 So, good morning to everyone. Welcome to this first session on innovation, investment, and productivity. You don't have to vote for the posters now. You can wait until the end of the session. And I see very much our contribution here on this session as setting the stage in a sense for everything that will follow. We want to understand better the long-term drivers of prosperity in advanced economies across production factors. So we'll do it in two steps. We'll first have a look at employment with David's paper. And then we'll have a look into investment and capital accumulation. And that's with Thomas' paper. And we want to have a thorough discussion on facts. So that any further policy discussion will be informed by facts. So that's really the way I see the contribution of this panel. One important feature of both papers, actually, is that both have an important cross-sectional dimension which will allow us, hopefully, to also have a discussion on the distributional implications of productivity growth across sectors. So that's what's in the papers, but also possibly across individuals and across space, which will maybe also shed light on the political dynamics which were hinted at by Ben Bernanke in his speech yesterday night. So as I said, we have two papers to start the discussion. So the first paper is by David Otter, who's one of the leading labor economists, four professors of economics at the MIT, associate head of the MIT Department of Economics. David will be commented upon by Dietmar Harhof, who is a director at the Max Planck Institute for Innovation and Competition, and a honorary professor for entrepreneurship and innovation at the Ludwigs Maximilian Universität in Munich. Am I right? Yes. Okay. And also chairing the commission of experts for research and innovation for the German federal government. And the second paper will be presented by Thomas Philippon, Thomas is a professor of finance at New York University. He's also serving on the Monetary Policy Advisory Panel at the Fed New York. And if I'm right, you're also a member of the French Prudential and Resolution Authority, which we're not discussing this morning. And Thomas will be discussed by Janice Eberle, who's been the James and Helen Russell distinguished professor of finance at the Kellogg School of Management at Northwestern University, and also served as assistant secretary to the Treasury and chief economist for the Department of the Treasury from 11 to 13. So a great set of presenters and discussants, and we start with David. David, you have 25 minutes. Thank you so much. Do I go to the podium? I hope. Oh, sure. Why not? Great. Thank you. Good morning. It is a delight to be here. Thanks to the European Central Bank. Thank you to President Draghi, to Ben Bernanke. This is, it's an honor to be of this, contributed, I'm looking forward to all the discussions. This is joint work with Anna Salamons of Utrecht in the Netherlands. And the question of our paper is, does productivity growth threaten employment or the shorthand term we're using for this is the robotic apocalypse or if you like the roboclips? And so a subtene of our paper is roboclips now. There is a longstanding concern that automation threatens employment. We can look back a couple hundred years. Of course, you're all familiar with the leadites, skilled weavers in the 19th century who rose up against the power frame, potentially threatening their artisanal jobs, correctly in fact. But you don't have to go very far back in history to see examples. U.S. Secretary of Labor James Davis in 1927 warned about the scrappage of men and was concerned that labor needs a new outlet. In 1964, President Johnson formed a blue ribbon commission on technology automation and economic progress. The productivity, they were dealing with a productivity problem at the time. The productivity problem, as you all know, was that productivity was rising too fast and they were concerned that there wouldn't be enough demand. So employment would fall. In 1982, Nobel laureate Vasily Leontief warned that the role of workers will inevitably diminish people will be akin to horses put out to pasture. And of course, there's no question right now that we are in an era of similar concern. An era of people asking, is the roboclips upon us? From the perspective of citizens and policymakers and intellectuals, this concern seems immediate. The more work is done by machines, the less work done by people potentially. And we even have the legendary fable of John Henry, the steel-driving man who faced off against the steam-powered hammer in laying railroad ties, and you'll remember he beat the steam-powered hammer, but then he laid down his hammer and he died. Professional economic opinion has always countered with three different arguments. One is that demand might be quite elastic. You might see employment growth in advancing sectors. So this figure from James Besson shows you the trajectory of employment in textiles, cottons, and fibers, and also primary iron and steel over more than a century. And you see in the initial phases of rising productivity, there's enormous consumer surges in demand, unmet need. Employment rises incredibly quickly with innovation. But then eventually you hit a peak where demand is somewhat saturated and further increases in labor productivity eventually cause these sectors to diminish. So one possibility is that demand will respond directly. Another, of course, going back to Colin Kark, is that rising consumer incomes will create new demands, will move the locus of consumption from goods to services, for example. Finally, going back to Bowenville in 1967, we may have sectoral reallocation. Advancing sectors, sectors that have rapid rise in productivity, may see falls in employment, but then we may see compensatory effects elsewhere. I would say that economists are losing confidence in these long-held theories. More and more we see economists saying this time could be different. Sort of exhibit the first horsemen of the robocalypse, I would say, is that a labor share of national income is falling across a large number of countries. Perhaps this is the first evidence that machines are in fact wholesale eliminating the value of labor in the economy. But you don't have to look at the aggregate data. We know all around us that we are in the age of brilliant machines, to use the words of Brynjolfsson and McAfee. Computers are managing financial portfolios. They're beating go players, thank you, Hal. The websites and drones are eliminating sales workers and warehouse workers, and robots are leaving the assembly lines, perhaps coming for our jobs. And theoretically, an emerging understanding makes clear this can happen in canonical models, technological, capital advancement, technological changes, always labor augmenting. And canonical models are always full employment models. So only good things can happen when productivity rises. But we now recognize machines can directly replace specific job tasks. They can substitute for some workers, potentially complement each other, complement others. And there are several recent sort of theories of labor immiseration. One has to do with market failures, with the failures of intergenerational investment, like work by Saxon, Kotlakov, or Bergadol. Another is what we call the no place to hide models, where there's just more and more task encroachment. And eventually, the last worker is highly productive until he's out of the job as well. The third, very sophisticated paper by Asimoglu and Restrepo, looks at two contravailing forces. On the one, you have machines successfully replacing codified tasks. On the other, new tasks are created as labor becomes abundant. Those two might cancel each other out. There might be a balanced growth path, or there might not. We can't be sure. The evidence does not yet strongly support the labor immiseration view, but it's early days. There's a vast literature that makes clear that computerization has been skill biased, but that's really not about the overall employment impacts of technological change. There is some recent evidence. A paper by Alexis, Alexopoulos and Cohen, finds that technological progress in the 10s and 20s, first decades of the century, was strongly employment creating. But that was the first decade of the previous century. So it may not be as relevant to now. Work by my co-author Anna Salamons and her co-authors finds that the employment reducing effects of so-called routine replacing technical change had been offset by compensatory demand and spillover effects across regions of Europe. Focusing specifically on robotics, which a lot of people are speaking about. In fact, you see many, many more articles about robots, and you'll see robots in the course of a day. A paper by Grayson Michaels finds that across Europe, industrial robots are raising wages, raising value added, and raising demand for skilled workers. But using a variant of the same data, Asimoglu and Restrepo, find that in the US, areas susceptible to robotization are seeing falling employment and falling wages. Our paper asks, in a very broad brushway, is recent labor augmenting technical progress eroding employment? Specifically, does productivity growth cause advancing industries, by which I mean industries experiencing rising productivity, to grow or to shrink? Do cross industry spillovers offset or augment these direct own industry effects? And what's the net effect if they do? Has this relationship changed in the 2000s? And is productivity growth skill biased? In other words, should be worried only about the number of jobs or also the type of jobs and who can access them? So our approach is very broad brush. We're gonna look across 19 countries, 28 industries in 37 years. We're gonna focus on overall productivity growth measured as either output per worker, value added per worker, or total factor productivity. And we're gonna look at employment by industry, employment to working age population, consumption just to corroborate the productivity effects, skill inputs within industries, and then sectoral shifts that may shift demand for skills. And we're gonna do that in the course of 17 minutes. So let me start with a big picture. At the aggregate level, this goes back to many people will know this, but Francis and Ramay is one. It appears that productivity growth is employment augmenting. In countries where we see rising labor productivity or rising total factor productivity, we see employment to population rates rise. We see that in the big five European countries in the US. We see that across continental Europe. We see that in Asia. So historically, this time series evidence says, well, look, we ought to be welcoming productivity growth. But that doesn't fit with many people's intuitions when they see their own jobs potentially replaced by automation. So we're gonna zoom in at the industry by country by year level. We're using data from the EU Clems for 1970 to 2007. I will say in the course of a long haul flight, my co-author Anna Salamon's also updated our analysis through 2014. So at the end of the talk, I'll tell you what we find. So 19 countries, 28 industries, we also use additional measures from the world input output tables to look at consumption responses. So the first question we wanna ask is, do advancing industries grow or shrink? To ask that question, we're gonna estimate a stacked first difference model. The log change in employment in industry I and country C in year T regressed on a bunch of control variables, dummy variables. And then the contemporaneous change in log labor productivity in that same country industry year. And we're gonna ask what happens to industry employment on average as we see productivity rising. Now you should say, well, what should happen? Well, three things could happen at least. In the kind of lump of labor world in which output demands are fixed, we should get a coefficient of negative one. Doubling of labor productivity should reduce employment by half. In the kind of demand surge model, it's possible that rising productivity would actually cause such a large consumption response that we would see employment rise. And the third is somewhere in between the kind of bound of view that we will see shrinkage in advancing sectors, possibly offset by elsewhere, possibly not. What does happen? What does happen is robustly clearly, no matter how we specify it, industries that see rising productivity see falling employment. That's true in different periods. It's true whether we measure productivity using gross output, using value added, using total factor productivity. Doesn't matter what set of fixed effects we put in. It's very clear, rising productivity lowers industry employment. In fact, that's true across all of the 28 industries that comprise the private sector economy in our data. So you can understand why people worry about the employment consequence of productivity growth because they see it right in front of them. Clearly, you can produce your way out of a job. You can become so productive that you're no longer needed. And that's the partial equilibrium effect that people are going to see. No one sees general equilibrium effects, right? We write them down, but you're not going to see, you don't say, well, rising output in my sector will create jobs elsewhere. It may or may not happen, but you're not going to see that. Now let me just quick reality check. Is there something mechanical about this? Perhaps we're looking at employment on one side, labor productivity on the other. Maybe they just negatively covariate because of measurement error. We checked that, we have instrumental variable strategy. But let me just show you the dual of this, which is the consumption response. And we see that 10% rise in labor productivity typically leads to about a 3% or 4% rise in consumption of industry outputs. So two things are happening, productivity is rising, output is rising, consumers are ultimately consuming more of the good, and employment is falling. So it's a convex combination of the two, what you might expect. And this reminds us of something that you all know by way of background, which is unbalanced growth. We've had enormous productivity growth in manufacturing, in primary sectors like mining, utilities and construction, much less so in education and health, in low tech services like restaurants, in high tech services like finance. And as productivity has growth has accelerated or accumulated in those sectors, employment has fallen. So the share across these 19 countries of manufacturing employment has fallen by 15 percentage points between 1970 and 2000, with growing employment in services and education and in health. So that leads us to ask, how do we put together at the aggregate level, we see this positive relationship between employment and productivity growth. At the industry level, we see a strongly and robustly negative effect. How are these two related? Well, perhaps their employment spillover is elsewhere. This could happen through rising final demand, i.e. rising consumer incomes, causes growth and demand for other sectors, or through inter-industry demand linkages. We're not, we don't model those. We're taking a reduced form approach in this paper, we'll look into that later. So we're gonna use simultaneously industry level and country level data to estimate the relationship between employment in my industry, productivity growth in my industry, and simultaneous productivity growth everywhere else in the economy. The leave out productivity growth, to ask whether we see countervailing effects. So this is what you see. These direct effects, the own industry productivity effects, are strongly negative. And that's true if we look at gross output and value added, if we look peak to peak or trough to trough. However, the spillover effects are positive and always strongly significant, and in fact almost identical and opposite in sign to the direct effects. And so the net effect is in every case weekly positive, not always statistically significant, saying that effectively these own industry effects are compensated by rising employment elsewhere in the economy. Productivity is good, but not for the workers in the sector in which it's occurring, at least not in terms of numerical employment. Probably is good for wages and many other things. Now that seems like good news, but you might say, well, it's pretty aggregate. Doesn't it matter where the productivity growth takes place? Is all productivity growth equal? Should I be equally happy about mining versus restaurants versus healthcare? And you would think that the spillovers and the direct effects would vary across sectors. One is if a sector is a larger piece of the economy, productivity growth, they are freeze more income for consumption. Product market competition may differ, so the price effects may differ and therefore the spillovers. Also international integration and production chains may mean that productivity growth and services has different effects from productivity growth in traded sectors. So we're gonna allow those direct effects and spillovers to differ by sector using five sectors, mining, utilities and construction primary, then manufacturing, education, health. And then we break services into two others, low tech, which are restaurants, hotels, etc., and then high tech, finance, business, telecoms. And this is just a variant of the model previously, but now we add more, less structure, so we can allow these effects to differ by sector. What we find is there really is important heterogeneity, both in the direct effects, the employment reducing effects, and the spillover effects. Interestingly, the smallest direct effects are in manufacturing, suggesting the demand is relatively elastic for manufactured goods. Prices fall and you buy an even larger television set. The demand is quite elastic, actually, in many services. Maybe people have finite demand for healthcare, although the evidence from the US is that they don't. The spillover effects also vary. And notably, we find almost no positive spillover from productivity growth in primary sectors, some from manufacturing, and very large effects from productivity growth in low tech services. Which is important, those are a big part of the economy. In some sense, that's good news, because there's potential there. You can't just add those up, because what happens depends on the size of those sectors, the productivity growth in those sectors. So we do that for you, we just take those coefficients, look at the cumulative productivity change in each of these sectors, the implied direct effects, the implied spillover effects, and ask, what do they imply for employment to population in net? Well, if you look at the direct effect, the Baumeau effect, we see that productivity growth has reduced employment, pro-topulation, or notionally, in every sector, most strongly in low tech services and in manufacturing. And manufacturing, because there's been so much productivity growth, in low tech services because demand is so inelastic. So there's a strong employment response. Adding those up, if there was only direct effect of rising productivity in own sectors, the aggregate effect would be to reduce the employment to population on average across these countries by about 15 percentage points. Quite substantial. But now if we look at the spillover effects, not surprisingly, they're positive, that doesn't have to be true, they're not constrained to be positive. The largest spillovers are coming from low tech services. And also for manufacturing, it has a modest spillover, but it's had enormous productivity growth. So if you put those together, the net effects are positive. We say, our estimates say, holding all else constant, rising productivity over these 37 years would be expected to raise employment to population rates, yours by about 6 percentage points. Notice, by the way, that the effect of mining utilities instruction is negative. There's no positive spillover. And then other sectors contribute a great deal. Education and health contributes very little, because as we know, there's been almost no productivity growth in those sectors. So now, and this is true if we look across countries as well. We do this exercise for the big countries in our data. And if you ask, well, how do these implied effects on employment population coming from rising productivity, how do they compare to what's actually occurred? Well, in most of the countries in our sample, overall employment to population has risen. But of course, it's risen because of rising female employment. And the changes coming from rising productivity are modest. They say that rising productivity has generally pushed for more employment. That's good. But it's not the predominant factor that's leading to rising employment. But in general, it's a tailwind that pushes towards more formal sector jobs. You say, well, what's the biggest predictor of rising employment? Well, of course, the biggest predictor of rising employment is rising population. And of course, that seems tautological. Big countries have more jobs. We know that's true. But it's not completely, it actually underscores an important economic point, which is that each person who enters the economy is a consumer. But if they're going to be a consumer, they're likely to need to be a worker as well. And so each person is both a supplier and a demander. And so it's not automatic, but it makes a lot of sense that as people enter the economy, employment rises. And over the time period, we look that the coefficient on population growth is about one with respect to overall employment growth. So now, two final questions I want to address. One, is this time period different? Have things changed? And two, should we worry about jobs, per se, or skills? So to ask if this time period is different, we do the same exercise, but now we allow the effects to differ by decade, both the direct effects and the spillover effects. And what you see is the direct effects appear pretty stable, maybe a little bit more negative in some periods. However, it is the case, if you look at the compare of the 70s, which is the blue bar, the 80s is the yellow bar, the 90s is the red bar, and 2000s is the green bar, the spillover effects appear more modest. And in fact, the net effect is weakly negative for the period of 2000 to 2007. So there's some evidence that the virtuous relationship between productivity growth and employment growth has broken down. However, I will note that was also true that the 1980s had a, the yellow bar, had a smaller spillover effect. So this period doesn't look especially different from the 1980s. Moreover, we did, using the different Clems data set, run this analysis for the most recent 17 years, 15 years. And what you find is if you look peak to peak or trough to trough, again, the net relationship between rising productivity and rising employment is positive. If you average over the Great Recession, everything goes to hell. But that's a, it's not clear how you interpret that. Overall, we don't find strong evidence to suggest that this, you know, that we should, that the productivity has become the enemy of employment. So now let me turn to, I think, what we think of as probably even more fundamental point and a little bit less self-evident, which is the relationship between productivity growth and skill demands. So labor productivity growth could affect skill demands in two different ways. One is it could just be skill biased, that where sectors rising productivity start eliminating low skill workers prior to middle skill workers prior to higher skill workers. Quantitatively, we do not find that's important. We know computerization has been skill biased, but we don't see evidence that where overall labor productivity is rising, firms, industries are differentially substituting away from one skill group towards another. But there's another channel that could be important, which is this unbalanced growth, that as the high advancing sector shrink, if their skill mix is different from the rest of the aggregate economy, that may shift aggregate demand for skills in a non-neutral way. So let's be clear, even if growth was balanced, it's still the case that the contraction and expansion of sectors causes economic pain. It's not Pareto improving except in a frictionless world. So still it means lots of labor reallocation. But then if those sectors differ in their skill mixture, that's not even Pareto improving in a frictionless world because it's going to shift aggregate demand towards four or against specific skill groups. And we know that the fastest productivity has occurred in primary sectors, manufacturing utilities and so on, and also, excuse me, utilities mining and so on, and also manufacturing, which means, and those are relatively low skill-intensive sectors, so you would tend to intuit that that unbalanced growth would be skill biased. And that is what we see if we just run through the same calculations and re-weight industries according to their initial skill employment shares, what we see is that the overall effect of rising productivity has been to increase the sort of aggregate demand for the top-terrace style of skills by about 10 percentage points in this period versus about five percentage points for low and medium skill workers. So even though productivity growth has been employment augmenting, it's much more shifted the locus of employment towards more skill-intensive sectors. If you look across countries, you see this broadly. In the US, what's interesting is that the greatest growth has been for low and high skill workers with the weakest effect on middle skill workers suggesting a form of employment polarization. So the productivity growth rising employment, the unbalanced sectoral reallocation that occurs has been skill shifting. So let me just summarize. So is productivity growth threatening employment? No, not so far. Employment is shrinking in advancing sectors, but the spillovers are offsetting in lagging sectors. The net effect is that productivity growth modestly contributes to rising employment of population and rising consumption. Let me be clear. This is the sort of class of models I was raised on. This had to, you know, the answer was, tautologically yes, these are full employment models. All productivity growth is labor augmenting. That had to occur. A new class of model says lots of things could happen. Machines and workers can compete directly. You can have a case occur where there's just no tasks in which labor is competitive. People become like horses. Our evidence suggests that the, our evidence does not constrain either of those cases to occur. So far we find that the neoclassical wisdom is holding. However, that doesn't eliminate the concern that people have in the sense it really is the case where productivity rises, employment contracts, and that leads to painful reallocations, even if the net effect is positive. Is it changing? You know, I said, you know, I said not robocalypse now, perhaps it's robocalypse later. The virtuous relationship may have weakened a bit in recent years, then it seems to be rebounding. I think the jury is out on that. The distribution of productivity growth across sectors matters. The largest spillovers we find are from services, and that's good news because as the robotics, as the robots come off factory floors, where are they gonna end up? They're gonna end up in restaurants. They're gonna end up in hotels. They're gonna end up in vehicle driving. If these estimates hold, those are the sectors that have the largest spillovers, that productivity growth could be on employment augmenting, although not in the sectors where it occurs. Finally, what's good news for employment is not necessarily good news for all workers. These skill impacts of sector reallocation are non-neutral, and this underscores something which I think many of us have been aware of for a long time, that the challenge is not the quantity of jobs, it's the quality of jobs. There are many high-skill jobs being created. There are many fewer workers who are qualified for that work. And meanwhile, the residual sector that's growing are many in-person services, food service, cleaning, security, home health aides, and those jobs are numerous, they're plentiful, but they are intrinsically low-wage because they use a generic skill set. And that is a concern that I think our paper underscores that the opportunity set is quite polarized and employment is being created, but the jobs that are the most appealing are the ones that are much less available. Okay, thank you very, very much. Look forward to the comments from Dietmar Harhoff. Thank you. Thank you very much, David. That's already a lot of food for discussion. We're very eager to listen to Dietmar's comments. The way we'll have the discussion is we have the two papers being presented and then commented, and then I'll open the discussion with the audience. I just give a chance for the authors to reply to the comments if you feel urged to object or comment on something Dietmar said you will have the floor, and then we'll open a general discussion after the two papers. So Dietmar, of course, yours. Yes, thank you. I hope objections will not be necessary, but let me start by thanking the organizers for having me and pointing out that they encouraged me to put on two hats because as you noted, there is a presentation on productivity and employment. There's one on investment, but the overarching title is also innovation, so I'll say a little bit about innovation as well to link these things together. And in order to do that, I'll provide you with a little bit of context because I'm talking a lot of people who do innovation and who also build the robots, not so much in the service sector yet, but in other areas. I'll comment extensively on this really great paper. I think it provides a wonderful foundation for the discussion, congratulations to the authors. And now I'll come back to the innovation point to productivity because despite of the hardship that productivity growth may cause, we wanna have it. And especially in Europe, as was pointed out yesterday by Mario Draghi, we haven't had all that much of it and we may wanna have more of it and there's some hard work to be done to have it. So innovation and productivity growth, Europe has not delivered. We heard that, I will not go into that. We have relatively stable structures in those sectors that deliver productivity growth. However, if you look more closely, these are challenged right now and these are challenged by very disruptive things. We call them digital transformation. Now the things they are largely driven by successful non-European players that by and in and of itself is not a problem. We have had international division of labor for a long time, but that puts many large European corporates and SMEs in follow-up positions across industries because this is disruptive change in the sense of a general purpose technology. So what we see right now, not just in the corporate sector, but also in the government sector, is lots of experiments. What we thought was the holy grail of innovation management and innovation processes has been lost. People are experimenting because they don't quite know how to deal with this drastic change. They deal with, they experiment with startups, with accelerators, venture capital open innovation, we'll hear later, I take it about ecosystems. So all of these things are in a state of experimentation right now because there is fundamental uncertainty among policy makers. So it's worse than that. Growth is not enough, but in some quarters we also have declining trust in science and experts of any kind that is not limited to the United States. You can show that in surveys. We respond, we have now the term inclusive growth that comes up. Citizen science participatory approaches, but those do not necessarily respond directly to the big challenges like installing new business model across the board. So I think that there are three short-term challenges in the digital realm that Europe faces, one among the SMEs, and there is evidence, I will not go into details here, that there is a little bit of a digital divide. This whole large corporations can handle this change, that the smaller ones, the SMEs, may have problems with that. We're working with startups and ecosystems, and ideally, of course, with market forces, and then there's a lot of digital infrastructure that we need and that isn't quite there yet. And whether governments as strategic innovators will be, an answer will certainly pop up in the panel discussion later on. This came up yesterday and brings us to the topic of the paper. The Guardian had these wonderful lists. You will see that central bankers are neither among the list of safest jobs, nor among the list of the least safe jobs. I'm not quite sure how that is to security detail and so forth, but the automation principle is. But what we have here is essentially an example for news that has been around over the past years. And all of this started, of course, with a study by Freyja Nosswald, who came up in 2013 with a study of 702 occupations and found that 47% of US workers would be at some risk if one looks at the possibility of automation in these jobs. There were other studies then by consultancies. The World Bank came out with one that had 57%. These are controversial, but they make their way into the press and right now they shape the image of digital change. And I think that's very important for us to realize that there is fear out there among people who read this. Now, there are more qualified studies. Anset Al, for example, look at the occupational patterns and come up with a figure of 9%, which is a big difference to the 57 in the World Bank. I think it's fair to say that from studies of this kind, we simply cannot come up with safe estimates as to what the replacement potential is. And therefore it is, I think, very, very good that David, Otter, and Salmond look at the evidence in a slightly more reliable way. However, it's looking at the past. And that is, of course, what we do here. Now, let's look at the setup. Both authors have made ample contributions to skill bias and routine bias change and actually David has led the conceptual development away from purely skill bias to routine bias. So there are many things to be commanded that predate this paper that are in the background. What we have here is a systematic attempt to estimate various effects in one model in order to allow for aggregation of the effects and of course in order to allow for comparison. Comparison across countries, across industries and time periods. They use the clams data. I think that's a prudent choice. They set this up as country by country by year stack first difference models. There's some experimentation in the background with IV approaches. They stick to the OLS estimation later on. All of this is in reduced form that comes with disadvantages but that is sort of by way of the territory that they choose. It's a very logical approach. I have a few quibbles but I will communicate most of them offline because they're really very few. You can talk about measurement issues. You could talk about or take more seriously the panel dimension here and apply dynamic panel estimators or take this to the time series. One could try to decompose the productivity growth meaningfully in different components. All of this is work that goes beyond what they want to do here and that can build on this particular paper. Now what are the main results of this conference paper? The first one is predictable. Own industry employment declines as labor productivity increases and this is a very stable finding that they test and where they have various robustness checks. They also show that there is a consumption response but nonetheless there's a negative impact at the industry level but not at the aggregate level and they go on and try to explore why that is. And the argument is that productivity growth has important spillover effects into other sectors and that these effects actually fully offset the negative internal effects and that the net impact on employment over population is weekly positive. So they speculate on the sources. David said that already that they do not go into that because they're somewhat restricted what the data can tell them here but income effects or inter-industry demand linkages are the most probable candidates here. They also point to important heterogeneity. Manufacturing has, David said, has the smallest own effect. I think calling it the least negative own effect makes it maybe a little bit clearer what is meant and low-tech services has the larger spillovers which is interesting. So of course this may all have changed over time and fortunately the data allows them to go after that and I've copied from the slides that David gave me before one picture that I find particularly telling and I have now chosen the version where the productivity is measured in value edit rather than in gross output and this picture looks a little bit more intimidating than the other one and maybe we wanna discuss why this one is not sort of the focus of the presentation before because what you see here is that things may have changed in the last decade on the consideration here or maybe over the whole time frame depending on what you wanna make out of this the standard errors in especially in the external effect estimators are fairly large. That is understood because that's sort of aggregating variables on the right hand side which makes them less precise and then you also get a relatively noisy overall effect but if you want you can interpret a time story into this particular picture. So the conclusions that the authors draw are then not quite so strong as in the abstract of the paper does productivity growth threaten employment not so far. Employment in advancing sector shrinks we have heard that. The net effect is productivity growth modestly contributes to rising employment to population as well as rising consumption. There is a concern about the 2000s. The distribution matters. Productivity growth in services produces the larger spillovers we have heard that and the authors then turn that into good news. Robotics may have potential to raise productivity in services. I would put a little warning sign on this. This is based mostly on a view of robotics we have from other sectors whether that is the case that's a sort of prediction into the future that's a little bit speculative. Nonetheless there's the virtuous story that David has described and then he comes I think to the gist and to the policy concerns that we should have. The impact of skills is non-neutral. Maybe we don't want to worry about the overall quantity of jobs but due to sectoral shifts and other factors there is an impact particularly for low and medium skill workers that should concern us. So growth is not enough to pick up what Ben Bernanke said yesterday. A minor point, the polarization differences are very interesting. You don't comment then extensively in the paper but it might be nice to hear what drives them. Okay let's come back to the asymmetry because I think that's a very important one. Growth is not enough. Education and human capital formation matter strongly and so that puts us firmly back in an institutional game in how to deal with skill formation and education. How do we teach? How do we generate reliable insights on learning? And I think there is much to be done in this particular field. Now technology itself may come a little bit to a rescue with MOOCs, with other forms of learning but then there's also very interesting observation that the market supplies interesting skill upgrading models. If you have gone into San Francisco and have gone to galvanize which is a organization that takes people in for three months and turns them into relatively capable Python programmers and I think that that is a very interesting short-term response that we should also look at here in Europe. So let me come back to innovation because in any case there's hardship here, there's need for prudent policy. We don't want to have, we don't want to forgo the productivity gain so what do we do? Here's a list of policy options that OECD drew up in the future of productivity in 2015. And I have sort of drawn up my own little list of what can be said positively or maybe not so positively for European institutions at this point in order to get at the added productivity and innovation. So I would note positively in response to we need improvement in funding of basic research, by the way that is very interesting because we had a long time period where policy advisors were arguing in favor of more applied work, applied research. So I think we're doing nicely there in Europe with the ERC, 6,000 or 6,500 grants out by now. I think that that is doing very nicely. I think that we have made progress on global mechanisms to coordinate investment such as anti-tax incentives. We put a dampener on patent boxes or we put OECD did in the course of its BEPS base erosion profit shifting considerations. We have harmonization of IPR systems in Europe to some degree. Unfortunately, the harmonized patent system is right now pending approval because there is a German case at the Supreme Court. OECD also advised to support diffusion from global frontier firms via trade and I don't have to tell this audience that that is unfortunately a little bit under pressure given political developments. It would be nice to have a comprehensive trade agreement with Japan, maybe one that follows CETA in terms of design and so forth which was after all acceptable for large groups of the population in Europe. So then there's a concern of not favoring applied over basic research. I think that there's some positive things to be said there. And of course the call for reducing barriers to firm entry and exit and that brings us back to the topic of the paper. For that we do need labor market reforms in order to be able to handle those sectoral shifts and in order to have increased worker mobility. Internationally, there is a little bit of a dampener there with Brexit of course. Okay, let me wrap up. I have 18 seconds. To David and Anna, this is a great analysis of the impact of productivity growth on employment. I think it's very informative and it also pulls out the implications regarding skilled buyers, routine buyers, sectoral shifts and so forth very, very nicely. Whatever is said, it may be a fortuitous relationship but technical change has the potential of creating disruptions and hardship. So apparently growth is not enough and to the Europeans, I can just say if you wanna have more of productivity growth, we still have some hard work ahead of us to get those innovation related policies and institutions up to shape. Thank you very much for your attention. Thank you very much, Dietmar. And by the way, thanks to both of you for sticking with the time allocation. That's as much time that we give back to the audience or that belongs to the audience actually. Dietmar, I was intrigued by your remark that central banking is not on the picture of the safe jobs in face of technical change. But thinking about it while you were speaking, I thought that was probably a good news because given that central banking is obviously a service industry and given the polarization result that David put forward, if we would be safe that would mean that we are either very high-skilled or low-skilled and that's not a question we want to ask ourselves. So thank you. David, you want to have a chance to respond to some of Dietmar's point before we move on? I'll only take one minute. First of all, I agree with what Dietmar said, that innovation is central. We're looking at the results of innovation, not the inputs into innovation or the causes. I agree, this is a very high-level analysis and there's lots of uncertainty. I do think that what's happening in the 2000s, we could underscore that more. We don't wanna raise alarm. My thesis advisor, Larry Katz, once said, I love history and I love science fiction, but I think that history is a better guide to the future than is science fiction and so there's always a danger to extrapolate too much from that. There's more to do here in terms of looking at the role of input output linkages, the role of trade integration and I think that's been extremely consequential for employment in the United States for the role of population aging and so I think we think this provides a foundation for looking at how these factors interact. I do just wanna flag the fry and Osborne results. They're extremely alarmist. They say, well, 43% of jobs are at risk of automation. Well, all jobs are at risk of automation. I mean, the truth is almost nothing that we do for a living now resembles what people did 100 years ago. 40% of US employment was in agriculture in 1900, now it's 2%, it would have been hard to predict that once agriculture went away, people would be doing search engine optimization for a living, but that's what occurred. So that is the most kind of mechanistic view of automation and really not representative of the dynamics. I think what is correct about that and what our paper underscores as well is even if the aggregate effects are positive, the reallocation is painful, especially if it moves quickly and it may not be neutral. Even in a frictionless world, it may shift skill demands and that's the pattern we have seen across the developed world over decades and I think there's every reason to continue to focus on the challenge of matching the set of human capital inputs to the labor market opportunities. Thank you. David Dietmar made a, what I found a very thoughtful remark on the sources of productivity, meaning labor productivity obviously can be TFP or capital deepening. Did you look into that? I mean, would you come to this? We use three different measures of productivity, which are labor productivity, raw labor productivity, value added and TFP. We didn't decompose where those are coming from and I think that's a very worthy topic. We wanted to use an encompassing measure rather than looking at specific episodes or incidents but obviously not all productivity growth needs to be the same. So Thomas, talking of capital deepening, for us yours. Well, good morning, thank you for being here. Thank you to the organizers for inviting us to write this paper. Thank you, the ECB, President Draghi and Benakure for chairing the panel. So this is a joint work with Robin Dutling and Hermann Gutierrez, both our PLG student. Robin in Amsterdam and Hermann with me at NYU. So the focus of the paper is to understand investment dynamics in Europe and in the US. So here on this graph, you have net investment relative to the stock of capital for Europe on the left and US on the right. So the series on the right for the US, actually that's part of a series of paper have been writing on the US trying to understand why we see this downward trend in investment in the US. And just to be clear, when I say investment, I mean all measured investment including intellectual property and intangible investment. So this includes structured equipment, software as well as R&D and intellectual property style investment. So there's this downward trend in the US which I find striking and we try to explain. So we have a series of papers on the US and some extent what we try to do in this paper is to replicate everything we've done in the US in Europe. Okay, so Europe needs to be defined. We're gonna use various data sources and some of them restrict the availability of some countries so when it says EU claims it means the eight largest EU economies and the UK we treat separately which is not in this picture here but throughout the Tokyo and think of it as the UK looks, it's like in between EU claims in the US a bit more like the US. So I can tell you everything we find about the UK later. So investment here includes intangible so it's not just physical investment. The other thing is we look both at growth and net investment in the aggregate. It's kind of robust, it doesn't matter which one you look at. Sometimes measures of depreciation at the more micro level are noisy. So we tend to focus on growth investment when we drill down to firm level investment. So you see in both cases this decline in investment, investment rate, okay. So the growth of capital is clearly much lower now than it was in the past. And the question is, is there a common cause for both or is the story different for the US and for Europe? And so I'm gonna show you that, I'm gonna explain to you what happened in the US, what happens in Europe and in fact the two stories are quite different. And the first place you can look at is what happened to profits over the same period of time. So profit in the US, the scale is a bit different. So this looks less volatile than this but that's because the scale is different. So profit in the US have of course taken a hit during the crisis but they've recovered extremely fast and in fact today they are, if you just accumulate the profits over a three year moving average, they've never been higher in history period, all right. So the profit rate, so this is measured as operating surplus over the capital stock. The profit rate of US firm has never been higher in a persistent way. In contrast, in Europe you see profit was kind of relatively stable and then it took a big hit during the crisis and has not recovered since. That already suggests that the reasons are gonna be different. One thing you can do is take the ratio of investment over profits and figure out how much of their cash flow do firms flow back in investment. Of course that's cyclical, you always see the dip in the crisis but in Europe there's no obvious trend, maybe a small downward trend but it's not very obvious. In the US on the other hand there's this continuous downward trend. So firms are plowing back less and less of their profits into investment. Of course by accounting that means what they are not putting in investment they are paying as dividends or shares by back. That's just by definition. And final point to show you that the story is different in Europe and the US, it's Tobin's Q. So Tobin's Q is gonna be core of our analysis so just a reminder, Tobin's Q is the value of the firm divided by the replacement cost of the stock of capital. And remember again the stock of capital include intangible investment, Brent and stuff like that. So if Tobin's Q is more than one that means you should invest because any extra dollar you put in capital is valued more than $1 by the market. And conversely if Tobin's Q is less than one you should divest. Comparing the level of Q between Europe and the US is a huge pain in the neck because of the treatment, in particular of the treatment of land because if you think about the firm you want to, of course you have the, you have the integral property, the equivalent structure, you also have the land that they own and it's not treated exactly the same way. So it's hard to compare in levels between the US and Europe. You have all the tables in the paper to make the comparison. But of course the time series are very clear. So Q is high in the US, in fact it's higher than it's ever been except at the peak of the internet bubble. In Europe on the other hand, Tobin's Q is pretty low, okay? So that sounds like a simple point but actually it has a very broad implications because broadly speaking there are two kinds of theories you're gonna write down about investment. Some theories are gonna explain why you would have low investment rate because you have low Tobin's Q. That includes every theory based on spread, risk-premia, weak aggregate demand, weak expected TFP growth, every single one of these works through low Q and low investment, okay? On the other hand, you could have low investment despite high Q and that suggests the gap and the gap could be explained by financial friction by intangible investment which is still not properly measured despite all the effort we are making in improving measurement. You still have mis-measurement in intangible or different type of investment function. Jan has a paper on that. Or, and that's the story I'm gonna push today, it's competition, okay? If you have changes in the degree of competition you typically move Q and investment in opposite direction. If competition goes down, profit rates go up, Q goes up because it capitalizes the rent but investment goes down, of course. And so what's the story? Where our story is that this is the EU, in other words, investment is in line with Q and this is the US, investment is low despite high Q because the US has become less competitive. That's what I'm gonna show you. All right, so just to show you that investment is in line with Q in Europe. This is what you predict using Turbin's Q. I showed you the two series separately before. So this is just if you regress one on the other. Try to predict investment at the very aggregate level for the whole EU. So it works pretty well. There's still this gap here, of course, suggesting some gap, of course. But then you look at the countries and the gap is entirely driven by Spain and Italy. Countries that we know have other types of constraints. If you remove Spain and Italy then there is no residual, okay? There is no residual. Investment is exactly where you would predict based on Q theory. In the US, on the other hand, you get that. This is what you would predict based on Q and this is what actual investment is. This gap is large, persistent and has a huge, I mean, if you believe the story I'm gonna tell you, that's massive welfare consequences. I'm gonna tell you that roughly speaking, K in the US right now is something like 5% or 4% below what it should be. That's massive implications for labor productivity and welfare. Okay, so why do I believe competition is the story? Okay, so now this picture doesn't look like much but you wouldn't believe the amount of work you need to create that picture. Okay, so on this picture you have the half-induln index for the US and for Europe in red and in green. So for the US it is not too hard because we have pretty good firm level data. This is based on the top 50 firm of each industry to avoid truncation effect for very small firms but it's robust to the way you define it. So that's the simple half-induln measure, not the one adjusting for common ownership which we can discuss later. Very strong and continuous increase in concentration in the US over the same period of time. Europe's is decreasing concentration. Now you might ask why do I have two curves for Europe? Well because you can define Europe in two ways. You can look country by country, that's the right curve. You can compute the half-induln for France and Germany and Italy separately and then take the average across countries, okay? So implicitly if you do that you're treating each country as one market. The polar extreme, of course, you can really believe in the single market integration in which case you would treat the entire Europe as one market. So for a particular industry, you would look at this industry but you would look at players in all of the European countries to compute the half-induln index. Now mechanically that half-induln index is gonna be lower than the red one. But the trends are very similar, okay? So my sense is that there's clearly a trend towards less concentration in Europe. Both at the control level and at the EU level. And if anything, if you think about where what we are doing at the same time since we are integrating the market, over time the green curve becomes more relevant relative to the red curve. So probably what we are doing is we're moving from the red to the green as we go along. So I think the decline in concentration, the increase in concentration in Europe is quite remarkable. And I'm gonna show you that this explains the different patterns of investment. Okay, so I won't spend too much time on the data but this is where the value added of the paper really is. This is the list of countries I mentioned earlier, the eight countries that are part of the core sample for Europe. So we're gonna use country industry data. We're gonna use standard claims to have a longer time series. And then we're gonna go to the firm level data where we have CompuStat Global and this monster here, Amadeus Orbis. And this is where we, this would not have been possible without the help of Steadman and Carolina because the big difference between Europe and the US now in terms of research is the firm level data so much better in the US. In Europe we have this Amadeus Orbis data set which is really of relative low quality and extremely difficult to use but as it turns out they have done an amazing amount of work to link and create long panels with actually correct firm level data and thanks to their work we can compute the half-indole for Europe over time. So essentially every time you see this firm level measure for Europe what we did is they run our code on their data. Okay, so we are forever grateful. And then some caveats about accounting standard but that's mostly for the level so it doesn't matter for what I'm gonna show. Okay, so I'm gonna show you two sets of regressions. The first is about the half-indole. So here as well as background remember I've done all of that for the US, okay? And for the US we show that in the time series you can explain where investment is rising concentration and then you can do the same at the industry level and at the firm level. In the US investment the gap between either profits or Q and investment the gap is coming from industries that have become more concentrated over time, okay? And within industries it's coming from the firm that have exceptionally high profitability within each industry, okay? So the question we had was is the same true in Europe? It's interesting because in Europe we don't see the overall trend, okay? So if anything concentration in the aggregate is going down in Europe but there is still dispersion across industry so you might ask is it the case that industry that have become relatively more competitive faster than others do they see more investment? So same exact regression we run in the US we are running now in Europe and this is the question you find on the half-indole, okay? So these are industry country regressions, okay? And you see the very clear impact of the half-indole on investment controlling for Q which is by the way this coefficient is very similar to the one we found in the US something of the order of four to five percent with the right scaling. So that's the impact of the half-indole. The other thing that's important obviously is the rising share of intangible, okay? So we know that intangible investment is different. It would accumulate differently. It has different depreciation rates and different adjustment costs and there's a sense in which we are still not measuring it perfectly well and therefore you think that in industries where there is relatively higher share of intangible you might miss more of the investment, okay? And indeed that's correct. It's not overwhelming in terms of how important it is but it is an important factor, okay? And both are very orthogonal. So in fact you can see when you put up both of them together in the same regression nothing changes. The points estimate are the same, the standard are the same. These are essentially orthogonal factors to a large extent. And the last thing I want to say also is we run these regressions. So this is for total investment which includes tangible and intangible. We also run them separately for intangible, separately for tangible and everything I'm showing you is robust to both cases. So this is at the industry level which is a natural place to check the impact of competition. If you want to look at credit constraints the more likely, the more natural specification you look at firm level, okay? So again the question is in Europe we see sometimes this residual is it accounted for by credit constraints or banking crisis, okay? And the answer is yes. So you can see it here in terms of the firm level regression. So again we have investment on the left. We have Tobin's Q which is always strong and significant. So, but then we have this extra bit here which is easier country in recession and do you have high leverage and do you have short maturity or I guess here long maturity, okay? And this by the way has been shown there is at least a couple of papers from the ECB showing that. And I know Luke is around here so I think he has went further showing that. So we are confirming stuff that has been found already which is that if you're in recession obviously it has an impact on you. If you enter the recession with a lot of debt then you will hit harder and if the debt is short term you also hit harder or conversely if you have long maturity it dampens the impact of the debt, okay? That's well known. So that's for all countries. It turns out this is mostly concentrated among the, I don't know how you call it, very, very country or non-core country. So this is the jeep's effect. So if you're in one of these countries and you have a lot of debt to refinance and then you have a negative impact on your investment if you have long maturity that dampens the effect, okay? That's exactly what you would expect. Again, going back to the macro now the question is if you put the pieces together do you get a consistent story? Yes. So green here, these are the time effect that you would get from the residual of the regression year by year, okay? So if you don't control for anything you have the green curve which is the strong deep line in investment in Europe, okay? If you just take into account Q and firm edge you almost get rid of the trend, okay? There's still a little bit here. So there's no obvious trend in the time effect anymore if you add the intangible share and the financial constraint which mostly hit in the jeeps then the scale again is different. So these varieties is also reduced and you can see there is absolutely no trend left, okay? Or in other words, in Europe I can explain investment at the firm level, at the industry level, at the country level for the Europe as a whole with just the standard Tobin's Q and a measure of financial constraint. In the US, it's the opposite. In fact, Tobin's Q is high. It's never been higher almost except in 99 and despite that investment is low. So what's going on? So we show in the other paper that it's really well explained by lack of competition and there's something that I wanted to do which is this concentration actually is not uniform across industries in the US and we looked in particular at some industries where concentration has gone up the most. So this is what we've done here. We've picked the US. We looked at the top five industry by increasing concentration in the US, okay? So the red curve here for the US, these are the industry that have become the most concentrated in the US and that's their average orphanage also. Mechanically, that goes up a lot. They were selected on that basis. Now we looked at the exact same industries in Europe and this is what happened. That's amazing. And then even more amazing, that's the investment rate in the US and in Europe. It's actually higher in Europe. Remember, this is the raw data, not controlling for anything. These guys are in crisis, their profits are lower and they're still investing more than the US counterpart. The only difference between these industries, they are the same technology. In one case, there is strong competition, in the other case, there is no strong competition. So that's my first big takeaway on competition. The second thing I'm gonna talk about is intangible investment. Jan is the world expert on that. So I just want to highlight things that are a bit different and more like the comparison with the US. Because that's one place where we know, of course, that the US was way ahead of Europe at least some years ago, okay? So we had a basic question. Actually, two basic questions. Does intangible explain part of the mismanagement of investment? Yes, it's there. It's significant, it's not the overwhelming factor but it's important factor. But the more important question from my perspective was, is Europe closing the gap with the US in terms of intangible investment? And the answer is yes. It's very hard to measure properly because it would be nice if a national account people could agree on depreciation rate between the two regions so we could compare the stocks as well. But this is what we've done. So the US is in red, Europe is in green and the difference is in blue. So this is using claims, you can see this. So there's a clear decline in the difference, okay? So catching up of Europe based on claims data where the depreciation is really a bit messy. This is based on compute stat global, so firm level data, same exact calculation. But in this case, it looks like the convergence is essentially done, okay? So according to claims, we've closed the gap but there's still a gap. According to compute stat global, the gap is gone to zero. Which one to believe, honestly, it's somewhere in between, I really believe that. I think these guys are too optimistic, these guys are too pessimistic because of the depreciation rates. I think we are somewhere in between. So we are closing the gap in terms of intangible investment with the US. But the thing that's amazing is how we do it, actually. So this one takes, I'm fine, okay? Takes one minute to explain because there is lots of information, okay? But that's a really cool picture, you really understand it. So these are vintages, okay? Each line is a vintage, okay? The question we're trying to understand is how, so both economies, both Europe and the US, obviously are ramping up their intangible investment, are becoming more intangible intensive. The question is, how do you achieve this transition, okay? In the aggregate, it's very similar. The capital share of intangible is gone up roughly the same amount in both cases, okay? But the way it happened at the micro level is very different and interestingly different. So these are the vintages. So this is by cohort of firms for when they were born, essentially, okay? And or when we classify as entering the day, okay? And this is for Europe and this is for the US. So for the US, what you see is that given a vintage, your intangible intensity is roughly stable over time, okay? And the way the US economy has achieved the increase in intangible is by each new cohort being more intensive than the previous one, okay? So the intangible intensification of the US is through the extensive margin, okay? Entry of high intensity intangible firms, okay? So cohort after cohort, the thing goes up, but each cohort is relatively flat over time. And this process mostly took place in the 1990s, okay? Only perhaps early 2000s, okay? So it started earlier than in Europe and it started and it was done differently. In Europe, on the other hand, so you can see it starts later, okay? And also it takes place within each firm, okay? So each cohort is becoming more intangible intensive, okay? It's the incumbents that are becoming more intangible intensive in Europe, okay? So I think there are two implications. One is, of course, it went fast in the US, presumably because these guys can adjust faster than the incumbents, so that explains some of the time lag in history. But if you look at now what's going on today and you compare where we are today, the two places look very similar in this dimension, although they achieve the dynamics in very different ways. All right, so let me wrap up. So what's the summary? We have weak investment in the US and in Europe. Our interpretation, however, is that the causes are different. In the US, it's structural. Firms have firm in most industries, except for the ones that are subject to foreign competition which we analyze in details in the other papers. But in most industries, firms have weak incentive to invest because nobody's threatening their market share. In Europe, investment is low for cyclical reasons. It's purely a leftover legacy of the Eurozone crisis. So if that view is correct, now that the risk premium are coming down, then we would expect a pickup of investment in Europe. In the US, on the other hand, I just can't possibly imagine what you would take to make firms invest. They have record high profit, record high valuation, zero funding cost, and they still don't invest. So give me a factor that would change that. I don't see one. Intangible, we are catching up. The catch-up was very different, as I showed you, income inverse with startups. Broadly speaking, I think this suggests a role for product market regulation and antitrust. Because at the end of the day, you might ask yourself, okay, why is it that we see these diverging trends between Europe and the US? So I'm going to show you two pictures. The first one is the OECD Product Market Regulation Index. And when I grew up as an economist, it was all about, oh, Europe is too regulated. Every OECD report every year was like, you should remove this product market regulation, okay? So we're always beating up on the European countries to do something about their product market regulation. And the truth is they did, okay? So these are the 80 open countries in our data. Remember, the OECD PMR indexes are every five years, I guess that's why you only have four of them here. So there were every single one of them, every single one of the eight countries was above the US in terms of product market regulation in 1997, okay? They went down over time. Today, the last vintages, every single one of these eight countries is below the US in terms of product market regulation. So I think that's an important factor. And the other one, of course, is antitrust. Antitrust is a complicated object to measure. So this is one attempt, okay? So here we have my hair findal for the US in red, okay? And in green, you have one measure of antitrust. And caveat, that's not obvious, that's the best one. This is just counting the number of action taken under section two of the Sherman Act, which is the act you use to prosecute antitrust in the US, okay? There are other measures, you can look at fine, you can look at people going to jail. This is one measure, which is you count how many Sherman Act chapter two are started, investigations, okay? And there is a clear negative question between the two, right? So over the past 10 years, essentially antitrust has decreased in the US, that is, that's my interpretation. Definitely, merger or approval, it's open bar, that's for sure, there's just merger always approved. So I think that one is clear. Antitrust, other measures might give you something different, fines have gone up, so maybe it's a bit more ambiguous. But that one is very personal to me because I moved to the US in exactly here, 1999, and I remember arriving in the US and I was amazed, everything was so cheap. Computers, laptop were half the price than I had in Paris, airline tickets were much cheaper and phones were much cheaper, okay? Today, it's the same, but when I fly back, because like just to give you one instance, if you have a cell phone like an iPhone, the minimum contract you're gonna pay per month in the US is $100, if you want to have given amount of data, the same exact contract in Paris is 40 euros. Thank you. So first I want to thank the organizers for putting this session together and for inviting me to discuss this very rich paper by Thomas and his co-authors. I recommend if you haven't read it to take a look, there's a lot of work and a lot of information there. A lot of work, especially on the data, which Thomas certainly didn't have time to discuss in detail and which I won't focus on as well. So I recommend the paper. I also want to comment to bring in some other work that's been done around in less than the last year from a number of authors, including people on the panel and in the audience, because this has been a rapidly evolving literature and moving pretty quickly. And I see in the paper Thomas presents today already through the cumulative effect of work that's been done in conversations that we've been having across a number of authors and papers. So the paper they present today has a pretty nuanced answer to a simple question of whether there is an investment gap in the U.S. and the U.K., the U.S. and the EU. And I write in the answer that they say yes and no and I realize I want to be clear that that doesn't mean yes and no in the com si com sa kind of way. It's a very precise yes and a very precise no. The, in particular, the answer is no for Europe. There's no investment gap and the answer is yes for the U.S. And much of the paper focuses on why it's yes and why it's no looking at each separately. And then of course at the end, one of the fascinating open questions is not only why in each region but also why are the answers different in Europe and in the U.S. So I will focus a bit on some work in the U.S. where Tomas has focused earlier and then return to his implications for the difference between the U.S. and the EU. So let me first look at the components of fixed investment in the U.S. which I think is illustrative of the issues that arise when you address why investment has been low. And Tomas commented a bit on the composition and I want to emphasize that. So the first line here is gross private domestic investment. So this is the broadest measure of investment and I've measured this lately different than Tomas did by looking relative to GDP. So you're scaling by output. And here you see a bit of a decline at the end but one should be immediately suspicious about the role of residential investment and overall investment in the U.S. So if you take out residential investment you see that the crisis is much more muted. And if you squint you might see some evidence of a decline but the decline is clearest when you look at equipment and structures. So the sharpest decline is in the fixed capital investment and the difference between the red and the green is this upward trend in I just put in intellectual property investment which we're now measuring in the national accounts and that accounts for much of that offsets much of the decline in equipment and structures. So this trend has led some researchers to look in more detail at the role of intangibles of which intellectual property is part but not all and whether how much of the decline in equipment and structures is actually accounted for by rising an increasing role of other forms of capital. So intellectual property is an important one but other forms of intangibles would include brand research and development which does which might not be measured as intellectual property and other measures but those are the main ones. This also makes clear that just looking at the time series is not going to resolve this issue because there's long run trends here and so both say concentration is going up as Thomas emphasized but intangibles are also going up. So looking at the cross section data and using that for identification is an important part of understanding what's happening in these data. So one thing Thomas didn't emphasize in his paper partially because of data limitations is that much of what we're seeing is an effect that occurs in investment after 2000. So this is based on looking at firm level data and you see something similar in the EU that when you put in all of the usual controls like Tobin's Q and cash flow and firm fixed effects the time effects have a very clear pattern. So you see on the top here the 70s and 80s there were positive time effects so that is investment overperforms. In the 1990s the time effects are nearly zero but these negative effects of the decline is really starts in 2000 and is an effect throughout the 2000s. It's exacerbated by the financial crisis but it starts well before then and other authors including some people here have emphasized that there's a same time pattern when you look at productivity that the productivity decline is actually initiated well before the financial crisis. So this again raises the issue of identification because there are many things that change during the 2000s so intangible investment picks up but there are many other changes as well. So it leads us to looking at the cross section. So this is a bit of a complicated chart but it bears some similarities to things that David Otter has worked on. I think it illuminates a bit of what's happening across industries. So let me explain what we have here. So the chart calculates the share of investment that's represented by industry and here I'm using a slightly different definition of the industry than one typically sees so these are the Fama French industries and the major difference relative to the NIPA or the national accounts industries is by breaking out high tech from manufacturing and from services. It's very difficult to look at manufacturing as a whole over this period because the high tech manufacturing industry is growing strongly and investing but the traditional old line manufacturing firms are declining and this industry classification breaks them out. So on the left you see that the investment is really shifting toward a set of industries that I've defined as non-tradables which is mostly energy and telecommunications and telecom here is not telecom devices so it's not phones it's telecom distribution. So the industries on the left to which investment is shifting are grounded industries or non-tradable industries where you're putting in pipelines and extraction and distribution towers and networks. And so much of the investment in the US and this goes back to the 1970s through 2015 has been in this literally fixed investment not just fixed investment as equipment but investment that's fixed in place and the decline in investment so where a capital has been shifting away or investment has been shifting away is in production sectors and the production sectors we're not surprised have seen a smaller share of investment over time and in particular manufacturing durables and non-durables. The surprising part of this chart I think is the part on the right where you might expect that high growth industries such as high tech and the investment industry here that's health is health devices. You might expect there to be rising investment in high growth industries as well so you might expect a U shape in this chart but instead on the right hand side you have the dog that doesn't bark. You have the high growth industries that are not investing and this is part of the puzzle this is where the gap resides that Thomas is emphasizing. The interesting comparison here is also to think about the similar chart for labor markets that David and his co-authors have drawn where you see the job polarization where there is actually the U shaped result where you see employment shifting to non-tradables on the left but also to high skilled sectors on the right but investment looks a bit different because you don't have the skill bias happening in investment that we traditionally saw in employment so investment looks a bit different than the labor market here. So when you see these gaps in investment given the measurement problems that we face that should really be our first stop in trying to explain them and there are surely measurement issues broadly here certainly Q is very difficult to measure but the strength of cash flow in the US enforces that we're probably not going to explain our investment puzzle entirely with mismeasured Q. And another place to go is to think perhaps we're just mismeasuring intangible investment that there is a lot of intangible investment there we're not picking it up and that would fill the gap in our investment regressions. The difficulty there is when you actually think through how that works it's not as obvious as it sounds. So the typical story for mismeasured intangible investment is that firms are actually investing say they're building software internally or they're developing brand internally. And so rather than measuring that as capital investment firms are expensing the labor expense and the materials expense that go into intangibles. But what that means is not only is investment too low but expenses are too high. That you shouldn't be counting that labor expense and that materials expense as a cost you should count it as an investment. And if that's right then there's two things that are mismeasured and when you look at net business savings so you look at business savings less investment those two effects would net out and there would be no effect on business savings. But in fact we see huge changes in business savings. In the direction, let me show you the chart this is net business savings and in the earlier years you see there it's significantly negative in the US which means that investment exceeds retained earnings essentially investment exceeds savings. And so what the negative numbers are the financing gap that firms fill with external finance. But on the right hand side you see that again since the early 2000s this number is significantly positive. So business saving, business is the corporate sector is a net source of savings to the rest of the economy. And that's not a story that's just mismeasured intangibles. This says that savings is very high and firms are not investing. So this, there's two points from this one is it's very consistent with Tomas story that financial frictions are not explaining low investment in the US. Cause it would be hard to square financial frictions with this much net business savings. And secondly it doesn't look as if the investment shortfall is entirely due to mismeasured intangibles. Looks like investment is low even relative to savings. Okay, so the other explanation is concentration. And this has been an interesting highlight out of recent work. Again David has done work on the labor share which highlights the role of concentration there. And the current paper show a similar result for investment that perhaps concentration can't explain the gaps of the decline that we see in labor share as well as an investment. This is a really intriguing idea and is very thought provoking. So in some way it's very simple. Because if there's market power or if there's decreasing returns to scale marginal Q is less than average Q. And what we put in all of our regressions is average Q. And so that says we're overestimating the incentive to invest and if there's market power then Q is going to overstate investment exactly as Tomas's charts show. So it seems like a very appealing explanation and I think there's a lot to it and so we should push harder on this and ask about it. So first do the data lineup. So I'm using the charts here from David's five co-author paper in the AER papers and proceedings there's a longer paper in this literature there's always another longer paper that goes through all of the data and much more detail. But here they're measuring concentration ratio. So it's similar to what Tomas is looking at in his data and I'm going to emphasize the higher curves in blue and green because those are concentration ratios measured off sales. And you see that they're rising for a broad range of industries so manufacturing is going up, services goes up quite sharply, retail trade, wholesale trade, a little more shallow increase, utilities and transportation. So a very widespread increase in concentration but what I also want to point out is that their data goes back to 1980. So for all of the work that Tomas has done on data the measures of concentration start in the late 1990s. And so for manufacturing, you see by the late 1990s most of the increase in concentration has already occurred and the same for services. The big increase in concentration in otters data is in the 1990s. So there's a bit of a puzzle here because remember the enforcement measure that Tomas showed spikes in the 1990s. So this immediately for an empirical economist starts raising endogeneity because you worry that when there was much more concentration that is there were more acquisitions than there was also more enforcement because what he's measuring is the number of cases. So you worry that what we're measuring is not so much the intensity of enforcement but just the number of acquisitions that were brought forward. Now, of course the number of acquisitions, the number of cases is endogenous as well. So this is not dispositive but we just want to be careful about the timing because it looks a little different in David's data than it does in Tomas's, okay. Let me skip David's finding in the interest of time he can advertise his own work during the discussion. So the question that we are left with is it looks like both concentration and intangibles have an effect on investment, have a negative effect on investment. So we have to ask what drives the increase in concentration so if we're going to worry about endogeneity and when you see that manufacturing is one of the main industries with rising concentration then the issue that arises to me is reverse causality because what may be happening an alternative explanation is that in a low growth industry you would have had low investment anyway and then that drives consolidation and so the concentration is following from the low growth and the low investment rather than low investment following from concentration. Now these are not mutually exclusive but before we implement policy we certainly want to understand causality and there are many reasons to think that in these sectors, so here's manufacturing as a share of total value in the largest firms, manufacturing is dramatically shrinking over time whereas you're seeing high tech is the sort of blue-gray color is rising over time and other industries taken as a whole are rising over time as well. You see similar patterns in the number of firms so it's not only the firm value where you have the shrinkage and consolidation but also in the number of firms which you would expect. Similarly, new entry is in high tech firms and not in manufacturing firms so in particular, the largest and most active industries for new entry is business services, business services that's the same sector as Microsoft and Facebook. So there's much more activity there. So final question is concentration seems highly correlated with the decline in investment but in order to take policy action we wanna know what drives concentration it's clearly an endogenous variable so is concentration ultimately the culprit or is concentration a byproduct of other things happening in the industry and the industry equilibrium and there is an interesting correlation between intangibles and the industry concentration because if firms are building market power one way to do that is through building brand and also through intellectual property which are both intangibles, intangible investments so these are not mutually exclusive. You've run out of time substantially, sorry to say that. So I look forward to the discussion because there are many open questions. Thomas, do you want to give a first reaction to Jenny's comments? Maybe we can cancel the two minus five plus five and yeah, just then I'll say one thing, only one thing, I'm gonna advertise my own notes. So I agree with the endogeneity issue which is it's plausible that if you see weak growth going forward, you're not going to invest and you're gonna try to consolidate and that's exactly the point of the other paper we have on the US where we come up with three different identification strategy to exactly get that the causal effect of concentration on investment so if you're interested then I can just recommend this paper. Thank you Thomas, before I open the floor does any of you want to come back or to comment on two cross comment on each other's paper? Just one, just one short comment. I think that the increase in intangibles also has some roots in the harmonization of these systems that I mentioned at the end of my presentation. It makes it much easier for businesses to go international to expand brands, patent families and other things and that adds of course to the cost of doing so and that's what you kept sharing the data. One question if I may, I wanted to ask you Thomas is about uncertainty which features in your paper but something we know about investment dynamics and in the Tobin's Q literature and Jadis you've written about it is that investment can be lumpy, adjustment costs can be non-linear, I mean marginal adjustment costs can be non-linear which can go somewhere towards explaining why firms are not investing and the more uncertainty, the more the value of waiting that's something we've learned. So I mean would that fit into the story that there would be either more lumpiness because it's when coming with intangible investment in the US or maybe more uncertainty in Europe holding back investments which is not what you have in your conclusion but that's something that you could imagine is happening that uncertainty would be also would be creating some value of waiting. Yeah, so absolutely. There's no question that you could have a value of delay if you have more uncertainty. My sense from looking at the data is that political uncertainty uncertainty about the future of the Eurozone was a huge factor in Europe. My sense is it's fully priced in Tobin's Q. So one of the reasons Tobin's Q works so well is because every time you have a spike in uncertainty about a country or development in the Eurozone you see Q going down. So I think it's working pretty well in that. And that by the way is exactly what the theory would predict because uncertainty is just one type of risk premium and every single type of risk premium works in Q. The adjustment cost theory is a bit different so about how fast can the industry pick up after the risk premium go down. So that's more like, yeah, if we believe there are these strong nonlinearity it might take more time than we think for investment to pick up in Europe. For the US I just don't buy the uncertainty story at all, period. No, we can debate it but that's my view. And the other thing is the more you believe in intangible the less you should believe in uncertainty because the thing with intangible investment especially in human capital is there is no gap between the timing of investment and that of cash flow. You pay people as you go along when to write software. So in that world uncertainty is the second order issue so I don't need to explain anything in the US. I think uncertainty in the US is shortcut from tax cut. Okay, thank you. Thank you, Thomas. Let me open the floor for discussion. Please, so we have 20 minutes or so. Let me remind you that this discussion is webcasted. So all of you may more or less know each other in the room. This is also for the benefit of all those who are watching us. So please introduce yourself. So, Kristin. So the mic is coming. So I will take like two, three questions and then hand it back to the speakers. Kristin Forbes from MIT. So two very thought provoking papers. I have one question for each. So first, David, you focus on the effect of productivity growth on labor market quantities on employment. Have you also done any work when you look at the impact on wages? For example, in the sectors that have higher productivity growth, do you also get some of that in the form of higher wage growth, even if less employment? And then what happens to the labor share in those sectors or do most of the gains go to capital? Thomas and actually Janice also. Neither of you mentioned in all of your comments the fact that US companies have roughly two and a half trillion dollars of cash kept abroad, at least some of which is probably held there because companies are hoping to see some sort of temporary permanent reduction in the tax rate for when they bring that money back into the US. Could that affect any of your estimates of why some of that is not being used for investment or affect your estimates of towing queue and it will affect some of your end results? So Richard, please. Richard Baldwin, President of the Center for Economic Policy Research. David, I was a little surprised not to hear the word China in your entire presentation. International Factors, are the fixed effects picking up the China shock or in particular what I worried is that the very rapid drop in the price of manufacturing due to China's emergence is getting into your productivity numbers and thus you're ascribing some things to productivity which you probably ascribe to China. And it's this basic idea that American workers are competing against China abroad and robots at home and it's very difficult to sort out which is which and I'd just like you to come on that. Thank you, Sylvester. Yeah, Benewa. I have one question to David and one remark to Thomas. My question to David is the following. Well, you showed that it's not only lower my middle classes but it's also coming to the professionals including central bankers. And what you see nowadays that with artificial intelligence that FinTech, order tech, legal tech and if you see the dropping of numbers or corporate finance members, auditors, et cetera. It's hard. So that means that we have to educate. I'm Sylvester Ivory, I'm a senior manager at Tilbury University so we confronted with that. Five to 10 years we have to have a different educational system including data science and artificial intelligence for all our professionals. So that's my question. What do you think is it the rather five to 10 years? My remark to Thomas has to do with one explanation which I was missing. That's the difference between, in the total skew explanation, difference between the stakeholder model which is typical for Europe and I know although maybe you can call it and the shareholder model which is typical for the US. You see nowadays that we are confronted in Europe especially in the Netherlands with a loss of hostile takeover by private equity firms. Unilever, Aksu, Nobel, we can continue and why because they are focused on value creation in the long run in servicing their stakeholders which is completely different model way like in the US and especially with private equity firms. So Thomas maybe, well you could include that in one of your intangibles in your, it's very difficult but maybe you can give your opinion about it. So let me take one last question for this round and there will be another round, Marco Boutier. Thank you. Thank you very much. I have a question across papers. Who relevant results in the first place? You have in David finds lower spillovers in the recent years and you have in Tuma higher concentration in the US, especially in the US. Now, can we have a relation between these two? I mean if you have higher concentration, industry concentration, you have more appropriation of the productivity gains so possibly less lowers spillovers, going hand in hand with lower level of investment. So does this story convince you? And the second point then would be immediately following, is that if it is true that the good news for Europe is that according to Tuma there is lower, let's say investment due to cyclical conditions rather than structural conditions, maybe once these cyclical conditions are over and hopefully that is going to be the case soon, then you may have more positive spillovers in Europe. So maybe you have a virtuous cycle there. Is it daydreaming? Okay, back to the speakers. David, do you want to start? Sure, should I try to answer all five questions? Yep, whichever, whichever. 42. So first on the question of wage bill and labor share. So we didn't in this paper look at the wage bill component. It's a valid question. But it follows naturally. We know that the relative wage of skilled workers have been rising. We know that there's not a within industry shift towards skilled workers in the sectors have greater productivity, but we know the wage bill share must be rising in aggregate and that partly is a function of the cross sectoral shifts. So it definitely our story where this unbalanced growth is raising demand for skilled workers would be one where the wage bill share of skilled workers is rising. You asked about the relationship that to labor share and this is something that many are working on myself included looking at the change in concentration and what we know in aggregate that labor share has been falling across the developed and developing world, even China. It is associated with rising industry concentration and we conjecture that also the same phenomenon of what we call superstar firms, which are firms that are capital intensive and highly profitable and they command a growing share of output that that is also related to the concentration of wages. And that's a conjecture. It's not something we confirm the work by Jay Song and Nick Bloom and others shows this sort of the between firm component of rising wage inequality also visible in Germany, high wage firms, highly skilled workers increasing covariance between them. We hypothesize that this mechanism is leading to wage concentration. And so it's important to distinguish that I said jokingly about the falling labor share being the first horsemen of the Robocalypse. It's not the case that labor share is falling on average across firms and unweighted average. It's that the firms larger firms or profitable firms have lower labor share and their weight in aggregate output is rising. And that's the key contributor. The question on China. So I didn't mention that because our data don't directly speak to that but of course that has been one of the key labor market disruptions for the US particularly in the 2000s. And I think it's contributed a lot to our fall employment to population ratios. You might say, well, why hasn't the same thing happened in Europe? Well, to a substantial degree Chinese goods in Europe substituted for imports that were coming from elsewhere already that many of those manufacturers were already not being produced in Europe. The US being so large and actually being a relatively low skill country by the advanced country standards produced a lot more textiles and leather goods and dolls than has been produced here in quite a while. But obviously China is extremely important and that also raises the question when we think about the declining spillovers that could partly have to do with import substitution. And that's something that's on our agenda to look at. And even the productivity gains could, as you said, be exaggerated by import competition. This is something that Sue Hausman has worked on a lot to show that some of what looks like productivity gains are actually just substitution of imported products that are not correctly accounted for in national accounts. On the question about central bankers being automated, I was amused by the friar and Osborne to say that apparently clergy also have a 20% chance of being automated. I was wondering how that works. It's low, but still it's not zero. So clearly, that's some automation. Because productivity hasn't been rising at clerical work as far as I can tell. The bandwidth of the deity is pretty stable. So I think what appears to be the case from a lot of data that I've seen actually is not the rising importance of technical skills, but it's actually interpersonal skills, problem-solving skills, managerial skills, interactive skills. Work from Sweden shows that there is rising return to non-cognitive interactive traits. And in the US, it's not the case that the highest tech occupations, the pure mass science occupations are growing. It's the jobs that combine management with technical expertise. So I actually think human comparative advantage lies in expertise, judgment, and creativity. And those, if we just wanna be better calculators, we're gonna, that's a race we're not gonna win. It's really our comparative advantage is the flexibility and the ability to bring both our refined expertise, but also our panoply of insights and our sensitivity to direct situations. So I think it's not that education needs to be just stemmier and stemmier. It needs to be broad and allow people to be adaptive. So let me end up quickly. So this question about spillovers, again, I think the China component may be important part of that. We really need to do the input-output linkages on that. I think it's also, and on the final question about the relationship with concentration and spillovers, I think that's quite plausible that in other words, rising concentration means that we have sort of superstar firms that are very profitable. They're not investing strongly. They're holding onto a lot of cash and that would tend to reduce the spillovers. So I don't wanna offer a grand unifying theory, but I do think this nexus between concentration of economic activity, which may not just be antitrust, by the way, which may be that information-intensive goods, improved search, give strong market dominance to firms with a cost advantage or a technological advantage. They have effects not just on the share of labor, but where the investment money goes and the investment output goes and also which sets of workers benefit. And I think that remains a very first-order topic that I'm hard to see many researchers are exploring making progress on. Thank you. Dietmar, do you want to say something? I just wanted to make one comment that is related to what David just said. If we have an increasing importance of a financing mechanism via venture capital and so forth, then we also have a selection of business models into scalability. That's the first thing venture capitalists ask you, is this scalable? That means trying to get away from high labor shares. So I would like to see your, the latest paper, the super firm paper, broken out with respect to financing form and sort of the origin of the company because I would expect some selectivity in there that may not be as strong here in Europe as we just built these sectors of financing. Just one quick comment to that. The irony is in the finance sector, labor share is rising. You might think it'd be just the opposite, right? All the automation, but in fact, a lot of the rents are accruing to workers and the importance of sort of sales and personal services is rising. I'm not saying that's a healthy development, but it's the opposite of what intuition would suggest where you would just think, well, it's all robo-investors and lights out. But if you need expertise and judgment and creativity to do these deals and find out which teams should be financed, then that will be in line with what you said before. Okay, so quickly then. So Christine's question on, yeah, so I guess there are two parts. What is a measurement issue? And when we define it to be skewed, you net out financial assets and liabilities. So you don't pick up any kind of excess cash they have as part of the valuation. But there's a deeper issue, of course, of firms operating abroad. So when we measure investment in the national accounts, in the industry account, at the firm level, they don't represent the same thing. Sometimes it's measured domestic investment. Sometimes it's firm consolidated. So that's why we do both and we check that none of the results we emphasize is driven by firms that say invest only abroad. And the last point that's very important, which again I didn't have time to do justice to, is when we look at Herfindoll indexes, then there's a big issue when competition comes from abroad. And so we've created now, almost finished completing a new set of Herfindoll measures that actually adjust for fine competition, okay? Otherwise you get nonsensical results. So all of that is taken care of. China, of course, which are, is not just for the labor share, it's big time for investment. And in fact, in the paper that I mentioned earlier, China is one of the instruments we use to look at an exogenous shift in competition. Shareholders and stakeholders, Sebastian's question. Yes, big time. In fact, we have a paper specifically on that, on the US, because, but I can't give you the details, but the one thing that's important in the US, which is tend to be a bit more US specific, is when the typical measure of Herfindoll is defined by firms. But if two firms are the same owner, then should you count them as one firm or two firms? Now you might think it's not a big deal because of this person in shareholders, except that in the US, there's rising concentration of large money managers. So now, oftentimes in an industry, among the top five firms, you're gonna have large money managers owning 20% of each. Are they really gonna compete that much? That's an open question. It turns out in the data, the answer is that it has a strong impact. And if you build a Herfindoll index that adjusts for this overlap, actually, first, it predicts better what's actually going on in terms of investment. And two, of course, it's concentration is rising even faster. And so then I tend to think, I tend to think more in terms of the horizon of the shareholders and what factors could be pushing towards short-termism versus non-short-termism. And that's a very complicated question. But we do find some evidence of increased short-termism. I'm not sure it's all private equity. I think it's also, well, I guess it depends when you put the hedge funds in there. But activists, you know. So, yes, I think it's there. And we don't see that much of that in Europe so far. And Marko's point, I agree with what David said, that, yeah, I think there's probably a link. And in fact, he connects to what John was saying about the labor share. I mean, you know, it's in equilibrium, you know, the market is peer or W and then the labor share is W over P. So if one goes up, the other one goes down. Is that as simple as that? So for sure, there is a link between the two. I think the tricky question is to understand changes in the technology at the same time. Because you have two things shifting at the same time. You have the competitive capital share, which moving because of intangible investments. And then you have the credit market rent. And we observe the review of the two and the challenge is to separate them. Then? Just a quick comment. I completely agree with Thomas's response on the cash held abroad. I think it's well accounted for in their methodology. On the shareholders versus stakeholders question, it comes out very dramatically in the chart I showed on net business savings. Where you see the accumulation of net business savings and as Thomas's paper showed, one allocation of that is to share repurchases and dividend payouts. And so that likely has much to do with the point that you emphasize, but it's very dramatic in the data for the US. The last point on how these things tie together with concentration, I think is a very interesting one. But it could go in many ways because you could see that concentration might limit the spillovers across firms. So if there's less dynamism in an industry because there's more monopoly power, it could play the role of limiting the kind of spillovers and why we see that reduction that you report in the 2000s. On the other hand, if the concentration is resulting from acquisitions and so you're seeing consolidation and note that Q theory is silent on whether firms invest by new capital investment or by acquisitions either could be the result of high Q. So if the result of high Q is that there's acquisitions, then that might be a way in which the spillovers are actually implemented across firms is you take the low Q less dynamic firm is incorporated into the high Q more dynamic firm. So I think that's an empirical question that's outstanding. Thank you, Janice. We are going to close it now because we run out of time. So I apologize to a lot of you who wanted to ask more questions. I'm sorry about that. You have a chance to ask questions in the next session. So that's a legacy I hand over to Peter. I think we've done a very good job at sowing the seeds for the policy discussion that which Peter will lead in a moment after the coffee break. And we've also made the case that this conference cannot be automated, definitely. So thank you very much.