2016 Financial Markets Conference—Policy Session 1: Getting a Grip on an Individual Firm's Liquidity

The financial crisis highlighted the importance of liquidity and led to new minimum standards. Yet our understanding of liquidity at the firm level is primitive. This session will explore perspectives on how financial firms can measure and manage liquidity.

Transcript

Paul Kupiec: Welcome to the first policy session, Getting a Grip on an Individual Firm's Liquidity. In today's session, we're going to have one academic paper, and the paper is "Trading Liquidity and Funding Liquidity in Fixed Income Markets: Implications of Market Microstructure Invariance." And the author/presenter is Professor Albert "Pete" Kyle, who is the Charles E. Smith Chair Professor of Finance at the University of Maryland, Robert H. [Smith] School of Business.

Now, Pete is well known in the profession, he's a well-known market microstructure guy, he is a very, very deep thinker on liquidity, and the paper is a very good and interesting paper. I look forward very much to Pete's talk today.

We'll have two discussants for Pete's paper. The first discussion is Gregory McGreevey, who is the CEO of Fixed Income Research at Invesco. Mr. McGreevey has extensive experience in the Treasury markets and financial markets managing liquidity, and I'm sure he's going to have a lot of good things to say this morning.

The second discussant is Jai Sooklal, who is a partner at Oliver Wyman, and Jai comes to us before Oliver Wyman by way of the New York Federal Reserve Banking Supervision Department, where he was a very senior person in the liquidity risk part of that group. Before that, he worked in financial markets at various financial firms.

The author and the discussants' bios are in the sheet, and I encourage you to look at them. I'm Paul Kupiec, I'm a resident scholar at the American Enterprise Institute, and I have background in banking in various ways.

Now, I'm supposed to do an intro, and to do that, if you read Pete's paper, I feel the only way I can really try to do this is to channel my inner Rod Serling—and those of you who may or may not remember, Rod Serling was the very famous host of a show called The Twilight Zone in the early '60s, a very spooky science fiction show that I barely remember, but some of you may not remember at all, they made movies about it—but let's just put ourselves in that frame of mind for a minute.

So as Rod would always start out: Imagine a world where the price of a unit of trading risk is identical for each and every market. Now a unit of trading risk is really a random variable with a fixed probability distribution, but let's just imagine this world where it costs the same to trade a unit of risk.

But there's a catch: markets in this world do not all run at the same speed. There's always a catch in The Twilight Zone. So if we watch trading in each and every market with our own eyes, and we measure time using our own wristwatches, it's really hard to figure out what we're seeing. It's very complicated, things seem kind of fuzzy, they're kind of a mess. But maybe, just maybe—if we adjust our time clocks so that some clocks move very fast and others move very slow—then maybe we can make more sense out of this.

So we're going to change time, we're going to do a time deformation-type process. What if the rate at which you can trade risk depends on something called "business time" and each security keeps its own "business time clock"? So you've seen that famous picture before, every market has its own clock, they don't run at the same rate. It reminds me of a wall we used to have at the FDIC [Federal Deposit Insurance Corporation] that had Tokyo, London, and Paris, and Washington, DC, on it, and all of them were in sync except one was 20 minutes fast, and I could never figure out what was going on with that clock. And nobody ever fixed it!

So, a similar idea actually has worked before, and some of you may have heard of Einstein's special theory of relativity. I'm not saying Pete has invented the same thing, but it's along the same lines and maybe it will be. But the whole idea here is there are some things that are constant, but time-frames-of-reference kind of change depending on where you are and the laws of motion.

So, how time deforms things—and if you think about things, taking that into account—has actually yielded pay dirt in the past.

So to start the story, as it happened, one strange day more than 16 years ago, Professor Kyle woke up and found himself in this strange new world. And he was faced with a challenge to get back to his own home and his own time: to figure out the expected cost of trading a bazillion-dollar trade in the U.S. Treasury market, using only his slide rule, traveling southbound on a slow-moving float, on Bourbon Street in New Orleans, at 5:00 p.m. at Mardi Gras.

Now, luckily for Pete, he brought along his favorite thinking cap—and there's Pete's favorite thinking cap. So for the next hour and a half, do not look at your wristwatches or read your cell phone emails, but listen to Professor Kyle and our two experts. But do send me some questions on Pigeonhole—because Pigeonhole is the only thing that will work in this Twilight Zone.

Pete, you're up.

Albert "Pete" Kyle: Thank you. So, while I am waiting for the slides to get up—this analogy with the theory of relativity, I think there's a better analogy, and the analogy is with fluid mechanics. Indeed, a lot of our current thinking about what I'm going to talk about comes from the way physicists think about fluid mechanics is very appropriate in the context of getting a grip on liquidity, because it's about the flow of money or the flow of trades through a system.

This paper is actually joint work with Anna Obizhaeva, and indeed, the whole market microstructure invariance agenda is something that we have been working on for the last six years or so. She is a professor at the New Economic School in Moscow right now, used to be at the University of Maryland.

Let me give you the main idea of what I want to talk about. I want to take this concept of market microstructure invariance, which I will summarize in a few minutes, and discuss this illiquidity measure that we call 1/LL being a measure of liquidity.

And then a related measure, that we call "business time" or the "better rival rate"—we call that gamma. I want to just talk about what that theory is, and how it works.

And then, I want to claim that that theory is basically a theory of both trading liquidity and funding liquidity. And it can deal with trading liquidity and funding liquidity both because it has consistently dealt with the time dimension that Paul was talking about.

So I want to claim that this illiquidity measure—1/L—is a measure of both trading liquidity and funding liquidity because it has dealt with time appropriately.

Then the next part of the talk, what I want to do is extrapolate some empirical estimates that Anna Obizhaeva and I made from the stock markets. We estimated liquidity for individual stocks, and we claimed that this invariance theory should apply to all markets, not just markets for stocks. So we want to take the empirical estimates that we have for stocks, and then apply them to the bond market. And to apply it to the bond market, what I'm going to do is pick two polar opposite cases; I'm going to find the most liquid part of the bond market, which would be 10-year Treasuries, and apply it there. And then I'm going to find one of the most illiquid corners of the bond market, which would be—let's call it individual, off-the-run corporate bonds—and apply it there, and see what the differences are.

Essentially, this liquidity measure has a scaling law built into it, and if we look at things that are very, very different, and we have a scaling law, we can sort of get an idea of whether the scaling law works. And that would be the intuitive check on whether this extrapolation seems to be economically sensible. I didn't put it on the slide, but I'm going to also talk about the flash crash and the flash rally as we go along.

So, what are the results going to be? Let's begin with some empirical results that apply to this invariance idea. We find that a typical stock—that would be one that's kind of in the middle of the S&P 500—has institutional traders placing what we call bets. And you can think of a bet as a meta-order, it's a decision by an institutional trader to buy an institutional-sized quantity. We find in a typical stock that the average meta-order, or average bet, is the size of about $470,000, that the average transactions cost for such a bet is about 43 basis points, and that the typical stock would have about 85 bets a day, adding up to a trading volume of maybe 40 million a day.

What we want to do is extrapolate that to Treasury bonds, and we find that the average bet size that we get in Treasury bonds, or Treasury securities—Treasury notes, really—the 10-year Treasury market has an average bet size of 20 million, much bigger than the average stock. An average transactions cost of about one basis point, with about 8,900 bets per day. So that's saying that the market for 10-year Treasuries operates more than a hundred times faster than the market for a typical stock.

So if you put yourself in the time warp that Paul was talking about, you have to speed up your clock by a factor of 100 to trade 10-year Treasuries in a logically consistent manner. On the other hand, if you look at corporate bonds, things are very, very slow. We think that a typical corporate bond has about three bets per day, that's going to reduce the average bet size to $400,000 and imply a much bigger measure of illiquidity greater by a factor of 50, and we've got to use a round number here of about 55 basis points rather than one basis point.

It turns out this "business time" measure I'm going to talk about operates at a speed that's kind of illiquidity squared. So what does that mean? That means that things in the market for an individual corporate bond operate about 3,000 times slower than things happen in the market for 10-year Treasuries. So a couple of minutes in the market for 10-year Treasuries would correspond to days—and even weeks or months—in the market for illiquid, fixed-income instruments. This reminds me of what William Dudley said last night: "How long does it take to value a CDO?" And the answer was, "Three weeks." Well, that's a slow-motion time scale. How long does it take to value a 10-year Treasury? It takes about three seconds.

So we think that if you take the idea even further and apply it to commercial banks, you're talking about an even slower operation, just superslow motion. And I'm going to talk about the flash crash and the flash rally, and talk about how they could have resulted from very large sales or purchases executed extremely rapidly over a short period of time.

OK, here's a basic formula, I can just show you these two formulas—I write papers that have 200 equations in them sometimes, so this is being easy on you—so this illiquidity measure is the main thing I want to talk about. We know that illiquidity probably has something to do with volume and volatility. If you look at this formula, in the denominator you see P times V; that is dollar trading volume per day in any security. For a typical stock it would be 40 million a day. For 10-year Treasuries we'd calculate it as approximately 168 billion a day.

In the numerator, you see sigma squared, that's the returns variance, so for a typical stock it's 2 percent per day—0.02 squared—and for 10-year Treasuries it's maybe a half a percent a day, 0.0050 (50 basis points) squared. And the only security-specific information you need, according to our simple theory here, is dollar volume and volatility to calculate a measure of liquidity.

We've got two other numbers in there, C and m squared; C is the cost of a trade, that's the thing that's going to be constant across stocks, it does not have a subscript. And m squared is another scaling constant that's related to a bunch of different factors that makes things measure what we want them to measure. The way most people measure liquidity is they start with the standard deviation of returns. And if you start with the standard deviation of returns and measure liquidity the way most people do it, you're left with a time dimension that you didn't deal with. And that's true of every liquidity measure I know of, except for what's called a square root model, which actually is consistent with this theory here.

And what we've done here, is we've dealt with that time dimension, so the second equation you'll look at has got gamma in it, it's called gamma. Gamma is going to translate into the number of bets that arrive per day in an asset. And it's going to be related intuitively to trading volume, but intuitively it also should be related to volatility. If there's more volatility in an asset, you have more risk being exchanged, just like you would have if there's more volatility or more dollar volume—each means more risk being exchanged. So the right way to think about it consistently is to multiply them together. So in the numerator there, you'll see P times V times sigma, dollar volume times volatility. That gives you this consistent measure.

But it turns out you need to raise that to the two-thirds power. So let me talk briefly about why you need these exponents of one-third and two-thirds, and the answer has to do with leverage. Leverage has a lot to do with liquidity. Many of you may have heard of the Modigliani-Miller theorem. The basic idea is if a company, let's say buys back a bunch of shares or pays a cash dividend—let's say pays a cash dividend—that's equal to half the value of the company, probably what's going to happen, other things [being] equal, is the share price will fall by 50 percent, but the volatility of returns in the company's stock will double. And that would mean if you didn't take the one-third power in this illiquidity measure, the numerator would increase by a factor of four, because sigma has double so sigma squared multiplies by four.

And even though you are exchanging the same risks, you're exchanging them to more leverage, so the dollar volume drops by a factor of two. So that means the whole thing would increase by a factor of eight if you didn't take the cube root. Taking the cube root increases it by a factor of two, and that's what should happen. The cost of trading a stock, when measured in basis points—if you leverage it up—should increase proportionately with the amount of leverage.

So, in a nutshell, the reason you have the one-third power is because of a leverage argument.

Now, the other thing to notice about this measure 1/L is that it's dimensionless. If you look at both the numerator and the denominator, they're measured in dollars per unit of time. This C number that you see is a certain number of dollars; we're going to call it $2,000. Price times volume is dollars per unit of time; C times sigma squared is also dollars per unit of time, so the numerator and the denominator have units that cancel, and you get a dimensionless measure of illiquidity—which we claim is necessary and correct for measuring something like transactions costs and basis points. Basis points are essentially dimensionless; they're a fraction of the value traded. Fifty basis points is like a fraction of one over 200, and it doesn't have any dimensions.

This 1/L is our liquidity measure, and it goes along with this idea of "business time," which measures the speed at which the market is operating, and the speed at which the market is operating has an exponent of two-thirds. And if you think about that, you want it to be measured in units of time. And to get the units of time to be consistent you need to take the two-thirds power, and that's what's done in this formula.

I didn't mention it, but the numbers C and m are numbers that we are using to try to scale things so that they measure what we want them to measure. We want 1/L to measure the average transactions cost in a market of trading and institutional-sized quantity, so we need some kind of scaling. And we want gamma to measure the number of bets that arrive into the market. And then this number C is measured in dollars. We think of it as the cost of a bet, measured in dollars—the average cost of a bet measured in dollars. We want to scale things so that the average bet size coming into the market—if you look at the fourth bullet point—P times Q bar is the average size of a bet measured in dollars, happens to be equal to CL. So C and m squared are scaling parameters that make this happen.

So now I'm going to just summarize here—you don't have to look at the slide that carefully—summarize some results for stocks. We have this other paper where we try to estimate essentially trading liquidity for stocks. This invariance theory implies that you ought to be able to write a transactions cost function to measure trading liquidity in the form you see on the very first line there. It's 1/L, the measure of liquidity that's specific to stock, times some function of the size of the bet. And we scale the size of the bet as a multiple of the average bet in that particular market. That's what Z means as a multiple of the average bet in that particular market. It has that form, and we estimate that form, and we find numbers that when I—we get different numbers, but I reinterpret them for a formulation in this paper—they are roughly equivalent to the cost of a bet is about $2,000, and this constant m squared is about a fourth. And so we can plug those two numbers into that 1/L formula, and then all you need is volume and volatility in the market you're interested in, and you have a measure of liquidity.

We also estimate a particular functional form for this function F that I'm not going to go into, but if you look at the bottom line there we find that the average transactions cost for a typical stock is 43 basis points with an average bet size of about $470,000, and with a median bet size of only about $133,000 because we find that the distribution of the bet sizes is a log-normal distribution with a very fat tail—the log variance is about 2.53.

So, that's the results from this other paper. Now, what we're going to do is take the results from this other paper and apply them to the bond market. But before we do, I want to talk about the relationship between funding liquidity and trading liquidity. So, what about funding liquidity? Well, here's an equation that's kind of...I think the way most people think of funding liquidity—certainly the way I used to think about it—is pretty simplistic, and it makes a lot of sense in some sense, but we think it's wrong. We think the way to measure funding liquidity is, what's the appropriate haircut to put on a financing transaction? So you could think of what's the appropriate margin for a futures contract as well, but let's say I'm going to finance some bonds in the repo market, and I want the haircut on the financing transaction to be sufficiently high so that I'm protected against a default resulting, let's say from a decline in the value of the bonds. Well, what most people would do would be to say let's look at the daily standard deviation of returns—that's sigma—let's look at how often we're going to mark these bonds to market and repost the collateral. That's typically one day, so delta T would be a day, but it could be a week or a month, or whatever. And then, let's multiply that by some number of standard deviations of safety, maybe three, four standard deviations, and that will give us a haircut.

We think that, in some level, that's pretty reasonable. But what's unreasonable about it is that this theory fails to take account of how time interacts with funding liquidity. When I talked about trading liquidity previously, these are institutional-sized trades, and they're taking place over time. You don't go in and trade a gigantic quantity instantaneously, you go in and you trade it over time. And when you trade it over time you trade it at the speed that that market actually operates at. So if the market has lots of bets arriving, you execute the trade very quickly. If the market has very few bets arriving, you're going to execute the trade very slowly when you convert it into calendar time.

So, in a funding transaction, you've got to do things like mark things to market. You've got to value the collateral, you then have to resolve disputes because in a no-liquid market, when things start deteriorating people are going to start arguing about the value of the collateral, and that's going to slow things down and it's going to take place in the speed with which things are happening in that particular market. And then if the collateral does default, someone's going to seize possession of the collateral and then liquidate it, but they're not going to liquidate it instantly. They might engage in a fire sale, but they're still going to do a time trade-off where they're spreading the liquidation out over time and executing it at a speed that's appropriate for that particular market.

So what we think is that funding liquidity needs to think about time, and if you were liquidating a position in Treasuries you would do it much more quickly than if you were liquidating a position in illiquid corporate bonds, or in the case of the financial crisis, obscure mortgage-backed securities. So what we think is that you need to think about the volatility (sigma)—that was the right idea, but you need to think about it in business time. And it turns out, we can say, "Well, what is the volatility from one bet arrival to the next in a market?"  And the answer is, it's proportional to this illiquidity measure, it's proportional to 1/L.

So, if you're thinking about volatility, and you're thinking about it in these new time units, you use the same illiquidity measure (1/L) as that you would you to think about funding liquidity. So if valuation of collateral, marking to market, resolving disputes,
liquidating collateral—all those things are taking place in business time, then you should be using the volatility in business time, which is m/L or something proportional to L—multiply that by the square root of the horizon over which these things are taking place, and that would give you the first step in coming up with a good haircut.

We can modify our trading cost function to introduce the idea of urgency. And that idea is the more rapidly you do a trade, the more temporary price impact you're going to have, and the more expensive it's going to be. We have a theoretical paper in which we do this, and it turns out that it's just a simple proportionality that comes out, so that if you execute a trade 10 times faster, you have 10 times more temporary price impact than you would otherwise have. Other people have estimated this empirically and found a square root, but we like the linear formulations.

So if you use that linear formulation as a foundation for thinking about what would happen in a fire sale, if you speed up the liquidation because you think the collateral's value might change on you, what happens? You result in a haircut that takes account of two things. It takes account of all this marking to market, and arguing and fighting activity that takes place in business time—and that's proportional to 1/L—and then you have another component that's—if you have to liquidate the collateral, what is the transactions cost of actually liquidating the collateral? Well, guess what? That's proportional to 1/L, too.

So this liquidity measure, 1/L, we think is a measure both of trading liquidity and of funding liquidity.

OK, so that's the theory in the paper. So what we do next is we say, let's take these numbers that we estimated from stocks and let's apply them to bonds, apply them to Treasury bonds first—the Treasury market 10-year notes—first, and then we'll apply them to illiquid corporate bonds.

When we apply them to Treasuries what we do is we calculate a total volume for the entire Treasury market by adding together 10-year, 5-year, and 2-years—we add together both Burger Tech volume and Treasury bond futures volume, and we weight them appropriately so that we get a 10-year equivalent from this little table of $168 billion a day in volume in 10-year Treasuries. And then we put that into our formula and we get that 1/L, the transactions cost for an average institutional-size trade, and Treasuries works out to 1 over 10,000, so it works out to almost exactly, coincidentally, one basis point. And what does that mean? That means the average size of an institutional trade in Treasuries should be, according to this theory—this is purely a theoretical calculation here—should be about $20 million. And the number of bets per day in Treasuries should be about 8,900—and these are kind of round numbers—more than 100 times faster than a typical stock. And that 1/L of one basis point means that the average transactions cost should be, for an institutional trade, should be about one basis point.

By the way, it doesn't mean that average transactions costs for a $20 million trade is one basis point; it's the average transactions cost, not the transaction cost of the average [trade].

So we calculated an implied transactions cost for trades of different sizes and put them in this little table here. There's a bid-ask spread component as well, so the very small trades—it's mostly bid-ask spread costs. But if you go up to the larger trades, you'll find that like a four-standard deviation event is like a $3 billion trade, and it would have a transactions cost of 10 basis points and probably would move the market by about—not 10 basis points in yields, but 10 basis points in returns—and move it from a 99 to 99.10. In a gigantic, $8 billion plus bet, it would move the market about 25 basis points, according to our calculations. But just from this simple formula, 1/L.

So let's apply this to the flash crash and the flash rally. In the flash crash the reports that the government agencies did there found that a $4 billion sale hit the market, was executed over 20 minutes; and during that time the S&P E-mini went down about 5 percent and then came back up again. We think that that was...and basically, according to the way that particular trader had behaved previously, that was about 15 times faster than a normal speed. So what we've assumed here is that the temporary price impact was about 15 times greater than you would expect.

In the flash rally for bonds, the market rose like 120 basis points, and then fell again, all in 12 minutes—which the official report there called it a round-trip. So what we've done is we've said, let's ask what would a flash crash—or what we would call a pseudo flash crash—in bonds look like? Which of these two events was bigger, the flash crash or the flash rally?

So what we did is we took the flash crash—transformed business time, transformed volatility, took into account volume, recalculated 1/L—and figured out what an equivalent event would look like in bonds, and what an equivalent event would look like in bonds would be—with purchases rather than sales—would be 5.6 billion arriving in the market, executed 15 times faster than normal, which would be over 48 minutes...driving prices up or down—may be reversed here—by 277 basis points.

So, that's much bigger than what happened in the flash rally. So what we think would have had to happened in the flash rally is to scale these numbers down a bit using invariance and taking into account a speed of execution of large bets that we think is appropriate. We think that 2 billion, purchased over 20 minutes at about 20 times the normal speed, would have driven prices up and then they would have fallen by about 130 basis points.

So the moral here is that we think of the Treasury market as being very, very liquid, but if people don't respect the time dimension of that market, which is one where things are operating already pretty fast, but if you put pressure on it and operate it even faster, you could knock it temporarily out of equilibrium, so to speak, and create a big dislocation, and we think it would take about 2 billion of very rapid trading over about 20 minutes to do this. And that's based on our invariance numbers that are estimated from stocks.

Let's think about funding liquidity in Treasuries. Well, funding liquidity in Treasuries might—this may be where our theory doesn't work very well. That is, the theory would imply that you should be marking things to market and liquidating collateral and resolving collateral disputes like every 10 minutes or something. Typically, I think, funding liquidity is something where people like to do things on a daily basis, where you get a valuation for assets at the end of every calendar day. But the theory suggests that the calendar days might be too slow for Treasuries; things should be operating on an intra-day basis. The fact that they don't might explain why haircuts would be a little bit higher.

Let's turn to corporate bonds. Corporate bonds, we think, have very, very low liquidity, and partly it's because there are very few bets taking place per day. So in a typical corporate bond that's off the run—the trading volume has died down to a very low number—we're going to use three bets per day instead of 8,900; we're going to assume the volatility is 50 basis points per day (I don't know if that's a reasonable number, but it's supposed to not take into account interest rate risk, we're going to assume that that's been hedged out). And we get an estimate of something like 55 or 58 basis points—I'm going to round it to 55 for discussion—and that suggests the typical institutional betting in corporate bonds is about $342,000; if you read Larry Harris's nice paper on corporate bonds you'll find that they do put them into buckets based on size, and some of them are a hundred thousand, some of them are a million or more—and we think that $342,000 is a reasonable number to get. And we find an average transactions cost—it's really from extrapolation—of about 55 basis points; that's also, we think, rather consistent with the numbers reported in Larry's paper for institutional-sized trades.

But our transactions cost model implies much, much lower transactions costs for very small trades. We think the transactions cost should be like 5 basis points, instead, Larry reports they're like 80 basis points. And we think that that's telling us that something's wrong with the corporate bond market, in terms of the way it provides transactions cost to smaller traders. And Larry makes the same point in his paper, so invariance theory here is confirming what Larry said.

And by the way: the numbers I reported on stocks are very similar to numbers reported by Chester Spatt and Larry and Jim Angel, in their paper looking at transactions costs in stocks. So we think the numbers are probably reasonable.

So, what about a fire sale in a corporate bond? Well, a fire sale in a corporate bond is going to take place in slow motion. You might execute things, say, twice as fast—or maybe the market's unsettled, and so the cost of liquidating a typical bet might be about 110 basis points, but it might take place over a very long period of time, you know, like a month or more. And during that time prices can change, so if you need a three-standard deviation cushion to make the funding transaction safe, we would suggest a haircut of something like 860 basis points, which is not too different, I think, from the haircuts that you saw in the tri-party repo market, before the financial crisis.

OK, let's think about whether this [is] a reasonable way to think about it, that illiquid fixed-income assets—funding liquidity events—take place in slow motion. You know, the BSAM hedge funds, it seems like I was reading about those week after week, and they kind of, in the summer of 2007—this wasn't something that happened in like five minutes, this was something that was stretched out over a period of a couple of months—and eventually Bear Stearns took over the collateral in their hedge funds, and it probably led to the collapse of Bear Stearns, in my opinion, but at any rate...it took place in slow motion.

Similarly, with Long-Term Capital in 1998, it was called a crisis, but it again was unfolding in slow motion, over the summer and into the fall, as those positions were liquidated. The London Whale trades—that was also something that was unfolding in slow motion. And the reason we think these things were happening in slow motion is that while some of the instruments that some of these funds were trading were quite liquid, many of them were very illiquid, and with illiquid instruments the collateral disputes and marking things to market takes weeks, it doesn't take minutes like it would with Treasuries.

[I've] got a couple of other points here; invariance implies that if you chop bond issues up into tiny little pieces like the bond market does, and like William Dudley was talking about last night—and if you do that, the average size of a bet is not going to fall as much as the average size of the issue. So the ownership of the issue becomes more concentrated in the hands of a small number of owners. And that's going to make borrowing it more difficult for shorting it...and that's going to make the market less liquid, because the short-selling mechanism is not going to make the prices accurate and it provides a justification for a "pay as you go" swap. So we think that pay-as-you-go swaps get around the cost associated with selling short illiquid instruments, and so pay-as-you-go swaps are natural inventions that deal with this problem. And invariance lets you quantify how big it is.

If you take this idea of funding liquidity and apply it to banking...well, the assets that a typical commercial bank has that are commercial loans, and loans to real estate developers—loans that are very idiosyncratic—we think those loans are even less liquid than corporate bonds, and so if you're thinking about bank regulation, if you're thinking about bank capital, if you're thinking about stress tests, you have to take into account that things in banks happen in slow motion. And that illiquidity (the gamma) for the bank's assets is going to be very close to zero. As a practical matter, you're going to be holding things until horizon.

And so, to conclude—and I'll just stop here—invariance implies dramatic differences in liquidity across fixed-income measures; 1/L takes that into account. There's maybe 50 times greater difference in liquidity between Treasuries and corporates, and we need more research on this.

Gregory McGreevey: First of all, thanks so much for having us here today. I'm going to hopefully either put you at ease or get you awfully excited. I'm not going to have any slides, and certainly not going to have any differential equations—which I know for some of you that might be a little bit much. But I really wanted to give you maybe four things that I wanted to cover today; first was really to give you a little bit of backdrop from a practitioner's standpoint, in the markets day in and day out, about liquidity and maybe a little bit of the backdrop, some of that I think will be pretty familiar with those that are sitting around the room here.

Then wanted to go from there to systematic risk, because this liquidity issue which I think we'd all agree, is real, it's something that needs to have more work done on it. Is it real or perceived?

The third thing is really talk about maybe some suggestions from at least our standpoint that regulators, as they're dealing with this issue, and the industry might be able to take into consideration.

And then finally, I'm going to spend a couple of minutes on how as a firm we manage liquidity on a daily basis.

So I think, as Dennis [Lockhart] started off, liquidity is one of those things that's very, very difficult to measure. I think we'd all agree with that as we look at it from a fixed-income standpoint, where we separate credit risk and liquidity risk, it's one of those things that's a very difficult question to answer. So why is it an issue now, first of all—and I'm going to go through these in a minute, when you look at some of the data that's out there from the marketplace, the data's conflicting as to how broad of an issue this is. So that's kind of point one, too.

The growth of the fixed-income markets has expanded, so if you look at the high-yield markets, it's doubled over the last seven years. You look at the investment-grade market, it's more than doubled. So we now have a $7 trillion market, which is about a little over two times from where it was not that long ago.

And at the same time, you've got shrinkage of bank balance sheets, and so you don't have to be a rocket scientist to put those two things together and say we probably have to have an issue that we're going to address as an industry. The question really is, against those two things—because they're vulnerabilities, at the end of the day, that we have to deal with during the next crisis, if you will, and I'll talk about that.

So, why are we at the situation that we're in—and this will be a little bit redundant for those in the audience—but there's really structural and cyclical issues, as we look at them. The structural issues are regulatory based. We know what those are, there are Basel requirements to increase capital requirements, there's risk sharing in structured assets, there's a decline in the whole shadow banking marketplace, there's the leverage test and the liquidity test, all of which serve to reduce what has been the normal source of liquidity in the corporate bond market—and that's bank balance sheets, and the size of that.

And then there's cyclical issues, which post the great financial crisis, and central banks across the globe and their need to be able to provide support for economic formation, have done a couple of things: one, they've reduced interest rates, as we all know, to very low levels. We've got Japan and Europe at negative interest rates out to five and seven years, and at the same time, we've had a lot of quantitative easing, which has crowded out the impact of those investors that are in the room trying to find available assets with which to invest.

You couple that cyclical issue—some may say it's a structural issue—with aging demographics, and you have an interesting need in those aging demographics for income and yield. So the combination of those has really resulted in net inventories today being down about 80 percent from where they were a mere six or seven years ago, and at the same time the bond market's grown, if you will.

So how big of an issue is it? Let me give you a couple of statistics, at least from our perspective—and, you know, these might be a little more simplistic but we think they're pretty practical at the end of the day. So if we looked at bid-ask spreads—that's just the cost of transacting kind of buy and sell orders overall—if you look at that today, they're largely in line with where they were precrisis, and that's pretty much in every market that's out there. This is TRACE [Trade Reporting and Compliance Engine] data, so it really is very representative [of] the industry as a whole, and that would exist whether I looked at the high-yield market, the investment-grade market, the bank loan market—pick a segment of the fixed-income market.

So that would tell you, at least from that measure, while transactions costs you think would go up, they really haven't, at the end of the day. And I think that's widely supported in the industry.

Trading volumes have risen in most markets. Now, part of that's the advent of just the increase that I mentioned, and corporate bond issuance. And the number of trades have increased, so a little bit supporting that maybe liquidity isn't that big of an issue. Now, if you look at the other side of the equation, the average trade size [is] much smaller today, actually, in investment-grade corporates today it's about 750,000. If you look back, say, precrisis, it was about 1,200,000; it doesn't sound like a big differential, but it is at the end of the day, and probably the most significant thing is the turnover, which is really then combining, "Well, how much trade activity is out there, relative to the size of the market?" It's down about 80 percent, and that's probably the thing that we as a company that has to manage that liquidity for our investors at the end of the day, and hopefully those in the audience, would be most alarmed with.

So the question is kind of mixed, I guess, if you look at those measures, some would say liquidity is not that big of a deal, some would say it's a very significant deal. So I think the broader question is, is it really a systemic issue, at the end of the day? And what we did is [we] went back and looked over 30 years at the mutual fund market—the reason we selected mutual funds is it's right now one of the areas of focus from a regulatory standpoint, the retail market is very different from the institutional market, typically the flows in that market are more volatile—and so we looked at the corporate bond segments of the mutual fund market and did this 30-year study to look at a couple of different things. One, on an annual basis, to look at what the actual level of outflows would be for the industry overall, relative to the available cash on hand. So we wanted to look at how well could the industry meet those withdrawals from not having to sell securities, so this would be things like cash on hand, coupon payments, and maturities of bonds, etc.

And when we looked at those markets, with particular focus in the high-yield markets, what we were able to see is a couple of really interesting things. Number one, the biggest outflow in the high-yield market didn't occur during the great financial crisis, which I think if I would have asked that question of the audience, they probably would have said, what was the biggest outflow? I would have probably thought before the study the great financial crisis probably would have been that time period. It was actually the junk bond market in 1990, where you had about 22 percent outflow within the industry. The great financial crisis only saw about 6 percent of high-yield outflows—there's a whole bunch of different reasons for that—and the average of the five worst time periods was about 12 ½ percent. Now, that compared to about 27 percent for the industry of average liquidity in high-yield funds, so that again would be cash on hand, [and] a variety of things related to that.

So we looked at that and said, reasonably, that the high-yield market—and the same analogy could be made, or the same study was done, on the investment-grade market—that they were able to meet their withdrawals. So what we did is took a little bit deeper dive on it, and said, "Well, let's look at just cash on hand." Because sometimes in a difficult market maybe you don't get your interest payments, maybe you don't get your maturities, maybe you don't get certain redemptions that you anticipate. So we looked at a fund level across the industry, cash on hand at the end of a month relative to what withdrawals were going to be, are redemptions in the subsequent month. And we looked at this roughly over that 30-year time period.

A couple of interesting conclusions that came out of that—and this is all public, so for those want to get a little more detail about the study that we did, we certainly can make that available—so over that 30-year time period, only two months happened where cash at the beginning of the month was insufficient to cover redemptions. So we looked at that and said that, on the surface, didn't seem to be all that accurate, even though the data was the data. So we then stressed the environment and said, "Well, what if redemptions were twice as great? In terms of what we saw, what would that look like?" And out of that roughly 360-month time period there were only 12 months that cash on hand was not sufficient to cover what the redemptions were going to be.

So we looked at that and came up with a couple of conclusions: one, that the mutual funds have proven to be not a source of systemic risk, if you will...that it didn't seem, at the end of the day, that based on the historical evidence that mutual funds as a source of systematic risk seemed a little bit overdone. But we also know that when you look at the market today, and you look at overall liquidity as it relates to some of the things we talked about before, we don't want to be naive in just concluding that history is going to repeat itself, so we definitely think that something needs to get done.

So let's go through, maybe talk about a couple of considerations, and we'll talk about how we manage liquidity risk at the end of the day. I think one of the big problems is we think about liquidity risk as just the definition of how you define it. And I think if we asked everybody in this room to define liquidity risk, I would guess that we would have at least more than 10 different definitions of how to account for it. One of the best things that we've looked at was a report that the IMF [International Monetary Fund] did, where they described liquidity in a more qualitative way that had really five different components: it was tightness, immediacy, depth, breadth, and resiliency. I'm not going to go through each of those items, but I think it really captured, at the end of the day, the things that we think you need to look at: How liquid is the market? How quickly can you transact? How much of an impact are you going to have on your prices? Things like that. So some of it is a commonsense approach to liquidity, at the end of the day.

So as we look at the various things that have come out of the SEC [Securities and Exchange Commission], we're very supportive of a number of those things, and there are some things, frankly, that we think maybe there's a little bit more work that needs to be done. So what are we very supportive of? First, I think as an industry and certainly as a firm, [we're] very supportive of advisers having to put in place a very reasonable liquidity management program that has a governance that's approved by their various boards, or whatever their governance within their company is going to be. So we definitely think that's a great idea, clear guidelines that limit illiquid assets, we're very supportive of, for a whole bunch of different reasons. Very supportive of enhancements to disclose the amount of liquidity risk that's out there, we think disclosure is a very good thing overall.

And I think this whole issue of swing pricing, I think as a firm and as an industry we think there's definitely some merits to that, but there's a whole bunch of different things that we think more work needs to be done. Now, where, maybe, there's a little bit more focus, I think that regulators in dealing with this issue could place on it is the classification of the three-day liquid asset requirement we think is overly burdensome at the end of the day, and I think it is going to be highly ineffective, at least from our perspective. The classification of liquidity into six buckets, I think, is also something that is going to be very practically difficult to implement at the end of the day, and I think it's going to be hard for the industry to really get its arms around that, and so we worry at the end of the day that as a firm we'll talk about a little bit how we manage liquidity, that we may have to abandon some of the things that have worked for us during very difficult market environments, and managing that liquidity risk to something that's maybe kind of a "one size fits all."

So what are some of those unintended consequences, and then I'll talk about how we manage liquidity risk. And these again, maybe the other side of things that maybe the regulators hopefully can be thinking about, and the industry can.

One of the things we worry about in the current path is that the regulators are going to pick winners and losers. As one of the larger firms out there, just because that may benefit us doesn't mean that it necessarily is the right thing at the end of the day. I think large firms that have access to, but limited capacity or inventory, there is at the street are going to have the ability to transact at a different level than firms that are smaller. Clearly, there's going to be...the regulators, at least in their current form, may have some ability to pick the winners and losers from a product standpoint. I think you're going to see an increase in multisector products at the expense of single-sector products, and for those firms out there that want to control or manage their own asset allocation, that may not be a great thing at the end of the day.

And then, clearly—depending on how some of the liquidity issues come out—I think you're going to...if in their current form, lead to greater concentration of portfolios in liquid names, because they're in essence going to be somewhat forced to go into those names, which may reduce the amount of alpha opportunities, or returns overall that may take place for investors, especially during the time when rates are low, and credit spreads across the globe are pretty tight right now.

So how do we manage liquidity, and then I'll wrap up from a conclusion standpoint. So I think there's a "belt and suspenders" approach that we look at, and certainly in no way do we think that we're doing everything right in all facets, but really, the first line of defense for us is with our portfolio managers. We spend a lot of time talking to them that they have to know what's in their portfolios, and have to make they're managing all risk, liquidity risk being a big portion of that. So that's the first line of defense.

We have CIOs [chief information officers] within our various organizations that spend a lot of time doing peer reviews. One of the things that they'll look at right now, because of the things that we talked about earlier, is liquidity risk, and make sure that that liquidity risk is being managed. And then we have an independent risk group that also is going to be looking at our various funds, especially those in fixed income, and looking at how much liquidity risk we're taking; those things combined, and we'll at least have an ability to have good discussions.

Now, in addition to that, on a quantitative basis, we go through and look at withdrawals of all of our phones, what our historical withdrawals have been, and then our stress withdrawals, and make sure that we're able to meet the liquidity requirements. Not in a seven- or eight-standard deviation event, but in an event that we think is somewhat stressed, that we're going to have liquidity to be able to meet that. Liquidity, for us, in that definition, is cash on hand, or the ability to get interest payments or principal payments for those securities that we know are above a certain rating threshold.

So, the combination of that—if we looked at a particular product where we didn't think that we had enough liquidity, there's a couple of things that we would force our portfolio managers to do. One is to increase the cash position overall in their fund, or in some case go out and get letters of credit that they would have to put into their product. And we did that in one of our particular products on the retail side, as a result of it.

So, that's how we manage liquidity at the end of the day. Hopefully, I went through a number of things fairly quickly, I guess. From our standpoint, we are strong supporters of the SEC, we're strong supporters of the Fed and other regulatory bodies who are trying to deal with this important issue. So despite our own studies, and despite coming from the industry, we're well aware that this is an issue that needs a lot of eyes and a lot of discussions. It's enormously complex, it's multidimensional, and it's really going to require a lot of thoughtful analysis. We want to be able to hopefully—from a firm standpoint, or from an industry standpoint—be able to at least provide our thoughts and perspectives, and hopefully there's some question and answers that we can begin to address in a minute.

Jai Sooklal: Good morning, thank you all for the invite to be here from the Atlanta Fed. What I'm going to do is pull it up a bit and talk generically about financial institutions and how we link market risk, early availability of liquidity in the markets, to an individual firm's assessment of its liquidity. If you think of yourself in a regulator's shoes, or if you're in a bank management position—or an investor—one of the things that is obviously key is the whole concept of market liquidity, and the liquidity of the firm (your ability to generate cash flows to meet demands when needed).

So, high level, what I'm going to talk about quickly in the next few minutes is to give some thoughts on some of the market issues that we just heard from Greg and from Pete. I'll tell you about some of the current measurement practices that have evolved over the industry in terms of how firms are measuring their own liquidity positions; I'll tell you about some of the challenges firms are facing in the measurement of those liquidity positions. And then, just toward the end if we have time, I thought I would throw in two illustrative cases I picked at random to illustrate the complexity of the issues, and why there's probably a whole lot more to be done in this space. Then I will conclude with some few high-level thoughts.

So as I said, in terms of measuring and getting a grip on a firm's liquidity as the theme is asking us to do: market liquidity is obviously critical to the assessment, so in the simple diagram that I put out here, when one is thinking about liquidity assessment, I think it's obvious that you need not just to think about assets or positions that exist on a firm's balance sheet, but obviously you have to tie that into market analysis as well, given that there's obviously going to be huge pools of ostensibly high-quality liquid assets that firms are holding to meet needs and the need to monetize those assets. Interestingly enough—and to tie it into specifically to somewhat Pete's work is pointing out—these pools being held by firms are typically sized to meet 30-day liquidity needs, so this whole notion of immediacy of cash flow and speed of liquidation is critical to the assessment of a firm's liquidity.

So Pete's paper, to me—and based on also what we heard from Greg—raised some interesting questions for all concerned, all stakeholders, whether it be your regulator or bank manager, an investor, to just give serious thought to some interesting questions. So if we take it that the corporate bond market—or the other way around, the Treasury market—is 55 times more liquid and operates at a speed that's 3,000 times faster than less-liquid corporate bonds, it really enforces the notion that policymakers had, I think, when in defining highly liquid assets, there is an inclination to lean more toward sovereign-quality collateral.

So having been part of, in my prior life, associated with some of the discussions around the LCR [liquidity coverage ratio] creation and the NSFR [net stable funding ratio] creation, and what sort of positions supervisors should be looking at—there was a natural tendency that you would expect banks to hold more sovereign-quality bonds...and I'm not saying U.S. Treasury bonds, I'm saying "sovereign," broadly speaking, so that it picks up sovereign quality in foreign jurisdictions. It would be interesting, by the way, to get a view from Pete and his paper, and we've talked some about some of these issues: Greek liquid, Greek bonds, and sovereign bonds, how does that liquidity translate versus U.S. Treasury or gilts or bonds, for instance. So not every sovereign bond, I would argue, is necessarily going to have this liquidity advantage.

Having said so, I think the analysis ought to raise some interesting questions about ultimate calibration of liquidity that supervisory bodies have set up, and which, as you will see in a minute or two I point out that firms are largely following. So, when you think about, for instance, the Basel or the U.S. implementation of the LCR standard, as much as 40 percent of a bank's liquid assets can be held in nonsovereign securities. That is a huge amount when you think about—and this is all public—you take any bank's financial, or some of the G-SIFIs financial statements, you see numbers 400, 500 billion dollars of liquid assets just at one institution. So you think across the spectrum of G-SIFIs [global systemically important financial institutions], and when you start talking about 40 percent of assets held in nonsovereign securities, there's a tremendous expectation that tends to build, then our question about how exactly will liquidity and cash be generated? How easy, in fact, will it be to monetize some of those securities?

The standards that are out there as well also don't make distinctions between things like on-the-run visits, off-the-run Treasuries, so a Treasury is a Treasury. And again, Pete's work focused on the most liquid part of the Treasury market, what if an institution is holding another type of Treasury bond, how does that impact the firm's liquidity?

And similarly as well, in its wisdom I think that the regulatory bodies did have a cut-off in terms of the quality of corporate bonds that could be allowed into, or would be allowed into, a liquid asset pool. So it's not the most illiquid bonds, but having said so there's a fairly broad range of collateral that's allowable, and so there are questions about the mix of HQLA assets, the high-quality liquid assets, and the assumption about a bank's ability to monetize those, especially given as I said that these needs were meant to be covering cash needs in a very short window of time.

So it begs another question, a bigger question, which Bill Dudley started to address last night, which is about the role of the lender of last resort. If you have...just think about the magnitude of monetization that has to happen in a very short period, if the scenario that regulators and banks—think about a combined idiosyncratic and market shock happening simultaneously, and you have huge pools of assets to be liquidated. Where is that liquidity going to come from? And will, in fact, the official sector, even though it's been saying that it should not be looked to as that ultimate lender in such situations, maybe there's a question there about whether or not, from a practical perspective of a lender of last resort, it's not ultimately going to be the way that this liquidity is generated.

Having said so, there are obviously complications, I can see the official sector easily taking sovereign-quality bonds. When you get to wrestle 1,000 equity securities, is the official sector internationally going to be willing to lend against those sorts of collateral to the extent a bank or financial institution is holding those sorts of assets in their liquid asset pool?

So a lot of questions for us to think about. There is also a question, I think, about the operationalization of the monetization process, so does the market infrastructure exist to support a rapid liquidation of assets, monetization of assets, to the extent there were investors and liquidity available to be taken out by the institutions? So just to link Pete and Greg's comments in to the broader assessment of a firm's individual liquidity.

So what I'll do next, I was asked to talk some about current practices that have evolved, so for the next few minutes I will talk about a firm's practices in measuring their liquidity, and then some about current challenges that exist around those practices.

So, high level: we're going to look at firm liquidity, and as I think has become clear by now, when assessing a firm's liquidity and particularly given the nature of the complexities of businesses and how those businesses may or may not be subject to different regulations and different supervisory bodies, it should be clear that consolidated level analysis is completely insufficient, that you really need to go to a much lower and more granular level of analysis. That takes you basically down to the legal entity level, and certainly when you start crossing jurisdictions as well, you have to start thinking about currency type issues and jurisdictional issues.

We'll get to some more of that in moment. Firms have largely adopted stress testing techniques, so this all started, I think, in the capital space and has quickly moved into the liquidity space as well. So there has been significant evolution and continuous evolution in the stress testing space. Stress testing over a variety of scenarios and over time horizons as well.

Critically, and where there's also probably a lot more work to be done: intra-company flows and intra-company exposures, including if you think of complex transactions like derivatives, trades spanning globally prime brokerage businesses, which operate with a global clientele—you get to quickly realize that intra-company cash dynamics and dependencies also have to be sorted out. So that's the space that I think firms are continuing to spend a lot of time evolving, and also I think supervisors are also on top of that pretty closely.

So what are firms doing in terms of their current practices with the measurement of liquidity, how do they assess their own liquidity? By and large, firms have adopted the cash flow measures that the regulatory community has suggested and put out publicly. At this point in time in the cycle, there is not a whole lot of new, innovative thinking about liquidity measurement. I think, instead—this is not meant to be a ding on the industry or anything—I think innovative thinking is now shifting to, how do we meet the regulation and what sort of product creation do we go through, how do we create new products that meet client needs as well as satisfy the regulatory needs?

So there's not a whole lot of thinking away from let's look at cash flows and stress cash flows and determine our needs over various time horizons. Material legal entities are certainly now being captured, I think, prodded by regulatory questions; the largest legal entities within firms, particularly broker-dealer entities and main banking units, are now definitely in scope and being analyzed. I think what has to happen, and given some of the trading and the way cash is moved around, ostensibly as a tax optimization, is often a big driver of how things are moving through the system.

And then there's largely been an abandonment of more traditional measures, the cash flow measures, just because it was clear there during the crisis those measures were not capturing liquidity risk very well. Market stress testing has largely been abandoned. I think people have taken the regulatory definitions and just assumed that the depth of market will be there, so as we talked about, there is a big question on that.

In terms of some of the challenges: data availability, good time-series data has been very difficult for firms. Often the data just was not tracked, it was not captured, analytics therefore are sometimes weak and lacking. There's also a big question around calibration of stress elements, largely because to the extent data is available, the stripping out of the impact of official sector support, for example, during the financial crisis, has not really been attempted by a whole lot of institutions, and so that creates a complication because whether or not or how easy it is to try to isolate out the impact of official sector intervention is...it's a very difficult question. So how it impacts the data, to the extent the data is collected, is another area.

Risk capture: by and large firms have recognized their particular areas of vulnerability. I think where there is more work, potentially, it's just how things are linked and the interconnectedness across businesses. Organizational structures: Greg talked about a first and a second line; those are key.

And then the systemic interconnectedness is something that really is in its infancy in terms of good analysis, and so that is also an area potentially that makes some of these liquidity assumptions questionable.

I'll quickly throw in some illustrative cases, and with a minute I'll just put up a graph—a simple diagram, this just illustrates, real quick, for the prime brokerage business, for instance. Two clients, one looking to take a margin loan [and] buy securities, which it's financed in the repo market—those same securities may actually be borrowed by a second client, and sold short. In terms of understanding the liquidity implications of a structure like this, one has to make many different assumptions about the timing of when these transactions may unwind, and the synchronization in order to get to reasonable assumptions. Being able to call exactly how all those pieces may play out, and the impact they have on liquidity—obviously, a huge challenge.

So that's on the security side. On the banking side, if we look at what has happened in the banking system since 2008—this is all Federal Reserve public data—you can see the huge growth in surge deposits, and so with that comes a lot of questions about where those deposits [are] coming from, what's the potential behavior of the depositors, and in a stress will those deposits really be there and be sticky?

So, quick conclusion: firms' liquidity measurements are convergent around regulatory-defined cash flow approaches—not a whole lot of innovative thinking as yet in this liquidity cycle around new and innovative approaches. Monetization of assets in private markets remain to be tested, and I would say there are a lot of calibration and certainties around the risk modeling.

Having said so, though, I will say that there has been a tremendous amount of improvement in the resilience of the banking sector, no question about it. Firms are much more liquid and resilient than they were precrisis. At least there's one good element to this. So, thank you.

Kupiec: So, thank you to both our discussants and to Pete, and, as I mentioned, Pigeonhole was working in the Twilight Zone—working very well. And we have a number of great questions, and what I'm going to try to do is group them a little bit. Some of them are a little bit technical about the model, which are very interesting questions, I think, and others a little bit more policy oriented. So we'll try to get them together, and I hope all of our panelists will help answer them.

So before I start, the first active question is kind of a technical model question. And when I'm reading, Pete, your 2016 forthcoming Econometrica paper, what I find in the details of the model is that you're really talking about modeling idiosyncratic risk, and idiosyncratic risk is a risk that doesn't really have a price. And then when we move to this paper here, we're dealing with Treasury markets, and you kind of think of Treasury markets as being—not having idiosyncratic risk, right? Treasury is a whole signal about the whole economy, there's nothing idiosyncratic about...maybe the off-the-runs have idiosyncratic risk or something, but yet you apply your model of—it's essentially a model of idiosyncratic risk—to all these different securities, and you get surprisingly good ballpark results. Like they kind of match the numbers.

Can you talk a little bit about systematic risk versus idiosyncratic risk, and how you think that should play into this in the future?

Kyle: Yeah, I think that the risk aversion that most people talk about as being important in finance—I remember Alan Greenspan used to talk about it—but finance professors build their finance courses around risk aversion. As a practical matter, I think it shows up in the volume and volatility in the market. So if—let's say, if Alan Greenspan says traders are becoming more risk averse, what I hear from that is that volatility in the market is increasing, because in order to get those traders to absorb bigger positions you have to offer them a better risk premium, which means a bigger price concession and therefore prices will move more.

Similarly, if you look at it more from the academic perspective, I think that you have models in finance that say there's some people that demand liquidity and want to trade; other people, who supply it, might be influenced by risk aversion or might not be. In the case of idiosyncratic risks, maybe it's they're not influenced so much by risk aversion but they're influenced by the possibility that the orders that they're trading against might have private information in them. But if they're trading, say, aggregate stock market risk, or aggregate bond market risk, it might be the case that there's just an exchange of risk going on, and so the volatility of the whole stock market—the VIX index—that's going to be a kind of measure of how much risk aversion is operating in the market right now. And with idiosyncratic risk it's probably not driven by risk aversion so much as by information content of the trade.

Kupiec: OK, I'm going to mark that one as answered. Maybe if you don't think it was, you can, we can take it up at the...but here's a related question: the numerator of gamma looks a lot like a VaR—right?—of an average transaction. And the problem with VaR is it changes rapidly over time, because sigma changes; do gamma and 1/L have the same problems as VaR, practical problems?

Kyle: Well, it's certainly the case that if you look at gamma and you look at 1/L in a time series, and compare that with measures of liquidity that people have put together in the empirical finance literature, you're going to find that most of the time-series variation in our measure of liquidity is going to come from standard deviation of returns changing. So during the financial crisis, assets across the board became very illiquid, and if you ask yourself why that happened, well, maybe some of it was due to the dollar volume going down in some assets, but most of it is due to volatility increasing. So these models don't allow you to predict what's going to happen in the future, you can't really predict liquidity from them, nor can you predict returns from them, but they do give you an estimate of what the transactions cost would be if you tried to transact reasonably soon at a reasonably normal rate.

On the other hand, you can be surprised, you might go into the market and try to transact, and then conditions suddenly become unsettled, and you find that things are worse than you thought, typically because you'll see volatility going up.

Kupiec: We're going to knock that one off as answered, too. There's a number along here—and this one is...let me pick up one that is for Pete but for everybody. And it's harder to use this than you might think, sitting up here! I'll find it in a second...

We'll throw this one right to Mr. McGreevey, and I'll find the next one. So, Mr. McGreevey mentioned liquidity disclosure is a good thing. Have you considered that disclosing declining liquidity could actually create a run on liquidity?

McGreevey: Yeah, so first of all, thanks for the question. Thanks for making it anonymous, too, I like that. That's fantastic.

So, I think from our standpoint, I don't know if we necessarily think it would create a run, we certainly had some conversations about that. The presumption is that if you disclose liquidity it's going to get into the hands of individuals who may use that unintendedly. So I guess we would take a little bit different view, that if we can come up with something that's reasonable in terms of disclosure—give people something that's practical at the end of the day, they can use that to be able to compare liquidity within given products and within funds, and that may lead to them deciding that they may go to another fund. But the scenario of creating a run on liquidity we think would be largely overblown and at the end of the day, we kind of think that there's more benefits than there would be issues on the other side of this. So hence why we'd be for the liquidity disclosure of some form.

Kupiec: Anybody else want to weigh in?

Sooklal: So having been part of discussions around disclosure issues I think one of the beliefs on the other side of the argument is that if you have ex ante a lot of good disclosure, that hopefully the investing community and your counterparts, your lenders, whomever, will have a better understanding of the firm. It will also force internally within institutions harder thought about where there might be vulnerabilities. And since you're going to be held to a public disclosure standard that market discipline should be beneficial to the firms themselves as well as to the investors and others who may be deeply investing and taking that risk. So there is that other view, which is that ex ante, if you get that information out, stakeholders really fully understand that disclosure in fact could be a good thing.

By the way, the other thing that we've seen repeatedly is that any time there's a lack of information in the marketplace, and you think the European bank crisis of three or four years ago, you think more specific firm crises that have happened, anytime there is a lack of information in the marketplace coming from the institution, there tends to be a flight away from that institution and market makers and others start creating their own "information," so the other side of this question is whether or not better information actually does in fact bleed down to the benefit of the firm itself.

Kyle: So I've got a little bit of perspective on that question. If I'm investing in a mutual fund that has a lot of liquidity, what that liquidity is telling me is that if everybody else runs I'm going to be the one that's not running and therefore I'm going to be the one that is buying the assets that the other people are selling. So I don't necessarily view a mutual fund as having a lot of liquidity as a good thing for a long-term investor in that fund. I would much rather see it have low liquidity, see the assets marked down, see the other investors leave at low prices, so there's more money for me, the long-term investor in the fund.

Kupiec: Thanks, so I have two questions that are very much the same, and of course, very important. I'll pick the shortest one, but they're the same. You suggest, Pete, that the last crash was caused by about a two bazillion, or billion, or however you say it, trade. Precrisis, those sizes were common. Has liquidity changed since the crash? And the other question, which has actually more votes, was, "have you looked at data over different periods?" Why don't you go last? Why don't we have the two practitioners tell us what they think about "Has liquidity changed since the crash?"

Sooklal: I can offer one thought, which is that Greg talked about bid-offer spreads and the TRACE data; I think the concern with some of what you're saying there is the TRACE data picks up the reported trades, so for the thousands of other trades potentially that may not have happened when bid-offer spreads between buyers and sellers might have been potentially much wider—they're not captured in the data centers, a natural bias when you look at bid-offer spreads.

The turnover ratio is actually also interesting because—I was having this discussion with someone as recently as last week about the numerator and the denominator impact—and some of the turnover data is actually suggesting what intuitively would be the opposite of what one might expect, which is that trading in highly liquid sovereign bonds is actually lower than trading in corporate bonds. And I am not sure that is really the case, I think there are some dynamics happening in the marketplace around ownership of the bonds, the amount of Treasury securities that have been issued, the fact that banks are holding a whole ton of them...and so there may be less liquid, or less trading [and that] is all impacting the dynamics, so I'm not sure I would make a broad-brush statement that liquidity has actually crashed.

Kupiec: Greg, could you weigh in a little bit?

McGreevey: Yeah, I probably, I think some of that I covered a little bit in the brief discussion, but I think it's a really tough question to answer. I don't want to be vague, but it really depends on the factors that you look at overall. Transactions costs clearly have not increased, they probably are at the same, as I mentioned, precrisis levels; that's just marketplace data at the end of the day. The ability to transact in the marketplace—at least, what it feels like on the trading floor each day—we're able to get trades done, sometimes it's going to take more work and more effort. There is no doubt that turnover is substantially less, it's a fraction of what it was, and bank balance sheets and broker balance sheets are substantially less. So, the result of that, clearly, is, it makes it a little bit more difficult at the end of the day to get your trades done.

When the market volatility comes up—and you saw that in the first couple of months of this year—even with the outflows (and I'll use the bank loan marketplace, and I wasn't able to be here last night, but I know that was a topic of part of the conversation) that market on the retail side, which is about 10 percent of the market, has declined about 30 percent, and those redemptions during a pretty difficult time were able to be met. Doesn't mean you're not going to have a little bit of widening of bid-ask spreads, it doesn't mean that at the end of the day it doesn't have a cost at the end of the day to the fund, but I think liquidity was able to be transacted.

So it depends on which measure you look at, at the end of the day to decide whether it's lower today or about the same as what it was precrisis.

Kupiec: OK, Pete, bring it back to 1/L.

Kyle: Well, I think that—as I said earlier—during the crisis, volatility went up a lot, and that had a huge tendency to decrease liquidity, to make 1/L go up. Volume maybe is a little bit lower now than precrisis, but if you go back far enough I think volume hasn't changed that much either, and so if you look for a long time trend in liquidity, it's probably not there. The markets for Treasuries have been very liquid for a very long time and the markets for corporates have been very illiquid for a very long time.

Now, the question referred to $2 billion; I think the flash crash and the flash rally were the result of very rapid movements of positions in a market that was already kind of stressed. So, the very rapid speed, if you had tried that previously in markets on a day that was, say, a bad day, you could have gotten a flash crash there, too. So I don't think the markets have changed, but I would entertain one possibility, and that one possibility might be that electronic trading systems that people use—they use algorithms now—and we might be a bit early in our understanding of how to use algorithms wisely, so one time out of let's say, a hundred thousand, somebody's going to do something that's just not a wise use of an algorithm. In the old days, when it was a human being doing it—every now and then a human being is going to do something that's not wise, either—I've seen that happen before, but in the case of these algorithms it might be that they just sped up their trading very rapidly for reasons that the human being who is using them didn't quite understand. And that resulted in very rapid movements in prices, and very high transactions costs, and what we call a flash crash or a flash rally.

Kupiec: OK, thanks, Pete. So we have another policy question here, which—if I press this right—will show up. Do central banks need to move away from constructive ambiguity to a much clearer posture on lender of last resort? The Bank of England is going there, the Fed—not so much.

Kyle: While everybody else is thinking of it—and we hope to hear from the others—but I think central banks should make it very clear that they will accept very safe but very illiquid collateral for lending securities, and that they will not take unsafe collateral. So something that's rated AAA, AA, but extremely illiquid—and the bank can verify that the ratings agencies aren't really messing up on the rating—that should be taken as collateral from anyone. And I think the central banks should have a continuous interaction with the market to make sure the mechanism is working.

McGreevey: The only thing I'd add to this, maybe if you just step back, because embedded in that question is a lot of central bank activity during and right after the financial crisis. My own view is that the Fed, in its various forms, did a phenomenal job of dealing with an unbelievably challenging situation, and just where we were at within our financial systems, and we spend day in and day out time trying to look at economic formation, all the nuances of Fed activity and other things related to interest rates. So I know I could only, if I were there, screw things up...so I think it was phenomenal.

Now, specific to the question, I think any communication and transparency and clarity is always a good thing, so I would say to the extent that whatever policy they decide to do, that policy getting out to the marketplace on what they're going to use for collateral and the requirements around that, is kind of a commonsense approach, and I think it's something we'd support.

Kupiec: Thanks. Let's move on to another interesting question—well, we have some great questions here—Volcker rule limits banks' ability to inventory bonds, so it is forcing faster time on securities—so is it forcing faster time on securities whose gamma is slow? If so, what are the consequences?

Kyle: I think that it probably is. So, there's one example I didn't talk about, and that would be doing arbitrage between on-the-run and off-the-run Treasury bonds so that on-the-run bonds have a very fast gamma, the off-the-runs very slow. And so the Volcker rule is kind of discouraging prop trading, and so what that is going to do is widen the spreads between on-the-run and off-the-run bonds, and maybe make the spreads a bit more volatile.

It's not just the Volcker rule, it's really the restrictions on leverage that are amplifying that, and so I think that's kind of what you'd expect to see. You would clearly also expect to see less arbitrage with, say, corporate bonds or illiquid mortgage-backed securities against other instruments as well, both because of leverage restrictions and because of the Volcker rule.

But to me, the biggest issue is these very safe assets that have, let's call it a low gamma—or even a pretty high gamma—blunt rules that are designed to cover everything that don't respect the tremendous variation in gamma are likely to backfire. So the worst issues are the things that actually have a reasonably high—the worst problems are likely to occur in the things that have a reasonably high gamma.

McGreevey: The only thing I'd add to that is, I think when you look at the data—and I totally agree, it's broader than the Volcker rule—but you look at turnover, and part of that's the size of the market, it's gone down kind of dramatically, if you will. If you look at bid-ask, you look at on- and off-the-run, those things necessarily don't suggest that there's necessarily forcing faster times on securities. So I think we know the impact of it is lower bank balance sheets, and I'm not convinced that we therefore can conclude that it leads to faster times when you look at the data.

Kupiec: OK, I'm going to say we finished that one, too. So now, I'm not sure I know what they're referring to exactly but I'm going to assume you guys do: can you explain why the SEC three-day liquidity would not work?

McGreevey: Yeah, so for those that aren't familiar with it, the three-day liquidity requirement is a requirement that a fund is going to have to meet its liquidity—[it's] somewhat undefined, but that liquidity is the redemptions that historically have existed without a move, if you will, or a significant move, in its NAV. And so, you can disintuitively think through that, and it has a number of different things—what redemptions rates should I use on my assets overall, what does it mean to not have an impact, if you will, on the NAV in a material way, and it is over all kind of market environments? It requires you then, to have a variety of other things, in terms of disclosures and things like that.

So we understand what the SEC is trying to get at, we just think that the way that it's worded right now, just pragmatically, is going to be difficult if not impossible and could lead to maybe some other issues, as it relates to underpinning what they're really trying to get at, at the end of the day.

Kupiec: I think we're going to mark that one done. Here's an interesting one, too—and this is open for certainly all of you—I'll read it up here since it just moved on me: Have market structure changes, like circuit breakers that have taken place at different speeds, in correlated markets (cash versus futures), do they contribute to volatility and reduce liquidity? What's your views on that?

Kyle: Well, my view is that the circuit breakers they tried in China didn't work, and so they got rid of them very quickly, because the bands that they had placed on the market seemed to be attractive barriers, not barriers that the price bounced up on—off of—or barriers that slowed down price fluctuations. It seemed like as the market approached the limits, they would quickly move to the limits, and then panic would set in. So, they just didn't work.

Kupiec: Greg? Jay?

Sooklal: So I was going to say something similar. Along those lines, I think once there's this thought of the rapid decline, people almost look to the limits, and I don't know that the limits have really in the past, as circuit breakers, proven effective, and ultimately, determining the level that the particular securities would go to. So I'm not sure it has really...

McGreevey: Yeah, I think if you look at CBO [Congressional Budget Office] data, and other things, the China situation, I think people know where those kind of circuit breakers are, if there's going to be a or at least have some intuitive understanding of it, and so I think the research that I've seen out there would conclude that it does lead to a little bit greater prospect for volatility. China's a unique situation because of the outflow situation; they were trying to deal with post-reduction in their currency valuation, but it's one of those things that sounds good [but] at the end of the day, it's complicated, it probably leads to some unintended consequences.

Kyle: So I would just say one case where you do want a circuit breaker is when a piece of the machinery of the system has failed, either an equipment failure or a firm has gone insolvent and the clearing and settlement system is dubious—then you may want to stop trading briefly. You know, like the World Trade Center—that would be an example where we wisely stopped trading.

Kupiec: We'll try to get in one more quick last question: Isn't the real liquidity problem in bond mutual funds the promise of daily liquidity when the underlying assets are liquid? Aren't requirements to hold cash and cash equivalents simply a palliative?

McGreevey: Well, I think when you look at the data—maybe I'll start on the mutual funds side—there's lots of things that go into managing a mutual fund, and part of it is the credit risk, and all against what objective you're trying to manage, and the expectations for those that are investing in the fund...so, even if you had a completely illiquid, you didn't have daily liquidity, there's still a need at the end of the day for being able to have some level of liquidity at the end of the day. We do that in a lot of our hedge funds that have lockups and a whole bunch of other things.

So I'm not convinced it's sequitur to say that because I have daily liquidity and all of my assets aren't daily that I can't have a product that still has the promise for liquidity at the end of the day, because not everybody typically is going to be heading out the door at the same time, etc., and there's a way to be able to manage those redemptions, I think, for those funds who—and those managers who—know what they're doing. So, again, I understand the question, I maybe have a little bit different slant on it.

Kupiec: OK, why don't...since it's my job to bring us back from the Twilight Zone, back to the real break time in the real world, why don't we pick up this conversation outside at break—I think we have a break next—and let's thank our speaker and our panelists for a great session.