Paper presented by Itay Goldstein, Professor of Finance, University of Pennsylvania


Itay Goldstein: OK, thank you very much. It's really great to be here. It's an amazing location and the morning sessions were very interesting. What I will do in the next 25 to 30 minutes is present the paper that we have written about flows and potential for fragility in corporate bond funds. I will give you the motivation and summarize the empirical findings that we have and then talk a little bit about the implications. I should say that I have two coauthors on this paper. One of them is Hao Jiung who is in Michigan State University, and the other one is David Ng who is in Cornell University.

The focus of the study is mutual funds that hold corporate bonds, so these are corporate bond funds. They are holding corporate bonds which are considered to a large extent to be illiquid, but then these are open-end funds, meaning that investors can take money out of these funds basically every day. There are many reasons why this is an interesting and important topic to think about. One potential reason is just the fact that they have grown tremendously over the last years.

And you can see here in the red columns the total net assets they hold going up to about $1.8 trillion. A lot of the spike has come up during the crisis and after the crisis, and people talk about monetary policy that was leading to this trend, the constraints on banks that were leading to these trends, and others as well, but at the end of the day, we now have many corporate bonds that are held by open-end mutual funds.

If you are wondering what happens to this particular type of fund relative to all other bond funds, you can see it here in this graph and the blue area shows you corporate bond mutual funds as a fraction of all fixed-income mutual funds, and this has also been trending up quite significantly. So it's not just the growth in bond funds that has been generating it, but within bond funds there is a shift more and more toward the corporate bond funds.

So, as you probably know, many commentators have looked at this and basically started talking about concerns for fragility. There was a paper about two years ago by Anil Kashyap, Hyun Shin, and others, basically saying that, the fact that we have so many assets held by corporate bond mutual funds, and the fact that monetary policy is so loose and will eventually reverse, is basically giving the materials for potential fragility. What happens when monetary policy is tightened up, people would like to take the money out of these funds, and this will potentially have adverse implications for bond prices for the economy. They conducted some basic analysis that kind of suggests that these forces are there.

At the end of the day when we look at it, the thing that we would like to really understand at a more basic level is, what are the patterns of flows in corporate bond funds? How do people decide when to take money out of these funds? How do they respond to performance? We don't have that much evidence, and this paper is trying to fill the gap.

When you look at the literature on mutual funds, there is a huge amount of research on equity mutual funds, and in particular, the relationship between flows and performance in equity mutual funds. It is such a huge literature that there is a survey by Susan Christoffersen, David Musto, and Russ Wermers dedicated to this particular topic—only the flow to performance relationship in equity mutual funds. They cite many papers. They talk about the variance fact that people have found in this literature. One thing they say at the end of their paper is, unfortunately, there is very little research on flows in bond mutual funds and we need more research on that. So we are providing at least one piece of research in this paper.

Basically, at face value what we do is, we look at the flows in about 1,660 actively managed corporate bond mutual funds in the period between '92 and 2014. Then we look at the sensitivity of flow to performance, the relationship between flow and performance, and how this is different relative to equity funds. And as you will see, the main theme behind our analysis is liquidity and I will talk about this theme in a minute. So we are going to link the patterns in corporate bond mutual funds to the underlying liquidity, and we will draw conclusions based on that.

So the very basic finding that we have, but this is by no means the main thing that we do, the very basic finding that we have is that the flow/performance relationship in corporate bond funds is different than in equity mutual funds. The one feature that everyone has documented in the context of equity mutual funds is the fact that you have a convex flow/performance relationship. What does that mean? It means that you have greater sensitivity of inflows to good performance than of outflows to bad performance. This is the convex shape. What we show here is that across various specifications, at the end of the day you don't see this convexity in corporate bond funds, and if anything you tend to find more concavity in corporate bond funds.

So this is the picture to tease your appetite, hopefully. If you're thinking about equity funds, which you see here, in our sample we can still replicate this convex shape, whereby you see much more sensitivity of inflow to good performance than outflow to bad performance. Then when you look at corporate bond funds, you tend to see the opposite. The blue line is the main estimate. You tend to see the opposite, whereby the outflows are more sensitive to bad performance than inflows are to good performance, and this is kind of the motivating fact behind many of the tests that we are going to conduct later on.

The main story that we are going to try to develop and test is a story of underlying illiquidity. Basically what happens in mutual funds, they hold assets and they offer investors the ability to redeem their shares at any day, so every day you can come and say I want to take my money out and you will get your money out as of the net asset value [NAV] at 4:00 p.m. on that day. Now, sometimes, there is a liquidity mismatch here because the mutual funds are holding fairly illiquid assets, and as a result, when investors are promised liquidity on a daily basis, you have a mismatch between the liquidity of the assets and the liquidity of the liabilities, the liquidity of potential redemptions.

We have looked at that in a previous [article] that was published in the Journal of Financial Economics in 2010 in the context of equity mutual funds. We were talking about the differences in the level of liquidity across different equity funds and how this potentially creates liquidity mismatch and potentially affects outflows. But the basic idea is once you have this liquidity mismatch, the greater is the liquidity mismatch, the greater is the illiquidity of the underlying assets, potentially you create a first-mover advantage at the level of the fund, whereby when people come in to take their money out, they are getting prices that are not yet reflecting all the future transactions costs, liquidations costs, trading costs. As a result, they are getting a better price. Whatever negative externalities they're generating, they're imposing on those investors who stay in the fund. That potentially leads to a first-mover advantage.

You can see it here in this illustrative picture. So let's say that we are now at day one, investors in a mutual fund decide that they want to take their money out, they want to redeem their shares, they will submit an order to the fund—they can do it up until 4:00 p.m. that day. Then they will get their money out as of the net asset value at 4 pm of that day. The problem will arise if this net asset value is not fully anticipating all the future transactions, liquidation costs, and so on that the fund will have to bear in order to accommodate these redemptions.

What will the fund have to do? The fund will have to start trading, burying commissions, bid-ask spread, potential price impact. The fund will have to deviate from the desired portfolio. The fund will have to liquidate assets based on liquidity needs rather than information. All these are going to potentially reduce the performance of the fund going forward.

Now the fact that the net asset value was quoted as of 4:00 p.m. on the day of the redemption, and the fund is conducting those trades later on essentially implies that people who took their money out on that day are getting a better price and they are imposing some of the negative externalities on those who stay in the fund. That is the potential for the first-mover advantage.

Now this first-mover advantage is going to be stronger in those cases where the liquidity mismatch is more severe—when the fund is holding more illiquid assets. Because then, there is a mismatch between the liquidity of the redemptions and the illiquidity of the assets. I mean, clearly if you are a mutual fund that invests in S&P 500 there is really nothing to talk about. Investors are going to come in and take the money out. The fund will have to maybe conduct some selling and stuff like that in the future, but this is not going to create much in terms of price impact, deviation from the desired portfolio, and so on and so forth. At the end of the day, the externality you impose on those who stay in the fund is not that big.

But if you are thinking about the corporate bond mutual funds that invest in emerging markets, and now you have this underlying portfolio which is very liquid, and people come to you and say we want the money out today, and you quote them whatever price you have even though the bonds have not been trading in recent days, and so on and so forth, then essentially what happens is the net asset value that those investors can get will be higher than the "fair net asset value." Those who stay in the fund will have to suffer those liquidation costs and they will get a worse deal. That creates this first-mover advantage.

What this is doing is potentially amplifying the reaction of flows to bad performance. Because if I hold my money in a fund that has illiquid assets and I see some bad news, bad chart, bad performance, whatever, and I think people are going to take their money out, the fact that this is an illiquid fund is going to amplify my incentive to take my money out because I now say, if others are going to do it they are going to impose the negative externalities on me. As a result, I want to take my money out as well.

So basically at the end of the day what we would expect is that the sensitivity of outflows to negative performance is going to be stronger in those funds that hold more illiquid assets where the liquidity mismatch is stronger. We have shown evidence on that in a previous paper as I mentioned in the context of equity mutual funds, so the sensitivity of outflows to bad performance in illiquid equity mutual funds is stronger than in liquid equity mutual funds. But, at the end of the day, equity mutual funds are by and large fairly liquid. I think the problem is going to be amplified in corporate bond mutual funds, and this is why we look at them here.

This may be one explanation for the difference in the sensitivity of flow to performance between corporate bond mutual funds and equity mutual funds, just given the fact that corporate bond mutual funds are so much more illiquid than equity mutual funds. But clearly, we want to dig deeper and the tests that we will conduct will zoom in on this liquidity aspect and look across different corporate bond funds and over time, at corporate bond funds and see whether liquidity can indeed explain this excess sensitivity of outflow to bad performance.

These are the main tests that we are going to conduct here. Again, the first hypothesis is essentially what I already showed you, given that corporate bonds hold more illiquid assets than equity funds, then the sensitivity of outflow to poor performance in corporate bond funds is going to be stronger than in equity funds and this is what will lead to this so-called concave flow-performance relationship relative to the convex flow-performance relationship that you see in equity funds. Now, zooming in on the corporate bond funds, again, we will expect to see variation across funds depending on the level of liquidity, and over time depending on the level of liquidity in the corporate bond market.

Hypothesis two talks about the over time and basically what we say is that when the corporate bond market is more illiquid, the sensitivity of outflow to bad performance in corporate bond funds will strengthen.

Hypothesis three talks about the cross-fund dimension, comparing across funds—funds that have more liquid assets will generate weaker externalities, and as a result, the sensitivity of outflow to bad performance in these funds, those that hold more liquid assets, will be weaker than in those that hold more illiquid assets.

Finally, the last hypothesis has to do with the clientele, and this is a bit more subtle. I will not have time to fully explain it here. The idea is that taking your money out of the fund in anticipation that others are going to do that is kind of a coordination failure. If you look at theoretical models, and we have done one of them in the previous paper, the prediction is going to be that this tendency for coordination failure is going to be more severe when you have small retail investors rather than large institutional investors. As a result, this effect of illiquidity on the sensitivity of outflow to bad performance is there only for the retail oriented funds, and not that much for the institutional oriented funds. These are the main hypotheses that we are going to take to the data. And again, all of them highlight this connection between illiquidity and the sensitivity of outflow to bad performance, and they're based on models of coordination problems.

I have a little time so I'm not going too much to get into the technical details, but let me just say that we construct flows in the usual way that is done in the literature. We have to think about measures of performance because at the end of the day we look at the sensitivity of flow to performance and how this is affected by illiquidity. Thinking about performance is a bit more difficult in the context of corporate bond funds because we don't have that many studies and we don't have a consensus on how to do it. We measure alpha relative to the performance of the bond market and the equity market. But at the end of the day we also conduct robustness checks, and you know, the big picture is still there.

The first regression that I will show you...this is the kind of regressions that we have. We are basically explaining flows on the left-hand side. Flows are going to be a function of a, which is a measure of performance. The key thing we are focusing on is the interaction between a and -a. So this will be this coefficient b2 here. When this is positive, it means that you have more sensitivity on the downside than on the upside. When this is negative, it means the opposite. So, b2 being positive implies a concave flow-performance relationship, and b2 being negative implies a convex flow-performance relationship.

You can see here, this is the comparison between the equity funds and the corporate bond funds. And as you can see, that in the corporate bond funds b2 is positive and in the context of stock funds it is negative. Again, this is essentially the picture I showed you in the beginning but now it's a regression analysis imposing linearity and so on to get statistical significance. But you can see that the coefficient is positive for corporate bond funds and negative for equity funds, suggesting that the sensitivity of outflow to bad performance is greater than the sensitivity of inflows to good performance in corporate bond funds. The opposite holds for equity funds. This is sort of this extra sensitivity on the downside for the corporate bond funds, which is what got us motivated. I'll tell you that it basically holds across subsamples, it holds when you put in fund fixed effects. As far as I can tell, the result is pretty robust to many different cuts of the data and many dimensions.

Let me move on to the second test. The second test is basically looking at changes in liquidity over time. And as I mentioned, what we expect to see greater sensitivity of outflow to bad performance at times when the corporate bond market is more illiquid. We use different measures of illiquidity of the corporate bond market, linking it VIX. All of them are based on literature, linking it to VIX, to the TED spread, having a particular measure of the illiquidity of bonds based on a paper that was published a few years ago. Across all the different specifications, at the end of the day what you see is fairly similar. These are only -a funds, so what you would like to see here is that the interaction between a and the illiquid period is going to be positive, suggesting that among those funds that have -a, there will be greater sensitivity to a when we are in an illiquid period. So in those times when the bond market in general is not that liquid, investors are going to respond more to bad performance in their redemption decisions.

The next thing is to look at the proxies for illiquidity at the level of the fund. There's a lot of discussion on that in the paper. We have different measures of liquidity, we talk about endogenous issues, I don't have time to cover all of it here. I will tell you that we measure liquidity in various ways. First of all, we look at the cash holding of the fund, we look at the cash holding and the government bond, we use different sources for cash holding based on CRSP and then based on N-SAR filings. We also measure liquidity based on liquidity measures of the underlying bonds and we employ different measures here like Roll ('84) and inter-quartile range of bond prices and again, all of these are explained in the paper.

At the end of the day the conclusion that comes out of here is still the same—that the higher the illiquidity, the stronger will be the sensitivity of outflow to bad performance. Funds can essentially undo this effect to some extent by increasing their liquidity. When you choose to not have that much liquidity, the sensitivity of outflows to bad performance is going to go up.

OK, let me skip those tests, I want to leave a few minutes for the conclusion. I will tell you that we do some of the tests on the last hypothesis that I mentioned, which is the difference between institutional and individual investors. As I said, from the point of view of coordination failures, you would expect to see more coordination failures among the retail investors rather than the institutional investors. People can come with alternative stories that will push in the other direction, but the evidence here by and large is consistent with this coordination angle. When you are thinking about the institutional oriented funds versus the retail oriented funds, at the end of the day our story therefore is the retail oriented funds are not for the institutional oriented funds. When you think about the effect of liquidity on the sensitivity of outflow to bad performance, the story is therefore the retail oriented funds are not for the institutional oriented funds.

In case you're curious about other asset classes, we conducted general analysis on Treasury bond funds and muni bond funds. In general, you see that in Treasury bond funds the shape is convex just like in equity. In muni bond funds the shape is concave just like in corporate bond funds. This also lines up with the liquidity story because munis are illiquid like corporate bonds and Treasuries are more liquid, closer to equity. So the underlying story of liquidity seems to carry through other asset classes as well.

One picture that I will show you about aggregates, and then I will move on to sort of summarize the findings and talk about conclusions. All of the analysis that you have seen so far is at the level of the fund. It's panel analysis, you have a cross-section of funds over time. These are all comparisons across funds over time. We didn't say anything about systemic risk. I should emphasize, and I will emphasize it again in the end, our paper doesn't really deal with systemic risk. At the end of the day I think that finding evidence for the systemic risk is incredibly hard with the data that we have.

Just to see if these results carry to the aggregate we did this analysis in the late section in the paper to think about what is the sensitivity of flow to performance at the level of the sector as a whole, and we compare the corporate bond funds here with the stock funds here. Interestingly, and this has been pointed out before, when you think about stock funds, and you think about the sensitivity of flow to performance in the aggregate, it's completely flat. It seems like in the aggregate, people take money from one fund and put it in the other, it's kind of flat when you think about the sector as a whole. But when we look at our entire sample over time, we do find that the concavity in corporate bond funds to also carry out in the aggregate. So there seems to be in our study, some action in the aggregate, but again, this is just one test toward the end and is not the focus of the paper. Let me skip the other aggregate analysis that we do in order to just leave time for the conclusion and implications.

So, taking the result at face value, this is what we show. If you think about previous literature, there is a lot of literature in the context of equity funds showing that there is a convex flow-performance relationship, there is greater sensitivity of inflow to good performance than sensitivity of outflow to bad performance. People have looked at it from many angles, criticizing it, supporting it, that's the literature. We bring this type of analysis to the context of corporate bond funds that have become very important and interesting recently. We show that in general they behave differently. There seems to be greater sensitivity of outflow to bad performance than sensitivity of inflow to good performance.

We link this to the underlying illiquidity, and we show that the sensitivity of outflow to bad performance is higher once you have more illiquid funds. The funds are holding less cash, they are holding more illiquid bonds, and this sensitivity is higher in illiquid periods. We also do interactions between the illiquidity of the fund and the illiquidity of the period. It even strengthens things further, I just didn't show it to you. This is mostly holding for the retail oriented funds, not for the institutional oriented funds. As I mentioned, all these line up with the coordination story that I mentioned in the beginning. It seems like there is this kind of fragility at the level of the fund, whereby the way that the NAV is calculated does create an incentive for people to come early when they think that other people are going to come.

What are the implications? I do think that this evidence and other evidence from the other paper showed there is this force. This force exists in mutual funds and creates some first-mover advantage that could generate fragility. Though this is the kind of thing that once you say it, everyone in the industry jumps at you. People are concerned about regulation. Let me put a caveat on this, and I think we put a pretty strong caveat in the paper, that these results in themselves do not necessarily call for regulation. Because when you think about this kind of fragility, this kind of fragility is generated at the level of the fund, and funds could potentially do many things to alleviate it.

For example, they could hold cash buffers, they could change the formula for NAV calculation. For example, the swing pricing that is now being discussed is one such example, whereby essentially if many people are redeeming their shares on one day, the price is going to go down. That is essentially trying to reduce this first-mover advantage. The fact that funds are considering it suggests that they do think this forces underlying dare, and they are thinking of ways to alleviate it. You can put restrictions on the redemptions like redemptions fees or reduce the frequency by which you can redeem. There are many things that funds can do.

At the end of the day, I do think regulators should be aware of these patterns and this is for two reasons. One is, to the extent that funds do not have all the incentives to reduce this fragility, in the extent that there is externality across funds, then maybe regulators would like to take a closer look. This has not so far been strongly established. There are papers showing the effect of flows on equity prices in the context of equity mutual funds. So, maybe this externality does exist to some extent across funds but in general, this is something to look more into.

The other thing for which I think regulators should be aware of it, and this is where I will finish, is at the end of the day I think that the way we think about the financial system, we have to think of the financial system as a whole. We can't live in a world where we put all these restrictions on banks, but mutual funds are different so they don't have any restrictions. We have to think about equilibrium forces, because once you start putting restrictions on banks, all this money is going to flow elsewhere and risks are going to pop up elsewhere. People will say, well, there hasn't been evidence of a big run in mutual funds. That's right, I am aware of it. But, if it happens, it will happen once, and then there will be a new regulation that will change it completely. I think what we want to do is understand ahead of time what the structures of the system are and make it resilient ahead of time.

We need to think of the financial system as a whole, not just in sectors. We need to understand that if we impose more restrictions and regulations on one sector, activity is going to move across to other sectors and we want to take this into account. In that sense, I think that FSOC [Financial Stability Oversight Council] and their attempt to think about the system as a whole is useful, but we clearly need more work to understand exactly what are the externalities across funds and what type of regulation, if any, we want to have.

Sean Collins: Thanks, Paula, it's an honor to be here and to have an opportunity to comment on this important topic and this interesting paper. So, first of all, usual disclaimer applies. This paper lays out a, I grant you, Itay and I have a slight difference of opinion on what they actually say in the paper. I contend that more or less this is what they're saying, he says it's a bit of a stretch, and I'm willing to live with that. I will tell you if it's a bit of a stretch about what's in the paper, this is not by any stretch, different from ways people are reading this paper in the public. We hear these comments all the time from regulators and international financial institutions that this paper represents this view. If it's not true and he's walking back from that, that's great. I'm happy to hear that.

So let me put this out as a bit of a caricature about how this paper lays out a hypothesis of the world. Bond prices fall. After bond prices fall, because of a concave performance-flow relationship, fund investors redeem massively. To meet redemptions, funds sell bonds, bond prices fall further in the future, potentially creating a downward spiral.

This is an old story. You can find examples of this all the way back to 1940. Here are two examples: one from Time magazine in 1959, and one from Henry Kaufman in a Kansas City Fed Economic Review article in 1994. These kind of dire predictions obviously didn't pan out. While this is a bit of an old story, there are two things that are new in this paper. First, as the paper suggests, the first-mover's not panicky, irrational investors, it's more super- or even uber-rational investors who are making very sharp movements, on the potential of a gain or a loss avoidance. Second, the paper introduces a possibility of a concave performance-flow relationship. The idea that bond fund investors react more sharply to down than to up markets. So, what's the evidence on the paper's main hypothesis?

I thought I would show you a little bit of data. This paper asks whether individual funds that have relatively worse performance than peer funds have relatively larger outflows than their peer funds. To illustrate, here is some data, fund-by-fund data for December 2015 for high-yield funds. December flows to individual funds are plotted against December returns for these funds. High-yield funds I'm picking because commentators have suggested that they're toward more the illiquid end of the spectrum, and therefore ought to be the kinds of funds that Itay is talking about in his paper.

December catches two market events. First, we had the Third Avenue high-yield episode, and we had the first Fed tightening in several years. So if you fit a line through the data, the dashed line, what you see is there's definitely a positive relationship. Investors do redeem from funds that underperform their peers. But do investors react more strongly to bad relative performance than to good relative performance—what the paper calls a concave performance-flow relationship?

If you allow the line running through the data to bend up or down, it likes to bend down a little bit, perhaps suggesting that investors do, in fact, redeem more heavily from individual funds that perform poorly relative to their peers. In other words, these fund-by-fund data are not inconsistent with the main findings in the paper, but does this have any relevance for bond market liquidity? It's unclear. For example, in December, some funds had rather strong outflows, but others, although having negative returns had positive inflows, so those guys have negative returns, pretty substantial, but they have positive inflows. Why is that important? It could be that money is simply flowing out of more underperforming funds into less underperforming funds, on net balancing out, and having little if any effect on the high-yield market. So presumably, what's most important for bond market liquidity is how aggregate flows to bond funds react to bond market returns. Consequently, it helps to sum up the fund-by-fund flows to get to aggregate flows across all such funds on, say, a monthly basis.

So what happens when you do that? When you do that you get a rather different picture. This slide looks at how aggregate flows to high-yield bond funds, the vertical axis respond to Morningstar high-yield category returns, the horizontal axis on a month-by-month basis. Each dot represents one month of aggregate flows to high-yield funds for the period 2008 to early 2016. Aggregate flows are scaled by aggregate fund assets. As before, if you run a line through the data you see a positive relationship. Fund shareholders tend to redeem, in aggregate, when the high-yield market declines. Presumably, if the hypotheses in the paper are correct, if we allow the line in this chart to bend, it should bend downward, the same as with the fund-by-fund data. It doesn't do that.

The line tends to bend upwards, suggesting that when returns get really bad, in aggregate, investors just sit there. So, you may be thinking to yourself, but what you need to do is throw out that outlier that's at -15 percent, that guy right there. The problem is that dot is key, that dot is October 2008. So if the first-mover hypothesis is true, you would suspect that this ought to be the perfect case where this hypothesis would play out. The fact that flows were virtually nonexistent, and you had massive negative returns that month, poses a pretty significant challenge I think for the first-mover hypothesis. And so in case you're wondering, is this a lead-lag relationship? Is this a lead-lag issue? The answer is no. You see that the other two months, November and September 2008, also this effect is going on. Very negative returns, very modest outflows, if at all.

So do bond fund investors, in aggregate, redeem massively? That's the issue it seems to me that's important for bond market liquidity. I think the evidence from aggregate fund flows is pretty consistent. Investors in all kinds of long-term funds, over long stretches of history, appear to redeem only modestly even in the face of severe economic shocks. So here's an example for corporate bond funds. The dashed white lines indicate, roughly speaking, the funds within a given month had the highest inflows and the greatest outflows relative to their assets. In any given month, some funds see inflows and others see outflows. For example, December 2015, funds are seeing inflows. The center yellow line adds up the flows in and out of all such funds in a given month. What you see is a fair bit of flow variation in the vertical direction, that's in the cross-sectional direction across funds at any point in time. But that tends to balance out, in a given month, so that flows in this direction, in aggregate, are very modest month to month. For example, in May and June 2013, the "taper tantrum" right there, outflows from corporate bond funds total just 2.6 percent of their assets, and that was the biggest monthly outflow since 2000.

So why are bond fund flows so modest despite significant shocks? Well, there are probably a number of reasons. One is that long-term mutual funds are held almost entirely by households, not institutional investors who trade on every wiggle in the market. As the slide shows, 95 percent of the assets in long-term mutual funds are held by households. These are investors who are not staring at their Bloomberg screens. In fact, unlike many of you, they don't even have Bloomberg screens. These are people that have day jobs that demand their attention and bosses who also demand their attention.

In addition, households, generally speaking, are saving for long-term goals. The right-hand pie chart shows that over half of the assets in long-term mutual funds are in retirement accounts. This means that a large fraction of fund shareholders have very long investment horizons, perhaps 10 to 40 years. That gives fund shareholders the ability to look through market cycles, which Donald Morgan of the New York Fed pointed out in 1994, in the same case the Fed volume review where Henry Kaufman raised his concerns.

There are likely other factors at work as well. Your fund may have a redemption fee. Also, most funds and some 401(k) plans have restrictions that limit trading, for example, the Federal Reserve's thrift plan has a 14-day round-tripping restriction. ICI's 401(k) plan has a 30-day round-tripping restriction. Taxes, if you redeem shares outside of a tax advantage, tax-deferred account, you may incur a current tax liability. That may make you think twice about redeeming.

This is something I've clipped from the website of the Fed's thrift plan. It's another reason that fund shareholders may not trade all that much. Financial market experts tell them not to. They tell them to ignore short-term trends and focus on the long-term goals.

So finally, how strong is the evidence that fund redemptions move bond prices meaningfully? It'd be surprising if fund sales of portfolio securities had no market impact whatsoever. But suggestions that there might be large and destabilizing effects from fund outflows seem overstated. One reason is that funds are not the only players in the bond market. And by many measures they aren't a particularly large player. For example, as this slide shows, although bond fund assets have grown substantially in the past several years, corporate bond funds still hold less than 10 percent of the U.S. corporate bonds outstanding. Also, depending on what bond fund category you look at, fund share of trading volume is generally as low or even lower than their share of the relevant bond market.

Consequently, it'd be surprising if the relatively modest outflows that are likely to arise from corporate bond funds from even very large market shocks were to create destabilizing effects. In fact, the evidence of strong spillovers from bond fund flows to bond prices or yields, in aggregate, has been pretty elusive, I would say. This paper does some exploratory work on whether outflows from corporate bond funds create meaningful spillover effects on corporate bond returns or credit spreads. Itay didn't really talk about this evidence so I don't want to belabor it, but I would say when you read the paper, recognize, he does at the back do some exploratory work using vector autoregressions. There's a fundamental chicken-and-egg problem with those vector autoregressions—you don't know whether funds are causing flows, or flows are causing returns, and I can tell you in standard econometric packages, those packages make an underlying assumption about which way, whether the chicken came out of the egg or vice versa. If you change the ordering, and all that is, is a matter of a little bit of typing, you get a completely different result. Bear that in mind when you read those results.

I just want to say that again, Itay didn't belabor this so I don't want to either, but if you look at the results they do have a bit of results in their paper where they suggest there could be meaningful effects of outflows from bond funds on to corporate bond spreads or corporate yields. He will tell you that that applies only to individual funds. If that's the case, that's great. If people read that, though, as being an indication that outflows are causing significant effects in the bond market in general, my reading of the paper is that those effects are pretty small.

I just want to offer one last thing before I close. This might perhaps be slightly controversial, but I'd like to suggest reconsidering the research agenda in this area. Perhaps the most enduring and salient fact about long-term fund investors that shows up in the data is that they just trade very little even in the face of significant market shocks. One might even ask whether they trade too little. So the research challenge, it seems to me, should be to build models that explain this. One possibility might be to start with some of the recent models of optimal inattention where investors don't trade much, for example, because they have day jobs. If we can build such models we might be able to better gauge whether long-term funds pose risks. For example, if it turns out that life-cycle savings behavior is a key factor explaining this result, that seems to me that that's unlikely to change soon or quickly. So, I'll stop there.

Goldstein: So, thank you for your comments. I didn't expect them to be nicer, given the response of the industry. Actually, I don't think we disagree on that much. I think you disagree more with the response to the paper than with the paper itself. Let me just point out a few things. When you look at slides 7/8 and you say that in the aggregate the relationship becomes concave, there are two things I would like to highlight. First of all, this is not the hypothesis that we have. Our hypotheses are all at the level of the individual fund and this is where we did most of the analysis. We have one graph in the paper that I showed toward the end that looked at the aggregate, but this is really not where we do all the tests. The other thing I would like to highlight here is, we do show that in the aggregate it is still concave when you look at the whole sample, so you focus on part of the sample, particular funds, and a particular period. As a result, I am less sure what exactly we want to take out of it.

I have comments regarding the VAR analysis but I think that the way you have done it is essentially misinterpreting the way that we did it, but this is something that we don't need to bore the audience with. The main thing I would like to highlight is at the end of the day, I don't think we have a strong disagreement. It's fairly clear that we did not have a big run on mutual funds. As I said, if we do have it, I think it will happen once and then the system will be immediately changed. As a result, what we are trying to do here is detect some potential underlying failures in the resilience of the system, and this is why all of the tests that we do look at this effect of liquidity on outflows and the sensitivity of outflows to performance, and I think when you take all of the tests together they do show that there is some first-mover advantage at the level of the fund, not in the market as a whole. This is something I think people should be aware of and pay attention to, both in the industry and more generally.

Paula Tkac: OK, so let's turn to audience questions. I'm going to group some of these together. The first one I want to talk about is actually related to a comment I had initially.

If we look back at the fall of 2008 and we think about what happened with money market mutual funds, it was large outflows from institutional funds that were moving from prime to government, and really the retail outflows were not significant. Part of your analysis contrasts institutional flows or institutional funds in retail, you find something a little bit different. So here is a question that gets to the heart of that. The effect of illiquidity on sensitivity to performance is greater for institutional funds, you find, but is the level of sensitivity greater for retail or institutional funds? So maybe you could talk a little bit about the breakdown that you do, and maybe Sean, you could whatever insight you have on the open-end fund sector and how we want to think about institutional versus retail investors.

Goldstein: I think what Sean is going to say is that the way we measure institutional versus retail investors is off, and I agree that there are problems with it. They are the same problems that apply for the money market funds. But at the end of the day, I do think that the measure that we have captures some difference in the nature of these funds. I think underlying this there are two things going on. One thing that is going on, which Sean mentioned in his discussion, is the fact that institutions may be more tuned to the market. The other thing going on is this force of coordination and externalities I believe applies less to them because they are bigger. Those two go in opposite directions. The best that I can say is in the events of 2008 with the money market funds, one effect dominated the other, and in our data set the other effect dominates. That's my interpretation.

Tkac: Sean, do you have anything to add to that?

Collins: Yeah. I'm not really sure how to answer the question. Part of the reason is I had the pie charts up there. The pie charts say virtually all of the assets in long-term mutual funds are retail money. There's almost no institutional money, what you would think of as institutional money: hedge funds, DB plans, sovereign wealth funds, there's a little bit. What's in most of institutional share classes in mutual funds is household money that's there by virtue of 401(k) or IRA accounts. They get institutional pricing because they come to a fund through omnibus accounts, which are bundled. They may come through a broker dealer where there are bundled trades as well. That's in institutional share classes. So again, it's hard for me to know how to answer that question.

Tkac: OK, so the next set of questions follows off on that and our discussions this morning. So we talk about market liquidity and price impact and clearly that's at the heart of your paper, Itay, and so this question asks, why wouldn't the effective trades on bond prices be transitory especially for retail investors? It kind of picks up on Pete's [Kyle] question this morning, about wouldn't you prefer your open-end funds to have low liquidity, certainly you're not limiting your results and your return results by holding a lot of cash. And so, how do we think about this from the investor perspective? And whether or not what you're talking about with respect to fragility really is a cost that the vast majority of open-end fund investors would care about, and then kind of how do you weigh that against what may be the magnitude of this first-mover advantage could possibly be?

Goldstein: That is absolutely a good point. Now as I pointed out in my response to Sean, we did not conduct analysis on the effect of flows on prices and how persistent it is, but these studies have been conducted in the context of equity. I mean, there's one study by Coval and Stafford, for example, that I think people who work in academia on mutual funds they know very well, and they have shown that there is price impact and the price impact is long lasting and it's pretty significant. Many papers have followed up on this and showed the same thing in the space of equity.

So, my conjecture is you will be able to find similar price impacts also in the context of bonds, especially given that the markets are so much more illiquid, pricing uncertainties and stuff like that, but as far as I know this study has not been conducted. I think that is a very important thing to study. We know there are many market frictions that generate this kind of persistence. It won't be new to the literature to find something like that.

Harris: I might say I'd like to answer that question with a question. This is a question I'd love to ask people like Pete Kyle and Larry Harris if I had time. Perhaps afterwards.

Tkac: Let's go to one now that's gotten a lot of votes and one that actually came up earlier today. Why do you think, let's assume there's a first-mover advantage and that this is a source of fragility. Let's take all that as a given. Why are funds set up this way? Why not offer some products that have monthly or quarterly liquidity as opposed to daily liquidity?

Goldstein: Right, so first of all, I do think many people do think that this idea of having an open-end fund where you can take your money out every day and then on the other side having very illiquid assets is not a very good idea. I think everyone in the room will agree there has to be some limit to it. I know that in Germany, for example, we used to have open-end funds that invested in real estate and that caused a big run, and they backed out from it. They don't do it anymore. Clearly, at some point the liquidity mismatch between the two has to be addressed. I don't think that's a very controversial thing to say. When you are holding very illiquid assets and people can take their money out every day, it is potentially a source for a problem. That is something that we need to think about.

Having said that, there is some role for liquidity creation. Banks exist for many years because they provide liquidity and even though they create some risk for a run, people are willing to live with it because of the useful role of it. Same is true for mutual funds. There is some benefit for liquidity creation but you always have to assess the benefit against the cost. As I mentioned, I think funds can do many things in order to address the issues and I've talked to many people who work in it. I think they all agree that this mismatch is there and they worry about it, they think about it, and they constantly think about ways to reduce the fragility. The question is, whether they need to do more.

One last thing about the ETFs [exchange-traded funds]. ETFs are very interesting here. The way they pay the investors is different in the sense that when there is a redemption, they will immediately have to sort of sell the assets and create new securities and so on and so forth. In some sense, they kind of bring it to market all the time and maybe this reduces the first-mover advantage to some extent, but I haven't completely thought about it.

Tkac: OK, so Sean, why don't we have monthly liquidity?

Collins: I think this is probably one area where Itay and I agree almost completely. I'm not convinced there's run risks, systemic risks in these kinds of funds. Put that aside. If you thought that there was a problem, there are things you can do, and Itay mentions a lot of those things in the paper. You could use bid pricing for bond funds. A lot of people do that. You can have redemption fees, swing pricing is a suggestion that's been out there in the SEC's liquidity proposal. That's a possibility as well. As I said, a lot of funds do have frequent trading restrictions and they monitor for frequent trading.

Why don't we have lockups? I think the answer is we used to have them. There used to be these things called interval funds. People don't like them. Part of the reason is it's really hard to make them work with 401(k) plans.

Tkac: Why is that?

Harris: 401(k) plans work by typically having scheduled purchases that come every couple of weeks. You might have somebody that says, "I need to roll my IRA/401(k) because I'm leaving now." Do you want to put them in the position of saying, "Well, no, we know you're getting fired from your job, but please wait two months before you can roll your account." The company may want to roll them out of the account right now. So I think generally what our members would say is, if we felt like there were things we would need to do, that wouldn't be our first port of call. There would be other things.

Tkac: We're going to do one last question, and I'll try to give you plenty of time for discussion afterwards. This one actually again ties back to what we were talking about this morning. I think I know what you are likely to say, but, what if we actually had better pricing and a more liquid transparent bond market? It seems to me that would start to ameliorate some of the issues that you're raising.

Goldstein: It's all a matter of backward-looking pricing rather than forward-looking pricing. To the extent that you can have more liquidity and more frequent pricing, you can alleviate this problem. If I take my money out of the mutual fund today and I get the fair price, and I don't impose any externalities on those behind me in the line, then there is no problem. Then it's just a market. The only problem is when these assets are illiquid and there is uncertainty about pricing and then there is this potential externality. When you think about swing pricing it's kind of a way to make pricing better by recognizing that when many people take money out on a given day, there will be downward pressure on the price so we want to account for it immediately. That's what they are trying to do. So I think definitely, having better models and better ways to come up with prices every day, that's a very important direction to go in.

Tkac: Or in the extreme to lessen price impact from these bond trades. It's a nice example to me of regulation on one side could help out another side. Maybe we wouldn't need so much swing pricing if you, in fact, had better pricing and better liquidity.

And with that, I want to thank my panel presenters here and we'll move on to the next session.