Conferences & Events

The dynamics of asset returns have always had both a systematic component, arising from the overall direction of the economy, and an idiosyncratic component, arising from the quality of individual investments. As the asset returns have become much more interdependent in recent years, identifying the dominant sources of risk and their impact on the shape of the asset distributions is critical for market participants and policymakers.

Transcript

David Zervos: Hi, my name is David Zervos. I'm the chief market strategist at Jefferies. It's a pleasure to be back at the Atlanta Fed conference, which I think I've been at three or four times, and unfortunately missed last year. So I'm very excited this year to be here, listening to all the panelists and the discussions and the dinners. It seems that the panel that I've been asked to moderate, if I read the synopsis correctly, is something in a more glib way, maybe: did QE [quantitative easing] make all asset correlations go to one? And I'm tempted to say that there is a lot behind the statement that Fed policy, or monetary policy in general, has driven a lot of correlations into the market that were not there before, and certainly made a lot of assets that were not correlated, correlated.

We have two guests—we had three, but one...we'll just have to have his spirit here from the Board of Governors, given that he's one of the QE people. We have two people here who are going to speak on correlations, and we have a little bit more time for questions so we can open it up to some other things, and I'll try to fire some questions in there for these guys as well. I'm going to let them each come up and give their thoughts on correlation, and risk hedging around correlation and the likes in the markets. And then we'll tackle them with some questions at the end.

So first up, from the Atlanta Fed himself—Nikolay, please.

Nikolay Gospodinov: It's a great honor to be here and part of this panel, at my favorite conference. I'll talk about asset co-movements, but my focus will be less on the observed patterns of these co-movements and more on the underlying structure and risk factors that are behind, that the driving these co-movements. So most of my discussion will be along these lines.

The type of co-movements I'll be talking about are within asset classes and across asset classes, and across international markets. We did an asset class, for example—I'll use as an example small caps and large caps in the S&P 500—so you can think of this as co-movements across sectors in the equity space. For the across asset classes, I'll talk a little bit about these bond/stock rotations and the "risk-on/risk-off" view of markets. And I'll talk a little bit about international markets as well, but it's important also to point out that these are co-movements that I'll be considering not only across different markets but also across different parts of the distribution. So some of them will be related to movements in the second moment, such as covariance and correlation. Some of them will be related to tail risk and movements along the whole distribution.

So there are multiple sources of risk that are driving these asset co-movements, and I'll categorize them in terms of the horizon. And then we'll do long-run, fundamental, and structural factors that are driving these co-movements. There will be some transitory factors, and they will be reflected mostly in the asset risk premium. And there will be some spurious co-movement as well at the very short end. So I'll elaborate on all these points in my next slide—and I don't have to convince this audience about the importance of these co-movements for the asset allocation portfolio management, but they're also important for policymakers as well.

So again, the basic picture is that there will be a multidimensional factor structure that is driving all these asset co-movements. What we observe are the asset co-movements, or some measure of them, such as correlation. And what we would like to do basically is to infer some information about these hidden, unobserved risk factors that are driving these co-movements. So this is basically the main point of my talk.

So in this—I don't know whether it's visible—but in this overloaded table here I'm trying to summarize some of these observations to provide this crude taxonomy. I'm classifying the co-movements by frequency. I'll start with low frequency, medium frequency, and high frequency. And for each frequency I'll talk about the source of the co-movement, the type of the co-movement, and the effect on economic agents.

So for the first row, for example, low frequency...these are the factors that are of business cycle duration, and I'll talk a little bit even lower frequency than that, related to demographic factors. Geopolitical factors are also part of this story. This covers a long span of data, so in a sense it allows us to extract the signal quite reliably, quite accurately. So these co-movements are stable, and we use them for policy purposes. The long-term investors are also interested in these types of co-movements.

There is no clear delineation between these transitory and low-frequency factors. But this medium frequency, you can think of roughly as one- to three-year duration. This will be transient factors, as they may turn out to be long-term factors eventually. Or some of the long-term factors may turn out to be transitory, if there is regime switching. So they are very difficult to distinguish, and I'll give you a few examples to elaborate on this point, but you can think again of this as shorter-term market fluctuations, some political events that don't turn into geopolitical risk, some behavioral factors as well. So put them under the category of sentiment. And these medium-frequency movements, they force some short-term asset reallocations and induce some movements in the risk premium, and they're somewhat stable, so they're not quite stable. So it takes time before we decide whether this is the risk factor that we're interested in or not.

And finally—and this is the most challenging category—this is the high-frequency movement that we observe on a daily basis. We read about this in the paper. And if it was difficult to distinguish between fundamental and market-specific factors, it's even more difficult to distinguish now these market-specific factors for some spurious short-term movements in these fluctuations—they are largely unstable. And I'll give you a few examples in which it's possible that there is spurious time variability so we can confuse the signal with the noise in this type of...and despite this, there's a tendency to ascribe some fundamental structure to this type of co-movements.

So this is basically the framework that I wanted to lay out to organize our thinking about the problem. So I'll start with an example—by the way, my examples will start in reverse order, so the most challenging case, to me at least, is the high-frequency case, so I'll start with an example in high-frequency. It's a fairly recent example—it's developing now in real time. This is the 60-day rolling correlation of S&P 100/600 returns—so large caps and small caps—and if the vertical axis is not visible, it goes from 0.5 to 1. So it started around 0.5 in 2000 and it increased close to 0.95, and it stayed at high level for a long time,. And only recently, toward the end of last year and the beginning of this year, there was a sharp, sharp drop in this correlation. It dropped from 0.9 to close to 0.5.

And these are headlines from some newspapers that pointed out this correlation—either for this particular example (for small versus large caps), or some cross-asset correlations as well. And the correlation dropped to 0.5 then recovered, so it went back up again. I'm not sure whether it's visible, but it's now up to 0.75. So the question is whether this is a stable change, stable breakdown in this correlation, or is this something related to noise and this type of thing—this is what we are trying to investigate—and is it possible to ascribe any fundamental or transient structure to this type of correlation?

And the last question that I'm asking, whether this co-movement is a genuine feature of the data or statistical fiction. I just want to make the point that in many cases these transient co-movements, and completely spurious co-movements, appear observationally equivalent. Just to illustrate this point with a graph. So I'm replicating exactly the same graph from the previous slide. On the left—this is the actual data—and on the right I simulate data that calibrates all the statistics to the actual data. So the mean and the unconditional covariance of S&P 100/600 returns are the same. The only difference is that on the right graph, I have constant correlation.

So again, we observe quite a bit of time variability, even though the underlying structure is a constant covariance structure. I observe quite a bit of variability. I observe similar drops in these correlations, and they arise purely from the fact that these co-movements are not observed, they are measured. There is a statistical measure behind it—this is a correlation measure here. So estimation error can overlap over this 60-day horizon, and can lead to this time variability—even though the underlying structure is constant, and if there are movements in the tail as well, they will generate this type of drops that we observe in the actual data. So you have to stare really hard at these two graphs in order to distinguish them reliably. To me, they look exactly the same.

So by saying that, in no way am I suggesting that there is no drop in the correlation. It is just that we have to approach just a little bit more cautiously, and we have to take into account this estimation uncertainty and model uncertainty when we are making decisions for asset allocations of this type.

So now I'm going to go to an example that was very widely discussed in the press. This is the oil shock in 2014 to 2016. So again, if the horizontal axis is not visible, it goes from the beginning of 2014 and it goes all the way until April of this year. The red line is always the log of oil prices—so the oil prices over this period between 2014 and 2016 dropped by 50 percent or more. And I plotted these oil price with several other assets, such as five-year, five-year forward breakeven inflation, the U.S. dollar index, the S&P 500—this is a one-year change for this index—and then some high-yield bond index.

So you see that, before 2014, the correlation was kind of weak. Then it became very, very strong. Sometime in the beginning of 2015 to the middle of 2016, it exhibited very strong correlation with these assets, and in these oval regions I put basically the period in which we believe that this correlation broke down again. So there was weak correlation, then a very strong correlation, and then a very weak correlation towards the end of the sample.

So the question is, was this co-movement due to some fundamental factors, or was it due to some market-specific or technical factors, changing market structure? In order to answer this question, I'm going to focus on only the top left graph here—this is the five-year, five-year forward breakeven inflation. There was a lot of discussion of that during this period 2014, 2015, there was a growing concern that inflation expectations are declining, and this should be a concern. And just to put this in the background—so five-year, five-year forward, this computed from index-adjusted bonds and nominal bonds. And five-year, five-year forward is the expected inflation over a five-year period, five years from now. So it's a very long-term inflation expectation. All the short-term variations are washed out, basically. So the premise of the argument basically is that if this is due to fundamental factors, these inflation expectations should be affected. If they are not affected, then it means that it's more likely to conclude that these are transient or market-specific factors.

Breakeven inflation is not a pure measure of inflation expectations. It contains also liquidity and risk premium. With some of my colleagues here in the room, we did the decomposition that decomposed this observed breakeven inflation into inflation expectations, liquidity, and inflation risk premium. Inflation expectations turned out to be fairly stable over this period. Most of the dynamics came from changes in the inflation risk premium. But inflation risk premium changes were low-frequency movements, whereas the short-term variations that we observe over this period 2014–16 came mostly from what we call this liquidity premium. The liquidity premium basically is a very general term here. It captures liquidity effects in the TIPS [Treasury inflation-protected securities] market. It captures some specificities in the TIPS market—this market is index-adjusted, it contains some seasonal care, et cetera, et cetera. But it also captures these general market effects, such as forced liquidations during this period, reallocations, hedgings, et cetera, et cetera.

So here I'm plotting basically only the liquidity premium, with the oil price. And you see that most of the correlation is picked up by this liquidity premium only. So even though we suspect that during this period this correlation was driven mostly by technical factors, there is still concern about financial stability. If the markets of the different assets are driven by such high volatility assets (such as oil price), this should be a concern for policymakers.

Also there is some evidence that some fundamental—I will call structural—factors were at play as well. So there is a recent paper by the Bank for International Settlements that argues that during this period of zero lower bound, oil and equity returns have become much more sensitive to macroeconomic news and policy announcements than the period before that. And there is also some evidence that the transmission mechanism of propagating these oil shocks through the system has changed due to the increased role of this domestic oil production. So in these latest computations, Lutz Kilian concluded that the positive net effect on consumers as a result of this oil shock was almost fully offset by the negative effect on oil production—so the effect was around zero.

And the last thing that I want to point out is that it is possible that this co-movement is purely coincidental. I just want to bring this point with this graph. So in this graph I plotted two series. The question is, if you cannot read the text below this, whether you see any co-movements in these two series or not. Honestly, I see quite a bit of co-movements. There are local trends in both of the series—when one of them decreases the other one decreases as well. The truth, however, is that these two series are generated as completely independent processes. The co-movement here is driven by the persistence of both the series, so what I did, basically, was I generated these two processes as independent processes—no co-movement between them whatsoever—but they match their persistence to the persistence of the observed oil prices, and observed breakeven inflation.

Both oil prices, if we work with levels and breakeven inflations, are highly persistent processes. So persistence drives this spurious co-movement. And in this particular example that I give you, if you see any commonality this is completely spurious. None of this exists, because this is how we generated the data.

So the lesson, basically, here is that we always have to compute returns when we're studying co-movements and common variation in this. This is typically done in equity space, so we always use returns in equity space—this is not quite the case when we go to cross-asset computation. Often we use bond yields, which are very, very persistent. We use oil prices and levels. We use VIX. We use risk premia and term premia—all these variables are very persistent, so if you're interested in co-movement we have to transform these variables first, extract the signal, and then if you want to get back to levels then you have to integrate them again.

One challenge—two of the challenges, I already pointed out. The third challenge, basically, when we're looking at these co-movements, is that we typically look at the second moments—covariance and correlation. There is very interesting co-movement in the higher moments as well, throughout the whole distribution—particularly in the tails—and this is still kind of underappreciated and understudied.

In the remaining five to ten minutes, I will talk a little bit more about a longer-term perspective on these co-movements and how to extract common factors from, common variation from, these expected returns. And it looks like a natural question. John Cochrane pointed this out a few years back, that this question—what is the factor structure of expected returns across asset classes?—is a fundamental question in finance. The theory provides very little guidance, so we typically model each market separately. And it's difficult to figure out what is exactly the common structure across the different assets. So it becomes an empirical question.

But even when we do this empirically, there is still tension between theory and empirical evidence on price economic risks. So I'll give you an example. For asset price it models for equity returns, so when we take these models of equity returns to the data—including both traded factors, such as Fama-French factor, the market factor, or macroeconomic factors, such as nontraded factors—all of these models are strongly rejected by the data.

So, to me, this is not a concern. In fact, the models are just approximations of a very complex reality. So we intentionally misspecified the models, we intentionally constructed them as partial maps of something, to emphasize particular aspects of these phenomena—or just because the underlying structure is completely unknown, or unknowable, to us. So the fact that the models are rejected by the data is not a concern. But once you account for this estimation and model uncertainty, what happens is that there's almost no evidence of price macroeconomic risk.

And all of these asset pricing models—fairly sophisticated asset pricing models—are about modeling this relationship. And my view is that there is a mismatch between the frequency for which the models are designed—and this is low-frequency—and the frequency at which the models are evaluated empirically. This is monthly frequency—and to many of you, probably the monthly frequency is a very low frequency. To academic researchers this is a kind of high frequency—so what happens is that at monthly frequency, the returns are very volatile. And we're trying to explain them with very smooth macroeconomic factors, and our statistical models are unsuccessful to do this.

So what I'm going to do here is to take a model-free approach and try to match the frequency of the macroeconomic factors with asset returns to see whether there is some interesting risk factor structure in these returns. So the first graph that I'm producing here is using four assets. This is at daily frequency. It goes from 2000 to the end of April this year. These are stock returns, bond returns, commodity returns, and currency returns—so again, they are all in returns. I extract the common factor in these variations, and then I integrate the series and smooth the series to get to this graph here. And what you can see—in fact, you cannot see it here, but there were supposed to be shaded areas that indicate the NBER [National Bureau of Economic Research] recessions—so all the drops that you see in the series, they should have shaded areas here.

So there is a very strong business cycle co-movement in these asset returns. Again, it's not surprising, probably, to many of you, but in light of this comment on the previous slide when I said that most of the asset pricing models don't find evidence of price macroeconomic risk, this is reassuring that there is a very strong business cycle variation in asset returns.

I have these four assets. The largest loadings are on the stock index, on the stock returns, the bonds and mortgages. There is very small weight on the dollar index. So on this next slide, I superimposed—the blue line is the same as in the previous graph, computed from U.S. domestic assets, and the red line is international stock returns. So we'll do the same exercise: so I collect five international stock returns. They're all converted to U.S. dollars, but it turns out that there is very little difference, whether they're using local currency or U.S. dollars. And partly this is explained by the small weight of the U.S. dollar that was on the previous slide.

So you take the S&P 500 again, UK index, Japan, Germany, and this emerging market index. I again construct the common factor in these returns and I superimposed this. So to me, again, it was kind of surprising to see this very strong correlation between the global factor and this domestic U.S. dollar factor.

So again, to many of you this may not be a surprise. But this lends support to the finding in the recent literature that there is substantial international market integration after 2000, and this figure kind of confirms this. It's interesting that the largest loadings are not in the U.S. stock index, but on the UK, and Germany—and emerging markets, in fact. The Nikkei index contains a very large idiosyncratic component, so it doesn't contribute much to this common movement.

And this last figure here: I superimposed again a third line—this is the thick, sort of orange line—which is computed from completely different data. This is a collaboration with the New York Fed, my colleagues at the New York Fed. We compute something like a turning point business indicator, and it's computed purely from labor data—a large cross-section of labor data. And what we do is smooth a little bit this index, and we're interested in the turning points. And again, it's very unfortunate that these shaded areas for the NBER recessions don't show up here, but this index provides very good guidance. When it turns, it indicates some recession coming—or expansion, if it turns in the other direction.

But what, again, is striking to me is this very, very strong relationship between the asset returns and these business cycle indicators. So there is a very strong business cycle component in asset returns. And this is sort of, again...it's an informal approach, a model-free approach, but it provides very strong indirect evidence that macroeconomic risk is important for asset returns—it plays a big role for this.

Is there any common variation in asset prices, even at lower frequency than the business cycle? So if you think of a business cycle between six to ten years, this is even at lower frequency. And to look at this I'm going to look at some demographic trends, and whether they are somehow related to what we observe in asset returns as well. The usefulness of these low-frequency demographic variables for stock returns has been established in the literature, and typically the argument goes that they affect directly the savings rates and the risk preferences of savers. And they can explain and also predict these low-frequency co-movements that we observe in stock market valuation ratios.

So dividend price ratio and earnings price ratio are very slow-moving variables. They're very persistent—not like the stock returns—and it turns out that these demographic trends provide some information about these valuation ratios as well. In a different strand of literature, demographics also explains the low-frequency movements in the labor productivity, as well as technology shocks.

So what I'm going to do here—and this is going to be my last picture—is, I'm collecting kind of old data on stock and bond returns, valuation ratios. The stock and bond returns are already returns—there's no persistence for the valuation ratios. I'm taking the changes to get rid of this persistence that I was talking about before, and I take demographic population data for the period from '46 to 2016.

The usefulness of this demographic data is also that it provides projections for these demographic trends until 2060. So I do exactly the same exercise as before. For the blue line—I think it's the blue line—the blue line is the common factor from asset returns. So stock returns, bond returns, valuation ratio—again, I extract the common factor, I smooth this common factor, and this is the picture of this common factor from 1946 to 2016.

And the red line is this proxy for the demographic trend. The most stable proxy is believed to be this middle-young ratio. This is the ratio of people age 40-49, over people age 20-29. And this is the red line here, without smoothing. And again, there are only three or four effective observations here on this graph—there are three turning points. But again, it's just striking to see this strong co-movement between the demographic variables and the stock returns, so there's obviously a very low-frequency component in stock returns as well.

And the useful thing about this projection is that I can look—and again, this is a very simplistic model. I don't want to put too much weight on this, but I have projections for these demographic trends. So what it is saying basically is that until 2020 this middle-young ratio will continue to drop. And the argument is basically that this will exert some downward pressure on stock valuation ratios, so we sort of expect, purely based on this simplistic model, that the asset returns over this period until 2020 will remain low—including the interest rates.

After 2020, the middle-young ratio is projected to start to increase again until 2040. And again, during this period it is believed that this down pressure will diminish, and even reverse. And I don't plot what's going to happen beyond 2040, but again: from '24 to '26, this middle-young ratio starts to decrease, purely from these demographic trends.

So this is basically the way I'm thinking about this problem. Aagain, there is a multidimensional factor structure that is underlying all these observed asset co-movements. Can we learn something about the underlying risk factors by observing these asset co-movements? So I'm arguing "yes" at low-frequency—we can identify some very reliable low-frequency components related to demographic trends, business cycles. At the short end, it's a little bit trickier, but still with a little bit of work one could isolate the spurious variation from this market-specific variation in the...

So these are open questions to the panel, in principle. There are many interesting things that one could say, right? There is no way we can approach or explore this topic in full generalities. There are many things that I left out of this presentation, this discussion, but it's interesting to spend a little bit more time to identify what are the underlying drivers of these co-movements.

Is the co-movement just a direction between, let's say, S&P 100/600? Or is there co-movement in volatility, is there co-movement in the tail? Et cetera, et cetera. So we have to incorporate information from the whole cross-sectional distribution base to answer these questions. Also, for these transitory co-movements—they are short-lived by definition, so again, I'm curious to hear from market participants. So if I see a sharp drop in these correlations, do you take an immediate action or do you wait a little bit for a particular trend to form, and how long this is going to persist?

I have in the paper that is coming with this presentation in which I have a numerical exercise in which I do optimal mean variance portfolio allocation versus fixed-weight portfolio allocation that was discussed also this morning. I compute the sharp ratio, taken out of a sample period and see which one is doing better. And it turns out that this fixed-weight, in which either you give 1 over N—with N being the number of assets—weights to each asset, is performing quite well. And the reason for this is that there is a lot of estimation error in this optimal mean variance portfolio allocation. Again, this is fine as long as we explicitly acknowledge this estimation error. We incorporate this into our decisions—this is fine. But if you ignore this estimation error, you can spuriously conclude that something is behaving better than the other.

There are some interesting feedbacks, and Dave started with this, basically the concept that there are interesting feedbacks between the observed co-movements and the underlying financial or policy landscape. I'm not going to say much about this, but it affects also whether these co-movements affect these passive/active investment strategies, or it's vice-versa because people switch to passive investment strategies, this affected, somehow, the correlations between these assets. So there are interesting indulgence feedbacks that I'm leaving open.

And I just want to conclude with this interesting fact: factor investing has become very popular, so it is interesting to look at the dependence structure of these factors as well. But I'm going to basically close the circle with this last statement. This is coming from a paper by Richard Roll in 2013, who claims that if you construct the spread factors—which are basically mimicking portfolio of this underlying risk across highly heterogeneous asset classes. They provide probably the best proxies for this underlying risk that I'm talking about, and they're very likely to span the whole factor space.

So my whole discussion was how to identify these risk factors. Once you identify the risk factors, maybe you can construct this mimicking portfolio, that mimics the dynamics with traded factors, that mimics the dynamics of the underlying macroeconomic factors, and then you can use this for investing. This is exactly what this factor investing is doing. And I'll finish here.

Zervos: Thank you, Nikolay.

[applause]

Michael Mendelson: I'd like to thank the Atlanta Fed for inviting myself and Ernst Schaumburg to prepare this presentation and to maybe answer a few questions on the panel, I hope. I know we're missing one of our panelists today, but certainly systemic risk for asset managers is a lively topic and I hope we can get some questions on that anyways.

Nikolay was talking about trying to understand the co-movements of assets. And as asset managers—and I'm going to try to give a practitioner's view of some of these issues today—as you know, we obviously care about the covariance of assets. And in particular, and part of the case I'll make is that, that diagonal term—the variances—is often what is more telling about what our actions are. And our questions about the off-diagonal terms—which are some of the questions that he's been asking, like spurious changes—lead us often to not be as reactive to those correlation changes as you might guess. In asset management these days, I think they recommend that we linger on the disclosure pages. [laughter] And I'm not sure, and I don't want to upset our compliance department.

I know, it's a little bit of small type, so I apologize. Okay, so I'm going to talk about a few types of asset managers, the first being the asset owners. We heard from some asset owners earlier—Satish [Swamy] gave an interesting presentation about the history of the University of California System's money management.

But there are obviously other agents—and "agents" is really the right word, there are people like AQR Capital. We are an asset manager and also a levered fund manager. We act on behalf of the asset owners, and it's really the asset owners, at the end of the day, that are most important in this whole story. They're really the ones who, ultimately, determine allocation pretty much at every level, with some exceptions. I would say that the levered fund managers may determine the magnitude of leverage to some extent—to the extent that's important—but the really important actions based on changes in the market environment. And the open question, I think, partly for today is—do we act on correlation changes or not?—really the asset owners, either through their direct actions or who they hire and fire as managers, and what mandates they provide.

So let's start with a little bit about—again, we're going to go back to pension history, and talk about how U.S. public plan allocation has changed. This represents today, trillions of dollars of investment. Not included—corporate DB plan investment, which has changed, and possibly changed in somewhat different ways, actually. There's, I would say, some divergence right now between asset allocation, but I'm going to give you two different sets of data here. The top set of data is from Callan Associates, which is a major pension plan consulting firm. And you know, it was out of a risk that that industry really grew, so that's tracing back to 1974.

And as you can see here, the equity allocations have risen somewhat over this period from 1990 to 2015. The bottom chart is over a longer period of time, and I'll get to that in a second, but notice that the allocations don't change quickly. And in the alternatives bucket, which has grown decently—don't think of that alternatives bucket as being a different animal. In many ways it is a different animal, but the reality is that alternatives bucket has a lot of equity exposure, okay? So it's sort of intentional that we made that blue and light blue, let's call it.

The allocation to riskier assets—to equities—has grown in public plan allocation. I would say it's declined some in corporate plan allocations, and the corporate plans are not growing, the public plans are growing—in total assets, too. On the bottom is from Pew, and it's a little rougher data, but it goes back farther and so it's very interesting. What it shows is that in the early '50s, trust allocations—pension allocations—were almost all bonds. And there were a variety of reasons for that. You can see that it changes a lot, so two things then from this: one is, plan allocations can change very substantially, but they don't change fast—and that's why I included the top one.

Why plan allocations change that much is a little bit murky. I've spent some time trying to understand it. In particular, if you look at '62 to '72, it's really in the late '60s, early '70s that allocations from equities went from very little to a decent amount. And I tried to understand the drivers of that, and there isn't an obvious driver. That's one of the things that's really interesting, and that sometimes leads me to think that the reasons are not so much a regulatory reason in this case, or an accounting reason, but perhaps a performance reason—and that's something that we don't often hear about.

But one question for today—and unfortunately it looks like somehow the...no? Okay, it's not in there. Okay, so the question is, as this changes on the left, are they driven by correlation changes—and the primary correlation that's going to matter for these funds is stocks and bonds, because the allocations are mainly long only, to stocks and bonds. And I'm going to have to switch back and forth, because the graphs aren't showing up on the left here, but you can see that we went through some sort of structural shift around '01 or so, where stock/bonds correlation went from positive to significantly negative. But—again, switching back and forth here—it does not appear that this is the driver of any big change.

I think that's kind of interesting. We have had a large change—and persistent change—in the correlation of stocks and bonds. I for one don't think it's permanent, but it has not really led to a shift—at least not directly because of changes in allocation.

However, what's going to happen to investment portfolios when this happens? The correlation changes—meaning that the total risk of the same allocation, if the volatilities are constant—is going to shrink, okay? And so potentially, this might at least be a factor in the gradual increasing of equity allocations in the last 15 years. So they may change. I think the way correlations at least appear potentially to impact asset allocation at the big bulk level that's really important, may be sort of a second order thing. It's more not because we as pension managers—or "the pension managers." I shouldn't say "we," I'm an agent—the pension managers don't necessarily observe this and then act on it, but they react to the consequences of it, and there's a lot of lag in that.

Another thing that's shifted dramatically over my career is the international allocation of plan sponsors. It's gone from very, very little to really quite a bit. And for a U.S. plan—remember, in other countries these numbers can even be higher—but if you were fully globalized you would still have a lot of U.S., right? So you're never going to have—even if you just looked at your plan as completely global—you're never going to have much more than half anyways outside the U.S.

But you can see that asset allocation has become pretty global over the last 35 years. And it's maybe at its end state right now—or at least, for now it is—but we see fairly large allocations. And yet at the same time, the correlations within equity markets across the world have risen—and that's the purple line—and the correlations across global bond markets have also risen some. This would mean that diversification globally, because of correlation changes, is probably marginally less valuable today than it was at the beginning of the movement. So why has it continued? It's continued because U.S. plan sponsors were so underinvested globally, that even where correlations are today there's still a lot of value.

So this was really more of a very long catching up of allocation to probably better allocation than it was a response to what appears to be some long-term change in the correlations. And we can speculate why—it may just be because of the increased globalization of the world. That makes a lot of sense. That should make global investing slightly less useful than it was many years ago, and that's probably true. But it's still way better than being hugely home-bias allocators.

So, again—I think correlations have changed, but it looks as if asset allocation has shifted in a direction counter to that. But it's merely because that's not why asset allocation is shifting.

Another type of situation: active equity managers. So active managers have had a lot of stress recently. We know there's been a big move to passive that accelerated in the last few years and is creating a lot of business pressures. There are always equity managers that are underperforming their benchmarks. Their benchmarks, of course, tend to be within their own asset class, so their concerns are not correlations across asset classes, but instead we hear a lot of verbiage about correlations within an asset class, and that high correlations in the last few years have somehow led to underperformance. I think our view is more this—that on the left, you see the dispersion of returns, sort of the cross-sectional dispersion of returns, within a market, within a benchmark, okay? Let's say in this case, Russell 1000, which is a common large cap benchmark. And you can see that dispersion is very low—that generally means lower opportunity. Now, if there was a transaction cost-free world, it should mean not so much that more managers underperform, but that maybe how much they underperform—and how much the other ones outperform—shrinks.

But we live in a world with transaction costs, and so I think a world with low opportunity, low gross opportunity, based on the dispersion of returns, probably leads to relatively more underperformance than a world of high opportunity—or high variance, really is more likely. And so while we've seen high correlations in recent times, I think that if we look at this sort of opportunity—at least what we're going to assert today is opportunity—for asset managers to outperform, which is the dispersion. It certainly looks far more related, long term, to the volatility of returns rather than the correlation, that the low-volatility environment that we have been in is one of low opportunity and maybe more explanatory.

Levered funds—I'm going to give you two more examples—levered funds, one, here is a place, I think, where correlation matters more—in particular, where levered funds hedge. When you hedge, you have to care more about the correlation—is that hedge really a hedge? And I'm going to pick one example: market neutral equity portfolios in August of 2007, which some of you may be familiar with that event, where there was a general liquidation of these levered portfolios. And an unlevered return is what I showed you here over a four-day period—it got as bad as negative 8 percent for an unlevered return. But all of these strategies are levered, so the returns were significantly worse than negative 8 percent.

At the same time, we can look at the basket of longs—let's say at the beginning of this period—and the basket of shorts, at the beginning of this period. And every manager does this differently, but there's a lot of commonality in how they do it. And look at the return correlations of these two baskets over short periods of time. Now, of course there's not...normally I wouldn't look at correlations over a five-day period, but in this case it's so compelling. We know what happened. You can see that the correlations of the longs and the shorts got to basically zero, which is not how it's supposed to work when you have a long and short portfolio, which means obviously things got wildly risky. And when does that happen? When something goes wildly wrong.

So in this case we could look at the data and see this high-frequency shift in correlation. But here it's not the shift in correlation that created the problem, it's the problem that created the shift in the correlation. And I would think that a lot of times these very high-frequency things are events that we can identify. Not always, but I think often they are identifiable events. When they are identifiable events, then people are going to act on them not because the correlations change but because of the event.

So if we look at what's the ramification of having this correlation shift over a multi-day period...well, when those things happened in hedged funds—I'm not going call it "hedge funds" because it doesn't necessarily mean that—but hedged funds, it's usually going to be accompanied with major losses. What's going to happen after a major loss? After a major loss, it's going to create disinvestment from asset owners, so the result of this correlation shift is an action by asset owners. But it's not because the correlation shift, of course—it's because of the losses.

And that tends to be often what is acted on, is performance rather than a forward-looking understanding of shifts and correlation, because, to be honest with you, we don't have great confidence in our estimation of forward-looking correlation.

Now I'm going to look at the last example, which is a risk-parity type of strategy. I'm going to include this because this is what I actually happen to do as my day job. Risk-parity strategies are long-only, levered strategies that are asset allocation strategies that merely seek to balance risk across different types of assets. If I'm going to hold as much risk in stocks as I do in bonds, I have to hold more bonds than I do stocks. This is merely a graph of an idealized version of a risk-parity strategy—not one anybody actually does, but it doesn't matter in this case.

You notice the green line is exposure to bonds, and let's just take the purple line—that's exposure to stocks. More volatile asset, less exposure, vice versa. But, as volatilities and correlations change, if I were to just do it on paper—not how I do it in reality—but if I would just do it on paper, I would have shifts around in my asset allocation. It's not how anybody actually does it in reality, right? Things are getting much more measured, but just bear with me for a second.

Now I'm going to compare that to one same thing, only I'm not going to really try to best forecast my correlations—and correlations tend to be somewhat persistent, so I can forecast them some. I'm just going to say they're constant. So I see my correlations are constant in my risk model that I'm using to create those exposures, but my volatilities are not. And so those changes are changes in volatility.

What is the net ratio of the exposures I get using a model that forecasts fairly rapid updates in correlation structure, versus one that only forecasts volatility and assumes correlations never change? What you can see from this graph is that the ratio of exposures with one set-up versus the other—they don't change that much, meaning that even though the exposures change a lot, which is this big squiggle, the differences are not driven by correlation changes. The differences are driven by volatility changes.

And I think that that ends up being the case in many cases in asset management. We are often driven by changes in volatility, in risk levels, in sort of the bulk risk level. We are less frequently driven by changes in correlation, partly because we don't have that much confidence in our understanding of how correlations will change. So for example, today, even though we've now lived for about 15 years in an environment where stocks and bonds have been negatively correlated—I should say "stocks and government bonds"—have been negatively correlated. You do not see an enormous amount of direct activity because of that, at first order, okay? Because we don't know if that will continue, and there isn't necessarily a belief that that is the way things will be. If it did, it would tell people they should take more risk or what appears to be more risk.

On the other hand, there is a second-order effect. And the second-order effect is on how we observe the impact of reduced correlations or of changes in correlations. And that then affects performance, realized volatility, and asset allocators do tend to react to that. And what I believe we may be seeing today is an actual increase in risky assets in leverage—a little bit, not very much, but in leverage a little bit—in response to that. Thank you.

Zervos: Thanks, Michael.

[applause]

So, we have a lot of questions coming in. Please keep them coming. There's some with a lot of votes, some with zero votes.

I'm just going to throw one back at you, Michael, because you just ended on risk parity, and I wanted to maybe have you spend a second more on that since it's a subject near and dear to my heart, and a strategy that we've spent a lot of time on at Jefferies over the last seven or eight years. You make the comment that this is not about correlation, it's about volatility, and I think you're exactly right. But this 15-year stock/bond correlation has created a huge industry in risk parity—not just at your firm, but at Bridgewater, at individual asset managers that have their own risk parity strategies built in them. I know the guys down in Texas do it, I know a lot of other pension funds do it internally—not just where you are, there's foreign firms that are doing it.

Whatever happened—and I'll postulate this, and you can tell me whether you think it's right or not—something in the reaction function of central banks changed pretty dramatically in the last 15 years. We went to a back-stopping vision of central banks as opposed to central banks that were trying to earn the credibility that they had lost in previous decades. And that backstopping mentality created this correlation that everybody felt, "Hey, if things get really messy, they're just going to cut rates. So I have a big stock portfolio: what's my hedge? I just lever up a position in the front end of the bond market, or in the middle of the yield curve—and lo and behold, I'll be protected." The amount of people that believe that trade cannot go wrong has skyrocketed. It has certainly skyrocketed in the last 15 years, but really accelerated postcrisis.

And Q1 of '16—so Q1 of last year—was a perfect example of stock market going down 10 percent. Everybody was scared about China. What happened? Five-year yields fell 100 basis points, and the Fed—which was talking about four rate hikes—basically said, "Oh, by the way: we ain't doing nothing for a while." Which they didn't. They didn't do anything until December. It was risk parity perfection.

You said you're not so confident in that correlation going forward. Maybe you could tell me—are you losing faith in the 15-year trade, of risk parity being a dominant strategy out there?

Mendelson: Well, there's a lot in there, and the good news is we're going to have some controversy.

Zervos: Well, you got me thinking!

Mendelson: [laughter] So, risk parity strategies—first, to try to parameterize them. I don't know how many of you out there have seen the occasional media noise about them, partly because any time markets fall they look at strategies that may include a component of risk targeting as being potential sellers or buyers in a market that's moving. That effect is really enormously smaller—I guess "enormously smaller" is not a way to put that...

Zervos: [laughter] Significantly.

Mendelson: ...significantly smaller than anybody would ever believe, because the total dollars managed to risk parity styles I think is about $150 billion—which is quite a bit less than a lot of the press reports, but we're pretty closely in touch with that market and I think that's where it is. Most of that money is not volatility-targeted. Some of that money is volatility-targeted, so the amount of actual trading in bulk based on changes in volatility is surprisingly small. We've published some stuff on this, but its calculations are pretty straightforward, and there really is not a lot of market importance to changing risk-parity allocations. That I can state pretty categorically.

With, though, the more interesting question: so why are people pursued, to a limited extent—and it can only be a limited extent, because it's not an infinite capacity strategy—why do risk parity people adopt that in part? And I think it is important that people understand that you do not need to think of bonds and stocks hedging each other. I do not think of bonds and stocks as hedging each other. It's wonderful if they do, but I don't think we have much reason to believe that that is the persistent, long-term behavior of them. I look at: we can always explain stuff—our whole industry, everybody is great at rationalizing and explaining, here's what's happened, this is why these numbers came the way they did. I have no idea what markets are going to do in the future, and I know I've lived through periods where stocks and bonds are very, very positively correlated, and very negatively correlated. They just have to be, for diversification to work, for risk diversification to work. They just have to not be super highly positively correlated, which never really happens anyways.

So if bonds and stocks were plus 0.3 correlated, or plus 0.4 correlated—and they have been in the past—risk diversification is still a big benefit, just like global stock diversification is a big benefit even though global stock markets are much more correlated to each other. The fact that stocks and bonds have been negatively correlated, though, has allowed a lot of people—and you allude to this—a lot of people to believe somehow that risk parity (or risk diversification, more broadly) is about this hedging property. It is not, and it's important for investors to understand that is not why you risk diversify. You do not need long portfolios to have offsetting movements. They just have to be different movements, okay?

And you're right—some people do think that, but I don't think that's the logic behind risk parity, and I don't think that is something that any investor should count on, regardless of whether they're risk-parity investors.

Zervos: I'll just make a follow-up statement that I think it would be wishful thinking if that were the case. My guess is that's not the case. I also would guess and venture that in stated risk-parity funds your numbers are correct. But inside of every macro hedge fund out there—inside of many other strategies—lurks a risk parity trade that is much, much bigger than the number you're describing, and that significant levered bond portfolios have been used by many, many people across our industry to hedge riskier assets. And I'm specifically saying, government bond or eurodollar futures.

Mendelson: You mean a long stock?

Zervos: Long, volatility hedged.

Mendelson: That's not risk parity. But as a long-term trader, what I can tell you is I'm always petrified by trades with a lot of basis risk. And that has a ton of basis risk. It's not risk parity, but I don't deny what you're saying. That's something people should not rely on.

Zervos: Although it's been a great trade for a long time. [laughter] Just sayin'. Let's take one of the questions out there, the maximum vote question: does financial regulation encourage financial institutions to overreact to the co-movements that we're discussing? I think, Nikolay, that's right up your alley. So is financial regulation a big driver of the overreaction, the higher frequency stuff, that you were alluding to in your discussion?

Gospodinov: I'm not sure how much I can say about this, but at least—over this period again—this paper that I referred to by the Bank for International Settlements that asset returns have become more sensitive, more responsive to macroeconomic news, to regulatory changes, to policy announcements. And I think this is true, right? Whether this is the cause for this type of co-movements—honestly, I don't know, and I would like to...this was part of the open questions that I left at the end. I'm interested in these feedback effects, and to isolate these effects is fairly difficult to say. It seems that this is the case, by just observing these co-movements. But without further work I cannot make a...

Mendelson: The only observation I have is that: we sometimes fear—systemic risk regulation of asset managers is a little bit of a hot topic—and we worry more about how the outside world perceives how we would react to sudden changes, when the reality is we probably react a lot less than people think ("we" meaning as an asset management industry), and asset owners barely react at all. And we worry about how the perception of what we may do may lead to regulation of things that we aren't going to do anyways, or wouldn't do anyways.

Zervos: I'm going to switch to another popular question here, which was kind of what I alluded to in my opening remark. Actually this I think has the most votes: how important is monetary policy as a common factor across asset classes, and how might the renormalization—which was discussed in the previous panel—affect these correlations, and potentially returns, going forward? You want to take a stab at that, Nikolay?

Gospodinov: Yes. I'll go back to my graphs to answer this question. So, again, from these graphs it's obvious that asset returns are highly correlated with the business cycle, right? And we know that monetary policy reacts to this business cycle, naturally. So these are all associations—right?—so we observe these co-movements between asset prices, the business cycle, and monetary policy—what is the cause for all these co-movements? Again, it is very difficult to say. It requires a structural model that will decompose these monetary policy shocks and see there. But on the surface it appears that there is a strong relationship.

Zervos: Michael?

Mendelson: Yes. There's clearly a huge impact of monetary policy, and as a result you see more than a cottage industry of people out there consulting on it, forecasting it, acting on it. At the end of the day, in a lot of ways the asset owners are more observers of it. They're affected by it, and their agents are more affected by it, but not the decision makers. And I think there is a fraction of the industry that is trying to act on those things, but it's only a fraction of it. I don't think it's the bulk part of the industry. We just observe and live with the consequences.

Zervos: So one comment I'd make on that is, I think we learned a lot about central bank behavior over the course of the crisis. And maybe we even learned a lot about it prior to the crisis, with some of the things that were done in '03, or in response to the tech bubble pop, or even in '98 with the backstopping of markets in that crisis. That backstopping mentality is very much ingrained in markets today and has created I think a cottage industry in people wanting to feel more comfortable that there's somebody taking care of them. It's migrated over to places like Europe, where Mario Draghi tells you he'll do anything and everything to make sure—or, "whatever it takes," I guess was the quote.

So this whole sort of "tender loving care" central bank theory has really built up over the course of time. And I think that's made its way into a heck of a lot of correlations and probably kept your stock/bond correlation a lot more negative for a lot longer than maybe people would have thought.

And that gets us back to that risk parity discussion. I think there's a lot to that, and the question, of course, would be: on the way out—which is what this question asks—does that change? And I would answer no. Just because the Fed is taking rates back up, or normalizing the balance sheet, doesn't mean they're there to not backstop the market anymore if everything went wrong. If anything, they've proven their behavior pretty consistently. At least most central banks—whether it's the BOJ [Bank of Japan], the ECB [European Central Bank], or the Fed lately—have all done that. So I think it's not that much of an issue on the way out, and we learned a lot on the way in about the behavior.

Mendelson: There are two ways of thinking about what asset owners and asset managers think about. One is, are we looking at the world over the last decade plus, and saying that there's this backstop, and therefore we're more comfortable taking risk, because there's this backstop? Or are we observing asset movements over the last decade and saying, "Geez: stocks have done pretty well, volatility has been pretty low—I'm comfortable"? Just think—how does the money management world work? Is that inducing people to then take more risk, potentially, because of performance, which may have been the result of this? Or is it because they're saying, "Look, I think the Fed's there as a backstop"?

I think it's more the former. I think it's more because of performance, because that is what asset owners are focused on. I think there's a certain fraction of it that also look at what makes them comfortable and what's not, but I think it is hugely driven by what's actually happened in markets. And so I look at it like, if that's the cause of markets doing what they've done, then that's the cause of changes in asset allocation, But it's really that the direct decision has more been made on what risk and performance have realized over the last three years, five years, ten years.

Zervos: I would just put a lot more on the policy. That's personal taste.

Mendelson: The policy may be what underlies those movements, but when we think of asset owners—the state pension plan of XYZ or something—and what's actually making them make those decisions, I think it is less about trying to be macroeconomists.

Zervos: Well, let's go back to the Q1 of 2016 example again. We had three to four rate hikes priced in. We had every Fed president and governor telling us they were on their way up. And China does a little dipsy-do again like they did in August of '15. Stocks are down 10 percent and all of sudden the reaction function comes back to its beautiful risk parity place, which is, "Hey, we're not going to do anything for a while." I think maybe that's just some sort of bias confirmation that I get, but...

Mendelson: Because it's not the time scale that asset allocators work over.

Zervos: It's definitely the time scale that a large section of the hedge fund community operates under.

Mendelson: The hedge fund community—and which we're part of that—we're small compared to the rest of the world. And if you think about what we do in the hedge fund community—one of the things our firm does, is that you have specific mandates. And your mandate may be, like half the hedge fund world, is long-short equity guys, which lean long and are low leverage. I don't care what happens with rate policy or whatever, they may change their leverage around—they're still just going to be managing stocks, okay? And they're not going to then shift assets into buying commodities or what have you. There are macro funds that are more the way you think. The macro fund universe is tiny compared to the hedge fund universe, and the hedge fund universe is relatively small compared to the overall asset ownership out there. So I think that's true, in that slug you're saying. And those guys are highly active and responsive. It just, in the overall meaning of things, is fairly small.

Zervos: I'll take the other side of that again, and say that the high-yield bond fund managers—long-only equity managers—they sit there and they look at what happens in Q1 of '16. And do they come out in Q2 of '16 a little more comfortable, with a view that they've got a backstop? Absolutely.

Mendelson: But that doesn't change much of what they do.

Zervos: I think it does. I think it keeps them from spitting at the bottom, again. It keeps them from overreacting to a crisis, and maybe—we're sitting here with the VIX at 10 percent, and one of the lowest volatility periods we've had in my career in stocks. And we just have a lot of people pretty comfortable. The question is, why are they comfortable? Are they comfortable because we have a new president that's bringing in a lot of uncertainty to the market? Probably not. But they might be comfortable that if everything goes wrong, Jim and his colleagues are going to turn around and do what they've been doing for the last—not just this past quarter, but take it all the way back to '98. Take it back to the...I don't know what you'd call it—the original sin, maybe? But the real backstop beginning, which is kind of the beginning of your sort of 15-year-odd trend, of...

Mendelson: Right. But let's say our firm manages close to $100 billion of long-only equity, okay? So we do that, and we do an alternative. So in the long-only part of our business, the equity part—how much of that is fully invested today in the benchmarks? All of it. How much was, in the first quarter of '16? All of it. How much was it at any point in time? It's always the same, because we are agents for the asset owners, and the asset owners give us a mandate to manage long-only equities. And it's when the agents change it that asset allocations changes—or, I'm sorry. When the asset owners change it, that changes it, and the asset owners move slowly. Maybe appropriately slow, but they move slowly. And so, yes—we look at the markets every day and we say, "Geez, this thing's scary, this is not scary." It has much less effect, though, on what our actual portfolios look like than what our mentality looks like.

Zervos: I think we're just going to agree to disagree. [laughter] So let's do another question here: are asset owners reacting to changing correlation, or driving those correlations? I push the button, but nothing seems to happen. I don't know what I'm doing. So maybe...I'm technologically inept.

Mendelson: It's useful for somebody my age to be able to see the questions, so I can forget them.

Zervos: I'll try to say it slowly: are asset owners reacting to changing correlations, or are they actually driving those correlations? For example, did increased international allocations of U.S. pensions drive higher global correlations? Was that the actual driver? So, are we look...there we go, yes. Nikolay.

Gospodinov: I think Mike can give you a better answer on this.

Mendelson: [laughter] No, I can give a more controversial answer. [laughter] No, I think a little bit of both. I do think that increased global allocations matter. And then when you have large investors, large pension funds, invested in smallish markets—let's say like emerging markets, in particular. When there are shifts in asset allocation, and I've already claimed that those happen slowly, but they do exist, when they do it's going to have an impact. It will definitely have an impact.

And when there's an accident in a market—and we know emerging markets in particular are prone to accidents—when there's an accident, you will see a very sudden shift from investment. So you'll see there can be quick changes, not necessarily immediate allocations, but pending decisions all change. And so that can have a fairly quick effect—over a year or two time scale—on actual investment, and definitely have a big impact.

So maybe to a lesser extent in developed markets, but for sure we affect the markets. And obviously those markets affect us, and they affect us—my claim has been today, they affect us by us observing the effectiveness of international diversification, the effect on our risk, and then people act accordingly.

Gospodinov: Let me say something: I think that asset owners should react to undermine risk, right? So as we saw in this discussion, correlations are moving for different reasons, right? And they may react to different risk factors differently, so unless we identify the underlying risk for this particular...it's just...not the proper type of behavior to react, to observe correlations.

Zervos: Are either of you guys surprised at how—we'll take this question in a different direction. So are either of you guys surprised at how well emerging markets have done as the Fed raised rates these three times, brought the balance sheet into play, and the dollar strengthened over the course of the last two or three years now? It's a little bit stagnant, but emerging markets seem to be doing just fine right now.

Mendelson: Which is great. It has been a little bit of a soft period for them, for the last few years. But I think the last year or so, it's been a little better than maybe we would expect from that type of thing. Again, I don't know how much effect it has on what actions are, but it certainly could have been a lot worse.

Zervos: Prior correlation patterns would have said things could've got a lot uglier, right?

Mendelson: Yes.

Zervos: So, is this a new correlation? Is this asset owners that are driving a new correlation, that are just saying, "I want the EM no matter what"? How do you think about interpreting that, Nikolay? If you had to kind of dig deep into the rationale for why EM seems to have held up better than many people would have thought.

Gospodinov: Well, again, you know...it might be related to this discussion with the policy environment right now, and this reach for high yield [mindset], right? So it could be explained along these lines, and again—I'm not sure what are the underlying factors, whether I would take this at face value. So if I observe that over the last year, there was this co-movement, whether I will say that, yes—there is something structural in this co-movement.

But there are substantial movements recently that we observe just purely because of lack of investment opportunities, or reach for high yield. And it could be explained along these lines, I suppose, but again my interest lies—if this truly happens to be the case—my interest lies: yes, what kind of risk factor did they see, in order to adopt this particular strategy? And this is something that I cannot answer, obviously.

Mendelson: One thing you left out of that event is the commodities debacle didn't help with that, but...

Zervos: Which seems to have somewhat stabilized, although it's accelerating a little bit.

Mendelson: Somewhat stabilized, right. But I think the relative resilience of those markets has led asset owners to at least not take that sort of disinvestment direction, which is helpful.

Zervos: I guess where I was going to go was kind of where we were in the discussion before, which I—maybe it's the policy variable, again, that's in there that the Fed does seem very cognizant of these risks. And the dollar, I think, made it into statements—Federal Reserve statements—and speeches much more often than many people would have thought, given the typical dichotomy between Treasury and Fed on the dollar. And that gave people comfort that it really would never become too much of a bloodbath, which kind of worked out in '16.

Let's go to another question here: since the early '90s there have been periods of co-movement. This predates QE. Might co-movement just be a phenomenon of market stress or uncertainty or general factors that cannot be categorized? So, really, are we overthinking this policy variable, which I'm claiming is maybe big? But maybe you want to say, Michael in particular, it's not nearly as big as we think it is, that there's just a lot of other stuff that Nikolay has in his charts that are just so much more important than policy.

Mendelson: Right. That's why I look at the work that he's doing, and say, "Okay, if understanding this stuff better might give us more confidence in understanding how to act on these things...." But we observe this, and we look at it saying, "Boy, there are an awful lot of factors out there in the world." And you can postulate, there's QE, and then I look, "Well, what about before QE?" And I think this question's right. This is why we don't take strong action on a particular point of view of "this is why correlations are why they are, and then therefore we should allocate assets this way." It's for exactly this reason.

Gospodinov: Well, there are some recent, reliable, uncertainty changes. I'm pretty sure that one could correlate these co-movements with these uncertainty changes. So it was definitely this uncertainty that is driving a lot of these asset co-movements. What exactly is this? We can separate this into financial uncertainty, macro uncertainty, policy uncertainty, et cetera, et cetera. But I think this is something, again, that we need to understand a little bit better. But it's a big factor. It's a big factor.

Zervos: Policy, not just straight-up certainty, market stress...

Gospodinov: No, I would say uncertainty. Well, uncertainty—it could be uncertainty around policy, but there's uncertainty, not...

Zervos: So you've got to think about breaking it out between policy related drivers of co-movement and just standard, real, demand shocks, supply shocks—real side economic drivers of co-movement. What do you think the percentage is? Break it out: is it 80/20, is it 60/40, is it 90/10? We all look for numbers, so why not make one up? [laughter]

Mendelson: I'm totally making it up, obviously. But I think that policy has been enormously important—I'm going to give it 60 percent, where I'm going to...

Zervos: Wow—I would have totally missed that on you. I would have low-balled you at like 25.

Mendelson: And normally I would say that, but I think that we have been through this time of that. But I don't know how that is going forward. I only care about going forward.

Zervos: Yes, you...you got me confused. [laughter] Nikolay: whatcha got? C'mon, give us a number—just throw one out there. [laughter] Well, I see a question on here that is near and dear to my heart, and I'm going to answer it because it's not a good question. [laughter] Well, there's nothing wrong with this question. And also it's very exciting for me to have "spoos and blues" announced at a Fed conference as a strategy, which is basically buying S&P futures and then having a volatility-weighted hedge in, back on eurodollar futures. For those that don't know eurodollar futures—they trade whites, reds, greens, blues, golds. They go out 10 years. I think blues are the three-year forward, one-year rate. And so you just lever them, and that's your hedge. Wow, there you go. [laughter] Somebody needed a hedge back there.

I call that the poor man's Bridgewater trade, with zero fees. [laughter] So sorry, or maybe it's even the poor man's AQR trade—you guys have made a fortune out of that. I do that with one blue contract and one spoo contract on the Merc.

No, it's not good. It's tailor-made for positive if they're both going up. So if both stocks are going up and eurodollar futures are going up, or yields are going down, this is a wonderful thing, which is kind of what's happened in the last eight years. Rates have stayed pretty low. Every time stocks went down, yields went down. Eurodollar futures rallied, you had a nice hedge—even though you don't like to think about it as a hedge, Michael.

So it's been a great trade. But if we go back to the 1970s, and we think about a world where stocks are going down and bond yields are going up, in the old correlation prior to 15 years ago, this is an absolutely awful trade that would blow you up and destroy your grandmother's pension. So I think you have to be very careful. The negative side of it is actually the nice side of it, like what we saw in Q1 of 2016—which is, stocks go down, and then eurodollar futures rally, or yields go down, as a protection for your trade. So I wanted to make that distinction. I think we have time for one more question before we break for lunch.

Mendelson: David, one thing...

Zervos: Yes, please—interrupt me any time.

Mendelson: Be wary of basis risk. That's what this is.

Zervos: You can hit the bit on both of those in 30 seconds. Let's do this—let's stay on the international theme—and I really thought your chart on demographics, Nikolay, was a fantastic chart. And I actually have no idea how to interpret that. Why is it what it is, why is having more 49-year-olds than 29-year-olds is so good for stocks. If anybody's got a great story for that, I'd love to hear it. But maybe you can tackle that if you've got a good idea for it. But with greater international integration in financial markets, will global demographics—rather than this local country demographic story—be important as a long-run driver of asset returns?

Gospodinov: It's an interesting question. So first, the story for this demographic is, basically, the demographic is affecting savings rates and risk preferences. And this is how the story goes, but this is an interesting question. So we haven't thought about this carefully. At the same time, I would say that, globally, the demographic is exhibiting similar trends, right? Especially over the developed world. Whether the emerging markets will help in this direction quite a bit, I don't think so. It's a substantial drag on not only these low-frequency movements in asset returns, but as [St. Louis Fed] President [James] Bullard pointed out this morning, it's affecting labor productivity and stuff like that. So the global factor cannot overcome—at least in the near future—this strong demographic structure that we are observing in the U.S. and the rest of the developed world. But this is my take on this. But it's an interesting question—I haven't thought about this.

Zervos: You have an answer? No? Me either. My only thought was, maybe there's just less lazy millennials, so it's better? I don't know, but maybe that's not a less...because you have less 29-year-olds, and more...I don't know, I'm just trying to think of why it would go...

Mendelson: I don't know much about 29-year-olds, or 49-year-olds, so... [laughter]

Zervos: I'm 50, so... well, we have two minutes to spare. I'm not going to run over, I know everybody hates when we run over. Thanks to our two panelists, Nikolay and Michael, for giving great presentations. Thank you all for being a great audience—great questions.

[applause]