• 00:20    |    
    So, I wanted to talk now about how I measure the macroeconomic contractions and the sort of rare disasters, which pertain to the macro economy. I want ton talk about the measurement data.
  • 00:35    |    
    I permanently find that people interrupting to make critical remarks or whatever, is fine.
  • 00:46    |    
    This is conceptually looking at the two kinds of macro variables. So, one is personal consumer expenditure per capita, that's C, and the other is real pern capita GDP.
  • 01:00    |    
    Now, conceptually I thought that the consumption would be more relevant, in terms of asset pricing, but I don't think you can be too rigid on that becausen the way you actually measure it's more like a personal consumer expenditure, it's not consumption.
  • 01:11    |    
    And this is in more measurement error in the C data than in the GDP data, so I think it might be more useful to look at both macro concepts rather than justn looking at one or the other. So it's dependant if you look at both, and of course, we have more data on the GDP, so that is reflected in some of the results.
  • 01:33    |    
    So, the basic procedure if I talk about personal consumer expenditure per capita is I have some annual time series on that, so for the U.S., it started 1969n and goes through 2006.
  • 01:46    |    
    The counterpart of the Dower jumps, which in the model corresponds to rare disaster, and mentioned before in the data, it tends to be that a really badn macroevent, cumulates over more than one year, so the Great Depression is a peak in 1929, in terms of both consumption and GDP, and a trough after which the growth rate becomes substantiallyn positive is in 1933.
  • 02:15    |    
    So, basically a four year period, where an accumulative percentage decline, is the number shown here, so this 1929 to 33, this is for personal consumern expenditure, it's a decline by 21%, so that's a four year cumulative proportionate decline or for GDP over that period would be 29%.
  • 02:40    |    
    So the empirical counterpart to what are the jumps downward, the disaster shocks in the model, are these kind of peak-to-trough measure declines, neither Cn or GDP.
  • 02:54    |    
    So, we're formulating this more rigorously in this ongoing work that I mentioned before, with Nakamura and Steinsson, where we sort of look at all the annualn data and then sometimes it happens that the negative annual ones cumulate to something pretty big.
  • 03:13    |    
    But in terms of this first set of results, I just isolated peak-to-trough periods in the data for consumption of GDP for various countries, which accumulatedn to some amount. Arbitrarily, I looked at contractions that were at least 10%, in either consumption or GDP.
  • 03:31    |    
    So actually in the first study 2006, the threshold was 15%, so I looked at events that were downward of 15% or more. And in the more recent analysis, In pushed the threshold down so that looking at contractions that are 10% or more in this cumulative sense as being quote, unquote "depressions" or "rare disasters".
  • 03:55    |    
    And it turns out that in terms of the asset pricing results, that the difference in the threshold of 15 versus 10%, actually doesn't matter much, which isn fortunate since the threshold really was arbitrary. So these are examples pertaining to per capita consumer spending the consumption of big declines.
  • 04:22    |    
    Twenty-four countries have the necessary long-term data on consumption, starting at least in 1914. In this first set of results we found 95 events weren accumulated in terms of peak-to-trough declines of 10% or more.
  • 04:36    |    
    And I brought data that a bit some more data and finding some things that were missed, so instead of 95 it´s actually 99 now, in terms of the data, it'sn still 24 countries; anyway, it's around 100 disaster events gaged by consumption, which is about 4 per country, over roughly a century, in many countries it's 1870 to 2006, so it's 136 years orn something like that.
  • 05:04    |    
    So four per country is roughly how many you see, the average contraction size, given that I'm looking at things that are 10% or more, turned out to be aboutn 22% per consuming spending per person.
  • 05:19    |    
    For the GDP, there are 36 countries with the necessary data, we found 152 of these GDP depressions, one of which is the U.S. Great Depression from 1929 ton 33, so that's one out of a hundred and fifty something.
  • 05:35    |    
    Turns out that's 157 now rather than 152, but basically the same story. The average contraction size for GDP is similar to that for consumption, 21% to 22%,n so it's not actually a big difference.
  • 05:50    |    
    The usual view in terms of business cycles, which is certainly true for the post WWII U.S. data or OECD, is that if you measure by consumption it looks moren stable proportionally than GDP.
  • 06:05    |    
    But that's not actually true for these big contractions and actually in the wartime ones, which are a subset, consumption tends to contract propotionallyn more than GDP in the really bad events.
  • 06:16    |    
    And part of that is because the government is taking a lot more of the pie during the wars by spending a lot on the military, and of course, militaryn spending is not part of personal consumer expenditure, so that's kind of an extra effect downward on consumption.
  • 06:29    |    
    But overall there are kind of similar. So these are some examples, for the U.S., I mentioned that there were two contractions gaged by 10% or more decline inn consumption.
  • 06:44    |    
    This is the one with the trough in 1921, which Jose Ursua studied in great detail because we think this is about the great influenza epidemic, so he's beenn assembling data across countries on how severe the epidemic was, based particularly on mortality, and whether that matches up with the macro experiences is a question, I don't know the answern to that.
  • 07:03    |    
    Of course, he has to disentangle this from WWI, because WWI was related in terms of spreading the epidemic, but it looks like the timing is different. In might also have mentioned that you may have seen some of the work I've done on religion and political economy, it's another kind of research topic very different from this one.
  • 07:25    |    
    But at some point, I saw a list of the famous people who had died because of the influenza epidemic and Max Weber was there in 1920, he was sort of a mainn figure in sociology and religion, so I took that as a sign that I should be studying the great influenza epidemics, because this figure related to this other line of research had died becausen of this disease.
  • 07:46    |    
    Also Woodrow Wilson suffered from the flu in 1919, and some people think that had to do with the Treaty of Versailles at the end of the WWI, that Wilson wasn kind of incapacitated, even though he didn't die of the disease, people think that it would've been a very different world if he hadn't been suffering from the flu. I don't know whether that´sn true but that's the story.
  • 08:04    |    
    So these are just some examples of other declines based on consumption out of the 100, or so, for consumption, and a hundred and fifty-something forn GDP.
  • 08:23    |    
    So for France you see a decline of 58% in consumption during WWII. France is a little different from some other countries during the war because they didn'tn have as much physical destruction as some places because they surrendered so quickly.
  • 08:36    |    
    But they were occupied and the Germans were basically taking resources from France, so it´s a clear factor of influence. We don´t have long-term data forn China, in terms of long-term information.
  • 08:49    |    
    Taiwan has a very good National Account study going back to, I think it's 1901 or something, so in some sense, Taiwan might be interesting, particularlyn because they're not in the China data but the experiences are not exactly the same, particularly in terms of the role of Japan.
  • 09:05    |    
    But you can see Taiwan has a kind of decrease starting really before WWII, in terms of other conflicts in Asia before WWII. Here laying the 68% which isn quite remarkable, and then you see Japan there.
  • 09:29    |    
    So this is the data set with respect to disasters, macro disasters, gaged by C or GDP, which I'm then going to use to gage the probability of disaster andn the frequency distribution of disaster sizes, which then filters back into the equity premium results.
  • 09:45    |    
    Audience:I was thinking of a variable that might affect all your presentation, and that is culture, cultural matters. We start from the social that man is an cultivated resource, as Julian Simon said.
  • 10:11    |    
    ... well how different consequences that are cultural, that these correlators that have reached taken.
  • 11:00    |    
    Let me give you two small examples that I don't know if this takes on savings too, or is it as Japan before 1868, was the fuel of society and the reason In took it was war but not in economics, after 1868 through 1945, they increasingly took great risks in the economic field, specially being a country without natural resources, very importantn natural resources.
  • 11:01    |    
    Although the American occupation from 1945 until now, it became the second largest economy in the world, through risk-taking. Now, take Korea, the sellingn rates of Korea in 1955, when it was completely destroyed, was higher than the selling rates of the United States at the same time, when there was peace in the States, there was no war, Vietnamn hadn't still started.
  • 11:34    |    
    So, all this is cultural and the variable of culture, I think, might make more reliable the projections or even the study of data of past times, regardingn risk-taking, savings, returns and so on.
  • 11:58    |    
    Let me ask you another question because is related to the present question...
  • 12:3.5    |    
    Robert J. Barro:Maybe I should respond to that one first. You know, in some other work I've done about determinants of long-grown growth, I've beenn sympathetic to the idea that this cultural influences might matter a lot. I tried to quantify that, tried to measure things, particularly in regard to the role of religion as one element ofn culture.
  • 12:28    |    
    So I've tried to incorporate this kind of feature, which I think is in sympathy with what you were saying and you know, you have some results, and I wouldn'tn say that they were that dramatic, but there were some results.
  • 12:42    |    
    The way I would enter into this analysis is through the preferences that would enter as you suggested as the willingness to take on risks, so that would ben this parameter I called Gamma, a higher value would mean less willingness to absorb risk and that would certainly matter for the results.
  • 12:58    |    
    And it could also matter in terms of attitudes about savings, so that would be about this other parameter about willingness to substitute consumption overn time, which I mentioned in more recent versions of this model, is conceptually distinguished from the risk.
  • 13:16    |    
    There's two parameters and those could be culturally influenced and they might differ, now the way I'm implementing this at the moment, is to pretend liken the various countries and time periods are the same in that regard.
  • 13:28    |    
    So, I'm thinkning of it as though these preference parameters, which could have a cultural basis, in what I'm doing at the moment these are thought to be then same across countries and over time, so that's the sense in which I'm not admitting these cultural differences or other differences that could affect risk attitudes or saving attitudes.
  • 13:47    |    
    So that could be incorporated into this framework; either differences across countries or shifts in certain points in time, and then one could see how thatn mattered, but what's being estimated here then is that some kind of average of those parameters in terms of the results.
  • 14:04    |    
    I would have to interpret it as some kind of international average of those, because I'm not allowing for these differences, which could be culturally basedn or based on other things.
  • 14:13    |    
    There are other things that can affect preferences or tax systems can affect willingness to absorb risk or to save. So those are all factors that could ben brought in but they're not in this basic framework.
  • 14:32    |    
    Audience:I'll make you another question: How do you define rare?
  • 14:36    |    
    Robert J. Barro:I don't really have to define rare, what I'm really defining is how big of a contraction do I consider to be a depression, so initially, in an mechanical sense, the thing I'm taking as a definition is that a depression is at least a 10% decline or a 15% decline in consumption or GDP.
  • 14:57    |    
    Once I've said that, then the model delivers how rare or common these events are, it tells me what the probability per year is of entering in one of thesen events and it turns out, that's a number of something like 3% per year, if I have this 10% threshold, so I don't have to define that 3%, I start with how big a contraction do I consider to ben large.
  • 15:21    |    
    Then I can also consider what difference does it make if I have a different threshold, if instead of saying 10%, I said 15, does that matter for the results?n It turns out it's not too sensitive to that.
  • 15:31    |    
    So in that sense I don't have to define separately what is rare, although it turns out that if you're thinking about big contractions, they are rare inn history so...
  • 15:45    |    
    Audience:How big is big?
  • 15:46    |    
    Robert J. Barro:The arbitrariness is in the sense I say in the threshold, so I took 10%, usually that's considered to be a big decline. The U.S. for example,n has not had a 10% decline in GDP since the 1930s, so it's very unusual there.
  • 16:02    |    
    Other parts of the world outside of the OECD have had a lot more declines of that maginitude in the recent years, so you know.
  • 16:11    |    
    Okay, this is just showing me where the big events concentrate, in terms of the history, and of course, there are these global events that dominate, althoughn some of the events are more country specific.
  • 16:26    |    
    I mentioned before, if WWII is really the biggest world disaster event, so if I look at consumption, for which I had the data of for 24 countries, GDP forn 36, you can see that there are 23 events for consumption, with an average size of 34%, which is enormous, and similarly for GDP there are 25 events of contractions in GDP of at least 10% andn the average sizes are remarkable; 36%.
  • 16:55    |    
    So WWII is definitely the biggest disaster, so second to that, pretty close, is the Great Depression of the early 1930s and WWI. You have similar experiencesn there, roughly comparable in terms of global significance, I mentioned this in the early 1920s, which we think reflects the great influenza epidemic.
  • 17:20    |    
    Turns out that there are eleven events there, quite a few, not as much as I just mentioned, but eleven events with an average contraction size of 18%, so itn looks like it's pretty important.
  • 17:30    |    
    As I mentioned also, we don't have natural disasters that are big enough to show up here. You might have thought that some typhoon or something in Southn America would've been big enough...
  • 17:42    |    
    Audience:Earthquakes?
  • 17:42    |    
    Robert J. Barro:Earthquakes, none of that is big enough in terms of these country level data that we´ve assembled. If we focus on cities, we'd probably findn more, but of course that's not the data that what we're looking at.
  • 17:54    |    
    So the only kind of natural disaster, if you want to call it that, that's in the sample, is the great influenza epidemic; and maybe that's not even a naturaln disaster.
  • 18:03    |    
    Post WWII period, the OECD is remarkably tranquil, remember the sample ends in 2006, this is going to change by next year. So in the OECD we only have ninen events, with an average size of 14%.
  • 18:23    |    
    The biggest OECD event that shows up for either consumption of GDP, is Finland in the early 1990s, which had a major banking crisis, probably had beenn brought up by the collapse of the Soviet Union, which affected the trade relationship between Finland and Russia in a substantial way.
  • 18:41    |    
    It also affected Sweden, there's been a lot of discussion recently on Sweeden in terms of having nationalized banks and stuff, it's the same crisis exceptn that it was bigger in Finland in the early 1990s.
  • 18:53    |    
    But the main thing is that it's very calm, I think five of these nine events are in Iceland, which you might not want to count. Basically it's been veryn little in terms of disaster events in the OECD.
  • 19:07    |    
    Until now, there's no question of the current crisis is going to add significantly to this list of OECD events since the WWII, that's already clear, thatn there are a number of countries that are going to show in the list of contractions of 10% or more, the only question is how many.
  • 19:24    |    
    It will probably include a place like Japan, for example, but we don't know yet as to how many. Outside of the OECD this period is not that calm. So postn WWII there are 29 cases of consumption declines, so this includes a lot of things in Latin America and also some cases in Asia.
  • 19:46    |    
    So the tranquil appearance for the OECD does not carry over to the kind of rest of the world. Again, the data set is limited by the data criteria that I'ven mentioned before. We do have this set of countries in Latin America and in Asia as part of this sample.
  • 20:05    |    
    Audience:Excuse me but I'm a little bit amused, just what strikes me, looking at the list of episodes, is that the wars and influenza, particularly the wars,n we assume that they're not caused by an economic factor or competitions.
  • 20:30    |    
    If we assume that, then we can say that all these events are exogenous, the Great Depression would we consider it exogenous as well? I don't know, it justn struck me to ask that question because it's a totally different kind of event...
  • 20:49    |    
    Robert J. Barro:I mean, you might ask whether it was independent of government policy.
  • 20:54    |    
    Audience:Right, that would be my question.
  • 20:53    |    
    Robert J. Barro:And also whether it was independent of institutions, I think is very unlikely the Great Depression was independent of that, but that'sn probably true of the wars too.
  • 21:02    |    
    You know, I think the main reason we have the European Union and the Euro, is to reduce the probability of a war that looks like WWII. I think that was then main reason France and Germany got together, I think that's really why we have that, not for trade or financial doing, but because we don't want WWII again.
  • 21:21    |    
    So if that vision is right, then the war probability is certainly not exogenous with respect of that institution. Of course, it may not be correct, it mightn be that forcing Germany and France to sort of share all this stuff and might produce conflict in the long run, maybe it goes in the wrong way.
  • 21:37    |    
    It's very likely that the war probability is independent of a lot of these other things. The influenza epidemic was obviously spread by WWI, so in some sensen is not exogenous with respect to anything that mattered to WWI. But, yet the current flu thing now is actually the same strain as the 1918 epidemic.
  • 22:02    |    
    I don't know if you're familiar with the pattern of the great influenza epidemic, but it started out in the spring of 1918, and it was very mild, basicallyn just like the current one and sort of disappeared. And it was when it came back in a different form during winter, then it became very... an epidemic that killed lot's of people.
  • 22:25    |    
    So that's why we might be worried about the current situation, because I think, people have sort of gotten bored with the influenza thing; figured it's goingn to be mild and, most likely that would be correct, but actually the pattern is very similar to the 1918 epidemic and it's the same kind of strain.
  • 22:45    |    
    So I don't know, exogenous depends, I think, on what question, I mean, I'm tryng to ask whether this kind of framework, with the incidents of disasters thatn we saw sort of works in term of some of the financial pricing patterns that we see.
  • 22:58    |    
    And I think for that purpose we also need to know when these events are exogenous with respect to some other things, but I want to know about designingn government monetary policy or fiscal policy or institutions, then it's going to be central whether those things affect the probability size of disasters.
  • 23:17    |    
    If you think about evaluating Greenspan as the Federal Reserve Chief, I think the main thing is not what you do year by year, in terms of price stability,n but was it true that he was a hero in the sense of mitigating the 1987 stock market crash effects, the macro effects, did he really have success in 1998, with the various crises that occurredn at the same time there.
  • 23:45    |    
    Because I think the reason he was applauded as a hero, was that he seemed to have somehow worked through those really big potential crises and it didn't haven been a big deal for macroeconomics.
  • 23:56    |    
    So if it's true that he was successful there, then that's very important with respect to disaster probability and sizes, of course, now he's not popularn anymore, so now people go back in the same record and say, well he didn't do so well here and so on, back when he did set the stage for 2007- 2008 problems.
  • 24:11    |    
    So, I'm not saying I know the answer as to what it is, but those kinds of questions make these disasters probabilities and sizes clearly endogenous withn respect to some interesting things.
  • 24:24    |    
    You know, so now Bernanke is going to go down in history as a hero or a villain depending on how things work out, it's the same.
  • 24:36    |    
    Audience:Bubbles, these successive bubbles in the market, like the technological bubble three years ago, and the real estate bubble, and recent market stock,n are they predictable?
  • 24:53    |    
    Robert J. Barro:Well, predictable certainly not, but I'm not sure the 2001 thing was a bubble. You know, I think the internet and related innovations weren very hard to evaluate, in terms of the commercial implications of those things.
  • 25:20    |    
    And I think the stock market price revolution going through up to the year 2000 was a way to try to cope with the sort of whole new world with respect ton commercial prospects.
  • 25:22    |    
    I don't think, except exposed, that that was a bubble, but that's a different story. I think the housing prices, for example, in the U.S. were clearlyn influenced by certain aspects of policy, credit markets.
  • 25:35    |    
    And I think you can criticize some of those policies that encouraged too much risk taking, in terms of the real estate development section and the way then things were priced. I think those are very different cases. Anyway, that's what I'm supposed to talk about tomorrow.
  • 25:56    |    
    Audience:Tomorrow there'll be a different group of people.
  • 25:59    |    
    Robert J. Barro:I get to say anything tomorrow.
  • 26:03    |    
    Audience:About this meaningful, take into account the number of years that the recession was?
  • 26:17    |    
    Robert J. Barro:Well this one doesn't, I've done it in two ways, this is just comparing the levels of per capita GDP or consumption, peak-to-trough. Now, youn might say that really there´s a regular average growth as positive, and this is understating the extent of the decline and it should be somehow relative to what the trend would've been.
  • 26:39    |    
    So I've also done it that way and I in terms of some of our current work, I've also taken account of that. So that gives you bigger depressions, if youn adjust for the trend and then the trend effect has to do, as you suggest, with a number of years, the duration of the event. These don't take account of that.
  • 26:54    |    
    Audience:Does consumption include durable assets like automobiles, housing?
  • 27:06    |    
    Robert J. Barro:This is personal consumer expenditure which is non-durables and services, and consumer durables but not including residential investment,n which is classed with gross investment. To the extent that we can, we've divided it up; consumer durables versus non-durables and services, and I have some results there.
  • 27:27    |    
    But the problem is that for many of the countries in time periods we can't do that division. I would prefer to get something closer to consumption, somethingn like non-durables and services, plus the service grow under durables.
  • 27:42    |    
    But that doesn't seem to be feasible for most of these cases. I have some other data on that.
  • 27:56    |    
    So, let me go on a bit further, I'm going to skip over most of this, the main point in this is that if you look at contractions based on consumption versusn GDP, there's actually a lot of similarity in the proportionate changes and you can´t say a lot regularly about the the timing of the changes either. There are some diffrences between the warn and the non war.
  • 28:19    |    
    The war disasters tend to have proportionally more contraction in consumption, as I mentioned before, and if you look at the non war disasters, then proportionate declines in consumption and GDP, if you sort of match them up by the timing, turns out to be similar.
  • 28:33    |    
    But I think I won't take the time to go through this part.
  • 28:40    |    
    This shows you what the distribution of the crises looks like. So on this data we have 95 consumption disasters, now it's 99 but this is 95, and 152 GDPn disasters, now it´s 157.
  • 28:57    |    
    So this shows you the frequency distribution of the sizes, these are just histograms of the sizes of the disasters. So each individual disaster was measuredn in the way it described, if you look at the horizontal axis under C, you can see the proportionate decline in consumption during each disaster; 0.1 being 10% going up as high as almost 70%,n that's the B.
  • 29:23    |    
    And the vertical shows you how many there were of each, so this is about the frequency distribution of sizes of disasters. And on the right, you have then GDP, there are more events but the form of the distribution doesn't look that different; one versus the other.
  • 29:38    |    
    So, this was just what was asked about the duration, this is the distribution of the durations measured in this peak-to-trough manner. So the averagen duration is something like 3.5 years, in 1929 to 33 there's a four year contraction, these are all annual data, not quarterly or something like that.
  • 29:58    |    
    So this is the distribution of the durations, which I'm not willing to do anything with it on this analysis, but this is what it looks like, and it's not son different based on consumption versus GDP.
  • 30:09    |    
    Now, in some current work, trying to analize the form of this frequency distribution, so as you kind of look at it, it looks like what's called a Power Lawn Density, where as you look at bigger and bigger sizes, the probability of seeing something that big is declining in a way that looks kind of like a Power Law or Pareto distribution.
  • 30:36    |    
    So we've been in this unpublished working paper that was distributed, we´re trying to analize the form of that distribution. So it's something like this kindn of distribution except this alpha isn't quite right, but don't worry about it.
  • 30:53    |    
    So again, Bis the contraction, and this is about normal relative to the disaster, so a big number here means that the contraction is large, so this is somen converted form from the B itself, which is more what enters into the analysis
  • 31:11    |    
    So a power law says the probability of seeing something really far out, is diminishing as you're looking at bigger sizes, and the way it diminishes is inn this power law way which has a key exponent alpha .
  • 31:24    |    
    So if alpha is really large, it means that the thing attenuates very rapidly and then the tail is going to be thin. But if alpha is smaller, then you getn thicker tails and that's what's going to matter a lot for the equity premium, for people worrying about really bad events.
  • 31:44    |    
    This parameter alpha, is going to be really important. So basically, I've done the analysis two ways; one is, just take this histogram and say, well thatn tells you the probability of all these different sizes, the observed frequency in the data, and from that you can compute the various expectations that appear in the formulas in the equityn premium, the ones that involve the thing B.
  • 32:06    |    
    Another way to do it, is to fit a certain probability density function, which has some parameters and here the P parameter is alpha. And if you know the formn of the probability distribution you can calculate all the moments, all the expectations, that enter into the equity premium.
  • 32:26    |    
    And then you can do it that way. And one advantage of this approach is it allows some possibility of going even further out in the tail than you've ever seenn in the data.
  • 32:41    |    
    So the probability of seeing something like 60-70% is very low, but those big disasters also matter a lot in terms of the required risk premium. So if youn have something like what the probability of being 80%, if you used the observed data you'd say the probability was zero because we haven't seen any events like that.
  • 32:59    |    
    Whereas in this form, you're going to get some small probability and that actually mattered in terms of the results somewhat. So that's the approach I'mn taking now in that working paper. In the results here I just used the histogram.
  • 33:17    |    
    Audience:Do you have any micro data in terms of surveys to analize if there is any information about the expectations or to analyze the magnitudes of thesen events in something like potential contractions, duration or something like that?
  • 33:35    |    
    Robert J. Barro:There are data like that, for example for the U.S., of relatively recent periods, questions like what's the chance that this is going to falln by 10% or 20%.
  • 33:46    |    
    Alternately you can look at financial instruments, which pay off only when those things happen, so imagine you had a put on the U.S. stock market, and it'sn out of money and it only pays off if the market declines by 10% or 20%.
  • 34:02    |    
    So the prices on those instruments tell you something about some combination of what the probability is that people think that this will happen, and how badn they view it if it does happen.
  • 34:13    |    
    So some other people are taking the approach of backing out from like the financial prices data, what the probability is that people saw in the disasters,n and an advantage of that is that you don't have to then assume the probabilities are constant, they could be moving around.
  • 34:31    |    
    This advantage is that those data are only available relatively recently, and for a small number of countries, you don't see a lot of far out of the moneyn options that are actually traded, if you look at the U.S. S and P; 500 to 100; you only see volume on stuff that's like 10% out of the money, you don't see 20% being traded.
  • 34:59    |    
    I think it's an interesting approach and some people are doing it, but it has some limitations also. The survey part, I don't know whether, you might be ablen to survey all that, I don't mind results with surveys.
  • 35:31    |    
    I tend to be skeptical about all that relative to looking at financial prices, but I'm not saying you coudn't do something with that kind of survey. Then question would be how extensive is it when it does relate to the events that you really care about.
  • 35:26    |    
    Audience:Have you seen any of them just to check how different it is from the results you are getting in the paper?
  • 35:35    |    
    Robert J. Barro:Look at a little bit of that with the U.S., because some people are doing this option price stuff, it looks like there's probably somen consistency, but I wouldn't say that results are definitive at this point.
  • 35:50    |    
    I know of at least a couple of papers where people are trying to back out from the futures prices or options prices, what the probability looks like.
  • 35:58    |    
    But you know, that would be for the U.S. in 2008 or something, and I mean, it is true that the prices on that went up a lot, so this is related also to thisn volatility rates, the so-called vicks, which went up astronomically toward the end of 2008.
  • 36:17    |    
    And implicit in that is that people started worrying a lot more about potentially big declines in the stock market, so it's clear the direction of that inn this framework, a higher disaster probability.
  • 36:35    |    
    But I think it's still up in the air as to how these things correspond. People are doing some similar work with respect to exchange rates, and I guess then options data are better for looking at exchange rates rather than stock price index options.
  • 36:47    |    
    We can take a similar approach to trying to look at exchange rates across countries I guess it would apply to Guatemala, as I understand the situation.
  • 36:56    |    
    I mean you have a country like, say Japan, which historically had low interest rates, and then some other country has higher interest rates and then it looksn like there was a great deal to borrow with low interest rates and invest at a high interest rate.
  • 37:10    |    
    So this story would say that in a disaster, this low interest rate currency, like the Japanese Yen, is going to do particularly well, it's going ton appreciate like crazy in a disaster, and that's what people are worried about and that's why it looks like there's a big return from the so-called carry trade; when you're doing this businessn of borrowing at low interest rates and investing in high interest rates.
  • 37:32    |    
    So, I guess Guatemala would be analogous, if we thought about the distinction between the U.S. interest rates and higher Guatemalan rates; and then then argument on this disaster approach would be that if something really terrible happens, it's Guatemala that's going to go to hell and you're going to really be doing badly in that sense.
  • 37:52    |    
    So there are a couple of papers trying to take that approach to see if it can explain this so-called carry trade, which is related to this interest raten arbitrage across countries.
  • 38:02    |    
    I'm not doing that work but some other people are working on that.
  • 38:15    |    
    Okay, so this is how they backed out the disaster probability, so I'm pretending now that this is a constant, so you can think of it the average disastern probability across country and across time.
  • 38:27    |    
    So think about being in a situation where you are not currently in a disaster, it's a little complicated because disasters don't occur in an instant of time,n they have this duration in the data, averaging 3.5 years.
  • 38:38    |    
    So if you're in a normal year, at some point you might enter into a disaster state, and the numerator is how many times that happened; the number ofn disasters and if you take that relative to the number of years that are not disaster years, which remember to have a fine duration of 3.5 years, the number of normalcy years, which is all then years that are not disasters.
  • 39:03    |    
    So the ratio of those two is the counterpart of the disaster probability, which is moving from normalcy to a disaster state. So that's the thing here whichn ends up being about 3.6% per year for C and similarly for GDP.
  • 39:16    |    
    This is what disasters define as, at least, a 10% contraction. So before I had at least 15%, and then it turns out that you obviously get a smaller numbern and it's more like 2% per year. But that's a probability for disaster of a bigger size, that is 15% or more.
  • 39:29    |    
    So this is 3.6% per year, that's about three times per century, the typical country would be predicted entering into one of these 10% declines. So the U.S.n is a little more favorable in that, in terms of the results.
  • 39:51    |    
    Audience:The large intrusion of over pricing financial markets today, is going to speed up or slow down?
  • 40:01    |    
    Robert J. Barro:It's a good question. So that's like culture, that's another thing that influences either preferences or these disaster probabilities. Theren are a lot of good questions about things that might cause changes in disaster probabilities and sizes.
  • 40:18    |    
    I don't have the answers to that right now. You know, it looked like there was a great moderation, particularly in the OECD, it looked like things weren really calm, and that's why people weren´t so interested in rare disasters research, until last year.
  • 40:33    |    
    Things now have seem to have changed. But, I think the view before was that whatever intervention there was from the OECD governments, including monetaryn policies especially, the view was that that had reduced the disaster probabilities.
  • 40:51    |    
    Part of that would be whether the war probabilities had gone down, and I think people thought that it had. If you look at it today, I think people are non longer so optimistic, in terms of, particularly, non war disaster probabilities.
  • 41:04    |    
    So here I'm not assesing that, I'm thinking of this as a constant, across the whole time period, back to 1870, and across the countries, so that gives men enough observations, enough realizations of disasters, so I can actually pin down this number from realizations.
  • 41:20    |    
    But part of that constraint is that in order to do that, I have to pretend like the structure was the same over time and across countries. But I said thisn other approach of using financial asset prices or surveys, would allow you to entertain without time varying Ps or cross sectionally varying Ps and it would be some discipline on being able ton do that.
  • 41:45    |    
    I mean some people have reported this number just using the U.S. data, which I think is ridiculous, I mean basically in most of the analysis I have onen realization, because most people don't count 1921 as a realization.
  • 42:00    |    
    There was a problem in the data actually, people are mostly using the wrong data, with a period right after WWI, after 1920, they don´t usually count that,n they only count the depression, there´s only one event.
  • 42:14    |    
    And then you take a probability, obviously is one over the number of years that you've got, it's not very interesting.
  • 42:21    |    
    This is just from the duration, it´s a probability of exiting the disaster state and moving into what you might call normalcy, turns out to be 30% per year,n that's what corresponds to having an average duration of disaster of about 3.5 years.
  • 42:38    |    
    So again, the theory has these jumps for disasters, the data have these durations, and I've ignored that in this application, and I've argued in some waysn that I don't think this is a serious problem, there are problems that are more serious.
  • 42:55    |    
    But some people think this is a big deal. When you make a bunch of assumptions, it's good to have some that don't matter, because some people say, hey youn make a stupid assumption, then you can show that that doesn't matter, and you gain credibility.
  • 43:10    |    
    But there are some other assumptions that I think are more serious, like the assumption that the disaster probability is a constant, and I think that's an problem. The assumption that disasters can one day occur are kind of permanent.
  • 43:22    |    
    It starts from that level, as opposed to having recoveries, that I think is important, I don't have that in this at the moment but we're adding it to then current work. This, I think, is not a big deal but some people disagree with that.
  • 43:34    |    
    Okay, so now, I've tried to... I'm using this model to see if this works for the equity premium, there are certain parameters that I need that don't actuallyn matter for the equity premium.
  • 43:54    |    
    So, one is what's the expected growth rate, as long as there's a constant it doesn't actually matter what it is. So I've used an expected growth rate of 2%n per year, that's per capita, that doesn't matter.
  • 44:03    |    
    This normal business fluctuation stop sigma quantitatively doesn't matter. I've used 0.2, that really just doesn't count, if I'd made it 0.03 wouldn't haven been any different.
  • 44:14    |    
    I haven't introduced this parameter theta. This is in a setting that distinguishes intertemporal substitution of consumption from risk aversion. One overn this is about the willingness to substitute consumption over time, so that means that that's a number 2 which is pretty big, but I think reasonable.
  • 44:35    |    
    That doesn't matter for the equity premium. The equity premium, the key parameter is the coefficient of relative risk-aversion, which I call gamma.The raten of time preference also is not important for the equity premium, it matters for other things.
  • 44:47    |    
    Basically, the approach I've taken is the following, I´ve looked at the average real return on short-term government bills, which is around one percent, I´ven thought about the risk-free- rate as being a constant equal to that, and then I figured out what does the rate of time preference have to be to deliver that in the model.
  • 45:10    |    
    So this is not a test of the model, I'm just rigging it to fit the average level of rates of return, in terms of the risk-free-rate. Then the question is,n what value of this gammacoefficient of risk aversion with an attitude towards risk, do I need to fit the equity premium given all this other stuff? That's the key test of this model at thisn point
  • 45:33    |    
    Now, the observed equity premium in the long-term average is about 7% per year, but actually in the data the stock returns are what's called levered returnsn because corporations typically have a mix, of fixed financing, bond financing, and equity financing, and that does affect the riskiness associated with the equity piece.
  • 45:54    |    
    So I've made some adjustments for that, based on what seems to be a plausible thick equity ratio. And the upshot of that is what you really want in the modeln is an equity premium of about 5%, rather than 7, because in the model, the model is talking hypothetically about corporate finance with no bonds; unlevered equity finance.
  • 46:18    |    
    And the equity premium for that should be lower, than the observed levered stuff. So actually, you want the model to deliver an equity premium of 5% as then argument, not in 7, there are some adjustments there.
  • 46:33    |    
    That means that the risk-free-rate is 1% and the one expected rate of return of unlevered equity is about 6%. That's when you declare success, when a modeln delivers that. That's the claim.
  • 46:49    |    
    Audience:So the equity premium is still than more than possible?
  • 46:54    |    
    Robert J. Barro:Well, we don't know yet, let me talk about this slide and you can tell me if you believe any of this.
  • 47:03    |    
    Alright so the halfline, this is for the consumption data, the bottom stuff is about GDP data. So for the consumption data, my baseline specifications, I'mn looking at contractions that are at least 10%, that's where I have the 95 consumption contractions.
  • 47:19    |    
    What turns out to fit is a coefficient for risk of about 3.5. So I've assumed here that this gamma is 3.5. Of course the idea, what it's supposed to be is an reasonable number and then we can think about whether that´s correct.
  • 47:35    |    
    So, if I had 95 disaster events in this particular sample, but there's a duration, thirty and forty-three of the years are in disaster states, if you thinkn kind of the duration. The disaster probability is the number I mentioned before, 3.6% that's for disasters of size 10% or more.
  • 47:58    |    
    The pie is about exiting disasters, I'm not going to use that, so let me forget about it. The average size of disasters is 22%, that's what I mentionedn before.
  • 48:10    |    
    So these are the expectations of the two key moments that appeared in the form of the equity premium, so that's based on the histogram for the disaster sizesn Bs
  • 48:21    |    
    given the value of gammathat I´ve now postulated, which is 3.5. So that's what those things turn out to be and from that in the model I can compute what isn the expected return on equity, this is from the model, and what is the equity premium.
  • 48:35    |    
    So, it turns out that the expected return on equity is that 0.059, is about 6% per year. I´ve rigged the whole thing in terms of time reference to get then risk-free rate to be 1% so that gives you an unlevered equity premium of about 5%. So that's about right.
  • 48:53    |    
    That's what you need to match the equity premium here, you need, given the things that I get from the data about disaster probability and sizes, you need an risk attitude characterized by a gamma of 3.5, that fits, that's the claim.
  • 49:09    |    
    Audience:The 3.5 came from your intuition?
  • 49:12    |    
    Robert J. Barro:No, no, no. I'm looking at different values of gamma, if you raise gamma, you're going to predict a higher equity premium in the model. I'mn asking what value of gamma would match, roughly speaking the observed unlevered equity premium, which is about 5%.
  • 49:29    |    
    So gamma 3.5 is the answer to that. That didn't come from anywhere,except now you can say, based on maybe other evidence, do you think that the risk-aversionn coefficient of 3.5 is reasonable.
  • 49:43    |    
    But here the 3.5 is the number you need to fit the equity premium given all the other stuff, and most important which is the probability of size distributionn of disasters, which is what I've got for the long-term data.
  • 49:58    |    
    So, the other rows tell you what happens if you make some changes from the baseline specifications. So for example, you can see how much difference does itn maket if this coefficient for risk is 3.0 rather than 3.5, I'm just illustrating.
  • 50:15    |    
    Well, then you get an expected return of 4.2, equity premium of 3.2.,so it's noticeably too low. And if you go anything much below 3, you clearly don't fitn at all.
  • 50:27    |    
    So you need risk aversion of at least 3 to get in the right ballpark. And on the other side, if you go much above 3.5, well you really cannot go above 4, ifn you go above 4 you get a much too large of an equity premium model.
  • 50:42    |    
    So actually, it's fairly well confined in terms of gamma that you need, it has to be roughly between 3 and 4 in these results. Otherwise the equity premiumn is either way too low or way too high.
  • 50:59    |    
    The OECD versus the non-OECD matters somewhat more with the consumption results than with the GDP actually. But even for the OECD countries you can getn pretty much the right answer, so it doesn't really depend on having the non-OECD countries in the sample.
  • 51:16    |    
    So then, I looked at what´s the consequence of having different thresholds, which I mentioned the 10% was arbitrary, some might yell " well these are big".n The previous work assumed 15%, also arbitrary; so this shows you what happens if you have higher and higher thresholds.
  • 51:36    |    
    It's not that sensitive to that because it's really the big disaster events that matter in terms of having enough risk to matter in the pricing. So if you gon from 10% which is this baseline, to 15%, you see the next variable difference.
  • 51:55    |    
    So if you go from 10% the probability of diaster is like 3.5%, 3.6. If you go down to 15% you lose a lot of events between 10 and 15, so the disastern probability falls to about 2%, but of course, the average size is much larger because you are truncating at a much higher value.
  • 52:15    |    
    And it turns out that losing the small ones doesn't make that much difference, in terms of what people really worry about in this model. So actually, thisn doesn't change too much. So just looking at the really big events is what has the most explanatory power here.
  • 52:36    |    
    So you can also look at, suppose that you get rid of all the wartime observations, which was mentioned before, well it's systematically true that then wartimes ones are the bigger ones. If you remember from WWII that's particularly true.
  • 52:48    |    
    So throwing out the war observations is like throwing out the upper tail of the beads, and that has a substantial effect on the results. So if you look onlyn at the non war observations, there are originally 95 disaster events, in this sample, if you look at non war you get 66, so you get about two thirds of the total, and only one third is war, butn those are the biggest ones.
  • 53:11    |    
    So that actually makes a big difference; you can't really explain the equity premium if you throw out the war; or put in other way, in a much higher riskn aversion coefficient, you need gammato be 9, which is like what Hall was saying, you don't get out of bed in the morning. So a gammaof 9 is, I think, ridiculous.
  • 53:31    |    
    That's what you would need if you only had the non war observations in the data, which includes the Great Depression, but throws out all the WWII and WWI inn particular. I won't bother with the rest, I'll put it aside, but...
  • 53:46    |    
    So the results are robust to a certain set of changes, but no to things that throw out the biggest disaster realizations.
  • 53:53    |    
    Audience:You seem to have a higher premium for OECD than non-OECD countries, based on the GDP data, right?
  • 54:03    |    
    Robert J. Barro:I don't have a good story for that, it's different, the GDP, you see, is different from the consumption. I don't have a good story for that,n I think it is a coincidence; but I don't have a good story for that.
  • 54:20    |    
    I would´ve expected to get more than an equity premium in a non-OECD, I would have thought it was sort of riskier. And you see that in the consumption datan but I don't know why you don't see that here.
  • 54:32    |    
    So I was focusing on the fact that if you only look at the OECD, you could still get into the right ballpark for the equity premium , we don't need then non-OECD, but that pattern I don't have at this time.
  • 54:45    |    
    Audience:So the equity premium is really a production of war?
  • 54:51    |    
    Robert J. Barro:Well, you need the biggest events and those turn out to be disproportionally the wartimes, even though they're smaller.
  • 54:58    |    
    Audience:Or revolution...? Or revolution too?
  • 55:03    |    
    Robert J. Barro:Well, that can be classified with wars. Like one of the revolutions is the Portugal revolution on 1974, that's called a wartime event in thisn classification. You know, war is somewhat subjective; I treated as though the United States was not at war currently, for example, it's only kind of major wars.
  • 55:18    |    
    So for most of the cases, that's the World War is, in terms of the Spanish Civil War and Japan fighting with Russia, that was another war.
  • 55:35    |    
    Audience:I'm thinking of war that some governments make on their own people, especially against entrepreneurs.
  • 55:43    |    
    Robert J. Barro:Chile in the mid 1970s is classified as a wartime observation, that's called a coup...
  • 55:49    |    
    Audience:Venezuela today?
  • 55:51    |    
    Robert J. Barro:Well I'm not sure how to...
  • 55:53    |    
    Audience:It's not a shooting war yet...
  • 55:59    |    
    Well, the Portugal Revolution is tricky, because I think hardly anybody got killed in the Portugal Revolution in 1974, it's more a massive of event thatn involved coercion and it was very important, I would call it war but, you see, we have casualty rate numbers, but it's not that relative.
  • 56:18    |    
    Portugal was involved in a long collonial war before that.
  • 56:25    |    
    Right, but it didn't have any disasters there so... The big disaster event for Portugal after WWII is this revolution period, that's why it matters how wen classify that. Okay, so this just repeats what I just went through with.
  • 56:40    |    
    This is some discussion about extensions of this baseline research, which, either I or other people, are carrying out currently. So one is about this linkagen between stock market crashes and depressions.
  • 57:01    |    
    One quick issue that comes up there is the following; in using the financial returns data of what I've described so far, I really use that only to measuren what was the average equity premium, and particularly to get this target of 5% for the unlevered equity premium that I said the model has to replicate.
  • 57:17    |    
    I didn't use the financial returns data up until that point for any other purpose, so if you think about the underlying asset pricing, there are somen predictions there that involved, basically the co-variance between the macro variables and the returns data, particularly, the stock returns.
  • 57:35    |    
    And I mentioned before the reason that stocks look particularly bad is because in the model stocks do badly during disasters. But I'm not using the datan about stock returns during disasters to verify whether that's correct or not.
  • 57:52    |    
    I'm just assuming that it's true in getting the results that I've discussed so far. Partly of what this other topic is, is looking at the actual pattern ofn stock returns, disasters and other periods, and seeing whether it relates to the macro contractions in a way that works in terms of the asset pricing.
  • 58:11    |    
    So that's part of what's in this additional study. I mentioned this great influenza epidemic, we´re doing a lot of research here. Jose Ursua, my student, isn particularly doing a lot of work in terms of data for that period, in terms of the macro contractions, but more particularly about the spread of the disease in the mortality incidents byn country and time period.
  • 58:35    |    
    Jose is particularly putting an amazing amount of work in terms of generating the underlying data, which I discussed. So instead of talking about it in 15n minutes, I should've taken days to talk about the data to do justice to this.
  • 58:54    |    
    He really does know every expert now in 36 countries, some of whom are still alive. We called up this guy in Austria named Kalssel, who was the main Austriann long-term GDP person, because there's always one person. And he's like ninety-something.
  • 59:12    |    
    We called him up to ask him something and he said "I don't know you should talk to my student", who turned out to be 72-years or something. But Ursua reallyn knows all these people, it's a different person of course, for each country.
  • 59:27    |    
    There was a discussion about consumption breaking down between durables versus non-durables and services. We have that for about a less than a third of then disaster observations on consumption.
  • 59:44    |    
    By looking at the data where we have it, we made the argument that the error by looking at total consumer expenditure rather than something more liken consumption, is not going to be that large, we argued that it was about 3 percentage points on average.
  • 59:57    |    
    So we're looking at contractions of at least 10%, or at least 15%, so we've made an argument that could be lightly errored in terms of this consumptionn discrepancy, is probably no more than three percentage points or so.
  • 01:00:12    |    
    I mentioned this other data that I'm going to evaluate with Steinsson, who were at Columbia, so this is using a whole time series for consumption and GDPn for every country.
  • 01:00:24    |    
    And one of the things we can do with that is estimate the fact that after a big disaster events, particularly wars, it tends systematically to ben recovered in the sense of abnormal growth reward, that definately is in the data, it's not in the model that I described. So that's one of the things I think the model is really missing.
  • 01:00:41    |    
    This analysis was allowable, it also allows for other things that I discussed before and I discussed this last one.
  • 01:00:52    |    
    Audience:Just a question about that recovery time, do you think there's a possibility of having like a similar market model where after the disaster,n probably some people have limitations to go to the financial markets and probably that can solve part of the puzzle?
  • 01:01:16    |    
    Robert J. Barro:Maybe but, I mean, you have to think about the history of the equity premium puzzle, because Mehra and Prescott started with an representative agent sort of a perfect markets model, like the type that I have, and they conclude it doesn't work.
  • 01:01:31    |    
    So, of course there's been a lot of literature since then trying to come up with all kinds of different explanations, so one set of that literature isn along the lines that you just suggested, I think, about some kind of incomplete credit markets and maybe separate labor market risks, that are not, you can't somehow insure against.
  • 01:01:52    |    
    So, there's been a whole set of papers like that, but I think the reason that whole literature bloomed was because of the original Mehra-Prescott findingn which was that the basic model doesn't work in a very stark, large sense.
  • 01:02:06    |    
    So, I'm taking a different approach, I'm sort of going back to where they were, as to representing each model with one additional feature which is thisn Rietz's type of rare disasters.
  • 01:02:16    |    
    And then I'm planning that that gets into the right ballpark. If that's right, I think that it reduces the motivation to do these other things.
  • 01:02:24    |    
    I'm not saying that they might not matter. Certainly like Constantinides, for example, it doesn´t work like that. He doesn't like this sort of stuff, andn he says what you say, and maybe he's right.
  • 01:02:35    |    
    And certainly there can be a role for that, along with this it could complement, they don't have to be one or the other, and also this business aboutn variations in the long run growth rate, which might matter for example for Japan that had gone down, that might also be a moment of that I don't have.
  • 01:02:54    |    
    So I think it's possible that that kind of approach might matter and it's not the way I'm thinking about it, I'm trying to think about can I go back ton the representative agent model and get a lot closer to what I see related.
  • 01:03:07    |    
    So I think Reitz was unfairly treated, and by the way, he has done nothing since then, in terms of research. He had this 1988 paper, he runs this Iowan betting market, at the Business School at the University of Iowa, which figures what´s the odds of the elections.
  • 01:03:25    |    
    That's what he mainly does, I mean, he did this one piece of work, which I think was very innovative,
  • 01:03:37    |    
    Audience:Why's your feeling when you're working with a set of countries that when you introduce periods of Africa, we don't have the data, we might nevern have the data, but do you think that Latin America...
  • 01:03:45    |    
    Robert J. Barro:South Africa's in the data.
  • 01:03:47    |    
    Audience:Yeah, but South Africa, is there, but... I mean what's your...
  • 01:03:51    |    
    Robert J. Barro:Each of them is almost in the data, you can get GDP numbers back to 1923 or something, each of this all goes there, but I don't think youn can go through WWI. WWI is the problem, it's the same thing with Turkey. I don't know, I'm so sorry, go ahead.
  • 01:04:03    |    
    Audience:I mean what would be the effects of putting these... we don't know right, what would be the effects of including for example, developingn countries where probably crises are more prevalent?
  • 01:04:22    |    
    Robert J. Barro:Yeah, financial markets are much less developed, you're not going to see those data, but you can still think of a model as hypotheticallyn applying, even in the absence of this explicit financial markets.
  • 01:04:34    |    
    Audience:I was thinking of something like the 80's crisis...that's really liked reduced life expectancy and probably like GDP growth and all that. Son there are some crises that...
  • 01:04:42    |    
    Robert J. Barro:Russia would help for that because Russia's life expectancy has gone way down, but that's more related to alcohol, but it's quiten strengthing the numbers for Russia. Russia would be a very interesting case to hear actually, and I think it is feasible.
  • 01:04:55    |    
    But Russia has like 5 different episodes, each of which is a separate PhD dissertation, but I think it is feasible; all the periods you can sort ofn estimate stuff, and I think Turkey Ottoman Empire is feasible.
  • 01:05:13    |    
    Actually the Ottoman Empire dated before 1914, fine, it's the period from 1914 to 1923 that's hard. I don't think you can do much with any more Africann countries. You might be able to do Egypt.
  • 01:05:29    |    
    Audience:India or...
  • 01:05:31    |    
    Robert J. Barro:India is in the sample. India is included. Indian data are very good because they were managed by the British. No, the Indian data aren quite substantial. Even though they've gone long market back to 100 years.
  • 01:05:48    |    
    Let me try to say something of what I'm doing in the stock market stuff, I won't be able to finish it. I'll just sketch some of these results, working atn stock market crashes. So again, I didn´t use this data up until now, I only looked at stock returns to calibrate what was the equity premium.
  • 01:06:11    |    
    Samuelson has this famous quote from Newsweek magazine, "Denigrating the predictive power of the stocks market". It's a very clever quote, but completelyn misleading.
  • 01:06:26    |    
    Think about it this way, suppose you see a stock market crash and suppose half the time it doesn't matter for the macro economy, nothing happens, and halfn the time there's a depression, wouldn't you think the stock market crash data were very relevant?
  • 01:06:40    |    
    I mean, being able to predict half the time right about the depression coming is extraordinary, given that depressions don't occur very often. So that'sn why I think it is a very misleading quote.
  • 01:06:51    |    
    Because he makes it sound like, you know, stocks market are really vulnerable and they don't have anything to do with the real economy, anyway.
  • 01:07:04    |    
    So, part of what this project was, was to look at how much information there is in terms of stock market returns, such as in 2008, for predicting whethern there's going to be a depression.
  • 01:07:10    |    
    So that turned out to be a very timely application, because you can use this to look at the stock market performance in 2008, and say what the probabilityn that various countries are going to experience a depression, in the sense of a 10% or more decline in GDP or consumption, so that was a very interesting application.
  • 01:07:30    |    
    But you can also extend the research related to the equity premium by trying to look at this covariance pattern between stock returns and eithern consumption or GDP growth, which are the two measures of the macro.
  • 01:07:43    |    
    So this is just discussing the data, the stock return data became more important here so we tried harder to get more stock market data. We have this onen source, Dimson, which added some earlier data on stock returns, but those guys are really obnoxious and they won't give us their data and all their stuff, they're very difficult to dealn with.
  • 01:08:09    |    
    And they like Madison because some people use their data but there are certain places where stock markets are closed, and of course the data are missing,n but they never have missing data, because they always make up the numbers that are missing.
  • 01:08:23    |    
    If you ever look at the stock returns you gotta be really careful when looking at the Dimson stuff, but they are useful for some of the earlier data thatn are not in the global financial data, we have a longer time series for some countries, something like 7 or 8 countries that have longer, earlier time series.
  • 01:08:42    |    
    And we recently added Mexico, Argentina, and I think we'll be able to add Brazil in terms of the longer term stock return data. So, it measures stockn market crashes in a way analogous to what we did for depressions, which I'll discuss more in detail.
  • 01:09:00    |    
    The baseline sample, there's 29 countries that have long-term data, on both stock returns and the macro variables, so it's a somewhat smaller sample,n because here we're insisting on the stock return data to go back at least to the 1930s. The problem is you lose too many countries, if you insist on having longer term stock coverage as an trade-off.
  • 01:09:25    |    
    So, if you insisted on going back to 1914, you'd leave out a lot more countries, I wasn´t sure what's the best way to deal with that. So here's ann illustration with part of a methodology, this is the U.S. history.
  • 01:09:46    |    
    This defines stock market crashes using annual data here, as cumulative declines which add up to a rate of return of -25% or worse. So we have real ratesn of return annually on stock market, and if you get a cumulative decline it could be 1 year or 2 years or 3 years of -25% or worse, that's here called the stock market crash.
  • 01:10:09    |    
    And of course you can look at different thresholds for that. So that's analogous to the 10% threshold for the macrovariables. So here we have a 25%n threshold for a negative return. Of course the normal return on stock markets is 0.08 per year, so -25% is really much lower than normal.
  • 01:10:29    |    
    Particularly, if you look at it over two to three years, the normal return should've been something like, say 20% over three years, so instead, you´ren getting -25%, so relative to normal is particularly low.
  • 01:10:41    |    
    So this is the U.S. history goes back to 1870, it shows the stock market crashes coming from the annual data for the U.S., and it´s going to match this upn with what was going on with the macro economy based on either consumption or GDP.
  • 01:10:56    |    
    So, up until 2008, the basic research study was 2006, there were seven stock market crashes in the U.S. by this definition. Actually, the main one that'sn missed by insisting on annual data is 1987.
  • 01:11:17    |    
    The global crash where it was a very sharp, short-term decline in the U.S. and elsewhere, but then it turned out to be temporary and for the U.S., then year as a whole 1987 was nothing special.
  • 01:11:28    |    
    That's the main thing that's missed by looking at the annual data for the U.S. The biggest stock market crash, so this is in real terms, is 1929 to 31n associated with the Great Depression, 55% negative cumulative of real return.
  • 01:11:46    |    
    So this is really from the beggining of 1929 to the end of 1931, so that's really three years, accumulates to -55. That's the worst for the U.S. usingn annual data. Of the seven stock market crashes, and this relates to Samuelson's quote, .
  • 01:12:10    |    
    if you match it up with the macro contractions, I'll focus on the last column, which is sort of averaging the consumption and GDP experiences.
  • 01:12:20    |    
    Since I don't have too much time I'll focus on the last column. There are only two macro contractions that you would call depressions for the U.S., then ones I was talking about before, one of them is the trough in 1929, and the other one in 1933.
  • 01:12:36    |    
    So averaging the consumption GDP, the first one is a contraction of 16%, and the second one is 25%. So those kind of match up with those two stock marketn crashes, the ones that are shaded with the purple. The other five do not have depressions defined as 10% or more contractions.
  • 01:12:58    |    
    So that's the sense in which you can think about the stock market crash is predicting a depression, it predicts seven out of the last two depressions,n that would be the analogue through the Samuelson quote, he was talking about recession which would be pretty mild, not depression.
  • 01:13:16    |    
    If you look at the other five, you can see kind of what they are, the recent is this so called internet bubble, stock market decline of 42% and a minorn recession in 2001, it's what's true.
  • 01:13:33    |    
    1973-74 is another stock market crash, 49% it's a more serious recession, but it's not a depression; 1937 is another recession, but not big enough to calln a depression; 1907 is another recession, almost big enough to call it a depression.
  • 01:13:56    |    
    That's the recession that made the reputation of JP Morgan, he came to sort of rescue the banking system, before the Federal Reserve existed, preminishedn war based on that set.
  • 01:13:57    |    
    If the Federal Reserve had never existed we would have better outcome, we would've had something like the 1907 banking crisis, which was not so good, butn not nearly as bad as a depression.
  • 01:14:18    |    
    So 1907 was a stock market crash, almost a depression based on a... so there´s clearly some association between stock market crash and depressions, butn only 2 out of the 7 have what we call depressions.
  • 01:14:36    |    
    And then you have 2008, the U.S. return was -37% in the calendar year, and then it did really bad in the first two months of 2009, it rebounded and triedn bringing it up.
  • 01:14:48    |    
    So actually today, it's about the same point as it was at the beggining of the year, January 2009. So it's about the same point right now as it was at then beggining of the year, overall.
  • 01:14:59    |    
    Audience:This is breakdown equity premium, long-term equity premium, and short-term equity premium?
  • 01:15:12    |    
    Robert J. Barro:I thought you were going to say something about wars and non wars which you might before, because that does matter a lot actually, then predictive power of stock returns for depressions depends a lot, and whether there's a war or not.
  • 01:15:24    |    
    I thought that you were going to say that, and I was going to agree with you. It depends on some other things, like whether or not there's a financialn crisis and we´re sort of looking at more things, but I don't know about this short versus long term, I don't know how to bring that in here.
  • 01:15:38    |    
    Audience:Can I give you a short example?
  • 01:15:42    |    
    Robert J. Barro:Okay
  • 01:15:44    |    
    When you invest in Venezuela today, you'd want to have a return of capital as fast as possible because in that situation, they change the rules of then game everyday, whereas if you invest the same amount in Switzerland you would be aiming at the return of your capital long-term, because it's rules will be the same.
  • 01:16:06    |    
    Robert J. Barro:But I don't think the short-term return of Venezuela is all that secure either. It's like musical chairs, you don't know when the thing isn going to explode. I don't think is necessarily true that in short-term the stock return in Venezuela is more secure than in the longer term, I'm trying to argue that. It could be then opposite.
  • 01:16:28    |    
    I'm almost out of time, maybe I should say a few more things about this... I won't have time to finish this...Let me go through a few more of thesen results.
  • 01:16:43    |    
    Audience:Could we say that based on the U.S. data that if there's a stock market crash there´s a 2/7 probability of a depression is...
  • 01:16:50    |    
    Robert J. Barro:That's the spirit of what I'm trying to do, not just look at the U.S. The other thing I neglected to mention; there are no depressions inn the U.S. data without associated stock market crash, that's not true for the broader sample, but for the U.S., if you knew there was no stock market crash, you´d know that there's non depression.
  • 01:17:15    |    
    On the other hand, there are several, five, stock market crashes, without depression, so stock market crashes are certainly not a perfect predictor of an depression, but you ought to worry about it more than normal when you see...
  • 01:17:32    |    
    Audience:Isn't the stock market crash a consequence of the depression? Like the depression can predict a stock market crash?
  • 01:17:39    |    
    Robert J. Barro:That's a causation kind of question, which is very interesting, but suppose that I am in January of 2009, and suppose the only thing In know, the only thing I´m looking at is that the stock market fell by 40%.
  • 01:17:52    |    
    And then, in terms of the timing, what usually is true, not always, is that the stock market crash is usually earlier in time than the depression, if youn think about accumulating for several years, usually the timing is that you know about the stock market crash before you know there is a depression.
  • 01:18:08    |    
    So in terms of predictive content, I'm trying to ask how much would you know just by looking at the stock market crash about whether there's probablyn going to be a depression, but you can also bring in other factors, but just by knowing that there's going to be a lot of information.
  • 01:18:24    |    
    So, I'm not saying there's a causation here, I'm not saying that the stock market crash caused the depression, I'm not saying that at all, it's probablyn mostly a signal, a symptom of other stuff that's affecting both, the stock market and the real economy.
  • 01:18:37    |    
    So this is what the broader set looks like. So now, I'm looking at 29 countries where I have both, long-term stock return data and the macro variables.n There's some arbitrariness about how you match the things. So you see, for the U.S. actually it's pretty straight forward.
  • 01:18:57    |    
    So for the U.S. I've matched the 1929 to 31 stock market crash with a macro contraction which is 1929 to 33. That's sort of a common pattern where then stock market crashes earlier but there's sort of roughly, otherwise coincidence, matched in time.
  • 01:19:19    |    
    So that's a common but not universal pattern. And then if you take the other one, I have a stock market crash from 1916 to 20; and a 1917 to 21 macron contraction, so that's also a similar kind of pattern.
  • 01:19:32    |    
    So for the U.S. it wasn't too hard to match those, but in many other cases, it's not so straight forward and I'm not sure exactly what the right way is ton do it, but of course, we did it anyway.
  • 01:19:45    |    
    So now I'm taking 29 countries, not just the U.S.; the U.S. was just an example of the methodology, we take all the 29 countries, there are 69 cases ofn matched stock market crashes and depressions, where the stock market crashes are defined as cumulative return over one or more years, of -0.25 or worse.
  • 01:20:09    |    
    And a depression is a contraction by 0.10 or more, 10% or more, I'm averaging the GDP and consumption results, which I think is roughly the best way ofn doing this. So there are 69 matched cases where stock market crashes are related to depressions, but there's not always this nice pattern like we saw for the U.S.
  • 01:20:33    |    
    In fact a lot of times when the pattern is messed up, it´s because of price controls, particularly associated with the wartime events. Germany turns outn to have some of the biggest price controls during WWII, but the Nazi Government actually instituted price controls in 1936, before the war, and then eventually...
  • 01:20:54    |    
    So that was holding down the inflation, but eventually they put price controls on the stock market numbers and they didn't want their return to look toon negative so they were holding it up; and consequently, when the price was freed up there was a tremendous, like a 90%, decline in the stock prices.
  • 01:21:13    |    
    But that messes up the timing, not necessarily the cumulative chain but the timing, so I think, like for Germany in WWII, if you take a long window then stock market crash and the economic contraction are probably reasonable in terms of how to measure it. But the timing is all wrong because of these controls, but I believe it is true.
  • 01:21:34    |    
    And there are some other measurement issues, which I think are important, mess up some of the time. In Portugal the stock market closed between 1974 andn 1977, so I can compute the cumulative stock return over the whole period which turns out to be an amazing -97% in real terms, but we don't have the timings because the market's closed.
  • 01:21:59    |    
    So when France's stock market closed as part of German occupation, but not the whole thing, so there are a number of cases like that which I think cann mess up by timing. Anyway I have these 69 cases.
  • 01:22:14    |    
    But a fairly flexible view would have matched the timing of the stock returns and macro thing. This flexible thing is critical for this and has somen arbitrariness in it.
  • 01:22:25    |    
    So the U.S. had no cases of depressions without stock market crashes, but the broader sample has 27 of those; 27 depressions not associated with stockn market crashes, and these are examples of those.
  • 01:22:44    |    
    To some of these, these are all negative numbers for stock returns, I just don't have the minus signs. So some of these have pretty big stock marketn declines, but they don't make the threshold of 25%, so these are cases, none of which are up for the U.S.
  • 01:23:02    |    
    South Africa is in this particular sample because they have long-term stock return data that go with the macro variabes
  • 01:23:15    |    
    Then the final category, which is of interest, is there are 157 stock market crashes not associated with depressions, so that's like the five U.S. cases,n which Samuelson was mentioning, and it's certainly true, there are a lot of stock market crashes without depressions.
  • 01:23:31    |    
    So, I have these three categories I care about, and of course the fourth category is neither a stock market crash nor a depression, which is not of an particular interest to me actually, it doesn´t matter too much, given that I ´m a negative person, I only look at bad things in one place or the other.
  • 01:23:47    |    
    Well this is something about what the statistics look like, let me summarize the kind of results on this graph which is preliminary because we have somen more results that are no incorporated in this.
  • 01:24:05    |    
    So I'm looking here at the size of the stock market crash, so I'm looking at things that are 25% or more, but it matters how big the crash is, so this isn kind of the associated probability, let´s take the non war sample here on the left.
  • 01:24:22    |    
    So if you see a stock market crash of 25% or more, the red bar says what´s the probability of seeing a depression, associated with that of at least 10% orn more. That's what the red bar says.
  • 01:24:36    |    
    The blue bar says what we see at the stock market crash of 25% or more, what's the probability of a bigger depression; 20% or more, of course that's an small number, and then the black is 30% or more, that's more like the U.S. Great Depression.
  • 01:24:53    |    
    So these are the kind of conditional probabilities of depression of various sizes, assuming that the only thing I know about is how big the stock marketn crash is, this could be implemented because often, you know the stock market crash before you know what the macro results are. Like today on the U.S., I think that's true.
  • 01:25:13    |    
    If you look at bigger and bigger stock market crashes, of course, the probability of depression goes up, the conditional probability of depression, so ifn you have a stock market crash of 50% or more, the conditional probability of a 10% or more depression is almost 40%, so as you would expect, bigger stock market crash raises the odds.
  • 01:25:40    |    
    Now, wartimes ones are here and non war over here, they're completely different in this case, but in many other respects they're not so different.
  • 01:25:49    |    
    But for this exercise the war examples are totally different. If you have a war situation with a stock market crash, the odds for a depression aren extremely high.
  • 01:26:02    |    
    Because basically a stock market crash during the war is telling you you're not doing so well during the war basically, that's really bad for the macron economy. I focused on this side because applictations of this today, I'm mainly looking at non war environments.
  • 01:26:21    |    
    So again, I'm thinking about the U.S. as not being involved in a war, that's relevant to this sample. So the yellow arrow is roughly speaking where then U.S. is today, and many other countries, most countries have stock market crashes, a lot of them had 40% and it varies.
  • 01:26:44    |    
    So the conditional probability of a 10% depression, given that is of 30% which is very high, I mean, the other side of 30% is that 70% probable you won'tn have a depression.
  • 01:26:53    |    
    That's unfavorable if you're looking at it in aghast. But 30% is enormously higher than usual probability of a depression which is a trivial number for an typical year.
  • 01:27:04    |    
    Audience:But 30% is 2/7.
  • 01:27:08    |    
    Robert J. Barro:Oh it's exactly the same as the U.S.? Okay, yeah it is 2/7, so the U.S. experienced this with this bigger universe.
  • 01:27:20    |    
    If you asked today what's the conditional probability of 20% depression, is about 9%, so that's of course much smaller but still not trivial. I wasn talking with a reporter once and I had this result about the 30% and that was in this Wall Street Journal column, and you know, journals like things like that, like 30% chance of an depression.
  • 01:27:46    |    
    So this reporter was asking me "what do you really think is going to happen...?" And I said, well the best I can gage from what I know is that then probability of a depression is 30%, and she said "yeah, but is it going to happen or isn't it?" And... you know I can't talk about the probabilities, not admissible.
  • 01:28:08    |    
    So war and non war obviously matters, we're looking at whether other things matter, like if the thing is accompanied by a financial crisis that does raisen the depression odds.
  • 01:28:20    |    
    So if you look at non war results accompanied by a financial crisis, then it does somewhat raise the odds, another thing is whether it´s a global versus an localized crisis, that doesn't seem to matter very much for this odds actually.
  • 01:28:32    |    
    And thinking about that you might observe that we're already looking at the stock market price movement, which itself is related to whether it's local orn global, and I think that's why once we've got that, it doesn't seem to change the information about whether it´s local or global in terms of crisis, but we're still looking at that in mostn surveyed cases.
  • 01:28:52    |    
    So I have to wrap up. Let me just give you one more... One other point is that in reverse, if you see a big depression it´s almost sure to be accompaniedn by a stock market crash, so that´s like the U.S., all the depressions, all two, had stock market crashes.
  • 01:29:14    |    
    So that does generalize that the bigger depression you see, the more likely it's true that it was accompanied by a stock market crash.
  • 01:29:21    |    
    A 25 to 35% stock market crash, you see a big depression, the odds that it's accompanied by a crash is something like 70 or 80%. Putting it the other way,n if there's no stock market crash you can feel pretty safe that it's not going to be an associated depression.
  • 01:29:44    |    
    So the final thing I did here, which I won´t go through in detail, is I tried to ask the question, going back to where I did the results on the equityn premium, within the Lucas type model, so that was not based on looking at the actual stock returns that you saw during disaster events or other times.
  • 01:30:10    |    
    This is asking, suppose I try to look at the pattern about stock market returns and depressions, did they co-vary in the right way, as the theory implied,n to get the right answer with respect to the equity premium.
  • 01:30:27    |    
    The basic answer to that is if you take this flexible view, don't insist on a particular timing always being the same, if you take the matched cases as In constructed them, it does work basically, and it does have the right covariance.
  • 01:30:42    |    
    But if you insist on having a fixed timing, like stock returns based on macro contractions a year later or two years and two years and whatever, thatn doesn't work, that doesn't work; the timing is not that fixed in terms of how the stock crashes relate to the depressions.
  • 01:31:01    |    
    So I'm not sure about the final answer on this.
  • 01:31:04    |    
    Audience:So we may conclude that you are in favor of making love and not war?
  • 01:31:10    |    
    Robert J. Barro:You didn´t need the seminar for that. Okay, I'll stop at this point since I'm out of time.



Share

Video:

Embed:

Macroeconomic Crises and Disasters

16 de julio de 2009   | Vistas: 7 |  

About this video

In the second session of this four-part series, Robert J. Barro continues his discussion on his current research into the effects of macroeconomic crises and disasters in different countries throughout history. Barro’s ongoing research demonstrates how different economic indicators fluctuate through time during significant historical events. According to Barro, wars, epidemics, and other rare disasters affect both a country's GDP and its per capita personal consumer expenditure. He also analyzes the interplay between depressions and stock-market crashes and explains how it is now possible to foresee how an economy will behave during future disasters and how indicators around the world will be affected. Finally, Barro explains how some variables, like consumer preferences and culture, may affect the outcome of a crisis.


IDEAS DE LA LIBERTAD

Nuestra misión es la enseñanza y difusión de los principios éticos, jurídicos y económicos de una sociedad de personas libres y responsables.

Universidad Francisco Marroquín