Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
========================================================================

# Baby Busts and Growth Booms : Demographic Change and the Macroeconomy

Authors: Daron Acemoglu, David Autor, Keelan Beirne, Andrew Scott  
Discussant: None  
Video: https://www.youtube.com/watch?v=w6ObGDwelyg&t=5435s


## Talk (01:30:35 – 01:56:01)

[01:30:35] >> I have that it's u half past so we can go ahead and get started when you're ready. So, thank you. the the next paper is the the schedule that you all have baby bus and growth booms demographic  
[01:30:48] change in the macroeconomy. So thank you Keon and thank you Andrew for um engaging in the chat if you can during during the talk.  
[01:30:56] >> Yeah. Uh thank you to the organizers for putting together such a such a great um schedule and thanks for including our paper. I'm excited to present. Uh this is joint work with my illustrious co-authors Rowan Asamoglu, David Aer and  
[01:31:09] Andrew Scott who's here and uh looking forward to hearing everyone's comments and thoughts and if there's any clarifying questions please just don't don't hesitate to stop. Okay, so it's going to come as no surprise to this audience, but uh we're in the midst of a  
[01:31:24] really unprecedented demographic transition. And this plot that I'm showing here shows you the dynamics since 1950 of the birth rate. birth rate is measured here as births per 100 people. Okay? And I'm showing you this both aggregated across different  
[01:31:38] continents and also for a select few countries. And two general trends that I want to highlight in this graph is first that this this transition has been going on for quite a while. People are beginning to get worried about it now, policy makers and academics alike. But  
[01:31:53] this has been going on since for the last 80 years now. This these these uh these trends take a little bit of time to to affect the economy. But the the trend is is quite old. The second thing to note is that the timing of this  
[01:32:06] transition is very heterogeneous. So even if we look at the period between say 1950 and 1970, we see there's a large amount of variation in birth rates both across different regions of the globe and also within a different region  
[01:32:20] or within a different country. Say let's pick South Korea here. There's quite a bit of variation over time. Okay. So the point I think to make is that we are all worried about the future demographic transition but there's actually quite a bit of past experience in dem  
[01:32:33] demographic transitions in the data. Okay. So the defining feature of the future demographic transition is that birth rates will be much lower. So and what do lower birth rates lead to? They lead to slower population growth and an  
[01:32:47] aging of the workforce. Okay. Both of these channels at least has been commonly thought of to uh slow growth and reduce GDP. So there's a lot of international organizations who are worried about this. In the paper we quote the IMF, World Bank and also you  
[01:33:02] know popular news articles like the New York Times and the Atlantic and it seems everybody's a bit worried about this and I think the logic is you know it's pretty reasonable. So for example if you think about an semi-indogenous growth model you know growth comes from ideas  
[01:33:16] are non-rival. A larger population means more ideas, more growth. And similarly for aging, if we think that young people have a higher propensity to innovate or uh you know start a business while a patent, then we might think that an older population generally also leads to  
[01:33:30] slower growth. Okay. The question that we're going to ask in the data given what we saw in the previous slide is that what does the data say from this past experience of demographic transitions that we have observed? And what we're going to find is somewhat surprising I think to a lot of people  
[01:33:43] but also very reassuring. Not only are these lower birth rates not a disaster, but actually they're going to increase incomes and increase incomes by quite a bit. And for all the analysis that we're going to do, we're going to run regressions both across countries and across commuting zones within the US.  
[01:33:58] And we're going to find very very similar results. Okay? So we're going to do that wherever the data allows us to do those uh two parallel analyses. Okay.  
[01:34:07] Now this is a a quite surprising finding. How can this happen? Our story is one of uh techn induced technological responses in particular smaller a scarcity of workers scarcity of labor  
[01:34:22] creates upward pressure on wages and that induces um firms to choose to substitute towards technology. Okay. So it induces firms to invest in more technology. Um this is I think  
[01:34:35] beautifully traced out by Deron's 2010 paper where he shows in general if workers and technology are substitutes then increases in wages due to labor scarcity will naturally uh lead firms to increase their investment in technology.  
[01:34:50] And we're going to show evidence of this both across countries and commuting zones. I mean by uh by growth accounting if GDP is rising while the workforce is shrinking then it must be the case that either investment or TFP should rise.  
[01:35:02] What we're going to find is that actually both rise and actually the capital stock even rises a little bit despite the fact that the number of people who are using that capital stock is now much lower. Um we're going to find there's naturally laborsaving patenting. So we're going to say there  
[01:35:17] are certain types of patents that we can Xanti say look like labor saving patents for example automation patents and we're going to see that uh the share of patents in these laborsaving categories does indeed rise when labor becomes more scarce. What we're also going to see is  
[01:35:30] that labor intensive sectors contract. I mean it's kind of as you would expect when labor is more scarce labor intensive sectors contract. That means that capital intensive sectors rise and also we find that R&D intensive sectors  
[01:35:44] expand even more strongly than capital intensive sectors. Okay. Now, of course, there are some other channels that could generate um an effect or uh generate a negative relationship between birth rates and GDP growth such as, you know,  
[01:35:59] the quality of education that we just saw, female labor force participation or something to do with uh availability of children to work in agriculture.  
[01:36:09] I have undersorted here just to save space in the slides, but what I really want to say is that there is evidence of these channels. these channels are real but they're just insufficient in quantitative magnitude to explain the results that we have. So we we basically run the same regressions with these as  
[01:36:22] an outcome and we don't see the consistent kind of patterns in the data that we do see when we when we look at growth. Okay.  
[01:36:30] Finally, as I said on the last slide, um there's two dimensions to uh or two smaller birth rates. they lead to both an increase in population growth and an  
[01:36:43] increase in aging. Okay. And so with just birth rates, we cannot separately identify the these two channels. And so in the last part of the paper, what we tried to do is we used some different variation which is debts from the Second World War to try and tease out the  
[01:36:58] different effects of population growth and aging. And what we find is that the scarcity of younger workers is key to stimulating technical change. Okay, any questions so far?  
[01:37:11] Okay, I'll keep going. Okay, just as a road map of what I'm going to do. So, uh the first part of the talk I'm going to walk you through in detail. This is the most important. How do we estimate the response of incomes to falling birth rates and next I'm going to show you as  
[01:37:26] I promised how we think that this can be explained by technological responses and the evidence we have for that in the data. And then finally, time permitting, I'm going to try and show you how we can distinguish uh population versus the effect of population size versus the  
[01:37:39] effect of age structure using debts in the second world war. Okay, very good.  
[01:37:46] So the model that we have in mind is um one in which GDP growth is related to the size of the workforce um with a constant elasticity. So on the left hand side here we have uh uh the growth in  
[01:38:01] say GDP per worker and on the right hand side we have the size of the labor force just the entire labor force. We can generalize this uh equation to account for many different age age groups or lagged effects of the lab uh the size of  
[01:38:14] the labor force. The end result is the same. Okay. However, we can't run this this structural equation. We can't estimate this structural equation directly because the size of the labor force is indogenous in particular the size of the labor force and population  
[01:38:29] in general is determined by three different factors. Okay, one is the number of birth rates or the number of births. Two is migration. People might move in and out of a given population and three are death rates. Okay, and these migration decisions in particular  
[01:38:43] are indogenous. When I decided to move from Ireland to the US, that was surely shaped by the economic conditions in both countries. Okay. And so we when we run uh regressions of population on GDP growth, you know, we just can't estimate  
[01:38:56] the causal effect from this alone. Okay. Uh however, we can take population dynamics seriously and the you know um this is a little bit of a simplistic model here,  
[01:39:09] but we in the paper we really try and take these population dynamics seriously and acknowledge that population shifts come from either birth rates or migration or death rates. Here I'm going to uh bundle migration and death rates into this composite exit rate. What we find is that the growth in population  
[01:39:24] between 92 periods is just the number of deaths less the number of births divided by the initial population. And so it's just the birth rate minus the death rate. Okay? And so we can put that into our structural equation and we find that  
[01:39:37] GDP growth is related to the level of the birth rate. This is the birth rate measured as a share of the total population. Okay. and modulo sum controls and here we have a an error term which is the structural error in  
[01:39:52] the GDP equation plus the uh unobserved determinance of migration etc etc and deaths okay now how are we going to identify this equation so the key feature of the data  
[01:40:06] that we're going to exploit is the fact that birth rates today don't affect the labor market today it takes people 20 years to grow up and be able to enter the labor market and so our identifying assumption is that past birth rates are  
[01:40:18] not co caused by the shocks that occur in later years. So you know quite literally birth rates 20 years ago have not been caused by the shocks to that uh change GDP today. So if you imagine you  
[01:40:32] know you might imagine well parents make fertility decisions based on their forecast of uh of future economic conditions if you you know quite a cynical view of parenting but you know it's it's at least somewhat plausible  
[01:40:46] and you know you could think that is actually fine for our story because even if fertility is indogenous our exclusion restriction still holds basically under a kind of markov assumption for GDP or in particular if I'm a parent who's deciding whether or not to have children  
[01:41:01] today and I'm trying to forecast the economic conditions for those children in say 30 years from now in 206 that indogenous uh fertility choice is okay as long as I don't have extra information about 2056 shops that are  
[01:41:15] not available to an econometrician who comes along in 24 46 with all the next 20 years of data available to them. So we think this is pretty reasonable that you know uh while parents may make fertility decisions indogenously they're  
[01:41:29] not oracles who can are amazing at forecasting future GDP. Okay. Finally, it's important to note what we can identify with this strategy. So we're going to estimate impulse responses or event studies to higher birth rates in  
[01:41:41] the past. Okay. And I'm going to show you that these birth rates in the 20 years after they happen, they have essentially no effect on um on the next uh a 20 years after the birth rate, there's no effect on GDP growth. So this  
[01:41:54] is a nice placebo test for us. Um what do these impulse responses identify?  
[01:42:00] They identify the entire reduced form effect of birth rates inclusive of all endogenous population dynamics that these birth rates uh cause. Okay. So as long as the birth rates are random, but they can affect the population structure  
[01:42:14] in a myriad of ways, that's all captured in this reduced form response. Okay? So I just wanted to be wanted to be clear on that. Things like the demographic dividend where you might have a home shape response, all of that should be in the reduced form response. Okay, any  
[01:42:28] questions on our identification strategy before I move on.  
[01:42:33] Okay, so let me jump into the results. So here's our main result across countries.  
[01:42:39] So what I'm plotting here is the impulse response of GDP per worker. GDP per worker what we define it as is actually GDP per working age adult GDP per worker for short. And it's GDP divided by the size of the population aged between 20  
[01:42:52] and 70. This is just to take out the mechanical effect of a you know a younger population means more dependence more dependence and more people in the in the denominator of GDP per capita. We can use GDP per capita and it's the  
[01:43:07] same. So what this is plotting is the response of GDP per worker to a birth rate shock at t minus 20. This is the horizon. T is t is equal to 0 is when the people entered the labor force. t minus 20 is when the birth rate shock  
[01:43:21] happens. And we see as I said in the first 20 years after the birth rate shock happens there is no effect on economic outcomes. Then as soon as these people are old enough to enter into the labor force what happens? Well, with more people entering the labor force, we  
[01:43:35] see GDP per worker declines. And even 70 years, 50 years after they've entered the labor force or 70 years after they've been born, presumably when they're starting to leave the labor force, the effect size persists. Okay?  
[01:43:48] And the effect effect size is quite large. a 1 percentage point increase in birth rates which is you know very very large but also in the support of our data over the long run leads to a 30% higher GDP  
[01:44:02] about 70 years after the birth rate happens. So 1% decline in in in birth rates leads to 30% higher GDP. So it's important to remember that these impulse responses are to a rise in birth rates.  
[01:44:13] So then if we want to think about the effect of the actual birth rates change that happened we have to flip the science. Okay, so we can also do this across commuting zones and across commuting zones we don't observe GDP but what we do observe is w microlevel wages  
[01:44:28] and so we have a method to construct uh wage series that uh account for the change in composition of workers across commuting zones and we find very similar results across commuting zones we uh have more power so our standard errors are are smaller and our estimates are  
[01:44:42] more precise one thing to note is that we use variation in birth rates that happen sufficiently far in the past. So we can see uh their effect over time.  
[01:44:54] For communing zones we use birth rates between 1914 and 1970. Across countries we have only have them from 1950 to 1970. Okay. And across communing zones the magnitude of the effect is much smaller. It's about a half of the  
[01:45:07] magnitude across countries. Okay. And so what explains this difference in magnitude and are these magnitudes plausible? And the answer to that question is actually the we have one answer to both of those questions. And  
[01:45:21] so the answer is that when we estimate the effect of aggregate income on the left hand side which is average income plus the size of log average income plus log size of the population we find essentially no effect on the size of total income even though the size of the  
[01:45:35] workforce is shrinking dramatically. So what this means is that whatever those technology responses are, they're sufficient to keep aggregate income constant as the magnitude of the population shrinks. And what's happening is that across uh across communing zone,  
[01:45:49] the size of the population decline is just smaller. This is explained both by the fact that there's migration responses which might dampen the effects of birth rates and actually what's more important which is somewhat surprising is that birth rate shocks are less persistent across communing zones than  
[01:46:04] they are across countries. Okay. So one thing that stands to reason is from growth accounting that if GDP is rising and the labor force is shrinking then capital or TFP must be responding to compensate. Okay. And so now I'm  
[01:46:18] going to talk about these capital and TFP responses.  
[01:46:22] Okay. So before I jump into the uh technology responses it's important to have an example in our mind of what is this mechan this labor saving mechanism we have in mind. Okay. So think about a labor intensive industry, say a  
[01:46:36] restaurant, and somewhere in the 1980s they have a pretty profitable business because they hire a bunch of say high school or college age kids to staff the restaurant and they have a very labor intensive production, but they can provide very good service.  
[01:46:50] And then suppose that the birth rate declines and these college kids are no longer so abundant and the wage that they pay rises. The wage they must pay the kids rises. And then essentially we have uh two decisions that the firm can  
[01:47:04] make to adjust. Decision number one is they can invest in technology such that they don't need so many workers anymore.  
[01:47:11] So this could be as simple as making their processes more um more streamlined to as complicated as you know introducing those screens that you can order yourself. You know different there's a bunch of different types of uh labor saving technical change that that  
[01:47:26] firm could invest in. Another way in this this economy could adjust which is kind of equivalent at the macro level is rather than that entrepreneur and using that physical capital to open a restaurant they could instead  
[01:47:39] pursue some other activity like say have a startup or manufacturing firm. Okay.  
[01:47:46] And in that case economic activity shifts away from this labor intensive sector towards sectors that are less labor intensive and more capital intensive and maybe even uh R&D intensive. We're going to find evidence of both of these channels. Okay. So,  
[01:48:00] first I'm going to show you the changes in industrial competition. So, here I'm showing you the changes in the share of either exports for countries or employment for commuting zones in high-tech sectors. Okay? And what we see  
[01:48:13] is that 20 years after uh birth rates rise across both countries and across commuting zones, the intensity of production in high-tech sectors falls.  
[01:48:23] Or on the flip side, if birth rates fell, then the intensity of production in high-tech sectors would rise. Okay, these are both share of total exports that a country has and the share of employment. Okay, so this is more about the composition of economic activity  
[01:48:37] than the size of the pie that I showed you uh that I showed you before. Okay, the second thing we can look at is look at uh how economic activity shifts based on the factor intensity of an industry.  
[01:48:52] So here we can only do this within the US across country commuting zones and the way we classify industries based on factor intensity is as follows. So we can see the factor intensity of industries at the aggregate level using  
[01:49:05] the BEA accounts and we call an industry labor intensive if the labor share at the aggregate level is above the median over all industries in the US and we do the same for R&D intensive. we can see uh compensation to R&D in those accounts  
[01:49:20] and we can classify those industries and then run this same regression but but uh using the share of labor intensive or R&D intensive as an outcome of employment as an outcome and what we find is that in response to an increase  
[01:49:34] in birth rates so more abundant labor labor intensive industries naturally expand and quite interestingly R&D intensive industries contract and contract quite a lot and this is especially start considering these R&D intensive  
[01:49:47] industries are a much smaller share of the economy than labor intensive industries. Okay. So there's a general reallocation of economic activity away from labor intensive uh when birth rates fall there's a reallocation of economic  
[01:50:01] activity away from labor intensive industries and towards R&D intensive industries. Okay. And then finally the simpler margin of adjustment is just uh those firms themselves invest in technologies that save labor. We've got  
[01:50:15] many three different measures of uh labor saving technical change labor saving technical change in patent data.  
[01:50:22] One is patents that are classified as ICT patents. Another is patents that are we can classify as automation patents.  
[01:50:30] And then a final one is actually patents that end up being used by industries that are labor intensive. For all of these measures, we see that when birth rates increase or when labor becomes more abundant, the intensity of these  
[01:50:43] labor intensive patenting patenting goes down. So their share in the total patent pool in these locations shrinks. This is especially surprising considering that I I just showed you those labor intensive sectors are expanding but their share of  
[01:50:56] patents is actually contracting and this I think is very indicative of our laborsaving technical change mechanism.  
[01:51:04] Okay, very good. No question. Any questions so far on the technology responses? Okay, so as I I just showed you um following lower birth rates, um higher TF leads to  
[01:51:18] higher TFP and a higher capital stock across countries. I didn't include this slide in the deck, but we can show it either of these must increase just by growth accounting. It turns out that both increase. economic activity shifts  
[01:51:32] away from labor intensive sectors and towards high-tech sectors and there's also a shift of patenting towards laborsaving technologies. Okay, so with the last 10 minutes let me talk  
[01:51:47] about how we can disentangle the effect of the population size versus the effect of age structure on GDP growth. Okay.  
[01:51:56] So, uh as I said before, birth rates move both population and the age structure. However, with one instrument and two indogenous variables, we can never separately identify the two. This is just the nature of these reduced form  
[01:52:09] responses. They include both effects. And so, how are we going to separate separately identify the two? Well, we're going to rely on some somewhat morbid variation, and that's deaths in during the Second World War. And in particular, we're going to distinguish between  
[01:52:22] military debts, which are concentrated among young men, and civilian debts, which we believe are plausibly uniform across all ages. And so we think that um in the data, it's true that when there's more military debts, the population gets older, whereas when there's civilian  
[01:52:37] debts, the population gets smaller, but the age structure says stays the same.  
[01:52:40] And as I said, this is something we can test in the data.  
[01:52:45] And then we're going to estimate this this regression which is just an event study of GDP growth since 1935 and mo both military debts which is the number of military debts divided by population in 1935 and the civilian  
[01:52:59] death rate which is the number of civilian debts divided by the population in 1935. And we're going to control as we do in all our regressions for the initial size of GDP. Okay. So our sample is 22 countries. So it's much smaller  
[01:53:13] here uh because we need to see what debts in the second world war and it runs from 1910 and 1995. Okay. And so what we're going to find is that before the second world war in these early  
[01:53:26] periods there's no discernable effect of these uh debts on economic outcomes which we think is a reassuring pre-trend. However after the war we see a very interesting pattern. First is that more military debts which increases  
[01:53:41] aging leads to much higher GDP per worker even as late as 1995. So a 1 percentage point increase in military debts which is very very large leads to  
[01:53:53] a 10% higher GDP by 1995. By contrast civilian debts have the opposite effect although it's lower uh smaller in magnitude. So what this is saying is that while population growth  
[01:54:06] may indeed be good for economic growth, birth rates have another effect which means that they increase aging and that in that effect is sufficiently strong to outweigh the negative effect of lower  
[01:54:19] population. Okay. And so with that I'm going to conclude.  
[01:54:24] So there's general alarm among economists, demographers, and policy makers and in the media that low birth rates are going to be really bad for the economy and they're going to cause a slowdown in economic growth. When we look at the past data, we find exactly  
[01:54:39] the opposite. So using the last seven decades of data, we find that uh baby booms decrease GDP per worker and a baby bust or lower birth rates increase GDP per worker. Our interpretation is that  
[01:54:53] this is due to laborsaving technical change where wages increase and firms choose to substitute towards technology and the response is labor saving and output increasing technology. Okay. Uh  
[01:55:06] when we analyze debts in the second world war we find that our responses are really driven by the scarcity of younger workers. that is when birth rates increase aging that makes younger workers more scarce and that is the key  
[01:55:20] force that induces this labor saving technical change and alternative mechanisms which I haven't included here but I have in the appendix slides in case anyone is interested um they just cannot explain our findings you know  
[01:55:32] they they there's definitely evidence of th of those channels in other data they're just not uh able to quantitatively explain our findings okay and So a final word about our paper is just that there's a lot more to do and  
[01:55:47] you know these this uh our paper has showed you what has happened in the past when birth rates have declined but they it can't say what can happen in the future. However any model that you come up with that you use to project into the  

## Q&A (01:56:01 – 02:11:31)

[01:56:01] future it should be able to explain the observed data and the observed data includes our estimates. Okay so with that I'm going to conclude. Thank you very much and I look forward to everyone's questions. Fantastic. Thank you so much, Kaylin. Uh, we have a few  
[01:56:15] questions lined up, so Diego, please go ahead.  
[01:56:24] >> Thank you. I hope you can hear me. Um, >> yeah, I can hear you now.  
[01:56:28] >> Fantastic. I wanted to ask two questions if you allow me. Number one is I'm just wondering why birth rates. I understand that they are useful for understanding for for um for the process of having an  
[01:56:42] exogenous labor force but birth rate birth rates are the can are the ones that drive the demographic dividend as well as population agent. So how can you distinguish between those two effects that are caused by declining birth  
[01:56:57] rates? That's question number one. And question number two is how will these dynamics play out for different countries or if the model has something to say about that? I'm thinking about middle or low income countries that will  
[01:57:11] experience aging at a really fast pace. Um will they be able to also gain from the positive impact that your model predicts from  
[01:57:24] birth per rates into capital formation and TFP.  
[01:57:28] Thank you. Yeah. So on the first question about why birth rates. So a number of things. First is that birth rates are actually the by far the most important driver of population structure. So if you do a simulation  
[01:57:41] where you keep um longevity or mortality constant uh since 1950 but vary birth rates, you can create a decomposition of how much of the population is due to mortality changes versus birth rates.  
[01:57:54] And you find that birth rates explain the bulk of population structure changes at least for our uh relatively crude measures. Um the second question about demographic dividend is our estimates are the reduced form  
[01:58:08] effect estimate of birth rates. They include both channels and so you know it's just impossible to separate the two. We believe that birth rates because uh as I said population changes have to be due to either birth rates or migration rates or death rates. We  
[01:58:23] believe the latter two to be indogenous and so birth rates are really the first thing we should look at. Um but you're absolutely right. We cannot separate the difference between aging and population growth with uh with birth rates alone.  
[01:58:35] And we tried to take a first step towards this with the second world war debts. Um I mean you know the sample is limited statistical power is limited but I still find them really quite plausible. Um or at least very thoughtprovoking. The second question  
[01:58:48] about low-income countries. So haven't done it here but we did do some uh work on the hetrogenity and we didn't find that there was too much hetrogenity. I mean you see you still see see the effects across or within the US but it's  
[01:59:02] actually true that the effect sizes if anything are larger for uh lower income countries um which is somewhat surprising but uh it's true I guess they have less uh manufacturing/hight  
[01:59:15] intensive production and so there's more uh scope for this laborsaving technical change but it's not something that we've explored in so much detail but if anything the evidence suggests that for low-income countries uh the effect sizes  
[01:59:28] will be even bigger. >> Great. Thank you. Uh let's go through other questions and then we can have some back and forth as is relevant. So u  
[01:59:42] >> uh great uh thanks uh super interesting. So that uh version of your result uh on commuting zone uh is very analogous to the literature of urban uh economics on  
[01:59:54] you know elomeration externality and the you know maybe this is like what people want to believe but my understanding is that there's a robust you know pattern that there's positive association between population density and uh the  
[02:00:07] productivity so is is that result so I have couple of concrete questions one concrete question is obviously your the difference of your design is using you're using birth shocks which I'm not  
[02:00:20] aware that urban ecologists have tried in the past. So if you do just like OS uh do you get the positive results and you get the opposite results once you do  
[02:00:32] the look at the birth effects and the um uh yeah and I guess I have a kind of related um concern about identification strategy which is in urban areas you know the fertility  
[02:00:46] rate is lower and that's you know also I think a robust pattern because there's a you fertility and working trade-offs.  
[02:00:56] So, uh, and I think you mentioned a couple of things about, you know, controlling for future expectation and so on. I'm I'm wondering whether, you know, you have some comments on that.  
[02:01:09] >> Yeah, thank you very much. Let me deal with the second question first because it's fresh in my mind. Um, so we have in the paper we have a bunch and I should have mentioned this a table that explores a lot of robustness. One of those is control for urbanization share  
[02:01:23] or uh um population density and we find that the estimates are basically unchanged when we do that. So we don't think that this is being explained by that. And the other thing to note is that um you know you might think that  
[02:01:36] these high birth rates are concentrated in say the more rural areas the US and the lower birth rates are on the coastal areas. That's actually not true. If you plot a map of birth rates, which maybe I should have included, it's basically like a bunch of different dots and  
[02:01:49] they're also not that persistent over time in the US. Um, so there's not a whole lot of evidence of huge spatial correlation in these birth rates. Um, that kind of is linked with the first question about density and uh the  
[02:02:03] difference between OS and um IV. So across commuting zones, the difference between OS and IV is very very important precisely because migration is such a big deal. And if you run OS across communing zones, you get a coefficient of basically zero, which I think makes  
[02:02:17] sense because if wages rise, people will move in to a in into a location to exploit the higher wages. Um, if you do it across countries, you actually can get something with OS. And this is just presumably because  
[02:02:31] migration is just not that as big across countries, and so birth rate variation dominates. But we still think that the birth rate variation is the part of the variation that we should be looking at because that's the variation that's uh truly indogenous. So I hope that  
[02:02:45] answered your question. If not, you should let me know and there's an active discussion in the chat  
[02:02:57] too as well. So thank you uh Andrew for that. So why don't we move ahead to James?  
[02:03:06] >> Thanks. Um so yeah my question is um about sort of the short versus long run of the theory that you're proposing. You kind of you motivated at the beginning um this like laborsaving technological change theory as an alternative to  
[02:03:20] semi-indogenous growth. Um and that is a very kind of long run theory. Whereas you know in the time span of that kind of theory I would view your empirical results as pretty short run. Um also there are you know there's a one  
[02:03:32] particular paper that comes to mind by uh Connor Walsh and Michael Peters uh in which they they have a semi-indogenous growth model but they study the transition of population decline and they they actually find that on the short run you kind of do get some  
[02:03:46] positive growth along the transition and then in the long run you get the the semi-indogenous growth um effect dominating where you just get persistently low growth. So my question for you though is like do you view your theory as something that's like going to persist over a few decades and then in  
[02:04:00] the long run we get to the world of semi-indogenous growth or do you think this is something that's like permanently going to be offsetting that?  
[02:04:07] >> Yeah, that's a great question. So first our empirical estimates are basically growing over time and there's there's no evidence that they're going to going to turn back towards zero. So it seems like this is at least in the data a long run response if you define the long run as  
[02:04:20] say 80 70 80 years. Um the model we have is uh we actually have a static model in the in the paper but um you know you could make it dynamic and it's really about steady states. I think it's true that you know this so our second world  
[02:04:35] war estimates while a bit spec uh speculative they do suggest that this uh semi-indogenous growth mechanism is real it's just quantitatively dominant dominated by the laborsaving technical change mechanism and these laborsaving  
[02:04:49] technical change models are really just a multis- sector extension of a kind of semi-indogenous style growth model. Um and we think that yeah this this effect on aging is going to last into the long run. Lower birth rates does mean a uh a  
[02:05:04] older uh stationary distribution for the population. And so yeah we think that uh in the long run there'll be older workers, wages will be higher for younger workers and that will in induce a more advanced technological choice. So  
[02:05:18] at least in the theory and in the data there's nothing to suggest that these effects are temporary. Of course, without seeing more data, we can never say for sure, but we think that there's no reason to believe that these would be temporary, but it's a it's an excellent point and, you know, definitely a  
[02:05:32] fruitful act uh area for future research.  
[02:05:37] >> Excellent. Thank you, Adrian. >> Yeah, thanks. Super provocative paper, very interesting. Uh, I have a question about the laborsaving technical change  
[02:05:48] mechanism. So how much do we know that is kind of a some a phenomenon that's happening locally versus globally? And because I think if I understand correctly your results can be  
[02:06:02] rationalized with something like this if it's more like a kind of a local phenomenon. Um but you could imagine that a lot of those innovations or so on that are developed in response to falling birth rates can actually give  
[02:06:14] hues pretty quickly and so so and and if so how how can you interpret your results in that light?  
[02:06:24] >> I see. So so by local you mean correct me if I'm wrong but I'm understanding that you mean geographically locally rather. So geographically local as in in the unit of observation that you have which is whatever country or you know  
[02:06:36] within the counties in the US and so on. >> Yeah. So that's a that's a good point and uh you know there's always a missing intercept problem and I guess that yeah I think in in general it may be the case that uh you might be somewhat worried  
[02:06:51] about uh uh technology at the global level but to the extent that our results are very continuous across community zones which are more much smaller unit of aggregation than say countries and we still see the same effect across  
[02:07:04] countries and across commuting zones. Um I think that makes us pretty confident that uh this labor saving me technological change mechanism is is real in the data. Um and yeah I think our moments that we estimate these  
[02:07:19] cross-sectional moments are still uh very informative of how uh technical change works. Of course if you tried to simulate a model there would also be an aggregate intercept that's related to the aggregate level of global  
[02:07:32] technology. But I think we could you know there's an argument to be made that you could just you know our model or our regression estimates should be used to discipline that model. Um >> but if I understand correctly the effects at the global level in this missing intercept would be even bigger  
[02:07:47] because to the extent that there's any diffusion at all of labor saving technology across places that's something in your missing intercept.  
[02:07:53] Actually, you know, if we wanted a theory of this, this would this would be an even bigger effect than the 30 percentage point in increase.  
[02:08:03] >> I think that's exactly right. Yeah. Yeah. Um Yeah, it would be even bigger.  
[02:08:07] Uh yeah, I guess. Yeah. I don't I can't think of any uh thing that will push in the other direction at least within the laborsaving technical change um way of  
[02:08:21] thinking about the world. Yeah. Great. We have uh time for one more uh comment question. Larry, I you're muted still.  
[02:08:34] >> Yeah. Hi. Uh very interesting paper. So here's my perhaps a tangential concern.  
[02:08:42] >> In the US last year, we saved 1.5% of national income. Our net domestic investment rate was 6.5. So foreigners were investing $3.33 for every dollar Americans were  
[02:08:56] investing in our country. So uh if we have um this aging and we also have in  
[02:09:06] our country very high fiscal uh uh you know obligations due to uh our pay are basically you know our Ponzi scheme and  
[02:09:19] consequently um we end up not owning the the capital in our country as Americans but foreigners are owning it. So we're looking more like Jamaica where all the hotels are owned by foreigners and the Jamaicans are earning $6,000 in this  
[02:09:33] servant economy they're living in. So if you have um in your model uh let's say you know technological change and and capital becomes produces a higher return and this produces uh more imported  
[02:09:48] capital uh into let's say the US. So it could be the GDP per capita is not really the right measure for thinking about is this going to help the country.  
[02:09:58] It could be the GDP per capita goes up but but income per capita goes down for if we think about uh Americans national income.  
[02:10:08] >> Yeah. No, I think that's a that's actually an excellent point and I should have said that the object of analysis is not a country's welfare. I'm actually uh a trade economist and so I know that GDP doesn't in include both transfers but  
[02:10:21] also the terms of trade, right? Um that's an advantage for our cross country estimates because there's no price effects in there. But um it also means that what you're measuring is productivity and production. You're not measuring welfare. And I totally agree  
[02:10:34] with you that uh this is not saying much. This cannot say anything about welfare without some extrapolation. But I think it's pretty reasonable that if your productivity goes up by 30%, you you probably do gain. But yeah, the the object of analysis is really um  
[02:10:49] productivity. That's why we're focusing on this labor saving technical change mechanism. Another thing that you brought up that I think is interesting is thinking about the future of say the US. You know, we can say stuff about what happened within the support of our data. You know, a US sovereign default  
[02:11:04] never happened in the support of our data. So that's not to say that you know this is these birth rate these things that we've seen in the past are definitely going what's going to happen in the future. It's just that whatever the story is that about what's going to happen in the future it should respect  
[02:11:18] what we have seen in the evidence that we've seen in the past. Um you know for example the economic mechanism of labor saving technical change that the evidence of our paper suggests that's a real feature of demographic change.