# Industrial Policy with Ernest Liu | Markus Academy | Ep. 128

https://www.youtube.com/watch?v=QfPTdjYJR8U

[00:01] nice to see you all again and welcome back to another webinar organized by Princeton for everyone worldwide
[00:08] we are here today to talk about industrial policy with Ernest Lou also from Princeton University
[00:12] hi Ernest good to see you
[00:14] good to see you Marcus thank you
[00:15] thanks a lot for telling us today something about industrial policy
[00:21] before we pass on the mic to you a few opening remarks uh by myself
[00:25] industrial policies big in the news
[00:30] this I took from a recent IMF working paper and you can see how dramatically essentially industrial policy is in the news
[00:39] so it's way more mentioned and it's actually one of the big issues people talk about in the policy circles
[00:45] if you look at a recent paper by Danny Danny rodri and others you can also see the indust industrial policy is also in action
[00:52] so there's way more implementation of industrial policy in the recent years as well
[00:55] so it's not only in the news but it's also applied excessively or to a large extent uh in
[01:04] in the real world by various policy makers
[01:06] just think of the inflation reduction act and many many other chip act and many many other measures which you can see as a industrial policy action
[01:15] Now what is industrial policy and what are the motives
[01:19] of course one motive is to correct for some market failures
[01:21] so one would be externalities instead of using beian taxes you try to uh correct some other externalities
[01:31] there could be some agglomeration failures which the market doesn't doesn't fully internalize or you know in particular some infant industry where you promote some new industries or also on the science side DARPA in United States where there's a lot of innovation going on which Le the groundwork which leads to subsequent industries
[01:51] there might be some increasing returns to scales economies of scales which you cannot exploit fully because inoodle firms might not have the financial capacity not the access to financing to set this up think about large language models you
[02:04] know that's essentially only few companies can actually do that or there might be also coordination failures.
[02:08] which need to be corrected some firms might need the action of other firms and they can't coordinate to invest in the same direction at the same time and so the action would be to bring players together or it might be that we need some public inputs some public goods some activity specific ones and you can think of Cil is a very important public good whereas the system is resilience.
[02:32] think about Supply chains and you would like to have some resilience if something would go down if something doesn't work that you can actually substitute and scale up some Alternatives very easily.
[02:45] you might also want to do some industrial policy for geopolitical reasons because there might you might be subject to blackmailing if there's Market power and so certain choke points you want to avoid and you know might be some national securities and the sanctions which might also fall in a broad sense of industrial policies and
[03:05] Finally there also Regional policies.
[03:07] Which are uh you know to some extent industrial policies.
[03:10] So all of these motivations lead governments and to intervene and uh conduct certain industrial policies.
[03:19] But the key essentially is that you need some good governance uh because industrial policy invites essentially some grony capitalism.
[03:26] So the question is which firm should you favor which sector should you favor and by how much.
[03:31] That's an information problem.
[03:33] How should the government find out which firm is the innovative one which changes the economic landscape down the road.
[03:42] And you know will in the future will the good jobs be really be with the manufacturing sector with the industries for industrial policy or will we moved much more to the service sector and can industrial policy be extended to the service sector as well.
[03:52] But in general the industrial policy invites a lot of lobbing effort and Martin wolf in a reason article in the financial times put it nicely.
[03:57] He says governments may fail to pick the winners while losers.
[04:07] May succeed in picking governments.
[04:09] I think that puts it very nicely.
[04:11] How you know the dangers of industrial policies.
[04:15] The other question is, should you, you favor some upstream firms or downstream firms or within the network?
[04:18] I think interest will tell us more about that.
[04:22] And should you favor some micro innovations or should you favor some moon shoots or you know a mission-driven innovation?
[04:30] Some broader sense, where should should you subsidize individual R&D projects or should you just buy the end product of something?
[04:40] But in general, it's essentially industrial policy is not promoting competition.
[04:42] It's is favoring one or two firms or one or two sectors.
[04:48] And the question is, when is this justified and can governments do a good job?
[04:52] And what governance structure do we need in order to ensure that?
[04:58] Now, if you put this whole thing in internal context, it has of course implications.
[05:02] If you favor certain industries, it will all the industry will be more competitive at.
[05:07] International landscape and that actually will lead for a tit for tat relationship with other countries and other trading blocks.
[05:13] Some countries will favor certain industries.
[05:15] Others will then put on some tariffs because I think there's unfavorable treatment of their own industries if they don't put up tariffs.
[05:24] And this could undermine all the international globalization uh which we have experienced the last few decades.
[05:28] So that's essentially what some of these challenges and uh that's essentially what's going on now.
[05:36] The poll questions ear put forward are the following ones and let's see how you voted.
[05:41] Um the first question is very much related what we discussed.
[05:44] Can policy makers really successfully identify market failures and select the right sector to promote that?
[05:56] And the answers are what you have given us is 38% thought yes the government can do it they're very good at that and 62% no the government cannot do it.
[06:05] So the majority almost two-thirds say the
[06:09] government can actually not do it now.
[06:12] some of industrial policy is regional.
[06:14] competition great incentive to use.
[06:16] industry policy to accumulate economic leverage.
[06:18] this is a negative zero or positive sum gain.
[06:21] negative says 33% Z.
[06:25] said 29% and positive sum games 38%.
[06:29] it's roughly equally split.
[06:32] positive sum game gets the edge with 38%.
[06:34] and the third question is is the Revival in interest in industrial policy among academics but also among policy makers is it a Fed is it here to stay or it's simply recasting of ongoing ongoing policy debates like trade policy R&D policy.
[06:50] it's just a different way of phrasing things.
[06:54] it's just a Fed 7% set so it is here to stay.
[06:57] 37% say or it's simply just recasting some ongoing policy debates 56% said so.
[07:05] 7% 37% majority 56% that it's just a
[07:09] recasting of existing debates correcting Market fails we had before.
[07:15] so with these answers uh to Ernest questions uh I pause on the phone to Ernest who will Enlighten us about industrial policy and he will give his perspective attention also to these questions thanks again Ernest for doing it.
[07:32] great um thank you all for being here.
[07:34] thank you Marcus for providing this platform as a public good.
[07:39] I've benefited uh very much as an audience from past episodes so it's great to um have this opportunity to present here and the answers to the poll questions are surprisingly even split which I'm very happy to see.
[07:51] so part of Marcus introduction um also anticipated my opening spiel about how do think about industrial policy.
[07:59] so um in the context of this talk I'm going to define Industrial policy as the collective intervention by the government to
[08:11] promote or to push resources into certain key economic sectors of Technology with some kind of aggregate um objective in mind aggregate efficiency or economic growth to achieve to potentially selective intervention.
[08:26] now it's a um historically it's been widely adopted in many parts of the world especially in East Asia so many of the so-called success stories come from episodes in East Asia so in Japan in the 60s and 70s South Korea 70s 80s modern day China but there's also this revival as Markus shown of the interest in policy circles in the US and Europe.
[08:49] now industrial policy as part of uh the introduction that Marcus gave it takes many forms historically um it facilitates the formation of Bates or the chos in South Korea in China they take the form through state owned Enterprises occupying many of the key sectors in the Chinese economy they can take form of directed or subsidized credit to certain sectors or R&D.
[09:12] subsidies and more relevant in the US.
[09:15] the chip for America Act or the inflation reduction act for GRE.
[09:19] Innovation essentially nowadays it's become an umbrella term to cover government interventions that Target the supply side or selective intervention on the production side.
[09:30] historically it's primarily been um um the Narrative of industrial policy is to focus on correcting market failures or externalities for example taking advantage of learning by doing or inant industry and as part of the modern Revival it's being used or at least proposed to potentially address apply chain resilience to facilitate technolog uh technological growth and international competition or building up of international power through geopolitics and so on.
[10:00] so how to conduct industrial policy that is um a question that I will hope to um share some light on in the talk today.
[10:10] now conceptually because economic actors
[10:13] firms sectors Technologies countries.
[10:16] they all linked with each other through various connections through inut outp.
[10:20] linkages through spillovers through trade policy that Target one sector will unambiguously have Ripple effects do these linkages that affect many other other sectors.
[10:29] so it is important to think about these Network linkages when trying to identify the so-called strategic or pillar Industries or the key Technologies to promote.
[10:39] in fact hman who is an early Pioneer on thinking about industrial policy emphasize the importance of both input out linkages and trade linkages for thinking about both Economic Development and also building National powering and international stage.
[10:56] so in thinking about how to conduct industrial policy one need to recognize a key difficulty conceptually in assessing how effective a policy intervention is.
[11:07] it's just very difficult to evaluate whether industrial policy is successful that is what we can do um.
[11:15] what policy makers can do empirically is to do careful cross industry studies.
[11:21] compared before after outcomes using the defen diff strategy to compare outcomes between say treated and control industries.
[11:27] but such Studies by Design do not inform us about aggregated effects.
[11:33] so for instance you know as an outcome of the chips for America Act if we see that the US developed semiconductor industry because of the ACT do we view that as a success of the policy or not.
[11:45] in general you know these policies are designed with some aggregate goal in mind.
[11:49] so the fact that semiconductor as an indust you grow because of the ACT that's by design it doesn't uh Point towards success or failure of the policy design.
[11:59] to to evaluate whether a policy package or some kind of intervention is effective the proper kind of factor will be an otherwise identical economy that didn't promote the semiconductor industry.
[12:11] and such a kind of factor does an existing practice so that's where economic theory
[12:18] comes in that is you know we can use economic theory to think through the general equilibrium effects and to think through what are the aate implications of selecting specific sectors.
[12:30] so because you know um in general even with that conceptual difficulty addressed using economic theory we still need to think about carefully about what are the information required to implement some kind of theory in vieing uh in trying to understand industrial policy in of data.
[12:47] so for instance if if economic theory says we should intervene and promote sectors with high externalities but we can't measure externalities or it's hard to measure there are lots of measurement error then the wouldn't be very useful.
[13:01] so when writing down economic theories that guide industrial policy you know we as academics need to bear in mind on having empirical handles that are useful for policy prescription what are the observables in the data that can guide us with policy prescription and in this
[13:19] talk that empirical handle will be some kind of network characteristics.
[13:22] there are also political economy considerations um such as objectives or incentives policy makers that I will briefly touch up.
[13:30] so in this talk I'm going to um cover a selection of academic works that I've been part of in trying to identify different or distinct economic mechanisms that make certain sectors strategic in a network.
[13:48] and uh each of these uh studies I'm going to talk about will aim to derive some policy relevant measure from the data that we can actually measure in the data that's not so susceptible to measurement error.
[14:01] now this indal policy is a large literature with a long tradition.
[14:06] so the work that I've been doing built on many of the prior work and gets inspired from many of the contemporaneous work as well but the talk will mostly focus on the work that I have been involved in.
[14:16] so in um I'm
[14:19] going to talk about two papers that um relate close more closely to Economic Development and then two pieces of work that are closer to technology adoption or R&D that are potentially more relevant for as advanced economies like the us or the Europe so let's start by looking back in history.
[14:40] let's think about South Korea in the 1970s.
[14:44] what this figure shows is a visualization of the input output table of South Korea in 1970 so each column and each row represent an industry.
[14:56] The Columns are the input supplying industry and the rows are the input using industry and the size of the cell are drawn in proportion to the fraction of the column Industries output that's being used that's being sent to the row industry so this cell for instance this this particular column industry sends a lot of output to this row industry okay
[15:21] So, uh, the I industry in the row and the I industry in the column is the same industry.
[15:26] Hence, there's a strong diagonal here, and industries here are organized by standardized industrial code.
[15:31] So, typically, you're going to have agriculture sector at the beginning, and then light manufacturing such as textiles at the next group, then heavy chemicals, and then services in the, in, in, in towards the end.
[15:44] So, what do we learn from this input-output table?
[15:47] It seems like just a random figure that's completely chaotic.
[15:52] I'm going to do two simple operations to the table to try to bring out some pattern.
[15:57] First, I'm going to remove the small entries so that only big cells will remain.
[16:02] Second, I'm going to reorder the sectors according to upstreamness.
[16:07] So I'm going to first show, um, to make the most upstream industries appear first and most downstream industries appear last according to a notion of extremeness that's that's widely used in the literature.
[16:20] Un trust.
[16:22] at all so once I do those two operations
[16:25] rearrange industries by upstreamers and then we move small entries
[16:29] this is the remaining input outa table now I want to highlight a few features
[16:34] so how do we read this well think about the most Upstream industry
[16:38] this column is applying to many of the industries that are all relatively upstream
[16:43] and if we look across the rows the most the last roow which is the most Downstream industry it's buying from all the industries before it in a sense that the downstream industri is buy inputs from many of the upstream and this this network or this input out table is highly asymmetric it's almost lower triangular which means when once we order industries by upstreamness
[17:05] Upstream sectors don't buy from Downstream Downstream buys from upstream
[17:09] and there's this hierarchical structure in a sense that relatively more Upstream sectors Supply to Downstream sectors that are closer to it so you know this part is very sparse yet this part closer to the diagonal is very de okay so this
[17:25] is the hierarchical structure.
[17:26] Illustrated when we buy uh after we organize Industries.
[17:29] Can I ask picture look if we had a very perfect down.
[17:32] Upstream Downstream structure so no.
[17:34] Downstream firm would you know ever produce for an up.
[17:39] Firm just um so so if it's a vertical production chain meaning one Supply to two.
[17:47] Two to three 3 to four the only cell that would be activated will be off diagonal.
[17:52] Okay and if you know if one can supply to 2 three4 two can supply to 3 4 56.
[17:59] It would be perfectly lower triangular.
[18:02] Okay so this is much closer to a lower triangular Matrix and uh that's a feature.
[18:07] That's a it's a feature of South Korea shown here but it's a feature of many real world inab table.
[18:12] So some of the theory will take advantage of that feature.
[18:14] So why do we look at South Kore.
[18:17] 1970s well it turns out between 1973 1979 that's when South Korea actually.
[18:26] implemented this large scale industrial policy episode called The Heavy chemical Industry Drive where they dumped resources onto many uh onto six broad strategic sectors and a lot of the these are large scale interventions you can think of that as a big push intervention and many of the subsequent conglomerates were formed as a outcome of of intervention during this period so South Korea essentially subsidized steel nonfer metal ship building machinery Electronics petrochemicals it's called The Heavy chemical industry because of the industries targeted in this figure I'm highlighting the rows that correspond to industries that were targeted as you can see all the target industries were very Upstream okay so the heavy chemical Industries occupy uh only in the in in the first half of of of this input app table the heavy chemical Industry Drive HC HCI Drive cut targeted Upstream.
[19:27] sectors and uh
[19:29] that looking back in history that's actually a feature of also many of the East Asian interventions so one country that I had better data for in the context of China China also provided significant subsidies to these Upstream sectors so we uh look at sectoral value added that's accounted for by state own firms the fact that you know many sectors were occupied by state Lo firms which perceive explicit and implicit sub subsidies from the government that's a feature of sectors being targeted by industrial policy so here I'm plotting State Lo shares against upstreamness you can see Upstream sectors disproportionately have more State on firms so receive more resources from government so it's a common feature not only Enterprises are less productive than non-state own Enterprises how does this show up it doesn't show up here in this figure I guess it does not show up added for
[20:30] is this part of the value add of the
[20:34] firms yeah so here we are only
[20:37] calculating value added just regarding
[20:39] the fact that these firms may be
[20:41] inefficient so stateum Enterprise may be
[20:44] a very inefficient way for the
[20:46] government to promote certain sectors
[20:48] but it is a way to promote certain
[20:50] sectors regardless and what I don't have
[20:52] on the slides is if you so this figure
[20:55] is drawn using 2007 which you know
[20:58] that's why in many sectors the state own
[21:00] share is not very high it's high only in
[21:03] the very Upstream sectors heavy
[21:05] chemicals if you draw the same picture
[21:07] in the year 2000 or 1998 it would be a
[21:10] flat line that's about 60 70% state
[21:13] owned across many many sectors and the
[21:16] fact that you know it comes it goes from
[21:18] a uniformly high State own share economy
[21:21] to a state being more present in
[21:23] Upstream sectors it's not only due to
[21:26] selective privatization but in fact many
[21:29] new newly established plants in these
[21:32] Upstream sectors in early 2000s the
[21:35] state is not only selectively
[21:36] privatizing the downstream sectors but
[21:39] they also building new plant in these
[21:40] Upstream sectors it's an Evidence of
[21:43] active government intervention in in
[21:45] these sectors okay so question the
[21:49] audience Upstream Downstream
[21:51] categorization in terms of industry or
[21:53] firms within
[21:54] Industries yeah all of these are in
[21:56] terms of Industry so in the context of
[21:59] both so South Korea you know each each
[22:02] column and each row where these tiny dot
[22:05] represent would be a sector I don't
[22:07] remember the exact level of aggregation
[22:09] in in China it's about 120 sectors in
[22:12] South Korea I don't remember the exact
[22:14] number but similar order of magnitude in
[22:16] terms of number of Industries so I'm
[22:18] only
[22:20] me only manufacturing or Does it include
[22:22] Services as
[22:25] well so in the when I show a picture of
[22:28] input UPA table includes all Industries
[22:31] meaning Agriculture and services because
[22:33] all of those information are present in
[22:35] the input UPA table when I use firm
[22:38] level data in the context of China it's
[22:40] based on manufacturing surveys so in
[22:43] this figure only manufacting firms are
[22:46] present thanks great okay so the
[22:49] starting point of the first study I'm
[22:51] going to show is to try to understand
[22:54] can this type of intervention make sense
[22:57] you know a pick e because these seem to
[23:00] be success stories of industrial policy
[23:02] can the selective targeting of Upstream
[23:04] sectors benefit the aggregate
[23:07] economy so I will not show you the
[23:10] details of a model I'm going to lay out
[23:12] the economic intuition using a simple
[23:15] example and then show you some empirical
[23:17] measures in a data so let's consider a
[23:19] vertical production chain where there
[23:21] are three sectors of the economy an
[23:23] upstream sector think of it as producing
[23:24] Iron and Steel a Midstream sector too
[23:28] think of it producing machines and
[23:29] downstream produce textiles which is
[23:31] what the consumer actually care about so
[23:35] each sector uses the input from the
[23:37] previous sector as well and combine with
[23:38] labor in order to produce the
[23:40] output now Suppose there are some Market
[23:43] imperfections along these intermediate
[23:45] transactions so think of the market
[23:47] imperfections as Financial constraints
[23:49] but you can micropile it using
[23:51] Contracting fractions or markups and so
[23:54] on that is the textile producers they
[23:56] would like to buy more machines given
[23:59] the prevailing market price but they
[24:01] couldn't because of some Financial wedge
[24:03] so they're using inefficiently low
[24:05] number of machines per unit uh
[24:08] production and let's suppose the machine
[24:10] producers would like to buy more irons
[24:12] but they also Financial constraint in
[24:13] doing so what should the government do
[24:15] in this case well a tax put answer which
[24:19] is also a trivial answer is for the
[24:20] government to intervene and give
[24:23] subsidies and taxes to exactly offset
[24:25] the market imperfection then the first
[24:27] workare the hold that's a trival answer
[24:30] because of first s theorem that's also
[24:33] not very practical answer because that
[24:35] answer relies on accurately measuring
[24:37] the market imperfection if you can't
[24:39] measure the market imperfection with
[24:42] enough Precision government intervention
[24:44] may actually backfire and actually
[24:46] create more harm than good so instead
[24:49] I'm going to consider different exercise
[24:51] not what the government should do
[24:52] optimally but let's think about you know
[24:54] given that there are Market
[24:56] imperfections in a decentralized economy
[24:59] if the what's the B for Buck if the
[25:02] government can provide a margin of
[25:03] subsidy so I'm going to think about an
[25:05] inefficient decentralized economy and
[25:08] the government can collect some taxes
[25:10] from the consumer and provide in terms
[25:12] of production subsidies by targeting
[25:14] each sector only in a small way as a
[25:16] first approximation around a
[25:18] decentralized economy which sectors do
[25:21] the subsidy have the highest General
[25:23] equilibrium effects that's which sector
[25:25] we should prioritize in promoting
[25:27] according to this note notion of which
[25:29] sector to
[25:30] Target so Market imperfections in
[25:33] economy distort the use of inputs right
[25:35] so when taile producers they're
[25:37] constrained the production machines
[25:39] become inefficiently low and when
[25:41] machine producers are constrained the
[25:43] production of Iron and Steel become
[25:45] inefficiently low and labor which are
[25:48] the prod ultimate production resources
[25:49] they have to go somewhere they have
[25:51] markets have to clear so in this sense
[25:53] the the distortionary effects compound
[25:55] in a way that you know two little labor
[25:59] is used in Upstream production and too
[26:01] much may be used in a downstream
[26:03] production and this this notion of which
[26:06] sector is too inefficiently small and to
[26:09] which sector is too inefficiently big is
[26:12] robust to the size of Market Perfections
[26:14] no matter how large are the market
[26:16] imperfections as long as they exist
[26:19] marginally the B for the buck is always
[26:21] highest if policy tries to promote
[26:24] upstream and make it slightly bigger or
[26:26] tax Downstream and make it slightly
[26:27] smaller
[26:29] so in this sense subsidizing Upstream
[26:32] sectors create first order welfare gains
[26:35] now how effective what's the actual B
[26:37] for the buck depends on the size of the
[26:39] distortions of Market imperfections but
[26:41] the prescription that the first dollar
[26:43] going to Upstream will create aggregate
[26:45] gains that that is a robust prediction
[26:48] that doesn't really depends on how we
[26:50] how how large are the marketting
[26:52] Perfection so that's a feature that
[26:54] we're going to bear in mind when we go
[26:55] to the data yep you import distortionary
[26:58] taxes at the downstream firm in order to
[27:00] raise revenue to subsidize the Upstream
[27:03] firm
[27:05] so do you have some Le tax no I don't do
[27:09] that I think about tax as non Distortion
[27:11] I can create a lumpsum tax of a dollar
[27:14] in give it to different sectors the only
[27:16] thing that disto the use of the only
[27:19] thing that distort the economy is the
[27:20] tax itself not raise not from Raising
[27:22] fiscal Revenue just to simplify the the
[27:27] analysis so we can generalize this into
[27:30] just a general Network structure okay
[27:32] let's talk about the difficulty of
[27:34] measurement a bit later let's suppose
[27:36] that we can um we can measure everything
[27:39] in the economy meaning we can observe
[27:41] the input output structure of the
[27:43] economy and we can observe how large are
[27:45] the market
[27:46] imperfections in that context we can
[27:48] derive a measure what I call Distortion
[27:51] centrality which measures the ratio
[27:53] between how large a sector should be
[27:56] versus how large a sector actually is in
[27:58] equilibrium and this Gap this Ratio or
[28:01] the Distortion centrality has basically
[28:04] captures where the policy should
[28:06] intervene should spend the first dollar
[28:09] so this notion of distorted centrality
[28:11] it's a scaler per sector and has several
[28:14] intuitive properties that are go
[28:16] question I just want to understand so
[28:19] Theta is a matrix and D is is a matrix
[28:23] or is it a vector yeah D is a matrix in
[28:26] the sense that every pairwise
[28:29] interaction can have its own distortions
[28:31] some Goods may be more pledgeable other
[28:33] Goods less pledgeable it doesn't quite
[28:35] yeah so it's it's cast it in a general
[28:38] way to have paraly
[28:40] distortions Okay so let's Suppose there
[28:42] are no distortions then a d will be
[28:44] Zero D will be identically one so I'm
[28:47] defining the distortions as the as the
[28:50] wedges so one is the case where there
[28:52] are no distortions
[28:55] yeah okay and just can you explain the
[28:57] beta Prime that's a vector that the prim
[28:59] is just a transpose is this correct
[29:02] or uh it essentially captures each so
[29:07] for each
[29:08] sector I calculate what is the fraction
[29:11] of output sent to the consumer versus
[29:13] what's the fraction of output sent to
[29:14] each individual producer this is the
[29:16] fraction sent to the consumer this is
[29:18] the Matrix encapsulating the fraction
[29:21] sent to other producers so if you go
[29:23] back to your simple example essentially
[29:25] the beta for the first two industries
[29:26] will be zero it's only lost yeah it will
[29:29] be 0
[29:30] 01 in the in the example yeah okay but
[29:35] um the sum of thetas plus beta should
[29:39] add up to one because the total fraction
[29:41] of output is one so some there's some
[29:44] accounting uh identity that need to hold
[29:47] okay so this object Distortion
[29:51] centrality it's a vector it's a scaler
[29:54] per sector and at a sector level it
[29:57] captures the marginal social value of
[29:59] policy subsidies assuming that we can
[30:02] Finance the the subsidy using a lumsum
[30:04] tax that's not distortionary and this
[30:06] marginal social value encapsulates all
[30:08] the gener effects so if this number the
[30:12] number is normalized so that if it's
[30:14] bigger than one subsidizing the sector
[30:16] raises aggregate output if it's below
[30:18] one subsidizing the sector reduces
[30:20] aggregate output so the first result is
[30:24] this object actually average to one
[30:26] across sectors what what does averaging
[30:28] mean here well by average what I mean is
[30:32] um if you average the the the the the
[30:34] centrality using sectoral value added as
[30:37] a fraction of GDP as the weights then
[30:40] that's a notion of weighted average by
[30:42] value
[30:43] added this implies that uniformly
[30:46] promoting all sectors is ineffective or
[30:48] randomly selecting sectors in
[30:51] expectation will have zero aggregate
[30:54] effects now in an economy without any
[30:57] distortion
[30:58] the distor centrality will be
[31:00] identically equal to one across all
[31:02] sectors which mean which means
[31:04] subsidizing any sector on the margin has
[31:06] zero aggregate impact that's also the
[31:08] first rare theorem in the presence of
[31:11] distortions the Distortion centrality
[31:13] would be in general have some dispersion
[31:14] around one it will be different from one
[31:16] in a generic sector and they will be
[31:19] higher than one in some sectors and
[31:20] lower than one in some other sectors and
[31:23] Industrial policy raises aggate output
[31:26] if it promotes sectors or some sectors
[31:29] where the centrality is above
[31:31] one now this is the first there's no
[31:35] uncertainty here the expectations means
[31:37] average here it's not an expectations
[31:38] over
[31:40] probabilities right right so I'm using
[31:42] sectoral value added share as the
[31:44] measure to define the expectation as
[31:48] well as the cence that will come uh
[31:50] right now so this measure essentially
[31:54] one can use that to evaluate government
[31:56] spending because it captures the bend
[31:59] for the buck of government spending so
[32:01] if we multiply government spending in
[32:03] dollar amount per sector with the distor
[32:06] centrality that's the aggate impact of
[32:07] spending in those sectors and if we add
[32:10] up the effect across different sectors
[32:12] we get the aggregate effect and because
[32:15] it averages to one so the overall impact
[32:18] on GDP to first order is simply a cross-
[32:22] sector C variance or a regression
[32:25] coefficient in some normalized way if we
[32:27] regress
[32:28] government spending across sectors on
[32:30] this notion of distorted centrality that
[32:33] regression coefficient tells you what's
[32:34] the argate effect so this this simple
[32:37] coari formula you know to the extent
[32:40] that we can measure this St centrality
[32:42] in different sectors and we can measure
[32:44] government spending VI different
[32:45] industrial policy program we can use
[32:47] that to evaluate what's the aggregate
[32:49] effect of industrial policy I just ask
[32:53] this Distortion centrality how is it
[32:55] connected to the network centrality
[32:57] concept
[32:59] yeah so it's uh right right next point
[33:02] in the sense that sectors with high
[33:04] distor centrality tend to
[33:07] supply disproportionately more to
[33:10] sectors facing loss of distortions
[33:13] directly or indirectly so a central
[33:16] sector in a network is a well-connected
[33:18] sector a distortion Central sector is a
[33:21] sector do apply to many distorted
[33:23] sectors directly or indirectly so I have
[33:26] high distorted centrality if I Supply a
[33:28] lot to you and you have high Distortion
[33:30] centrality and you face Distortion when
[33:33] you want to buy from me so the and
[33:36] trality is like for each sector we're
[33:38] tracing out its value added how does my
[33:40] value added travel through different
[33:42] producers before reaching the fighter
[33:44] consumer each time it travels and and
[33:47] and meets some Distortion it picks up
[33:49] picks that up that's the notion of dis
[33:51] and tral here that's why it tend to be
[33:53] high in Upstream sectors in a vertical
[33:55] economy Upstream always say
[33:57] unambiguously High Distortion inal so in
[34:00] a sense it's the same concept just
[34:02] applied to
[34:03] distortions correct exactly that's why
[34:07] it's called
[34:09] distality
[34:11] okay okay um how do we measure it
[34:15] well the beta and Theta or you know you
[34:19] can directly read off from input out
[34:21] table and we need this whole Matrix of
[34:23] distortions which capture bilateral
[34:25] fuctions some sense
[34:28] now I have met you know in general you
[34:30] know the distortions are very hard to
[34:32] get so a key um critique of industrial
[34:36] policy that's very popular in the 90s as
[34:39] part of the Washington consensus against
[34:42] you know government picking winners is
[34:44] how do bureaucrats know where to target
[34:47] how do they actually even measure
[34:49] distortions but I have shown you the
[34:52] vertical example where even though the
[34:55] magnitude of distortion will determine
[34:57] what's the aggregate effect of
[35:00] intervention the ranking uh of which
[35:03] sector should be picked first doesn't
[35:05] really depend on the size of distortion
[35:07] in a vertical example and we can
[35:09] actually generalize that to a
[35:11] hierarchical network so you know Marcus
[35:14] asked what does a vertical change look
[35:16] like in an inut table the way I Define
[35:19] the input table a vertical would be SE
[35:22] one Supply to two two Supply to three 3
[35:24] to four four to five hence the linkages
[35:26] are all off Di
[35:28] in a hierarchical network is one where
[35:31] one supplies to 2 3 4 two supplies to
[35:33] 345 and so on so it's a it's a lower
[35:36] triangular Network that's very sparse
[35:39] when you get close to the lower uh
[35:42] bottom left corner in that Network
[35:44] Distortion centrality also tends to
[35:47] correlate with upstreamness hence they
[35:49] can be measured with some sense of
[35:51] robustness that is you need very
[35:53] perverse way perverse distribution of
[35:55] distortions for thisor centrality not to
[35:58] align with network in ways that are
[36:00] formalized in the paper so I've shown
[36:02] you this hierarchical network that's hyp
[36:04] markets you have a
[36:06] question just going back to the previous
[36:08] slide in a sense new result is it an
[36:11] approximation around no Distortion as a
[36:14] linear L linear or linear approximation
[36:16] or is it it holds more generally for big
[36:18] distortions as well is there big
[36:20] difference where Distortion is smaller
[36:22] yeah in terms of
[36:24] approximation y so it's an approximation
[36:27] around the observe decentralized
[36:29] equilibrium with potentially large
[36:31] distortions but without government
[36:33] intervention oh so I'm saying suppose
[36:36] you have a world without government
[36:38] intervention but it's very distorted
[36:40] what's the B for the back of a first
[36:42] dollar government intervention that's
[36:44] that's the approximation but it's
[36:46] essentially not a non-distortion
[36:48] benchmark L around it you linearize
[36:50] about the existing economy without
[36:51] government intervention and then you say
[36:53] I change government intervention how
[36:54] does this in a linear way change things
[36:57] correct correct another way to answer is
[37:01] you know if we approximate around an
[37:03] economy without Distortion then any
[37:06] government intervention will have Zero
[37:07] aggate Effect to first order because
[37:10] yeah because you know that's the person
[37:13] there so this is a figure of the
[37:16] hierarchical network but if you remember
[37:18] the South Korea input AB table from the
[37:20] very beginning it looks a lot like the
[37:21] hierarchical network and South targeted
[37:24] all the right sectors things themselves
[37:26] okay this is the hierarchical network
[37:29] illustration using China's input daa
[37:31] table in 2007 it's also fairly
[37:34] hierarchical and in the case of china
[37:37] can you go back to China's why is the
[37:39] horizontal line suddenly what what is
[37:41] that sorry yeah yeah this is a
[37:44] distraction um different countries
[37:47] account for uh Capital formation
[37:49] differently so if you think about
[37:51] Capital formation in a steady state
[37:53] economy you know some sectors have
[37:56] separate accounting tables for for
[37:58] Capital formation some countries treat
[38:01] Capital formation as part of
[38:02] intermediate inputs and so on this
[38:05] sector here the horizontal line is
[38:07] construction so in China during a period
[38:09] where there's a lot of construction and
[38:11] all other sectors use the output of the
[38:14] construction sector it's treated all as
[38:16] intermediate inputs where part of that
[38:18] should go to Capital formation because
[38:21] the only part that should be treated as
[38:22] intermediate input should be the
[38:24] depreciated part of the capital and I
[38:26] think that's just an AR of a bad form of
[38:29] accounting so that artifact doesn't show
[38:32] up in other
[38:33] countries by the way there's one
[38:35] question the audience from Ram aaria he
[38:38] wants to know when you do this input
[38:39] output m is only within the country that
[38:42] Imports exports are not
[38:44] included is this
[38:46] correct um so right now I'm thinking
[38:50] about viewing this as Clos economy we
[38:52] can think about input output linkages
[38:54] but more on uh to follow but for now
[38:56] think that de close to
[39:00] col so in China um where I have access
[39:04] to fir level Manu data I can use IO
[39:07] methods structural IO methods to
[39:09] estimate firm level wedges I treat those
[39:12] as distortions and then I arate to the
[39:14] sector level but recall that you know
[39:17] the notion of distor centrality doesn't
[39:19] really depend on how I measure the
[39:21] Distortion at least the ranking of
[39:23] distortion Cal doesn't really depend on
[39:24] how I measure it if I set Distortion to
[39:26] be con across all sector Pairs and
[39:29] basically get back the same answer in
[39:31] terms of ranking of this or cality
[39:33] across it so this table on the left
[39:36] shows the industries that have the
[39:38] highest disor centralities you know
[39:40] these are all Heavy chemical sectors
[39:41] that's essentially the same set of
[39:43] sectors targeted promoted by South Korea
[39:45] in the 70s on the right these are the
[39:48] sectors with the lowest disor centrality
[39:50] those that should be taxed these are all
[39:52] Downstream um light manufacturing Goods
[39:55] I'm restricting the analysis with the
[39:57] fror goes because that's where I have
[39:59] firm level
[40:00] data so using firm level data um what I
[40:05] found is Industries with higher
[40:08] distortions C trali these firms in those
[40:11] Industries tend to pay lower interest
[40:13] rates they tend to have higher debt uh
[40:15] debt to Capital ratio they tend to pay
[40:17] lower taxes receive more explicit tax
[40:20] breaks these in thees also tend even
[40:22] more State own firms that receive more
[40:24] subsidies so you know all across these
[40:27] different policy platforms of potential
[40:30] ways for the government to intervene one
[40:32] do see that firms in the high centrality
[40:35] sectors seem to be favored okay so in
[40:38] that sense you know these policies the
[40:41] cross- sector variation in taxes
[40:43] interests and subsid to State firms May
[40:46] generate aggregate benefit and one can
[40:49] evaluate those aggregate benefit using
[40:52] this coar formula where we turn that
[40:54] Cross actor variation in interest and
[40:57] tax
[40:57] into some dollar amount of government
[40:59] spending what I found is you know across
[41:02] these policy platforms altogether they
[41:04] accountable for about 6.7% of
[41:07] GDP and if we change sectoral targeting
[41:12] targeting by thigh consumption share or
[41:14] value added or even export intensity or
[41:17] in in inter intermediate expenditure
[41:19] shares these are ter of their policy
[41:22] Target wouldn't do as well as the data
[41:24] that is the government really need to
[41:25] Target Upstream sectors to generate as
[41:28] much gain as as what's observed in a
[41:30] data okay what's L dlw
[41:34] sorry yeah this is the deuer and
[41:36] warzinski method of specification of
[41:38] picking distortions but in the paper I
[41:40] try to pick distortions in many other
[41:43] ways yeah yeah
[41:46] so oh yeah exactly so it's it's their
[41:50] method applied to Chinese
[41:52] numbers but all of these uh quantitative
[41:56] effects of GDP gain coming from
[41:58] industrial policy they all scale with
[42:01] the how much variance is there in distor
[42:03] centrality if you know the variance
[42:06] these numbers are basically almost
[42:08] proportionate to that because the the
[42:10] centrality isn't so sensitive to exactly
[42:12] how things are measure and different
[42:14] specifications just differ in this
[42:17] variance can ask the financial indust is
[42:21] it also part of the input output
[42:24] table and is it although
[42:28] yeah it is um different countries
[42:30] account for it differently so it's a
[42:32] little hard to think
[42:34] about uh exactly where it is in the US
[42:37] it's fairly Downstream so um most of the
[42:40] service sectors are fairly Downstream
[42:42] it's not as Downstream as food
[42:44] consumption but it's so it's in between
[42:46] food consumption and heavy chemicals I
[42:49] would
[42:50] say yeah in China I don't remember what
[42:53] the number
[42:54] is and how much does it how fine the
[42:58] input output table is now if you group
[43:00] more Industries into one bigger industry
[43:03] and you don't have such a fine Gran
[43:07] definition uh yeah does this
[43:10] matter yeah so it it definitely matters
[43:13] in in two senses one um this
[43:17] formula this is derived using num
[43:21] parametric production function so if you
[43:23] have you know s over just gen General uh
[43:29] jointly concave function over different
[43:31] inputs this result holds because it's a
[43:34] first order proximation result so you
[43:36] can you can deal with nonparametric
[43:37] production functions so in that sense
[43:40] you know the level of elastic
[43:42] subtitution across different level
[43:44] aggregation that doesn't quite matter
[43:47] but mismeasurement could still be an
[43:49] issue if you only have very aggregate
[43:51] data I can't go from this level
[43:55] aggregation to finer level because
[43:57] that's not available but in the paper
[43:59] what I do is I try to bundle sectors
[44:02] together to make it more aggregated and
[44:04] results are fairly robust to
[44:07] that
[44:10] okay okay um so that is a paper where I
[44:14] consider a closed economy so in a closed
[44:16] economy you know subsidizing Upstream
[44:18] sectors is welfare
[44:21] enhancing um what what's happened in
[44:24] practice in both China and South Korea I
[44:26] mean from multi-region economies and
[44:28] Industrial policies are often enacted by
[44:30] local governments so there could be a
[44:33] lot of political economy considerations
[44:35] with terms of trade considerations so
[44:38] different local governments are
[44:39] essentially treating their domestic
[44:41] economy like open economies in a
[44:43] presence of cross region trade and input
[44:45] output there may be misalignment between
[44:48] local and Central incentives so that's
[44:50] the topic of study of this followup
[44:53] paper to illustrate the main idea think
[44:56] of you know this illustration with two
[44:58] two regions in the economy the if the
[45:01] two regions are not trading with each
[45:02] other here the three sectors are metal
[45:04] conrete and housing a central planner
[45:07] will want to subsidize the metal sector
[45:09] a local planner also want to subsidize
[45:11] the metal sector if they don't
[45:13] trade now imagine a world where each
[45:16] region to build concrete you need to
[45:18] import the metal from the other
[45:20] region then local planner doesn't have
[45:22] incentive to subsidize metal because for
[45:24] the local planner to subsidize metal
[45:27] it's only going to affect relative
[45:29] prices or the so-called terms of trade
[45:31] in a way that harms domestic economy
[45:33] you're essentially sending more metal to
[45:35] the other economy and buying the same
[45:37] amount of metal from the other economy
[45:40] so from a local production perspective
[45:42] you want to subsidize concrete and
[45:43] actually tax metal
[45:45] production okay so when regions trade
[45:48] that really differs on you know local
[45:51] policy versus what the central planner
[45:53] would want and the real world is
[45:55] somewhere in between so paper
[45:57] essentially we so let me just make sure
[46:00] that if the first region subsidizes
[46:02] metal it is benefiting the other region
[46:05] the other regions
[46:06] concrete correct so the so you want to
[46:11] essentially impose exit tariffs from
[46:14] metal not going to corre region correct
[46:18] so so if these are actually different
[46:20] countries their incentive is to to to
[46:23] impose export tax or actually import
[46:27] tariffs import tariffs and Export tax
[46:30] would generate some kind of learner
[46:32] symmetry they would have the same
[46:33] generate equilibrium effects but here
[46:36] you know when within a given within a
[46:38] given within a given country one cannot
[46:41] impose tariffs but one can use
[46:43] industrial policy to serve as T now you
[46:47] know think about different economies
[46:49] within Europe within a free trade zone I
[46:52] understand industrial policy is also um
[46:54] hard to implement within Europe but but
[46:57] if different countries can give
[46:58] subsidies to favor certain industries
[47:01] they can essentially use that to as a
[47:04] way to manipulate terms of trade and and
[47:06] uh uh and and and uh bypass this no
[47:10] tariff restriction makes some sense but
[47:13] whether you have export tariffs or
[47:14] import tariffs it might be equivalent
[47:16] but it it's not equivalent who gets the
[47:18] revenue from the tariffs I
[47:21] guess right right so it's equivalent if
[47:23] there's some kind of Lum transfer but
[47:25] I'm I'm not thinking how the tariffs are
[47:28] redistribute and in in this study we
[47:30] don't think about tariffs at all we only
[47:31] think
[47:34] policy okay so what the paper does is um
[47:37] we write down a model with cross region
[47:39] trade importable linkage and Ed Mark
[47:42] imperfections as in the previous paper
[47:45] and we derive essentially the local and
[47:47] Central counterpart to the Distortion
[47:50] centrality here we call them local and
[47:52] Central indices for intervention where
[47:54] the central index captures
[47:57] the the local index captures the effect
[48:00] on local welfare per unit subsidy
[48:02] financed by taxing local agents the
[48:05] central index captures the effect of a
[48:08] national welfare some weighted sense a
[48:10] weighted average of local Welfare by
[48:12] taxing nationally in proportion to their
[48:14] income but otherwise it's very similar
[48:16] to the previous paper the information
[48:18] required here would be we need some kind
[48:22] of cross region input out a table so in
[48:24] the context of China we created this
[48:26] with we we we obtain this carvings by
[48:29] industry input have a table so it's a
[48:31] gigantic Matrix 1300 by 1300 Matrix and
[48:35] we estimate firm level wedges from
[48:37] production data but again it's more of
[48:40] the input output structure that's
[48:42] driving this rather than where the
[48:43] distortions are the driving the
[48:45] variation okay what do we do with this
[48:48] so we have this index that differ by
[48:50] region by industry capturing incentive
[48:52] for Central planner and incentive for
[48:54] local
[48:55] planner so first we want to you know
[48:57] just to demonstrate how they differ I'm
[48:59] going to show you regression and show
[49:01] you a concrete example the central
[49:04] planner they want to subsidize Upstream
[49:06] Industries the most of variation is not
[49:08] across region but across Industries it's
[49:11] famous in the previous paper the local
[49:13] planners they want to subsidize sectors
[49:16] that are Upstream but only to local
[49:18] production they don't want to subsidize
[49:20] industries that send inputs to to
[49:22] elsewhere so they want subsize sectors
[49:24] that actually export little so if you
[49:26] progress the share of in share of output
[49:29] sold to other regions as inputs well
[49:32] that's a Upstream sector so Central
[49:35] planner has extended to subsidize it
[49:37] local planner has extended to tax
[49:39] it what's a specific example well we can
[49:42] see it by comparing metals and concrete
[49:46] metal is highly tradable it's an
[49:47] upstream input concrete is also a fairly
[49:50] Upstream input but it's not really
[49:52] tradable so if we compare Beijing and
[49:55] Shanghai in terms of Metal Products
[49:57] Central planner would like to subsidize
[49:59] them local planner in Beijing would like
[50:01] to tax metal production because a lot of
[50:04] the Beijing metal is sold outside of
[50:06] Beijing as an intermediate
[50:08] inputs concrete is Upstream to local
[50:11] production so Central planner doesn't
[50:13] have incentive to subsidize it local
[50:15] planners in both Shanghai and Beijing
[50:17] want to subsidize it so that's where the
[50:19] variation comes from okay we're going to
[50:22] collect um interventions by different
[50:25] levels of government
[50:27] you know we're going to compare do two
[50:29] kinds of empirical exercises first kind
[50:31] we're going to compare central
[50:33] government intervention and local
[50:34] government
[50:36] intervention government intervention
[50:37] take many forms State no firm subsidies
[50:40] they received fiveyear plans you know
[50:43] different levels of government Central
[50:45] or provincial they they they publish
[50:47] their own fiveyear plans which lay out a
[50:49] set of strategic industries that um the
[50:52] economy should promote over the next
[50:54] five years they also set up special econ
[50:56] ecomic zones with designated industrial
[50:59] focus and we're going to compare Central
[51:01] and local and and provincial we're going
[51:03] to show that Central intervention tend
[51:05] to align with the central index and the
[51:07] local strategic Industries tend to align
[51:09] with the local index number two we're
[51:12] going to look across provinces in the
[51:14] local targeting what we're going to find
[51:17] is in provinces that are more fiscally
[51:20] independent from the central government
[51:23] they tend to Target strategic Industries
[51:25] that's better for the eles whereas
[51:27] provinces that are less developed and
[51:29] rely more on the central government for
[51:31] fiscal transfers they tend to Target
[51:34] sectors that are more aligned to the
[51:35] central government so this is
[51:37] essentially a formalization of what I
[51:39] mentioned you know if we compare the the
[51:42] officials in China they knew your paper
[51:44] already before you wrote it is this fair
[51:47] to
[51:47] say or they understood the mechanism yes
[51:51] they understood in some sense yes if you
[51:53] if you read if you read policy um
[51:55] documents
[51:57] uh is my internet okay from from my
[51:59] perspective Marcus I can hear your voice
[52:01] but your Zoom is frozen can you see me
[52:03] moving or yes it works fine on my side
[52:05] okay good okay so if you read policy
[52:09] documents you do see mentions of network
[52:11] linkages subsidizing in a sector
[52:14] generate you know these are the pillar
[52:16] Industries because it's going to
[52:17] generate a lot of spillovers and the
[52:19] language describing specific Industries
[52:22] differ between the Central and local if
[52:24] you look at the fiveyear plans why they
[52:26] designate specific sectors language seem
[52:29] to suggest some of this terms of trade
[52:32] or externality to other regions are part
[52:34] of the
[52:35] consideration so State own firms some
[52:37] are centrally owned some are affili with
[52:40] provincial governments provincial State
[52:42] own firms tend to align with local index
[52:44] Central tend to align with the central
[52:45] index and if we regress the local state
[52:48] shares on the local index Province by
[52:51] province and look at the pattern across
[52:54] provinces in that correlation we see
[52:56] that correlation is stronger in areas
[52:58] that receive less fiscal transfer from
[53:00] the central government in terms of the
[53:03] fiveyear plans you know the central
[53:04] 5year plans Target Central index local
[53:09] the provincial fiveyear plans don't
[53:11] really targ I mean is correlated but
[53:13] insignificant with both Central and
[53:15] local but if you look across different
[53:18] provinces in the East Co in the coastal
[53:20] region that are more financially more
[53:22] developed their fiveyear plans tend to
[53:25] Target the local index so then more West
[53:27] and Northeast and more less economic
[53:30] development region their fiveyear PL
[53:32] seem to more support the central index
[53:34] these are also regions that receive more
[53:36] fiscal transfers so on the map you know
[53:39] if we look at which which local which
[53:42] provinces have local policies that spit
[53:44] generate positive spill over to others
[53:47] these are all the less developed region
[53:49] and the more developed region actually
[53:51] create negative spillovers on the rest
[53:53] of China and if we look at which sector
[53:56] which regions receive positive
[53:58] spillovers these are the regions that
[53:59] actually benefit from local industrial
[54:02] policy in in other areas okay so those
[54:06] were two studies based on you know
[54:08] developing economies thinking about how
[54:11] to use industrial policy to correctable
[54:13] Market
[54:14] imperfections um I'm going to switch
[54:16] gears and maybe speed up a little bit
[54:19] how much time do I have take your
[54:22] time okay try to do it in like
[54:26] maybe 15 18 minutes yes
[54:30] that's great okay so I'm going to switch
[54:33] gears to think about industrial policy
[54:35] for you know technology for growth and
[54:38] maybe for the green transition which are
[54:40] more relevant for developed economies so
[54:43] this is recent work uh with Som we think
[54:46] about Innovation uh Network and R&D
[54:49] allocation so this is a paper where we
[54:51] want to understand how should economies
[54:53] allocate R&D resources across technology
[54:57] to fix ideas is this is an illustration
[54:59] drawn using the patent citation Network
[55:02] you should note is a on-dit patent
[55:04] sector so this is semiconductors
[55:06] Pharmaceuticals basic chemicals there
[55:08] are these links across sectors capturing
[55:11] citation relationships which are proxies
[55:13] for knowledge spillovers we want to
[55:16] understand you know when there are
[55:17] cross- seor spillovers how much
[55:19] resources should the society allocate to
[55:21] semiconductor versus Pharmaceuticals so
[55:24] just make sure this is us data now
[55:27] oh U we are going to look both with us
[55:31] and cross country um this figure I
[55:33] believe is drawn using us patterns and
[55:36] the spillovers is not citations of
[55:39] patterns by some other field it's the
[55:41] spillovers of the
[55:43] industries the links between the
[55:46] different so uh here I'm going to Define
[55:51] um a sector as a technology I'm not
[55:53] going to think about SIC code I'm only
[55:55] going to think about patent
[55:57] classifications and these are citation
[55:59] shares across patent classifications oh
[56:01] citation shares
[56:04] okay so basically we're going to look at
[56:07] the network and think about how
[56:08] countries should allocate and compare
[56:11] with how countries are actually
[56:12] allocating and see whether some
[56:14] countries are doing better than others
[56:16] and maybe just to think about whether
[56:18] you know ch's act or inflation reduction
[56:21] act have some rationale based on this
[56:24] analysis to do this erors the
[56:27] connections is not directional are they
[56:29] directional or they they are they are
[56:32] directed the direction is just drawn
[56:34] there is in a very small way yeah but
[56:36] but there are directions in the
[56:38] arrows so to to to actually talk about
[56:41] this I need to formally introduce the
[56:43] model otherwise the analysis wouldn't
[56:45] make much sense but I'll set up the
[56:47] model in one slide and go through result
[56:49] in one
[56:51] slide so it's an economy that consumers
[56:54] are forward looking at each point in
[56:56] time they consume a bundle of different
[56:58] Goods beta ey captures how important is
[57:01] technology I in a consumption bundle and
[57:04] the consumption goods are produced
[57:06] linearly from labor based on technology
[57:08] so Q is the technology in in sector I is
[57:12] the level of knowledge in in sector I so
[57:15] this knowledge can be improved through
[57:17] R&D and this is the Innovation
[57:20] production function so at any given
[57:22] point in time Q is the stock or the
[57:25] state variable
[57:27] um the flow of new innovation is the
[57:29] product between how much resource we
[57:31] spend on innovation in a sector and The
[57:33] Innovation efficiency in that sector The
[57:36] Innovation efficiency in the sector
[57:37] depends on some exogenous component but
[57:40] also endogenously on the level of
[57:41] knowledge in all other sectors of the
[57:44] economy Omega J captures how important
[57:47] is technology J for sector I knowledge
[57:51] and the collection of this Omega is what
[57:53] defines an innovation Network
[57:56] so this is a law of motion that governs
[57:59] how flow converts into stock so in
[58:01] simpler terms in the absence of cross-
[58:04] sector spillovers this law of motion
[58:07] collaps down to something very simple
[58:08] that is the growth rate of knowledge in
[58:10] a sector is log linear in the amount of
[58:12] resources we spent in that sector but in
[58:16] general with cross- sector spillovers
[58:18] growth rate of knowledge in the sector
[58:19] not only depends on R&D in that sector
[58:22] but also on the relative stage of
[58:23] knowledge between all other economy all
[58:25] other sect
[58:26] sectors and the key question is given a
[58:30] stock of production and R&D resources
[58:32] where production resources we allocate
[58:34] to produce R&D resources we allocate to
[58:37] do R&D how should the society optimally
[58:40] allocate our cross sectors that's the
[58:42] question that's the abstract question
[58:43] we're trying to
[58:45] answer ask you the the linkages
[58:47] themselves are given exort justly
[58:49] there's no investment that I can promote
[58:52] spillovers from one sector to another
[58:54] sector is this correct
[58:57] yeah in the Baseline model in everything
[58:58] I'm going to show that's the case so
[59:00] this these omegas These elasticities are
[59:03] exogenous in the paper we characterize a
[59:06] case where these elasticities can be
[59:09] indulgently changed which essentially is
[59:12] just by replacing this function with
[59:14] some general aggregator over knowledge
[59:16] so that if you define the network based
[59:18] on local elasticities that then a
[59:21] general aggregator would capture that so
[59:23] the local elasticity would change then
[59:25] the result I'm to show you in particular
[59:27] what I call a sufficient statistic that
[59:29] guides the empirical analysis it will be
[59:32] valid as a first approximation around a
[59:34] balanced growth path where you're
[59:36] defined using the local elasticity so in
[59:38] that sense the the logic I'm going to
[59:41] use in the analysis is General but what
[59:44] the numbers I'm going to show you treat
[59:46] it as first order approximation if you
[59:48] deviate from this long linear
[59:51] environment so in general you know how
[59:53] do we allocate resources across sectors
[59:55] it's a in abstract is a very hard
[59:57] problem because anywhere you locate R&D
[01:00:00] is going to have influence the growth of
[01:00:02] knowledge in that sector that has going
[01:00:04] to have long run consequences on many
[01:00:06] other sectors of the economy then
[01:00:08] abstract is a very hard problem now we
[01:00:10] picked the log linear formulation so
[01:00:12] that this problem is a very intuitive
[01:00:14] solution but as I in response to
[01:00:16] Marcus's question the solution I
[01:00:19] characterized has some generality in
[01:00:22] it but here in this environment
[01:00:26] spillovers are immediate no there's no
[01:00:27] delay in
[01:00:29] spillovers the spillovers are not
[01:00:32] immediate that
[01:00:34] is the efficiency of
[01:00:38] research uh in sector I depends on the
[01:00:41] knowledge of sector J which is a stock
[01:00:43] so to spill over you know number of
[01:00:46] mathematicians today doesn't benefit my
[01:00:48] research efficiency it's the math
[01:00:50] theorems how many math results are out
[01:00:52] there that helps my writing this paper
[01:00:57] so for any initial state of knowledge
[01:00:59] the optimal allocation of resources in
[01:01:01] this lock linear environment is timeing
[01:01:03] variant the worker the labor allocation
[01:01:05] should follow a beta which captures how
[01:01:08] important is technology in a consumption
[01:01:10] bundle so if we as consumers value
[01:01:13] textiles more workers should go to
[01:01:16] textiles the allocation of Rd resources
[01:01:18] is different it should follow a vector
[01:01:21] gamma which should sum to one and in
[01:01:23] proportional terms it's equal to the
[01:01:25] product between
[01:01:26] Vector capturing how consumers value
[01:01:28] different technology in the consumption
[01:01:29] bundle and the Le of inverse of a
[01:01:32] discounted version of The Innovation
[01:01:34] Network what's the intuition we can
[01:01:37] expand this leant of inverse in in in
[01:01:40] into a power series of
[01:01:42] Matrix so the social planner is thinking
[01:01:44] about you know if we allocate more
[01:01:46] resources of R&D in a sector it's going
[01:01:49] to raise that technology in the sector
[01:01:51] then benefit a consumer according to
[01:01:53] Beta so beta times the identity Matrix
[01:01:55] that's direct effect but now that sector
[01:01:58] has higher technology in the future it's
[01:02:00] going to benefit future researching
[01:02:02] other sectors and then other sectors
[01:02:04] benefiting other sectors those are all
[01:02:06] the network effects those effects occur
[01:02:08] in the future so they're discounted and
[01:02:10] the relevant discount rate is how we
[01:02:13] discount future consumption versus how
[01:02:15] quickly those future benefits
[01:02:18] materialize so you can see that if the
[01:02:21] society is very impatient if the
[01:02:23] discount rate goes to Infinity all of
[01:02:25] these Network effect go to zero so for a
[01:02:28] myopic planner the optimal R&D
[01:02:30] allocation coincide with the consumer
[01:02:32] preferences we should all do research in
[01:02:34] what we produce we should all do
[01:02:36] research in improving
[01:02:37] textiles If instead the society is
[01:02:40] infinitely patient the network terms
[01:02:42] become more and more important and the
[01:02:44] original beta term becomes less and less
[01:02:46] important in the limit the optimal
[01:02:49] allocation converges to the igen vector
[01:02:51] centrality of the network meaning a
[01:02:53] sector is more Central if generates l
[01:02:56] still risk to other Central sectors
[01:02:58] turns out that's also the growth
[01:03:00] maximizing allocation so in the you know
[01:03:03] if we're infinitely patient we should do
[01:03:04] research with the only goal in mind
[01:03:07] that's the long run grow for any finer
[01:03:10] level of patients in between these two
[01:03:12] extremes the optimal allocation is a
[01:03:15] weighted average in a vector sense in
[01:03:17] Matrix equation sense between consumer
[01:03:20] preferences and the vector that maximize
[01:03:23] growth but in general it's always in
[01:03:24] between
[01:03:26] okay next is the result that enables us
[01:03:29] to take this result take this model to
[01:03:31] the data that is suppose we can measure
[01:03:33] the optimal allocation for an economy
[01:03:36] and we observe real world R&D allocation
[01:03:38] in the data we can calculate what's the
[01:03:42] wre gain if we reallocate R&D
[01:03:45] optimally so this terming red is also
[01:03:48] known as relative entropy of one
[01:03:50] distribution from another it measures
[01:03:52] how far is the real world allocation
[01:03:54] away from the so called optimal one it's
[01:03:56] a notion of misallocation across sectors
[01:04:00] and with a handful of parameters it
[01:04:02] converts into a notion of welfare loss
[01:04:04] due to r&
[01:04:06] allocation so to operationalizing the
[01:04:09] data we're going to try to measure
[01:04:10] optimal allocation and actual allocation
[01:04:13] one elephant in the room that's a
[01:04:15] feature in the data is this is a closed
[01:04:18] economy in the data there are lots of
[01:04:20] spillovers that come from a bra so I'm
[01:04:21] going to incorporate that feature now
[01:04:24] essentially the require modification is
[01:04:27] to do Rd the Rd efficiency in sector I
[01:04:30] depends not only on domestic knowledge
[01:04:33] but also on foreign knowledge so the
[01:04:35] elasticity is no longer just Omega J but
[01:04:38] it has foreign domestic components and
[01:04:40] foreign components but I'm still going
[01:04:43] to continue to think about a domestic
[01:04:45] planner trying to all to ma maximize
[01:04:47] domestic welfare taking all the future
[01:04:50] forign knowledge as as as
[01:04:52] given turns out the intuition is very
[01:04:55] Sim IL to in the closed economy case
[01:04:58] it's just that the relevant elasticity
[01:05:00] for domestic spillovers is no longer
[01:05:02] Omega but only the domestic components
[01:05:05] of that so it's the product between
[01:05:07] Omega IG and
[01:05:08] xig the implication of that is for an
[01:05:11] economy that's reliant on on own
[01:05:14] domestic knowledge as I was showing the
[01:05:16] data it's the US and Japan they should
[01:05:18] do R&D as if they're very patient
[01:05:21] whereas for an economy that benefits a
[01:05:23] lot of from forming spillovers it's as
[01:05:26] domestic knowledge doesn't matter as
[01:05:27] much that's isomorphic to a world where
[01:05:31] they have more impatient planners so
[01:05:33] they should allocate R&D more towards
[01:05:35] sectors that they directly
[01:05:38] produce and we also generalized this
[01:05:40] welfare calculation taking into account
[01:05:43] the fact that you know if the US mess up
[01:05:45] its R&D allocation because it's future
[01:05:48] R&D built on its fire R&D R&D mation has
[01:05:52] long consequences Canada free right on
[01:05:55] the US so domestic arting misallocation
[01:05:58] is not that Ware consequential so we
[01:06:01] need to make this open economy
[01:06:03] adjustment to take account of that and
[01:06:05] that's all I want to show in theory now
[01:06:07] I'm going to operationalize this by
[01:06:09] trying to measure optimal allocation of
[01:06:11] data and compare with the actual so
[01:06:14] we're going to the the main object is
[01:06:17] this Innovation Network we're going to
[01:06:19] measure using patent citation so among
[01:06:22] all the citations made by patents in a
[01:06:25] technology High what fraction actually
[01:06:27] goes to patents in a technology J we're
[01:06:30] going to rely mostly on this
[01:06:31] International patent data from Google
[01:06:33] patents that combine patent data from
[01:06:36] more than 20 major patent offices around
[01:06:38] the world so no matter where the
[01:06:40] inventor is if you file a patter with us
[01:06:42] Japan China European PDO one of 20 major
[01:06:46] offices you are in the data center and
[01:06:48] we augment that with production data and
[01:06:51] R&D informational firm level and
[01:06:53] aggregate to the technology level let me
[01:06:56] first show you some summary stats of
[01:06:57] this Innovation Network just to
[01:06:59] visualize it and then look at the
[01:07:01] optimal
[01:07:02] allocation on the right these are the
[01:07:04] top 10 sectors with the highest
[01:07:06] centrality these are you know you can
[01:07:08] think of the centrality of a sector
[01:07:10] capturing the contribution of R&D in
[01:07:12] that sector on Long Run growth so these
[01:07:16] are you know most Central sectors are
[01:07:17] medical science Computing basic electric
[01:07:20] element which is semiconductors electric
[01:07:23] communication techniques and so on and
[01:07:25] figure on the left plots the centrality
[01:07:28] sorted by sectors so the most Central
[01:07:30] sectors are much much more Central and
[01:07:33] ourd in those sectors are much more
[01:07:35] important for growth than the peripheral
[01:07:39] sectors visualization of the netw for
[01:07:42] for sometimes in certain Fields you site
[01:07:45] very easily in other fields you're very
[01:07:47] you know careful not to site too many uh
[01:07:50] yeah even within economics you correct
[01:07:53] for this and
[01:07:56] yeah
[01:07:57] so um we we Tred to do various definite
[01:08:01] various ways to measure this um
[01:08:04] definition so we can weigh the citations
[01:08:06] by but remember this is defined at the
[01:08:10] technology to technology level we're not
[01:08:12] really using patent level on at once we
[01:08:16] aggregate but we can we underlying
[01:08:18] patent buy the total citations or to buy
[01:08:21] the stock market value from Cog at all
[01:08:24] we can also use different scaling that
[01:08:26] is you know these things don't have to
[01:08:28] sum up to one but you know the whole
[01:08:30] Matrix we have to have some spectral
[01:08:32] radius bounded by one we do various
[01:08:35] things with different ways to measure it
[01:08:37] I
[01:08:39] if so if certain if different sectors
[01:08:41] have different propensity to file for
[01:08:44] patents but if the influential patents
[01:08:46] receive all the citations once the field
[01:08:49] generate spillovers our measurement will
[01:08:51] be fine if certain Fields just don't
[01:08:54] receive citation
[01:08:56] then that's hard to characterize because
[01:08:57] that's happens at the field level and
[01:08:59] this is constructed at the field level
[01:09:01] so we don't have good ways to character
[01:09:02] that character
[01:09:08] that okay this is a visualization of the
[01:09:11] citation Network you know the most
[01:09:12] Central sectors are heavily cited by all
[01:09:14] others the peripheral sectors are not
[01:09:17] very not cited very heavily by others so
[01:09:19] network is very
[01:09:21] asymmetric at the country sector to
[01:09:23] Country sector level us and Japan have
[01:09:25] the most number of patents and the
[01:09:27] distinction is US creates a lot of
[01:09:29] spillovers to other economies Japan is
[01:09:31] relatively isolated doesn't site other
[01:09:34] as much doesn't get cited as much and
[01:09:37] finally the the important piece of
[01:09:38] information is when the countryes site
[01:09:42] how frequently do they site their own
[01:09:44] their own pattern so us and Japan 70% of
[01:09:47] citations are towards domestic patents
[01:09:50] all other countries side domestic
[01:09:52] patents less than 50% of time and this
[01:09:54] is 2010
[01:09:55] so you know to first order you can think
[01:09:57] of us and Japan as close to being closed
[01:10:00] economies all the economies are open
[01:10:03] econom so these are the ingredients
[01:10:06] towards the measurement of the optimal
[01:10:08] allocation which I'm showing you now
[01:10:10] this is the optimal allocation of the US
[01:10:12] R&D across three digit patent sectors
[01:10:16] the Medical Science us should allocate
[01:10:18] 6.5 to 7% semiconductor 6% Follow by
[01:10:23] Computing automobiles
[01:10:25] I'm going to overlay on top of that
[01:10:27] optimal allocation by four other major
[01:10:29] economies Japan China South Korea
[01:10:31] Germany compared to the us all these
[01:10:34] economies should allocate less to
[01:10:35] Medical
[01:10:37] Science Germany and Japan should
[01:10:39] allocate more to Automobiles and South
[01:10:41] Korea should allocate more to uh
[01:10:44] consumer
[01:10:45] electronics so the variation optimal
[01:10:48] comes from mainly not the Innovation
[01:10:50] Network because that's very similar
[01:10:51] across countries it comes from
[01:10:53] differences in production structure and
[01:10:55] different in the degree of B spill overs
[01:10:57] thatc because germanyy is producing a
[01:10:59] lot of cars it should also correct have
[01:11:02] more
[01:11:03] allocation that that's exactly
[01:11:07] right this figure shows you how actual
[01:11:10] allocation compare with the optimal
[01:11:12] allocation actual on the Y optimal on
[01:11:15] the x axis and with with line of best
[01:11:18] fit you see across all of these economy
[01:11:21] sectors that should have more resources
[01:11:23] do end up getting more R&D resources
[01:11:25] this is not obvious you know it's not
[01:11:28] mechanical in the data because the Y and
[01:11:30] xaxis come from completely data sources
[01:11:33] XX come from patent citations YX come
[01:11:36] from aggregation of firm level R&D these
[01:11:38] are different data
[01:11:40] sources and in particular some economies
[01:11:43] the optimal allocation the line of fit
[01:11:45] is indistinguishable from 45 degree line
[01:11:48] although a line of fit being 45 degree
[01:11:51] doesn't imply the country is allocating
[01:11:53] resources optimally there's still a lot
[01:11:55] of dispersion around the 45 degree if
[01:11:57] you move allocation vertically towards
[01:12:00] the 45 degree to close the gap that
[01:12:02] improves welfare and the relative
[01:12:04] entropy measure of how far is the y axis
[01:12:08] away from the x-axis tells you what's
[01:12:10] the degree of inefficiency which is what
[01:12:13] the theory provides in how to interpret
[01:12:16] these dispersions and is what we're
[01:12:18] going to show you
[01:12:19] next this figure essentially plots the
[01:12:22] relative entropy which is a notion of
[01:12:24] R&D missile location that's comparable
[01:12:26] across countries Japan South Korea
[01:12:29] Germany and the US have the most um
[01:12:32] efficiently allocated R&D across
[01:12:34] Technologies Russia had very bad
[01:12:36] allocation uh the last column is
[01:12:39] European Union by aggregating the EU
[01:12:41] together treating it as a single economy
[01:12:43] but it seems to me that smaller
[01:12:45] countries like the Netherlands or
[01:12:46] Austria they're much less efficient is
[01:12:49] there a size component to it too or not
[01:12:53] no uh it's not up
[01:12:57] so this is directly coming from the data
[01:13:01] um country size doesn't necessarily
[01:13:04] correlate with um degree of inefficiency
[01:13:07] the European Union for example would be
[01:13:10] larger than each individual Country
[01:13:12] combin and this is not lower so in in
[01:13:16] this talk I don't have time to talk
[01:13:17] about what could be the predictor um but
[01:13:20] I I don't have slid prepared to show
[01:13:22] that but what we find is essentially
[01:13:25] what's the source of inefficiency in R&D
[01:13:28] if each firm only conducts one
[01:13:30] Innovation it's going to ignore all the
[01:13:32] spillovers but if the firm conducts one
[01:13:35] Innovation and its own itself future in
[01:13:38] the future Builds on it then the firm
[01:13:40] will internalize Innovation so in that
[01:13:42] sense you know Samsung son seens Google
[01:13:46] Apple all these firms that do
[01:13:47] fundamental research in ways that
[01:13:49] benefit its own future research that
[01:13:51] helps correct for some of the
[01:13:53] externalities so what we find in data is
[01:13:56] in countries where Innovation activities
[01:13:57] are more concentrated a few firms that
[01:13:59] hand for most patents allocation tend to
[01:14:02] be more efficient that's one very strong
[01:14:04] predictor but anyway coming back to the
[01:14:07] slides we
[01:14:09] can compute what's the welfare loss due
[01:14:12] to the misallocation so you know in
[01:14:15] economies that are more self-sufficient
[01:14:17] and rely Less on foring the gray bars
[01:14:20] are comparatively taller compared to the
[01:14:22] red in economies where they benefit from
[01:14:25] forign the gray bars are lower so
[01:14:27] domestic misallocation matter less for
[01:14:30] welfare given the same degree of
[01:14:32] misallocation in table form in the US
[01:14:35] 20110 misallocation account for
[01:14:38] correcting for that can account for 8%
[01:14:40] consumption equivalent welfare gains
[01:14:43] Germany 4% South uh Japan 5 5.6% higher
[01:14:47] in
[01:14:49] Russia now the the the the model also
[01:14:52] has predictive power about which sectors
[01:14:54] are over under allocated in particular
[01:14:57] we can zoom in on key industries of
[01:14:59] interest for example semiconductors and
[01:15:02] compare how the US does compared to East
[01:15:05] Asian economies so in semiconductors us
[01:15:08] under allocates by 21% Japan does just
[01:15:11] about right China South Korea over
[01:15:13] allocates more aggressively provides you
[01:15:16] know potential support for the relevance
[01:15:18] of chips for mea Act and the recent
[01:15:21] policies to support semiconduct these
[01:15:23] numbers are before the chip Act
[01:15:26] or before these are all before the yeah
[01:15:30] okay yeah so um so first you know it it
[01:15:34] takes a while for chip act to to
[01:15:36] actually generate any effect and it
[01:15:38] takes a while for firms R&D data to
[01:15:40] actually materialize and we we don't
[01:15:42] have that that that
[01:15:44] Precision in terms of green Innovation
[01:15:47] us also under allocates relative to East
[01:15:49] Asian economies notice that this is not
[01:15:51] about we're not taking into account any
[01:15:54] environmental nowadays this is purely
[01:15:57] based on how important is Green
[01:15:58] Technology in the consumption bundle and
[01:16:01] also how much citations Green Technology
[01:16:04] receive from other sectors that's
[01:16:06] important in the consumption bundle us
[01:16:08] underfunds about
[01:16:10] 25% okay let me briefly talk about you
[01:16:14] one slide to talk about this um last
[01:16:17] last topic and then I'll conclude so you
[01:16:20] know in in in this other paper we think
[01:16:23] about how to use industrial policy to
[01:16:26] correct for coordination failure in a
[01:16:28] dynamic context in particular we want to
[01:16:31] think about it in in terms of green
[01:16:32] transition Queen trans Green Technology
[01:16:34] along the supply chain now this green
[01:16:36] transition really requires um just the
[01:16:39] switching along entire supply chain
[01:16:41] because if we think about transitioning
[01:16:43] from gasoline cars to electri Vehicles
[01:16:46] the ladder requires batteries the
[01:16:47] production itself is emission
[01:16:50] intensive so while there's broad
[01:16:53] consensus worldwide on the need to speed
[01:16:55] up the transition countries tend to
[01:16:57] diverge on how to achieve a goal in
[01:16:59] Europe most of the policy tend to focus
[01:17:01] on carbon tax or Captain trade in the US
[01:17:04] US seems to be more open to targeted
[01:17:06] industrial policy or to facilitate the
[01:17:09] green transition so in this paper we in
[01:17:12] in this paper briefly discussed we build
[01:17:14] a dynamic model of transition which
[01:17:17] features strategic complimentarity of
[01:17:19] the following form think about an EV
[01:17:21] producer if we take into account the
[01:17:24] Environmental cost right we're thinking
[01:17:26] about you know maybe green production is
[01:17:29] cheaper once environmental costs are
[01:17:31] taken into account for the EV producer
[01:17:34] if producing battery is still very dirty
[01:17:37] and generates a lot of prod uh pollution
[01:17:41] then it's actually environmentally
[01:17:42] expensive to produce an EV likewise if
[01:17:46] there are no demand for from EV
[01:17:48] producers there are no EV producers
[01:17:50] there's no incentive to actually
[01:17:52] greenify the production of batteries
[01:17:54] because there's no market for it so
[01:17:56] there's strategic complimentarity along
[01:17:58] the supply chain but it's very different
[01:18:01] from Big push where all the
[01:18:02] externalities come from demand here the
[01:18:04] externality come from both the cost side
[01:18:06] and the demand side the demand
[01:18:08] externality travel from Downstream to
[01:18:10] Upstream that is if you you know more EV
[01:18:14] producers create demand for battery
[01:18:16] producers to greenify more battery
[01:18:18] producers that become green reduces the
[01:18:21] total social cost of producing
[01:18:23] electrical Vehicles generating
[01:18:24] incentives so the directional nature of
[01:18:27] these two different forms of
[01:18:29] externalities Al on the supply chain
[01:18:30] generates several new insights
[01:18:32] highlighted by this paper that is
[01:18:35] implementing a social optimal requires
[01:18:37] both a carbon tax and targeted subsidies
[01:18:40] so uniform or just Captain trade or
[01:18:42] carbon tax doesn't work second as
[01:18:45] opposed to Big push Theory which says
[01:18:48] you know the government need to use
[01:18:49] industrial policy to imple to intervene
[01:18:52] in many sectors simultaneously to create
[01:18:54] a big push maybe for a sustained period
[01:18:57] of time in order to have any long effect
[01:18:59] that's no longer true in this sitting
[01:19:01] small and temporary subsidies what we
[01:19:03] call small nudges to key sectors can
[01:19:06] generate longer impact and what are
[01:19:08] those key sectors well if the subsidies
[01:19:11] are limited we show they should
[01:19:13] primarily Target Downstream sectors
[01:19:15] because at the downstream Market size
[01:19:18] created from Downstream sectors travel
[01:19:20] to Upstream inputs one for one 1% more
[01:19:24] EV producers create 1% more demand for
[01:19:28] for for batteries on the other hand the
[01:19:31] the the other side of incentive travel
[01:19:33] less than one for one because EV uses
[01:19:36] not only batteries but also labor and
[01:19:38] other inputs so the cost reduction from
[01:19:41] batteries doesn't translate one for one
[01:19:43] into cost reduction in in in EVS we also
[01:19:47] created several examples to show misir
[01:19:50] just to make sure I'm fully understand
[01:19:51] because earlier in the first part of the
[01:19:52] talk you said oh we should actually
[01:19:53] intervene at top of the supply chain now
[01:19:56] we say we should intervene at the bottom
[01:19:58] what's
[01:20:01] fundamentally so um which I I'll I I'll
[01:20:04] talk about that distinction again when I
[01:20:06] conclude that you know policy presion
[01:20:08] really depends crucially on the
[01:20:10] underlying mechanism in the first paper
[01:20:13] it's the accumulation of distortionary
[01:20:15] effects due to Market imperfections in a
[01:20:17] network in that world policy has the
[01:20:21] highest band for the buck to subsidize
[01:20:23] to expand the most distorted sectors
[01:20:25] which is
[01:20:26] Upstream here the market failure comes
[01:20:29] from coordination it's not really so
[01:20:32] coordination and it really depends on
[01:20:34] you know if you give a sector a
[01:20:37] subsidy how what's the impact of
[01:20:39] additional decentralized incentives can
[01:20:42] be created from the outcome of that
[01:20:44] subsidy so it really depends on how much
[01:20:47] does demand externality travel through
[01:20:49] the network versus how much does the
[01:20:51] cost reduction travel through the
[01:20:53] network these are very different
[01:20:55] economic forces hence they call for
[01:20:57] different
[01:20:58] intervention and we show that
[01:20:59] misdirected or delayed policy
[01:21:02] intervention can permanently derail the
[01:21:04] green transition so that's a recent
[01:21:06] working paper um with the question you
[01:21:09] know the carbon tax versus industrial
[01:21:11] policy or supply side policy depends of
[01:21:14] course a lot on their fiscal space as
[01:21:16] well many European countries probably
[01:21:19] cannot do this huge subsidy because they
[01:21:22] get into fiscal difficulties you don't
[01:21:24] consider these
[01:21:26] aspects yeah that that's an aspect that
[01:21:28] um that so far um I have not considering
[01:21:32] any of my work and there are also you
[01:21:34] know just many other practical
[01:21:36] difficulties with policy intervention as
[01:21:38] you mentioned at the beginning you know
[01:21:40] one uh practical inter problem is
[01:21:44] there's a lot of bureaucracy or
[01:21:46] corruption or or law being you know how
[01:21:49] do we ensure that given even a b a
[01:21:53] benevolent planet
[01:21:55] a sound um economic plan with a policy
[01:21:58] Target how do we ensure the policy funds
[01:22:00] go to the right actors all of these
[01:22:03] practical difficulties with Implement
[01:22:05] with intervention both in the actual
[01:22:08] giving out the subsidy versus raising
[01:22:09] the fisal revenue versus political
[01:22:12] feasibility of P policy my work has so
[01:22:16] far not touched up on those forces
[01:22:18] instead I treat those implementation
[01:22:21] issues aside and only think about if we
[01:22:23] can collect L lumsum tax and conduct
[01:22:26] intervention to different sectors what
[01:22:29] are the policy prescriptions coming from
[01:22:31] first principles interacted with the
[01:22:34] network which this discussion is a great
[01:22:37] um segue into my conclusion in that how
[01:22:41] to conduct industrial policy depends
[01:22:42] crucially on the underlying mechanism we
[01:22:45] have seen you know accumulation Market
[01:22:47] imperfections Target Upstream with
[01:22:50] cross- sector externalities we have to
[01:22:52] be careful about the incentives of the
[01:22:54] local planner and different political
[01:22:56] economy considerations with knowledge
[01:22:58] spillovers we want to Target Central
[01:23:01] sectors but only to a certain extent
[01:23:03] governed by how forward looking the
[01:23:05] society is and how much do domestic
[01:23:07] spillovers actually matter with
[01:23:09] coordination failures at least in the
[01:23:11] context of Technology adoption we want
[01:23:13] to subsidize Downstream to generate the
[01:23:15] most incentives along the supply chain
[01:23:19] so one observation that um I want to
[01:23:22] make here is none of these mechanis Ms
[01:23:24] are new to economics all of these are
[01:23:27] very old mechanisms have been studied
[01:23:29] just not in a network context until
[01:23:31] recently so you know the fact that these
[01:23:34] mechanisms that these old well study
[01:23:37] mechanisms interact with networks and
[01:23:40] have so many new predictions in ways
[01:23:42] that can connect your data and that's
[01:23:44] just super exciting so you know
[01:23:46] different inter different mechanisms
[01:23:48] interact with network and that calls for
[01:23:50] new models and new Theory and new ways
[01:23:52] to measure the mean in the data and
[01:23:54] really there's no a single umbrella one
[01:23:57] size F all prescription on which sector
[01:23:59] to to Target but that's also the beauty
[01:24:01] of why the subject is so interesting so
[01:24:04] that's that's where I want to
[01:24:06] conclude great thanks a lot uh we ran
[01:24:10] over big time this time but uh that
[01:24:12] wasn't very exciting I apologize no no
[01:24:15] um perhaps a final word what's about in
[01:24:19] Indulgence Network foration that you try
[01:24:21] to influence the network itself
[01:24:24] um that probably some policies along
[01:24:27] that line how would you categorize
[01:24:29] them this these type of
[01:24:32] policies great great question that's
[01:24:36] something of interest to many people
[01:24:38] including myself I haven't done any work
[01:24:40] on that so I I don't know
[01:24:43] okay future work so so yeah so so so in
[01:24:47] the sense you know I I gave a shortcut
[01:24:49] is answer in the R&D Network also the
[01:24:52] production Network the network can be
[01:24:54] endogenous but as a local approximation
[01:24:56] those C are not that big but I have not
[01:25:00] done any analysis where the endogenous
[01:25:02] Network structure is the focus of the
[01:25:04] analysis that's a that's a great area
[01:25:07] though okay great thanks a lot Ernest it
[01:25:11] was fascinating and U we learned more
[01:25:15] essentially there are a lot of
[01:25:16] challenges ahead to really do the if we
[01:25:19] want to do industrial policy how to do
[01:25:20] it right and we have to take all this
[01:25:23] ordera mechanisms into account where to
[01:25:25] intervene what's the best way to
[01:25:27] intervene and then on top of it there
[01:25:29] will be all the political governance
[01:25:31] issues we have to think about thanks
[01:25:34] again and uh thanks to all of you uh for
[01:25:37] the audience and hope to see you next
[01:25:39] time soon again and um I wish you a
[01:25:44] great weekend bye-bye see you soon
[01:25:46] Ernest great thank you bye bye