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

# The Global Transition – The Impact of Demographics and AI on Economic Power

Authors: Seth G. Benzell, Laurence J. Kotlikoff, Victor Yifan Yifan. Ye, Plaid  
Discussant: None  
Video: https://www.youtube.com/watch?v=w6ObGDwelyg&t=13230s


## Talk (03:40:30 – 04:11:36)

[03:40:39] next paper is the global transition impact of demographics and AI on economic power and Victor is presenting.  
[03:40:46] So you have 30 minutes. >> Uh awesome. Can everyone hear me?  
[03:40:51] >> Yep. Cool. Well uh good afternoon. Um thank you all for uh being here. Then I'm presenting joint work with my two co-authors uh Seth Benpenzel and Larry Cotlikov on the impact of demographics  
[03:41:05] of AI on uh global economic power. Quick acknowledgements views expressed in this presentation do not reflect any of the views of uh any of the institutions that either me or my coursers are affiliated with.  
[03:41:20] as um we've kind of been talking a lot about there's um pretty significant uh decline in global fertility and I think specifically even though global fertility has declined since the 60s the  
[03:41:34] rate it at which it has declined more recently has been perhaps faster than what people expected. So figure one shows the global fertility rate um  
[03:41:46] between the 2017 versus the 2024 um UN demographics. This is the the medium variant. And as you can see seven years is not a very long time in by demographic standards, but the revision  
[03:42:00] the size of the revision is is pretty striking and negative.  
[03:42:07] So needless to say, a major demographic revision is going to have a lot of very significant kind of force order uh macroeconomic uh implications. And one I think topic  
[03:42:21] of maybe particular recent interest is how a decline in the quantity of labor is going to interact with perhaps labor substituting uh AI or automation technologies. And that's precisely the  
[03:42:35] question that we are um trying to answer.  
[03:42:39] The way that we will try to answer this question is by utilizing a large scale 17 region global overlapping generations model comprising over 98% of world GDP  
[03:42:53] and about 99% of world population. Our uh quote unquote experiment is to calibrate and run the model twice. So we calibrate the model to a set of  
[03:43:04] identical macro fiscal targets but using two sets of different demographic projections. So under respectively the UN's projections as of 2017 and 2024.  
[03:43:16] In so doing any difference in what the model projects we know for sure that it exclusively comes from the difference in demographics and and and nothing else.  
[03:43:27] The way that we model AI is in each region for the firms in each region, we allow them to adopt a frontier technology that is  
[03:43:39] moving at a rate that is um faster than what the historical pace of share shifting technological advance I guess was. Um and then finally we do two additional  
[03:43:52] demographic sensitivity analyses. uh one concerning US immigration and the other one using the low fertility variant uh of the UN demographics.  
[03:44:01] To give everyone here a quick um preview of our results, unsurprisingly, under either set of demographics, the model projects pretty substantial global aging effects. We  
[03:44:13] project a very significant uh decline in the world real world interest rates through the end of the century due to aging and and essentially a a global savings club.  
[03:44:27] Pensions are hit pretty hard. Um assuming that uh per capita pensions are indexed to GNI, we project that the global average payroll tax rate needs to roughly double by end of century.  
[03:44:40] Moving from the 2017 to the 2024 projections, we project will lower global output in 2100 by roughly 10%.  
[03:44:49] And I think somewhat strikingly, it will leave China and the US by 2100 with roughly equal um GDP shares of of of global output. And for some context, under 2017 demographics, uh China is  
[03:45:02] projected to be more than twice uh US GDP by end of century. So it's a it's a very significant kind of shift.  
[03:45:09] The US benefits a lot from AI technologies. Um so under our preferred definition of transformative AI and I'm happy to to expand that in a couple of slides.  
[03:45:20] The um US share of war output increases to about a quarter by end of century. So for reference that's levels that were last observed for the US um immediately  
[03:45:32] after the uh the second world war. That all being said, our US results all depend very critically on um immigration assumptions and want to later as before going into the results in more  
[03:45:48] detail. Just wanted to uh give a very quick overview of our model. We split the entire global economy into 17 regions.  
[03:45:59] Um each of these regions contains 101 overlapping generations. uh comprising three skill groups. So low, middle and high. Each agent in each generation, each skill group they work and save and  
[03:46:13] they consume and we assume they have perfect foresight. Um we have a calibrated region specific labor force entry and retirement and we also have idiosyncratic mortality at all  
[03:46:24] ages. So um and and um requests are are all uh unintentional.  
[03:46:32] The model features a single global good that is produced using capital and the three um aphformentioned types of labor and each region also has a a government that aggregates state and local and  
[03:46:46] federal. Um and it does all the things that governments typically do, taxes, transfers um and then pays pensions.  
[03:46:56] And um in in in all the scenarios in our current analysis, we assume that through time they hold their uh respective debt to GDP ratios constant.  
[03:47:08] So quick look at what regions are in our model. We have the US, we have China, India, we aggregate Western Europe and um Japan, South Korea, uh etc.  
[03:47:20] For demographics, we in our model, we use the fertility, mortality, and migration rates, which are all age and region specific. And >> Victor, before you go ahead, just just a quick question on the last point you  
[03:47:34] made about holding that to GDP constant. So, so you've got this big demographic imbalance like you pointed out and you've got this imbalance pension system. So, how do you So, what do you do to hold the debt to GDP ratio constant? Is it is it like  
[03:47:48] government spending goes down or what? >> Um so we we hold um all the government spending here is indexed to GNI. So we we we we basically use taxes as the  
[03:47:59] lever. So we assume that um given each region's current mix of income and consumption taxes, we keep that relative ratio constant and then we adjust the overall rates up and down to to to balance that.  
[03:48:12] >> I see. So it's like US tax rates are much higher than they currently are because >> we're able to >> you know population gets older less income.  
[03:48:20] >> Okay. >> You know >> so >> also pay payroll taxes are separately adjusting.  
[03:48:30] >> Yes. >> Yeah.  
[03:48:32] >> Well payroll taxes are keeps pensions uh >> to balance social security.  
[03:48:36] >> Exactly. >> You're not running down the trust. Nope.  
[03:48:39] >> So, so per per retirey benefits are indexed to to GNI. So, and the way the reason we do that is to isolate the effect of of demographic changes from from you know benefit level changes.  
[03:48:52] Um so um the model is calibrated to the our two um runs of the model calibrated precisely to identical fiscal and macro targets. um those are derived from the  
[03:49:04] UN IMF um central bank data and demographics is is is the only difference. We do have exogenous regional um TFP catchup that's not something that we can generate from the  
[03:49:17] model itself. Uh those we calibrate to uh Mueller stock and Watson which is a paper uh projecting long run uh GDP per capita growth. So we we take those numbers and then we we use them to infer um regional specific region specific TFB  
[03:49:32] catchup. And then as mentioned before uh pension benefits per retirey are we index them to per capita GNI and this is just to um it precisely isolate out the demographic  
[03:49:45] in terms of AI. Um every year for firms every region so the US, China etc. We give them option of choosing between two production technologies. They can always stay on their quote unquote legacy technology.  
[03:49:59] So as of 2017 that that never goes away that's always in the back pocket. Or they can choose a gradually advancing through time more capital intensive frontier technology. So technology that has larger capital shares. Um in this  
[03:50:13] scenar we also define as as relatively speaking higher um high school labor shares.  
[03:50:20] and um the rate at which that frontier technology evolves we calibrate to um the uh US's historical experience between 1980 and 2017. So in other words, what we're doing is we're looking  
[03:50:35] at how have capital and labor shares shifted in the past 30 35 years and then we project those rates forward by another 30 35 years um in in in our  
[03:50:46] baseline and our definition of AI or transformative AI the more extreme scenario is basically we take that process and then we make it much more aggressive we make it much more capital  
[03:50:58] intensive um and um uh firms design on technology based on profitability. So every year the firms look at two technologies, see which one produces more output and then you know add equilibrium factor prices and then make  
[03:51:13] a determination kind of from there. This is just a quick um overview of what the um shares in the production function look like. So these are all for the US  
[03:51:26] for the front technology in 2050. Um our baseline calibration uh has the US at 34% capital share in 2017. So our our baseline business as usual scenario entails the US capital share increasing  
[03:51:40] by about three percentage points um for the next 30 years which is consistent with historical kind of experience. And then for the transformative AI scenario the the more extreme AI scenario our frontier technology capital share by  
[03:51:54] 2015 increases to 60%. So it's it's it's pretty um pretty significant. We assume that the capital shares are coming proportionately from the three types of labor. So as you can see the the 37 to 60 it takes about 10 percentage points  
[03:52:08] away from high skill labor. It takes about six percentage points away from uh low skill labor.  
[03:52:14] So how different are the two demographics or how different is 2017 versus 2024? U pretty pretty significantly. So this is uh global population. So from the 2017 to 2024 UN  
[03:52:28] projections um 2100 global population is projected to decrease by about 7% so about 700 million people and um the dotted lines that we extend we basically we have an assumption where  
[03:52:42] um after 2100 demographics will will gradually stabilize through time.  
[03:52:46] was full stabilization by um 2150 the negative revision in overall you know population is not equal across  
[03:53:00] regions uh obviously China I think is really hit the hardest in terms of of population count um between 2017 and 2024 its projected 2100 population is revised down about 40%. So it goes down from a little bit over a billion people  
[03:53:14] to a little bit over 600 million people. The US's population projections are actually revised upward. And the reason for this is not fertility. The US is still projected to have lower fertility  
[03:53:25] but um there was um a very significant increase in the rate of assumed net immigration which is why the immigration scenario is is so important.  
[03:53:38] So that's kind of the stage. So what did this revision do to global GDP? Um the main takeaway from this table is if you  
[03:53:50] look at the 2017 UN demographics by 2100 our model projects that the US will claim 11% of global GDP by 2100 while China has 25.6 so 26 close to 26%. So  
[03:54:03] the so under 2017 demographics we project that China's economy in 2100 is going to be more than twice as large as the US. 2024 completely changes that. So now the US is 14.4, China's 14.9. Those  
[03:54:16] are roughly kind of equal. So demographics is um a very large negative for China's economic power. It should be mentioned and um there was you know um Dur's paper cut earlier today. Uh we do  
[03:54:29] also find that output per worker is effectively unchanged by the changing demographic projections. So this relative share shift is coming almost entirely from number of people  
[03:54:41] essentially or number of workers. Um we can also show this view in um GDP levels. So these are uh 2017 US dollar uh trillions of 2017 US dollars. um  
[03:54:55] global GDP goes down by 10% um and um US GDP goes up by um by 15 uh trillion 2017 and 2100 between the two uh demographic  
[03:55:08] wages and China's GDP goes down by by a huge amount by um more than hundred trillion dollars to >> to clarify Victor. So so when you're looking at this you're saying in your model by 2100 what are relative GDP shares?  
[03:55:22] >> Yes. And this is GDP level. So yeah. >> And but how much do the general equilibrium effects matter here? Because it's like all relative GDP. So is it is this just basically plugging in the difference in the population sizes in UN  
[03:55:35] projections 2017 versus 2024? >> Yes.  
[03:55:38] >> So this is just saying there was like a 50% revision in the number of Chinese people by 2100 in those projections.  
[03:55:46] >> It is like the the vast majority of this effect. There are some general equilibrium effects. So for example um a place like subs Africa. So because the world is aging and the number of workers is is smaller this kind of shifts around capital demand. So if you have countries  
[03:56:00] where there's a lot of population they actually in relative terms they suffer from from having less uh capital. So this is why like subs Africa subsar Africa which also has slightly fewer people but um the demographic  
[03:56:15] projections are are are not explaining all of subsar Africa. This is also because as these you know more developed countries age they well they supply more capital but they also um due to the technological adoption they demand more capital and that's being sucked away  
[03:56:29] from the countries that have a ton of people.  
[03:56:32] >> So in your model there isn't there equalization of capital labor ratios.  
[03:56:38] Aren't the capital labor ratios sort of relatively stable? Um because the only thing that's can that can drive this is the interest rate and that's kind of equalized across countries.  
[03:56:50] >> The interest rate is equalized across countries. Yes. The >> So then how can capital intensity change as interest rates decline >> in because we we give the countries the option of the two technologies. So in when when when countries age they're  
[03:57:03] more likely to go on to the the newer more capital intensive technology.  
[03:57:08] >> Okay. I see. So that's that's that's that's that's the that's like most of the remaining the non-demographic part of these results. Okay.  
[03:57:18] Um global interest rates. So um due to um basically a global savings glut we do project that global interest rates are um declining pretty  
[03:57:29] dramatically through the century. Um this is essentially all coming through the ratio of the world's holding retirees increasing significantly. um relative to workers. Um and the demographic revision does not change  
[03:57:44] this by by very much. These are all real rates. So this is there's no sense of inflation or anything here. Um and and because the 2024 world is even older, the 2024 projected rates are are indeed  
[03:57:58] slightly lower than 2017, but the demographics are not a not a huge driver of of of interest rates.  
[03:58:06] Um pensions are very stressed in in both cases pensions are very stressed but in particular the 2024 demographics have an incremental stress on on pensions. Um we  
[03:58:18] estimate that the as a share of global GDP world pension outlays um are projected to um roughly double between uh this is sorry this is between um just 2017 and and 2100. So this is not not  
[03:58:33] the two vintages. Two vintages are are are are relatively similar. So they're about 1 point uh 3 percentage points apart in terms of the the u the cover payroll rate. The thing to also read here is um  
[03:58:47] China's cover payroll rate goes up pretty dramatically. Goes up by by 10 percentage points. The US is 2100 rate however is very little changed. And then the reason is you know going back to this different this combination of  
[03:58:59] updates in fertility and immigration um the US pension results here depend critically on the assumption that the US is going to be importing a lot of people.  
[03:59:10] So moving on to the AI um scenarios. So this table contrasts um so remind um just to remind everyone so in our model we have two different technologies for  
[03:59:25] each country and then one is less cap intensive one is 2017 one one is the more cap intensive one. This table shows um which year the different regions move onto the new technology um from the  
[03:59:37] legacy one under our three um different AI scenarios. So a lot of regions uh will adopt the frontier technology immediately and they never come back. So the US for example, Western Europe um  
[03:59:51] and as the technology becomes more capital intensive many regions will actually delay adopting this technology and the reason for this is essentially there is so in the short run there is only so much  
[04:00:05] capital to go around. Over the long run, savings rates will adopt. But if we immediately give countries a much more capital intense technology, the countries that have the cheapest capital and the most expensive labor will adopt  
[04:00:19] those technologies first. But once like once the US moves on to the transformative AI, it's going to suck up the world's capital and people in China, people in Europe, they will find it they don't want to invest in their uh  
[04:00:33] domestic economies anymore. They want to invest in the US. And because of that it becomes less attractive for those regions in relative terms to to move on to the the TI technology. So that's why for example China's adoption of the the  
[04:00:48] baseline um automation is in 2024 but China will actually wait until 2042 to adopt the the transformative AI.  
[04:00:57] Um so um for and this is needless to say this is very good news for the US economy um for  
[04:01:11] under the 24 demographics the more updated version so with the transformative technology we project will essentially double US share world GDP uh in both 2050 and 2100 relative to  
[04:01:24] the the business as usual baseline. So um and the reason is basically that the US has very expensive labor. The US has has um cheap capital and um  
[04:01:37] um for context I think these are so the last time the US comprised a quarter of global GDP was around 1950 so immediately after uh World War II.  
[04:01:47] In terms of GDP levels, China also gains um under the TI but it's global shares are essentially negatively impacted because they have to delay adoption due  
[04:01:58] to um shortrun uh capital uh scarcity and it does not so for the US for for China it does not um it is not able to reverse the um negative GDP impacts of  
[04:02:11] the the 40% less people for um what does AI do to global interest rates? So  
[04:02:23] if we have a so assuming a a frontier um technology that has 60% capital share by 2050 which is very aggressive this is more than sufficient to fully reverse  
[04:02:34] the global savings club. So um by 2050 the TI technology is projected to raise global interest rates from about 3% to over 6%. So it is it like um a TI  
[04:02:49] technology of the magnitude that we model is more than sufficient to keep interest rates at above 6% for essentially the the rest of the century.  
[04:02:58] There is some um household response here as well. I think this is worth highlighting that part of the reason why the interest rate response here is so sharp is because as households in these regions as they um they have perfect  
[04:03:11] foresight they will see that in the future there is going to be more output and they will prefer to pull that consumption forwards through time. So essentially as they're seeing the future economic abundance um coming from TAI  
[04:03:26] they are trying to actually save less immediately and um this is um so that this kind of adds to the interest rate kind of problem. Uh Ron uh  
[04:03:39] I see your comment. So I I think that's a really fair point. So our our kind of model here like we're not we're not trying to like second guessess like you know fiscal decisions like that are um or like potential fiscal  
[04:03:54] decisions that may be made in the future but I think it is on a on a practical level it's very difficult to imagine that countries will have a 40% payroll tax although it should be mentioned that like per calibration the effective European rate right now is already  
[04:04:09] exceeding 20%. So um a lot of countries already have very high kind of payroll tax rates >> but but I mean if you're not trying to second guess what governments are doing I mean what they are doing is running deficits right and and they're not keeping that to GDP ratio constant at  
[04:04:24] the moment at least and for a while and I mean CBO does not project a fiscal consolidation by 2050.  
[04:04:31] >> Yes. and and and and for for a graph like this, this matters a lot because right now you you know in your baseline with that transformative AI you you have declining interest rates but if governments continue to issue debt at this pace even in your model that  
[04:04:46] reverses that prediction. >> Yes, that that that's that's 100% fair.  
[04:04:51] Um and I I I know um this is this is like I think one of the two reasons like you know it bridges like the gap between these findings and what you have in in your paper which is well one that we so  
[04:05:05] like we're implicitly assuming that China will still save a lot um in the at least in the media future and that is going to push down global capital prices. That's something that I guess we will have to see what what kind of happens there. And then the other point  
[04:05:19] as you mentioned is that we we we do have this like you know very conservative kind of fiscal like uh culture rule like we're we're basically saying you know we're not we're not going to go crazy with debt and we can we can simulate this with maybe a more  
[04:05:34] realistic US debt scenario. That's something that we could do to maybe try to bridge some of that.  
[04:05:38] >> Well just to be clear this is pretty pretty crazy to begin with with ending up with 40% payroll tax rates.  
[04:05:45] uh you know I know that's pretty uh expansive fiscal policy uh across the transition. So um and a related point you said you said it's kind of neutral but one thing about  
[04:06:00] very high payroll tax rates is they do affect savings incentives quite a bit because you're you're keeping benefits constant and low and and raising payroll tax so lowering income during work life and that's affecting the savings  
[04:06:14] behavior. Right? So if you wanted to do something more neutral, you could say that doesn't affect directly the savings incentive, you could you could you could change government spending, say, right?  
[04:06:23] And I think that that's kind of what often is done in those models to try to to try to not affect the life cycle savings decisions.  
[04:06:31] >> Um yes, and we actually um this is in the appendix paper, but we have um we have a scenario that kind of partially maybe gets award to say like Adrian. So we we instead we we keep the the 2017 levels of the committed expenditures by  
[04:06:44] 2050 fixed which is often much much lower than um what expenditures would be needed to keep the benefits constant and that that that does raise global GDP. So that's that's that's the mechanism that you're referring to which is running 40%  
[04:06:58] of payroll taxes is is not good for growth.  
[04:07:02] Um so um and then oh sorry the the last point about the air scenarios is that um on the on the global level and again you know due to the fact that in the short run the pool of capital that is  
[04:07:17] accessible is is limited and actually shrinking due to the households wanting to consume. um even a very aggressive technology is in our projections is not actually going to increase global GDP  
[04:07:30] that much until the second half of the century. So in our in in in this projection actually by 2050 global GDP is lower under the TI scenario and this is just because all the capital is being sucked up to developed countries. So subsort of Africa is really suffering um  
[04:07:44] you know India is is is falling behind. So um those dynamics are are are also in place.  
[04:07:53] Um there is some fiscal relief at least for the US coming from TI. Um we project that under TI the effective US tax burden on on total income. So this is essentially the incidence of of um all  
[04:08:08] spending on all income roughly halves by 2100 but it doesn't really help with uh with pensions. So assuming similar levels of um  
[04:08:20] um shares of of of or like assuming that the pensions are are are funded at at a similar rate with a specific payroll tax, TI will nonetheless still significantly increase um statuto kind  
[04:08:33] of payroll rates. So um in for example um oh this should say per capita but in in places like western Europe and China maintaining per capita output index benefits um in the TI scenario would  
[04:08:46] require labor tax rates that are approaching 100%. So um when we when we do the TI scenarios we do actually have I think we we we cap payroll taxes at at 45% because and the reason for that is because a lot of these places already  
[04:09:01] have an income tax that's about 40% average. So pushing it beyond that would would push the model dangerously toward the point where tax rates are 100%.  
[04:09:10] Um so and and I think the the conclusion we draw from that is assuming that AI does advance very aggressively that this is I mean a general point I think cess and I have have made in other papers um financing of of of pension  
[04:09:25] systems would almost by necessity have to shift toward either consumption or if there is enough gross capital taxes.  
[04:09:34] Um so have a yeah we have like 10 minutes almost so I want to just quickly kind of talk through the two uh sensitivity analyses. So the first one very simple um all the other countries  
[04:09:46] are not changed we but we we turn off US immigration. So for for the US only we run um the UN variant with zero migration. This does mean that global population is changing by a little bit.  
[04:09:58] We we chose this way of doing it to essentially precisely isolate the just the the first order immigration effects on the US economy and those are pretty huge. The 2024 UN demographics uh double  
[04:10:11] the assumed US net migr uh migration and if we take that away US population 2100 is only about 270 million compared to 420 million in the the 2024 base case  
[04:10:23] and the US share global GDP falls pretty far. So it falls down from uh roughly 14% to only about 9%. So um basically yeah so um around around 30 to  
[04:10:37] 40% of US GDP in 2100 in our projection is attributable to net um immigration inflows. Um in either way the output losses are just confined to the US. Um  
[04:10:51] the global response of this effect is pretty small.  
[04:10:56] are um maybe once in our other um sensitive analyses basically the same setup we just run it with the low variant of infertility instead of the medium variant. Um this kind of further shrinks global GDP by about a third. Um  
[04:11:10] that being said in relative terms this is good for countries like the US because you know as we were discussing just now uh the US results are are essentially driven by immigration. So if everyone's fertility is going down but  
[04:11:22] the immigration paths are are relatively speaking less less changed then that is good for the US economy and um that is pretty much it comments questions  

## Q&A (04:11:36 – 04:20:21)

[04:11:36] great so thanks so much so we're just just in time um and so we'll open up the floor for questions Maybe as we wait for questions when um  
[04:11:57] one comment I had a bad the modeling is so so I guess um labor supply in your model is exogenous. Is that right? So, so you don't have this sort of indogenous labor  
[04:12:10] uh response that this tipped me up when I I saw your 100% tax rates on labor and and I would I would assume that you're on the other side of the lefer curve at this point. But >> yes, >> but but in your model that's not a  
[04:12:24] concern, right? You still get you you still actually get just as much of a tax base.  
[04:12:29] >> So labor supply is indogenous. Um but the per our calibration, it is fairly rigid. So I I think the the way I think I would maybe think about that is um  
[04:12:43] basically >> yeah the income effects overcome the substitution effects.  
[04:12:48] >> Exactly. Yeah. >> Better than I could.  
[04:12:52] >> I see. So so you don't have a peak to you actually don't have a a peak to that Africa. If I just thought about how much tax revenue can I generate out of a certain tax rate.  
[04:13:00] >> No, we we don't. We have a and and with the um right we we don't have a peak. We have CS utility function and consumption leisure >> and u you know if you look at Russia's  
[04:13:13] payroll tax rate it's already 32%. >> So the idea that you can't live with that kind of a tax rate it's not the case. Well, I mean I think 40% maybe not so so crazy, but I guess 100% at this at  
[04:13:26] this point to the other side of the Africa.  
[04:13:31] >> We're we're around 40% at the end here. >> Right. Right.  
[04:13:34] >> And uh Yeah. And then with the uh transformative AI, it's 24%. So it it keeps it down.  
[04:13:44] >> Is it or not? Maybe I'm wrong in that. Maybe it stays up with even with transformative AI. I guess it does, right?  
[04:13:51] >> Yeah. I think agent directly like the really high rates, they do negatively impact GDP, but they don't I think like per our calibration, they don't they don't collapse the system. They're just people people are definitely working less hard at some point, but but yes,  
[04:14:04] but the LER curve doesn't peak. >> Yeah. The other thing to keep in mind here is that when the US uh in the these scenarios ends up in the best case scenario for the US with 25% of world GDP and China's down to 14 at the end of  
[04:14:19] the century rather than what we projected back in 2017 basically that China would be 25% of world GDP and we would be 11%. That was our earlier  
[04:14:30] study. uh we have to realize that under this new study a lot of the world's a lot of the GDP in the US is being is owned by foreigners right a lot of the  
[04:14:42] the capital is owned by the foreign owners because we're calibrating to US saving behavior which is terrible which is really you know we don't save like the Chinese or the Indians or the Asians  
[04:14:55] in general save so a lot of the capital ends up um we we on massive c current account deficits along the transition.  
[04:15:04] >> Right? But is it accurate to say that we shouldn't think of this as  
[04:15:17] um I mean I would think of this as an analytic projection that lets you look at sort of what if this and that happened to the demography but not an  
[04:15:28] attempt to sort of forecast the future which I imagine would look quite different because just the idea that the outlook for the demography of the world could change so dramatically in seven  
[04:15:42] years and then be perfectly believable as you know a sort of fixed and believable projection thereafter is kind of strange but of course we do that sort of thing all the time. What's what  
[04:15:57] makes it seem so strange here is because you're also looking at what how the world looked in 2017. Well, how was it going to look in 2033? Personally, I expect fertility rates to go up again  
[04:16:09] myself, but I'm in a small minority in that group.  
[04:16:17] >> Well, I you know, my response would be uh you just keep running the model and seeing where you project at the moment because we you don't know what's going to happen in the world. And that's our best uh you know, it's our best understanding  
[04:16:32] at the time. And uh I hope you're right on that one, Ron.  
[04:16:46] >> Ron, why do you expect for TD to bounce back?  
[04:16:51] >> Oh, I think we're in a transitory period of adjustment of our cultural values, our our public institutions. We have insane pension  
[04:17:04] public pension policies all around the world. Uh there's going to be a generational uh resistance to this. The dead weight loss effects of rising tax rates are  
[04:17:17] going to become important. And uh I think um for the most part with some exceptions I think most younger people want to have more like two two children and they feel like it's not very  
[04:17:31] possible under current uh economic conditions and I think a piece of this is population aging crowding out u public expenditure on children uh higher  
[04:17:45] education that we forcing children to pay for more themselves. Um the working conditions of uh young families as they contemplate having another child and so  
[04:17:58] on. I think these things are going to change. But uh can I point to any evidence that that's so? Not really.  
[04:18:08] Except uh looking at uh current survey responses on uh how many children people would like to have. But but do you think people will internalize the effect that having children has on things like  
[04:18:22] sustainability of pensions? Because if it's just about their own incentives, isn't it the case that we're just continuing to tax them more? And like what is the direct incentive to them of having more kids?  
[04:18:33] >> Well, you could imagine all kinds of policy changes. For example, you could make pensions depend on uh a person's fertility.  
[04:18:41] >> There have been proposals to do that sort of thing. I I mean I think governments are increasingly are to turn towards uh you know stronger measures to try to  
[04:18:56] get people to have more kids but I think mostly uh I think there'll be increasing realization of the situation and I think that will lead to policy changes and to  
[04:19:10] changes in individual attitudes and changes in our concept of what old age is and what what an old person is. Uh the next paper I think is going to be  
[04:19:24] exploring some of the possibilities that just arise from the fact that our our healthy lifespan is going up together with our life expectancy. I think it's going to be intolerable to society to  
[04:19:36] see healthy, vigorous, cognitively intact older people uh you know imposing taxes on the younger population so they can enjoy 30, 40 years of leisure. I  
[04:19:49] just don't think that's a a socially, culturally, politically, humanely sustainable situation.  
[04:19:59] So yeah, those are my views. >> Very interesting.  
[04:20:07] >> Any other questions, comments? >> So thanks a lot Victor um Larry for