 So, good morning everyone, welcome you to lecture number five, of course, Collective Dynamics of Firms. I would argue that we have not reached the critical mass here, right? So, I mean, it's a bit, although up to you. If you have an interest at IT2, then you may want to show this by your personal attendance. If you don't show it by your personal attendance, then what can I do, right? The only thing I can recommend is, maybe you want to watch a video of last year, or you want to read the slides yourself or something. Why should I waste my time talking to you, right? I mean, this, of course, is addressed to the wrong people because you are here, but I assume who's from MTech? So, okay, why don't you talk to the other MTech students about this? That would be my recommendation, right? You sit and say, okay, well, we have two options now. This professor cancels his commitment and then we have to see how we get the credit points or we team up and just attend the course, right? These are the two options, but if you just leave it like this, then there will be no improvement, of course. So, what's your comment on this? Yeah? Okay, yeah, there's nothing against regular economic courses, but what I do not understand is, so there are 19 people subscribed to this course, right? That means I, if I go and look into the number of registered students, or then I would assume I should teach, right? This is a good question, why do I record the videos? Because my personal feeling is that it largely improves the grades. Basically, this is a service to you. So, I mean, you do not recognize it now, and you in particular are not because you are here, but then when you prepare for the exam, right? So then you have no clue what was the message of slide number, and then the nice thing is you go to these videos, there's a slider that shows you precisely the slide, and then you go there, and then you can listen to my comments. That's why we do it, because we learn that people, by having this additional possibility to repeat the lecture, do better in the exams. That's the reason for this. I mean, I also understand that many other professors from the M-Tech, but also other professors I meet in ETH are against this, because that's the effect, right? I'm not sure how we handle this. Yeah, but you see, I know in economics it's all about incentives and the right incentive, okay? I think the only incentive I accept is, that's an interesting course. I would like to learn about it, right? What other incentives should I set? Should I pay you for coming here or what? Okay, I have a complete different opinion on this. I think you are here because you like the course and you are interested in the topic, right? That's the issue. So let me do it like this. I define the critical math of this course as five people. That's something I will also send out today. If there are less than five attendance, we skip it, right? Yeah, I don't know. I mean, then it's up to the students to inform me, okay, now we are ready to come to the lecture. We will still upload the slides because we have prepared all the material. We will upload the literature, the self-study talks, everything, but I'm not here available to teach. Then you can, if you really want to see the course, then you may want to pick a video from last year, but just as a word of warning, so we improve the course every year. There is no clear relation between the material we hand out this year and the video we have recorded last year because every year, even if the student does not assume this, we improve the course. We sit together with the student assistant, with the assistant of the course, which is probably, and improve the slides, right? That means you may find material that was not covered last year and so on. Then it's all up to you. I will send out this email today, and then you can coordinate, right? If there are no five people attending this lecture, no one will stay here and give the presentation, right? No? It's really up to you. Okay. You might still be able to do the exam at the end, of course. But then by means of your own self-study effort. Okay. Let me conclude what we did so far. So today we are finishing our empirical part and we'll start with a modeling from next week on. That means by today we have learned all the phenomena that we would like to describe in terms of models and then next week we try to find out what kind of agent-based models do we need to reproduce the style that's fact we have learned about. We will do the step-by-step. We start with a very simple model and then we recognize what needs to be added in order to capture it. So you learned about the stylized facts of distribution. There was one related to the size distribution. It's a skew distribution where we were not specific. You remember why we were not specific because different data sets argue for different skew distributions. This is not a mistake. That is reality. Second, distribution was about the growth rate. We were very specific because all the available data points into the same direction. I recall that even the growth rates of GDP or of university expenditure on R&D was following these growth rates. That's a remarkable fact. Then we also talked about parameter dependencies. There was stylized fact number three, which told us that the growth volatility depends on the size. This is not to be mixed up with the growth rate. That's a very important thing. If you have to recapitulate the stylized fact in the exam, then you hopefully notice the difference. Today we talk about two other stylized facts. The first one is about dependency on aid, and the second one is related to entry and exit dynamics of firms. Let me go to the first slide that addresses the age dependency. If someone will close the door, I don't mind. You have enough imagination to guess that there is an age dependency on the growth rate. This is via the size. Usually firms start with a small size, and they grow in population dynamics. If you are young as a firm, then you have a smaller size. If you are older, if you have survived, then you have a bigger size in some sense. That means we can guess already about this. What the stylized fact number four tells us is that the older the firm is, the smaller the growth rate is. This is not only a size-dependent effect. You could then, of course, say, how is age correlated with size, and then you map it down to the size, but that's wrong. Why is that wrong? I've written it here. The growth rate does not depend on size. But the growth rate depends on age. We cannot simply substitute it. Even if we have a logical argument that there is this dependency, we cannot simply substitute age by size. Because the growth rate does not depend on size. But it depends on age. That's a different place. In some sense, but you can be whatever five-man company for 20 years. This is just your optimal size. You are consulting rich clients about how to invest their money. Five people are just enough to do this. There is no need for you to grow over time. That's what we say. If you have not grown over time, then this stylized fact tells us that you will likely not grow in the future. Because the older you are, the more likely you have already reached an optimal size and then stays at that. You could say it like this, that's correct. Because what we said on the previous slide is that we do not see the mean growth rate depending on the size. But for individual firms, there might be correlations. These correlations do not really point into a direction that we can call this a stylized fact. You've got the point here. I'm not going to discuss this in more detail here. Instead, we're looking into something differently. Namely, how does the distribution of growth rate change over time? We learned that the distribution of growth rate is a Laplacian distribution. This was confirmed only for the one-year growth rate. I did not mention this explicitly, but that's an underlying assumption. It holds for one year. But if you look now into larger time intervals, how will the growth rate distribution change? Will we still find a Laplacian distribution? We can see it here. The thick black line is the one-year growth rate. You see it here. This is the number of years. Then you see how the shape of the distribution changes if we go for larger time intervals, like 16 years growth rate is the maximum here. This depends on the data sets that we have available. What we notice here is, of course, that the picture becomes a lot more weak. We cannot clearly identify a double-exponential Laplacian distribution if we go for larger times, because it's very broad. No one really knows. Please recall, a double-exponential distribution in this plot has to be a straight line, like this. You see that this already does not hold for smaller firms. That's something we have already discussed last time, because the volatility depends on the size. That's the average here. You have a huge fluctuation here. There is a straight line, at least in the vicinity of the mean. But this here is no longer a straight line, clearly not. That's what you see here. You also see these bends here. The second question was how we found a dependency of the volatility on the size. I hope that everyone is able to write this down in a very simple equation. The volatility is proportional to the size, to the power of minus B or B. Then we talked lengthily about the fact that the B is obviously 0.2 for different growth processes. Oh, sorry. No, that was the wrong direction. Here we go. We recall from the previous lecture that in this plot this has to be a straight line, where the slope is equal to the beta. Here you see that the beta depends on time. This one is a log-log plot. This one? Here, I mean, sigma proportional to x to the minus beta, that's the straight line here. Are you talking about this picture or that picture? About this picture. It's a double exponential distribution. Of course, it's a real straight line, only if we go for a log-normal plot, where this axis is a log plot and this is a normal plot. This is not a log-log plot, that's a log-normal plot. But this is a log-log plot. The picture you are referring to. That's a log-normal plot, growth rates. That's the log axis. The semi-log plot, yes. As we discussed last week. What you see here is the beta depends on time. If you go for larger times, then the beta is no longer 0.2, but it decreases. Which is equivalent also to the picture on this side. Here, you see if we do the scaling, we still get a nicely scaled distribution. That means the scaling relation that we were talking about last week still holds. You see this here nicely fitting towards the master curve. Now, let us think of the growth volatility, dependent on size but dependent on time. This basically is how does the sigma depend across time. This is for five different time periods. That is addressed here. Here are the five different time periods. Let us first look into the thick black line but that's the smallest size over time. You see here, this is clearly not a linear curve. In fact, what it is, it's something like proportional to square root of t, approximately. Whereas for the large firms, these are the larger size classes, and this dependency how the sigma squared develops over time is a linear relationship. What does it mean that this linear relationship does not hold for smaller firms? It means that in particular, if you look here into this non-linear part in the beginning, it points to some anti-correlations, which is very clear. If you are a small firm and you grow in one year by 50%, this is not a surprise because if you have four people or three people and you hire one more, then you easily reach this kind of growth rate. 25, 50% is not a surprise. But because you are a small firm, you can probably not keep this growth rate for the next year and the next year. You will probably not hire the next, but the upper next year and so on. That means there is what we see in anti-correlation. If you hire this year as a small firm, the likelihood that you also hire next year is smaller. The same for me. I run a small company, a share. I run a company, but if I look into my share, there are like 15, 20 people. Of course, if I hire a postdoc this year or two postdocs this year, the likelihood that I hire two postdocs next year is much smaller. But then, if time passes on, then of course after three, four years, the likelihood that I hire one or two new postdocs is of course increasing. That's what it says here for this, but for large firms, we do not see this. We see that Sigma square, the volatility of the growth rate, simply almost linearly increases time. Again, this picture refers to this picture here. And because we understood the time dependence here, we were able to scale this to the master curve. That's somehow the visual proof that we understood how the different variables depend on each other. Okay, now let me come to the second important topic of this lecture. It's about entry and exit dynamics. This is actually one of the most important issues in the collective dynamics of companies. One issue is about the performance of established companies, but another even equally important issue is the number of firms constant. Is it decreasing? Is it increasing over time? Entry means new firms enter the market. Exit means firms are dying. They leave the market. They can die on various reasons. There can be a merger, for example. Company A and Company B put merge in order to form a larger company C. Then I see in my statistics two firms died and one firm was born. Or your big company buys a small company. Then I see one company disappears from the market and another company had a huge growth rate. If I just look into growth dynamics, I cannot really distinguish these events. Therefore it makes sense to look into the independent statistics. We are not talking about merger and acquisition data here. That's probably covered in other courses. We just talk about entry and exit means a number of firms entering the market and exiting the market. So here on this slide, we have defined the entry and exit rate. These again are ratios. Why do I mention this? Because a rate, for me, depends on my education. Of course, a rate is number of events per time interval. But what they call rate here is in fact the ratio. The percentage of firms that enter the market is given that there are already N firms in the market at time t. The exit rate is defined accordingly. What we usually then observe is the number of firms at the time t plus one and the number of firms at a time t. This, of course, is a net rate, entry and exit. In most cases, if we see that's a number of changes, that's the number of firms' changes, then we see the difference is a net rate. From the data, we cannot tell how many firms have entered the market and how many firms have exiting market. We can only tell that the net rate is either positive or negative. But going to data, census data, for example, we also know precisely these numbers. If I see the difference is one, then it can be 100 minus 99, but it can also be 10 minus 9. We need to know these numbers in order to get a better idea about the dynamics. The net rate by itself has not the full information. Let me first argue a bit with you about the meaning of entry and exit rate. We also come to this in a few other slides where we talk about Mr. Schumpeter later. This entry and exit dynamics is also dubbed as turnover. I think there are other courses, at least for the MTech students, which cover this phenomenon of turnover. Did you enter any of these other courses? You already talked about turnover rate. That means you have already an idea. You can see the turnover process as a selection process. If new firms enter the market, then one assumption could be they have a new product, a new technology, they then propose to the market, which will probably survive, and firms who are already established and not adaptive enough, they have to leave this. That means turnover is good because the innovation then propagates in the market and all those firms who are not able to adapt to the innovation are simply kicked out. This is one argument that is usually used. People think turnover rate is good. We will look into this. Entry and exit are two different processes. There is some correlation, but there is also some anti-correlation. The rest of these slides are exactly talking about this. Here I am talking about turnover as a general phenomenon. Usually people think turnover is good because they put some pressure, some selection pressure on the market and badly performing firms or firms who are not adaptive enough or kicked out and new firms get a new chance. But we can go a bit further, and that is probably all that is discussed in this lecture about business dynamics. We can see turnover as an indicator of a decline in the firm performance. There is a saying about the reds that leave the sinking ship. Have you heard about this? That is a German proverb. The reds are leaving the sinking ship. When the reds march out, it is very clear that the ship is about to sink. That is something you see here as well. If the turnover in managers increases, then there is a strong correlation to the fact that firms are dying. You can even follow this in the NSATZ when they tell you that the famous CEO left the so-and-so company. The general conclusion is that there is something wrong. Maybe there are strong tensions, maybe there are other issues, and of course this person will not ruin his or her reputation. They better go early and leave it to the others to manage the closure of the firm. All these kind of things. That means there is a little bit more information involved in this turnover data. What we also have to say is that there is no clear pattern about entry and exit rates. It really depends on the different industries. It depends on general dynamics, on business cycles and other things. Sometimes you argue that it is related to profitability. The firm had to leave the market because it was no longer profitable. That can be an argument, but the entry rate is usually covering firms that are never profitable. That's something I think we put on the next slide here. You cannot correlate the entry rate with the profitability. That's what I wanted to say. Because new firms are likely never profitable. There is no clear correlation between entry and profitability, but there is a clear correlation between entry and innovation rate. That is what we are also going to discuss with Mr. Schumpeter's theory afterwards. Here are a few arguments that I gave for this. We assume that the new firms are the carriers of some innovations. If you have a large number of new firms entering the market, you have a high innovation rate. This is not true, as we will see afterwards. There is a positive correlation, but there is not the conclusion that the new firms have to be the source of innovation. You see that's an issue of interpretation. Although established firms can be the source of innovation, it doesn't exclude this because of this correlation. That's the important thing. You got it. Here is some data from Switzerland. I copied it from a newspaper. It could be also in parallel. I want to say established firms can innovate and new firms entering the market can innovate. If I have a large entry rate, I usually have a large innovation rate because new firms are entering the market, but it doesn't mean that the old firms never do anything. That's the conclusion. Let's look into this data. It's of interest. If you want to launch a startup company, maybe you are thinking about this, then you can have an office in Technopark. As far as I'm told, Technopark is always booked out. It means all the new firms have difficulties to find nice places. Here we see that half of all new firms only reach the five-year limit. That means half of all firms die. Here we see also how they die. After the first year, 80% are still in the market. That means 20% have died, 30% have died in the second year. Then 35% total in the third year and 40% in the fourth year. Again, another 10% or 11% to be precise in the fifth year. This indicates how difficult it is to establish a firm in the market. There is, of course, a strong dependency according to the different sectors. We see, for example, if you talk about consulting or if you want to run an insurance company or something like this, you are a broker for insurances, then there is a small entry barrier only. That means we see lots of companies established this way. We also see lots of companies that leave this way. Here, if you go for another sector like construction area, you have to invest a lot initially. That means the entry barrier is quite high. This means that once you have got the money together to establish a construction company, you are in a better situation. Not 50% of these companies will die, it is a bit lower. The last bullet point here tells us about the importance of the new firms. Usually politicians tend to think in terms of employment or unemployment, which implicitly affects tax income and all these kinds of things. Here you see, if you are able to establish a firm that survives in the market, then usually these firms have strong growth rates. Employment grows at 25% after three years and at 52% after five years. If you survive, then you have already put 50% more of your initial size in terms of employees. Therefore it is important to have new firms, although in terms of having new jobs, that is the message of the last bullet point. Then this year, different from last year, we have some more tables included. I don't think that we had them last year, but I'm not so sure. This is the usual way, econometrics places the results. We have to learn how to deal with this. I'm not particularly fond of looking into big tables, but that's simply part of the research tradition in this area. We have to look into this. In case you want to know the details, in the notes there is the paper and you find the paper also when you go to the Moodle platform under literature. If you want to read precisely about the interpretation of these numbers, please go and look into this. I'm not able to cover all the details of this table here, but I also do not want to avoid this table. We look into a few interesting things here. What Mr. Dunn did here was comparing entry rates and exit rates here, and he had a data set available that refers to the US manufacturing industries. This is data from the US manufacturing industries over different time intervals, which is quite useful. It refers to the US different time intervals here. Let us first look into all firms and smallest firms deleted. You see that this is basically a percentage, but the most important thing is the difference. The smallest firms are defined, I think it's written in the note here, and those who together produce one percent of the number of the industries output. Let's assume 10 firms enter the market. Nine of these are one-man companies and one is a company of 20 people. Then Mr. Dunn said, I probably calculate the entry and the exit rate wrong if I give the one-man company the same weight as the 10 or 20-man company. Therefore, he compared these two numbers. He didn't say you should exclude this. He said, what's the effect of neglecting this? That's a different thing. He doesn't manipulate the data as some people might want to do. He presents this and says, that's the effect. You can decide yourself whether you want to have the one-man company in or not. Here you see the impact in the entry and the exit rate. My suggestion is that we drop the numbers of all firms and just look into smallest firms deleted, because this gives us a better feeling of the impact in the industry. Here you also see the market share. In the market share, these small firms don't do a lot. All small firms together have only 0.03 percent impact. Therefore, it's probably reasonable to leave them out. That's the first thing that we want to see. Then the second one, if you just look for these, then you see here for the time interval between 67 and 82, it's almost a more or less constant entry rate. Please recall that this is for five years. 40 percent that you see is for five years. You want to get the annual rate and you divide by five. That's the first important thing. Here by the way, it's only three years. It means four years. This is four years, this is five years. Am I right? Five, five, five and four. Before we argue what happens in 1963, we better look into the time interval. Then compare it. You see it's almost constant over time. That's an interesting thing. The market share is also almost comparable to the new firms. The exit rate is also almost comparable if you see it here. If you compare these numbers with these numbers, you see the US manufacturing industry has steadily grown because in all these time intervals, you see the entry rate exceeds the exit rate. You can guess the net rate as positive and then you see the growth of this. If you look a bit further into the impact of the new firms coming to the markets, they compare the market share and the relative size. That is given here in these two numbers, market share and relative size. Let us just look into the exit rate of the relative size. Here, of course, you see that the number of small firms generates some impact if you compare these relative size here and relative size there. Because you have so many small firms, one of my companies, for example, then of course the average relative size becomes very small because you have only few large firms and many small firms. The relative size of these firms is quite small. That is wrong. If you neglect all the one-man companies, for example, then you see that the relative size of the market share or the market share is quite considerable. What you can see here is that where is the market share here? It is about 13% or 18% of the market share that is occupied by the new firms. I forget about the one-man companies. This is quite a remarkable number. You can divide it, of course, the time interval, but still it is a large number. That means new firms have not the lion share on the market, but they have a considerable number. You can even see it here when they drop out. Of course, they have increased their market share because they have grown in this time period maybe. Here, when you look into the relative size, you cannot ignore the part of the market that is occupied by firms dropping out or by new firms entering the market. This is what the number tells us. Yes, please. The relative size is the total number of firms. It is the number of new firms entering the market, 100, divided by the total number of firms, 1000, then we have 10%. That's the relative size. It's a percentage of what accounts for the new firms. That's the idea. Of course, you can have a small firm, but a big market share. Therefore, the market share is not identical to the relative size. The market share measures your impact on the market, whereas the relative size just measures how much the firms occupy in numbers. That's the idea. What do you want to say? You are arguing that you want to see larger numbers here, or what do you want to say? What does it have to do with the crisis? You would assume there is a crisis and more firms drop out. Is this what you want to say? Let's go back to this in a later slide, actually. There is an underlying business cycle, and you see clearly the correlation to the business cycle. There are even slides that I show later on this. The response is not that immediate. You have to see what firms we are talking about. We are talking about the US manufacturing sector. If there is an oil crisis, then you do not just lock the door and leave. There is a period in which you respond to the oil crisis before you have to exit the market. There is not a one-to-one correlation between these kinds of events. We will look into this later on. Let us have a 10-minute break here, and then we continue. The lecture. As I said, we have some tables included here. A good question for the exam is for someone that you are presented some part of such a table and that you should draw some conclusions of it. Just to give you a hint on the fact that you should learn to interpret these numbers yourself. This is also the reason why we give you the literature. In case you have questions, then you can look up what was the intention or the interpretation given by the authors of that study. Let me go now to a bit more of theory. I already mentioned Mr. Schumpeter's theory, which is behind the dynamics of entry and exit rate. I think I only have like three or four pictures of persons here in this lecture, and these are always very famous ones. Mr. Schumpeter was a real famous and important economist. He did a lot of work in economics, which we do not want to discuss here. He also thought about the future of democratic societies. Will this be maintained? What's the future of capitalism? He had a very broad intention. What we are interested in here are these contributions to the theory of firms and to entrepreneurship. He developed two theories, which are dubbed as Mark I and Mark II. Mark I was developed while he was in Europe and Mark II was developed while he was in the US, and both of them point into opposite directions. That's the interesting thing. In Europe, in this theory called Mark I, Schumpeter put forward the argument that innovation is driven by new companies entering the market. He referred a lot to this entrepreneurial spirit in German Unternehmergeister, which is important to get some dynamics in the economy. The new firms enter the market, the challenges, the established firms. There's a competition like David and Goliath. Then David Wins and Goliath have to leave. That's the underlying idea. Then he went to the US, he became a professor there, and he completely changed his mind. He said, no new firms may have good ideas, but they do not have the resources to develop these good ideas. The resources have the big company. It's great that you have this idea about audio compression, but only a company like Sony is able to make something out of this in order to affect the market and to change the product landscape. This was the second idea. He said, capital is important, equipment is important, and therefore the big companies with their big research and development divisions are the most responsible ones. Today we do not look at this as either or, but as well as. Both of these theories hold, dependent on what sector of the market you look at and what data you have available. Here's a bit more detail on his hypothesis of creative destruction. That's a nice buzzword, independent of this particular notion. You destroy something in order to develop room for something new. There's always something good if a whole industry crashes or is teared down. That's the idea. Here, in particular, as I said, he assumes that the new firms are the sources of innovation, and the old firms cannot cope with the pace of development because they are less efficient. They are too large to adapt to something new before they get the point that they are already behind everything. That was the idea. Of course, I mean, this has a lot of impact in the industry, if, for example, some technology changes. That was also his argument. The example that he took was, there was a whole industry of horse carriages. Then the automobile was invented and then the whole industry of horse carriages became completely obsolete and the whole market was taken over by automobiles. This turmoil was then dealt with the term market turbulence. Market turbulence has two meanings here. Market turbulence can be related to the end-to-end exodynamics of the firms, but it is also related to the fact in which firms change the market share. If you think of the mobile communication market, then there is a market turbulence which can be seen by the fact that a firm like Nokia who has ruled large part of the market now is heavily decreasing in market share. That's also a market turbulence. You cannot trace this down to a new company entering the market, but you can certainly trace it down to a big company supporting a small company like Google with the Android phones and so on. There's a clear relation. It's not always that you can identify the small company that now had a cascade established to turn over the market. Here we have another view on entry and exit dynamics. It shows us the end-to-end exit rates for different countries in Europe in the time from 1990 to 2000. Here you can look into established economies which are characterized by lower entry rates, but also lower exit rates. That's an established mature economy like Western Germany, but also Italy or Denmark or whatever. There are on the other end countries that made a major transition in their political regime, from socialist country into a country of free market capitalism. We're not arguing whether this transition was good or bad. That's not the point. What we are arguing is that this political transition is reflected in a huge number of new firms established simply because before the economic conditions and also the legal conditions to run your own company were much worse. That is what the story behind it is. Now, because you have lowered the legal entry per year, so everyone could find a company if people had the power to do so or if they wanted to spend the effort on this, you can reflect it. At the same time, you see that for the established companies exit rate is almost equal to the entry rate. That means there's not a huge net rate growth in the economy. Here it's very different. What can we conclude from this? The market was empty. That's the only conclusion we can do. Therefore, very many companies could enter the market and very few had to leave. This picture of course changes when you really come to the period of strong competition. Then, of course, the free market. Capitalism between, by the way, all the models then puts a lot of effort on all these established companies. Then you will see that, of course, in the long term the exit rate increases and then balances the entry rate if these companies come to a saturated growth. Now let us put this into Starlight's fact. What is the hard fact or the Starlight's fact that we can really deduce from these entry and exit numbers? It's a very weak statement that we can make only. There is a positive correlation between entry rate and exit rate. That's not too much what we claim here. If very many firms enter the market then very many firms exit the market. That's the only thing we say. This keeps the economic growth bounded. Can you imagine what it would mean if there was an anti-correlation? If very many firms exit the market, very few firms leave the market then it basically means that the economy is exploding. You would not have enough workforce, you would not have enough capital and so on. This has a number of implications that we have marked here on the slide. We have already seen that the survival rate of the new firms entering the market is quite low. Remember that 50% in Switzerland after five years that is a low survival rate of new firms entering the market. Therefore you have this positive correlation because most of them do not make it. That's the implication of this. There are further correlations towards the business cycle. Of course there is also a correlation towards the product life cycle. Do you still remember what the product life cycle was? Maybe from the core systems dynamics and complexity? You don't remember it. Can you explain to us what the product life cycle was? More or less like this. There is a life cycle if you have the iPhone, a new product. In the beginning you have a strong growth because everyone wants to buy this product. Then you get a saturation and then instead of the iPhone the beta phone is invented. The iPhone dies and the beta phone is taken over. That's the idea. If you are in the early phase of the iPhone and then you are the company who writes all the nice apps for this, then you grow of course. But if you enter the market with your iPhone and with your Apple App business at a time when no one cares about the iPhone then of course you are going down. This is the meaning. What we see here the last sentence is because of these correlations Schumpeter's creative destruction is not validated by empirical evidence. Why is this not the case? How do we see this? Who wants to comment on this? The constructive creative destruction hypothesis included the idea all the new firms come up with these bright new products and business ideas and they challenge the established firms to kick out of the market. Why don't we see this here? What's the reason for this? Yeah, please. Right, exactly. This is the underlying assumption a new firm is better than the old one so the new firm should survive and the old one should disappear. That's not the case. We do not see this. We see new firms enter the market that is exactly the reason. That means if you have to argue in the exam why don't we see Schumpeter's hypothesis at work then that's the argument. Yeah, that's true. We can, yes, we can put this this is not explicitly said here we can put this argument because there is empirical evidence of this but the Starlight Effect as such does not have this. Why is this the case? Let's think of this example of the horse carriers that are overtaken by automobiles. Then you have the established firm and the established firm also leaves the market. Why is this? Because a complete new technology takes over. That is the reason. You are right with your statement the exit rate is the highest for the new firms that's correct but it depends on what scenario you are looking at therefore this Starlight Effect is formulated a bit weaker than what you said but your comment is of course correct. Here we have a table that shows us different countries and compares the entry rate both in terms of the number of firms so these are all percentages but in one case we refer to the firm number and in the other case we refer to the market share. We can go through this now where Belgium they made a distinction between manufacturing firms and services this is also very important to have this distinction because as we discussed during the break manufacturing firms for example depend on issues like the oil price available of natural resources these kind of things the immigration laws of Switzerland or Canada all these kind of things whereas a service company has other boundary conditions you can still run a business and then use the internet to let the people in India program your stuff so they do not need to be physically there in Canada that means you will not see the same dynamics for the manufacturing firms and for the service firms that's important to understand you will see dependencies on the business side but usually not on the other hand what we also discussed before when we talked about insurance business what are those construction business of course service companies have a lower entry barrier because they do not need to put up a lot of physical things whereas manufacturing firms usually have a higher entry barrier therefore the pattern becomes much more complicated this is what you see here there is a high number of firms a relative number that's percentage as I said compared to manufacturing you know why this is the reason because the barriers are different if you look into the exit rates and you see that this matches nicely you have a high exit rate it's a positive correlation high entry rate, high exit rate that was the only claim that we made in the Starbucks fact by the way now I recognized that I wanted to do it differently so I wanted to do it like this and then we can use the cursor and then this is what I wanted to do I rarely use this because we usually do not have tables and then I forget about this so this is something that we can look up here in the number of firms entry and exit rate how this is correlated here we see of course that the evidence is across eight different countries these are all very mature economies Korea I mean South Korea these are all established economies the time interval is different that means if you see something like this here where the exit rate in manufacturing is larger than the entry rate that's what you see there it has well to do with the business cycle it has well to do with the conditions in Belgium between 1980 and 84 you see the data is from various intervals here and it's not always some are covering the first part of the business cycle and the second part therefore you cannot really compare all these numbers note this you should see the correlations here you should also see if you look into the market here the impact of new firms entering the market you see it's always between it's about 3% that was the same number that we have seen before when we looked into the done paper fluctuates a bit here it's much smaller then if you want to argue why it is 1.1% in Norway whereas Germany 3% in Germany or 5% or 6% in Portugal then you have to look into how the economy is structured you have to think what is the major business in Norway then I come up with my small insurance company what might be the impact on the market share compared to stud oil for example that's not a lot this is then if you argue like this and you understand why some of the numbers are high and some of the numbers are low last table here this again from Mr. Dunn he looked into correlations of entry and exit rates remember that the start-up factor is about correlations between entry and exit rates he has given the exit rates of different time intervals that was this US manufacturing industries data set that we talked about these are the exit rates this is similar but let us just look into this now he looks into how it is the exit rate of this time interval correlated to the entry rate of another time interval here in the first part he just calculates the correlations what he sees is there is always a positive correlation between the exit rate and the entry rate that was the meaning of the stylized fact then the first confusing thing appears if you look here for example how can the exit rate in 1977 be correlated to the entry rate in 1963 of course if the company stayed for like 15 years there there is a correlation around how can the entry rate what was the other example I looked into the exit rate the correlation between the exit rate in 1963 and the entry rate in 1977 in order to understand these numbers you have to think of what is the meaning of a correlation a correlation is not a causality it simply tells if there was a period in time like 15 years ago where the exit rate was high then of course the market was somehow turned over and this allows later for new firms entering the market that is why we see a positive correlation but there is not a one to one causality because of this that happens then you ask yourself why you cannot really understand this causality that something that happens earlier affects something that happens later is clear but that something that happens later affects something that happens earlier but it refers to the conditions that are created under which these entry and exit rates can change so you see that style as fact number one but now Mr. Dunn did something differently he did a correction for fixed industry effect does anyone have an idea what a fixed industry effect is for example the oil price the legal conditions these kind of things environmental laws all these kind of things and then you see that these positive correlations turn mainly into negative correlations so and then you see that the whole thing is much more determined by subtle developments in the industry as a whole I think in two lectures or so we have a model where we look into different correlations where the growth rate of a firm depends on the growth rate of the industry but also on the growth rate of the economy as a whole so there is a correlation if you are in a booming country like Lithuania then of course there is an effect of the overall growth of the growth rate of the industry on the entry and exit rate of your specific business which is maybe in construction and here you see these correlations turn into negative correlations that means there is a high entry rate but a low exit rate or the other way around there is a low entry rate but a high exit rate and this refers to conditions inside the industry you understand what I mean that means one of the driving factors here is indeed the boundary conditions of the overall conditions of the industry and of the economy as a whole that is what you see here that is why these correlations turn into negative one if you control for the fixed industry effects so now we move to the papers not written by the economists but by the physicists now everything changes, no tables anymore no numbers, just these plots you can guess from these figures who did it you can also see from the arguments who did it of course Mr. Dunn was very careful in controlling for example for the fixed industry effects what we see here is the data set we already discussed where we have seen the nice distribution as you recall and here is the total number of companies plotted over time you see these are I think they were 20 years and then you see how the total number of companies in the S manufacturing sector oscillates you see this, it's a total number of companies and of course you see a business cycle here you can clearly determine the business cycle if you look into this and you see how the new entry entrance and the firms dying contribute to this you clearly see that of course there is an anti correlation if this total number goes down then the number of dying companies is quite large but you see that the number of new firms entering the market is not synchronous in time you see this coming back to our starlight's fact there's a positive correlation between entry and exit we can say well that might be true but not for any instance in time that's a different thing and the starlight fact didn't claim this but if I look into a particular year then I see of course that there might be a high exit rate and a low entry rate or the other way around but if I ever reach this over time then I see a very clear correlation between the two that's important if I go for time series analysis then I see this kind of anti correlation that by the way was also shown in the table of Mr. Dunn because there are a few more effects that we have to take into account if we just take the gross picture and we see the positive correlation we control for certain things like the business cycle and we see a different picture you get this point, right? here they looked into the firm size distribution so this is it for all companies this is for the dying companies and this is for the new companies again the size is measured in this example in terms of sales that's the sales number what they claim here it's a bit difficult to see is that for the new firms that is the lowest one, this one they claim that there is a log normal distribution we should believe them I hope someone has checked this we can check it here by looking into this parabola like shape looks a bit like this but for these two distributions there is a slight difference they did not confirm that's a log normal distribution but an exponential distribution of certain order polynomial we are not going into the details here what I just want to say is that if you look into if you look into if you look into specific parts like the dying companies you may find a different distribution but if you aggregate over the whole industry then you see this clear log normal pattern I'll come to this in a moment with another example as well the surprising thing is here what does this mean? this is size sales this means 10 to the 10 US dollars you have a new company entering the market we are always thinking of the one man company this is a huge company this is one of the largest companies that you can think of in this data set what's the meaning of that? and then we see that a large company a very large company is dying how do you comment on this? let's assume we have the exam and you should explain why there are big companies entering the market what's the reason for this and how does this refer to Starlight's fact number 4 where we talked about age and size absolutely correct maybe it's Lehman Brothers but maybe it's a part of the newly formed company this is the reason why we see this the data set doesn't tell us whether the company really died the company has died in a legal sense but it's now part of a large pharmaceutical company or something like this that's the idea mainly these things can be explained through merger and acquisitions I think I mentioned this here as well change of name is another thing that you cannot detect in the data set if you for legal reason change your name so then a big company dies and a big new company is born that's what the data tells us but then of course you can track down these cases say what company are we talking about and then you look it up and say well there is a reason for this this is not what these people were interested in but if you are an economist or you do your PhD in management science then you certainly go more into detail these people were already happy after they found the lognoma distribution for the firm that's also nice but there is more in this okay so this tells us the dying probability dependent on the size it is not a surprise that firms with a smaller size have a larger dying probability we already talked about this firms of a smaller size have a larger volatility in their growth that's what you already remember from style aspect number two firms of a smaller size are usually firms that are also younger there is a correlation between the age and the growth rate which is all the negative that means if you are a small firm and a young firm you may not make it it means you have a high probability to die remember the 50% in five years for Switzerland so if you reach a mature size then this probability remains almost constant that's the message here so it is not zero because there are these instances that big firms collapse because they have bet on the wrong horses in the financial business for example but they can also die because there is merger and acquisition it's a bit difficult to see that this is constant some people would argue for something like this I'm not going to comment on this if you are not the physicists but the economists they do in front of such a picture you would probably go for economic or management arguments that it is either like this and either like that that means you as a person who gets this degree in MTech you would probably formulate a hypothesis that is built on economic theories and economic insights or insights from the management side you would not simply go and plot the data and then say well we think it's linear or we think it's constant you wouldn't do that you see this is a major difference between doing science and economics and management science and doing econophysics I mean I'm a physicist myself I can say this in econophysics people have all these nice insights like this one but they have no clue about the meaning that's the problem here and at some point you appreciate that you do not just plot things and sell it to the audience but that you already have a story to tell and that's the major difference and that's why I'm in this department and not in the physics department I'm also a professor in the physics department but that's a completely different mindset alright so here we have a few other evidence numbers here I'm not going to read this here let us look into this exam because I think it's important to understand this before we close we already heard about the log normal distribution of all firms then we heard about the log normal let me go back to the slide here the log normal distribution of new firms entering the market we didn't hear about the log normal distribution of firms leaving the market these are the dying firms here we didn't hear about that instead we heard about a third order polynomial no one knows precisely what the difference is but probably there is a difference that's another issue but here you see the people claim that the dying firms also have also follow a log normal distribution in size that is a claim here that's not the same as Louis Amaral and co-workers claimed it's not the same but it's in line with other arguments that we had there's a skew distribution for all firms so maybe there should be all the skew distribution for dying firms and for newborn firms so that's what you see in the data but in fact, as these people argue here this is aggregated data they look at it over I don't know time interval of 20 years we can go back to the Amaral paper and then we see they have average over 20 years and when they average over 20 years then they may see a log normal distribution the question is how did we get the log normal distribution in fact, firms are born and die every single day in the year and the question is this picture that emerges over 20 years does it hold if we go for day to day events that was the question that was asked here and these people argue that the day to day underlying distribution is a different one if you aggregate all these distributions here then you may end up in a log normal distribution but if I talk about the monthly events for example or the weekly events then it's not very clear that we get a log normal distribution also for these time intervals so their argument is actually out of the blue I have to say but that's my personal comment so it's this one so they assume that if we disaggregate the data it can be described by this kind of distribution so what is it? what kind of distribution is it? it's a power law i to the minus r r is the exponent and i is our usual x value and that's the normalization here and the normalization applies up to a value m which defines some sort of size class okay and now they are trying to test this for the disaggregated data against some simulation value they have the assumption that there is you see that's an indirect proof they have the assumption that the data follows this distribution so why is it a truncated power law because the m gives us a value at which we truncate this and now they do the following they say let's assume that this is the distribution the data follows then we do the following we generate artificial data from this distribution we take the disaggregated data from this data set we plot this against each other and that's what they do here that's what's also called a q-q plot these kind of things that means if the aggregate if the simulated data which is here on the x-axis matches the real data about dying firms then we should see a straight line a diagonal exactly at 45 degrees yes please there are many comments on this so if these were my graduate students let's assume that then I would tell them if you want to show that then you have to use the square that's the first thing they have plotted this in a rectangle and they have even used different scales for the axis so that means they made it very difficult for us to recognize that there is a 45-degree line so now you can argue did they do this on purpose or did they do this because they do not recognize yeah okay I'm not going to comment on this I just met one of the co-authors yesterday yeah that's the idea you would probably if you really want to show this you take a square right you make sure that the two axes are precisely the same and then you really mark the straight line is this your comment? no then you would see a correlation but it would not be the same yes yes in a q-plot it should be on this 45-degree line yes please yes that's also correct that's also correct there can be a difference there can be also a difference in the normalization of course but let me finish with this so that means the argument and that's an important argument the following the aggregated data over 20 years may show us a log-normal distribution but we do not know how this log-normal distribution is accumulated so therefore their proposal here is so they didn't defend explains as well why they had this so they say okay the daily data the disaggregated data follows a very different distribution namely a truncated power law right and then they took the disaggregated data simulated data from a truncated power law and showed us in this straight line that there is a match the match doesn't hold for all data as you see not for the large size not for the small size it holds only for the middle so you can have lots of arguments against this if you want to the reason why I'm showing you this is another one it is known and this is all the how the people started here that this describes the exit rate of biological species there is outside if you have a look at these posters it's about ecology how species survive and die that's what the people on the poster outside tell you and we know already from dying species that the number of species dying follows this kind of distribution and we know that the R is 2 and then the people said okay if species die like this truncated power law was R over 2 let's look if firms also die according to this truncated power law with R equals 2 and the result is yes they do that's the result here also confirmed for 8 different OECD countries and so on so that is a nice message if you think of firms dying then I mean the birth of the firm like of course the birth in a biological setting it's a small firm in the beginning it grows over time likely to an optimal size and then it dies so there are all these biological arguments and this is another argument that confirms that this kind of life cycle of a firm really holds that's the message from this if you look into the disaggregated data you see that firms die over time and this is the distribution I already made it without finishing let me just come to one last remark because next week I'm not here and Pavlin is going to teach and I don't want to leave this for Pavlin the last style in fact we're talking about is survival of firms we implicitly discuss this already so we now take a variable i this is 1 if the firm has survived and 0 if the firm has died so we already know that h plays a role and size plays a role but we do not know what's the precise function of h and sizes so we name it d d depends on these two variables and then we draw a random number from a distribution and we say if the random number is larger than the d then the firm survives and is smaller than the firm dies that means that d defines a threshold at which the firm dies the threshold is very high then it's likely that the firm dies if the threshold is very low it's likely that it survives this is simply a definition and now we are looking into the expectation and that's the probability that this random number e is larger than the threshold that we have just defined without knowing what the dependence is and this probability is simply the cumulative distribution function of the d you see this we are asking what's the probability that some value e is larger than d and this is a definition of a cumulative distribution function namely f of d we have no clue how this distribution function looks like we don't know how the d looks like we also don't know how the f looks like but we are not really interested in this we just say if we could know this cumulative distribution function the style aspects number 6 says that the cumulative distribution function should increase with x that means your survival will be larger if you have a larger size or the other way around if you have a small size then you might likely die you have a small size because you were just born you have also a small age and then you are also dying that's the idea behind it you understand the argument we did not define how the threshold looked like we just said it depends on the age and the size and we did not define the cumulative distribution function but this argument tells us we are not interested if we just look into the tale of this and we say that this is positive and in the same way we argue that the survival rate increases with age without specifying the f that's important so you see our statement about survival dependent on age and size is weak weak in the sense that we did not specify any concrete distribution or how the threshold is defined but the assumption is if we define a threshold we are able to empirically determine the cumulative distribution function and we will probably see this kind of dependence okay so if we combine this and we see that the growth patterns of the smaller and the larger firms differ why that? because if you have a smaller firm then a younger firm then the probability that you also die is larger so you can read this at home I think this slide is not so important so that's our self-study talks and these are the questions so by now I would expect that you know all these six, seven starlight effects that you can name them that you can argue why these starlight effects exist what is the real meaning of this why can't we be more specific about certain things that's also important some of these starlight effects are not very precise some are very precise what's the reason that some of them are not precise like the correlation of entry and exit rate we discussed that we see this in the data but only if we do not look into the reason for this if we for example subtract the fixed industry effects which we did in this example and we see that this correlation is broken that's a very important thing so that means it's not there for all times and under all conditions there are reasons for this that are somehow hidden in this positive correlation and they may refer to the industry for example or to the business cycle that's only on average so this kind of argumentation you should be really able to do with this I stop as I said Pavling will take will give this lecture but only if there are at least five people that's our threshold from now on you keep it in mind you talk to the other people thank you very much for your attention and I see you in two weeks