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.
=

# Charting the Uncharted: Oil Sanctions and Dark Shipping (segment of Oxford lecture)

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=ezzVhAA-ABc&t=1541s


## Talk (00:25:41 – 00:44:43)

[00:25:41] Okay, so this was a very very high level introduction to many of the ideas of geoeconomics and you are saying okay fine that looks fantastic as a TED talk but what is the meat give me the real  
[00:25:54] substance so I'm going to tell you about one paper is my paper with Franchesco I don't think it's the best paper in the field but it's the paper I understand the best because I wrote it okay so you need to stick with it now on the other  
[00:26:07] hand I think it's a paper that illustrates very well why the economics is important, what it can give you interesting insights into the world economy and in addition to it where I can use machine learning because one of the goals in my life is to use machine  
[00:26:22] learning in everything that I can. Okay, I mentioned before already that the western countries basically the United States, Canada, Australia, the European Union and the United Kingdom have used repeatedly all sanctions over  
[00:26:36] the last decade or so to accept political power in other countries.  
[00:26:40] Basically, it has been on Iran. So, they stop their nuclear program. Syria, we actually just leave those sanctions because you know the Syrians were the Hassad family were not the people you want to marry into. Venezuela with Tabitha and Maduro and later Russia with  
[00:26:55] a war. Problem is just because I stop you or I pretend to stop you from using or doing some type of export that doesn't imply it's going to happen. Okay. All  
[00:27:08] western countries spend incredible amount of time and effort stopping the import. And I don't know that much about Oxford, but I'm pretty sure if I put my mind to it in the next 60 minutes, I  
[00:27:22] will be able to find at a very reasonable price very high quality coine in this campus. Okay. So just because there is action doesn't imply this doesn't  
[00:27:34] happen. Okay. In particular, what has happened is that the sanctions imposed on these countries had led to the development is what is called dark  
[00:27:47] shipping. So what the heck is dark shipping? The international maritime organization mandated many years ago that every ship in the world over 300 tons which is pretty much every single  
[00:28:00] ship of any importance needs to have this transceiver.  
[00:28:04] This transceiver sends a signal to a satellite with your location, direction, drop, etc. and sends a signal back of where the other ships around you are.  
[00:28:15] Why is this so important? Because you are navigating in the middle of the high sea. It's dark at night in the middle of a storm. The radar is not working very well. You may want you may bump into someone. Okay? So even if this was not  
[00:28:28] mandated, this is a very good idea. If you take any sailing class, at least in the US, the first thing they tell you is get one of those. It's around $700 on Amazon for a basic unit, top-of-the-line unit, $2,000. So, this is very  
[00:28:41] reasonable price. Okay. There also to be the case that this signal is sent to a satellite and the satellite also sends the signal to a public database and Franchesco and I with our co-authors  
[00:28:54] downloaded a database public available and we look at it and what you see is the following. You see ships living over there in the south of Nigeria where there is a lot of oil and they're sending the signal. they're sending the  
[00:29:09] signal and the signal disappears and then the signal reappears over here. Now from time to time these things fail fine and they should fail like 0.1% of times and what you can see  
[00:29:24] is one oil tanker another oil tanker another oil tanker dropping the signal.  
[00:29:29] What is happening? There is illegal smuggling of oil. That tells you that theoretically we could use this information from the database to figure  
[00:29:41] it out how much oil is smuggling is out there. Why do you care? First of all to know how effective sanctions are but  
[00:29:54] second and that goes to the point of what your economics is useful about.  
[00:30:00] Imagine that you are a producer of oil and I really really prevent you from producing oil. What do you think is going to happen with the price of oil in the international market? Go up. I'm the Bank of England. That's an inflationary  
[00:30:13] pressure. What should I do? Well, I may want to increase my interest rates to control inflation. But let's suppose that I'm telling him, no, no, no, you cannot export oil, but he's illegally exporting it to her. What is going to happen with the  
[00:30:28] international price of oil? Who knows? Maybe go up, maybe go down. Maybe in fact may go down because he's selling it cheaper to her because he's illegal. But if that's the case, then we don't have an inflationary pressure. We  
[00:30:41] have an inflationary a deflationary shock. You see how important this is. If you are the governor of the Bank of England, you are keenly interested in knowing what are the consequences of oil  
[00:30:54] sanctions. And as we demonstrate in the paper, the general equilibrium effects of what is going to happen in particular, what is going to happen is that the Russians cannot export oil anymore to the western  
[00:31:09] countries, but they can export oil to China. The Chinese are tough negotiators. And the Chinese say, "Yes, we are going to get your sanction oil, but we are going to impose you quite a  
[00:31:23] haircut. They're only going to pay you 80% of the price that we should be paid on the market. What happens then? China gets cheaper oil. How does cheaper oil from China  
[00:31:36] translate? Well, in the China can produce goods cheaply. And now you get into the international trade. China is producing more cheaply. The imports that come to the United States are more  
[00:31:50] cheaply are cheaper. turns out to be the case you have a deflationary pressure in the US. So surprisingly enough the optimal policy from the perspective of the US is to impose sanctions on Russia not because it will stop Russia from  
[00:32:04] exporting but because in that way Russia could only export to China. China will be tough negotiators and a big chunk of that will be passed back to the United States through lower cost of Chinese imports. Unless of course you decide  
[00:32:18] that importing cheap goods is a bad idea as some person think. But you see why you need economics. You need economics because you need to think about bargaining. You need to think about genetic effects. You need to think about unintended  
[00:32:32] consequences. And this is what the economics can contribute to the discussion. Okay. Now you are thinking okay Jesus how in the world Franchesco and Jillian Al and you are going to be able to do this. This sounds immensely  
[00:32:47] complicated. Well, of course, it means immensely complicated. You know, the people who are smuggling oil don't want to tell you. First thing the Russians do when they invade Ukraine, they take out their statistics on oil exports. But the people who are buying oil illegally  
[00:33:01] don't want to tell you either. Maybe the Chinese don't care. You know, a country that imports a lot of illegal oil. South Koreans, everyone thinks that South Koreans are nice, you know, gentle people. Junai, you know, nice car. And yet South Koreans don't ask too many  
[00:33:16] questions about where the oil that arrives to the refineries come. So South Korea doesn't have much interest in this close. Now the dark ships themselves they are doing something illegal. They don't want to tell you  
[00:33:28] either. Now in some proprietary data that one could try to use for instance we can have satellite imaging. The problem is even if I have pretty much satellite photographs of any place in  
[00:33:41] the planet on demand it cost between$10 to $25 per square kilometer. If I go to my dean and I say you know give me $10 million for this research project even with a different financial environment  
[00:33:54] than the one we live in right now is not going to be to keep give me $10 million.  
[00:33:59] And the other thing we could try to do is for instance buy data from SP global on ownership and insurance data extremely expensive and who knows where this data come from. Okay. So this is what we are going to do and this is  
[00:34:13] super cool. Okay. We are going to go to the AS database and we are going to download every oil tanker  
[00:34:23] ship trip between 2017 to 2023. every single one of them. This is not sampling statistics. I'm not going to report the standard errors. This is the  
[00:34:35] population of the world. Okay? And the frequency is every two seconds. So, you know, if you are worried about how often do I see this observation, now this means we will have billions of observations. And yes, in the longer version of this paper, there's quite a  
[00:34:50] bit of discussion about how to do database management when you have billions and billions of observations.  
[00:34:56] Not today. another date and I have amazing information the international maritime organization number think about as the license plate the type of stop the draw the speed the heading the geographical code and what I'm going to  
[00:35:09] do we are going to do is we are going to develop a machine learning algorithm that is going to take facts about the characteristics of the ship and fact  
[00:35:21] about the trip itself and it's going to determine whether or Not this was a trip that involved oil sanction busting. This is what is known as a  
[00:35:36] problem of unsuper unsupervised learning. What do I mean by that? A standard problems in machine learning are called supervised. Basically I do the following. I'm trying to figure it out. Is this a photograph of a cat or a  
[00:35:50] dog? So what I do is the following. I get 10,000 photographs of cats, 10,000 photographs of dogs. I hire 100 undergrads and I say, "Is this a cat or a dog?" And they label it. And then I use those labels to train a machine learning algorithm. Over here, I cannot  
[00:36:05] do that. So what I need to do is design a system, a machine learning algorithm that will use some information and being able to figure it out in its own. This is a much more challenging problem from  
[00:36:18] the perspective of computer science. And if I had a little bit of time, I will tell you exactly how we do it, but I can't give you the punch line. We are going to look for things that are interesting. I'm going to look at things  
[00:36:32] like the vessel H. Okay, I'm not telling the machine learning algorithm older oil tankers are more likely to be involved in sanction busting. That will be cheating. I'm only telling the machine  
[00:36:46] learning algorithm. Think about the age of the vessel. Is this informative or not? And the machine learning algorithm will decide whether or not this is important and where the site will be a plus or a minus. Think about the number of vessels owned by a  
[00:37:01] commercial operator. Probably if you're into sanction busting, you try to operate few vessels, not a big corporation. Think about the flag state according to the Paris memorandum of understanding list. You are listed in  
[00:37:15] Panama. Maybe this is interesting and we are going to compute a very interesting average trip substitution score. Okay. So this is this is beautiful. This is the type of  
[00:37:29] data that we get. Okay. So we have this ship you know traveling in the middle of the sea. You know this is the wide open sea.  
[00:37:39] Oops. You know hundreds of miles to the left, hundreds of miles to the right.  
[00:37:44] and then it goes back and then it go back again. Okay, so this is not your small road in the middle of the coach walls when you know there's a car coming in the other direction and then you need to turn to the left so both of you can  
[00:37:57] happen. This is the middle of the opening or this one an oil tanker and then does this like an American tourist lost in a turnabout coming back. These guys are spoofing the  
[00:38:15] satellite to try to hide something funky. So we are going to look for these type of trips. Now how are we going to validate this? Are we getting something right or  
[00:38:29] not? Okay. So we are going to look at the following things. We are going to see a ship getting close to a refinery that is sanctioned and then the signal  
[00:38:42] goes down and then we forecast from the time where the signal goes down, how long will it take at the regular speed to get to the suspicious port and then we see that a little bit later the  
[00:38:56] signal goes back. So I know where the ship is at that time with a very high accuracy. And then I ask for the photograph of the satellite. I don't have enough money to ask for millions of photographs, but we have enough money to ask for a few thousand photographs. And  
[00:39:10] we name names. This is Roma. That's the number on K Island in the south of Iran on August 20, 2022. Loading oil in an illegal way. We  
[00:39:25] got this right. These guys were not supposed to be at that place at that moment. Okay. Sometimes what we see is that there is a ship coming in this direction, a ship coming in this  
[00:39:38] direction and then suddenly they shut down their signals and then a little bit later they separate each other. So we again forecast where they are going to meet and we look for a photograph and  
[00:39:52] then you have Abis and Sania Queen on the Persian Gulf on January 28, 2022. So we validate our findings. We know who is cheating. I have a few more  
[00:40:06] examples. The galaxy on K Island, Aseron in Cosmino Russia, the panda and the Elcaparana in Gibralar. And by doing this we are able  
[00:40:18] to actually let me skip this one get information about the total amount of oil tankers involving in funky fields around 555 of them. So around one quarter of the world fleet is involved  
[00:40:32] into doing these things and again going relatively fast we are able to find not only the amount of oil when this is happening and how it responds to the  
[00:40:44] imposing of sanctions. So this is Venezuela and each of these dots is places where there are funky transactions going on. The darker the color the more transactions. There is always a little bit of funky stuff going on in the water. There is also some  
[00:40:59] illegal smuggling. Sometimes it's not because you are doing sanction busting.  
[00:41:03] Sometimes it's just because you are not declaring for tax purposes and things like that. Look what happens after Senator Marco Rubio convinces the US government to impose sanctions on  
[00:41:14] Venezuela. And then in 2019, Russia a few each of these is a trip of a of a tanker. We just put the red line between the beginning and the  
[00:41:26] end of the of the gap. a few in 2021, not many. There were already some sanctions on Russia, but 2022, 2033, many more. And finally, this is super  
[00:41:39] interesting. Okay, 2021, a lot of illegal transactions in the Persian Gulf because this is of course the Iranian sanctioned oil. You see much fewer in 2022, 2023. Why?  
[00:41:54] Turns out to be the case that the Russians are giving the illegal smugglers a better deal than the Iranians and there are only that many oil tankers involved in the smuggling of sanctions. So the guys who are really  
[00:42:05] losing out of the sanctions on Russia is Iran. This is why the use of machine learning and this is why thinking about this in terms of machine learning is such a game changer. So we are actually  
[00:42:18] able to find an evolution of the world oil. This blue line is the legally exported oil in um by sea. This is the  
[00:42:31] illegally exported oil. And you can see it's a big chunk of it. In some months it's actually more illegally exported oil that legally exported oil. And we also have a little bit where  
[00:42:45] this oil is coming from. and where this oil is going. So where is the Russian oil going? By the way, this is a perfect example of a gravity model that you learn from international economics.  
[00:42:56] Russian oil is going to South Korea, it's going to Egypt, it's going to China. A lot of Iranian oil is going to United Arab Emirates, Malaysia, etc.  
[00:43:05] Syrian oil and Venezuela. And what we do in the rest of the paper and I'm not going to present it here because I'm running out of time is then we put all this data in the context of a generic view model of the  
[00:43:17] oil market in the world. We get some predictions of the model. We use those predictions to identify linear projections and a structural vector auto reggressions and we look at the effects of sanctions on objects of interest like  
[00:43:32] interest rates, inflation, output etc. And we can use those to think about optimal monetary policy responses to these type of consumptions. So suddenly by thinking about this geo economic  
[00:43:44] perspective we can come up with data on all sanctions on the general human effects and what is important to tell the bank of England how they should think about. So I hope this was an  
[00:43:57] example of how you can use this very abstract introduction I did at the beginning to actually deliver something that is useful for economics. Now many of you are thinking okay you brought  
[00:44:09] that paper. What next? So what I'm going to tell you is ideas for future research. All of those are free for you.  
[00:44:18] If you want to use any of those for a paper feel free to do it. You don't even need one. You don't need to sign. Except for that feel free no other price that's site five of my random papers okay I want to tell you that I  
[00:44:31] think it's very important to think about the stoastic dynamics that is very important to think about political economy that is very important to think about multipolar world and there is very important to think about economy so I can see over there hundreds of