 Thank you very much. It's a great honor to be here and I very much appreciate the invitation. And hopefully this research will be something that you can find helpful in your day-to-day work on intellectual property and I assume, you know, the consequences that that has on well-being. So as she was mentioned, my work over the last decade or so part of it has been devoted to thinking and studying carefully the process of technology diffusion. And so I've been producing many papers. This is the last one. It's shown with Marty Mestieri. And, you know, as with anything in life, you know, you always think that the last is the best. So I think this is the best, some of the best I've produced. So hopefully you will find it interesting too. The broad issues that I've tried to understand in the last decade or so are of two types. One is about the factors that drive cross-country difference in technology diffusion. We observe all these patterns in technology diffusion that are different across countries. Why are they different? And second, what are the consequences of these differences in technology diffusion? The consequences for anything you may want. Typically I focus on things that have to do with productivity growth or more generally income growth. So this paper belongs for sure to the second part, to the second question, and it belongs a little bit also to the first question. So I will present the paper. I will present it, you know, in a very light way. It's a relatively technical paper. I will try to provide you with some of the basic intuitions and a general idea of what we are doing. I will scheme on the methodological issues. I'm super happy to take questions on them. I can go as deep as you want on them, but as a first pass I will skip most of the technical issues. And then at the end of the paper I will speculate a little bit about things that may be explaining what I find. So as I mentioned, this is a joint work with my friend, Martin Mestieri, who is a professor at the Toulouse School of Economics. So the motivation for this paper is the following. There's lots of work out there that has tried to explain why there are rich and poor countries. There are many possible explanations, and probably people have mentioned all of the possible explanations and some other which may not be possible, but despite that have been mentioned. Now, we know a lot about, or we have many hypotheses, many items in the list of things that can explain why there are rich and poor countries. But there is much, much fewer efforts to try to explain how these differences in income have emerged. In particular, efforts to account for these differences in productivity growth. Over the last, you know, 25 years, over the last 100 years between countries that have focused exclusively on the accumulation of factors of production. That means the accumulation of physical capital and the accumulation of human capital as the two more basic factors that could explain differences in the rates at which countries grow, at which rich and poor countries grow. Those efforts, what they have concluded is that the accumulation of factors of production, so that is differences in the rate at which countries invest in physical and human capital, at most can explain about 10% of the differences that we have seen in productivity growth across countries. So if you focus just on these two things, rates of investment in physical capital and human capital, at most you can explain about 10% of the differences in productivity growth that we have for factors countries. So another way to put this is the following. If you go back 200 years, rich and poor countries, they were not that different. Income differences between rich and poor countries were relatively small 200 years ago. Today they are humongous. We don't know what has happened in between to explain why the difference in income between rich and poor countries have exploded the way they have exploded. That explosion of income difference is called the great divergence. We don't know what explains this great divergence. Note that if we want to be convinced that a theory explains cross-country difference in productivity today, it should explain why this cross-country difference in productivity has increased so much. If you tell me that something explains difference in productivity between poor and rich countries today, but this explanation is not able to explain me this great divergence, then that explanation cannot be right. Because 200 years ago we didn't have those differences and now we have them. So it's not sufficient to explain me why we have the differences today. You have to explain me why they have arise these differences over the last 200 years. That's basically what we try to do in this paper. We try to explain why these differences have emerged over the last 200 years in income or in productivity. And in particular we try to explore what role technology has played. And by technology here I mean technology diffusion. So we basically are going to do the following. I will tell you in a second what we do, but let me first talk a little bit about what I mean by technology. If you look at the definition of technology, technology is the way in which we produce things. It's the techniques, the machines, the processes by which we transform inputs into outputs. Now when you take that definition and you try to think about why technology affects productivity, there are basically two dimensions of technology that matter. The first dimension is whether you have a technology or not. If you don't have a technology then it's not going to affect your productivity. People in Papua and New Guinea, they don't have internet. The internet is not going to impact their productivity. It's just going to have no effect. If they have internet then internet may affect their productivity. That's the first dimension, whether you have it or not. Has the technology arrived to your country or not? But there is a second dimension of technology which matters. Which is once you have it, how many people, how many companies are using it? Or how intensively they are using it? If people in Papua and New Guinea have internet but it's only available in one building in the whole country, then the impact on productivity probably is going to be small. If everybody has it, then the impact in productivity may be larger because the internet brings new ways to organize production, new information, new inputs. And if we all can benefit from it, then we're going to potentially have a greater impact on productivity than if very few people can benefit from it. So these two components, as I was saying, one has to do with how long it takes for technologies to arrive to countries after they have been invented, so they are option-luck. And the other component of technology which is relevant is once the technology has arrived to a country, eventually how many people are going to be using it? What's the penetration rate of the technology in the country? Okay? So what I'm going to do today is the following. I'm going to, I want to understand how technology has changed over time and how that changing technology over time may help us understand this great divergence, this movement from a situation where income was similar across countries to one where income has been very different. Okay? And if you step down for a second and you think about what do I need to do to make that journey? Okay? You will realize that I need to do three things. The first thing I need to do is to measure technology. Okay? I need to measure these two dimensions of technology. Okay? And in particular, I'm going to do that for 25 major technologies that have been invented during the last 200 years. Okay? And I need to do it for many countries because otherwise I won't be able to talk about something that happens in the world. And so I'm going to do it for a large cross-section of countries that includes in an unbalanced way 150 countries. Okay? So I'm going to be measuring for each technology and country, I'm going to be measuring how long it took for the technology to arrive to the country and eventually when it has diffused how deeply it has penetrated. Okay? So that will give me some information about the history of technology diffusion. Okay? So that will be my first step. Now the second step is consist in taking the data and studying how the process of technology diffusion has changed over time. And in particular, how it has changed over time across countries. Okay? So I want to understand whether these adoption lags and these penetration rates have diminished over time, have increased over time in rich countries relative to poor countries. So I want to understand how the diffusion process of technologies has been evolving over the last 200 years. Okay? And to do that, I need to first understand how it changes for each technology in each country. And once I've summarized this into these two parameters, I want to study how these two parameters have changed over time in different groups of countries. And that will give me an idea of the evolution of the process of technology diffusion in the world over the last 200 years. Which is pretty cool. Okay? That's step number two. And then step number three, once I understand that, I want to draw the implications that this evolution of diffusion has had for productivity dynamics. Okay? So at the end of the day, I want to explain why in 1800 we were similar in terms of productivity and today we are very different. So I want to take my information about how technology has changed over time, and I want to simulate the implications that these dynamics for technology have for the dynamics of productivity growth. The change in productivity growth over time for different countries. And I will compare these dynamics of productivity growth that I simulate with the actual ones. And so, from that exercise, one possible answer is, well, you know, what you have studied Diego so hardly is irrelevant. So, you know, we see nothing going on in productivity growth after you feed in the information that you have obtained from the diffusion patterns. That's perfectly possible. You can infer, you know, like in good movies, you know, typically they have a good ending. So, you know, I'm here because the movie has a good ending. Okay? And that's not what happens. What happens is that actually when I take the information from the dynamics of technology over the last 200 years, and I simulate what these dynamics of technology imply for the dynamics of productivity growth, that's something significant and substantial. Okay? So I can come back to this question of the great divergence, speak in the eye to the question and say, great divergence, you really are driven by technology. Okay? And so that will be my conclusion today. My conclusion is these great divergence that historians have studied at the end of the day is a great divergence in technology diffusion. Okay? And I will explain to you exactly how that is. Okay? So that's the plan for today. Okay? Now, I want to, any question at this point before I start digging? Yes? Oh, yes, of course. That's an easy one. Yes? I just want to know how you will make your equation to have which parameters you can include in your model to explain the impact of the technology on growth. That's a good question. I will, when time comes, I will try to provide you some information about that. Okay? Not an easy answer. The answer is in the paper. If you want to really understand the answer, you need to read the paper. I will not do justice to the paper, but I will try to give you some intuitions about how we, that part actually is not trivial at all. That's one of the beautiful things about this paper. Okay? Feel free to ask any question, okay? I'm used to teaching MBA students, which enjoy talking more than listening, so. I miss that when I'm in a quiet audience. So, let me give you, so again, we are going to start from technology, okay? So let me show you what's my input, what's my raw material, okay? Here, I'm plotting two cores, okay? They are both about the diffusion of computers in two countries, in the U.S. and in Vietnam, okay? And this, by the way, is typical of the type of, these shapes are typical of the diffusion patterns that I have in, that I find in the technologies in my dataset and across the different countries, okay? So typically, when you plot the log of measure of technology, which typically is either an input that embodies the technology, like computers, or an output produced with the technology, okay? And you scale it by a variable like GDP, okay? And, you know, there are many reasons why we do this. I also try other things, and the results are very robust. So unless there are some specific questions, I will not spend too much time talking about the scaling of the technology measures. So when you plot that, the shapes are very, very steep at the beginning, and then over time, you know, the slope of the curve weakly flattens, and then the curves typically converge to a linear trend, okay? So it's very much what you can see in these plots. Now, after staring to this course for many years, me and my colleague, my, you know, partner in crime on this project in the early stages at Barahobain, we realize that this course had some additional property that is very, very helpful. And is that to a first approximation, to a first pass, this course, when you compare the course, the diffusion course, of a given technology across different countries, they have the same shape. So as you can see in this plot, you can move from the US to Vietnam in terms of computer diffusion by having a horizontal shifter and a vertical shifter, okay? That means you can take the US, plug it on top of Vietnam, and you're going to get something that resembles very much to the Vietnam curve, okay? If you put the curve in the right position, once you allow for a horizontal and a vertical shift, okay? That property, it's kind of interesting, okay? Because allows you to reduce cross-country differences in technology diffusion of a given technology to just these two parameters, horizontal shifter and a vertical shifter. So, you know, as a pure data analysis technique, seems like a great breakthrough, okay? And we tested whether that was true or not in the data, statistically, and we found that it was true in a very large number of the cases, okay? So it seems like a good approximation. Now, you can see that what that implies is that cross-country difference technology diffusion for a given technology boils down to these horizontal shifters, which are basically shifts in time in the moment in which the technology starts to grow very, very fast. And then, and those moments, by the way, you know, could very naturally be interpreted as differences in adoption lags, okay? In the time that it takes for a technology to arrive to the country, okay? So given that initially this curve starts very steep and you start seeing this steep portion later on in time, the most natural interpretation is, well, you know, that's because the technology has started diffusing later in Vietnam than in the U.S., okay? So that's the interpretation of this horizontal shifter. The vertical shifter, you can see that here, the curves are asymptoting to the same horizontal trend and so this vertical shifter, what it informs you is about approximately the long-run penetration rate. So once the technology has fully diffused, once we have gone through this very concave shape and we are converging to this linear trend, how intensively are we going to be using the technology in one country relative to the other, okay? That's basically this vertical distance in the long run, okay? So that has to do with the differences in the penetration rate of the technology, okay? Once it has fully diffused. And so that's basically the graphical motivation for how we are going to extract information from this diffusion course across countries, okay? Parenthesis, of course, what we do is a bit more sophisticated than just plotting things and, you know, eyeballing two numbers, okay? It's a bit more sophisticated. We have, you know, just to give you some technical information about what we do and that, you know, it's a good, you know, links to the question that was asked before about how we draw the implications for productivity of these differences in diffusion. What we have in the model is we develop a model. What we have in the paper is a model, an economic model of diffusion that basically inform us about how differences in the lack with which technologies arrive affect the shape of the diffusion curve and the same thing for the adoption margins, okay? So we have an economic model. It's a full-blown model. So it's a general equilibrium model that captures the effects of aggregate demand on the demand for the technology, captures the effect of differing wages on the four different cost of production on the demand for the technology, captures the effects of technological change with new varieties of the technology from the shape for the diffusion of the technology, captures a wide range of mechanisms which potentially are going to impact the shape of the diffusion of the technology. And then what the model gives us is, it gives us at the micro level, it gives us this equation that contains all these other terms, but there are two important elements which are this adoption lack and this intensity of use, okay? Based on these differences in the shape of the diffusion course, we can estimate these parameters, these adoption lacks and this intensity of use for each country and each technology, okay? And then the beautiful thing about this model is that it's a general equilibrium model so we can aggregate things up and it provides us a theory for how technology impacts productivity growth, okay? So the model itself will tell us how these adoption lacks and how these penetration rates will translate into productivity growth after making some very minimal calibrations which I will discuss later on, okay? So that's the answer to the question posed before. So the microstructure that we developed to estimate the diffusion course by itself provides us with an aggregation of the impacts that technology diffusion has on productivity growth, okay? So we really need to do nothing on top of what we have done to estimate this diffusion course to draw the aggregate implications, okay? And I will be more specific about exactly what we need to calibrate in the model in a few slides. Yes, there was a question. Just to be clear, the vertical axis is the number of PCs per unit GDPs they're divided by? In logs, yeah. When you take logs, and probably lawyers do not know that, but when you take logs a division becomes a subtraction. I was just asking about whether... It's a dash, not a slash. That's my point. When you take logs, that's a math class. So it's the log of PCS per GDP. It's very good. But taking logs is kind of important because it provides you this, I mean, important just to feed the data nicely. Okay, so this is what I'm going to skip. So, what's the data? So, as it was mentioned before, we're going to use my data, okay? Which is publicly available. It's been available for seven years now, I think. And this data covers the diffusion of many technologies. Over the last 225 years, okay? In a wide range of countries. I mean, total is about 150 countries. Now we're going to focus on 25 major technologies. Why? Well, basically because these are the ones for which we have better data. The estimation relies on these nonlinearities. So if we just have information about the later part of the diffusion, we will not be able to make progress. So those technologies are not used. We want to have relatively wide coverage of countries. So for some technology which will have fewer countries, we don't use them, okay? And interestingly, we want also to have the technologies relatively, the invention dates of the technology covering relatively evenly the last 200 years because we are interested in understanding how the adoption patterns change with the invention date. So if we have big gaps, then, you know, it's harder to make this type of inference, okay? Now, the first technology that we have in sample that we are going to use are ring spindles, which was invented in 1775. And the last one is the internet, which, you know, we claim it was invented in 1983, but of course, you know, their fathers are certainly as Aristotle, okay? So this is the list of technologies, the full list of technologies, ordered by invention year, okay? So you can see that these technologies, they cover a wide range of sectors. They are, you know, as I just mentioned, they are the invention dates, you know, they are evenly spread over the last 200 years or so. And they are all relatively major technologies, okay? These are not things which are relatively minor, specific, most of them are significant technologies that have the potential to impact significantly the economy, okay? So what this table reports, I mean, this table has many numbers, okay? So it's a bit confusing, but, you know, we do that purpose just for people to invite us to give seminars. So what this table reports is, it reports information about the adoption lacks, the horizontal shifters that I've described before, okay? And there are some interesting patterns that I would like to point out. It's a pity that the clicker, I'm not sure if the red works. It works here, but it's not working on the screen, but I will try pointing with my finger. So this column means it reports the average adoption lack across countries for each technology, okay? So you can see that the numbers are large. So the average adoption lack for the full sample is 44 years, okay? So on average, technology diffuses slowly, but there is much more than that. So if you look, the most striking fact in this table is that if you look at the average adoption lack by technology, so that's the column mean, you will see that this average adoption lack declines dramatically in a relatively linear way, okay? So all technologies diffuse very slowly. It would take a long time before I arrive to countries. And slowly, but surely they have diffused faster and faster, and so now they arrive to countries on average much quicker than they used to, okay? That's a very interesting fact. This is the first time, actually, previous paper that we wrote in 2010, we published in 2010, is the first time that that was documented, okay? Note that this is not just about the internet, cell phones and PCs. I mean, this is something that has gone on for 200 years, okay? People didn't notice, but that's why you need this data set to understand those patterns. But you can see that this is something linear. I mean, the first technologies, on average, it would take 100 years to arrive to places, then it would take only 50, then it would take only 25, and now it takes less than 10, okay? This has gone on for two centuries. Now, another thing which is kind of interesting is that and I will show you another plot that represents that more nicely, but if you look at the last column, that's the interquartile range, okay? So this is the distance in adoption lags between the countries that are in the 25th percentile of adoption and the country that is 75th percentile of adoption, okay? So these are countries that are relatively fast adopters versus countries that are relatively slow adopters. So when you look at the distance, that gives you a sense of how different are the adoption lags across countries for any given technology, okay? So on average, you know, the interquartile range is 46 years, so that's, you know, pretty large difference in adoption lags across countries. Interesting fact is that this interquartile range also has been diminishing significantly over the last 200 years and here again, the reduction has been very, very dramatic. Initially, it was close to 100 years, the interquartile range. Quickly, drop, you know, 50, 25, 10, and now it's, you know, three years for the latest technologies. So cross-country difference in adoption lags have narrow, narrow, narrow in a way that we haven't seen for almost anything in the world other than, for example, you know, things like, you know, life expectancy or infant mortality, okay? Probably even more than that, I mean, because you can see that now differences are like minimal, three years. What it takes for internet to arrive to the country in the 25th percentile versus the country in the 75th percentile. There's nothing that you have three years of infant mortality or three years of life expectancy difference between these two types of countries, okay? So very, very striking fact. This is a graphical representation of that, okay? So here I'm plotting the distribution for each technology. Each technology is plotted in the year in which it was invented and for each technology I'm plotting the mean adoption lag distribution of the adoption lags between the 95 percent confidence interval, okay? So the countries that are in the 2.5 percentile, the countries that are in the 97.5 percentile, okay? So this is like the extremes of the distribution. You can see how the distribution of adoption lags have been narrowing, narrowing, narrowing, and at the end of the sample it's like almost none, okay? There's been dramatic convergence in adoption lags, okay? So that motivates the title of the paper. You know, technologies have arrived everywhere. It will take forever for them to arrive to certain countries. Now they arrive everywhere very quickly, okay? Any questions? Yes. I haven't said anything about the adoption margin, the intensive margin, the penetration rates, nothing yet. This is just the adoption lags, okay? But I'm going to say right now. So what has happened to the intensive margin? Well, the way we measure the intensive margin is... So the intensive margin, you know, each technology is measured in different units. So the way to measure it that makes sense is to measure it relative to some benchmark of...to some country benchmark for that same technology, okay? So one approach would be, you know, I want to measure the intensive margin relative to the US, okay? For each technology, okay? That would be one possible way of doing it. We don't do it that way. We do it relative to something a bit more cumbersome, but which is going to be more helpful for the empirical analysis that we do later on. And the benchmark is the average of the group of rich countries that Madison call Western countries. So these are basically the US, Canada, Australia, I guess, Japan, and Europe. So some countries in Western Europe. So it's like 17 rich countries, okay? Why we do it that way? Well, we do it that way because we're going to later look at differences of productivity growth in the Western countries versus the rest, okay? That's going to be like the baseline analysis we're going to conduct, okay? So now, by defining now Western versus non-Western, we're already setting everything in terms of what we're going to do later on, okay? The benchmark is not very important because the important thing is the dispersion, okay? Dispersion between, in the intensity of use between rich countries and poor countries, okay? Now, because of the way these things enter into the production function, it's also helpful to measure it in logs. So this is why you have here negative numbers, okay? The logarithm of a number which is more than one, as you know, is negative, okay? So a negative number here means that there is a percentage gap with respect to the Western countries, okay? So here, the important facts are the following. When you look at, for example, the gap between the 10 percentile and the 19th percentile, if you look at the bottom row, which is represent all technologies, you see that the gap between the 10 to 90 gap in logarithms is 1.9, which when you take the exponential, this is about the factor of 6, okay? And here, the intensive margin goes one for one with TFP. So we're talking that differences in the intensive margin can give us potentially factor of 6 differences in productivity across countries. So this is big potatoes, okay? This is a lot. Difference in productivity levels across countries. Right now, they are a factor of 50, 40, factor of 6, something significant, okay? Yes? Which production function are you applying? No, I haven't addressed that. I've been hiding that under the rock explicitly. So the production function, it's something that comes out from the model, okay? It's a very sophisticated underlying model, but the other production function is very simple. It's just a coped Douglas production function. It's a coped Douglas production function. It's a coped Douglas, but where the TFP level is endogenous, okay? So that's what makes it nicer. And it depends on the adoption lags, and it depends on how intensively technologies are used in the country. But the underlying function is just a coped Douglas production function. It's actually a coped Douglas. We don't even have capital. We have, like, you know, a version of coped Douglas where we have, you know, intermediate goods rather than capital. But yeah, okay? I would know I would be happy to answer more questions about that. Can I ask just for my understanding and maybe I'm just a bit slow, are you comparing here penetration rates at any given point in time? No, this is, okay. No, no, no, no, no. That's a good answer. I want to clarify. So what you can see from the plot is that you could compare the penetration rate at each point in time, okay? But that's not very relevant, because the distance between these two lines, it's endogenous to the adoption lag, endogenous to the factors that generate the shape. So the way to do this is by looking at the long-run penetration rates. So once a technology fully diffuses in the limit, what would be the distance that I would see between these two countries, okay? And the beautiful thing about my approach, I mean, you are going to ask me, how can you do that if you haven't seen the full diffusion curve in some countries? The beautiful thing about this approach is that we don't need to see everything because I have this model that fits well the data that tells me how these curves are going to evolve in the future, okay? So to the extent that my model fits well that part of the data, which it does, it does beautifully, I can measure these penetration rates quite accurately, okay? But that's a good question. So basically, you can think of this diffusion curve as having some factors that are going to affect the general shape and then two parameters that allow me to explain the cross-country differences. So it's just two parameters. They are fixed over time for a given technology, okay? Yes. What you're saying is we're trying to predict income or productivity differences by differences in the diffusion patterns, but that's completely in dodgers, right? It can be like you have a big simultaneous bias with anything that has related to openness to trace, structural changes. I mean, that affects simultaneously both the penetration rates, the adoption lives, as well as productivity differences. So how do you control for these limited variable biases? Yeah, I'm not trying to control for them. I'm trying to do something, I think, more sophisticated than that. So what I'm trying to do is I'm just trying to measure these two parameters, okay? I'm not taking a stand of what they are. I'm just measuring them. This is how all these factors that you have in your mind, other than income, okay? I'm controlling very well for variation in the slope of the angle course. I'm controlling very well for effect of demand on technology, okay? But you have all these factors that you have in your mind, we, you know, whatever. It doesn't matter. All those factors are going to affect technology by shifting these two parameters. I'm not trying to explain what's driving this variation. I'm just trying to measure it. No, no, no, no. I haven't explained to you what I'm going to do later. I haven't explained to you what I'm going to do later. Now, if you think, what you're thinking is because you don't know what I'm going to do later. I'm going to explain you what I'm going to do. So, just bear with me for one second and then I will tell you what I'm going to do, okay? More questions? So, there is this big variation in the intensity of use of technologies, okay? Now, how has it changed over time? Well, it's hard to see here because, you know, it's too many numbers. So this is why I plotted for you. Again, I have for each technology the year in which it was invented and then in that year, I plotted distribution of the intensity of uses, okay? So, on top, basically you have the top bar which is very close to zero, which is the normalization, okay? It makes sense, you know. Most countries are out there, so, you know, they are close to zero. That's fine. That's not very important. The bottom is the intensity of use of the countries that use less intensively the technology, okay? This is like the 2.5th bottom percentile. So, what you can see here is that initially, this difference in the intensity of use, and this is in logarithms, okay? So this is the log difference. Initially, the intensity of use, the difference across countries in the intensity of use was very small, or relatively small. And then over time, as new technologies arrive, the differences for the new technologies became bigger and bigger, okay? And so, when you... this line, this diagonal line that you can see there, the dash line, what represents is the regression line that you get by fitting the evolution of the intensive margin for the non-Western countries against their invention date, okay? So you can see how, if you assume that this line is linear, that this ship is linear, there's going to be a very systematic, very strong divergence in the intensity with which new technologies are used once they arrive to countries, okay? Now, you know, we could discuss a little bit whether this is linear or is non-linear, and I don't have any good answers. So, you know, the paper, we assume that it's linear, then we say, okay, what if it's non-linear? You know, what if it changes dramatically in 1880? And the results are very similar, so it's not very important. But you can see, I mean, the fact that stands out is that while for the adoption lags, there is dramatic convergence for the intensity of use of technologies, you know, they have diverged dramatically over the last 200 years, okay? So those are the two facts, the two takeaways from my first exercise, actually from my first two exercises, okay? And now I want to go to a third exercise, which is what was asked about, okay? Before going there, I want to do something preliminary, okay? And that will answer the question a little bit about the production function, okay? So, sorry to show you one equation, okay? But this is a simple one. So, in this model, when you are, you know, when the economy is what is called the state, the balanced growth path, the difference in productivity between the Western and the non-Western countries, the ratio of productivity, it's given by this expression, okay? And so, let me digest this for you. So basically, what this expression says is that the log gap in income between Western and non-Western countries is the sum of a term that has to do with the differences in the adoption lags, okay? So, if adoption lags are shorter in the West than in the non-West, then the West will be richer because it will incorporate faster, newer technologies, okay? And the gap in the intensity of use of technology. So, if technologies are used more intensively in the West, that means that income per capita will also be higher in the West, okay? So, these are the two factors that matter to explain cross-country difference in productivity in this model. Now, what we are going to do now and what we are going to do in the rest of the paper is to compute we are not going to regress, we are going to compute, okay? And this basically is the answer to the question that you had. Compute the implied productivity differences that arise from the patterns of adoption that we have estimated, okay? So, the first question that you can ask is, okay, so what difference in income should we have seen in 1820, you know, when the sample starts, okay? Given the difference that we observe in adoption lags and the difference that we observe in the intensity of use of technologies. To answer this question, I just need to calibrate two parameters. One is the growth rate of what we call in economics the wall technology frontier, the growth rate of technology for the wall before 1800, okay? And that's something easy to do, you know, you take Madison data, you look at what rate productivity growth in the wall grows in his data set, it grows about 0.2% per year, so I just plug this number, okay? Completely uncontroversial. The second parameter I have to calibrate is the capital share, which, you know, we all in economics know that it's one-third, okay? And so, you know, completely uncontroversial. So you just plug in these two numbers, you take the estimates that we have for the adoption lags and the intensity of use in 1820, okay? So the numbers that I just showed you that correspond to the feeder line in 1820, and you compute this ratio. You compute this ratio. What do you get? Well, what you get is that the simulation, the simulation predicts that income difference between Western and non-Western countries in 1820, they should have been of a factor of 90%. So Western countries should have been, based on their differences in technology, approximately 90% of the average non-Western countries, okay? What do we see in the data? Well, when you go to Madison and you look at the difference in productivity he has a 90% difference. There is nothing in what I've done that implies that the numbers are going to be similar, not just the same, just similar. Nothing. Absolutely nothing. This could have been zero and 200%. This could have failed in 200 different ways because I've taken some data that my arrays have put together for 10 years about the microdivision patterns of technologies in 200 countries. I've estimated this model to extract these two parameters and then I've fitted some lines to the evolution of those parameters across different country samples. And then I've plugged the estimates for 1820 into this simple equation in which I'm not getting anything. Technology could have been absolutely irrelevant for that matter. It could have gone the other way perfectly and it's working fine. So this is the first surprise to me. Here there is no magic. My hands are clean. Anyway, but this is just the beginning. This just shows you that the model is consistent with the fact that, not the model, the data. The data on technology is consistent with the fact that in 1800 income differences were not very large across countries. That's just the beginning. Now I want to go one step further. I want to now simulate, take the simulations of how technology has evolved, take the data of how technology has evolved and plug that into this production function. Which basically, again, the only thing I'm going to do is I'm going to simulate the evolution of these margins. So now I need to calibrate two more parameters. Well, one which actually I calibrated already, which is the date in which the industrial revolution starts. Well, I haven't calibrated. So industrial revolution starts in 1765 when Watson bends the steam engine. That's the first parameter. The second parameter is the growth rate of productivity in the world after the industrial revolution. And all economists know that this number is 2%. So again, it's a number which is absolutely in controversy. And then I just need to press the return key and simulate income growth in Western countries and non-Western countries. And this is what comes out. So in the blue line, in the top chart, you have the evolution of productivity growth in Western countries over the last 200 years. And in the green line, you have the evolution of non-Western countries. So this is, again, productivity growth. This is the growth rate. So before the industrial revolution, they were growing at the same rate, which was very low. It was 0.2% per year. Then the industrial revolution comes in 1765. It takes a while for technologies to arrive to most countries. So they don't start growing until 40, 50 years later. Western countries take off slowly as new technologies are slowly implemented. They initially represent a small fraction of all the technology. So though they are more productive and they arrive at a faster rate, they still represent a small chunk of the whole pie of technologies. So their impact on productivity is gradual. Productivity growth starts growing gradually. But there starts to be a gap with the poor countries where the new technologies have not even arrived yet. So you start having this acceleration of productivity growth in the rich countries. Before nothing happens in the poor countries, that creates some income divergence. The peak of productivity growth in the rich countries relative to the poor countries takes place around 1900. That's the point where rich countries are already growing at 2% per year, actually a bit more than 2% per year. Poor countries are starting to grow. That's the point where there is maximum gap in productivity growth. And after that, income growth is relatively constant in rich countries. In poor countries, it continues to grow until reaching a plateau of about 1.5% or so in the second half of the 20th century. Now note that despite the convergence in adoption lags, productivity growth is always higher in rich countries than in poor countries. And this is because of the divergence in the intensity of use of technologies. New technologies, when they arrive to poor countries, they arrive very, very fast, but they are going to have a smaller impact on productivity and gradually even smaller because they are going to be used less and less relative to the old technologies. That explains why productivity growth in the poor countries doesn't reach 2%. By year 2000, it will reach it eventually if we fix the divergence. Now, quantitatively, how does the model do? Well, here I have the growth rate of productivity for the data, which is the right column, the second column, and from the simulations. And you can see that the model does pretty well. It captures the fact that on average rich countries grow faster than poor countries throughout the sample, and approximately the magnitudes are correct. The difference implied by the model in the gap, the productivity gap between rich and poor countries is that that gap increased by a factor of 3.2. So basically, if the initial difference had been 1, now it would be 3.2. Of course, the initial difference was not 1, it was 1.9, so you have to multiply that. In the data, the number is not 3.2, it's 3.9. So basically, we are capturing that technology dynamics are generating 80% of the differences that we observe due to the great divergence. You can slice that in sub-samples and the model does great. This is a very ambitious exercise. So rather than looking at Western versus non-Western countries, we look at all the distribution of income. So we split countries by income quintiles, in each of three different points in time, 1820, 1930 and 2000. The blue line is the simulation, the green line is the data. So you can see that in the initial period, as I suggested, the model does a decent job in capturing not only the difference between Western and non-Western countries in productivity, but also the differences for each of the different income quintiles. Now, this is what you get in 1913. So the model and the data are one on top of the other, basically, and this is what you get in year 2000. So basically, this is not just rich versus poor, it's the different income quintiles at each point in time. You can do the same thing by continents and the model does a good job, the simulation does a good job in matching the data. Again, in all of these exercises, I'm not changing any parameter, I'm just feeding in the trends in technology diffusion that I've estimated for each of the different groups. The structural parameters, the few parameters that I have, they are the same, nothing changes. Okay. So, almost out of time? We're fine. We're fine? Are you sure? We still have 25 minutes. Oh, it's not too. Oh, sorry. I need a Swiss watch. So, any question until now, before I start my conjecture section? Yes. Why you don't change the parameters? Can you use? I'd like to know why we cannot change the parameters. Well, so, first, let me be clear about what changes and what doesn't change. So, I don't change. So, this model is so parsimonious that it really doesn't make much sense to change any of the parameters because let me tell you which parameters I've calibrated. The date of the industrial revolution, okay? The industrial revolution, you know, is just one, okay? So, the date of the industrial revolution is not going to change when I look at different countries. Okay? It's the industrial revolution that took place in England in the 18th century, okay? So, that doesn't change throughout the simulation. The rate at which the wall technology grows before the industrial revolution, okay? It's just a wall-wide parameter, so it doesn't make sense to change it. The rate at which the wall technology frontier grows after the industrial revolution, again, there's one parameter, so it doesn't make much sense to change it. And then the only parameter that you could convince me to change is the capital share, okay? Now, when people have looked at the capital share, they have seen that it's not that it's constant across countries, but there is no systematic relationship between the capital share and how rich a country is, okay? So, to a first approximation, you know, the most natural thing is to assume that it's constant at the average level, which is one-third, and, you know, tie your hands in that way, because, you know, that's not going to be a productive margin anyway. So, those are the parameters which are constant. Now, what does change? Well, what changes is the technology dynamics that I've estimated for each country, or for each group of countries. So, those change. So, I basically, you know, when I look here, for example, so here I look at... within the non-Western countries, I look at very poor countries, or super poor countries, okay? I look at countries in the bottom 25th percent of the income distribution, and countries in the bottom 10% of the income distribution, okay? So, when I... what I do is basically I order countries by income level, and I look at the countries in the bottom 25th percent of the distribution, and I feed these lines, these regression lines to measure the degree of convergence or divergence in the intensive margin in the adoption lags, okay? And I take those parameters. So, the only thing that changes are the technology parameters for each income group, okay? Which come from my estimates. So, those change for sure. That's basically what I want to change. The rest, you know, the rest is not, yes. Did I hear you say the change in productivity growth is because of the intensity of adoption? And if so, is that what you really mean to say? Yes, yes. So, what I said, and I think what I meant to say, is that the factor that has contributed to the divergence in productivity across countries, or in other words, that had generated these differences in the rates of productivity growth for poor countries versus rich countries, is the evolution of the intensive margin, okay? Let me show you why. So, again, I have two margins of technology diffusion. One is the adoption lags, which have converged, and the other is the intensive margin, which has diverged. Now, something that you can do is you can take those two margins and plug them separately. So, so far I've plugged them at the same time. I said, look, you know, what happens if I change them? Now you can say, okay, suppose that only one of them changed. Suppose that only the adoption lags changed. How good the income dynamics change with respect to the ones that we have seen or with respect to the ones that I've seen when I plug both of them? This is what you would get. So, here, the income dynamics for rich countries, they are the same, okay? That's not very interesting because I've normalized the intensive margin to be constant for rich countries all the time. So, you know, obviously you get the same number. But look at what happens for poor countries. For poor countries, as the green line, initially you have some divergence, okay, as we had before. But very quickly, because the only thing that goes on is that there is convergence in adoption lags, what happens is that it's still getting faster and faster in the newer technologies, they are adopted at the same level as the old ones, which were not very different from the rich countries. And so they are going to... the growth rate is going to, by 1900, is going to catch up with the rich countries and then is going to overshoot the rich countries and eventually will converge back to the rich countries. So the picture would have been very different. This is what you... If I tell you that the only thing that has happened in the world is that technologies now arrive faster and faster to poor countries, this is what you should expect. You should expect dramatic convergence in income, okay? This is what motivates the title of the paper, okay? The bottom plot, you see the difference in growth between the rich and the poor countries, implied by this... by the top figure, you see that the difference increases in the 19th century, less so than what we observe, but then it diminishes the growth difference... the gap in the growth rate diminishes in the 20th century, okay? So this is what happens if you just had the convergence in adoption lags. Now you can do the same thing if you just have the divergence in the intensive margin, and this is what happens. So if adoption lags remain constant at the initial level, and the only thing that happens is that the intensive margin diverges, then you're going to see that the difference in productivity growth between rich and poor countries becomes wider and much more protracted. So this would lead to humongous divergence. The great divergence is nothing as compared to this, okay? So really what has generated, given that technology explains the evolution of technology, the differential evolution of technology diffusion in rich and poor countries, explains 80% of the absurd differences in productivity growth. The margin that is driving, especially in the 20th century, is the intensive margin, okay? The fact that technologies have penetrated more and more to reach in the rich countries than in poor countries. And then you can ask me the question, why is that the case? And then you can try to think that this is due to trade or to institutions or to other things, okay? That's a separate question. That's the question which I think is interesting afterwards. What has generated this divergence in the intensive margin? Of course technology is something endogenous. Of course it is. I mean, I spend my life in technology because I was interested in endogenous technological change theories. So that means that technology is endogenous. The result of other things. Now the question is what, and in particular, what has been able to generate such a dramatic divergence in the intensity with which technologies are used once they arrive to countries? That's the question. So I've narrowed the question to something relatively concrete. And I think that's a very interesting step because of all the things that have been proposed to explain difference in productivity across countries, they are very, very few which have diverged so strikingly over the last 200 years. Okay? The world probably is more globalized now than ever. So openness to trade is not going to do it. Okay? There's probably been dramatic convergence in the quality of institutions. If you look at polity, definitely in the last 50 years, very surely in the last 100 years for sure. Which is the period over which we have seen this dramatic divergence in the intensive margin. There are some fixed factors like climate or geography. I mean those have not diverged or converged but that will not help us either because we need something that diverges. Okay? So what can have generated this divergence in the intensive margin? I mean that's not an easy question. I'm going to throw in a candidate, okay? Maybe wrong. This is just speculation. So this is not science. This is just, you know, what goes after science. I think that one thing that has diverged is technological knowledge. So what is technological knowledge? It's knowledge about the technology and about how to use it once you have it. Okay? I mean that's very hard to measure so this is why I'm conjecturing. But the reason why I think it has diverged is the following. There is complementarity between knowledge and technology adoption. How do you get technological knowledge? Well, you get technological knowledge by using the technology. So I've put together other data sets about technology going farther and farther in time just to amuse myself. And so I have one that, for example, measures technology adoption in different sectors in 1500, okay? And what technologies you have adopted in 1500 predicts very well the dates of creation of, for example, engineering schools in countries. Okay? Predicts it very, very well. Okay? Why? Well, because, you know, you had some advanced technologies in 1500 that the use and development of those technologies creates some knowledge that you want to, people want to have and to have it to institutionalize that knowledge you create a school, an engineering school. The same thing happens, for example, for other forms of schools. Okay? You know, sailing schools. You have some knowledge about, you know, naval technologies. You want to transmit it. You create a school. So, you know, that's technology leads to knowledge. I mean, I'm not meaning that this can't just be measured by schooling. I mean, this often is tacit. It's not something that can be measured by the creation of certain institutions, but often it's embodied in companies rather than in people, okay? So it's even more subtle than what I just described. But, you know, that knowledge, so knowledge comes from technology, but that knowledge is actually what makes you more prone to adopt new technologies. I mean, in all my work, you know, I've always found that how much you know about certain technology or certain sector is one of the best predictors about how likely you are to adopt subsequent technologies in that sector. Okay? And so that creates a vicious circle where if you don't know anything about, if you haven't used that technology because, you know, it's new to you, then you will not know anything about that sector, about that technology. And when a better technology comes down the road, then you will not be able to adopt it or to use it or if somebody brings it to your country, you will not know what to do with it. And so knowledge leads to technology and technology leads to knowledge. And so, you know, it's very difficult to get away from that circle, okay? And so the narrative that emerges from that logic is the following. You have the industrial revolution. The industrial revolution brings new opportunities, new technological opportunities, okay? Now, in countries that were familiar with the technology that came before the industrial revolution, then, you know, they were ready to use them. So they were ready to use them. They started adopting them and they adopted them earlier and they adopted them a bit more intensively than in countries that were not so familiar with the industrial revolution, with the technology that preceded the industrial revolution. But those advantages in the adoption lacks an intensive margin led to significant knowledge being created, okay? And that knowledge made these countries more prone to adopt faster and especially more intensively the newer technologies, okay? And that basically has gone on creating these divergence, okay? Now, all that has occurred despite the converges in adoption lacks, okay? So this rationale doesn't explain why adoption lacks have converged. But they could have converged for many reasons. I mean, now multinationals are everywhere. So every time, you know, multinational goes somewhere, they bring their technology and so the technology is already present in the country, you know, starts diffusing slowly. Or, you know, now there is always, in each country, there is some smart guy that has access to the internet, you know, when computers, when new computers arrive or whatever, you know, they can now order it from other places. Or they have, you know, there is more inequality. So now, you know, rich people come by, whatever fancy thing comes out before they were doing that in the past, okay? So there is a variety of reasons why the adoption lacks could have converged. I think the challenging element, the challenging finding to explain in these, in these, from this research is the intensive margin. And I think that my hypothesis is that it's technological knowledge. But, you know, it's just conjecture, yes? And then other statistics about how that technology is being and obviously for productivity, you know, the third element matters most, right? For example, the internet might have relatively easy and wide penetration. But, you know, not every type of internet use any of these to productivity increases on the same level as, for example, use of railroads or so. You know, very different technologies there. For example, railroads can measure the coverage of the country, the density and the number of populations. So I wonder whether there is a linkage here because arguably, I guess, it's likely, technically harder to develop an infrastructure such as railroads at the time than maybe coverage of internet. But then, in an internet, the type of internet use matters a lot for productivity, so how do you... That's an interesting point. So I think there are two questions in one, okay? One is about the measurement and the other is about whether there is any pattern in the measurement that is correlated with the invention date and that could deliver the finding that the intensive margin has diverged. So, just simplifying, I think the answer is just for the first one. I don't think so for the second one, okay? So just for the first one. I think you're right that for some technologies, you know, we do a better job capturing really the intensity with which the technology is used and therefore the impact on productivity, okay? So we typically do that better, I guess, when we are measuring an output rather than an input, okay? So for example, you know, for railways we have a ton of kilometers transported in railways, okay? So, you know, that's almost, I mean, some additional component that you could have to add, but that's basically what you would like to have to understand the impact on productivity, okay? For inputs, you know, we have, for example, you know, the number of ring spindles, okay? So, you know, then you can have some variation in how much output you produce with the ring spindles and so on. Now, so I understand that that's possible. Now, I don't think there is any correlation or any significant correlation between whether we have inputs or outputs and the invention data of the technologies. I mean, I just gave you two early technologies, one measuring input, the other measuring output, okay? When you look at newer technologies, sure, I mean, you have the internet, okay, but then you have things like, you know, a number of kidney transplants, okay, which is a measure of output. So, I mean, I don't think there is any significant correlation that can explain that. There are other things that you could think of, like, for example, you know, somebody suggested a measure of the complexity of the technology, which is something that you suggested. So, I haven't tried to correlate that formally with the measure of technology that we have, the time, the measure of the time with the complexity, because it's harder to measure the complexity of the technology. But, you know, we have some early technologies which are complex and some late technology which are complex and some early technologies which are simple and others that are simple later on. So, you know, we have, for example, you know, electric affordances which are relatively complex in the later period. And we have, again, simple technologies like, you know, like spindles or like, there's a few early on that are not that complicated. You know, cell phones, for example, you know, they require, people do not realize that, but they require significant infrastructure too. For some other papers, we are collecting information on the cost of setting up networks and they are quite significant. So, I mean, I think if you think, there are like two types of explanations about that you could use to explain this evolution of the intensity of use. Some have to do with features of the technologies that have changed over time in our sample or in the world. Nothing else underlying the trend, just the technology itself, the technology has changed. The others are the technologies hasn't changed, but there are some external factors that has changed or endogenous. I mean, knowledge is something endogenous. It's an external factor to the data that we have, but it's internal to technology. I'm not, I mean, I don't feel very strongly at this point about which one has more chances to succeed. I know, this is what I know. I know that many of the factors that people have proposed to explain the cross-country difference in productivity today, those have converged, okay? So, they would have a very, very hard time explaining why there were no differences in productivity in the 1800s, because the arguments would go out for theory. If institutions have converged, that means that income difference institutions should have generated 200 years ago. They would have been much larger than the ones that they generate today, and they don't. I mean, we didn't see that. So, yes. We have three minutes. Let me maybe ask two questions, and since we're short on time, maybe you can, you know, provide short answers. One is, you know, I think one, at least to me, obvious policy conclusion that comes out of that, and, you know, well-noting that, I think, you know, you stated that more as a hypothesis than defining, you know, your point about technological knowledge. You know, is, you know, fundamentally investments in human capital. How do you, I mean, first of all, would you agree with that? And secondly, you know, how do you contrast that? I mean, that was sort of tried before by the World Bank and others who in the 1960s and 70s, you know, invested heavily in the education systems and developing countries, and I think, you know, many people have sort of characterised that episode as a, you know, rather disappointing episode. The second question, you know, is essentially something you alluded to on the question of why do we learn anything by comparing different technologies in your data set. So you, for example, mentioned cell phones, I don't know the precise data, but it strikes me that cell phone is a technology that, you know, where you do see high penetration rates even in the poorest countries, and maybe that confirms your story, that just, well, maybe yes, there are some initial, you know, costs and knowledge of setting up the networks, but as far as the use of the cell phones it's not like building a nuclear power point. But my question is, I mean, could we learn something by comparing, let's say the various technologies in your data set that are still relatively recent, but, you know, I mean, from the one graph that I've seen, you know, one did notice quite significant differences in penetration rates, so these are my two questions. So, the first one I think is very interesting. So, I mean, so suppose that you take seriously my hypothesis, okay, which I don't take it super seriously yet. The question is what is this knowledge, where it sits, okay? And so there are various hypotheses. One is that knowledge is not in people. It's, you know, for example, embodied in companies. Okay, so, you know... Sorry, sorry, sorry. That's a kind of a company. No, I mean, seriously, like, how much can an engineer from Microsoft do, or from Apple do, from Google do? Well, probably not that much, but once they, you put it together with all the people that are there and have been there in the past, then they are sitting on an amazing body of knowledge. Okay, so if knowledge is embodied in companies, not in people, then, you know, education, per se, is not going to help you so much. You know, I alluded to technological knowledge. I mean, I don't know whether that's something that you learn in a school or not. I mean, maybe what you learn in a school is something that makes you a bit more ready to start acquiring technological knowledge once you work in a company that is familiar with technology. So if you don't have those companies, you don't have technology to learn from technology, then your technological knowledge, per se, is not going to be that helpful. So I don't know how technological knowledge is accumulated. I mean, my sense is, by using technology, maybe you can teach people how to learn technological knowledge. So that's my question to the first, my answer to the first question. The second question, I think that the only way to make progress with this data is to try to look at... I mean, one way is to slice technologies in different ways, and then look at features of the technologies that correlate with your hypothesis and see whether that has any chance to explain these patterns. So yes, I mean, I think that's interesting. I mean, cell phones, I don't want to make a big deal with cell phones. I mean, the beautiful thing I'll hide in 25 technologies, I don't have to rely on any single one of them, including cell phones, to make any case. But in my estimates, I don't see such a, you know, beautiful story with cell phones, because, you know, when I look at cell phones, cell phones, I think, is the last technology or the purest last technology. I still see a significant divergence. I think it's 24. 24, I mean, it's not as bad as 22, which is, you know, hard surgery, okay? But, you know, it's not like, you know, differences are minimal, okay? So... And, you know, we have extended these to data using until 2010, okay? And the numbers were very similar, so I mean... But again, I don't have to rely on any technology. That's the beautiful thing about having comprehensive or relatively comprehensive data set, which is that this is not about cell phones where, you know, spindles. It's about 25 technologies that cover a big chunk of our history of technology diffusion over the last 200 years.