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

# A Network Approach to Geopolitics

Authors: Ernest Liu, Stephen J. Redding, Bennett Smith-Worthington, David Y. Yang  
Discussant: Christopher Clayton  
Video: https://www.youtube.com/watch?v=U_ssYRB6uPs&t=192s


## Talk (00:03:12 – 00:37:10)

[00:03:14] He's here.  
[00:03:15] Since the group spans many areas, what I would like to do is have everybody introduce themselves very, very briefly.  
[00:03:22] State your name and affiliation and then we'll proceed to the first presentation.  
[00:03:27] Let me thank Jim for allowing us to have these meetings and also for supporting the whole area.  
[00:03:37] And finally, let me remind you that the NBR Code of Conduct for the conference applies to this group as well.  
[00:03:44] I'm a telemajorie from Stanford.  
[00:04:34] I'm a telemajorie from Stanford.  
[00:04:44] I'm a telemajorie from Stanford.  
[00:04:54] I'm a telemajorie from Stanford.  
[00:05:00] people there in the back.  
[00:05:03] Sorry.  
[00:05:03] Thank you.  
[00:05:04] Thank you.  
[00:05:05] So the first paper is a network approach to geopolitics, and earnestly, we'll be presenting.  
[00:05:22] I have my own click.  
[00:05:53] No?  
[00:06:00] It's yours, Mike.  
[00:06:08] No.  
[00:06:21] I'll try to grab it.  
[00:06:45] Okay, great.  
[00:06:54] Thanks very much for having this paper on the program.  
[00:06:57] It's new work with Steve, Bennett, and David.  
[00:07:01] The entire team is here, so we look forward to discussions afterwards.  
[00:07:06] We're going to talk about a network approach to geopolitics.  
[00:07:10] Our starting point is simple.  
[00:07:12] That is to recognize geopolitical issues are rarely bilateral in nature.  
[00:07:16] They almost always involve coordination, collaboration, strategic influence over a network of countries.  
[00:07:24] Examples abound.  
[00:07:26] So going back to the Cold War, the West and the Soviets fought for influence over African and South East Asian countries.  
[00:07:32] When Russia invaded Ukraine, you know, very quickly it became a coordination problem across two continents over sanctions, arms, energy.  
[00:07:41] And more recently, when Trump turned up the rhetoric of acquiring Greenland, within days, France and Germany reacted.  
[00:07:49] And soon after, a coalition of eight European countries issued a joint response to stand behind Dana's sovereignty.  
[00:07:56] The common thread in all of these issues, in fact, most of international issues we talked about, or read about in the news,  
[00:08:03] involve coalitions of countries trying to influence each other, influence third parties to coordinate, to interact, and to shape foreign policy.  
[00:08:12] So we are in this paper hoping to provide a framework to think about geopolitics using this network approach.  
[00:08:20] In this talk, we're going to first start with the model where countries would choose policies, or take policy actions,  
[00:08:28] and then trade off between the alignment of their action and their own domestic or national interest, as well as the alignment between their actions and the action of their allies over an interaction network.  
[00:08:41] Prior to countries taking these policy action choices, they can make costly investment to influence one another to shape other countries' incentive in taking different actions.  
[00:08:51] So in this setting, influence publicly through the network.  
[00:08:57] And try to influence Steve to take a higher or lower action.  
[00:09:00] Other countries who try to coordinate with Steve, their action in equilibrium will also be affected.  
[00:09:06] So bilateral influence will have cascading effects.  
[00:09:10] This model enables us to view geopolitical actions in the data as an equilibrium outcome.  
[00:09:16] So we're going to go to the data and try to recover this geopolitical network using revealed preferences.  
[00:09:22] To validate the network, we recover it with external data, with variables that we think should relate to bilateral relationships.  
[00:09:29] And the main payoff of having such a framework is to perform geopolitical counterfactuals.  
[00:09:34] You know, take an international issue, say China is contemplating taking military actions in the South China Sea.  
[00:09:40] Which are the countries that China would most like to influence?  
[00:09:44] And what would happen to regional outcomes, say between the conflict between Israel and Iran, when China switches preferences and take a stance that's closer to the US?  
[00:09:54] How would that affect the action of countries around the world?  
[00:09:57] So we can do that type of counterfactuals with this framework.  
[00:10:01] So let me start by describing the model.  
[00:10:06] Formally, there are n countries connected by a graph, g, of international relations.  
[00:10:12] This graph, g, it's a matrix. To start, we're going to assume this graph is symmetric for expositional simplicity.  
[00:10:19] So the graph or the network is a weighted undirected graph.  
[00:10:24] We're going to interpret this network as capturing long-standing bilateral relations, alliances, economic dependence.  
[00:10:32] Given a network, there's an international issue that arises.  
[00:10:36] Each country has some domestic interest or national interest over an international issue.  
[00:10:41] And then they can choose a political action.  
[00:10:45] So concretely, as an example, every country may have their own interest in whether to restrict China from having access to advanced semiconductor technologies.  
[00:10:54] And whether they actually restrict, that's the action.  
[00:10:57] Their own interest in restricting, that's the object stated.  
[00:11:02] Each country can exert costly influence over any other country to change their action.  
[00:11:07] So the US may want to pressure the Netherlands and Japan to stop exporting their technology to China.  
[00:11:13] We're going to study a game of perfect information.  
[00:11:16] And for now, we're going to focus a single international issue.  
[00:11:20] Whereas when we go to data later, we're going to recognize that many issues arise.  
[00:11:24] So to summarize, g is a network of bilateral relationships.  
[00:11:29] Theta are draws of international issues.  
[00:11:32] Closing down of Homoos, Iran, Israeli conflict.  
[00:11:36] Each of these instances would be separate copies of this issue.  
[00:11:40] We're going to study the model with a single issue and assuming that in the data there are many issues being drawn when we take the model to the data.  
[00:11:50] Let me first introduce the model without foreign influence.  
[00:11:54] So without influence, the model is a very simple network-based beauty contest game.  
[00:11:59] Every country I is choosing an action to minimize the distance between its action and its own domestic interest,  
[00:12:06] as well as between the distance of its action and the neighbor's action.  
[00:12:12] So in this beauty contest, the solution would be every country just takes the action as a weighted average of its own domestic interest  
[00:12:20] and the action of its network neighbors.  
[00:12:23] Of course, if Steve's action affects my action and Steve's action may affect David's action and so on,  
[00:12:31] so because of network interaction, in equilibrium the action of every country is a weighted average of its own interest,  
[00:12:38] the interest of its neighbors, its neighbor's neighbors, and so on.  
[00:12:41] That's the beauty contest.  
[00:12:45] We're going to introduce foreign influence or bilateral influence to this game.  
[00:12:50] Influence is bilateral in nature.  
[00:12:52] Every country in the world can choose to influence every other country and influence will be a number.  
[00:12:57] It can be a positive number or a negative number.  
[00:13:00] So in terms of payoffs, the first line is just the beauty contest payoff.  
[00:13:05] The third line here is the cost of country I exerting influence on other countries.  
[00:13:10] Country I can influence K to take a higher or lower action,  
[00:13:14] so this influence can be positive or negative.  
[00:13:17] There's a quadratic cost scaled by a parameter kappa of how costly it is to exert influence.  
[00:13:23] The second line specifies how a country having received foreign influence changes its incentive of taking different actions.  
[00:13:31] So when some countries ask me to take a higher action,  
[00:13:34] other countries ask me to take a lower action on net.  
[00:13:37] The net influence country I receive is this summation over all the influence received from different origins.  
[00:13:43] And if the net influence is positive, country I has incentive to take a higher action relative to beauty contest.  
[00:13:50] Vice versa if the net influence is negative.  
[00:13:54] The timing of the model is every country first chooses how much to influence each other,  
[00:13:59] then the game move on to a stage where they take influence as given and play a modified beauty contest game under perfect information.  
[00:14:10] So I'm going to solve the model via backward induction.  
[00:14:15] First, given bilateral influence in place, what action do countries take?  
[00:14:21] The best response looks very much like the beauty contest best response.  
[00:14:25] Every country's action is just a beauty contest component plus the net influence received.  
[00:14:31] So if I receive a lot of influence from foreign nations to take a higher action,  
[00:14:35] my action will be exactly the total net influence received on top of the beauty contest solution.  
[00:14:41] But of course, because other countries try to coordinate with me, these influence will have global effects.  
[00:14:47] So if I influence country K, other countries trying to coordinate with K,  
[00:14:53] other countries will change their actions as well.  
[00:14:56] And that's the consideration that give rise to the X&T influence stage,  
[00:15:00] where every country is choosing the set of influence it can exert.  
[00:15:04] Philippines may want to influence the US not because Philippine cares about coordinating with the US,  
[00:15:10] but because Philippine cares about coordinating with China and China coordinates with the US.  
[00:15:14] So those are the considerations that go into these first order conditions that give rise to the following.  
[00:15:20] I have greater incentive to influence Steve if my neighbors act out of line with me and those neighbors listen to Steve.  
[00:15:29] That's what this first order condition says.  
[00:15:31] So put all of that together, we have the equilibrium.  
[00:15:38] Equilibrium is a vector of actions across countries,  
[00:15:42] which depends on, which is equal to a matrix times the vector of domestic interest across countries.  
[00:15:49] So every country here, the action is a weighted average of the national interest of other countries around the world.  
[00:15:59] So US action depends on the national interest of US, China, its allies, enemies and so on.  
[00:16:06] And how the preferences of different parties influence each other's action is encoded by this matrix L,  
[00:16:15] which depends on the network and it depends on the cost of influence.  
[00:16:19] So model has intuitive properties.  
[00:16:22] If we shut down influence by making the cost of influence go into infinity, the model goes back to a beauty contest game.  
[00:16:29] Conversely, if we make influence as cheap as possible, well, cheap influence homogenizes the world  
[00:16:35] in the sense that all actions taken by countries converge to the global average of their domestic interest.  
[00:16:41] So equilibrium has the following implication.  
[00:16:46] How important a country is, how powerful it is, how pivotal it is, depends both on the network and the issue,  
[00:16:56] which captures disagreement across countries.  
[00:16:59] This is how important our countries to each other in terms of coordination, who's allies.  
[00:17:05] This is an issue that captures disagreement across countries.  
[00:17:11] So US may be the most important and the most central and pivotal along an issue that mainly loads on the conflict between US and China,  
[00:17:20] but US may be on the peripheral, not that important, on localized conflicts, say between India and Pakistan.  
[00:17:27] So same network, different issues, different set of pivotal countries.  
[00:17:32] So that's the model. I'm going to show you one theoretical implication of the model before we go to the data.  
[00:17:39] The implication is the following.  
[00:17:42] We would like to understand what kind of disagreements across countries lead to great divergencing actions across countries  
[00:17:49] and what kind of disagreements get smoothed out by coordination among countries.  
[00:17:55] The right mathematical object to look at to answer that question are the eigenvectors of this network.  
[00:18:02] Intuitively, each eigenvector correspond to a distinct geopolitical cleavage,  
[00:18:08] a particular mode of disagreements across countries.  
[00:18:11] So to give you some example when we go to the data, one eigenvector would capture the east-west divide,  
[00:18:17] where US loads on one side, China Russia loads heavily on the other side.  
[00:18:21] A second eigenvector may capture north-south divide,  
[00:18:24] a third may capture regional conflict between India and Pakistan,  
[00:18:27] a fourth may capture Iran, Israeli, and so on.  
[00:18:31] Those eigenvectors form an orthonormal basis of all possible international issues across this set of countries.  
[00:18:38] And any vector that capture disagreements across countries can be written as a linear combination of those eigenvectors.  
[00:18:46] So when China is contemplating taking over Taiwan,  
[00:18:50] there's a national interest that encodes each country's own preferences over that particular issue.  
[00:18:56] Are you on the side of China or are you on the side of US Taiwan?  
[00:19:00] That issue can be written as a linear combination of east-west divide, north-south divide, and so on.  
[00:19:07] Given this representation, the equilibrium action is the same linear combination over those geopolitical cleavages.  
[00:19:18] It's just that disagreements along certain cleavages get turned up and amplified  
[00:19:23] when they become disagreements in actions, whereas other cleavages get toned down.  
[00:19:29] And that scaling factor depends only on the eigenvalue of the eigenvector in determining the network.  
[00:19:36] So these eigenvalues are properties of the network that says how important is a cleavage for determining the network.  
[00:19:45] These can be formalized, but intuitively that comes from the following interpretation mathematically.  
[00:19:53] So eigenvectors is a basis of cross-country disagreements, and high eigenvalues,  
[00:19:59] those are eigenvectors that capture disagreements where network neighbors tend to agree with each other  
[00:20:05] and loads all the disagreement across blocks of weakly connected countries.  
[00:20:10] So in a world where US is very strongly connected with Canada and Europe,  
[00:20:15] China is very strongly connected with Russia and Iran,  
[00:20:18] an eigenvector that captures issues where US and its allies all agree with each other,  
[00:20:25] China and its allies all agree with each other, but taking the other side,  
[00:20:29] that eigenvector will have a high eigenvalue.  
[00:20:32] Conversely, small eigenvalues, those eigenvectors,  
[00:20:36] they capture issues where neighbors tend to disagree.  
[00:20:39] So a geopolitical cleavage that captures disagreements between US and France or Canada  
[00:20:44] will have a small eigenvalue.  
[00:20:46] Well, when neighbors tend to disagree, but they want to coordinate with each other,  
[00:20:50] their actions will not be so different, because differences in domestic interest  
[00:20:55] will be suppressed by coordination.  
[00:20:57] On the other hand, when neighbors agree but across blocks there's disagreement,  
[00:21:02] neighbors all try to coordinate with each other,  
[00:21:04] US coordinates with France and Canada taking one side,  
[00:21:07] China coordinates with Russia and Iran on the other side,  
[00:21:10] that results in an echo chamber that leads to big differences in actions.  
[00:21:14] That's the theoretical insight coming out of this.  
[00:21:19] So given the model, if we can measure the network G,  
[00:21:23] we can do a bunch of counterfactuals.  
[00:21:26] So one way to go forward would be to perhaps specify a priori what G should look like.  
[00:21:32] Say G or the network is the trade-by-little-trade relation,  
[00:21:36] or G is determined by who's a NATO versus not.  
[00:21:41] Once you have G, you can run counterfactuals to look at what are the kind of issues  
[00:21:45] that make certain countries pivotal,  
[00:21:47] given a particular set of preferences across countries  
[00:21:50] who would China or US would like to influence the most and so on.  
[00:21:55] We are not going to specify G a priori.  
[00:21:57] Instead, we're trying to learn G from a review preference.  
[00:22:01] Using the following strategy.  
[00:22:03] We start from observation that in the real world,  
[00:22:07] countries act on many international issues.  
[00:22:09] Every other week, there's some geopolitical news that come out.  
[00:22:15] Across countries, there may be some covariance in their preferences,  
[00:22:21] in their domestic interests.  
[00:22:23] There seems to be strong alignment between France and Germany and so on.  
[00:22:28] So given a covariance of countries domestic interests over international issues,  
[00:22:34] that would generate a covariance that's potentially observable  
[00:22:38] of countries actions across international issues.  
[00:22:41] With the following intuition.  
[00:22:43] If Steve and I share very strongly strong correlation in our domestic interests,  
[00:22:48] but we always act in polar opposite.  
[00:22:51] That means we cannot be very strongly connected.  
[00:22:54] Conversely, if Jesse and I have always the opposite interests,  
[00:22:58] but we always act very similarly,  
[00:23:00] then we must be very strongly connected either directly or indirectly.  
[00:23:04] So we're going to learn about network connection based on observing  
[00:23:08] how correlated countries are in their preferences or domestic institutions  
[00:23:12] versus how similar they act at the international stage.  
[00:23:17] So to operationalize that strategy, we need to impose one assumption.  
[00:23:23] To us, the restriction is natural that the same cleavage,  
[00:23:27] the same set of eigenvectors that organize a network,  
[00:23:30] also organize how interests covary.  
[00:23:33] Based on the fact that allies may have systematically similar or dissimilar  
[00:23:38] preferences due to shared media, ideology or history.  
[00:23:42] Formally, the assumption is graph stationarity,  
[00:23:45] meaning the covariance-covariance matrix where the issues are drawn  
[00:23:48] for domestic interests have the same eigenvector as a network.  
[00:23:52] This assumption is empirically testable in a sense that it imposes  
[00:23:56] no empirical restriction besides the fact that cross-country covariance,  
[00:24:00] in preferences, and in actions, they share the same principle component.  
[00:24:04] So we provide some evidence of that when we go do the data.  
[00:24:08] Under this stationarity assumption, we can recover the network  
[00:24:12] using exactly the intuition that I've mentioned.  
[00:24:15] That is, if one particular leverage, a geopolitical cleavage,  
[00:24:19] east-west divide, loads very significantly on their interests.  
[00:24:23] So in terms of cross-country differences in their domestic interests  
[00:24:27] over international issues, it's all loaded on east-west divide.  
[00:24:30] But that divide never shows up in countries' actions.  
[00:24:33] It doesn't explain a lot of variance in how countries co-variate in their actions.  
[00:24:37] That means that cleavage must not be very important in determining the network.  
[00:24:41] So putting all that together, we can recover the network.  
[00:24:45] Before I show you the data, let me tell you about  
[00:24:48] one extension that's useful in the data.  
[00:24:52] That is, we have so far assumed that G is a symmetric matrix.  
[00:24:56] It's an undirected network, but geopolitics is not.  
[00:24:59] U.S. is more important to Peru than Peru is to the U.S.  
[00:25:03] So to introduce network asymmetry, we're going to introduce one new object.  
[00:25:08] Each country's structural importance, which is to denote it as pie,  
[00:25:12] think of that as bigger countries as being more important.  
[00:25:16] In the model, what it translates to is more important countries have  
[00:25:20] disproportionate impact on their neighbors and themselves less sensitive to neighbors' actions.  
[00:25:25] And also it's very expensive to influence structurally important targets.  
[00:25:30] And it's expensive to project influence from less important countries  
[00:25:34] to more important countries.  
[00:25:38] So all of our earlier results go through once we adjust in some way the sense  
[00:25:43] of what even with this extension.  
[00:25:46] So now we're going to take a model to the data.  
[00:25:50] We're going to rely on two main data sources, capturing cross-country covariance  
[00:25:55] in their preferences and covariance in their actions.  
[00:25:58] To measure covariance in actions, we're going to rely on  
[00:26:01] United Nations General Assembly voting data.  
[00:26:04] So if two countries vote very similarly on a set of issues,  
[00:26:08] then two countries, we infer that they have high covariance in actions.  
[00:26:12] For preferences, that's harder to get.  
[00:26:15] So we're going to rely on the VDAM data set constructed by political scientists  
[00:26:19] to score countries based on various dimensions of domestic institutions.  
[00:26:25] Think freedom of press, gender equality, voting rights, and so on.  
[00:26:30] So when two countries share a lot of similarity across a wide range of metrics,  
[00:26:35] rating domestic institutions, we view that they have high covariance  
[00:26:40] between each other across international issues as well.  
[00:26:44] And in terms of structural importance, we're going to write it as an increasing  
[00:26:48] function of GDP and calibrate it by exploiting asymmetry in outward  
[00:26:53] versus inward influence implied by the model.  
[00:26:57] So with those construction in mind, we're going to show you three sets of  
[00:27:02] empirical results.  
[00:27:04] We have done the joint diagonalization and recovered the network.  
[00:27:08] We're going to describe dominant eigenvector and describe features of the network.  
[00:27:13] We're going to do some validation exercise to show that the network we recovered  
[00:27:17] from voting and scores of domestic institutions seem to relate to variables  
[00:27:22] that we think should capture international relations.  
[00:27:25] And then we're going to run some counterfactuals, changing China's preferences  
[00:27:28] towards the U.S., for example.  
[00:27:36] On this figure, we show you the dominant geopolitical cleavage over time  
[00:27:40] across six decades.  
[00:27:43] Now, in the data, there's a very salient East-West divide, where for the first  
[00:27:49] three centuries in the data between 1960s and 1980s, the East-West divide  
[00:27:54] is largely between U.S. and allies versus Russia and its allies.  
[00:27:59] It's the Cold War eigenvector.  
[00:28:01] And for the after 1990s, it turns into U.S. versus China.  
[00:28:08] Now, the East-West divide shows up as the dominant cleavage, the most important  
[00:28:12] cleavage in determining the network or international relations in all  
[00:28:16] decades except for 1970s.  
[00:28:19] That's the period of the taunt where denuclearization and the escalation  
[00:28:24] of Cold War makes that cleavage the second most important, not the most  
[00:28:28] important.  
[00:28:30] And notice that, you know, 1990, that's right after the fall of  
[00:28:33] Iron Curtain, Russia becomes much less salient in this eigenvector, in this  
[00:28:39] East-West eigenvector, in determining the network.  
[00:28:43] So we have a collection of eigenvectors for each decade.  
[00:28:49] We put eigenvector together, infer how important they are based on how  
[00:28:53] much they explain covariance of the actions versus covariance of the  
[00:28:56] imprevences.  
[00:28:58] So we construct a network for each decade.  
[00:29:02] This figure is meant to show you that the network seems to be fairly  
[00:29:05] persistent over time with the exception of 1990s, with the exception of a  
[00:29:10] trend break in the network after the fall of Iron Curtain.  
[00:29:15] So every country's connection seems to be reasonably persistent from  
[00:29:19] decade to decade except major reorganization of the network going  
[00:29:23] from the 80s to the 90s.  
[00:29:27] In this geopolitical network, it's a full matrix.  
[00:29:30] We're going to look at two particular roles to think about which are the  
[00:29:34] countries that weigh the U.S. very importantly, and which are the  
[00:29:38] countries that weigh China very importantly.  
[00:29:42] If you look across these two, you know, for the U.S., we pick up all the  
[00:29:45] usual allies.  
[00:29:47] In North America, in Western Europe, Australia, Japan, and South Korea.  
[00:29:53] The map on China is mostly Southeast Asia, many African countries,  
[00:29:57] some Middle East and Russia.  
[00:29:59] The more formal validation exercise would be to correlate the network with  
[00:30:06] other variables that we think should relate to international relations.  
[00:30:10] So political distance, distance in political institutions, gravity  
[00:30:16] variables on trade and distance, or whether there are cultural similarities  
[00:30:21] between pairs of countries.  
[00:30:23] The network recovered correlate very strongly with all these measures.  
[00:30:28] And a more powerful exercise of validation is the following.  
[00:30:33] We observe U.N. votes, right?  
[00:30:36] Countries vote on individual resolutions.  
[00:30:38] Yes, no, abstain.  
[00:30:40] We treat those votes as the vector of actions, use the model to  
[00:30:44] invert from those actions the underlying preferences.  
[00:30:47] So when countries vote to for or against sanctioning Iran, we take  
[00:30:52] those votes and back out their preferences and the implied influence  
[00:30:56] they would like to impose on all other countries in the world.  
[00:31:00] We collapse those bilateral influence down to country pair by year level,  
[00:31:05] and we show that this model implied influence predict very strongly  
[00:31:09] diplomatic visits, the country leader visits.  
[00:31:12] After controlling for country year fix effect, once there are  
[00:31:16] deviation in greater bilateral influence in that year between  
[00:31:20] that country pair within the model implied, roughly predicts 5%  
[00:31:24] more leader visits within that pair.  
[00:31:27] Finally, the main payoff of the paper is to run geopolitical counterfactors.  
[00:31:36] So recall that actions depend on network and depends on the  
[00:31:43] set of disagreements.  
[00:31:45] So we're going to run, we're going to show you this is the action  
[00:31:51] across countries represented in color when we load on the dominant  
[00:31:56] east-west divide, the dominant cleavage.  
[00:32:00] So the color of the country indicate whether a country is on one  
[00:32:04] side or on the other side of the cleavage.  
[00:32:07] You see U.S. and allies taking the blue side, China and  
[00:32:10] the red side and so on.  
[00:32:13] So that's the action implied by the dominant cleavage.  
[00:32:18] The counterfactual here is to hold everybody else the messy  
[00:32:22] interest constant, but change China's the messy interest to be fully  
[00:32:26] aligned with the U.S. That's the picture on the map.  
[00:32:29] I want to highlight three features.  
[00:32:31] One, even when China's interest is fully aligned with the U.S.,  
[00:32:36] China's action doesn't fully converge to that of the U.S.  
[00:32:39] Because China's action is shaped by the coordination to its allies  
[00:32:43] and its allies still have the preferences more towards the right.  
[00:32:47] Second, there's changing China's preferences, changes the actions  
[00:32:54] of a bunch of countries that used to coordinate strongly with China.  
[00:32:58] But the world is not fully homogenous because Russia and all these  
[00:33:02] other countries, Iran, they still have preferences that remain unchanged.  
[00:33:11] And finally, the more subtle feature is if you look at the color of the U.S.,  
[00:33:16] U.S. actually take an action that's less extreme.  
[00:33:19] So when China takes its position going from extreme right to the left to  
[00:33:25] align with the U.S. preferences, that actually makes U.S. action less extreme.  
[00:33:29] There's less need of taking extreme action and influence other countries.  
[00:33:34] So that's implication of this counterfactual.  
[00:33:39] A different counterfactual we can run is to think about what makes a country powerful.  
[00:33:45] The various notions of power, right?  
[00:33:48] One notion would be if we perturb a country's own preferences,  
[00:33:52] how much does that perturbation gets translated into changes in its own actions?  
[00:33:56] The freedom to express itself despite this geopolitical game that we're playing.  
[00:34:01] A second would be if we change the action of that country,  
[00:34:04] how much does it shift the action of all other countries?  
[00:34:07] How much influence outwards does it project?  
[00:34:10] So China and U.S. lying on are differentially important in those two different exercises.  
[00:34:16] Figure on the left, we are shifting each country's preferences one by one  
[00:34:21] and calculate how much does its own action change as we change the preference by the same unit.  
[00:34:28] U.S. has greater freedom to express its own other method of interest than China.  
[00:34:34] But in terms of moving others, China seems to be more influential than the U.S.  
[00:34:39] When Chinese preferences change, that induces more changes in action on others.  
[00:34:44] So this picture is shown using the dominant geopolitical cleavage as the disagreements across countries.  
[00:34:53] So if we look at the difference between China and the U.S.  
[00:34:57] and China on disagreements where U.S. and China are no longer the most important of pivotal,  
[00:35:05] this picture would change drastically.  
[00:35:08] I have two minutes. Let me show you a few other implications.  
[00:35:13] One, we can look at which countries receive the most influence in the 2010s.  
[00:35:19] The generating process, the covariance matrix that draws countries' preferences across international issues.  
[00:35:26] We draw from that estimated covariance matrix in the world,  
[00:35:29] calculate model implied influence and look at in the 2010s which countries receive the most influence from other countries.  
[00:35:37] Syria, Iran, Yemen, Venezuela, these countries are at the center of conflict,  
[00:35:42] and it seems to be in between or in pivotal bridges between blocks, they seem to receive the most influence.  
[00:35:49] And interestingly, U.S. receives a lot of foreign influence, China not nearly as much.  
[00:35:59] Another exercise would be to think about through what channel does influence operate?  
[00:36:05] If U.S. wants to influence another country, does most of the influence arise by U.S. preferring that country itself?  
[00:36:12] Or does U.S. generate most of the influence through indirect network connections?  
[00:36:16] How much of that influence generate through if I try to influence you and you change your action and affect all your allies?  
[00:36:24] So the black line, if we do this direct versus indirect influence decomposition,  
[00:36:30] on average across countries 40% of the influence come from indirect sources.  
[00:36:35] For U.S. Russia, China, the bigger economies, the hegemones, indirect influence account for lower fraction  
[00:36:43] because they're more directly connected to the rest of the world.  
[00:36:47] Okay, it's time. So let me conclude here.  
[00:36:51] This is a network approach to geopolitics.  
[00:36:55] It's a framework that enables us to think about geopolitical interactions as the equilibrium outcomes,  
[00:37:00] and we can perform geopolitical counterfactors. We're very excited about this new paper. Thank you.  
[00:37:04] Thanks so much. And the discussant is Chris Clayton.  

## Discussion (00:37:10 – 00:54:16)

[00:37:13] Okay. It's a pleasure to be here and get to discuss this really cool paper from Ernest and his co-authors.  
[00:37:28] So what's great about this paper is it's got this complicated looking structure to it.  
[00:37:36] But at the end of the day, it's a model that's set up very tractably to give closed form theory.  
[00:37:42] And you can take that closed form theory and use it to look at how each of these different countries  
[00:37:49] strategically influences each other within this beauty contest setup.  
[00:37:54] It goes from there and presents its key decomposition.  
[00:37:59] It shows that once you've got this characterization of not only how countries respond to their neighbors  
[00:38:07] within the beauty contest, but also without they actually then go and influence each other,  
[00:38:13] you can perform a decomposition.  
[00:38:15] Formally it's going to be a spectral decomposition to actually uncover the underlying geopolitical network  
[00:38:23] from the observed actions that you can see in the data.  
[00:38:27] And then the paper goes from there. That's a lot of these really cool graphs that Ernest showed you  
[00:38:33] and maps out what these different cleavages look like to see how they trace out over time.  
[00:38:40] So I think it's a really cool and interesting paper.  
[00:38:43] I'm going to start by just giving like a brief overview of the kind of approach that these authors are taking  
[00:38:52] in case it went a little fast during a short talk.  
[00:38:56] And then give a few comments for where I think some of the exciting places this paper can go.  
[00:39:03] So let me give a simple illustration here.  
[00:39:06] We've got two countries, one and two, so it's going to be much simpler than the more general setup the authors have.  
[00:39:15] They're going to play this two-stage game.  
[00:39:18] In the first stage, each country is going to influence potentially each other country.  
[00:39:24] And in the second stage, each country is going to choose an action.  
[00:39:28] So let me put in two symmetric countries.  
[00:39:31] So I'll just characterize one of them.  
[00:39:33] And then the other is symmetric.  
[00:39:35] Each country has a quadratic loss over divergence of its action from its preferences.  
[00:39:42] It has a quadratic loss of its divergence of its action from the other countries.  
[00:39:49] So this is the beauty contest element to it.  
[00:39:53] And then it's also got the influence capability.  
[00:39:56] Each of these two countries can put what I would usually call a wedge,  
[00:40:00] but there's a lot of interpretations into the decision problem of the other at a quadratic loss.  
[00:40:07] This is the X anti-influence game where they're going to shape each other's ex-posit portions.  
[00:40:14] Now what's nice about this is this beauty contest model.  
[00:40:18] It can be summarized by a matrix.  
[00:40:20] Here it's a really simple symmetric matrix capturing the weight that each of them puts in the beauty contest on the other section.  
[00:40:29] And of course in the general theory, this is a bad notation, I guess,  
[00:40:33] because they use G for the whole matrix and I use it for the off-to-hackinal components.  
[00:40:38] So sorry for that confusion.  
[00:40:41] So let's solve this out.  
[00:40:43] Simple.  
[00:40:44] It's like this is what's nice.  
[00:40:45] It's just a couple equations here.  
[00:40:48] But even in the paper it's all linear quadratic.  
[00:40:51] Once it's all linear quadratic, you can solve it all out with pen and paper.  
[00:40:55] That's one of the fun things actually about this paper.  
[00:40:58] Okay, so take the first order condition for the optimal action.  
[00:41:02] You want to align with your own preference, you want to align with your neighbor,  
[00:41:06] and you have some influence put in from your...  
[00:41:09] Okay, so your underlying action is going to be a weighted combination  
[00:41:14] between your preference, net of the influence put in by the other country,  
[00:41:19] plus the action of your neighbor.  
[00:41:23] Okay, where G is capturing the relative importance of the beauty contest element.  
[00:41:29] So solve it out nice.  
[00:41:31] You get a weighted average of your preference, net of influence,  
[00:41:35] and your neighbor's preference, net of influence.  
[00:41:40] Let me think you can go and you can solve out what is this actual influence.  
[00:41:44] Well, the simple envelope theorem in the ex ante stage,  
[00:41:48] all you care about is derivatives you're taking over your neighbor's action,  
[00:41:53] because you're optimally choosing your own action in the ex post stage.  
[00:41:57] So you get your influence you're placing, depends on what cost of influence,  
[00:42:02] and then how much you're shifting your neighbor's action  
[00:42:06] versus how much there is exceeds.  
[00:42:09] So like what's the intuition here?  
[00:42:11] If your neighbor is taking a larger action than you,  
[00:42:14] you want to bring them back down towards you  
[00:42:17] so that they align more with your action.  
[00:42:20] And so you're going to do that by putting attacks on their action within this setup.  
[00:42:25] And then you can go connect this of course, solve,  
[00:42:28] like from here, like you can solve out in closed forums.  
[00:42:31] The search is kind of uglier and uglier.  
[00:42:33] So this is where I decided to stop the solving out by hand.  
[00:42:40] Now, what can we do with this?  
[00:42:43] So we got this nice representation.  
[00:42:45] Let me take the limit where it's super expensive to influence the other country  
[00:42:50] so that we've just got the beauty contest element.  
[00:42:53] And of course what's great about the paper is it shows that like the simplified version  
[00:42:57] I'm going to do generalizes a lot in terms of the spectral decomposition insight.  
[00:43:03] So I'm really just giving here one example of it.  
[00:43:07] So then we can represent the action vector as a matrix multiplying the vector.  
[00:43:14] We're scaling for each of these two countries.  
[00:43:17] It's a nice symmetric matrix combining the weight you're putting on your own preference  
[00:43:23] and the weight through the graph that you're placing on your neighbor's friends.  
[00:43:28] Now what we're going to do is we're going to perform a decomposition of this,  
[00:43:33] re-representing the underlying actions as combinations of the underlying eigenvectors here.  
[00:43:41] What's this intuitively doing?  
[00:43:44] It's intuitively giving us a basis for these underlying actions  
[00:43:48] telling us how much these actions tend to coincide versus diverge from each other.  
[00:43:55] So for example, if we uncovered here within the basis an eigenvector that looked like 1 minus 1  
[00:44:01] that would be a kind of cleavage that holds one country apart from the other one.  
[00:44:08] Here we've got two by two symmetric matrix so it's going to be like trivial the eigenvectors that we get out of it.  
[00:44:15] Once you're in the kind of more general graph you get much more interesting.  
[00:44:21] Now what's nice here is to go back remember some of the properties of doing this eigenvector decomposition.  
[00:44:29] If we take a simple linear transformation of the underlying matrix we preserve the same underlying eigenvectors.  
[00:44:38] And what have they done?  
[00:44:40] Well in this simple example all we've done is done exactly that linear transformation of that underlying.  
[00:44:49] And so we're going to recover exactly the same eigenvectors from doing the decomposition on the action matrix  
[00:44:57] as we do from doing it on the graph.  
[00:45:02] And actually here they're going to be the kind of common vector 1 1 and the divergent vector 1 minus 1.  
[00:45:10] So it's the common component of the two actions and the divergent component of the two actions.  
[00:45:17] And then it gets much more interesting when you've got a bunch of these countries and a more complicated group.  
[00:45:22] So the paper's kind of big contribution here I think is to not only show this property as more general  
[00:45:30] but also to then conduct the empirical evaluation that Ernest walked you through very carefully.  
[00:45:37] I have very little comparative advantage on walking you back through the empirical decomposition.  
[00:45:43] So I will have rested on my trust in Ernest to do a really good job showing it to you all.  
[00:45:50] I'm glad I had rational expectations.  
[00:45:53] So let me dive right into my comments.  
[00:45:56] So first question I think really comes to mind.  
[00:46:00] What I like about this paper is that not only is every country able to be potentially have some significance within the network  
[00:46:11] but also every country can try to influence both themselves.  
[00:46:16] So that's like easy to get hidden in the equation is you can apply influence to other countries  
[00:46:22] but you can also apply influence to yourself.  
[00:46:25] So think of this as like the differentiation between offensive statecraft  
[00:46:30] where I try and get another country to change their behavior  
[00:46:33] and defensive statecraft where I try and prevent other countries from changing my own behavior.  
[00:46:40] So they've got both of these.  
[00:46:41] It can be like hard to spot in the paper but they've got both.  
[00:46:45] So on these two components the ability to be important in the network and to strategically influence others.  
[00:46:53] On the first question I couldn't agree for paper off show already documents evidence  
[00:46:59] that these indirect effects by the network are important.  
[00:47:03] And moreover you know even if you thought there's kind of a handful of small countries in the world  
[00:47:09] there's also a ton of small countries in the world  
[00:47:12] and you add up impact on a ton of small countries  
[00:47:16] you're going to get to a big impact overall.  
[00:47:19] The second one is where I think the paper has a lot to say on  
[00:47:23] and I'd be interested to see what it kind of more on what it has to say.  
[00:47:28] So for example let me take I think Ernest didn't show this or at least I missed it.  
[00:47:34] This is a cool paper from a counter fact or I'm sorry  
[00:47:38] cool figure from a counterfactual exercise where they think about  
[00:47:44] suppose that a country got a change in its preference  
[00:47:49] how much would that propagate through to its own behavior  
[00:47:53] versus the aggregate impact on its neighbors.  
[00:47:56] So what's cool is you see like the effect on itself  
[00:48:01] there are some outliers the US is like very kind of  
[00:48:05] it changes its preference it changes its action  
[00:48:09] but there's a lot of like a dissipation across countries  
[00:48:14] like even all these like collection of small medium big country like Great Britain is in here  
[00:48:21] they all translate like 25% into that change.  
[00:48:27] Now it's kind of more spread out when you go on the influence of others  
[00:48:32] like China is a big standout interestingly India  
[00:48:36] US is like you have to move all the way up here but it's there  
[00:48:40] but then you have a bit of a trick.  
[00:48:43] So my question here I guess is like on defensive statecraft  
[00:48:48] does my preference or does my ability to shape my own behavior  
[00:48:52] translate to changes in actions.  
[00:48:55] I think unquestionably here we've got kind of good evidence  
[00:48:59] that every country is pursuit you know can kind of shape its own behavior  
[00:49:05] through defensive statecraft.  
[00:49:07] On offensive statecraft I do think influence here seems to be more concentrated.  
[00:49:13] This isn't quite the right figure to be looking at for that question  
[00:49:18] because this is like if there's an exogenous preference shock what happens  
[00:49:23] whereas the offensive statecraft is more if I start trying to strategically influence  
[00:49:28] what happens but it's like a good kind of first pass eyeball  
[00:49:33] at what we might expect.  
[00:49:35] And so my like the first big set of questions I would really like to kind of hear more from  
[00:49:42] as this paper evolves is should we think of NS2 or like a very small end world  
[00:49:48] as of strategic offensive players or what I would often call a hegemon  
[00:49:54] as a good approximation to the world.  
[00:49:57] And this is like really helpful for especially people like me doing applied theory work  
[00:50:03] when I think about how I write and shape.  
[00:50:08] It's also like got me thinking about here we've got a model really of the network of actions being taken  
[00:50:16] that's kind of in some sense different from the model of influence  
[00:50:20] like you know the types of policies like sanctions export controls military capacity  
[00:50:28] the underlying network of that component of strategic influence  
[00:50:33] may actually end up looking different from the network of the end actions  
[00:50:38] that are trying to influence.  
[00:50:41] I would be you know my prior would be the US and China are particularly important  
[00:50:46] within that kind of network of strategic influence.  
[00:50:50] I'd be really interested to know how like it seems like the papers methodology is well suited  
[00:50:56] to make this type of differentiation.  
[00:50:59] I'd be very curious to hear what it's asked to say.  
[00:51:03] There's also a lot you know we see countries like the EU like the BRICS  
[00:51:09] trying to coordinate their strategies both in terms of their actions  
[00:51:14] and in terms of their influence.  
[00:51:16] I think that kind of idea of a lot of these countries aggregating up their strategic influence  
[00:51:22] to have a large impact may actually have a lot of bite.  
[00:51:26] And so I'd be curious even though you know maybe if small n is a good approximation  
[00:51:31] to individual countries and their strategic influence  
[00:51:35] how much actually coordinated influence ends up mattering within the network.  
[00:51:40] Okay I'm like running short on time.  
[00:51:43] So let me do my next two comments quickly and then we can talk more as we'll be useful.  
[00:51:49] Second this is I think really a model of an exogenous network structure  
[00:51:54] I think the paper already provides a bit of evidence that the network is evolving over time.  
[00:52:01] Now I personally view at least within the influence network  
[00:52:06] as there's a lot of endogeneity to this underlying network.  
[00:52:10] Like when we think about kind of underlying externalities like strategic complementarities  
[00:52:16] US dominance of the financial system, export controls containing technology  
[00:52:22] a lot of these are I think for the purpose of shaping the network  
[00:52:26] so that the US has more power more influence with it.  
[00:52:30] Like another extreme if a country puts itself into autarky  
[00:52:34] it sort of disconnects itself from the network.  
[00:52:38] You can only do us so much in one paper so I think a really great direction to go  
[00:52:44] maybe for the next papers to really start thinking seriously  
[00:52:48] as we endogenize this network and let countries strategically shape it  
[00:52:52] how do they affect its evolution?  
[00:52:57] And just like very quickly this is a beauty contest model  
[00:53:02] we want both the US to respect its underlying preference  
[00:53:07] and to align with China.  
[00:53:09] I might have written like if I had just like done my first past writing in this model  
[00:53:14] I might have written this as the US wants its action  
[00:53:18] and it also wants China to take its action.  
[00:53:22] Like this might have been my kind of theory of say like export controls to China  
[00:53:27] the US wants Sweden or the Netherlands or whatever European country  
[00:53:34] to actually go and reinforce its export controls  
[00:53:38] for the purpose of blocking China.  
[00:53:41] I don't know which of these two models is totally right though  
[00:53:45] and it depends on the context.  
[00:53:47] I'm here to know kind of what micro foundations are best represented  
[00:53:52] by each of these two different sets of preferences  
[00:53:55] as well as to know like do the models decomposition actually go through  
[00:54:00] with this type of setup.  
[00:54:02] Okay Jesse is telling me to get off the stage so it's a great paper  
[00:54:06] and I'm really excited to see where it goes from here.  
[00:54:09] Thanks so much and Ernest you want to just collect a few questions  

## Q&A (00:54:16 – 01:06:21)

[00:54:20] and then answer.  
[00:54:53] I'm curious could you talk a little bit about the assumption  
[00:54:57] and how much you need it.  
[00:55:04] I just want to correct you from the presentation  
[00:55:06] we sort of observed the issues data and invert them to recover this network of G.  
[00:55:11] I by the way got the impression from Chris's discussion  
[00:55:14] that it could also be about observing the geopolitical actions  
[00:55:17] in emerging loads.  
[00:55:19] My comment then is related to MLBs I think it's sort of  
[00:55:22] not immediately obvious to me that measuring data is easier than measuring  
[00:55:26] G. G is country by country thing.  
[00:55:28] There are lots of countries but there are also lots of lots of issues  
[00:55:31] that one could measure including not being able to.  
[00:55:34] It's the same for G.  
[00:55:39] My comment is related to what Chris said at the end  
[00:55:43] depending on the context of one of the micro foundations  
[00:55:46] because it seems to me that your framework would be great  
[00:55:49] for a serious job like global funding goods  
[00:55:51] or global policies where I had some ideas  
[00:55:55] like data and everybody else  
[00:55:58] who was saying to me and so if I'm more similar  
[00:56:01] then when you go and more similar culturally or historically  
[00:56:04] that's it would be at most of the other countries.  
[00:56:07] But in many other cases it's quite the opposite.  
[00:56:09] Countries that are very similar are for control Jerusalem.  
[00:56:12] There are really big countries where there are rival goods  
[00:56:15] but if I'm more similar maybe it should be a rush  
[00:56:17] I would like to find over it that I don't want you to do the same thing  
[00:56:20] because it would be like I want Jerusalem for myself.  
[00:56:23] So it would be a 13 that is connected  
[00:56:26] between common policy, I mean what that  
[00:56:30] maybe I wanted to follow my policy, my export policy.  
[00:56:33] So depending on the context when you go to the cigarette  
[00:56:36] and the trainer and the cigarette  
[00:56:38] you might find different outcomes  
[00:56:41] so different data depends on the tools  
[00:56:43] and the public tools is about policies.  
[00:56:46] Let me take these.  
[00:56:49] I'm at capacity of giving track.  
[00:56:53] So first thanks very much for the discussion.  
[00:56:56] All great comments.  
[00:56:58] Especially like the offensive versus defensive  
[00:57:02] statecraft interpretation of the influence.  
[00:57:05] I do want to mention that some of the figure  
[00:57:08] so how important a country is  
[00:57:11] how whether it's two hegemons versus  
[00:57:14] a bunch of smaller economies  
[00:57:16] it depends not only on the network but also on the issue.  
[00:57:19] For some of the counterfactual figures where we're showing  
[00:57:22] where we perturb preferences  
[00:57:24] US has more freedom to express its own interests  
[00:57:27] and China influence others  
[00:57:29] that's on the US China divide  
[00:57:31] that's the starting point where we perturb preferences.  
[00:57:34] If we start from a different divide  
[00:57:36] say India-Pakistan conflict  
[00:57:38] maybe US is not the country that shape actions the most.  
[00:57:42] So I think that's a feature  
[00:57:45] that we should highlight more in the paper  
[00:57:47] but great I think it's important to highlight that in the paper  
[00:57:51] and we're not doing that enough.  
[00:57:53] We are indeed thinking about a dynamic version of the paper  
[00:57:58] where this period's influence  
[00:58:00] shapes next period network in some ways  
[00:58:02] and that's in the works as a follow-up.  
[00:58:09] Emily's question on separability.  
[00:58:11] So I think the language in the  
[00:58:15] policy IR literature is issue linkage  
[00:58:18] that is you have many issues that are jointly negotiated on  
[00:58:22] and taking action.  
[00:58:24] So we have not analyzed that case in the setting  
[00:58:28] so we don't have much to say about that  
[00:58:30] but that would be a good direction to explore.  
[00:58:33] I agree.  
[00:58:35] My co-authors feel free to.  
[00:58:49] One can do this exercise in more refined ways  
[00:58:52] but I think there's one network on climate issues  
[00:58:55] there's another network on different set of issues  
[00:58:57] so for different issues network may themselves look differently  
[00:59:00] that would be another way to find the exercise we do.  
[00:59:07] On your point about similar countries may  
[00:59:16] on some issues may even more likely to diverge  
[00:59:20] that's built into the model as well  
[00:59:22] in the sense that we're not requiring  
[00:59:25] that countries that share preferences  
[00:59:27] must be more strongly connected.  
[00:59:29] We're saying the set of cleavages  
[00:59:32] that the preference covariance is formed based on  
[00:59:37] are the set of underlying basis for the network  
[00:59:40] but the loading can be negative can be weekly correlated.  
[00:59:44] So in that sense similar countries may have  
[00:59:47] very different preferences or connected countries  
[00:59:49] may have different preferences and so on  
[00:59:51] that's allowed in the network.  
[00:59:53] It's just that empirically when we recover  
[00:59:55] when countries intuitively what identifies the network  
[01:00:01] is if we always act very similarly  
[01:00:04] even though our preferences are not very similar  
[01:00:07] then the model says you must be very connected  
[01:00:09] directly or indirectly that's the idea  
[01:00:12] but there's no restriction built in.  
[01:00:14] So let me take a few more questions.  
[01:00:18] For example, when you put in a model  
[01:01:02] for example China and USA, are you moving first  
[01:01:06] so that you can say for example  
[01:01:09] if you're moving can it be more of a suggestion?  
[01:01:13] Yeah that would be an interesting extension.  
[01:01:15] In keeping the one-shot games  
[01:01:19] so we are working on an extension  
[01:01:21] where the network actually evolves  
[01:01:23] that's the dynamics we've been focusing on.  
[01:01:25] In keeping this one-shot game structure  
[01:01:27] there are various extensions that are interesting  
[01:01:29] that's been exploring other types of linear-curriotic games.  
[01:01:33] One would be Stackelberg moves.  
[01:01:36] In this multiplayer context you can also think about  
[01:01:39] a group of countries coordinating on their actions.  
[01:01:42] A subset of countries form an alliance  
[01:01:44] and collude on particular issues  
[01:01:46] that would be interesting too.  
[01:01:48] Those are the things that we could explore more we haven't done.  
[01:01:51] Just quickly related to that even in the static setting  
[01:01:53] there's a sense of which we have a feature  
[01:01:55] that's called the Anthos-Los Network  
[01:01:57] which is a pre-anthos-los-photomaker  
[01:01:59] in some way or two you could alternatively think about that  
[01:02:01] as investing to change the network.  
[01:02:03] There's kind of a conceptual issue that the strategic influence  
[01:02:06] is kind of a vital element of the network  
[01:02:08] because you've got this rich set of bilateral investment  
[01:02:10] which will make it so even within the static game  
[01:02:12] there's a flavor of it sort of changing the environment.  
[01:02:45] Let me take a few and I'll answer.  
[01:02:47] It's a bit of a detriment.  
[01:02:49] Okay.  
[01:03:43] My friend doesn't care.  
[01:03:45] In your case even if I don't think it's a problem  
[01:03:48] it's just the...  
[01:03:50] Is there any...  
[01:03:52] Can we possibly do that?  
[01:03:54] For example, this year I care about Jerusalem  
[01:03:57] but I'm trying to figure out if I don't have a partner  
[01:04:00] which is a really important issue.  
[01:04:02] Let me answer this first.  
[01:04:04] I'll go back to these two and take...  
[01:04:06] Ernest, do you take a minute or two to respond?  
[01:04:08] Yeah.  
[01:04:10] Okay.  
[01:04:12] The results rely a lot on a property called spectral decoupling.  
[01:04:17] That is the set of eigenvectors shaping covarious actions  
[01:04:21] is the same as those shaping covarious inferences.  
[01:04:24] To get that spectral decoupling one has to impose  
[01:04:27] some restrictions on the degree of asymmetry.  
[01:04:30] What we have is a form of quasi-symmetry.  
[01:04:33] That is we start from a symmetric network  
[01:04:35] but allow one vector to be the source of all asymmetry.  
[01:04:39] That's fully handled.  
[01:04:41] More generally, the spectral decoupling would fail.  
[01:04:44] So you can still do all the theory parts  
[01:04:47] but taking to data would require different identifications  
[01:04:50] that would be much more complicated.  
[01:04:53] The second is an interpretation question.  
[01:04:55] I think that's fully consistent with how we are viewing things.  
[01:04:58] So we are saying there's one network  
[01:05:00] where you want to go in with your allies  
[01:05:02] but you can draw different issues.  
[01:05:04] There will be issues where your ally can be strong about  
[01:05:07] you don't really care about  
[01:05:09] but if that issue gets voting in the UN  
[01:05:12] or if your foreign minister have to go to the press  
[01:05:15] and say something, comment on the issue  
[01:05:17] you may not care about that issue  
[01:05:19] but you may still want to support your ally  
[01:05:21] and coordinate with the ally.  
[01:05:23] So that's fully consistent with this setting.  
[01:05:25] On the measurement, World Value Survey  
[01:05:28] we have to think about whether that's feasible  
[01:05:31] in this particular setting but that's a great suggestion.  
[01:05:34] In the interpreting actions  
[01:05:37] the model is deliberately reduced form.  
[01:05:41] Actions and preferences  
[01:05:43] because we think the underlying economics  
[01:05:46] represented by this reduced form representation  
[01:05:48] can be fairly broad.  
[01:05:50] You can think about sanctions, military actions  
[01:05:52] versus when we go to data  
[01:05:55] the systematic data source of capturing  
[01:05:57] a panel of cross country actions  
[01:06:00] that would be a UN voting.  
[01:06:02] So that's the data that enables to do this empirically  
[01:06:05] but we feel the model can be interpreted more liberally.  
[01:06:08] Great, I think we have to take the rest of the questions over coffee  
[01:06:12] so thank you so much.  
[01:06:13] Thank you.  
[01:06:14] And we'll be back at 10.15 for the next paper.