Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable. ======================================================================== # Endogenous Technological Change Along the Demographic Transition Authors: Markus Pettersson Discussant: None Video: https://www.youtube.com/watch?v=w6ObGDwelyg&t=7891s ## Talk (02:11:31 – 02:41:48) [02:11:31] >> All right. Fantastic. Thank you so much uh Kalin. uh excellent paper. So now we're going to switch to the next paper where we're going to talk more about indogous techn technological change [02:11:45] along the demographic transition. So Marcus Peterson from Stockholm University, please take it away. [02:11:52] >> Thank you. Can you all hear me by the way? [02:11:54] >> Yes. >> Yeah. Perfect. [02:12:02] All right. Sorry. And you can see the slides. [02:12:09] >> Yeah, now we can. They're perfect. >> Perfect. All right. Yeah. So, thanks for for for uh having me and thank you for arranging the the conference. Um I was very happy about the last paper because [02:12:22] it's effectively a very good empirical complement of of what I doing. Uh this is also a paper about demographic change and and economic growth. So the start just start us off. Uh let's you know [02:12:36] state the fact that everyone here already knows the reason why we're here. [02:12:39] uh populations are are getting uh older uh everywhere and uh you know uh as I'm showing here with the rise in old age dependency ratios all across the world [02:12:52] and this growth uh in in in aging is seems to be spearheaded by high income countries uh with the rest of the world trailing not far behind and we know this is raising a lot of macroeconomic questions uh and in this paper we're [02:13:06] looking at two questions that are sort of underlying ing a lot of uh lot of these different macroeconomic concerns and that is uh one that uh what happens to upper per capita during this demographic transition and if anything [02:13:20] uh why so in order to to think about this uh a good starting point is to think about uh output per capita when you have cobless production in that case you can uh decompose uh output per capita into [02:13:35] three separate factors uh TFP or technology a capital output ratio and uh some type of employment rate. [02:13:43] Now uh conventional uh con conventional life cycle reasoning points to two separate separate mechanisms here to think about. One is that if we have an aging population so that the old age population rises, we know that uh the [02:13:58] share of workers are going to decline provided that there's not a huge increase in in the retirement age. Uh and so that is going to uh put a downward pressure on on output per capita. [02:14:09] But then if you think about the productivity of workers uh you have a if we have an increase in in in longevity people are going to save more uh in anticipation of longer retirement uh and so higher savings leads to higher investment and that is going to uh [02:14:23] increase the the capital stock and capital intensity and and therefore the productivity of workers which pushes uh GDP upwards and so the net effect here is effectively ambiguous right uh now uh my [02:14:37] take on this when I started this project was that well trying to get this uh pinpointing this effect empirically is um hard let's put it that way uh so I was very happy about the previous paper but I think the uh the a natural [02:14:50] complement to that is to think about quantitative theory uh because it allows us to take into account uh general equilibrium effects uh and different types of of channels that are sort of working against each other and what we [02:15:04] see in in calibrated uh sort of life cycle models to the advanced economies. [02:15:09] Uh what is typically found is that the employment effect dominates. Uh so if you look at for instance Kruger and Ludvig and Ludig and co-authors in their scenarios when they look at only these two channels uh the capital intensity and labor population ratio uh GDP per [02:15:23] capita declines uh on the order of 10% cumulative over the 21st century uh relative to this its initial trend. All right. And I mean there are a lot of different other mechanisms that have [02:15:36] been been uh proposed beyond these two. Uh but one thing that are sort of underexplored in this literature is the impact on this third term here this technology term. [02:15:48] So, and and it's a bit what we should think about here is that if we have an increase in savings uh and that increases investment, we shouldn't just think about an increase in in investment uh to be to to be allocated [02:16:02] towards physical capital. But it could also be that uh we increase uh investment in things like research and development uh other types of intellectual property products. uh effectively things that are core inputs [02:16:15] into the innovation process that that generates uh technological change. [02:16:21] Uh and the key issue and difference here between capital and technology is of course the fact that uh capital is is a rival good. Uh one a unit of a new machine can only be used by so many workers and so capital deepening only increases the productivity of workers in [02:16:36] so far as it increases the capital uh capital per worker ratio. Innovation on the other hand or technology is a non-rival good. So so it can be used by everybody at at any time at the same uh [02:16:49] at no added cost. And so we should think that R&D investment has a much uh larger sort of potential impact on uh on GDP from this aging process and from this increase in in savings. [02:17:04] Uh so this is in effect what this paper is doing. we're we're what I'm doing is that I'm quantifying this R&D channel uh in a general equilibrium setting and I do that by combining two sort of workhorse modeling traditions right so [02:17:18] on the household side I'm building an oval gene model in the Arab caut tradition and I'm merging that with a semi-indogenous growth model of of you know in the Chad Josian uh tradition which allows me to talk about how the population structure uh affects [02:17:33] technology over this demographic shift uh and I'm using that to both analyze uh the the demographic transition both qualitatively and quantitatively. Uh the qualitative uh analysis I won't talk about today but it's effectively what I [02:17:46] just told you uh in terms of of savings increasing TFP uh I'm not the first one to look at this but uh what what so so there are a lot of papers that that draw this connection [02:18:00] between demographics and and R&D based growth. uh what I'm what's different here is that I'm treating the uh the the demographic shift that we observe in in rich world countries uh as a as a [02:18:14] transition instead of a of a shift between two steady state population growth rates because if you think about uh semogenous growth aggregate growth or per capita growth is tied to the population growth rate. So if you model [02:18:27] uh population aging as a decline in steady state population growth, you mechanically impose that uh it's going to lower GDP growth. The problem here or like if you think about the US demographic transition since let's say [02:18:40] the second world war, most of that is characterized by two two sort of key features. So one is the temporary increase in fertility, the baby boom and the other is the uh steadily rising survival rates for people above the age [02:18:54] of 50. And neither of these, you know, you can argue that neither of these should affect steady state population growth very much. Uh and so it becomes in order to to get an idea of how large these quantitative effects are from [02:19:07] these temporary uh demographic shifts. Uh we need to sort of uh move away from from steady state comparative statics. [02:19:17] uh and these are of course important because they generate still temporary growth effects uh that leads to permanently higher level uh effects uh in this economy and those those transitions are of course uh over long periods of time in terms of decades and [02:19:31] and so on. Uh so that's the paper in a nutshell just to give you a quick quantitative preview. So I I simulate this model uh for a demographic transition path from [02:19:44] 1950 onwards for the US and what I did find is that output per capita goes up uh so this is like the annual average growth rate. So it goes up on the order of 0.16 percentage points per year uh [02:19:58] overall and uh in particular during the the first uh part during the 20th century it grows up by around 0.3 percentage points per year. Uh and the reason is precisely this technology channel that I'm adding to the model u [02:20:11] which is much stronger than than a capital deepening effect that you traditionally get from these types of OG models. [02:20:18] And just to put this number in perspective, you know, if you think about US long run growth as being around 1.82% per year, uh a 0.3 percentage point increase in or factor of of growth means that you know [02:20:32] roughly 15 to 20% of of use growth could be explained by changes in uh population size and structure alone. Uh so this is a pretty large effect. All right. [02:20:46] Right. So that's the paper in a nutshell. Uh so before uh let me uh skip the the underlying mechanisms and just go straight to the model and explain that in a bit before going more into [02:20:59] detail on the quantitative results. So what we have here is a uh as I said it's a it's a effectively a merger between a nog type model with a semi-indogenous growth model. So on the [02:21:12] supply side we have uh you know producers uh that that that you know perfectly competitive final for good producer that that hires intermediate inputs that are driven by monopolistic uh producers uh and then there's an R&D [02:21:26] sector that sort of generates new firms into this intermediate sector uh due to the profit motive that that monopolistically competitive firms have. [02:21:35] uh households are uh heterogeneous agent life cycle types of households. They consume save work over the life cycle. [02:21:44] Uh and the difference here relative to sort of standard exogenous growth models is that now they have effectively two types of assets to save in. So on the one hand they can save capital that goes directly into production but they can [02:21:56] also finance uh R&D in exchange for equity claims on uh whatever firms that that R&D sector innovates uh and then they earn returns and dividends from from those firms. [02:22:11] There's also government that sort of works a little bit in the background here uh in the sense that I'll use it as a calibration device with respect to taxes but also in order to capture sort of key uh fiscal policy adjustments such [02:22:24] as uh changes to the pension system that happens when a population gets older. [02:22:30] Uh but overall uh I just say that this is a model that I'm not trying to be super innovative uh within each of these separate building blocks. Uh the sort of core contribution here is the merger of [02:22:43] these two types of modeling strands. Uh and then I'm sort of deliberately being uh deliberately being pretty standard within each of these blocks. The key richness sort of goes into the household [02:22:55] sector in order to c properly capture the sort of the savings consumption decisions of the households uh because the entire causality of this model is driven by changes in the demographic structure that are then fed through the [02:23:09] household sector and then pretty much drives all the dynamics in the model. So that's the only the demographic block is the only source of exogenous variation uh in this model. Uh and that's modeled separately. I I take u age specific uh [02:23:24] fertility rates, mortality rates and and migration rates and just build the change in the population structure over time and which I then feed into the model. [02:23:34] All right. So a little bit about the the households and the production sector uh before moving on. Uh so households are um you know they have preference over consumption uh and leisure uh and the [02:23:48] idea of leaving uh the quest for future children. Uh and it I mean I should say here that you know my prior was that TFP would sort of switch the the the impact of the [02:24:02] demographic change into a positive rather than a net negative. And because of that, I sort of deliberately had a very conservative uh way of modeling things when whenever I had a choice. So here, for instance, I I allow for for people to have a warm globe quest [02:24:15] motive. Uh regardless of their income, whether they're rich or poor. Uh that is because that would is effectively working against whatever me the mechanism I suggested in the beginning. [02:24:25] So when longevity increases, the life cycle savings motive strengthens, but at the same time, this bequest motive is going to weaken. So, so that dampens a little bit the response that we're uh receiving. [02:24:37] The same goes for for hours worked. So, uh consumption and hours worked a model according to McCertie preferences uh that uh in which the the income effect from from an increase in wages effectively outweighs the substitution [02:24:50] effect. Right? So, uh, whenever there's going to be income growth in a model, people are going to respond by lowering hours worked, which is consistent with long run facts on, uh, on how ours have developed. And there's the same thing, [02:25:03] right? So, if GDP per capita goes up for some reason, people are going to respond by lowering hours and that sort of dampens the whole response in in a way. [02:25:13] So, households obviously they they choose consumption in hours. uh but they I also allow them to have an extensive margin uh retirement choice a discrete choice in which they they choose when to [02:25:25] to leave the the labor force. This gives us a little bit of action along the the extensive margin in the sense that when whenever longevity increases uh there there's not necessarily the case that the the share of retired people in the [02:25:38] economy are going to uh increase one for one uh because people might very well just increase work for another year or so. [02:25:46] So the way that that problem works is it's solved in a two-step process where where people enter some HJ uh and observe their assets, their productivity uh and the last period retirement uh state. Then they choose to retire by by [02:26:01] by choosing effectively which uh whether retirement or working is just giving them the highest value of of utility. [02:26:08] And then once that and so we assume that that's an absorbing state. So whenever you have retired, you cannot unretire uh for simplicity. [02:26:16] And then once that's done, the the the state is updated to to reflect uh this new retirement choice after which the households then make a standard consumption savings hours choice following a fairly uh standard beltman [02:26:29] equation. At the end of the period, people survive some probability s uh the whatever idiosyncratic productivity they have realizes to a new value and you enter hj plus one and the whole thing [02:26:41] runs over again. Um, right. So, quickly about firms as well. Uh, I'm going very quickly here because to me [02:26:55] to me it's very standard. If it's not, feel free to ask later on. Uh, I'll be happy to clarify things that are confusing. [02:27:02] Um but on the firm side we have a standard production function where where people where firms uh have a dog production function over labor and and a ces aggregate of capital inputs. Uh [02:27:15] there's a measured z of monopolists that that have a linear production function in the capital. So when you when you merge these two and consolidate them when you solve the optimization problem we get a pretty much a standard cob dogless function over capital which is [02:27:30] the sum of these uh intermediate inputs uh TFP which now is a function of intermediate the this measure of intermediate firms raised to the power parameters and and labor [02:27:43] and this uh TFP measure uh varies indogenously in the model uh by uh through an R&D process which looks like this. So this is pretty much the same standard way that [02:27:56] that Chad Jones had it in his uh I think his job market favor right. So so uh on the right hand side here you have uh on the last term here is the R&D production function. So this takes oh sorry so this [02:28:08] takes R&D investment units of final good as an input uh and and just to to sort of add a bit of um flexibility we allow that also for for duplication externality the stepping on toes effect in which people you know may work on the [02:28:23] same ideas but only one's going to be successful uh one in which a knowledge spillover so that the more the the more the process innovation process moves forward we may [02:28:36] become less or more productive in in in R&D depending on on the sign of this uh knowledge spillover parameter five and then there's a uh finally just an exogenous exit rate just to make sure that there's always some type of R&D [02:28:50] going on in the model bit of a technical term right so in a nutshell those are the key components of the model uh it has a bunch of other things um that are a bit [02:29:04] in the background from uh let me know if you have any interest in in hearing more about that later on. But otherwise, I'll I'll uh leave that for now. Uh and yeah, as I said, it's a semi-indogenous growth model. So, uh [02:29:17] growth rates in per capita or per worker and so on in steady state, they're all going to be pinned down by the population growth rate even though it has this rich OG structure uh in the household sector. [02:29:32] Cool. So that's very briefly the the model. Uh what I then do with that framework is that I I have a thought experiment or a computational experiment where we impose a demographic shock for the United States. Right? [02:29:46] So what I do is that I I assume that the model is instead in the year 1900 and then in the year 19001 people are surprised by the news that there's going to be a demographic transition in which all of these age specific fertility [02:30:00] rates, mortality rates and migration rates are going to change over time uh according to how it is in the data and and and in projections up until the year 20100. [02:30:10] After that there's some uh we're just going to keep those fixed until the model is converged into the to the new steady state. [02:30:17] The period of interest here is uh between 1950 uh and the year 20100. uh and the reason I'm putting the the steady state earlier than that is just that it allows me to put uh uh the steady state assumption little bit far [02:30:32] away from from period of interest so that I'm not you know uh so I don't have have uh the steady state assumption affecting my results in any major way but also because it allows me to to use the years in between as a calibration [02:30:46] phase of the model. So I'm uh I'm estimating several structural parameters in the model uh internally by simulating the model and then comparing uh these transition years to to aggregate out or aggregate macro variables in terms of [02:31:00] capital output ratio and so on. Uh one aspect here that I impose is that I impose the same long run growth rate in both the initial and final steady states. This is not a huge assumption, [02:31:14] but it allows me to because the the steady state growth rates would be fairly low anyways, but it allows me to cleanly identify anything that goes on due to this demographic shock in the model. Uh to attribute that into purely [02:31:26] transitory changes uh and so I don't have to sort of factor into account that you also have a long run trend that you sort of deliberately uh lowering by by construction. So everything here is [02:31:38] trans transitional and temporary. Uh temporary in the sense that you know it might go on for 100 years but still temporary. [02:31:48] >> So can I ask a quick clarifying question? Uh yeah >> in the long run are you assuming that the TFR is at the replacement level so that the population is constant? [02:31:59] >> Yes. Sorry. So so I'm imposing a constant population uh growth rate in India. Actually that's the next slide. [02:32:07] Good question. So, so, so I'm taking uh demographic data primarily from from the UN world population prospects. Point is that if you if you take the the last value from that in 2,100 and just roll that forward [02:32:20] until it converges to a steady population, it's pretty much a zero population growth rate. Uh so I'm imposing that as the the steady state trend. [02:32:29] uh which allows me to more easily uh identify anything because the whatever positive or negative growth rate that happens in the simulation is just going to be also the deviation from from trend. [02:32:43] Yes, we can discuss this later if you want, but I'm kind of wondering like kind of puzzled by like especially given you're using a part of like Chad Jones model uh like how it relates to his uh [02:32:56] uh empty planet paper which is like you know when the population growth is uh always negative then this innovation margin is going to stagnate in the long run. [02:33:07] But you're you're you're suggesting some optimistic view and like I'm wondering whether that's coming from your assumption that in the long run it's the population becomes constant. [02:33:17] >> Uh that's a good question. So u I'm not arguing against that paper at all. Uh I am not uh looking at that's the sort of the point here. I'm not looking at steady state you know long run uh [02:33:30] outcomes. Uh and the the point is that uh because things like rises in in old age uh survival rates shouldn't really affect uh long run population growth anyways that stuff that happens after [02:33:44] your childbearing years. Uh that is something that you know you you have to sort of analyze outside of a steady state assumption anyways. [02:33:53] So some so so to me this paper is very much a complement to Chad Jones's paper. [02:33:58] I'm not saying that. I'm not trying to argue against that in any way. I'm just not interested in in steady state outcomes. I'm primarily looking at things that happens uh that that shouldn't affect the population growth in steady state. [02:34:12] Right? So, uh let me so I'll I'll skip the calibration for now. uh in the interest of time, let me just uh briefly state that we calibrate the R&D production function uh to be consistent [02:34:26] with sort of long run data on on TFP growth and and R&D investment growth and doing so we get a knowledge spillover parameter that that's slightly below zero minus 0.07 which is effectively consistent with [02:34:39] this uh idea is getting harder to find paper boom and go authors. So, so the the model captures this idea that uh as as the research frontier uh is pushed forward uh ideas are just it's just the productivity of research is just becomes [02:34:54] lower and lower. Okay. So I take this model uh simulate this demographic transition and then I collect the growth rates from those uh transition paths. All right. And what I [02:35:08] then do is a pretty standard growth accounting exercise uh by by splitting up GDP per capita growth into uh well the TFP and the capital output ratio that you saw in the beginning plus uh a further decomposition of the labor [02:35:23] employment sorry the labor uh population ratio into sort of a raw headcount employment rate plus hours per worker uh in the economy plus an efficiency per hour worked term. [02:35:36] uh so doing that uh I get uh a baseline transition uh growth path which looks like this. [02:35:45] What you're observing here is the the the black line is the growth impact from this demographic shift uh throughout the transition. Uh and the the bars are effectively showing the u the the contribution of each uh individual [02:35:58] factor of production. And so uh to me looking at the overall development first the there are sort of three things that stand out. First there's the this period during the 20th century when uh when growth rates are [02:36:11] are positive and and fairly high. Uh then there's this sharp decline in in the growth rates during the early part of the 21st century after which it remains sort of around trend uh for the [02:36:23] rest of the uh the 21st century. uh and and if you look at you know as I said in the beginning these effects are kind of large so during the 20 20th century the the average growth impact is [02:36:37] 0.33 percentage points per year uh that peaks early in the or in the middle of the 90s in 95 at some around 0.5 percentage points. So the drop here from [02:36:50] the '9s down to uh I think this is of 2030 is on the order of 0.5 percentage points. So that sort of looks like this secular stagnation argument that current population aging are just depressing growth. Uh and you do get that decline [02:37:05] in the the growth impact. But I do think it's important here to to uh sort of uh specify that this decline is not necessarily uh [02:37:17] indicative of uh uh demographics being bad for growth, right? because what it is is a reversal back towards this long run trend which is zero from the above normal growth that it had induced in the first place. So it's effectively the [02:37:32] good times sort of fading out, right? Uh and and the the the sort of the the the factors of production that that contribute to these these growth dynamics are precisely what I said in [02:37:45] the beginning, right? So so TFP growth is is positive uh and fairly stable throughout. Uh the capital intensity, the red bars are are similarly positive also throughout and the main sort of negative impact on GDP per capita is the [02:37:59] the employment rate. So the the the fact that there just seemed to fewer workers per per capita in the economy >> just as a quick time check you have about two three minutes left. [02:38:12] >> Yep. Perfect. >> Great. Thanks. [02:38:14] >> Uh let me uh let me then finally finish off by just saying that after that I run a couple of of demographic counterfactuals in which I just say let's let's assume that the baby boom and this increase in old age survival [02:38:27] rates did not happen. Uh so I effectively just chop off the the hump shape in fertility that we observe in in in fertility during during the boom years and uh and pretty much eliminate [02:38:40] the entire uh increase in in adult life expectancy uh from 1950 onwards. [02:38:47] I then take those two scenarios, run them first one at a time and then run one scenario when when both of them are are active at the same time uh and decompose the the baseline output into different demographic factors. So in the [02:39:01] baseline uh GDP per per capita grows by a cumulative 27.5%. [02:39:06] Uh from 1950 to the year 2100 the the baby boom corresponds to about 8.2 points of that. the mortality the increase in old age mortality is around three points and the remainder is around [02:39:19] 14.7 points. So the baby boom and the increase in in old mortality can you know jointly explain roughly half of this increase uh with the mortality impact sort of slowing down at the end [02:39:33] but for the most part it it's it's roughly on the same order of magnitude as as the baby boom. [02:39:39] Uh and finally that the the other uh the remain the remaining factor here is you know it it's a result of the fact that you have a non-stationary population in the 1950 that's going to generate some transition dynamics anyways uh plus the fact that you still have some positive [02:39:53] population growth during this time which increases the population uh which generates uh TFP growth through through this scale affecting inside indogenous growth models. [02:40:05] Finally uh before uh closing off uh I run a scenario in which we shut down uh the the indogenous growth mechanism altogether. [02:40:17] So uh and that's a special case of the model and when we do that rerun the model the the growth rate effectively shifts down proportionately. So what we now find is that this positive 27.5 percentage point increase in GDP per [02:40:30] capita in a baseline becomes a negative 12.2% 2% uh when we when we remove this TFP channel and ju just as a sanity check I compare this to the Kruger and Ludy paper which I think is the the [02:40:42] closest uh comparable paper to mine and and during comparable years when I shut down the indogenous growth channel we effectively get the same uh response u when you compare that to the baseline that sign flips and becomes a positive [02:40:56] 6.3%. So whether or not demographic change during this period increases or decreases up per capita turns out to depend crucially on whether or not we sort of uh add this uh endogenous growth [02:41:10] channel uh in in the in this model framework and you can also see that this is purely driven by the TFP uh difference. So when you plot the difference in in GDP per capita growth and technology growth within between [02:41:25] these two scenarios, they effectively overlap. [02:41:29] All right. So I'll I'll close by that uh and just say that uh this paper sort of shows that the US demographic trends between 1950 and 2100 raises up per capita once we acknowledge the the [02:41:41] impact on technological progress. Uh so I'll leave it at that and you know be happy to take your questions. Thanks. ## Q&A (02:41:48 – 02:58:38) [02:41:48] >> Fantastic. Thank you so much uh Marcus. Excellent presentation. Thank you for staying right in time. So the floor is open for for questions Adrian. [02:42:03] Yeah. Uh super interesting. So thanks. So I have a question that's a bit related to you's question earlier about um about the mechanism and the role for population growth here. So, [02:42:17] so you see you have this risk structure on kind of consumption and savings. But I think for what matters for innovation is kind of the current level of population growth as opposed to something that has to do with say the age structure of the population. Let's [02:42:32] say suppose that we wanted to think about who's contributing to TFP increases at a point in time. Perhaps that would have to do with the a structure of the population or or something. But but here I think it's just a current population growth rate if [02:42:47] I'm not mistaken. And so and so then that's going to be so that's going to be pretty sensitive to assumptions about say fertility in the UN projections because I think if you take the median projections you know you you still see [02:43:01] that there's population growth towards 2100 but in the low fertility projections that's going to be totally reversed and also some of that is being sustained by momentum and I think by uh [02:43:14] um and by im immigration and So, so, so, so I just wonder how how sensitive are your results even to kind of fertility scenarios in a UN projection and more generally if you think about digging [02:43:26] into that box of what is the thing that like how could we be richer about the specification of um innovation that doesn't just rely on population growth [02:43:38] but on something else uh uh and and what the consequences of that would be. [02:43:45] Um yeah, that's a good question. So on the first one, I mean the um obviously the the as I said, this is a paper that focuses on demographic changes that do not really affect long run population growth. That was sort of [02:44:00] the the the whole uh premise of it. So if you're uh imposing a a model in which the fertility rate forces the the population growth rate to become negative in in the long run, you [02:44:13] know that then you're effectively in the Chad Jones AR paper uh scenario. Uh so that's a little bit of a different question here. I'm more interested in what happens when the population structure itself changes when longevity [02:44:27] generates people uh to to save more in for for retirement and so on. Um so I but but I >> so I understand that but the the indogenous growth mechanism itself in your model doesn't interact much with [02:44:41] the a structure at all I think because it's like it's all it's it's all about the current population growth rate at a point in time. That's what feeds into into TFP improvement. [02:44:52] No, no, it's it's a it's a function of of both the amount of of investment which is very much a function of the population age structure uh and of the population size and growth. So you have both the scale effect and an R&D [02:45:06] intensity effectively. Uh the R&D intensity works pretty much like a like a capital intensity uh channel, right? I think I have a slide where I can actually show that. Uh let's go here. So [02:45:19] when I I can decompose the overall TFP. So here's between the two steady states, the one in 1900 and the one in like 2,600. [02:45:28] There I can exploit u um steady state relationships to decompose that into an R&D intensity which is effectively the result of changes in household behavior, the saving rate and so on. And one factor which is just scale which is what [02:45:43] we think about when we think about semi-indulgence growth. And you see that uh you know changes in the population structure and their behavior is very much a factor in this uh this overall shift in in TFP. [02:45:58] >> Thanks. >> Great. Uh Larry, >> uh your microphone's not working. [02:46:11] >> Sorry. Uh lovely paper. um and very very clean and well presented. The um so I had a few questions one one of which just connects the last one which is maybe it's young people that are [02:46:26] producing the technological change. We have this 15year-old or no 25year-old telling us the world's about to um with 10% probability um disappear or at least [02:46:38] humans are. uh he's 20 25 um and if you look at all the tech bros are the ones who are really making the difference in AI. I think they're all quite young. So there's that that you [02:46:53] might add. I'm not saying you want to subtract anything to your cool paper, but you might want to add some things uh about techn you know other ways in which technology is impacted by the [02:47:05] demographics. But um and that's just um kind of repeating what was just said. [02:47:10] But the um the other question I have is if you're simulating the US, are you taking into account the fact I mean clearly people didn't know in 1950 that uh Nixon was going to double index social [02:47:25] security and that Medicare would be introduced by Johnson in whatever maybe 68. Uh you know we have these major increases in these Ponzi schemes that are being fiscal Ponzi schemes that [02:47:40] would impact if they they were anticipated that would impact the um path of saving and investment for far you know for forwardlooking life cycllers. So there's that question [02:47:54] but that's more relevant to the to the transition up till the present and then from that on that point on uh it's it's not uh that relevant for kind of projecting forward. uh the the last [02:48:09] question is or the second question is of about foreign capital flows whether you're introduc or incorporating that and I forget what you're doing with bequest whether we have a good chunk of [02:48:21] I think most of our bequests are coming involuntarily because I don't see any evidence whatsoever of altruism in in our country people uh if you look at um [02:48:34] the um the nuclear family, parents and adult children, they don't share uh their their consumptions do not move together whatsoever. They're not sharing their resources. Indeed, they don't even know [02:48:49] about what each one each other has. So, I think but they do leave their money to their kids if they haven't been able to spend it. So um anyway I was just wondering how how you're treating u [02:49:03] international capital flows bequests and also fiscal policy. [02:49:08] All right. Uh good questions alto together. So on the first question, the age structure of of innovators, right? [02:49:14] Uh given that I have uh R&D investment as sort of the input into the into the uh innovation process, that R&D uh investment is produced by the final [02:49:27] goods production sector which features households that have a hump-shaped uh productivity profile. So to the extent that you know uh the productivity profile of the population changes over [02:49:39] time uh that's already captured in indirectly through that I also have a version where I just use labor directly as the R&D input uh and the results are effectively the same. What matters there is whether or not the the production [02:49:54] profile of households that are innovators are necessarily different from the ones producing goods. uh and I haven't taken a stance on that because I think the the the empirical evidence seems to be a bit scarce on that point [02:50:07] with respect to pensions. uh so what I'm assuming is that uh pension or fiscal policy but the pension is the key uh aspect I'm assuming that uh the replacement rates [02:50:20] are effectively fixed over time and and so uh and that's going to generate tax changes uh to to balance the pension system budget uh when the population ages and this is something I mean there's perfect foresight so so people [02:50:33] do take this into account uh to to me to me that's that's sort of the most grow pessimistic version that I can have in this paper. I do not include an open economy channel uh for simplicity. I'm using this as a [02:50:46] sort of uh conservative benchmark because if you start thinking about capital flows or international a global economy, I would expect this TFP channel to be much much stronger due to the scale effect uh that's active in in in [02:51:00] semogous growth. So so I'm uh to me this is a conservative benchmark uh and and it simplifies my life a lot. [02:51:09] Uh, and then with bequest. So, so I'm only assuming that uh, so there's a warm globe request motive. People leave beests because they like it and then the government takes whatever beests are left over and redistributes that as lumpsum transfer to to the households. [02:51:22] Uh, and and again to me this is a conservative choice uh, modeling shops here because the it dampens the savings response for virtually all households in the model. [02:51:38] Great. Thank you. So, we are about at time. We're going to eat into the break a little bit. We will start the next paper in 10 minutes. But how about, you know, people want to take a break, please uh feel free. But why don't we ask Diego, you and Luis to each put [02:51:52] their questions on the table, and then Marcus, you can keep answering them uh as we go through the the break time. So, Diego, >> thank you. Thank you, Marcus. I don't want to rob your time. If you need to [02:52:06] rest that's totally fine or if you want to keep engaging maybe because you feel motivated about it happy to ask questions. [02:52:14] >> Yeah. >> Fantastic. So uh my question is the following and I'm just trying to see if I understood well the the chart showing the transition uh dynamics in your PowerPoint [02:52:28] um or your PDF. Um the result is that the demographic transition generates faster growth between the 1950s and the 2100s than if it didn't occur. [02:52:42] >> Yeah. >> The the net positive result for growth is positive. But say I am I am standing in 2026 >> and my population is about to start aging or it's already aging. So I I I I don't care that much about the past. I just want to know what's going to happen [02:52:57] to growth from now on. And I know my historical growth rate over the past decade has been X. Is that going to go down because of demographic dynamics? Is dynamics is demographic going to be a [02:53:10] limiting factor for growth now? Uh because that's that to me is more interesting for policy than to know what happened between the 1950s and 2100. [02:53:20] >> Yeah. Um, so >> if you can just hold off on answering that, let let's just let allow Yui and Luis to put their their questions on the table, then you can respond to them all. [02:53:32] You >> uh Yeah, great. Uh, so just I'm just going to follow up my previous question a little bit more. So I'm wondering like how uh much your result is going to change if you assume that in the long [02:53:46] run uh it's uh it's the economy is going to be on the balanced growth path of a constant population decline as John's uh AR paper is considering and you know it may be like that results don't change [02:54:00] much but I my understanding is that like that like where you're going to shoot is very different based on like the analytical characterization of what he shows in that paper. So I'm what like I think it's important to know like [02:54:13] whether you know your results hinges on on you know seeming like positive results hinges on that particular way of solving the problem uh rather than sort of you know quantitative results that [02:54:26] you have based on your data and uh inputs. [02:54:31] Great. Good, good question. Then Luis, do you want to jump in? [02:54:35] >> Yeah, thanks. Uh, I'll be quick. It's just a small almost clarification. I was just kind of wondering um what happens in in your model to the return on capital over the transition. I was kind of thinking how it squares or not with [02:54:49] the basically Adrian's paper and like their prediction that there's falling returns to wealth. Of course I understand as it was discussed that this depends on international capital flows and it's kind of not you can't really compare directly but still I was kind of curious. Thank you. [02:55:03] >> All right. So um on the first question if I understood correctly this was I mean the way to think about it is that um yes so the model shows that GDP growth [02:55:17] has has uh benefited positively over this entire period. If you were to look at it from this point on, uh you you when you saw the the sort of main figure, the the growth impact from demographics effectively remained around [02:55:31] zero for the remain from around 2030 up until the end of the century. So from from that viewpoint, from this framework, you wouldn't really expect it to impact growth either up or down. Um so so that that's the sort of short [02:55:45] answer on that on uh on the balance growth being negative. Um the the the you know I don't have a very good answer. I haven't run the model [02:55:58] with negative balance growth f uh population growth. Uh I could do that. [02:56:05] Uh the reason I and I'm saying this this again, the reason I did it this way is because when you uh if you were to simulate this type of model with uh a shift both in the long run population [02:56:18] growth rate and you have these temporary uh demographic shifts in terms of the population structure and longevity are going on. Uh it becomes a bit hard to disentangle what what is uh you know you have two effects that are sort of going on at the same time. The reason I did it [02:56:32] was just that this allows me to sort of isolate one of these effects. Uh and the re and the reason then being that there are other papers that are already looking at what happens to if if long run growth becomes or long run population growth becomes negative. So [02:56:46] uh I mean I wish I had a better answer but but I haven't uh looked at model with with negative population growth. U that's probably something I could look at at some point for sure. [02:56:58] And then the the third clarifying question. So the interest rates in this model they they decline as well. Uh in in the uh in the baseline economy the interest rate falls from around 4.2% in [02:57:13] the in 1950 to around 2.2% in in at the end of the century. So around 2 percentage points. Um when we look at the 21st century that drop is also around 1 percentage point. I don't remember the exact number but it's it's [02:57:26] in that order of magnitude. So uh uh so so so I replicate that qualitative pattern and I think most whole G models that do these exercises tend to to replicate that that type of behavior. [02:57:41] >> Great. Well, we have two or three minutes more if anyone wants to continue the the discussion informally. [02:58:07] Okay, I'm going to hand off the the uh moderator duties to Adrian for the next session. So, uh, Joe, if you want to just make sure you can share share [02:58:20] slides in the meantime while we're waiting for the top of the hour.