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

# International Friends and Enemies

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=wT0i4i-JMUY&t=0s


## Talk (00:00:00 – 00:27:06)

[00:00:10] i would like to give the floor to errands new on well the slides are already up so uh the floor is yours for the next 20 minutes all right hi everyone um thanks very  
[00:00:24] much for having this paper on the program it's joint work with benny kleiman who's a phd student at princeton and with my senior colleague stephen redding so the broad motivation of this paper is  
[00:00:37] really by two related questions so over the last few decades we have seen rapid economic growth both in china but also in other emerging economies and as a result of that we've seen drastic change in the relative economic size  
[00:00:51] of nations and a classic question in trade is what's the impact of such economic growth on the income or welfare of other countries around the globe or the trading partners and a related question in political  
[00:01:05] economy is whether such changes in economic size heightens political tension so there's some parallels in terms of the current u.s china tension but also at the beginning of the 20th century the tension between germany and uk but  
[00:01:19] also in asian greece between the athens and spartans so we contribute to both of these questions providing both the new theory and also some evidence so in particular we developed bilateral what we call  
[00:01:34] friends and enemies measure of how countries income and welfare are exposed to foreign productivity shocks or growth so the outcome of the analysis will be two matrices where each row or each column represents a country  
[00:01:48] and an entry of the matrix captures how does one countries productivity shock affect the income or welfare of another country so these are sufficient statistics that can be computed directly using trade data  
[00:02:01] and our approach is based on the quantitative trade model so we develop these sufficient statistics using the leading class of trade models characterized by a constant trade elasticity and what we do is essentially we  
[00:02:14] linearize this class of trade models so that we can represent everything in terms of linear algebra and just some matrices as sufficient statistics so because we do linearization our sufficient statistics are sort of exact  
[00:02:27] for small productivity changes and in fact we actually both analytically and numerically show you that our linearization provides almost the exact answer to the full nonlinear solution  
[00:02:40] of this class of trade models and we will talk about intuition for why linear linearization basically gives you the full solution to what's called hat algebra in trade now the advantage of our linear approach  
[00:02:53] is first it's extremely interpretable so it reveals underlying economic mechanisms behind the quantitative trade results which could be perhaps seen as kind of black boxy if one were just to apply the trip model and do quantitative work  
[00:03:08] we review underlying mechanisms and we are extremely computationally fast so computing counter factuals using trade models is not that slow unless you want to compute millions and millions of counterfactuals which is exactly what we do  
[00:03:22] and specifically what we do is we look at 140 by 140 countries each each year so each year we have 140 by 140 calorie fractions and we look through 42 43 years across time and see how bilateral  
[00:03:36] exposure changes over time and we do various decompositions so how does tfp shock in china affect the us and we decompose what's the role of say singapore canada or mexico in generating that exposure between china and the us  
[00:03:51] and so on we the the methodology we developed here holds for both the classic armington or eden courthouse model but also various extensions to it so the multi-sector version or input dapper linkage version or the econ geography version  
[00:04:06] the same kind of formula the same set of formula holds broadly for that kind of models so specifically what we do is we compute the first order effect of productivity shock from a country on the income and welfare in each other  
[00:04:21] country and we show that the trade flow the bilateral trade flow matrix essentially is sufficient to capture that effect in particular we derive from the bilateral trade flow matrix  
[00:04:32] three derivative matrices one of them is the expenditure share matrix so how does each country spend its income in terms of buying from other countries and the second is the income share matrix so for each country  
[00:04:46] what share of income does it derive from selling to particular markets and the third is the cross-substitution matrix which captures how does rising competitiveness in one country say china causes global consumers to  
[00:05:01] substitute away from the demand of another country say the u.s and in the eden courthouse or armenian model cross-substitution matrix is essentially the product between income share and expenditure share matrices  
[00:05:14] so we use this linear or matrix representation to reveal many underlying mechanisms uh in the in the talk i will showcase how we do these decompositions so we show that the general equilibrium income exposure of one country due to  
[00:05:28] another country's productivity growth can be decomposed into market size and substitution effects and the welfare exposure then depends on the incumbent exposure and cost of living effects it's easy to separate the whole effect  
[00:05:42] into partial and general equilibrium and as i mentioned we can also calculate the contribution of individual markets or individual sector in propagating this exposure between any pair of countries so after deriving  
[00:05:56] these sufficient statistics we're going to showcase how we use the toolbox using actual trade data we're going to sort of examine over time how has global exposure of income or welfare to each other's  
[00:06:10] productivity has changed over time and we're going to show that our linear resolution is almost exact with the is almost exact to the non-linear solution of trade models and finally we're going to show that as countries  
[00:06:24] become greater economic friends they also tend to become greater political friends measured by voting in country countries voting in the united nations general assembly and also measured by strategic rivalry indicators categorized by political  
[00:06:38] scientists okay that would be the last part of the talk i'm going to skip the literature review but i just want to point out that we contribute to this large literature on or growing literature on a sufficient statistic approach  
[00:06:51] in international trade and let me get into the derivation first let me start with a general armington model so what is an armington model we have a finite number of countries each country has some labor endowment  
[00:07:05] and we produce directly and linearly from labor and consumer there's a represent consumer in each country having homophobic preferences and they have a preference for variety that's why they trade with each other  
[00:07:20] so in armington models you know the welfare or the utility or real income of consumers in country n is equal to the nominal income divided by price index and the exact form of the price index depends on the  
[00:07:34] specification of the preferences and the price index depends on a vector of prices that consumers in country and faces in particular pni is the price that consumer and pays for good i  
[00:07:47] imported from country i which is equal to you know factor price in country i divided by productivity in country i multiplied by a iceberg trade cost okay and we are always going to use n as the  
[00:08:02] importer or consumer and i as the producer or exporter country so the market clearing condition that defines the general equilibrium armington model or essentially a budget balance condition is that  
[00:08:15] total income earned in country i is equal to the total payment by all of its uh by all of its consumers so how much does country n pays to country i where sni is the expenditure share  
[00:08:30] in country in country and on good eye where the expenditure share of course again depends on the relative prices of different goods that consumer and faces and it depends on the exact utility specification that  
[00:08:44] generates some expenditure function okay that's the structure of arming to model so we are going to totally differentiate this arming the model with respect to productivity changes and try to measure the response of  
[00:08:57] income or welfare changes so we take a total derivative and here's what we get you know the equation looks a bit complicated but it's actually extremely interpretable so let's think about you know country  
[00:09:11] h having a productivity improvement so think of h as china and i as the us how does a tft shock in country h affects the income in country i or in the us well when  
[00:09:25] china gets a productivity improvement chinese goods become cheaper so it causes consumers around the world to substitute towards chinese goods and away from goods produced by the u.s so each of these  
[00:09:38] third market is um is indexed by country uh is indexed by the number n so think of n as representing singapore malaysia canada mexico so as chinese goods gets cheaper consumers in all of those  
[00:09:52] countries shift towards chinese goods away from the u.s and the exact degree to which that affects u.s demand is captured by the cross-price elasticities so these datas are the cross price elasticity  
[00:10:05] in each of the markets around the world with respect to chinese prices on u.s demand and you know the effect of tfp shock mitigated by the cross-price elasticities affect income or factor prices that's the  
[00:10:19] partial equilibrium effect that's the first round of uh global exposure however as these factor prices changes that also affects production cost and then feedback into second and higher round of general equilibrium effects  
[00:10:33] okay so general equilibrium is essentially a fixed point to this equation now the final term that i haven't talked about is what we call a market size effect so because global consumers have different expenditure shares so think of you know canada consumers spend a lot of  
[00:10:48] their income on u.s goods more so than singapore so an increasing canadian income is really good for the u.s relative to the increase in singapore income and this first term captures that exposure  
[00:11:01] to to the u.s consumers okay so this is what we get by differentiating uh the market cleaning condition with respect to a productivity shock in order to get the exposure in terms of income  
[00:11:15] and in terms of welfare well the welfare change in a given country is just the income change net of the change in price index which depends on production costs around the globe weighted by the expenditure share of consumers in that country  
[00:11:29] so the cross price elasticities of course depend on the preference specification and in the classic ces preferences this cross price elasticity simplifies drastically  
[00:11:42] it just becomes the trade elasticity times the expenditure share so consumers in singapore are more elastic to chinese prices in terms of their demand to us if the trade justice is high or  
[00:11:57] if singapore consumers spend a high fraction of their income on chinese goods okay that's what this equation says so we can stack these equations into a vector form and that's our friends and representation  
[00:12:11] so this makes m captures all the cross price elasticities and captures the first round or the partial equilibrium substitution effect when you have productivity changes causes substitution and effect of income or factor prices  
[00:12:25] and then factor price feeding to higher rounds of substitution effect and income also affect market size that's the effect that i mentioned in the u.s gets a high fraction of income from selling to canada and canada gets richer that's good for the us  
[00:12:40] and in terms of welfare exposure again welfare effect depends on income exposure and the cost of living effect so here you know p is the income share matrix as is the expenditure shared matrix and  
[00:12:53] this cross-substitution matrix m is just the product between t and s so what's the intuition there well when china gets a productivity improvement that causes global consumers to substitute away from u.s goods  
[00:13:07] the extent to which consumers substitute away is captured by their expenditure share on chinese goods and how important is each market for us income is captured by the income share matrix so the substitution matrix is just a product between the income sharing and  
[00:13:22] expenditure so we need to add all the exposures due to each of those markets aggregate them in order to get the exposure between the u.s and china okay so we can we can sort of rearrange the term of these matrices  
[00:13:36] and write them in close form to get a bilateral representation of friends and enemies and that's what we're going to do now the theory can be applied if there's trading balance if we have actually trade cost changes and we can  
[00:13:51] handle small departures from constant trade elasticity and the theory extends to various extensions of this model so the multi-industry version or the cost no donors donaldson commander version of trademodels  
[00:14:03] or the input output version of calendar parallel as well as the econ geography version of this model so i want to walk through the intuition of the multi-industry version because we're going to apply that to the data  
[00:14:18] so essentially we have the same exact equation and same interpretation t is again the income share that one country derives from selling to other countries as is the expenditure share that one  
[00:14:31] country spends on goods produced by other countries and this m matrix captures this degree of substitution the difference between multi-industry and single industry version is a definition of market in a single industry version you know  
[00:14:46] consumers in singapore canada and mexico all substitute towards chinese goods and away from the u.s goods when china gets a productivity improvement in a multi-industry version and market is now defined as a country sector so think of it as  
[00:15:01] textile sector in singapore which or when consumers buy textile goods in singapore they switch to chinese textile and away from u.s exile and we have to add all of those effects across all industries and  
[00:15:14] all countries in order to capture the aggregate exposure between u.s and china so that's essentially the difference between single and multi-industry version of this model so we're going to take this matrix  
[00:15:27] representation to data and do a bunch of decompositions to showcase what this methodology can do okay we're going to use fairly standard international trade data  
[00:15:40] um across 43 countries uh 43 years 140 countries and we're going to supplement that with uh income and population distance data because we're going to estimate gravity equation in some of the specifications  
[00:15:54] so first we showed that the counter fracture that the linear counter factor our results generate is visually indistinguishable from the full results of the noun of the non-linear solution of the model okay so we plot  
[00:16:09] wealth exposure due to our linear approximation and due to head algebra they exactly line on a straight line and if we regress one on the other the slope coefficient is exactly one and r squared is exactly one  
[00:16:23] now why is that approximation so good well the non-linear solution in our linearized solution coincide with each other if countries are in authority or if countries are in free trade and real world matrices trade matrices is  
[00:16:38] exactly a weighted average of the identity matrix when countries are in otarki the free trade matrix which is basically saying exporter size is highly has high expenditure power in the trig flows and plus some noise okay we provide a  
[00:16:52] theoretical characterization of the quality of approximation by looking at the norm or eigenvalue of the hash and terms we show that the norm is tiny so that approximation is almost exact okay so we take the tool set to global  
[00:17:08] trade data and we compute bilateral welfare exposures across countries first of all the bilateral exposure is increasing over time consistent with increasing globalization second we also show that the dispersion  
[00:17:23] in the exposure term is also getting larger over time so this box and whisker plot shows the interquartile range and the fifth and 95th percentile so in any given year when one country gets productivity improvement in terms of the fifth  
[00:17:37] percentile welfare exposure among all other countries is actually negative so some countries lose out from productivity growth in other countries this is just to compare the effect of chinese productivity growth on u.s  
[00:17:51] germany and japan over time the broad conclusion is whenever there's chinese productivity growth it harms u.s income but benefits u.s welfare because u.s consumers buy so much chinese goods the price  
[00:18:06] index effect is shows up in the welfare but doesn't show up in the income equation so chinese growth is bad for u.s income but actually improves u.s welfare and relative standing of the u.s around the world is actually harmed by chinese productivity growth so  
[00:18:21] u.s share of world income relative to all other countries always declines when china gets a product to do growth now we use this bilateral exposure matrix to do a clustering or community detection using to maximize modularity score  
[00:18:35] essentially try to arrange countries into clusters to maximize within cluster linkages and minimizing cross-sector linkages in 1970s you know the world is centered around the u.s other big countries include germany and great britain  
[00:18:50] in 85 we see the rise of japan and the rising well-defined european cluster and asian cluster is centered around the u.s in 2000 um asia is centered around japan  
[00:19:04] there's a north american cluster and a european cluster centered around germany and 2012 the last year of a sample china actually has the biggest productivity exposure to all other countries  
[00:19:16] uh overtaking the us okay now we use this tool or this bilateral exposure to examining regional pockets pockets of regional economies so for example bilateral exposure in asia where the size of the  
[00:19:31] outgoing arrow captures the size of exposure one country has injured productivity growth on the welfare of another country so going from 1970 to 2012 we see the declining influence of japan in asia  
[00:19:43] rising influence of china and rising influence of singapore declining influence of malaysia we do the same for eastern europe around the fall of iron curtain we do the same for north america uh around  
[00:19:57] nafta okay we can decompose the importance of each individual market in shaping bilateral exposure so think about u.s exposure to due to chinese productivity shock well these are the important markets um that  
[00:20:11] that propagate that so why is singapore important well singapore has a high expenditure share on chinese goods and canada is important because u.s derives a lot of income from selling to canada okay  
[00:20:25] now even though we are considering a uniform productivity shock in a given source country it actually has heterogeneous impact on across industries due to different compared advantage or different industrial composition across countries here's a  
[00:20:40] particular interesting fact that is if we consider input alpha version of the model and think about the uniform chinese productivity gain across all countries in factory asia around you know regional economies in east asia  
[00:20:54] the sectors that gets harmed are all you know light manufacturing sectors textile office equipment and so on but electrical equipment and medical equipment these more high-tech manufacturing sectors actually gain so these sectors relatively expand  
[00:21:09] in east asia due to regional production linkages the pattern is drastically different if we look at resource rich or commodity exporters so they experience a form of dutch disease in general equilibrium so even though  
[00:21:22] china gets a uniform productivity gain these sectors the sectors in these countries the sectors that relatively expand are the primary sectors basic metal mining and agricultural factors so we do various decompositions in the  
[00:21:36] paper that i wouldn't have time to go into but we also can easily compare the welfare impact across different versions of trade models essentially using multi-industry or input output highly correlated with the single sector model but magnitude is a  
[00:21:51] slightly larger in terms of the input upper version okay now the final bits of the paper is we link the economic exposure matrix that we we come up with with political exposure political  
[00:22:06] distance across countries so what we do is we look at how countries vote in the united nations general assembly and we have essentially we use measures come up with uh by political scientists on  
[00:22:19] countries distance or similarity score in their voting we have a host of such measures based on u.n general assembly voting we also use indicator variables for whether countries are engaged in strategic rivalries which are also categorized by political scientists  
[00:22:34] based on contemporary persistence by perceptions by political decision makers of whether each other whether countries country pairs are competitors threats or enemies so we use both of these measures as the left-hand side variable  
[00:22:48] and we see whether the economic exposure that we have is predictive of these political exposures so of course this could be highly endogenous you know as countries get closer politically they trade more and  
[00:23:02] because they trade more that affects the welfare exposure measures so the strategy we come up with is we estimate and a gravity equation for each year and we use the changing coefficient of distance in gravity  
[00:23:16] to predict bilateral trade flows and use the predicted bilateral trade flow due to time varying coefficient of distance as the instrument variable for the welfare exposure to causally get effect off welfare exposure on political exposure  
[00:23:30] so for each pair of countries we compare with control for a country fix a country pair fix effect and time fix effect and we find that as countries get closer or in terms of more positive wealth exposure towards each other they also  
[00:23:44] become greater friends politically meaning they vote more alike in unga or you know united nation general assembly and they are less likely to engage in strategic rivalries  
[00:23:57] these results are robust across a wide range of specifications as well as different measures of wealth exposure using different versions of trade models and across both type of types of political measures  
[00:24:12] so let me conclude we developed a bilateral matrix representation of exposure to global shocks so the paper focus on foreign productivity shocks but the method holds more broadly for also for trade constructs  
[00:24:27] the methodology holds in the acr class of trade models with a constant trade elasticity as well as various extensions some may even break the assumptions underlying acr a representation is a linear linearization  
[00:24:41] but we show in the paper theoretically numerically that the linearized counter factor results basically gives you the full nonlinear solution in this class of trade models and we actually develop a theoretical bound for departure from the ces assumption so you can study  
[00:24:56] general unmeter model using the methodology as well the approach is vastly computationally superior to nonlinear head algebra and it's vastly more transparent you yield sufficient statistic that isolates underlying economic  
[00:25:10] mechanisms and now we apply that to the paper to do various decompositions and finally we show that as countries become greater economic friends they also tend to align better politically in terms of ung voting or in  
[00:25:25] terms of strategic rivalry measures thank you thank you so much ernst um this is uh we want to open the the discussion here if not immediately i had one question i  
[00:25:43] i understand from a practical point of view why you would rely on the uh the commodity trade flows that uh that are that are in the data that you described but  
[00:25:57] to what extent would you expect these patterns to be different uh if you could also incorporate service flows in trade flows and services and um well and i see that berthold  
[00:26:12] wants to join the discussion as well so go ahead let's collect the questions yeah thank you um i was surprised by the fact that your model does so well because it's a linear approximation and usually we think about  
[00:26:26] china as having caused big changes so i would have expected some nonlinear effects to go on can you comment on that or is it a sequence of many small steps  
[00:26:38] that amounts to a big step and you are looking at the small steps and then you can approximate them well um i mean i thought it was very nice but i was surprised how well it did  
[00:26:53] further uh questions okay um ernst why don't you go ahead david also  

## Discussion (00:27:06 – 00:33:34)

[00:27:06] had a question i saw who what's uh oh sorry david go ahead hi hi ernest very nice paper um i so i was trying to follow the  
[00:27:19] multi-industry one i was trying to think through whether that was um useful for thinking about gains to winners and losses to losers it just seems like now more than ever in trade it's obvious  
[00:27:34] that there are people like me who teach chinese graduate students and buy cheap chinese goods it's no no question that um i gain from trade with china there's a whole host of people who feel like  
[00:27:47] and probably do suffer it just seems like that within country heterogeneity you know you can always make that comment you know an aggregate model i was just trying to think to what extent can you guys get at that in this multi-industry version or whether you think maybe just this is  
[00:28:02] not the paper to think about that issue but overall very nice paper this is a clarifying comment on trying to be a helpful comment not a criticism okay let me let me get to these um  
[00:28:16] questions first on service flows to the extent that we observe bilateral trading service this same framework exactly applies and in the presence of non-tradables if we know sort of the expenditure share in  
[00:28:30] non-tradables then we can scale sort of our measures accordingly by the share of non-tradables um so this is more like a toolkit given whatever data you have if it's cross-country commodity flow  
[00:28:44] it can be applied if it's if you have data input linkages you can do the inputable version if you have services or data on expenditure sharing services you can do the service version as well so in the paper  
[00:28:56] uh we we do something quite crude in terms of computing services when we do the multi-industry as well as the impadable version we basically take the service here in the u.s and scale all the wealth and  
[00:29:09] measure accordingly by that so in response to bert's question um this is the comparison so we were very surprised by by the fact that the approximation does so well not  
[00:29:23] only for small shocks but also for large large shocks so even if you a country gains productivity by a thousand percent the approximation is almost exact so the graph i showed you the high r squared the fact that nonlinear solution  
[00:29:37] linear solution almost coincide hold even for large shocks we were very surprised by that that's why we did quite a bit of work to understand why that's the case i didn't really have time to go through just now  
[00:29:50] so this is the nonlinear equation this is the linear equation the difference in the first line is in the nonlinear equation it's taking logs of averages and we are doing averages of logs and  
[00:30:05] then of course the linearization involves second order terms and higher voter types the linear and the non-linear equation holds exactly or they are exactly equal if countries are in atarki if countries are in free trade  
[00:30:19] or if the deviation from those cases are exactly in terms of noises that cancel out because we're just doing log averages now the real world trait matrices has a strong diagonal component that's the authority matrix  
[00:30:33] the real-world trait matrix the exporter size really explains a lot of variation in the real world trade matrix evidence from the gravity equation that's the that's basically the free trade matrix under free trade the expenditure share on china in each  
[00:30:47] country is actually proportional to china's size so any residual from the weighted average of identity and free trade matrix is essentially noise and that partly speaks to why gravity equation does so well so if you you know this the world is a  
[00:31:02] sphere so if you take two random countries and plot a distribution of distance between two random countries that's essentially noise that gets cancelled out by the log by the log average so to formalize that you know  
[00:31:16] the linearization involves second order term and higher order terms and we can show that the norm of these hashes matrix in the second order term can be used to bound the size of second order term  
[00:31:29] and can actually be also used to bound the size of all higher order tracks so we compute those passions and we compute those norms we compute those bounds analytically we show that given real-world trait matrices our approximation is basically  
[00:31:44] exact so the non-linear effect in this class of trade models are very small so that it was a surprise for us we didn't know that before engaging in the project  
[00:31:55] so to answer david's question the the multi-industry version of the graphs i showed exactly gets at that uh at that question so in the paper we don't really  
[00:32:09] emphasize that so we think you know each country has a representative consumer and part of the consumer is employed in different sectors and so on but if you think that employment is sticky within sectors these cross-industry exposures exactly  
[00:32:23] gets at that question and the contrast here is huge in the sense that east asian economies you know just a uniform productivity change in china that benefits all of all chinese sectors benefits these high high-end new manufacturing sectors in  
[00:32:38] the east asian countries due to regional production chains and really harms the light manufacturing which china has especially the competitive advantage at but if you compare with the resource rich or commodity exporters really you know the sectors that  
[00:32:53] relatively benefit are the commodity sectors because the the light manufacturing sectors gets destroyed due to rising chinese productivity so if if we think that employment is sticky or if we don't no longer have a representative consumer  
[00:33:08] within each country this kind of decomposition exactly answers that question we're thinking about how to package that result better but currently this decomposition is how we touch upon that issue in the paper  
[00:33:22] all right thank you so much for this for this this interesting talk