Full transcript
0:06thanks for having me um this paper it
0:08it's I'm really excited to have it out
0:10I've been struggling with this work for
0:12I've I've tried to write this paper
0:13twice before in the last 10 years um and
0:17um the the current version which I'm
0:19very happy with my co-authors are all at
0:21dimensional where I've been involved for
0:22quite a while um it's about reversals
0:26and the return to liquidity provision um
0:28these things are are quite closely
0:30related I think um there's this fact
0:32that you know I we probably all know
0:33that if you look at shortterm
0:36performance one month out of stocks that
0:38have gone up over the last month they
0:39tend to underperform slightly those that
0:41have gone down over the last month um
0:43and it's fairly weak in the data um it's
0:46not very strong outside of micro caps
0:48and it's gotten a lot weaker post
0:50decimalization um so those are the
0:51reversals I'm talking about um they're
0:55related to uh liquidity provision and I
0:57think there's a really nice paper of
0:59Stephan Nel from 2012 that talks about
1:02how you can think about these as the the
1:04you know the reversal performance as
1:06aoxy for liquidity provision and the
1:08intuition the way you should think about
1:10that is that you go to the market
1:12demanding liquidity you want to sell a
1:15stock well there's someone who's going
1:17to come in and buy on the other side and
1:19and sometimes that person you're having
1:20to encourage to come in they're
1:22providing that liquidity and they expect
1:23to get compensated and as you're selling
1:25the stock they're partly being
1:27compensated by as you take that
1:29liquidity out of the market with your
1:31your selling you're pushing prices down
1:34um the buyer of the stock who's
1:36providing that liquidity for you expects
1:38part of their compensation to come from
1:39the fact that though you've pushed
1:41prices down as liquidity comes back to
1:43the market prices are going to recover a
1:45little bit and so they're going to be
1:46able to sell that stock on average at a
1:48slightly higher stock price than than
1:50they bought from you um so so that's the
1:53basic intuition Neel provides this this
1:57a picture kind of like this the the
1:58vertical line there shows the the the
2:01sample up to the end of Neel sample and
2:03then on the the right side it shows
2:04what's happened after that um but it's
2:07basically correlating the performance
2:08the profitability of these reversal
2:11strategies um with with vix so with with
2:14the level of Market volatility um which
2:16we know volatility is is um correlated
2:19both both in the time series and in the
2:21crosssection um with the cost of trading
2:24and so he provides this um sort of
2:26strong time series evidence that um
2:29reversals more profitable when it's
2:31expensive to trade and he links these
2:33things to to uh you know tries to
2:35directly link them into the reversals to
2:37liquidity
2:38provision um in terms of what I'm doing
2:42I think of this paper as sort of being
2:44what are the cross-sectional
2:46implications of that idea that the
2:48profitability of reversals is related to
2:51liquidity provision so Negal provides
2:52evidence over in the time series um but
2:55there's a lot of cross-sectional
2:57implications um it's not just about an
2:59agregate phenomena we think that if
3:01liquidity matters for the profitability
3:03of of of making markets um that we
3:05should see differences in reversals
3:08across strategies constructed using
3:10stocks with um different liquidities um
3:14and this isn't going to be just about
3:15magnitudes stepan's paper is all about
3:17magnitudes um I I have results about
3:20magnitudes as well but I think that some
3:21of the most interesting results are
3:22really about persistence um how long do
3:25these reversals last um and I think that
3:28what we're going to see in the data
3:29actually relates um strongly to some of
3:32Pete Kyle's work um on microstructure
3:35and variance and in particular on
3:36business time um the things just run at
3:39different rate calendar time might not
3:40be the appropriate way to look at um how
3:43fast some of these liquidity phenomena
3:45occur all right so in general um when I
3:48write a paper even when I read a paper
3:50I'm always encouraging my students when
3:52they're working on papers to think about
3:54what is it people should remember your
3:56paper for there's so many papers out
3:58there that if you ask someone about it
4:00they'll tell you something and it might
4:02not even be what the authors intended
4:03but generally we don't remember papers
4:05for whole papers we remember them for
4:07one thing um and so I I'm going to try
4:09and give you that up front and this is
4:10what I think you should remember about
4:11the paper if you only remember one thing
4:14okay so this is showing the performance
4:16of some reversal strategies and I'll
4:18I'll get to the details later but sort
4:20of big picture um the performance of
4:22these reversal strategies I'm showing
4:23winner minus loser spreads so you really
4:25think of this as you can think about
4:26this as negative momentum that's why
4:28they're going down EV eventually these
4:30would turn around um and I'm showing
4:32this for strategies constructed within
4:36stocks that are very different on some
4:38aspect of liquidity um and in particular
4:41in the left panel I'm showing you
4:42reversals among low and high volatility
4:45stocks and on the right hand panel I'm
4:47showing you differences between uh low
4:49and high turnover stocks okay so what
4:52you see on the left with the volatility
4:54is that there are massive differences in
4:57the performance or the you know how big
4:59these reversals are um across uh high
5:02and low volatility stocks really out for
5:05one and a half to two weeks okay so in
5:08in these pictures you can think about
5:09the portfolio is being formed at day
5:11Zero and then you're just holding it out
5:13over time um and I'm basically showing
5:16you three months of of of uh of trading
5:18days after that so what you see is in
5:21the first two weeks the high volatility
5:24reversals are much bigger than the low
5:26volatility reversals but there's some
5:28convergence after that
5:30um and on the right hand side it's very
5:33different there it's by low and high
5:34turnover and what you see is almost no
5:37difference in performance over the first
5:40again one and a half or two weeks but a
5:42very sharp Divergence after that um with
5:45the reversal completely done after less
5:49than two weeks for the high turnover
5:50stocks but for the low turnover stocks
5:53the reversal actually persist out for
5:55three full months which is a very long
5:57Horizon relative to how we typically
5:59think about um kind of reversal
6:03performance all right so um you know I'm
6:07writing about trying toink these things
6:09to liquidity and I I've shown you
6:11volatility and turnover those are two of
6:12the big variables I'm going to use uh
6:14I'm also going to be looking at size um
6:17why do I choose these variables uh well
6:20I think that just you know by casual
6:23introspection these should all be
6:25related to to to liquidity size small
6:28stocks we we we kind of all think is
6:29less liquid there's less attention and
6:31less Market making there um High
6:33volatility we know that volatility is
6:35correlated with the cost of trading and
6:37so should be correlated with something
6:39having to do with with uh liquidity
6:42provision and transaction costs um you
6:44can also think about it as driving
6:46inventory risk if there are market
6:47makers who are taking positions in
6:49individual stocks and exposing
6:51themselves to large levels of of risk
6:53that even though it's idiosyncratic in
6:55the technical sense they can't diversify
6:57away um how big the risk of the position
7:00is isn't just how many dollarss of the
7:01stock they hold but it's related to how
7:03volatile the stock is um and then
7:05turnover should matter as well turnover
7:08um again we if we think about trading
7:10time or turnover business time um the
7:14stocks that trade less kind of they
7:17things run slower um I tend to think
7:19about this in terms of inventory
7:21durations it's harder to if if you
7:24develop a position making markets it's
7:25harder for you to um to to work that
7:27position off your books if there less
7:30turnover for you to to to be working
7:32that um the other reason I've chosen
7:34these three variables really is again
7:36due to Pete Kyle um the the simplest
7:40sort of um empirical easy
7:43implementations of the Kyle model the
7:46thing that's most common in the
7:47literature is the amood measure which is
7:49um it's it's not a fully satisfying
7:51estimation of of of the KY Lambda but
7:54it's it's a pretty good approximation
7:56that's easily estim you can estimate it
7:59using very reibly daily data on on
8:02returns and and volumes um and so it's
8:04very commonly used and it turns out that
8:07these three variables um explain on
8:09average 96% of the cross-sectional
8:11variation in the amih hood measure um
8:14and and Pete's more recent work on on uh
8:17micr structure and variance also has
8:19suggested an entire class of U
8:22transaction cost estimators which
8:24basically are based on these three same
8:27variables all right so um
8:30the the main facts that are going to
8:31come out of this analysis is that
8:33reversals are um stronger and small
8:36stocks but I think the kind of
8:37surprising thing is how concentrated
8:39this is in really in in like the bottom
8:413% of the market by capitalization um
8:44even they don't look dramatically
8:46different between midcaps and large caps
8:49you do see stronger reversals in micro
8:51caps but it's really among the very
8:53small stocks uh we're going to see
8:55stronger um reversals among High
8:58volatility stocks um that's true even
9:01after controlling for differences in
9:03pre-formation spreads more volatile
9:05stocks tend to have more dramatic you
9:08know recent performance but even
9:10controlling for the size of differences
9:12in the pre-formation spreads uh past
9:15performance between winners and losers
9:16you see that the reversals are bigger
9:17for the higher volatility stocks um and
9:20then I think the most striking feature
9:22of the paper is just how persistent the
9:24reversals are um for the for the low
9:26turnover
9:28stocks all right so uh one other thing
9:30I'm going to do and it's not really the
9:33point of the paper but um we we are
9:36going to use a reversal refinement for a
9:38lot of our empirical tests um we're
9:40thinking about reversals as a lens to
9:42study liquidity um but sort of the
9:45literature basically suggests that the
9:48reversals come from Price movements that
9:51are unrelated to news okay and there's
9:54some obvious sources of news in past
9:57performance that we don't expect to get
10:00reversed that dramatically obscure how
10:03big these reversals are in the data the
10:06reversals look pretty weak and
10:08especially in the last 20 years but I
10:09think if you uh take out some of that uh
10:12news related returns and prices they
10:14just look dramatically stronger so the
10:16obvious things are that news when you
10:18buy a winner which or or if if you're
10:20going to trade a reversal and you sell a
10:21you buy a loser um the loser um we
10:25expect that the the loss to be somewhat
10:28reversed the next month in expectation
10:30but if it's a loser because it announced
10:32really poor earnings last month we don't
10:35expect that to get reversed if it's a
10:37loser because the industry it's in is
10:39down but it's outperformed its industry
10:41we don't expect that to be reversed
10:42either um and so you can do sort of a
10:44decomposition where you take out the
10:46post earnings announcement drift and the
10:48short run industry momentum from uh
10:50reversals and the reversals just look
10:52dramatically stronger um so here um the
10:55left hand side of the first panel rev is
10:58basically saying that the unconditional
11:00reversal strategy buy the stocks that
11:02were down the most last month sell the
11:03ones that were up the most last month
11:06has on average earn 31 basis points per
11:08month over the last 50 years um so a few
11:12percent per year uh it's not really
11:14highly statistically significant uh this
11:17has come down in the last 20 years so
11:19it's weakened these results um but if
11:22you adjust past performance for earnings
11:24announcements and for industry
11:26performance you get the right hand side
11:28that irx is industry relative reversals
11:30excluding earnings announcement returns
11:33um and there you see a a return of
11:36almost 110 basis points per month um and
11:39it's actually only got about half the
11:40volatility of a standard reversal
11:42strategy um and so it's got a sharp
11:44ratio that's roughly six times as high
11:47and if you do a decomposition regressing
11:49the returns of a standard reversal
11:51strategy onto these sort of what I think
11:54of as liquidity driven reversals and
11:56Industry momentum and post earnings
11:58announcement drift what you see is that
12:01the reversal is loading heavily on what
12:04I think of is liquidity driven reversals
12:06but it's taking very large short
12:08positions in post earnings announcement
12:10drift and Industry momentum and those
12:12headwinds are really obscuring how big
12:15the reversals actually are okay none of
12:18the the results I'm showing you really
12:20depend on their choice of using these
12:22industry relative uh earnings
12:24announcement adjusted reversals um but
12:27it just gives you a better length the
12:29the the magnitudes are bigger and it
12:30gives you sort of cleaner results on how
12:32to look at these
12:34things um all right so I've already
12:37shown you some of these pictures but but
12:39really a lot of what the paper is doing
12:41is forming portfolios and looking at how
12:44they perform over time um and um again
12:47we're going to be using size volatility
12:49and turnover as our sort of the the
12:52differences we want in stocks when we're
12:54constructing these reversal strategies
12:56uh and I'm going to show you results
12:58using different path performance Windows
13:01um so a lot of the literature has looked
13:03at monthly past performance some looks
13:05at shorter stuff I'm going to offer you
13:08uh reversal strategies based on stocks
13:10that are winners or losers over the
13:12previous one day one week or one month
13:15um there are advantages and
13:16disadvantages to each of these so the I
13:18for me the cleanest results in terms of
13:20what's going on come from only looking
13:22at one day of past performance um when
13:25you go out to 21 days you get smoother
13:27results the statistics are some somewhat
13:29better um but I'm looking at phenomena
13:33where reversals sometimes only last 5
13:35days and so if you're using 21 days of
13:38of look back to to look at past
13:40performance on something that only lasts
13:42a week it's really not very informative
13:45and it gives you some misleading results
13:47so so using more time in past
13:48performance gives you kind of smoother
13:50results but they're harder to interpret
13:52what's going on okay so first this is
13:56size here I'm looking within uh nysse
14:00quintiles based on market capitalization
14:03um and the the left is the one day the
14:06the right is the 21 days the middle is
14:07one week um so what you see here is that
14:11there's a dramatic difference between
14:13the magnitude of these reversals between
14:15the smallest NYSC quintile um which is
14:19really only about 3% of the market and
14:22the rest of the market there's really
14:23not big differences in how big these
14:26reversals are um outside of of
14:29you know the the real micro caps um a
14:33lot of this is driven by very big
14:34one-day effects um 60 basis points at a
14:38one- day Horizon among the these tiny
14:41stocks um when you look at
14:43volatility um I I guess I'll focus on
14:46the middle here it looks similar to the
14:47left but but a little more dramatic uh
14:49you see this kind of clear um increasing
14:54size of the short-term magnitude of
14:56these reversals as you go up the
14:58volatility quintile
15:00um but there's some convergence after
15:02several months um things start to get
15:06confusing when you get out to 21 days
15:08this is this is why I'm reluctant to do
15:09the 21 days for these it's because the
15:12the Cardinal ordering of where the
15:14reversals seem biggest switches when you
15:16get out to 21 days for volatility um and
15:19that's really because here I'm not
15:22controlling for correlations across the
15:24different liquidity variables I'm using
15:26I'll control for those in a second um
15:29but here it's hard to really interpret
15:32the right-hand panel because the high
15:34volatility stocks which I think of is
15:36having the bigger reversals we see that
15:37the green line in the middle panel those
15:40High volatility stocks also tend to turn
15:42over more and so even though the
15:44reversal that the the magnitude of that
15:46reversal is big from what's happened
15:49recently when you get out to Long
15:51Horizons uh the fact that these stocks
15:53turn over more means the reversals are
15:55less persistent and you're already
15:56starting to kind of when you when you
15:59look back that far you're not really
16:00getting a good reversal signal for the
16:03the high volatility stocks because they
16:04they tend to turn over quite a
16:06bit um and then finally there's the
16:09turnover plots here um again for the I
16:12think the middle panel is the easiest to
16:14interpret here you see basically across
16:17all five turnover quintiles things look
16:19identical for a week and a half or two
16:21in trading time um but then you get
16:23these this these divergences uh in in
16:26how persistent the reversals are with
16:28the lower term turnover stocks having
16:30more persistent reversals um and then um
16:34here the right hand panel is very
16:35dramatic and it um it's easier to
16:37interpret here here I'm using a 21-day
16:40signal well for the for the low turnover
16:43stocks where the reversal are very
16:45persistent not only are they reversals
16:48persisting and lasting longer but that
16:49past look back period is more
16:52informative the returns that happen 21
16:55days out still predict reversals for um
16:58for the the low turnover stocks and so
17:00that that increases the magnitudes there
17:02whereas for the high turnover
17:04stocks what was happening the first half
17:07of the previous month is just noise in
17:09sort of predicting the reversal for the
17:10the high turnover stocks because there
17:12the signal is very transient um and so
17:14the reversals get very
17:17weak um again I said that really we need
17:20to be careful because we shouldn't be
17:22looking at these things um sort of in a
17:25univariate sense when they're highly
17:27correlated the small stocks tend to be
17:30more volatile but they tend to trade
17:32less the high volatility and and high
17:34turnover stocks those two things are
17:37positively correlated with each other um
17:39and so um we have a bunch of tests which
17:41are a little more complicated that
17:43essentially attempt to control for two
17:46of the liquidity variables while uh
17:49getting dispersion in the third one
17:51before you construct your reversal
17:53strategies and I'm not going to go
17:55through all this I'm just going to give
17:56you one example of that so this is
17:59showing you within a size universe so
18:01we've controlled for size we're looking
18:03at all stocks of similar size we're on
18:06the left picking stocks that have almost
18:10identical
18:11turnover but have differences in uh in
18:15volatility so so there there that left
18:17hand panel is really trying to control
18:19we've got the same size the same
18:21turnover but differences in volatility
18:23how does that affect reversals and what
18:25you see is that you have these big
18:27differences again really over only over
18:29the first two weeks and then once we've
18:32controlled for turnover there aren't
18:33differences in persistence and so you
18:35get this sort of parallel performance
18:37outside of that twoe window where the
18:39volatility really
18:40matters on the right hand side socks of
18:43the same size same volatilities but
18:47differences in turnover um and there you
18:49see uh really you know almost identical
18:53performance for the first couple weeks
18:55um but then you get this Divergence um
18:57with the the reversal ending uh much
19:00sooner for the high turnover stocks and
19:02and continuing for the lower turnover
19:05stocks um for me as an academic one of
19:09the really interesting things about
19:10these results was that there were a
19:13bunch of results in the literature just
19:15sort of weird facts about reversals and
19:18momentum that were known um that can be
19:22understood simply in the context of the
19:24results I'm presenting here um and a
19:27couple of those results are stuff that
19:28that was in the literature because of me
19:30and so I was particularly interested
19:31about those um one of them which is
19:34obviously related to this work comes
19:36from my co-author um Mom dumit hat um so
19:39he has this RFS from a couple years ago
19:42um basically documenting that even
19:44though we see unconditionally short-term
19:47reversals that's the middle bar the
19:49negative performance if you Constructor
19:50reversals using all stocks um the point
19:53of his paper was really that there were
19:55some stocks where you should should not
19:57expect reversals you should expect
19:58momentum
19:59and it was the high turnover stocks um
20:02and I think that in the context of this
20:04and thinking about things in turnover
20:05business time it makes perfect sense um
20:08here I'm showing you the le- hand side
20:10of this is basically showing you the
20:12medhattan schmelling results but not in
20:15average performance over the next month
20:16but in what happens from portfolio
20:19formation on average um so in that
20:21left-hand panel we're choosing uh we're
20:25forming reversal strategies on the basis
20:26of Prior month's performance and we're
20:27looking at what happens over time and
20:29that vertical bar is the one month
20:31Horizon at which they measure their
20:33performance and they call the the Blue
20:35Line momentum and the yellow line strong
20:38reversals uh it's not really that
20:40there's no reversals in that there's
20:42instantly momentum in these high
20:44turnover stocks it's just that you're
20:46looking at a horizon you know one month
20:50past performance Windows a very long
20:51time to look back for a high turnover
20:53stock and measuring performance a month
20:56later is sort of a very long time to go
20:57forward to measure it and you're already
20:58into the momentum uh region for those
21:01High turnover stocks where for the low
21:02turnover stocks where the reversals are
21:04persistent uh you see much stronger uh
21:07reversals um I have a similar a sort of
21:10similar result from a transaction cost
21:12paper um there just this very weird fact
21:15or it seemed weird that uh industry
21:19relative reversals which are stronger
21:20than reversals look much stronger among
21:24low volatility stocks and these are all
21:27monthly strategies um this fact has been
21:30picked up on it was um there's a one of
21:33the early machine learning papers this
21:35kak Negal and Santos um finds that this
21:37industry relative reversals constructed
21:40among low volatility stocks is the
21:42single most important component of a
21:45stochastic discount Factor estimated by
21:47their machine learning techniques um
21:49there's some other people have picked up
21:51on this um and um so I was interested in
21:55looking at that and um here again the
21:58left hand here is showing that result
21:59it's showing industry relative reversals
22:01based on a month of past performance
22:03from portfolio formation and the blue
22:05line is the low volatility stocks the
22:07red line is the high volatility stocks
22:09and a month out the the low volatility
22:12reversals are much bigger than the the
22:14high volatility reversals but again
22:16there's this correlation the high
22:17volatility stocks tend to turn over more
22:20and so they have much less persistent
22:22reversals and if you go back to only a
22:24one we look back signal the right hand
22:27panel there you actually see much
22:29stronger industry relative reversals
22:32among the high volatility stocks than
22:33among the low volatility stocks the the
22:36problem is we were doing this simple
22:38thing of looking at monthly look backs
22:40and monthly look forwards we weren't
22:41paying enough attention to the frequency
22:44with which these phenomena actually
22:45operate and if you don't look at things
22:47at the right frequency you can get very
22:49misleading
22:50results um there's also kind of looking
22:55longer term this has relations to um
22:58results about momentum uh there papers
23:01from 10 years ago documenting that
23:03momentum's stronger among High
23:04volatility stocks um I also have some
23:07work saying that if you want to look at
23:09big momentum performance it's not based
23:11on what happened over the last six
23:12months it's based on what happened over
23:13the six months before
23:15that um if you look longer Horizon so
23:18this is showing the performance of
23:20winner minus loser strategies based on a
23:23just one month of past performance but
23:25holding them out fixing the portfolio
23:27and holding out to a year okay and what
23:30you see is among the high volatility
23:33stocks the green there that momentum
23:35sets in very
23:37quickly the the reversal ends quickly
23:39and momentum sets in for the very low
23:42volatility stocks moment it's shocking
23:45but momentum doesn't sort of kick in for
23:48seven months okay and this picture
23:51explains both the results you see the
23:54the high volatility stocks they have
23:56stronger momentum because momentum kicks
23:59in immediately whereas it takes a long
24:00time to set in for the low volatility
24:02stocks um if you're using past
24:05performance well when momentum doesn't
24:07set in for six months with some stocks
24:11you should expect that the longer
24:13Horizon past performance the stuff from
24:14before six months ago is going to matter
24:16more unconditionally okay so this
24:18explains both those results in the
24:19literature it also suggests a refinement
24:22which we can easily test if you look at
24:24this picture where does it say that
24:27momentum should differ among sort of
24:29recent history and and longer history
24:32well if you look at this it looks like
24:33momentum is relatively the same for the
24:36first six months of the next six months
24:38on the high volatility stocks but it
24:40looks much stronger out six months for
24:43the low volatility stocks from the high
24:44volatility stocks okay so the refinement
24:46says that my result on the differences
24:49in intermediate and recent Horizon past
24:52performance should be concentrated in
24:54the low volatility stocks and should be
24:56largely absent from the high volatility
24:58stocks um and if you can if if you test
25:01that um that's exactly what you see in
25:03the data um so the low column there
25:05shows you that um that there's really no
25:09even um there there's really no momentum
25:11it's it's negative momentum reversals
25:13out to six months among the low
25:14volatility stocks huge differences in
25:17performance long and or or intermediate
25:19and short Horizon past performance there
25:22but if you go out to the high volatility
25:23stocks there's really no statistical
25:25difference in the strength of momentum
25:27based on intermed immediate and longer
25:29term past performance um so
25:34um I guess there are a couple other
25:36things um I I've been working on since
25:39the first version of this paper um one
25:42thing is we also have some results uh
25:44directly related to liquidity um fairly
25:47recent results um that basically argue
25:50that um reversals are concentrated among
25:55um recent
25:56losers because there's less liquidity
25:59provision among those stocks because
26:01they're stocks that institutional
26:02investors shy away from um so so there's
26:05some evidence based on institutional
26:07Holdings that among these loser stocks
26:09um there's less institutional holding
26:11and we do see in the data um that
26:13reversals are much stronger among these
26:15loser stocks okay it turns out that um
26:20The Situation's more complicated than
26:22that because um if you look at what's
26:25driving the weakness of reversals among
26:28the recent
26:29losers it's not the component that I
26:31would identify with uh with liquidity um
26:34it's really Almost 100% driven by
26:36differences in the strength of post
26:38earnings announcement drift um which
26:41tends to be very strong for winners and
26:43weak for losers um and so the reversals
26:46are facing this uh post earnings
26:48announcement drift headwind among the
26:50the winners much more than among the
26:51losers which creates this big disparity
26:54in the strength of the reversals um
26:56among stocks that performed well or
26:58poorly over the last quarter um and then
27:02one other thing that's not in the paper
27:04but I'm I'm working on including in the
27:05next revision is is sort of what does
27:07this mean more practically for people
27:09who are actually constructing
27:12portfolios um I mean you could in theory
27:15try and you know trade this directly and
27:17be a liquidity provider um but I'm also
27:20interested in just how do we use these
27:22facts about liquidity to inform what
27:25we're doing anyways um so the the way I
27:28think about this is if you're doing any
27:31portfolio Management in in your life you
27:35are making decisions about when to buy
27:36and sell and what are these uh these
27:40signals I'm looking at really telling
27:41you well they're telling you about when
27:43liquidity is
27:44expensive when liquidity has been taken
27:46about out of the market and how long it
27:48takes you takes to come back um so it
27:51can help inform practitioners about if
27:53they want to buy something do they want
27:55to buy it right now or does it look like
27:58like um liquidity is very expensive on
28:01that stock liquidity has been taken out
28:02of the market if you come and demand
28:04liquidity in that stock right now it's
28:06going to be very expensive to trade and
28:08you could expect very short-term poor
28:10performance if you're buying at that
28:12time um and so I think about I've been
28:15thinking about how you use this
28:16information um to inform not the trades
28:19you actually make but the timing of the
28:22trades you make um should I delay this
28:24the the stock has entered the buy range
28:26for me should I delay buying it a month
28:28I want to hold it longterm but the
28:31because of these liquidity concerns and
28:33and how I can measure them based on on
28:36uh past performance metrics you might
28:38want to delay a stock you want to hold
28:40buying a stock you want to hold longterm
28:41delay buying it a month because um for
28:43liquidity reasons you expected to have
28:46uh very poor performance over um over
28:49the very short term um and and um you
28:52know that you can use information about
28:54the characteristics of the stock to try
28:56and forecast how these concerns are and
28:59how long they last um so that that's
29:02work in progress all right so in terms
29:04of of the conclusions um you know we see
29:07that liquidity I think expost it
29:09shouldn't be that surprising more less
29:11liquid stocks should cost more to trade
29:15and if if reversals are evidence of
29:17Market maker profitability we should
29:18expect small stocks to be more expensive
29:20um and the more volatile stocks to to
29:23have these bigger reversals and uh we
29:25also again the really dramatic thing I
29:28want you to remember is that you know
29:30these things last months and months in
29:32the low turnover stocks I think there's
29:34interesting further research there
29:36because I don't believe that the actual
29:39proximate market makers are it's taking
29:42three months for them to get inventory
29:44off their books when they when they have
29:46traded a stock but I think the market
29:48making process is much more complicated
29:50than academics often you know think
29:53about and I think that there's proximate
29:55market makers and then I think there's a
29:57whole range of different levels of
29:59Market making where the market makers
30:01themselves sell to some secondary
30:03liquidity provider who might hold the
30:05stock for a month but they're not going
30:06to hold it on their books a long time
30:08and there's a whole kind of diffusion of
30:10the portfolio Holdings that takes much
30:12longer Horizon uh than what you see just
30:15looking at Market maker inventory
30:17durations and I think it's it's uh a
30:19really interesting place for further
30:20study so thank
30:24you all right so good morning everyone
30:27uh first I would like to extend my
30:29gratitude to the organizers for such a
30:32wonderful opportunity to discuss Rob's
30:34paper this paper dives into a pivotal
30:37question uh the question is you know
30:40what expected returns uh investors uh do
30:44investors anticipate to pay uh in the
30:47capital market and for econom for E
30:50economists is is definitely one of the
30:52central ass pressing questions uh for
30:55investors on the other hand they they
30:58definitely also value this question
31:00because they they are very much
31:01concerned with you know the cost they're
31:03going to pay you know in order to gain
31:06liquidity I'll provide liquidity in the
31:08capital market and The Regulators of
31:10course they also you know think this
31:12question is a first order important
31:13question because they believe you know
31:16the market liquidity or the liquidity
31:18prision in the Capital Market play a
31:20vital role in maintaining Financial
31:23stability so however recently some may
31:27argue this question become less
31:29important the reason is people think you
31:32know the given this kind technological
31:34advancements and you know they they will
31:38this kind of like they will enhance this
31:39kind of Market liquidity however while
31:42you know technical advances advancements
31:45can enhance Market liquidity but
31:47actually the recent studies also found
31:50they can introduce new challenges for it
31:54one example would be the recent work
31:55myself with uh go and we show you know
31:59if we have this kind of AI power trading
32:02actually the market we we we are seeing
32:04this kind of like a new challenge for
32:06Market liquidity they may compromise
32:07Market liquidity uh additionally
32:11actually emphasized by the ICC chairman
32:13Gary gansler he emphasized the tech such
32:17Technologies actually can promote
32:20hurting behaviors among individual
32:23investors leading them you know make
32:26similar trading decisions
32:28uh driven by the same strong signals
32:32they they they gain from the you know
32:33using their technology and uh all given
32:36all this together we believe you know
32:38the value of this liquidity Pro
32:40provision and the market the level of
32:43Market IL liquidity remain significant
32:46in the Capital Market so this question
32:48asked by this paper still is a very
32:50important question so however you know
32:54to understand uh Market liquidity is
32:57very tricky uh why that's the case right
33:01so Grossman and Miller in their 1988
33:04paper s provide a good summary of the
33:08concept of Market liquidity they said
33:10Market liquidity is determined by both
33:13the demand and the supply of the of the
33:16uh immediacy in trading so it's
33:18basically it's like you you can you can
33:20imagine the Capital Market there are
33:22some like a intang intangible valuable
33:24Goods called you know IM uh immediacy
33:27trading and there's the supply side
33:29there's a demand side and the liquidity
33:31we say is like you know equilibrium
33:34outcome of this both
33:36forces and in in P Cal's 1985 paper pet
33:40also emphasized Market liquidity is an
33:44is not a simple con simple concept
33:46instead it's rather complex it is an
33:49abstract and a multi-dimensional con
33:51concept it is driven by numerous factors
33:54you know from both supply and demand
33:57side of the trading immediacy and
34:00pinning down it using just one number or
34:04one statistic is really tough and fully
34:07embracing this you know concept this
34:10Insight this paper uh offers you know
34:13study try to argue the different assets
34:17of um facets of the market liquidity may
34:20have distinct market
34:23prices so to be more specific the main
34:26objective of this paper is to estimate
34:28the expected returns from providing
34:32different types of liquidity to achieve
34:35this goal the authors take a two-step
34:37approach let me decompose slow down a
34:39little bit so first the the the the
34:41authors identify the component in the in
34:44shortterm reversal due to the liquidity
34:46provision it's called IR
34:49RX by removing this kind of a post
34:52earnings announcement drift and also
34:55this a short-term industry momentum why
34:58because this two are like two prominent
35:00training strategies it's like a based on
35:03this kind of new driven drift they want
35:05to remove this you know uh to identify
35:08this liquidity driven component in the
35:11short-term reversal so it's a very neat
35:14idea and once they construct this kind
35:17of liquidity driven component they
35:19further dissect this kind of liquidity
35:21driven return reversal to identify their
35:24sources in this paper particularly the f
35:27on the inventry risk measured by stock
35:30return volatility and the inventory
35:32duration measured by uh stock turnover
35:36so this is one of the main result of the
35:38paper in panel a shows this kind of
35:41performance of five different trading
35:42strategies so this R the First Column it
35:46is the standard shortterm reversal you
35:49can see this the average monthly return
35:52AIS return the magnitude and the
35:54significance both are weak right so the
35:57second column is this trading strategy
35:59based on this post earnings announcement
36:02draft and the third one is based on
36:04industry momentum both are positive and
36:06significant the idea of this paper is to
36:09remove the second column and third
36:11column from the First Column so they
36:13will re they will reach to the fifth
36:16column this IR RX so that's they call
36:19liquidity
36:20driven uh reversal you can see suddenly
36:23you compare the First Column and the
36:25fifth column the you know the a average
36:27return is much higher than mag both in
36:30magnitude and in
36:32significance okay so then the kind of in
36:35panel B they can say whether this kind
36:37of decomposition is valid in the sense
36:40you know this three components really
36:43statistically span the variation in the
36:45original standard reversal they say Yes
36:48actually see you can see the R square is
36:49pretty high it's close to 90% so it's a
36:52very nice decomposition now they they
36:54take this R this liquidity driven
36:57reversal
36:58and the kind of like UT further
37:00utilizing this cross-sectional variation
37:03the look at the stocks who has high
37:05volatility and stocks with low
37:08volatility
37:09so uh to interpret this cross-sectional
37:12variation they they use this stock uh
37:15return volatility as a proxy for the you
37:17know inventory risk this Market maker of
37:20this liquidity supplyer must Bear right
37:23if this is a inventory risk is higher
37:26they require higher returns to make
37:28provide this liquidity so you can see
37:30the right curve representing um uh the
37:34the stocks with high volatility you can
37:37see this reversal it's a faster and more
37:40dramat much more dramatic and if you do
37:43some back back of envelope calculation
37:46you you compare this kind of like sorry
37:49this kind of Gap around two weeks
37:51between these two lines is about you
37:53know 6% and based on this you can
37:56calculate back out you know what is this
37:58monthly access return uh investors have
38:02to pay for this inventory risk right is
38:06about 1.2% monthly so is a very big and
38:11and the next result is the if they do
38:14another
38:15cross-sectional decomposition so look
38:17they look look look like you know this
38:19kind of stocks with high turnover and
38:21the stocks with low turnover so this R
38:25curve represents the stocks with low
38:27turnover so it's a very striking result
38:29when you first look at this because you
38:31really don't expect you know the
38:32reversal will be so persistent but here
38:35is like fit to the story pretty well
38:38because due to this kind of like you
38:39know this uh uh duration story because
38:42for low turnover a longer duration
38:46inventory duration stocks there a
38:48suppliers they require higher
38:51compensation right for Bear this kind of
38:53risk providing this kind of liquidity
38:56and at the same time uh because it's
38:58longer duration it's lower turnover the
39:01news will be incorporated more slowly
39:04right so this can be seen actually uh if
39:07you write down a you know dynamic kind
39:09of model it's very clear so uh it's
39:12actually look striking but when you
39:15think about this you appreciate the
39:16economics behind it so a very nice
39:18result so my first comment is I like
39:22this kind of liquidity driven uh
39:24components in the reversal the idea is
39:26to remove this kind of news driven you
39:29know drifts from this uh
39:32reversal uh in this paper the uh the
39:35authors are focus on two right this post
39:38earning uh post earning announcement
39:40drift and uh industry momentum but my
39:44question is whether this to be enough to
39:46remove all the contaminations caused by
39:49this new driven drifts because for the
39:52past decades we SE the as presentent
39:54literature that there are other like you
39:56know cross Fromm across the industry uh
39:59news driven drifts uh at a firm level we
40:02know this famous paper by Cohan and
40:05frini right see think about this cross
40:08input output links new striven drift and
40:11cross industry you can go back to Harris
40:14Hall and his co-authors this kind of
40:16lead and lack industry drift and my my
40:20recent work with uh weiu my co-author so
40:23we identify and show this cross industry
40:26moment and view this competition
40:28horizontal competition Network so I'm
40:30thinking whether you know I would
40:32suggest the author think about also
40:34remove this component see whether they
40:37can make the sharp ratio of this irx
40:39even
40:40higher my second comment will be the
40:44identification of this inventory risk I
40:47think it's definitely there it's very
40:48important but how can we better identify
40:51an estimate this paper already made a
40:53huge progress on it but I think to give
40:56like a you know the the per better
40:59answer we should be more careful about
41:01this caal inference so because we all
41:04know right this a return volatility by
41:06the this is the measure this paper used
41:08to proxy this inventory risk and the
41:10return reversal both are highly
41:12endogenous when you think about the
41:14association in the cross-section of two
41:16hiding dodg variables you you may think
41:19this Association may not reflect the
41:22caal relation which means the inventory
41:25risk cause the you reversal actually
41:29here there's a concern about this
41:30reversal causality because you know when
41:33we when we thought in the crosssection
41:35according to stock volatility maybe
41:38mechanically we put this kind of stocks
41:40who have a short and a quick return
41:43reversal into the you know the the dell
41:46and the groups of high volatility so
41:48that's why you see this Association in
41:50the cross-section it doesn't really mean
41:53causal so my suggestion is actually
41:56there are some there more data comes out
41:59about this quantities about n trator
42:02behaviors one meas I think the authors
42:04can use is the monthly standard
42:06deviation of the retail investors other
42:09imbalances it's been used by some
42:11researchers in the literature and uh the
42:14idea is you know this kind of inventry
42:16risk yes is a volatility risk but uh
42:19volatility risk of the returns but there
42:21you want to take out focus on those
42:24really caused by this short-term no
42:26Trading right so basically this this
42:30data on the retail investors uh ordering
42:33balance can be useful for this to
42:34improve the result and the last comment
42:37of mine would be yes go back to the
42:40definition of the market
42:42liquidity right K made a lot of
42:45contribution on this and grossen and
42:47Miller so it is supply and demand and
42:51this inventry risk and the inventry
42:54duration considered the factors
42:56considered in this paper all focus on
42:58the supply side so when you think about
43:01the variation crossover variation of the
43:03factors from the supply side and their
43:06impact usually we need to think about
43:08whether we control we been doing a good
43:11job to control for the demount side so I
43:15know it's very challenging question but
43:17uh actually there are some advances in
43:19the literature for the past few years
43:22again I want to go back to this data
43:24this a retail
43:25investor uh order imbalance data so
43:28maybe the assets can use this kind of
43:30like you know retail investors absolute
43:33monthly order imbalances as a proxy for
43:37the trading intensity of noise traders
43:39to control that and to think about the
43:41variation on the supply side okay and uh
43:44also there are some literature uh they
43:47try to estimate the demand curve from
43:49the demand side of liquidity and uh go
43:53to you know coin Yogo 2019 and there
43:56some followup in improvements so
43:58basically the authors can also take out
44:00this kind price impact of the demand
44:02shocks liquidity shocks right from those
44:05papers and try to control for this those
44:08and a look at the variation on the
44:10supply side of the liquidity so that
44:12will give a better estimation and a
44:14better identification so let me uh
44:18summarize uh first this is a really
44:20great paper a significant Imperial
44:23contribution on this very important yet
44:25very challenging topic what I appreciate
44:28the most is this paper uh
44:31provide a novel and useful estimate of
44:35the expected returns from liquidity
44:37provision and most more importantly they
44:40offer this value very valuable
44:43perspective on the pricing of liquidity
44:46from various or Origins various sources
44:50of liquidity they just don't try to you
44:53know put one number say this is the
44:55return on this on Market liquidity they
44:57did very careful job fully embrac this
44:59kind of insights from Theory theoretical
45:02literature try to say this is the price
45:05this is a return for this type of
45:06liquidity this is a return for another
45:10type of liquidity so I think this is a
45:12very promising direction to go uh I have
45:15a few humble suggestions uh to further
45:18refine the metric for the liquidity
45:19provision component and uh sharpen the
45:22identification of the impact of unitary
45:24risk and uh control explore the factors
45:27influencing this kind of liquidity from
45:29the supply side oh s sorry from the
45:31demand side and the control for them
45:34when you think about variation from the
45:35supply side that's my discussion thank
45:38you very much thank
45:41you rob do you want to react and then
45:43we'll open it up for question sure so I
45:44mean there's a couple issues one is I I
45:47I think Winston is correct about some of
45:49the issues but um the struggle writing a
45:53paper is never to kind of put more in
45:55and do everything it's what you leave
45:57out and what's important and uh I would
45:59actually like to take more out of this
46:01paper not put more in at this point and
46:03it's it's not meant I mean I'm trying to
46:05document um I think some really
46:07interesting facts that aren't well known
46:09um and um I think trying to do
46:13everything um is should be done but
46:16maybe not in this paper maybe in
46:18follow-on papers um in terms of uh you
46:23know it's also not a theory paper I do
46:25motivate some stuff with Theory but this
46:27is really an empirical paper documenting
46:30some facts
46:32um I guess those are the main comments I
46:35mean I think that um doing a greater
46:37reversal refinement um might be good but
46:40it's you know I don't think that's
46:42really I think it's an interesting fact
46:44that you can so easily get a much
46:46stronger version of reversals but it's
46:47really not the point of the paper the
46:49paper is really about what different
46:51aspects uh what other characteristics
46:53predict the the size and and Persistence
46:56of these things
46:57um I guess the one question I had is
46:58about if it's really all um supply side
47:02because um the I mean turnover is one of
47:06the big things we're looking at and
47:07turnover seems like a demand side
47:09variable to me because um the market
47:11makers are not the ones you know we're
47:14only counting the the the the the trades
47:17as they're initiated which is you know
47:19by noise Traders or liquidity demanders
47:21and uh that doesn't seem like a supply
47:23side variable uh as much to me yeah yeah
47:28yeah I agree so but but my
47:31interpretation here there's kind of like
47:32a additional expect for return due to
47:34this kind of inv inventory duration is
47:36more from the uh risk bearing
47:39perspective of the suppliers that's why
47:41I say it's a supply side okay yeah yeah
47:43they have to hold longer yeah okay let's
47:46open it up for questions questions in
47:48the front Larry so the question is are
47:51these results obtained from uh closing
47:53prices or closing spread midpoints uh
47:57these are closing prices all right so
47:59for highly liquid stocks the spreads can
48:02be pretty large and if a stock closes up
48:05it's more likely to be included in those
48:08stocks that are liable to reverse and of
48:11course if they close down later they're
48:13going to reverse so if you uh would look
48:15at closing spread midpoints um you'll at
48:19least get some of what's happening in
48:20the first part of the sample the first
48:23part of the of these graphs that's a
48:25great suggestion we should definitely
48:26look at you're basically saying there's
48:28some bid ask bounce in here that's not
48:30real in terms of so the suggestion I
48:33guess the question I suggest so the
48:35other suggestion is that um there are
48:39two possibilities for explaining these
48:40results one is that it's a liquidity
48:43issue that we think is pushing on
48:45somebody's inventory not necessarily the
48:47market makers but somebody uh and the
48:50other possibility is that people are
48:52responding in systematic ways to
48:54information and it's somehow correlated
48:57and they forget about things and the
48:58pressure relaxes or something like that
49:01perhaps the way to discriminate between
49:03those two would be to see if there's any
49:05evidence that people are accumulating or
49:09or divesting in the period during which
49:12we're identifying the stocks that are
49:15going to go into these portfolios and we
49:17could do that by looking at um uh
49:21transaction data uh classifying them by
49:24whether the trades are taking place at
49:26at BS or offers and accumulating
49:28essentially what's called Trin uh and so
49:32uh that information might allow us to
49:34discriminate among these two
49:35possibilities and and perhaps give us a
49:38deeper understanding of whether it's
49:40really liquidity or whether it's um some
49:44other behavioral issue yeah so in the
49:46appendix I do have some plots showing
49:49what happens cumulative order Flows In
49:52The periods leading up to portfolio
49:54formation and it does look like there's
49:56some buildup of inventories in the way
49:58that would agree with um but I I'm it's
50:02you know
50:04those I'm just you know better than me
50:07it's based on Le and ready algorithm in
50:09the TAC data and I'm just not sure I
50:11mean there's a lot of noise in that
50:13signing of the order flow as well so but
50:16it is there thanks Larry questions Pete
50:19has a
50:21question yes so if if you're an asset
50:23manager and you're looking for a a way
50:25to make money by providing liquidity you
50:28know this paper is obviously very
50:29interesting but I I wanted to suggest a
50:32story and you tell me if you agree with
50:34it which is that it's relatively
50:36straightforward to implement a strategy
50:38that just trades against one day
50:39reversals or you know one month
50:41reversals it it's also little less
50:43straightforward but it's quite uh not
50:45too hard to implement a strategy of
50:47trading on post earnings announcement
50:49drift um but those the gist of the paper
50:52is that those two strategies kind of
50:55undermine one another um and therefore
50:57the the better strategy is to combine
50:59kind of combine the two together
51:01technology has changed and data
51:03availability has changed in such a way
51:05that combining the two together is much
51:06more feasible now than it would have
51:08been say 20 30 40 years ago so uh the
51:13time Trends must be showing that it's
51:15dramatically less popular in recent
51:17years than in past years is that's
51:19question so I mean we do show kind of
51:21Trends and things and we do a whole
51:23separate analysis post decimalization
51:25versus pred iation and um post
51:29decimalization like the reversals are
51:31gone but the irx um is still
51:34significantly profitable um though maybe
51:37only half the sharp ratio that it had in
51:39the early sample but it's um it's still
51:41a very significant thing in the last 20
51:43years uh you have less statistical power
51:46because you're looking at a relatively
51:47short sample but um um there's no
51:50question that um I mean I guess the way
51:53I think about it is that it's gotten
51:54cheaper to trade and it's you know I use
51:56post decimalization for the to include
52:00other things Winston mentioned like
52:02direct Market access and programmatic
52:04trading and um but I I think there's no
52:06question um the markets have gotten more
52:09liquid and liquidity provision
52:11strategies at least none of these
52:14liquidity provision strategies like you
52:15could not I I would not say you could
52:17try capture all of this even if you're
52:19providing liquidity there's some costs
52:21to you to doing that and you're making
52:23these liquidity provision returns net of
52:25the costs you incur and they're not
52:27nearly as profitable in practice as the
52:29very very high numbers I saw um and so
52:33um I think it's not surprising that as
52:35it becomes cheaper and you're incurring
52:36less of those additional costs the gross
52:39returns to liquidity provision have come
52:41down as the costs to providing that you
52:44incur providing it have gone down as
52:45well um
52:47but question in the middle great paper
52:51uh definitely very intrigued by the irx
52:53uh measure you have and you you you sort
52:54of uh drew the contrast between High Vol
52:56low Val I'm just wondering if you could
52:58tie that into cross-sectional dispersion
53:00like you know I mean we know that
53:01overall reversal phenomenon depends in
53:03cross-sectional dispersion as well so uh
53:06presumably that would uh uh
53:10exacerbate I haven't done this in this
53:12version of the paper but I have a table
53:14in the revision I'm working on um which
53:17explicitly tries to control for what I
53:19would call pre-formation spreads um so
53:21there's a critique of the sort of
53:23conditioning variables on momentum in
53:25the literature which is we see stronger
53:28momentum among High volatility stocks we
53:29see stronger momentum among small stocks
53:32but those stocks are more volatile and
53:34so when you look at the amount of
53:36momentum in the momentum strategy
53:38constructed among High volatility stocks
53:40it's much higher because the winners
53:42were bigger winners and the losers were
53:43bigger losers on average among those
53:45High volatility stocks um so I I have a
53:48a table in the paper now or in in the
53:51next version that will be coming out
53:53that explicitly controls for
53:55pre-formation spreads so among stocks of
53:57different volatilities low volatility
53:59High volatility you're looking at
54:00strategies that have on average the same
54:03amount of spread in that reversal signal
54:07um and you still see these dramatic
54:09differences in post formation
54:11performance and so while the critique
54:13has been made for momentum it doesn't
54:14seem to apply here I I'm not sure that's
54:17exactly the question you answered that
54:18you asked but that was what I
54:19interpreted it is being related to I
54:22meant that you know overall levels of
54:25cross-sectional vol
54:27determine the efficacy of reversals uh
54:30in general so if you made it regime
54:31dependent let's say you take the irx
54:33measure and you make make that you're
54:34asking about some interaction between
54:36the time series results and the
54:37crosssection exctly cross-section versus
54:39time series I have not done that um I I
54:41I hadn't even thought to do that but I
54:43could look in you know how how these
54:45results are different in different
54:48volatility regimes for kind of aggregate
54:49market and um I haven't done that I
54:52could do that questions right over here
54:54have you thought about using the trading
54:56volume on the day of the big price
54:58reaction to kind of proxy out
55:00information or as a proxy for new
55:03information I no I mean I'm using
55:05momentum as a I'm using that trading
55:09volume really over the previous quarter
55:11we've looked at different Horizons but
55:12the previous quarter is the primary
55:14version I'm using I haven't looked for
55:16spikes in trading um though um and those
55:22are often also coincident with these um
55:25earnings announcement days or or windows
55:28that that we are controlling for on the
55:30return side um but I have not explicitly
55:33used those as sort of a forecasting
55:35variable of how these strategies are
55:37going to do we do know that
55:40concurrently High volumes Associated on
55:42average with high returns um but I
55:45haven't looked at it here
55:48explicitly as a method for decomposing
55:50information yeah uh Ken hi uh thank you
55:53for the talk uh I have a quick question
55:55for WI
55:57you seem to equate retail Traders with
55:59noise Traders uh in your suggestions
56:03given the fact that institutions on
56:05average also don't outperform the market
56:08is that a distinction that
56:10matters uh so basically in reality not
56:14Traders not just retail Traders
56:17definitely there are some liquidity
56:18Traders can be insurance companies those
56:21those type of firms but it but here uh
56:24you know I think the retail their
56:26trading Behavior can be used as one of
56:29the as the instrument variable to kind
56:32of like you know instrument this kind of
56:33like uh this uh Supply a demand side
56:37movements so that's my suggestion it's
56:39it's a you should think that as an
56:41instrument ra rather than you know
56:43representation of the whole no tra