you can extend it for

1:11:32 developing economies as well.

1:11:33 Finally,

1:11:34 fragmentation is,

1:11:35 I think something that we don't take very seriously yet,

1:11:38 but we see it happening.

1:11:39 There will be a trade war between China and the US.

1:11:42 And where does that leave our discussion of uh

1:11:46 country convergence?

1:11:47 Should we be thinking about club convergence going back to that world

1:11:51 um uh so Shimito I mean that actually this is even

1:11:54 more important for you when you consider regional uh convergence.

1:11:57 Let me stop here.

1:11:58 Thank you.

1:12:04 OK,

1:12:05 terrific thank you Chris.

1:12:06 Um,

1:12:07 so we're gonna open it up,

1:12:08 uh,

1:12:09 in a moment for,

1:12:10 um,

1:12:11 some Q&A,

1:12:12 um,

1:12:12 and I'll,

1:12:13 uh,

1:12:14 give

1:12:14 each,

1:12:15 uh,

1:12:15 each presenter an opportunity to address Chris's

1:12:18 comments and the questions that come up.

1:12:20 Let me,

1:12:20 let me just,

1:12:21 uh,

1:12:22 put.

1:12:23 One or two,

1:12:24 general issues,

1:12:25 uh,

1:12:26 that came to my mind listening to the three excellent presentations.

1:12:30 Um,

1:12:31 Sumitra's central

1:12:32 finding about,

1:12:33 uh,

1:12:34 slowing,

1:12:35 uh,

1:12:35 regional convergence,

1:12:37 um,

1:12:38 is that a policy problem and what,

1:12:40 what are the policy remedies,

1:12:42 uh,

1:12:42 in your opinion if it is a policy problem?

1:12:46 Uh,

1:12:46 second,

1:12:47 for,

1:12:48 um,

1:12:48 uh,

1:12:49 for Marta,

1:12:50 um.

1:12:52 You have very strong results about short run uh gains to output and,

1:12:56 uh,

1:12:56 longer run costs

1:12:58 again,

1:12:58 sort of from a,

1:12:59 from a political economy perspective,

1:13:01 would you use this kind of a model,

1:13:03 uh,

1:13:04 to advocate certain policies whether on

1:13:06 uh visa reform or tax reform?

1:13:09 Uh,

1:13:09 how,

1:13:09 how do you think,

1:13:10 uh,

1:13:10 you would sell these very long run results

1:13:13 from a,

1:13:13 from a political economy perspective,

1:13:16 uh,

1:13:16 and,

1:13:16 uh,

1:13:17 third,

1:13:17 Alison,

1:13:18 um.

1:13:19 Your your results about,

1:13:21 uh,

1:13:22 the gains accruing uh

1:13:24 to the fixed factor,

1:13:25 uh,

1:13:26 entirely,

1:13:27 uh,

1:13:27 and the fact that,

1:13:28 um,

1:13:29 uh,

1:13:29 you know,

1:13:30 the non-farm sector doesn't really,

1:13:32 uh,

1:13:32 doesn't really benefit at all,

1:13:34 um,

1:13:34 uh,

1:13:35 what,

1:13:35 what policy remedies,

1:13:37 uh,

1:13:37 do you see as,

1:13:39 uh,

1:13:39 potential levers to,

1:13:40 uh,

1:13:41 to address those,

1:13:42 um,

1:13:42 uh,

1:13:42 those issues?

1:13:44 So let me,

1:13:44 let me turn it over,

1:13:45 uh,

1:13:46 now to questions and I'll,

1:13:47 I'll,

1:13:48 I'll turn back to the speakers.

1:13:49 Uh,

1:13:50 do people want to,

1:13:51 um,

1:13:51 uh,

1:13:52 intervene at this point?

1:13:58 Yeah.

1:14:08 Uh,

1:14:08 thanks,

1:14:08 Jonathan.

1:14:09 I'm Art Cray at the World Bank.

1:14:11 Um,

1:14:11 you,

1:14:12 Jonathan,

1:14:12 you asked the question I was going to ask Jomiro,

1:14:14 so I'll skip that one,

1:14:15 but I also have questions for Marta and for Alison.

1:14:19 So for Marta,

1:14:19 two quick questions on the paper.

1:14:21 The one is that,

1:14:23 um,

1:14:23 I think in the real world a lot of migration of inventors is because they already

1:14:28 know who their collaborators are in the other country and they move to where they are

1:14:32 and I wonder to what extent that changes the mechanisms and the model.

1:14:35 The second question has to do with the policy experiments that.

1:14:38 You look at,

1:14:39 so you showed us

1:14:40 a policy experiment in which one part of the world taxes innovation

1:14:44 and so in the end it's,

1:14:46 you know,

1:14:46 not that surprising that there's negative effects in the long term.

1:14:49 It'd be interesting to see what happens if you say had

1:14:52 fully integrated patent protection between the US and the EU.

1:14:56 You know,

1:14:56 how,

1:14:56 how much more positive would a policy like that be.

1:14:59 And for Alison,

1:15:00 two questions.

1:15:01 One is just a little one,

1:15:02 which is why don't people pump uphill out of canals

1:15:04 if there's such a high return to doing so?

1:15:06 The technology obviously is there because of all the groundwater pumping in India.

1:15:11 The second question is,

1:15:12 can you say more about the

1:15:14 history

1:15:15 of the placement of canals because I was thinking about this

1:15:18 in the sort of slightly cynical way and you think about,

1:15:21 you know,

1:15:21 these are big public works,

1:15:23 uh,

1:15:24 projects when they were built,

1:15:25 they were probably built under the influence of landowners and it's not

1:15:28 very surprising that they lined up in places that benefit the landowners.

1:15:32 Is that oversimplifying?

1:15:33 Is there something to that history that matters for the findings today?

1:15:38 Thanks,

1:15:39 Art.

1:15:39 And,

1:15:39 and I apologize for not recognizing your face.

1:15:42 The light completely blinds me from where I'm,

1:15:44 uh,

1:15:45 where I'm sitting.

1:15:45 Other,

1:15:45 other questions?

1:15:53 Are there any,

1:15:53 uh,

1:15:54 questions in the chat that are,

1:15:56 uh,

1:15:56 that are bubbling up?

1:15:58 OK,

1:15:59 so,

1:15:59 um,

1:16:00 why don't we,

1:16:00 uh,

1:16:01 give,

1:16:01 uh,

1:16:01 each speaker a chance to,

1:16:03 uh,

1:16:04 to respond,

1:16:05 Sumitra,

1:16:06 uh,

1:16:06 thanks,

1:16:06 Chris,

1:16:07 uh,

1:16:07 for all your kind and generous comments.

1:16:09 Um,

1:16:10 both points sort of well taken.

1:16:11 Um,

1:16:12 I'll just say that,

1:16:13 I mean,

1:16:13 we have to interpolate one way or the other to look at the trends.

1:16:17 Um,

1:16:17 the validation exercise that we did with the interpolation

1:16:20 was that we only interpolated the within country data.

1:16:24 And then we aggregated it up

1:16:26 to see whether or not we are matching the aggregate GDP that

1:16:30 you get from the Pen World table of the World Bank.

1:16:33 But if there are ways to kind of improve it further,

1:16:35 of course,

1:16:35 um,

1:16:36 um,

1:16:36 I would look for the,

1:16:37 um,

1:16:37 um,

1:16:38 detailed comments.

1:16:39 Um,

1:16:39 the cross-country things I I'm aware of the debate between sort of your paper,

1:16:44 and now there's a paper by Michael

1:16:45 Kramer that is claiming convergence and the Sandifer

1:16:48 Subramani and Patel paper.

1:16:50 Uh,

1:16:51 I thought it got resolved one way,

1:16:52 but maybe not,

1:16:53 and my reading is incorrect.

1:16:55 The thing is,

1:16:55 it's not even central to our paper.

1:16:57 Like,

1:16:58 I can rewrite the framing and say there has been

1:17:00 a lot of growth in the developing world and,

1:17:02 and,

1:17:03 and just focus on within-country convergence.

1:17:05 Um,

1:17:06 um,

1:17:06 but again,

1:17:07 would look,

1:17:07 um,

1:17:08 forward to talking more.

1:17:09 Um,

1:17:10 on the policy side,

1:17:11 um,

1:17:12 I mean,

1:17:12 let me even sort of extend that question

1:17:15 on the policy and politics,

1:17:17 um,

1:17:17 um,

1:17:17 problem of this.

1:17:19 Um,

1:17:19 I think that would depend on country by country,

1:17:22 like,

1:17:22 so I,

1:17:23 I mean the issue is that the,

1:17:24 the,

1:17:24 the regions that are getting left behind.

1:17:27 Uh,

1:17:27 uh,

1:17:28 uh,

1:17:28 uh,

1:17:29 what is happening to kind of people there and are people able to migrate to,

1:17:33 to regions with opportunities?

1:17:35 And of course,

1:17:35 like,

1:17:36 when you focus on countries like India and markets here,

1:17:39 there are migration restrictions.

1:17:40 So,

1:17:41 Bihar and UP,

1:17:42 for example,

1:17:43 tend to be the most unproductive states,

1:17:45 but also the

1:17:47 states with the most population and population density.

1:17:50 Um,

1:17:50 um,

1:17:50 and that's kind of part of the policy challenge.

1:17:53 But if you look at the US,

1:17:54 we are like,

1:17:55 I guess,

1:17:55 facing a completely different story here where the Midwest,

1:17:58 which is kind of falling behind,

1:18:00 um,

1:18:00 there are not too many people here,

1:18:02 but they do have an equal representation in terms of the politics.

1:18:06 And,

1:18:06 and so this,

1:18:08 this unequal sort of spatial growth in

1:18:10 countries depending on political structures will.

1:18:13 Um,

1:18:13 um,

1:18:14 kind of have impact

1:18:15 on political outcomes as well.

1:18:17 So,

1:18:18 of course,

1:18:18 we don't get into those issues in the paper,

1:18:20 but that's kind of what we have at the back of our minds.

1:18:25 Thank you,

1:18:25 um,

1:18:26 Mara.

1:18:30 Yeah thank you so much for the comments and the question and

1:18:34 uh

1:18:34 to start with uh from the comments by Chris I would I would say that

1:18:39 it is certainly an important assumption in this model that migrants

1:18:43 tend to

1:18:44 increase their productivity when they move and this is.

1:18:46 Because

1:18:47 these are migrants motivated by economic reasons,

1:18:50 so they move when they have good opportunities in the model

1:18:53 and that's sort of like reflects what I see

1:18:55 in the data that on average these migrants tend to

1:18:58 find more patterns after they move

1:19:00 but of course one could think of different settings where.

1:19:03 People including inventors move for different reasons like

1:19:07 refugees or as we were talking about the war in Ukraine that caused a

1:19:11 huge reallocation of people or as we also talked about

1:19:14 uh migration motivated by climate change

1:19:17 and so in this case it might be it might not

1:19:19 be necessarily the case that people actually increase productivity when they move

1:19:24 so 11 channel of this one looks at the effects

1:19:26 that come sort of like from the reallocation of people.

1:19:30 But on the other side,

1:19:31 there is also this important channel that comes from the knowledge

1:19:33 spillovers from the destination country back to the origin country.

1:19:37 And here again it really depends on the nature of migration.

1:19:41 This type of econo economic migrants that move for economic opportunities

1:19:46 are able to maintain a network in the country of origin,

1:19:49 but when we think about

1:19:50 refugees,

1:19:51 it might be the case that

1:19:52 even if they move to a different country and they may become more productive,

1:19:55 it might be very difficult for them to keep a network,

1:19:58 a professional network in their home country.

1:20:00 And so this connects to the questions about policy

1:20:03 where what I've done so far in this work is to

1:20:06 do a positive evaluation of some policies that these countries have implemented

1:20:11 to really answer the question about policy recommendations more deeply.

1:20:14 I think this would require

1:20:16 sort of solving

1:20:17 a planning problem to think about what the optimal policies are

1:20:21 and so in that sense I think

1:20:23 this is a great comment to think more carefully about

1:20:26 other types of policy completely for example instead of tax.

1:20:29 Where tax cuts,

1:20:30 what about,

1:20:31 uh,

1:20:31 a fully integrated patent protection or what

1:20:34 about temporary migration programs that could be,

1:20:37 could foster these knowledge spilovers

1:20:39 and this is,

1:20:40 this is again something that requires even uh

1:20:42 uh further analysis of these,

1:20:44 this

1:20:45 type of models

1:20:47 and,

1:20:47 and,

1:20:48 and related I think to to the other question about the fact that very often

1:20:51 people move

1:20:53 when they already know some people in their destination country.

1:20:55 I think this is,

1:20:56 this is true.

1:20:57 And this is something that I've seen the data where

1:21:00 actually most of these migrants move within

1:21:03 foreign branches of the same multinational that they already work for

1:21:06 but for the

1:21:08 um main channels of this model this is fine uh

1:21:11 for for.

1:21:12 For simplicity in my model I use sort of random matching of people in the economy,

1:21:17 but one can think about

1:21:19 uh directed matching where people know who they want to meet

1:21:22 and if anything I think this would make

1:21:24 the knowledge spilovers in the model even stronger.

1:21:28 Thank you.

1:21:30 Thank you,

1:21:30 Marta and Allison.

1:21:33 Uh,

1:21:33 great,

1:21:33 thank you all for those questions.

1:21:35 Um,

1:21:35 I guess

1:21:36 on the first point about kind of the effects of climate

1:21:38 on agricultural productivity,

1:21:40 I think one interesting,

1:21:41 um,

1:21:42 thing to think about in our paper is the result that the of the town growth,

1:21:46 the regional town growth as a result of the increase in agricultural productivity,

1:21:49 and

1:21:50 I think an interesting implication is that not only

1:21:52 are the outlying areas that might be experiencing drought.

1:21:55 Um,

1:21:55 going to be affected,

1:21:56 but that these towns that might be supported

1:21:58 by,

1:21:58 um,

1:21:59 the productivity and the demand of people in rural areas.

1:22:01 So thinking about I guess the,

1:22:03 the policy,

1:22:04 um,

1:22:04 ramifications of,

1:22:05 of needing of people

1:22:07 needing to move not just from extreme weather events,

1:22:09 uh,

1:22:09 disasters but also just changing,

1:22:11 uh,

1:22:12 economic realities of what what populations are able to be supported in what areas,

1:22:16 um.

1:22:17 On the,

1:22:18 the question about policy,

1:22:19 uh,

1:22:19 ramifications,

1:22:21 um,

1:22:21 so there's some other work by our co-authors Paul Novasad and Sam Asher looking at,

1:22:26 uh,

1:22:26 rural roads also in India

1:22:28 and looking at the places that get access to roads,

1:22:31 and they find that similarly,

1:22:32 um,

1:22:33 the roads essentially facilitate people to migrate to

1:22:35 other areas to search for other opportunity.

1:22:37 So in a lot of these,

1:22:38 um,

1:22:38 areas,

1:22:38 we're seeing

1:22:39 these infrastructure projects just leading to migration.

1:22:42 So I think,

1:22:42 um,

1:22:43 understanding more of those,

1:22:44 uh,

1:22:44 incentives to migrate,

1:22:45 like a lot of the work in Bangladesh,

1:22:47 um,

1:22:48 running

1:22:48 trials to incentivize people to move during the dry seasons,

1:22:52 or giving people vouchers to move to cities and understanding

1:22:54 what are the impediments to,

1:22:56 uh,

1:22:56 migration would be interesting to,

1:22:58 to understand more about,

1:22:59 um,

1:23:00 especially

1:23:00 making the destinations more appealing to actually migrate to and not,

1:23:04 um,

1:23:04 deter people from moving to cities and,

1:23:06 and living in,

1:23:06 in bad conditions.

1:23:08 Um,

1:23:08 and then lastly,

1:23:09 I'll just address the questions.

1:23:11 I,

1:23:11 I should know more about the answer of why don't people pump uphill from canals.

1:23:14 Um,

1:23:15 I imagine part of it has to do with the fact that these are,

1:23:18 many of these canals were built before that was a viable

1:23:21 option.

1:23:21 And I wonder how the,

1:23:22 um,

1:23:23 kind of gains accrue to landowners and how,

1:23:25 what capital is there to invest in those,

1:23:27 um,

1:23:27 you know,

1:23:27 how,

1:23:27 how expensive is it to,

1:23:28 to pump that water uphill and who's choosing to make that investment or not.

1:23:32 Um,

1:23:33 but it's a good question.

1:23:34 We should probably,

1:23:34 I should probably have a better answer for that.

1:23:36 Um,

1:23:37 and then lastly,

1:23:38 just on the,

1:23:38 the placement of canals,

1:23:39 we've we've definitely thought a lot about how,

1:23:41 how that could be endogenous.

1:23:43 Um,

1:23:43 there's actually another,

1:23:44 uh,

1:23:45 regression discontinuity strategy we use,

1:23:47 um,

1:23:48 where we look at

1:23:49 the,

1:23:49 the,

1:23:49 the government puts out polygons drawn around the canals called command areas,

1:23:53 which is the area that the canal should service.

1:23:56 And we do a spatial discontinuity looking

1:23:57 at places just inside the boundary versus.

1:23:59 Outside,

1:24:00 but we actually thought

1:24:01 those might be more likely to be,

1:24:03 um,

1:24:03 influenced by politics or,

1:24:05 or whoever is drawing those maps.

1:24:07 So we think that the elevation strategy is,

1:24:10 um,

1:24:10 more dependent on geographic fundamentals rather than

1:24:13 maybe other incentives.

1:24:14 Um,

1:24:15 and then we also do a test

1:24:16 where we look at

1:24:18 the longest and straightest canals.

1:24:19 Um,

1:24:20 this test has been done with some road construction,

1:24:22 um,

1:24:23 Uh,

1:24:23 as well to say if they needed to get water from point A to point B,

1:24:27 which places were treated along the way,

1:24:29 and especially when you're limited geographically and,

1:24:31 and where the canal can go to get from A to B,

1:24:34 it's more likely that those are,

1:24:35 uh,

1:24:36 treated

1:24:36 kind of by circumstance as opposed to endogenous to interest preconstruction.

1:24:41 Um,

1:24:42 yeah.

1:24:44 OK,

1:24:44 thank you to Shamitro,

1:24:45 to Martha,

1:24:46 to Alison,

1:24:47 and to Chris.

showAllTimestamps
no
transcript
OK, um, very, uh, warm welcome to this, uh, second session on, uh, growth and convergence. Um, we have, uh, 3 papers and they're all terrific. Uh, I applaud both the authors and the, uh, organizers of this conference for selecting them. They're really outstanding papers, um, and I very selfishly, I teach a course on economic growth at Georgetown and I'm. Already looking at what I can cannibalize from these lovely papers, so, uh, each, uh, presenter has, uh, 20 minutes, uh, and I've been told to, uh, uh, be rigorous in enforcement. So we will start with, uh, Chamiro and, uh, and then, uh, move on from there. Thank you. Is there So sorry. All right, uh, uh, thanks a lot to the organizers for including our paper in today's, uh, agenda. Um, so this is joint work with Elissa Genone who is at CRE in Barcelona and Can Cuno, our PhD student at Penn State, um. Uh, before I begin, I must say that parts of this project are sort of still work in progress. This has been a huge kind of data collection exercise, and I'm going to point out to those pieces, uh, where we should be slightly careful about interpreting the quantitative numbers. But just to begin, let's, I mean, what do we know about. Economic growth in the last 30 years, we know that there has been a lot of growth amongst the poorer countries in the world and on an average poorer countries have caught up with richer countries. Um, and then there's a little bit of evidence of what has happened within countries, uh, um, mostly the US and, and some in the UK. What we really don't know is what has happened to the evolution of spatial inequality, uh, uh, within countries over time in these last 30 years which has seen a lot of economic growth in the world, right? And, and with the evolution of spatial inequality in particular what has been the role of um um structural. Change in shaping this, this evolution. So that's going to be the focus of our paper, um, uh, a key contribution of this project is kind of building this new data set on the GDP per capita at subnational units over time so both kind of states and provinces of countries, but also at the city level. And then more importantly, something that did not exist at all is the labor force composition by sectors within each region of various countries of the world, right? So using this data set, we first kind of document three facts. One is for the average country in the world in the last 30 years, um, convergence between regions within a country has been falling over time. Um, the second is that the, the, the, this, this stall in uh spatial convergence is related to growth of services as the key driver of development in the last, um, um, 20 years or so. Um, and the, and the last fact, Is that much more than happened during the manufacturing phase in the services phase, um, especially kind of professional and business services, employment is getting very, very concentrated in few regions of, of, of countries. So, Then we were, so once we had the facts, we were thinking like, could, could a sort of canonical structural change model with space built in it explain these facts, or do we need some other ingredients? And in terms of the theory, I'm going to show you a very, very simple kind of canonical model of structural change with kind of space built into it, um, which is going to at least qualitatively match, um, um, our facts, um, very well. Um, we're contributing sort of to two large literatures. Uh, uh, uh, one is just the literature on the evolution of spatial income disparities, um, and the second, which is a more recent literature, which takes kind of structural transformation and space, uh, uh, more seriously, um, um. Um, on the first one, we are just expanding the evidence to, to many, many more countries than existed before. And in the second one, we are kind of highlighting, um, the role of agglomeration and services, uh, uh, which is going to be kind of a key feature of our, of our theoretical framework. So just to spend a little bit of time on, on the data, we are building on an existing data set by Giannoli, LaPorta Schleifer, uh, uh, but we have added many, many more countries that, that data set did not have. We have also went on and corrected some of the data. So for example, the data for India in that data set was missing for many, many years. Uh, uh, uh, uh, so we have updated it to the most recent year possible for our sample of countries. Um, from the Economist, we also have Data on the GDP at the level of cities, um, in the world. So sort of combining these two data sets, we have like a balanced panel on GDP per capita by region within a country, um, for 34 countries between 1980 and 2017, but we also have an unbalanced panel that starts much further back and essentially the reason is that for the advanced countries you, you, you can go back further, but for the developing countries you don't get data, um, going back far in time. We have added to this. This data on years of schooling by subnational region by year, um, in the world, um, and, um, the labor force composition by, by kind of broad sectors. We are, we are, so this summer we are making a, a, a new effort to kind of, so right now we have the labor force composition by just broad sector agriculture, manufacturing and services. Now we are making an effort to break down services into further subsectors so that we can get the tradable high skill part of the services in, in much more, um, richer detail. In terms of coverage, well, you can see that we're doing sort of a pretty good job in most continents, um, um, except for Africa and since I'm standing at the World Bank, um, here is a call for help, uh, uh, if any of you can help us add more countries in the African continent, that would be terrific. We have tried a lot. Like we write to like national statistical agencies in various countries. We sometimes get replies and other times don't. But, but, but Africa is one continent that is kind of, so we have done some robustness with night lights, etc. but it would still be useful to have actual employment and GDP figures at the subnational level. So let me start with the facts and so this is kind of your familiar convergence regression, except now we are doing it at the sub, uh, uh, uh, within countries. So you regress kind of growth in GDP per capita, uh, uh, for the next 10 years, um, for a given state or province on the initial level of GDP per capita, um, at time T0. So for a particular country. Uh, uh, and as must be familiar to all of you, if this beta is kind of less than 0, then that points towards convergence, and if it's positive, then that means the poorer regions are not catching up with the richer regions. So we run this regression for all the countries in our sample, country by country. So we have one beta C, uh, uh, at a given point in time. Uh, uh, uh, for each country and then for the figure I'm going to show you, we're going to average this across all countries, right? So, what this, so what this point is showing is the average beta C for all the 34 countries in our sample and essentially showing that in the 1980s, there was a lot of within-country convergence on an average in the world. And then over time there has been a sharp decline in the within-country convergence rates and by the 2000s, the within-country convergence process has basically come to a complete stall. Um, these are unweighted regressions. If you do population weighted regressions, the, the, the key fact that in the, in the 2000s, there is no within-country convergence. It's still true except that the decline happened, uh, much faster. Um, this is true for like 56% of our countries, but, but this covers like for, for most of the large countries by economic size or population, um, the within-country convergence rate is lower in, in, in the 2010s than it was in the 1980s. Um, um, this is also true if you do the same convergence regressions, not at the level of provinces or states, but by cities. Uh, uh, it's also true for conditional convergence, not only, um, unconditional convergence. It's true if you exclude India and China. Um, the one sort of big challenge is accounting for regional price differences, so we are collecting data we now have. Have it for India, China, Britain, uh, and the US, and for all these four countries, our facts are quite robust. We are trying to get it for more countries and hopefully we'll, we'll be able to say more. You can cut this by any dimension that you want, but like the big fact that in the 2000s for most countries of the world, there is very little within country convergence kind of holds. So, for the main story, let me just tell you, tell this to you through the lens of like two countries. So what I'm plotting here is GDP per capita of India and China relative to the US over time, right? So what this graph is showing is the enormous catch-up that US, uh, India and China have had with the US. But the key thing is that when you look what is happening within these countries, so now I'm plotting GDP per capita for regions within these countries arranged by deciles, right? So the orange line is the top decile regions in India and China, and the, the blue lines are the bottom decile, uh, regions within India and China. They used to be very close to each other in the 1990s and over time, um, um, the divergence has just increased, right? What we think is this is related to is, is a growing concentration of tradable high-skilled services. So, if you just focus on the national employment in sectors that is concentrated in the top regions for manufacturing in India, that has declined. That's consistent with deindustrialization in India, right? The top regions are de-industrializing. China has relatively been stagnant. Um, services overall, India has been stagnant. China has increased a little bit, but if you look at professional and business services, which is this high skill services, uh, India has gone from something like 39% to 47%. So 47% of the national high skill services in India. In terms of employment is just concentrated, uh, uh, uh, in these five states and you see sort of similar patterns, um, um, kind of in China. So if I, if I look, if I take this to the entire sample and if I regress, uh, the internal convergence rate on the services employment shares of course with development, the internal convergence rate is falling, but what it is falling, uh, uh, starkly with is the growth of services within these countries. So ideally once we have the full. Data set with the professional business services. I want to replace the x axis not by broad services but just by uh professional and business services which we have done for 22 countries where we have the data and the thing holds, uh, uh, but we want to sort of um do it for the full, full sample. Here's the same thing in regression so the Y variable is the within country convergence rates. All regressions have country fixed effects, so we are tracking the evolution of within country convergence rate over time within a country. It's of course rela related to development. So as development progresses, as employment moves out of agriculture, the within-country convergence rate falls, but it's not so much related to manufacturing than it is, um, um, to the growth of services, um, within these, within these countries. Now, one valid question to ask is maybe with development. There is no regional, there is no sizable regional inequality left and so this might just be reflecting kind of the solo lesson that once your regional inequality is down a threshold, uh, uh, there is, there are no gaps left to close, uh, uh, that is not fully true either. So now on the Y axis we. Have just the coefficient of variation in GDP per capita between regions, um, within a country. Of course, with the development, this sort of fell sharply, but then it kind of, when services hit kind of the 30% mark, uh, uh, this kind of stabilized at around 18% or 17%. And since then, um, these gaps between GDP per capita between regions haven't, haven't been closing. Um, So here is where we should be a little bit careful. So it turns out that even if you look at the EU, uh, they don't have a, they don't have good data on, uh, labor force composition by sectors before 2000. It depends on the country. For example, Italy, you don't get this data, which is why. I don't want you to focus on anything before 2000s, but after 2000, so what this, uh, chart is plotting is the, um, employment in, in each sector in the top region versus the bottom region. So lesson number one, services is much more concentrated than manufacturing or agriculture, right? The second is the growth in concentration in services, which is the slope of this line. Um, is much more for services than manufacturing or agriculture. So, not only is it more agglomerated in, in, in services, the rate of increase, um, in concentration, um, is, is, is also much higher. Now, we can't do, we, in terms of testing what other factors might be involved, we can't do much because you don't get a lot of sort of time varying controls by region within a country, but we can look at cross-country relationships. So if you put a time fixed effect and compare two different countries, one which has a lot of internal convergence and the other that does not, and see what factors would explain this, and I'm only showing some factors here. So this is like a kitchen sink regression. It turns out that it's services, employment share and the growth of productivity in services that seems to be the leading predictor of why certain countries have high internal convergence, um, than others and in particular, uh, uh, growth of services associated with less, um, internal, internal convergence. So, I'm going to sort of To see, to see whether this, all these facts can be rationalized by a simple sort of structural change model. I'm gonna explain this in words. The model is not complicated at all, right? So you, you take the most canonical structural change model where structural change happens because of non-hommothetic preferences, right? And you put it in a standard model of economic geography. So there are multiple regions, people choose which region they want to locate in. And they produce stuff and they, and, and they consume. The, the only departure from standard models is going to be so in standard structural change models, you have, so this is the sectoral growth rate of TFP in agriculture and manufacturing. Uh, uh, uh, it's all exogenous and, and, uh, uh, uh. Uh, uh, as a standard in the structural change models, what we have done for services is we have put in an additional kick that comes from agglomeration and services. So, growth rate of services productivity is, is due to an exogenous factor, but also because there are more people located in a particular region, um, um, in services. So the calibration is preliminary and, and pretty standard. So in the interest of time, I'm not going to focus much, but just show you this. Um, um, the, the red dots with the dashed lines is what the, the, the, the data says about the evolution of within-country convergence. That very simple model just matched to the initial concentration of employment shares in different regions can pretty much predict the decline in the uh, the decline in the within country convergence rate over time, um, um, at least qualitatively. And we can, of course, use the model to see what role this agglomeration force is playing, right? Right. So, what role is, is, is, is, is this factor playing here? So in the model, if you do this decomposition where you choose like some agglomeration force that is calibrated to the data versus setting it to zero, it turns out that you still get some decline in beta convergence, even if there was no agglomeration. But then the effect just multiplies, uh, uh, if there is agglomeration in services. So if I, if I excluded the agglomeration factor from the model, there are two pieces of the data that I won't be able to match. One is the speed of this decline, like it would be much slower without the agglomeration kick, and the second is increase in concentration, uh, uh, of services, employment in, in, in, in particular regions. So, kind of the big picture message of our, of our, of our research is this. So, one is, of course, like putting out the data set, which, of course, I'm assuming would be useful for many of you. Uh, but also to say this, I mean, the World Bank had this fantastic report showing that future growth is going to be driven by services. And if that is true, and, and, uh, and, and assuming that it's all sort of professional tradable high skill services, then spatial inequality is going to be a free gift of it, um, um, that we should be, um, kind of cognizant about. So thank you. Amazing. Thank you so much, Micro, and you, uh, and you, uh, are great on time. So, uh, let's move right on to, uh, Marta for, uh, the second paper. Great. So thank you so much for, to the organizers for inviting me to present this project that's called the Global Race for Talent, Brain Drain, Knowledge Transfer and Growth. And in this paper, I studied the effects of migration on innovation and growth and the starting point is that uh inventors are very highly mobile individuals. 22 Great. So to illustrate this, I want to start with this example where Jean Calvignac here is a prolific French inventor that appears in my data from the European Patent Office. And this is an example of a real patent that he filed in 1991 and from this we learned that at this time Calvignac lived in the city of Lagos in France. But also most of his collaborators at this time, that is the other co-inventors that contributed to this technology were also based in France. Then in 1998, Calvin Jaak moved to the US. And we know this thanks to his later patents such as this one from 2004, which shows that first at this time he lived in North Carolina. Second, most of his collaborators at this time were also based in the US. But most interestingly, even after moving to the US, Kaviac continued to work with some inventors based in France, and this is not an isolated example. In fact, there is a systematic flow of inventors across countries. To give a sense of this using patent data between 2001 and 2010, I observed that about 6% of European national inventors live in the US in any given year on average, but only about 0.4% of American inventors live in Europe in any given year on average, and the difference between these two numbers, the net migration is what I refer to as the brain drain. It has been a cause of concern and policy interventions both on the European and the American side. So in this project, I first ask how does inventors migration affect migrants' productivity, their collaboration networks, and their knowledge spillovers on local inventors. And second, I study the positive implications of some real world policies that have been implemented to manage its migration flows. So I asked what is the role of tax and migration policy in shaping migration flows, the innovative capacity of the economy, productivity and output. And in this paper, I answered these questions in 3 steps. The first step is to bring a theoretical framework to a literature that had so far been mostly empirical. So I built a two country dynamic general equilibrium innovation based growth model that has two important new features. One is that inventors in this model are allowed to move across countries and to return to their country of origin. And second, inventors here accumulate human capital by learning from others, but this learning happens inside endogenous interaction networks that produce knowledge spillovers both within and across countries. Now the key takeaway from this model is that migration has ambiguous effects on innovation and output in the short and in the long run, and this is because it produces several forces that have opposite signs. And the two main forces that come out of this model are on the one side, the effect of the brain drain, as it is partly offset on the other side by the positive benefits for knowledge spillovers. Now, given this ambiguous theoretical result, the next question is how can we quantify these different forces. And this takes me to the 2nd part of the project where I build a micro-level data set of migrant inventors between Europe and the US using data from the European Patent Office. Now here one of the challenges is that there is very scarce availability of data that consistently track individuals across countries, which is why here I focus on Europe and the US but more generally the theoretical insight from this model apply more broadly not just to individuals that invent technologies but also individuals that adopt technologies and so this model could apply more generally not just to advanced economies but also to developing countries. Now, the key findings from my data are that first, after migration, migrants, both European and Americans tend to increase their patent applications by 42% per year on average. Next, what I think is even more interesting in this data is to look at what happens to the collaboration network of these migrants. And here I find that after migration, migrants actually continue to work with inventors in their country of origin, but less frequently as they work more often with investors and desti inventors at destination than other migrants. So then the last question is what happens to these inventors in the country of origin after somebody that they work with moves away. And here what I find in the last point is that the local inventors at Origin also increase their patent applications by about 15% per year after their collaborator moves away. And this is what I will interpret through the lens of my model as the effects of the knowledge spillovers channeled by migrants on the productivity of the local inventors at origin. Finally, the last step in this paper is to bring together the theory and the data by calibrating the parameters of this model to match his empirical result and finally to study counterfactual policy exercises that resemble real world policies implemented either by European countries or the US to address these migration flows. So let me start with a brief overview of this model. On the human capital side, here, the population is exogenously split between inventors that produce technologies or ideas and workers that produce the final good. On the production side, these technologies or ideas are used to improve the quality of intermediate goods that are then sold to the final good aggregator. So that overall innovation here drives aggregate productivity growth through the improvement in intermediate quality, as is common in the innovation-based growth models. Now, the novelties that I bring to this framework are the following. I consider a two country version of this economy with two countries, A and B. And the goal is to describe the endogenous evolution of TFP ABC in each of the two economies. At the heart of the process of growth are inventors. And inventors here produce innovation, but on top of that, they are allowed to move across countries and to return to their country of origin. And second, they meet others in the economy and they learn from them. And this interaction and learning process generates knowledge spillovers both within and across countries. Finally, in this model, I also allow for some additional exogenous technology diffusion across countries to capture other sources of ideas flows such as foreign direct investment or trade in intermediate goods. And finally, these countries also trade the final good and share a common exogenous interest rate. So let me give you some details on this process of migration for these inventors. Inventors are born in a given country in this model, and they have heterogeneity across two key dimensions. The first, as Ufuk mentioned in his presentation, is that inventors have heterogeneous talent that I call Z. But in this model inventors also have heterogeneous idiosyncratic opportunities in different countries, and this is what I call their foreign productivity differential epsilon. And together these two variables determine how productive an inventory is. So an inventor in every period produces a certain bundle of ideas or technology Q that is equal to their talent when they live in their home country. But this is equal to the talency plus the foreign productivity differential epsilon if the inventor moves abroad. Now this epsilon here is meant to capture idiosyncratic reason why a certain inventor could be more productive in one place versus another, such as their individual field expertise that could be a good match with a firm in the other country in this period. So the important point here is that you can see that inventors will choose to move when they have a high draw for these idiosyncratic opportunities in the other location. But also these opportunities epsilon are allowed to move over time and if they become low enough, the inventor will choose to return to the country of origin. So here migration decisions are endogenous to the idiosyncratic productivity opportunities of these inventors in different countries. The second important piece here to the right of these slides is that their return to innovation for these inventors are also proportional to the aggregate TFP ABC in each location. And this means that a 2nd reason for inventors to move here is also that they may want to go to the country with the highest TFP, the technology frontier, where they earn more from their innovations. Finally, the third point is that inventor silent Z also evolves over time thanks to this learning process. And this works in a way such that with some probability lambda an inventor meets another individual in the economy and the result of this meeting is that both individuals become more productive. However, a novelty here is that the probability of meeting different individuals is different for locals and for immigrants, and this is due to matter frictions. And this liter frictions means that a certain inventor that is born in a certain country C and lives in some other country D has different probabilities of meaning inventors that live in different places and come from different countries. The important point of this structure is that first of all, individuals have a 3rd reason to move because when they move they can change their interaction network and look for better learning opportunities. But also the other important point is that when individuals move with some probability these migrants can still interact with inventors in the country of origin and with this meaning they transfer knowledge that makes the locals of origin more productive. So this is the way in which the model captures the idea that the knowledge spillovers channeled by migrants make the local inventors at origin more productive. So as a summary of the effects of migration in this model, if we think about the migration flow from country A to country B, which is the frontier. What we find is that there are now more inventors in B who are also more productive when they move because they're positively selected, although some of these migrants may crowd out some of the local inventors, and on, on the other side in Country A, there are fewer inventors because of the brain drain, but the migrants transfer knowledge to the locals at origin, making them more productive. And on top of that, the Country A benefits from higher innovation through the front at the frontier also through exogenous technology diffusion. Which is why overall the net effect of migration in this economy is ambiguous. And this motivates the need for an empirical and quantitative analysis in this model. So to do that I focus on an application of this model where location A is Europe and location B is the US and again this is due to data availability, but what could think of a different application and calibration of this model where one country is the US and the other country is, for example, a developing economy. Now, I use my patent data to calibrate the parameters of this model. And to do that, I first show this fact that represents the collaboration network of local and migrant inventors. So what I'm doing here is to group the inventors in my data into 4 categories that are locals and migrants of European and American origin. And here I plot on the y axis the fraction of collaborators of these inventors that belong to each of the four categories. So for example, the first bar on the left here shows that when I look at European local inventors, in orange, this means that they mostly work with other local European inventors. However, the second bar looking at the European inventors that move into the US shows in green that these migrants are much more likely to work with American inventors as well as other migrants, and the similar pattern is true for American inventors. Now, this is helpful to think about the network frictions in my model, and it says that inventors are more likely to meet and interact with other inventors that either live in the same country or also they have their same nationality. Now the next step here is to illustrate the evolution of patenting activity both for the migrant inventors as well as for their local collaborators. So to do this, the strategy here is to match every migrant in my sample with one placebo or control inventor who is not a migrant and not a direct collaborator of a migrant, but this control inventor comes from the same country of origin, appears for the same first year in the sample, and also has the same cumulative patent stock by the time of migration as the migrant as a proxy of productivity. And finally, as the last step, I collecting the data, the full network of collaborators in the country of origin, both for the real migrants and for the control group. And here is what I find. In this first figure I'm plotting here on the X axis, times 0 is the time when the real migrants move away. And the Y axis is showing the mean of patent applications per year. The blue line is the dependent applications of the real migrants, and the red line is for the control group. So what we see here is that the key point is that after migration, migrants tend to increase their patent applications relative to the control group. The next question is what happens then to the local collaborators of these migrants in the country of origin. That's what I show in the second figure where the X axis still shows the time 0, the time when the real migrants move away, but now in the blue line I'm tracking the patent applications of the collaborators of the real migrants in the country of origin, whereas the red line are the patent applications of the collaborators of the control group. And the key message here is that also the local collaborators of the migrants at origin see an increase in their patent applications after the migrant moves away relative to the patent applications of the collaborators of the control group. Now this, this fact is very important to calibrate the parameters of my model. Why is that the case? That is because I can replicate these two event studies from a simulated sample of inventors in my model. And I calibrate my parameter in such a way that the model simulated sample in red produces the same magnitude as the data event studies that we've just seen that here is in green. And the key figure here is the one for the local inventors on the right because in the model, the reason why the local collaborators of the migrants at origin see an increase in their patent applications is thanks to the knowledge spillovers that the migrants abroad channel to these inventors back in the country of origin. Now in the last few minutes, I want to show some policy exercises. Today I'm going to talk about a tax on inventors' profits that is allowed to be heterogeneous across countries and yet in my my calibration is initially higher in Europe and lower in the US. And in the paper I talk more also about migration policy in the US. Now the policy exercise that I do here is to study a a tax cut in the European Union for American immigrants and European returned migrants, and this is motivated by the fact that several European countries have implemented policies of this flavor starting from the Netherlands, Denmarks, more recently also Italy and other countries with the goal of trying to revert this brain drain. In my model this means solving the transitional dynamics from the initial equilibrium where the tax rate is higher in Europe to the new equilibrium where the tax rate in Europe is lower for Americans and returned migrants. Now in the next few figures, I'm ploying the results of this policy, so the X axis is always going to be the number of years since the policy implementation. This first figure here on the y axis is plotting the stock of migrants as a fraction of domestic inventors. So what we see here in terms of allocation of talent is that the green line shows that the stock of American migrants in Europe increases and this is because Americans know that now if they go to Europe they'll have a lower tax rate, but at the same time the stock of European migrants in the US, the red line is decreasing, and this is because European migrants know now that if they return to Europe they will have this lower tax rate. And the dashed line here is showing that uh the magnitude of this policy is such that the level of net migration that is the dash line is going down to zero in the long run. Now, what is the, the effect of this reallocation of talent on innovation? That's what we see in this figure, where now the y axis is an innovation plotted as a percentage of GDP. And what we see is that this reallocation of inventors towards Europe implies that there is an increasing innovation in Europe and a decline in innovation in the US. Now the last question is, what is the impact of this change in innovation on output in the two economies? So in this last figure I'm plotting outputs in Europe and the US relative to the counterfactual equilibrium where there is no policy change. So what we see here in the green line is that output in the US is declining, and this is because innovation in the US is declining and the US is the innovation and technology frontier to start with. But more interestingly, here in the red line we see that output in Europe increases for a few decades, but eventually, and this is thanks to higher innovation in Europe, but eventually at some point output in Europe starts to decline. And to understand why we need to look at the table of the right, where I provided the composition of the different forces that I discussed in the theoretical section, quantify the two different time horizons. And the one that I want to highlight here is given by the knowledge spillovers that are at the center of this model. And what's happening here is that this policy by reducing net migration flows to zero, it also reduces knowledge spillovers across countries and eventually this has a sizable negative effect on output in Europe, but this is also a force that takes a long time to become effective because it operates on the distribution of human capital in the economy that is a slow moving object. So to summarize, this paper presents a framework to study migration decision, knowledge networks and innovation and some new empirical evidence to think about the magnitude of these different forces. More importantly to me, the key takeaway from these policy exercises are the following. First is that um policies that try to change migration flows and reduce the brain drain. Actually generate several forces on the economy that have opposite sides and so they even they may even end up in backfiring and reducing output in the long run. On the other hand, I also highlight that knowledge spillovers are very important. They have a sizable quantitative effect on output. So perhaps if anything rather than focusing on policies that affect migration flows, one could think about policies that foster these knowledge spillovers across countries. And finally, I think we still have some open questions, especially relevant for this audience. Due to data availability, this, this paper focuses on Europe and the US and we need to have more research to understand, for example, how sizable these knowledge spillovers are for other countries, especially the developing countries. Thank you. OK, great. Very, very nicely done and, and right on time, uh, as well. Uh, let's now have the 3rd, uh, paper. Alison, uh, 20 minutes. Um, great. Thank you for having, um, our paper as part of the session. Um, this paper is with Sam Asher, Paul Novisad, and Doug Galen, who you'll hear more from later on the long-run development impacts of canal irrigation in India. So the motivating question of this paper is one of the oldest in development economics. How do gains to agricultural productivity affect broader growth and structural transformation? This question naturally arises from the observation that um Lewis put in 1954 that economies in which agriculture is stagnant do not show uh industrial development. So there's been a lot of theory, um, and exploration of exactly why and how, uh, gains in agricultural productivity are translated into, uh, into industrial development. Um, and more recently, there's been an empirical literature trying to quantify this effect, uh, much of it looking at the green revolution and the introduction of high yielding seed varieties, um, through the 20th century. Um, and in both the theory and in the empirics, um, the relationship is, is complicated, and we want to understand more about exactly when, um, technical advancement in agriculture translates into growth in the non-farm sector as opposed to crowding out growth in the non-farm sector. So in this paper, we add what we think is a unique angle by looking at the generally equilibrium effects of, uh, a massive agricultural productivity shock at high spatial resolution. So that allows us to look at exactly how and where, um, structural change might take place in the long run. So we're looking at the canal network in India, which is very large. It runs for 300,000 kilometers and serves, uh, about 1/5 of all villages in India. What's interesting about canals as a, as a shock to agricultural productivity is that unlike other technical advancements like seeds or fertilizers or mechanization. Um, canals are fixed in space. So when you build a canal, you introduce a long-run differential, um, in productivity for places that get access to canal water versus those that don't. Um, they don't diffuse across space through time like many other technologies do. Um, these canals were mostly built, uh, through the 19th and 20th century, and the majority of the canals in our data set were completed over 40 years ago. So we're going to be looking at outcomes in, uh, the long run in the modern cross section from this, uh, canal network that was built to deliver water primarily during the dry season, the rubby growing season. Um, to allow farmers to have, uh, better, more production in a year. So you're getting more production per hectare per farmer, and they're more resilient to the annual weather seasonality. Um, so as I mentioned, the canals, uh, are an old technology. They were invested heavily in by the British Raj and then also by the early independent government in India. They were actually the dominant source of irrigation before the rise of groundwater in the last 40 years, and they're still a very significant, um, source of irrigation for Indian farmers. Just so you have a picture of what these canals look like. These are not the Suez Canal or the Erie Canal. They're not built for massive transport. You can see here there's a boy standing in the middle of this canal, so it's fairly shallow, and the water level actually fluctuates such that sometimes these canals are empty, um, and the water level will go up and down as, as the, as the season changes. So don't picture big transport canals. These are really built for, uh, irrigation purposes for agriculture. So our goal in this paper is to estimate the impact of these, uh, getting access to irrigation from these canals in India. And we're going to do that in 3 ways, estimating 3 different local average treatment effects. The first is the direct treatment. So, in villages that get access to canal water versus those that don't, what are the agricultural and non-agricultural outcomes we see? Then we'll look at spillovers from those directly treated, um, areas into other villages that are proximal. And lastly, we'll look at the timing of canal construction to understand regional urbanization. So our gains from canal irrigation being invested in the region, um, into urban centers. To give you an idea of where we're going, uh, to cut to the chase, we're going to show that this increased access to canal irrigation, we can actually measure an increase in canal irrigated agricultural land that leads to increased agricultural productivity, that's been sustained, um, into the modern day. And that increase in agricultural productivity leads to an increase in migration. So, the population density in irrigated areas drastically increases, but we don't see any long-run, um, uh, gains in consumption. So, for the general population, places that are irrigated versus not irrigated, uh, their consumption are about equal, except for landowners. So landowners, um, who get their land irrigated have sustained gains, um, in the modern data. And we see no effects on rural industrialization, but we do see regional urban growth. So basically, in the long run, um, the story we're going to try to show here is that productivity gains introduced by these canals are equilibrated by labor flows across space, and any structural change they, uh, create is concentrated as, as urban growth. So to walk you through our empirical strategy in more detail, just consider two, villages that are similar, except for their elevation. So we have one that's slightly higher, maybe 10 m higher in elevation than the other. When the canal is constructed through the area, because canals require gravity to deliver water to their endpoint, the canal will only be able to feed the village that lies below the canal. So this is going to introduce a discontinuity, um, that allows us to do a regression discontinuity design based on the elevation of the villages relative to the canal placement. So in our first stage, we're going to see that access to this canal water is going to give, um, these irrigated villages higher agricultural productivity, and we'll test for any other non-farm outcomes. We'll then look at spillovers into those, uh, villages that lie above the canal that weren't directly treated, but are proximal to this boost in agricultural productivity by comparing to a third set of distant villages that are much farther from the canal. And then lastly, we'll look at, uh, regional urbanization to see if these gains from productivity in the treated villages, uh, result in growth of urban population. The data we're going to use for this, um, I won't go into detail here, but it's all administrative, publicly available data from India between 2011 and 2013. The one thing I'll highlight is that we don't have, um, a, a direct measure of crop yields at the village level. So we're going to use a satellite proxy, which is the enhanced vegetation index. So that essentially gives us, uh, a measure of the green up in every village during every season. So that's what we're going to use for our productivity measure. If you want to know any more about this data, you can check out the Shrug open data product that's, um, made available through Development Data Lab. There was just a second version of this data platform released last week, which was very exciting. Um, there's a lot of documentation there and happy to talk about it later. All right, so for our first, um, uh, identification strategy, we have a standard regression discontinuity, uh, specification where our running variable is elevation of villages relative to the nearest canal. So what I'm showing here on the, in the map, you'll see each polygon is a village, and the villages colored purple lie, um, below the nearest canal and the villages colored orange lie above the nearest canal. The gray villages are far enough away from a canal that they're excluded from our sample. And just for this district, I'm plotting in, um, on the Y axis of that plot, you see, it's the share of agricultural land that's irrigated by canals. So this is just to, to show you the discontinuity we were describing, um, the orange villages have much lower levels of irrigation than the, uh, purple villages. So that's the discontinuity that we want to exploit in this, um, design. So looking at a result with our full sample, um, here's this, uh, regression discontinuity, again with canal, um, irrigation as the outcome variable. And we see on the right-hand side, um, those are the villages that are below the canal that get access to irrigation, and on the left-hand side, those are above the canal that do not get access. So we see a jump at this boundary, um, and I'm going to show you a bunch more results here that just show you that jump at the boundary is on coefficient plots, but that's what we're measuring, um, with the, the, uh, design. So here is a coefficient plot showing irrigation outcomes. So the second one you'll see there is the same one we just looked at. It's percent of your agricultural area that is irrigated by canals. And we're plotting this as a normalized treatment effect. So all of these, uh, coefficients are in terms of standard deviation of their outcome variable, so that we can look, compare them all together. And we see that the villages that lie below the canals really do have, um, increased irrigation that's driven by canals. And we notably don't see, um, too well, which is groundwater irrigation or any other source of irrigation. There's no difference, um, at the boundary of in our regression discontinuity. And these irrigation outcomes lead to agricultural productivity gains. So villages below the canal have a larger share of, uh, their area that's cultivated. And they also have higher productivity, and that's going to be concentrated in the rubi, uh, season, the dry season, versus the Kari harvest season. So, that's what we would expect because these canals really were designed to deliver water during that dry season, and that's where we're detecting, uh, productivity gains. And these villages are also more likely to grow water-intensive crops. So we're pretty confident that this illustrates a first stage of, of, uh, what we're hypothesizing here is that the, um, canals really are bringing additional irrigation, which is creating long-run, uh, sustained differences in agricultural productivity. So the big question is, do these translate into any local non-farm economic outcomes? And the answer is, uh, there is one major one which is population density. So areas that are irrigated by canals have much higher population densities, but notably don't have, um. Any increase in non-farm employment. So we don't see an increase in industrialization, so no increase in manufacturing or services or even agro-processing employment directly in the villages that are serviced by the canals. Um, this increase in population density, I won't show the details here, but we have suggestive evidence that this really is driven by migration as opposed to changes in fertility or mortality patterns. Um, just to dig in on one result we showed here is there's, there's also no difference in consumption, but if we desegregate that by landowners and landless, so in the bottom graph here, you'll see, um, landowning households on the top and landless households on the bottom. So if you compare landowning households that have access to irrigation versus those that don't, they have, uh, higher consumption in the modern day, while landless households have about equal consumption across that, that discontinuity. And on the top, we just break it out by, um, the percentile of, uh, the size of landholdings. So the wealthiest households that own the largest land get the best returns. So the gains from this, uh, the increase in agricultural productivity are accruing to the fixed factor. So landowners are benefiting while landless laborers, um, are the, the gains from the productivity increase are, are eaten up by the population density increase. Um, these results are, are robust to several different specifications and checks. I won't go into detail for the sake of time, but happy to talk more about these later. Um, so the first question we have is if you have, you know, an increase in agricultural productivity in these local villages, there may be many reasons why this would spill over into other villages in the area. Um, the first could just be that as these villages are, are taking water from the canals, that allows the water table to run. And that groundwater recharge might lead to positive agricultural effects in nearby villages that don't get access to the canal water. Also, labor and goods could be flowing between these, um, villages and we might see, uh, a null result in the regression discontinuity. That just means the entire area proximal to a canal may be, uh, experiencing, um, economic gains. So we're going to test this with a regression where we hold out the above canal tree, uh, group. So those are the villages that are near the canal, but don't get treated by, um, getting access to the canal. And we'll compare those again to the villages that are below the canal and then to a third group of distant villages that are more than 15 kilometers away from the canal, and we're confident are not affected by the introduction of the canal. To create that, um, comparison group, we use an entropy balance weighting so that we ensure that characteristics, uh, like these geographic fundamentals are matching. So we try to match the, the distant village group to the villages that are treated by the canals. Um, now, these are a little bit hard to, hard to see, but I just want you to focus on the right-hand side. That's the coefficient that's estimating the spillover effect. So it's directly comparing the above, uh, villages that lie above the canal, but proximal to the canal to distant canals. And we mostly see, No spillovers in agricultural and irrigation outcomes. So there's no difference in agricultural productivity or percent of the land that's irrigated. There's a small spillover where villages near the canal are more likely to, um, grow water-intensive crops. And moving on to non-agricultural outcomes, again, we see mostly no spillover. So our, our villages that are near the canal look very similar to the villages that are far away, except for population density. So there's a small increase in population density, um, in the areas right near the canal, which mirrors the influx in population density we saw, um, from the regression discontinuity. But the, the major takeaway here is we don't see spillovers in rural manufacturing. So there's no increase in non-farm employment, in service sector employment, or manufacturing employment in areas that are in, that are proximal to the canal but not treated. So that leaves us to conclude that we don't have much evidence of rural industrialization. So our final, uh, set of results is going to look at regional urban growth. So our, our first two strategies really relied on the spatial distribution of the canal network and the villages, both the distance and the elevation and how those are all related. But we know that the gains in agricultural productivity might be invested into urban growth in a way that's totally orthogonal to that spatial relationship. So we could see over time as canals are, um, constructed in the, uh, in the region, that those gains are invested into town growth. So what we do is we have a panel of towns. So we have every town from the 2011 population census, and we have the population going back every decade until 2001. And so for each town, we draw a catchment area of a 20 kilometer radius around the town, and we look at the area. So that's what's in yellow, and then in blue, we have the area serviced by canals, um, over time as, as those change. So for every town, we can define a share of the area that's treated by canals and look at that change over time in a difference in difference analysis. So in this, uh, first result, we see that time 0 is when a town gets treated by canals in the catchment area surrounding it, and we see a significant increase in urban population. Um, for the periods after the town received access to canal water in those outlying areas. Um, we can also look at an outcome that is town appearance by size. So we can threshold town size and say, as soon as this town hits 5000 or 10,000 or, or 50,000, so at what, at what part of the size distribution is town growth sensitive to the, uh, canal construction? And what's important here is that we see it's concentrated in smaller towns. So, the smaller town sizes are more sensitive to, um, increases in canal irrigation in the areas surrounding it. So this tells us that small regional towns that might be very reliant on agro-processing or trade from agricultural goods are sensitive to growth when there's more productivity around them, while large cities uh that might be more diversified aren't going to be as affected by, by, um, the canal construction. So, kind of in summary, what we've seen from both the, uh, the regression, discontinuity and spillover analysis as well as this, uh, town panel of urban population data, is that there are large movements of people both from into rural areas that are treated by canals and into urban areas that are in the region. So the first question you might have is how big are these population movements. So if we take these coefficients and do a back of the envelope calculation, uh, we see that the effect is about. 5 million people living in urban areas and 29 million people living in rural areas, um, due to this canal network. So just for an order of magnitude, the most, um, significant migration event in Indian history is obviously the, the partition, which displaced about 17 million people. So we're talking about a large scale movement of people, um, obviously over a long period of time, but it's quite economically significant, um, to think about the spatial, uh, mobility of people, uh, just from these rural infrastructure projects. So to conclude, um, we'd see that these canals introduce a long-run difference in agricultural productivity. Again, these were built, um, through the 19th and 20th century, and we're now looking at the long-run effects in the modern cross section, and we still see sustained differences in agricultural productivity. Um, but those differences mostly were equilibrated by movements of labor, um, across space rather than between sectors in a highly localized way. So, the structural change that we can detect is not directly in rural areas, but rather in, in regional urban centers, um, and any sustained living standard changes we see are concentrated among landowners, uh, as opposed to landless laborers. So the implications for these findings are that, um, you know, generally in the long run, development entails substantial movement of people across space. Um, and as we rural infrastructure projects might not lead to the kind of rural industrialization. But rather relieving barriers to uh mobility and allowing people to move to cities might be an effective way to bring about structural change. Um, and one other thing I'll add is that, you know, this paper we, we deal with uh a shock to agricultural productivity that's driven by agricultural technology, but we also know agricultural productivity will change in the future, especially with, with climate change. Changing the patterns of what places are most productive, um, and which are less productive. So we can expect other shocks to agricultural productivity to inspire similar movements of people, um, across space. So it's important, uh, not definitely not the first paper to point to, to mobility due to climate change, but, um, to just in line with those other, other findings. So, um, that's all, and, uh, thank you very much. OK, that was, that was awesome as well. Thanks very much, uh, Allison. And so Chris Papageorge's going to, uh, discuss the three papers. Alright, thanks everyone. I'd like to thank the organizers for um the generous invitation to be part of this great event and for for taking a look at these three fantastic papers. I can really not do justice to, you know, the depth of these papers in 15 minutes, so I would be sending bilateral detailed comments to the authors and what I'll be doing here is just, just, uh, present to you what are my top highlights from my readings of these papers. I'd like to spend a few minutes in the end of zooming out and looking at country convergence and some of my own thoughts about some of the emerging challenges uh to developing economies that I see. Um, so let's start with, uh, Sumiro's, uh, paper. Which is a fantastic paper uh on specifically on the data side because it fills a gap, a major gap in the literature we have done work on convergence looking um at country level where the unit of information is uh the country a lot of work there. We have also some work that is at the individual level but we have very little at the regional level and this paper uh exactly uh fills that gap so the paper is very novel because it it constructs 600 regional income level data. Um, and it spans 34 countries for the period 1980 to 2015, so for those of you who have done work with data, this is an incredible achievement. Um, in the paper you can see how the authors are homogenizing, uh, different regions, and we are talking about Tanzania, Peru, and the US, very difficult, uh, task. Then, as, uh, Schmiro have shown they have used a, a very standard spatial structural transformation model, uh, to explain some of the, uh, main findings. Really the key finding of the paper is that there is a sharp slowdown, uh, within in within country regional convergence. Uh, and, and then they are using their model to provide a mechanism of why that's why they observe this, this fact, and the way they describe it then in the model is that this is basically due to the transition to services and uh and the economic activity being specially concentrated which results in this decline in regional convergence. Uh, so I have a few comments. Definitely the first one is on data, um, you know, uh, the devil is in the details on these constructions, and, uh, I have seen that the authors relied quite a bit on interpolations, and I've never been a fan of, uh, interpolating data. There is a lot of work. Uh, for example, from Angus Deaton and others that show that when we finally figure out actual data and we replace the interpolated data, many of the results are overturned. So I would say in my bilateral comments, uh, Schmitra, basically I guide you to some new, uh, now casting machine learning, uh, uh, empirics where basically they take into account all of the information that you have out there. To feel uh this data, basically the signal to noise that you will be obtaining from, uh, what I suggest I think is is higher than what you have right now. Um, the second issue that I wanted to bring up is, um, sort of, you know, I take a little bit of an issue of how quickly the authors come to the conclusion about country convergence and that the evidence is that we have actually country convergence, um, I have done quite a bit of work. I have a gel paper in 2020 with Paul Johnson. And we had looked at, we looked at the the evidence very carefully, um, and what we came out with is that the jury is still out so basically the way we explain some recent papers which focus this convergence because they look at shorter periods we call these episodes of convergence if you'd like I'll relate this comment actually to my final points about kind of zooming out and looking at uh cross country uh convergence. It's a fantastic paper let's move on to the next paper next paper takes us somewhere else on growth the one of the key production functions, the human capital specifically on talent and uh innovation migration basically the paper does a fantastic job in offering a new model of brain train brain gain it's a very complicated question. Um, the, the paper, um, complements the model with fantastic micro level data. Um, they look at basically patterns in the EU US migrant, uh, corridor. And finally they have they don't stop with positive analysis they do also some policy experiments, some key policy experiments one tax cuts for uh the migrant innovators and the second in increasing the cap of H-1B visas in the United States. So main findings, and I'll stick only to the policy here is not surprising the tax cuts have positive effects in the short run, negative effects in the long run. The increase in the cap of H-1B visas is good for everyone, the US, the, the European Union, the global growth. So two main comments I think, uh, Martha has already picked up on the first one. Um, you know, this is a fantastic tool, a framework, a template, and we look forward to having the published version and taking this and applying it to developing economies like the corridor being EU, I don't know, Africa or Asia. Uh, Europe, uh, so those would be very exciting ways by which we can bring in developing economies in this. I understand the reason why they went to EU US for, for, for data availability, but I think this is a very huge opportunity. Have uh to go in different corridors with developing economies then of course we can go further we can look at refugee uh innovators we can also you look at more specialized innovators uh with all the new reality of AI. The second, uh, comment I have is for to get some clarity from the authors, uh, to get them to think a little bit further on the assumptions made is a complicated model and assumptions are very important. One that I have a little bit difficulty understanding. Is whether that assumption made about the inventor's experience a productivity boost when they are moving abroad kind of pinning down the results and and whether uh I'm interpreting this correctly, you know, basically that's the conclusion of the model and policy implications depend on that assumption, something that we can discuss further in the Q&A, but again a fantastic uh uh paper Martha so finally we come to the last paper. Uh, Allison's paper, uh, on, uh, the process by which we grow the structural transformation, especially early stages on agricultural productivity, a fantastic paper, you know, very rich, uh, it took me forever to read, it's like 78 pages, so, uh, but, uh, but, uh, wonderful, so it careful, carefully documents how investment in irrigation using the India example, the canal example. Um, results in this positive return on productivity, but to me the most exciting part of the paper bear is the disaggregation, right? So you go from looking at the results at impact, which is the irrigated villages. You know then you have the connection with rural areas and then you go to regional urban economies and then you have possibly the global economy. This is a fantastic way of mapping out the impact and the ripple. Effects the total economy so I, I for me that was the most exciting part of the of the paper and for for that to happen of course the authors use different empirical methods, uh, a multi-sector location uh model by you know I think motivated by boosters and and so forth um so the findings are clear for me the biggest finding is on population movement. Uh, being so central to productivity, uh, gains and, uh, the key mechanisms for structural transformation, a lot of papers have been talking about this, but this paper documents, uh, quantitatively that impact. Uh, the numbers I had was 9 million, uh, drawn to cities and, uh, 32 million in rural areas. I think the numbers changed a little bit now is 5 and 29, but still, uh, massive, uh, impacts. Um, on my main comments is for future work mainly, uh, one, I understand the limitation of the paper to draw implications, um, at the aggregate level, and, uh, I understand that the limitation comes from measuring labor flows directly. I was wondering there and this is what I'm sending in my bilateral, uh, comments whether. You, you could do surveys to fill those gaps and we have come a long way on this. The World Bank is an expert on this um the fund is catching up. I think actually this surveys can be done and can be done effectively and go that extra distance to say something about aggregate effects, uh, which now I think you you are you come a little short. Um, the second I think is something that Alison picked up at the very end of the presentation which is exactly these shocks how do major shocks impact agricultural productivity, both economic but also climactic, and uh I'll come to this in my discussion in a minute. OK, so I'll spent the last 5 minutes I have on kind of zooming out and talk about um country convergence and before I get to my uh to the challenges that I worry most about uh facing developing economies I'll start with the current context. Most of the papers that have been written and trying to calculate country convergence stop at 2019, 2018, even before that 2015. Uh, a lot has happened and not in the right direction for developing economies. Uh, some people can call it transitional. There are all these shocks, the pandemic shock, Ukraine war, um, of financing tightness in the, in the global markets. But we also know that, you know, uh, impacts that seem transitory or are transitory for emerging markets and advanced economies are permanent for developing economies that, that worries me. But also this is in the back of what we think is a shift in the global economy for many decades we have been thinking about convergence when the economy, the global economy was doing fantastically maybe because of China and other factors but now we are getting into an era where the projections from the World Economic Outlook elsewhere show a truly a real global slowdown mainly because of China so. You know there is this the the so the initial condition for many of the developing economies given the shocks and given the new phase of, uh, global slowdown in growth is not very promising but this is not really what worries me mostly for developing economies are these emerging challenges and here I I'd like to talk about 3 and I have to move on here. OK this is it um climate AI and fragmentation you all know about climate, uh we have seen it in the fund uh how many of the countries come to us for financial assistance because of the shocks this demand is increasing at an increasing rate, um, so you know you wonder for example Alison how your work would change when you take into account this climatic changes both increasing, you know. Uh, frequency, intensity, duration of all these, uh, major shocks we had floods in India and so forth. I think, I think that that would be a very nice natural experiment for you guys. Um, second is AI, um, conceptual frameworks show that. You know robots, especially the low end robots, substitute for low skilled workers, and many of you know worked in developing economies there was this demographic dividend dream in Africa, right, that in 20 years, you know, we'd have the youngest population. You know this new robots that keep work basically in China uh where it should have been in Africa by real people it's something that worries me quite a bit so. Uh, Martha, I think it is very interesting to think about your work with AI migrants, of course, for your, uh, work is EU and US, but I mean you can extend it for developing economies as well. Finally, fragmentation is, I think something that we don't take very seriously yet, but we see it happening. There will be a trade war between China and the US. And where does that leave our discussion of uh country convergence? Should we be thinking about club convergence going back to that world um uh so Shimito I mean that actually this is even more important for you when you consider regional uh convergence. Let me stop here. Thank you. OK, terrific thank you Chris. Um, so we're gonna open it up, uh, in a moment for, um, some Q&A, um, and I'll, uh, give each, uh, each presenter an opportunity to address Chris's comments and the questions that come up. Let me, let me just, uh, put. One or two, general issues, uh, that came to my mind listening to the three excellent presentations. Um, Sumitra's central finding about, uh, slowing, uh, regional convergence, um, is that a policy problem and what, what are the policy remedies, uh, in your opinion if it is a policy problem? Uh, second, for, um, uh, for Marta, um. You have very strong results about short run uh gains to output and, uh, longer run costs again, sort of from a, from a political economy perspective, would you use this kind of a model, uh, to advocate certain policies whether on uh visa reform or tax reform? Uh, how, how do you think, uh, you would sell these very long run results from a, from a political economy perspective, uh, and, uh, third, Alison, um. Your your results about, uh, the gains accruing uh to the fixed factor, uh, entirely, uh, and the fact that, um, uh, you know, the non-farm sector doesn't really, uh, doesn't really benefit at all, um, uh, what, what policy remedies, uh, do you see as, uh, potential levers to, uh, to address those, um, uh, those issues? So let me, let me turn it over, uh, now to questions and I'll, I'll, I'll turn back to the speakers. Uh, do people want to, um, uh, intervene at this point? Yeah. Uh, thanks, Jonathan. I'm Art Cray at the World Bank. Um, you, Jonathan, you asked the question I was going to ask Jomiro, so I'll skip that one, but I also have questions for Marta and for Alison. So for Marta, two quick questions on the paper. The one is that, um, I think in the real world a lot of migration of inventors is because they already know who their collaborators are in the other country and they move to where they are and I wonder to what extent that changes the mechanisms and the model. The second question has to do with the policy experiments that. You look at, so you showed us a policy experiment in which one part of the world taxes innovation and so in the end it's, you know, not that surprising that there's negative effects in the long term. It'd be interesting to see what happens if you say had fully integrated patent protection between the US and the EU. You know, how, how much more positive would a policy like that be. And for Alison, two questions. One is just a little one, which is why don't people pump uphill out of canals if there's such a high return to doing so? The technology obviously is there because of all the groundwater pumping in India. The second question is, can you say more about the history of the placement of canals because I was thinking about this in the sort of slightly cynical way and you think about, you know, these are big public works, uh, projects when they were built, they were probably built under the influence of landowners and it's not very surprising that they lined up in places that benefit the landowners. Is that oversimplifying? Is there something to that history that matters for the findings today? Thanks, Art. And, and I apologize for not recognizing your face. The light completely blinds me from where I'm, uh, where I'm sitting. Other, other questions? Are there any, uh, questions in the chat that are, uh, that are bubbling up? OK, so, um, why don't we, uh, give, uh, each speaker a chance to, uh, to respond, Sumitra, uh, thanks, Chris, uh, for all your kind and generous comments. Um, both points sort of well taken. Um, I'll just say that, I mean, we have to interpolate one way or the other to look at the trends. Um, the validation exercise that we did with the interpolation was that we only interpolated the within country data. And then we aggregated it up to see whether or not we are matching the aggregate GDP that you get from the Pen World table of the World Bank. But if there are ways to kind of improve it further, of course, um, um, I would look for the, um, um, detailed comments. Um, the cross-country things I I'm aware of the debate between sort of your paper, and now there's a paper by Michael Kramer that is claiming convergence and the Sandifer Subramani and Patel paper. Uh, I thought it got resolved one way, but maybe not, and my reading is incorrect. The thing is, it's not even central to our paper. Like, I can rewrite the framing and say there has been a lot of growth in the developing world and, and, and just focus on within-country convergence. Um, um, but again, would look, um, forward to talking more. Um, on the policy side, um, I mean, let me even sort of extend that question on the policy and politics, um, um, problem of this. Um, I think that would depend on country by country, like, so I, I mean the issue is that the, the, the regions that are getting left behind. Uh, uh, uh, uh, what is happening to kind of people there and are people able to migrate to, to regions with opportunities? And of course, like, when you focus on countries like India and markets here, there are migration restrictions. So, Bihar and UP, for example, tend to be the most unproductive states, but also the states with the most population and population density. Um, um, and that's kind of part of the policy challenge. But if you look at the US, we are like, I guess, facing a completely different story here where the Midwest, which is kind of falling behind, um, there are not too many people here, but they do have an equal representation in terms of the politics. And, and so this, this unequal sort of spatial growth in countries depending on political structures will. Um, um, kind of have impact on political outcomes as well. So, of course, we don't get into those issues in the paper, but that's kind of what we have at the back of our minds. Thank you, um, Mara. Yeah thank you so much for the comments and the question and uh to start with uh from the comments by Chris I would I would say that it is certainly an important assumption in this model that migrants tend to increase their productivity when they move and this is. Because these are migrants motivated by economic reasons, so they move when they have good opportunities in the model and that's sort of like reflects what I see in the data that on average these migrants tend to find more patterns after they move but of course one could think of different settings where. People including inventors move for different reasons like refugees or as we were talking about the war in Ukraine that caused a huge reallocation of people or as we also talked about uh migration motivated by climate change and so in this case it might be it might not be necessarily the case that people actually increase productivity when they move so 11 channel of this one looks at the effects that come sort of like from the reallocation of people. But on the other side, there is also this important channel that comes from the knowledge spillovers from the destination country back to the origin country. And here again it really depends on the nature of migration. This type of econo economic migrants that move for economic opportunities are able to maintain a network in the country of origin, but when we think about refugees, it might be the case that even if they move to a different country and they may become more productive, it might be very difficult for them to keep a network, a professional network in their home country. And so this connects to the questions about policy where what I've done so far in this work is to do a positive evaluation of some policies that these countries have implemented to really answer the question about policy recommendations more deeply. I think this would require sort of solving a planning problem to think about what the optimal policies are and so in that sense I think this is a great comment to think more carefully about other types of policy completely for example instead of tax. Where tax cuts, what about, uh, a fully integrated patent protection or what about temporary migration programs that could be, could foster these knowledge spilovers and this is, this is again something that requires even uh uh further analysis of these, this type of models and, and, and related I think to to the other question about the fact that very often people move when they already know some people in their destination country. I think this is, this is true. And this is something that I've seen the data where actually most of these migrants move within foreign branches of the same multinational that they already work for but for the um main channels of this model this is fine uh for for. For simplicity in my model I use sort of random matching of people in the economy, but one can think about uh directed matching where people know who they want to meet and if anything I think this would make the knowledge spilovers in the model even stronger. Thank you. Thank you, Marta and Allison. Uh, great, thank you all for those questions. Um, I guess on the first point about kind of the effects of climate on agricultural productivity, I think one interesting, um, thing to think about in our paper is the result that the of the town growth, the regional town growth as a result of the increase in agricultural productivity, and I think an interesting implication is that not only are the outlying areas that might be experiencing drought. Um, going to be affected, but that these towns that might be supported by, um, the productivity and the demand of people in rural areas. So thinking about I guess the, the policy, um, ramifications of, of needing of people needing to move not just from extreme weather events, uh, disasters but also just changing, uh, economic realities of what what populations are able to be supported in what areas, um. On the, the question about policy, uh, ramifications, um, so there's some other work by our co-authors Paul Novasad and Sam Asher looking at, uh, rural roads also in India and looking at the places that get access to roads, and they find that similarly, um, the roads essentially facilitate people to migrate to other areas to search for other opportunity. So in a lot of these, um, areas, we're seeing these infrastructure projects just leading to migration. So I think, um, understanding more of those, uh, incentives to migrate, like a lot of the work in Bangladesh, um, running trials to incentivize people to move during the dry seasons, or giving people vouchers to move to cities and understanding what are the impediments to, uh, migration would be interesting to, to understand more about, um, especially making the destinations more appealing to actually migrate to and not, um, deter people from moving to cities and, and living in, in bad conditions. Um, and then lastly, I'll just address the questions. I, I should know more about the answer of why don't people pump uphill from canals. Um, I imagine part of it has to do with the fact that these are, many of these canals were built before that was a viable option. And I wonder how the, um, kind of gains accrue to landowners and how, what capital is there to invest in those, um, you know, how, how expensive is it to, to pump that water uphill and who's choosing to make that investment or not. Um, but it's a good question. We should probably, I should probably have a better answer for that. Um, and then lastly, just on the, the placement of canals, we've we've definitely thought a lot about how, how that could be endogenous. Um, there's actually another, uh, regression discontinuity strategy we use, um, where we look at the, the, the government puts out polygons drawn around the canals called command areas, which is the area that the canal should service. And we do a spatial discontinuity looking at places just inside the boundary versus. Outside, but we actually thought those might be more likely to be, um, influenced by politics or, or whoever is drawing those maps. So we think that the elevation strategy is, um, more dependent on geographic fundamentals rather than maybe other incentives. Um, and then we also do a test where we look at the longest and straightest canals. Um, this test has been done with some road construction, um, Uh, as well to say if they needed to get water from point A to point B, which places were treated along the way, and especially when you're limited geographically and, and where the canal can go to get from A to B, it's more likely that those are, uh, treated kind of by circumstance as opposed to endogenous to interest preconstruction. Um, yeah. OK, thank you to Shamitro, to Martha, to Alison, and to Chris.
showAllTranscripts
no
duration
PT1H24M54S
scene7File
worldbank/ABCDE23_Session_2
scene7Domain
https://worldbank.scene7.com/
scene7FileAvs
worldbank/ABCDE23_Session_2-AVS
title
ABCDE23 Session 2
description
ABCDE23 Session 2
showTimestampAndTranscript
yes
col-xs-12
col-sm-12
col-md-3
col-lg-3
col-xs-12
col-sm-12
col-md-7
col-lg-7
  • add-style
  • lp-body-content
lp-heading-top-medium
lp-heading-bottom-medium
col-xs-12
col-sm-12
col-md-2
col-lg-2