YouTube transcript (DKWVDvk1QYo) — Transcript
Full transcript
- 0:00Welcome to the meetings.
- 0:02Uh, into this particular session, I'm
- 0:04impressed to draw off this early to come
- 0:06out for this session and
- 0:09glad glad to have all of you here. We
- 0:11can talk for a little while on the
- 0:13subject, but you know, my plan is to
- 0:15leave plenty of time
- 0:17at the end for
- 0:19for Q&A. So,
- 0:20if you have any any thoughts along the
- 0:22way, throw them out and
- 0:24I'm sure it won't be
- 0:26unique.
- 0:27Um,
- 0:29So, this this topic is one I think that
- 0:32those those of you, presumably all of
- 0:34you since you're here, do work in in the
- 0:36area of corporate finance encounter
- 0:40problems associated with endogeneity and
- 0:42identification in in your empirical
- 0:45work. And so, I thought I would start
- 0:47like I do sometimes in a in a classroom
- 0:49setting. Start with a question
- 0:51and say, "Of among all of you who have
- 0:55submitted papers in in the field of
- 0:57corporate finance, how many of you
- 0:59have
- 1:00a referee or an editor say,
- 1:03"I don't buy your identification
- 1:05strategy." or something to that effect?
- 1:10That's all, really? Does that mean the
- 1:13rest of you haven't submitted papers yet
- 1:14to
- 1:15to this conference?
- 1:17I think you're hard-pressed
- 1:20in corporate finance to
- 1:23to do work without that
- 1:26endogeneity problem coming up without
- 1:29your identification strategy being
- 1:31criticized in some way. And so, it begs
- 1:34a lot of questions, I think, in terms of
- 1:37you know, how how how can we do this
- 1:39research in corporate finance in a
- 1:41meaningful way while still
- 1:45dealing with this endogeneity
- 1:47endogeneity problem. Now, a lot of
- 1:48people complain about how they're
- 1:50treated by referees and and editors on
- 1:53this issue and and and refer to
- 1:56so-called identification police. So, an
- 1:58alternative title for this talk might be
- 2:01eluding the long arm of identification
- 2:03police.
- 2:05But, the my idea today is not really to
- 2:08to talk about econometrics. There's been
- 2:12several very good tutorials at the FMA
- 2:14over years on that subject. I'm more
- 2:17interested in talking about sort of
- 2:20research design issues in empirical
- 2:22corporate finance with a particular eye
- 2:25towards expanding the types of studies
- 2:28that we do cuz I one of the concerns
- 2:31that I have is that this identification
- 2:34problem has has become so pervasive in
- 2:37our work that it's sort of narrowed down
- 2:40the the styles of papers that we think
- 2:42are acceptable to to journals now. And
- 2:45as a result, we're potentially limiting
- 2:47the scope of questions that we're asking
- 2:50in the field. And I I don't think that
- 2:51would be a good development if that's
- 2:53the case. So,
- 2:55uh what I'm going to do is is is the
- 2:57following. I'm just going to talk a
- 2:59little bit at at the start about the
- 3:01basics of of what the endogeneity
- 3:03problem is. I think everyone's pretty
- 3:05familiar with that as well as some of
- 3:08the econometric responses that that
- 3:10exist in the literature. And then I'm
- 3:12going to talk about those. Again, this
- 3:14isn't a talk about econometrics, but but
- 3:17more just to point out that none of
- 3:19these are are perfect solutions by any
- 3:22stretch. They all have uh
- 3:24some limitations to them. We're never
- 3:26completely solving the problem.
- 3:29Uh
- 3:30but what we have done is sort of
- 3:31narrowed down the the
- 3:34the set of papers I think that that we
- 3:36tend to do in empirical corporate
- 3:38finance. And that that raised these
- 3:40these concerns that I alluded to
- 3:43a couple minutes ago.
- 3:45And then what I'm going to do is is sort
- 3:46of apply some of these ideas in in the
- 3:49field of research that I've been doing
- 3:51more work on lately, which is capital
- 3:53structure research. So, I'm going to to
- 3:56talk about alternative different types
- 3:58of of of research designs, all of which
- 4:02I think are useful in terms of what
- 4:04we're really trying to do in the
- 4:06profession, which is expand the the set
- 4:08of knowledge that we have on a
- 4:10particular topic. None of them are
- 4:12perfect. Many of them will suffer from
- 4:15identification problems, but that
- 4:17doesn't mean they they they aren't
- 4:19useful. And so, I I I think uh this is
- 4:22this is where I'm going to try to
- 4:24convince you that we need to to
- 4:25re-expand the types of studies that that
- 4:28we're doing if we really want to get at
- 4:30some of the big questions that that we
- 4:32encounter in in corporate finance. And
- 4:34And I'll run through several examples
- 4:37from from the recent capital structure
- 4:39literature that I think
- 4:41uh illustrate the points that I'm trying
- 4:43to make, both in terms of what we can
- 4:45learn from those types of studies as
- 4:47well as what we what we can't learn, and
- 4:49how
- 4:51think about doing studies like that
- 4:54um in a way that uh will be viewed as
- 4:58useful by the journals and and therefore
- 5:00uh publishable in in uh in the
- 5:02literature. And then
- 5:05draw some conclusions from that.
- 5:07Okay, so the basic endogeneity problem I
- 5:10think everyone is is pretty familiar
- 5:11with. A lot of the research
- 5:14that we do in in corporate finance is
- 5:16really geared towards trying to
- 5:18establish some causal connection between
- 5:21some variable X1 in this case and and
- 5:24some
- 5:25dependent variable um
- 5:27that let's list as Y in in this
- 5:30particular case. So, you think of it in
- 5:32a regression framework, doesn't have to
- 5:34be a linear regression like this, but
- 5:36it's easier to represent this way.
- 5:39Um we have the we have the association
- 5:41between Y and X1 while controlling for
- 5:44some factor
- 5:46uh of of X variables that we think are
- 5:49are other possible determinants of Y.
- 5:52We're trying to establish whether access
- 5:55effects
- 5:56on on Y. So,
- 5:59common examples in the corporate finance
- 6:01literature, you know, does The question
- 6:03might be, does debt constrain
- 6:06investments on Y? So, we can we can test
- 6:09the association between debt levels and
- 6:11levels of investment.
- 6:13Uh do taxes affect capital structure?
- 6:16So, it's the association between some
- 6:18measure of the of the of the tax
- 6:21obligation that the firm is facing and
- 6:23the leverage choice that they make.
- 6:25Uh or does does governance affect value
- 6:31notable example that has a connection
- 6:33between ownership and some measure of
- 6:36value like like Tobin's Q.
- 6:38All right. So, that's These are typical
- 6:40tests, all of which suffer from this
- 6:43this identification problem because we
- 6:45don't know
- 6:47whether there's a causal connection
- 6:48between this variable X X sub one and
- 6:52the dependent variable or whether we've
- 6:53got some correlated variables. So,
- 6:56ultimately, the the issue is whether
- 6:58there is some some correlation between
- 7:00this X1 variable and the error term of
- 7:03this model. Now, obviously, if we had
- 7:06some idea what these omitted variables
- 7:09were, we'd put them in
- 7:11and and then we could take care of the
- 7:12problem. But, the issue is we don't know
- 7:15what they are.
- 7:16I mean, and it's very easy for a referee
- 7:20or an editor to say, "Well, there's
- 7:22something out there
- 7:23that's that's correlated both with your
- 7:26X1 variable and this dependent variable
- 7:29that you're trying to test for."
- 7:31And so, again, I just don't buy
- 7:33identification strategy. I don't buy
- 7:35that there is a a causal connection
- 7:37between the two. So, for example,
- 7:40you know, if we run if we run
- 7:41regressions on investment, on leverage,
- 7:44we will find that the higher is the
- 7:46leverage, the lower is the level of
- 7:48investment.
- 7:49Right? And that could be
- 7:51because debt constrains investment, or
- 7:54it could or it could just be that firms
- 7:56that have poor growth opportunities
- 7:59uh end up higher leverage. So, we would
- 8:02predict they shouldn't be in capital
- 8:03structure models.
- 8:05Right? And you simply can't perfectly
- 8:07control for growth opportunities. So,
- 8:09even if we have some proxies for growth
- 8:11opportunities on the right-hand side, if
- 8:13we can't perfectly control for them, the
- 8:16omitted part is is embedded in the error
- 8:18term, and we have this positive
- 8:19correlation between
- 8:21uh debt and and the error term,
- 8:24and we're stuck.
- 8:26So, it's What do we do about that? So,
- 8:28where I think the profession has has
- 8:30really made some some big strides in in
- 8:33the last 10-15 years is in understanding
- 8:36econometrics and in the application of
- 8:39these uh econometric methods to problems
- 8:42associated with with identification.
- 8:46Right? So,
- 8:47it's you know, it's fairly standard now
- 8:49that you see panel data techniques used,
- 8:52firm fixed effects, industry fixed
- 8:54effects, year fixed effects.
- 8:57Those are useful to a point. I mean,
- 9:00you've got some omitted variable that is
- 9:03firm-specific
- 9:05and doesn't vary through time, firm
- 9:07fixed effects are taken care of.
- 9:09All right, well, good for you if that's
- 9:11the case. But, the problem is we don't
- 9:13know that that's that's the nature of
- 9:15the omitted variable. So, if there's
- 9:17some firm-specific
- 9:19but time-varying
- 9:21variable that is that is the source of
- 9:23this endogeneity,
- 9:25the panel data techniques aren't aren't
- 9:27really going to do it for us. And And
- 9:29again, unless we really know what the
- 9:31omitted variable is, which we can't by
- 9:33definition, uh it it's it we're still
- 9:37going to subject at times to this this
- 9:39criticism. Now, what panel data
- 9:42techniques are
- 9:43useful for is that they're at least
- 9:45limiting the scope
- 9:47of the endogeneity problem.
- 9:49This is you can't be criticized
- 9:52for omitting something that is just some
- 9:54factor that isn't going to vary through
- 9:56time within a firm. You have to control
- 9:58for that. So, it's a bit of a solution
- 10:00in that sense, but it can never
- 10:03completely take care of the problem
- 10:05unless you really know precisely the the
- 10:07nature of the problem.
- 10:09All right. Another possibility is some
- 10:11sort of regression discontinuity
- 10:13technique.
- 10:14So, here here the idea would be that
- 10:18we have some sort of threshold event
- 10:21and when we cross this threshold, we can
- 10:23test whether there's a difference in
- 10:25behavior or something. So, an example
- 10:26would be uh if you're if you're trying
- 10:29to test
- 10:30uh whether creditor control has some
- 10:32impact on real investment decisions.
- 10:36Let's say something like Chava and
- 10:37Roberts in
- 10:38in their paper. So, you could you can
- 10:41observe a covenant violation.
- 10:43All right. And the question is when you
- 10:45cross this threshold and violate the
- 10:47covenant,
- 10:48all right, do you observe a difference
- 10:50in behavior? Now, the the nice advantage
- 10:52of the regression discontinuity
- 10:54technique is that right around that
- 10:56threshold,
- 10:57uh you can plausibly say that firms that
- 11:00just on either side of that threshold
- 11:03are roughly the same.
- 11:05And their their characteristics are are
- 11:07roughly the same. And if you can make
- 11:08that claim, all right, then this is a
- 11:10pretty solid way to try to deal with
- 11:13with the endogeneity criticism.
- 11:15Limitation here is that
- 11:17you don't come across that many clean
- 11:20threshold types of events that you can
- 11:22really use in in the context that that
- 11:25we study quite a lot.
- 11:26If you can find them, it's great.
- 11:29By all means, use them.
- 11:31Um but
- 11:34my experience is are difficult to find
- 11:37and therefore
- 11:38not often do you come across
- 11:41situations where that's really going to
- 11:42help you out with your identification
- 11:45problem.
- 11:46Third possibility is to surround to some
- 11:48sort of matching models and propensity
- 11:50score type of approach. And again,
- 11:53that's great if you can if you can
- 11:55identify what you think are are the most
- 11:58plausible set of factors that you can
- 12:00match up firms and do a difference in
- 12:02difference sort of approach. But again,
- 12:05almost by definition, these can't be
- 12:07perfect because we're talking about
- 12:09correlated omitted variables that we
- 12:11don't really know what they are.
- 12:13If we knew what they are, sure we could
- 12:14perfectly match on them and and and run
- 12:17this type of test, but not knowing what
- 12:20they are leaves us a little bit in
- 12:22limbo.
- 12:23Uh instrumental variables, not another
- 12:26possibility. Again, has some
- 12:28limitations. We need some some variable
- 12:30that's going to satisfy both the
- 12:32relevance criteria and the exclusion
- 12:36criteria.
- 12:37Relevance is pretty easy to demonstrate
- 12:39most of the time with with an IV, but
- 12:42it's also it's really easy to criticize
- 12:44instrumental variables on on the
- 12:47on the exclusion part. And so,
- 12:50uh they're difficult to find. Good
- 12:51instruments are really difficult to come
- 12:53across. They've got to be based on on
- 12:56underlying economics, but you know,
- 12:58there's no great test for
- 13:01for that instrument, and so you're
- 13:03always open to this criticism that
- 13:05you just got that instrument.
- 13:07And so,
- 13:09all of these techniques are useful.
- 13:11All of them should be done
- 13:13in different contexts, but we we can
- 13:16never use these techniques and think,
- 13:18"Okay, we solved
- 13:20the endogeneity problem." We don't We
- 13:22don't We don't solve it using
- 13:24econometrics.
- 13:25So So, where the profession is of headed
- 13:28more towards is is the fifth one, which
- 13:31is is natural experiments.
- 13:34All right. What's great about natural
- 13:35experiments is that you can identify
- 13:38some plausibly exogenous event that's
- 13:41truly exogenous, and you can do sort of
- 13:45a different diff sort of a different
- 13:47diff analysis and and claim causality in
- 13:51a pretty credible fashion.
- 13:53All right. And because you can do that
- 13:56and defend it much more easily than some
- 13:58of the others, I this has become sort of
- 14:01the holy grail of science in terms of
- 14:03doing empirical corporate finance. Uh
- 14:07but it it's not without problems, I
- 14:09think. And it's the the these are the
- 14:10concerns
- 14:12that I was alluding to before. I think
- 14:14it it I think the way which we started
- 14:16to head
- 14:18is is to have this general belief that
- 14:20every study's got to have some clean
- 14:22natural experiment or some some perfect
- 14:25instrument. Otherwise, the study's doing
- 14:28you know, just a quick check on SSRN
- 14:30that I did, you know, just
- 14:32putting natural experiment in the
- 14:34abstract of papers, I get over 1,600
- 14:37entries of of recent papers
- 14:40uh in the field, mostly in corporate
- 14:42finance where where you see that.
- 14:44Now, I want to
- 14:46make sure I'm clear that I I'm not at
- 14:48all criticizing natural experiments. I
- 14:51mean, they're they're great if you've
- 14:53got them and and they and they fit the
- 14:55question you're asking, but my concern
- 14:57is sort of twofold is that what you tend
- 15:00to see a little bit more often now than
- 15:03you than than you saw before is that
- 15:05authors are kind of starting with the
- 15:07experiment
- 15:08rather than with the question.
- 15:11That's not
- 15:12quite the way you want to you want to go
- 15:14about doing research. I'll say more
- 15:15about that in in a minute. In a way, it
- 15:18it it's kind of similar to
- 15:21uh concerns that have come up over the
- 15:23years as sources of data have changed.
- 15:27You know, back in the dark ages when I
- 15:29was getting my degree,
- 15:31there wasn't a whole lot of machine
- 15:34readable data.
- 15:35And now,
- 15:36most data is is machine readable.
- 15:39Um, and as a result, you saw over time,
- 15:43at least my perception is that you saw
- 15:45authors start to shy away from data
- 15:48collection data. They would say, "Well,
- 15:49they would do a study unless they could
- 15:51identify a readily readily available
- 15:54data source." And then you even to the
- 15:55point where
- 15:57uh, they're just be looking for data all
- 15:59the time, see the data in in good
- 16:01machine readable form, and say, "Well,
- 16:03let me figure out something to do
- 16:05with this data."
- 16:06Again, that's that's not really the way
- 16:08you want to conduct research. You want
- 16:09to you want to start um,
- 16:12with the question.
- 16:14Um, so it it alluded to sometimes as,
- 16:17you know, are
- 16:18are we reaching a point where we
- 16:21unnatural obsession with natural
- 16:23experiments. I think we've got to have
- 16:24these things all the time. And so that's
- 16:27one concern
- 16:29that is sort of changing the way that
- 16:30people go about doing research. A
- 16:32related concern though, which I think is
- 16:34the bigger one in terms of of the field
- 16:37or the profession in general, is that
- 16:40it can be very limiting in terms of the
- 16:42scope for the questions that you ask.
- 16:45And there's some big questions
- 16:47in our field like that should be
- 16:50addressed.
- 16:51And when these big questions come up,
- 16:54maybe there isn't a perfect natural
- 16:57experiment. Maybe there isn't a perfect
- 16:59instrumental variable that you can use.
- 17:01And if people think, "Well, unless I
- 17:03have that, I can't do this study," then
- 17:05those studies don't get done. But we're
- 17:08giving up a lot if that's the way we we
- 17:10approach research, I think, in the sense
- 17:12of
- 17:13really important questions that we could
- 17:15shed some light on, and just
- 17:17incrementally move towards a richer
- 17:20knowledge of somebody's questions. I'm
- 17:23going to talk about the capital
- 17:24structure research, and I'm going to try
- 17:27to
- 17:28highlight a little bit more what I what
- 17:29I mean by that.
- 17:31All right. So, all right. So, let me
- 17:33compare those concerns
- 17:35with what I think most of us in the room
- 17:38would would say is this how we would
- 17:40advise PhD students as to how to go
- 17:43about doing research. Right? First and
- 17:46foremost, say, "Well, you want to
- 17:47identify an interesting question."
- 17:50When students come and talk to me, it's
- 17:53frequently the first thing I'm saying,
- 17:54"Well, what's the question that you're
- 17:56asking people to follow?" And And I've
- 17:58noticed this even in terms of uh
- 18:02interviews uh at the meetings FMA or FA
- 18:05meetings or interviewing candidates, and
- 18:08you ask the candidate to
- 18:10launch into their spiel about about
- 18:12their research, and the first thing they
- 18:15do is they go straight to their
- 18:16experiment. Like, this is the experiment
- 18:18I'm I'm running.
- 18:20That's not really what most people want
- 18:22to hear. They don't want to They don't
- 18:23want to know what's what's your
- 18:24question. What's the question you're
- 18:26really trying to get at? And And this is
- 18:29the way I think, as I said, most of us
- 18:31would would probably advise students to
- 18:33do research, but first identifying an
- 18:37interesting question.
- 18:38All right. Once you've got that, all
- 18:41right, then you should be thinking about
- 18:43developing hypotheses from first
- 18:45principles, from the ground up.
- 18:48And develop a plausible, testable set of
- 18:51hypotheses.
- 18:53And then, and only then, I think, when
- 18:55I've gotten to reach that point, then
- 18:57you start thinking about the
- 18:59experimental design. Well, what's the
- 19:00best way
- 19:02that I could go about trying to shed
- 19:04some light on on this particular set of
- 19:07hypotheses?
- 19:09All right. And this This is the part,
- 19:10you know, just to reiterate what I said
- 19:12a couple minutes ago, like this is the
- 19:14part that that at times looks like it's
- 19:16getting split
- 19:17in in the profession. It's like too many
- 19:19studies
- 19:21are starting with the experiment
- 19:24uh and then uh and then moving on
- 19:27uh
- 19:28from there. All right, so once we
- 19:29identify this experimental design that
- 19:32we think is the best for say ISM and
- 19:34line on, you know, then we then we
- 19:35eventually get some results.
- 19:38We got to let the data speak, but we got
- 19:39to let the data speak in in in a way
- 19:42that we're well aware of what the
- 19:44limitations of what we're doing
- 19:46and this is
- 19:48I guess this is more of a criticism of
- 19:51uh referees than it is of authors at
- 19:53this point is that if you you tend to
- 19:55see a lot of referees
- 19:57uh
- 19:58who are unwilling to accept limitations.
- 20:01Uh say, "Well, if you don't have a
- 20:03perfect identification strategy, I'm
- 20:05going to reject your paper." That's
- 20:07already leading people to think, "Well,
- 20:09I got to have a perfect identification
- 20:11strategy, otherwise this this paper is
- 20:12not worth doing." I'm not sure that's
- 20:15the best
- 20:16uh development for our profession
- 20:18because there are a lot of things that
- 20:19we can learn from studies that are
- 20:22imperfect.
- 20:23All right, because let's be honest,
- 20:25every study is imperfect
- 20:28in some way.
- 20:29Uh and if the particular way is you
- 20:32can't necessarily identify causation in
- 20:35your study, it doesn't necessarily mean
- 20:38this that the study isn't worth doing,
- 20:40but it does mean that if you do the
- 20:42study and you fail to note those
- 20:44limitations, you probably deserve to be
- 20:46rejected at this point, right?
- 20:48You can you can provide evidence, you
- 20:51can interpret that evidence, you can
- 20:53understand that maybe you can't
- 20:56uh imply causation from that set of
- 20:58evidence. If all that is kind of laid
- 21:00out for the reader, then the reader can
- 21:02decide in the end, well, given all these
- 21:05limitations, do I feel like I'm learned
- 21:07something from from reading the study?
- 21:09And
- 21:09the answer is yes. And you learned
- 21:12enough,
- 21:13then the journal should be willing to
- 21:15accept that paper, and they do.
- 21:17And they do, and that's
- 21:19that's important.
- 21:21Um
- 21:23Let me let me just back up a second.
- 21:24There's one one more
- 21:27issue that I was I wanted to note on
- 21:28this particular slide.
- 21:31I'm using the example of business
- 21:33combination laws. Now, so so this is a
- 21:36natural experiment I think that has had
- 21:37a lot of success
- 21:39in the profession. And so let me be
- 21:41clear up front, I'm not criticizing this
- 21:44this natural experiment. It's a good
- 21:45one.
- 21:46The way it's useful
- 21:48is is in the sense that uh
- 21:51if you think about types of studies in
- 21:53corporate finance, one big class of
- 21:55studies is asking whether governance has
- 21:59some causal impact on value or or
- 22:03managerial action. You know, these are
- 22:06interesting and big questions, I think,
- 22:09but they're tough to get at
- 22:10from an identification standpoint. So,
- 22:14a clever natural experiment
- 22:17is the introduction of these of these
- 22:19business combination laws. And they they
- 22:21these laws are
- 22:23are sort of staged over time in in
- 22:26different states. So, it makes for a
- 22:27really nice exogenous shock to
- 22:31governance that we can then connect with
- 22:33some sort of outcome variable.
- 22:35And so, it's got a lot of the attributes
- 22:38that you're looking for in natural
- 22:40experiments. Now, it's also got
- 22:42it also raises some of the concerns that
- 22:44you always see in natural experiments as
- 22:47well, because none of these what we call
- 22:49natural experiments in the in the social
- 22:51science are never really pure natural
- 22:54experiments, because somebody had to
- 22:56decide
- 22:57that we're going to change these laws.
- 22:59And the laws are not purely exogenous,
- 23:01they're endogenous in some way. And
- 23:04maybe what's driving the change in the
- 23:06laws is correlated with your outcome
- 23:08variable. But these are authors who have
- 23:10done these these these studies of
- 23:11business combination laws have been well
- 23:13aware of those limitations and made a
- 23:16plausible case that uh they're they're
- 23:19not interfering with the
- 23:21interpretation of causation. Right? So
- 23:24it's So it's a pretty nice
- 23:26clean natural experiment. Um
- 23:29but it's probably one that I've seen
- 23:31repeated the most times of any sort of
- 23:34natural experiments. It's been repeated
- 23:36in some good ways.
- 23:38Right? But you start It starts to run
- 23:40out of juice after a while because, you
- 23:43know, you see some some big questions
- 23:44like does governance affect value? Does
- 23:46governance affect investment? Does
- 23:48governance affect innovation? These are
- 23:51all
- 23:52very useful applications of this natural
- 23:54experiment. But But what tends to happen
- 23:56in our profession
- 23:58is that everyone says, "Well, you need a
- 24:01clean natural experiment.
- 24:03I can see from these publications that
- 24:05the profession seems to accept this one
- 24:08as as a useful natural experiment. So
- 24:11look and I did with this one." Again,
- 24:13you're starting with the experiment then
- 24:16rather than the question and we start to
- 24:18get into some really marginal questions
- 24:20at this point, you know? You just keep
- 24:22following this path and you eventually
- 24:24start seeing studies that are like, you
- 24:26know, "What's What's the impact of
- 24:29of columns of two business combination
- 24:31laws and what the CEO had for lunch on
- 24:33Tuesday?" Well, yeah, I guess you can
- 24:36establish causation with this natural
- 24:38experiment, but now you're asking a
- 24:39question nobody cares about.
- 24:42So you have to worry about that a lot
- 24:44and you run into this problem precisely
- 24:46because you're starting with the
- 24:49experiment that you think will pass
- 24:51muster with journals instead of
- 24:53starting with the with what is the
- 24:56interesting and and and big question.
- 24:58All right, so now let me
- 25:00talk a bit about capital structure
- 25:03research as as an example of some of
- 25:05these these themes that that I'm talking
- 25:07about. So,
- 25:09um there's nothing particularly unique
- 25:12about capital structure research with
- 25:14respect to these identification
- 25:16problems. I'm choosing simply because
- 25:18I've been doing papers in this area, so
- 25:20I know this area a little bit more and
- 25:22you'll see in some of the examples I'm
- 25:24using in my own papers.
- 25:26It's not because I think these are the
- 25:28best papers in in that field, but I know
- 25:30these papers better and then they're
- 25:32representative of some of the points
- 25:34that that that I'm trying to make. But,
- 25:37there's some really fundamental
- 25:38questions in in capital structure
- 25:41research that amazingly enough after all
- 25:44these years we haven't done great job of
- 25:46answering them quite yet.
- 25:48We we don't really have
- 25:51a great idea of what are the primary
- 25:53determinants of
- 25:54of capital structure decisions. We have
- 25:56some pretty good theory
- 25:58about what sort of market frictions
- 26:00often matters,
- 26:02taxes, bankruptcy costs, agency costs,
- 26:06asymmetric information,
- 26:08uh but it's it's difficult. It's it's
- 26:11proven difficult to establish what
- 26:13factors are really first-order
- 26:14determinants of of capital structure
- 26:17decisions and part of the issue has
- 26:18always been
- 26:20uh proper identification. Now, another
- 26:22whole segment of this literature is this
- 26:24sort of hearing on static models
- 26:27versus dynamic models. I'm not really
- 26:29going to talk that much about that cuz
- 26:30it's not
- 26:31as directly relevant to to to topic
- 26:34today, but that's also a challenging
- 26:36aspect in this literature. Um
- 26:40but what you see and what what I want to
- 26:42talk about in in the rest of the time
- 26:45are a variety of different sort of
- 26:47research approaches in this field that I
- 26:50would argue are all very useful. They
- 26:53they all
- 26:54give us some information different
- 26:56aspects of the of the set of of the set
- 26:59of information that we need to
- 27:00ultimately draw some conclusions about
- 27:03about capital structure.
- 27:05Uh and only one of them
- 27:07falls in this category of of natural
- 27:10experiments. But
- 27:11what's important, I think, to to note is
- 27:14that even though all these are
- 27:16potentially useful,
- 27:18they all have different limitations,
- 27:20too.
- 27:21So, you're you're getting different
- 27:22pieces of information from them. So, you
- 27:26you can't view them as as
- 27:28being able to draw the same sort of
- 27:30inferences in terms of causation.
- 27:33Uh but you can view them sometimes as as
- 27:35telling you a little bit more about
- 27:37what's first order and what's not first
- 27:39order. And I'll I'll I'll be a little
- 27:40bit more precise about that in in a
- 27:42couple of minutes. So,
- 27:44So, the classic approach, of course, is
- 27:46just doing OLS type of regression.
- 27:48You've got leverage on off the hand
- 27:50side. You've got a set of your testable
- 27:53determinants on the right hand side and
- 27:55other control variables. Uh you might
- 27:57also
- 27:58estimate this using using panel
- 28:00regression techniques uh to try to
- 28:03control for fixed effects of some sort.
- 28:06And the literature has got tons of
- 28:08studies that that fit into this
- 28:11particular category. But there's proven
- 28:12to be two main problems. The main one is
- 28:15is the identification problem that we're
- 28:17we're primarily focused on. But as a
- 28:20practical matter, too, the second
- 28:21problem has been that it hasn't proven
- 28:24that fruitful. You know, we we we end up
- 28:26identifying some relatively vague
- 28:29factors that that seem to matter. You
- 28:32can say that leverage is associated with
- 28:34profitability,
- 28:36growth opportunities as measured by
- 28:38market-to-book ratios, usually. Uh firm
- 28:40size.
- 28:42That's great in a way, but when we think
- 28:44about relating that to capital structure
- 28:47theories, we find ourselves in this trap
- 28:50where it's consistent with multiple
- 28:52theories,
- 28:53uh even multiple classes of theories
- 28:56between static model, trade-off models,
- 28:58and pecking order models. They can be
- 29:00reconciled with this this set of
- 29:04observed determinants. And so, in the
- 29:06end,
- 29:07you feel like you don't learn that much
- 29:09at times with with with this literature.
- 29:12All right. So,
- 29:13the way the literature is progressed, I
- 29:15think, and this is this is a useful
- 29:17progression, is is to think about
- 29:19different types of empirical approaches
- 29:22that can shed some additional light.
- 29:24Now, one one type would be a natural
- 29:27experiment.
- 29:28All right. Can we Can we observe some
- 29:30exogenous shock to one of these
- 29:32hypothesized determinants while the
- 29:35others are held constant
- 29:37and see whether that shock seems to
- 29:39elicit a reaction from firms in in
- 29:42choice of capital structure.
- 29:44All right. That's one way to shed some
- 29:46light
- 29:47on
- 29:48a specific fact.
- 29:50All right. Another possibility
- 29:52is to conduct what's really more of
- 29:54descriptive data analysis type of of
- 29:57studies. Now,
- 29:59these next two, descriptive data and
- 30:01longitudinal types of studies, are are
- 30:03kind of similar in the sense that you
- 30:06would probably characterize them both,
- 30:08well, one certainly by the name, but
- 30:10both of them are are sort of
- 30:11descriptive.
- 30:14All right. And by that, I mean they're
- 30:16not generally set up as being specific
- 30:20hypotheses that are tested in sort of a
- 30:22traditional scientific way. All right.
- 30:25And and sometimes these kind of studies
- 30:27get a bad name. You see people use
- 30:29descriptive in a as a pejorative term.
- 30:33Oh, that study is just descriptive. Um
- 30:35but I would argue that this has a place
- 30:40in in the set of research approaches,
- 30:43uh and in particular in capital
- 30:45structure, I think it's been useful. I'm
- 30:46going to
- 30:47I'll give you a couple of examples of of
- 30:50how it they have been useful.
- 30:51Longitudinal studies are are similar in
- 30:54the sense of they they look over a long
- 30:56period of time whether capital structure
- 30:59has changed over that period of time and
- 31:01whether we can somehow connect that with
- 31:04changes in either firm or macroeconomic
- 31:07characteristics in a way that is
- 31:09consistent, you know, inconsistent with
- 31:11with what we think of as as the theory
- 31:14underlying it. So,
- 31:15the longitudinal models are are a little
- 31:18bit like natural experiments in that you
- 31:21you can have over a long period of time
- 31:23shocks to multiple
- 31:25possible determinants of capital
- 31:27structure. You can sort of think of it
- 31:29as sort of a horse race among these
- 31:31different factors as to what seems to be
- 31:33first-order and and and what's not.
- 31:36Another approach might be structural
- 31:38models,
- 31:39right, where we really try to get into
- 31:41the dynamics of of of
- 31:44capital structure policy over time. Can
- 31:47we specify some sort of objective
- 31:50function that the manager is maximizing,
- 31:53specify a set of exogenous determinants
- 31:56as well as a set of endogenous variables
- 31:58that be connected with uh with leverage.
- 32:01And so, in a sense, specify the nature
- 32:04of the endogeneity problem and then take
- 32:07this to the data through some structural
- 32:09estimation to see whether the data seem
- 32:11consistent with that uh with the
- 32:13predictions from that structural model.
- 32:16Lots of good examples of that in the
- 32:18recent literature as well.
- 32:20All right. And last and and certainly
- 32:22not least is that I think there's room
- 32:25in this literature for clinical or case
- 32:28study sort of approaches. Now, it's
- 32:30difficult, I think, to do a standalone
- 32:34clinical study.
- 32:35Uh I mean, years ago the JFE published
- 32:38lots of those. It's sort of backed away
- 32:40from that over time, but
- 32:43but I think clinical studies are very
- 32:45useful in conjunction with with some of
- 32:48the others. Now, I'll give you some
- 32:49examples of that how you might combine
- 32:52survey descriptive data analysis or
- 32:54longitudinal study with with uh clinical
- 32:57analysis of the subset of your data in a
- 32:59way that sheds a lot of light on what's
- 33:02on what's really going on. It really
- 33:04helps you tease out the identification
- 33:07issues as well.
- 33:09And so so let me give you
- 33:12a few different examples
- 33:14from the literature that that highlight
- 33:16the use of these different approaches.
- 33:18And good, so I want I want to ultimately
- 33:20make two points. One is that each of
- 33:21them are useful in shedding light on an
- 33:25aspect of of the capital structure
- 33:27problem.
- 33:29And then secondly, I'll make sure we
- 33:31understand that each of them also has
- 33:33limitations. Okay, so you have to be
- 33:35careful when you do these studies just
- 33:37to understand what you can and cannot
- 33:40conclude from the type of study that
- 33:42you've chosen to do.
- 33:44All right, so one approach as we
- 33:45mentioned is is a natural experiment.
- 33:48A good example of that are these studies
- 33:51that have that have looked at staggered
- 33:54changes in the income tax rates either
- 33:57across states within the US like the
- 33:59Hadlock paper or
- 34:01across different countries throughout
- 34:03the world like Lacho and and Shue at al.
- 34:07So it's sort of a classic natural
- 34:10experiment diff-in-diff kind of approach
- 34:12where you've got the change in in data,
- 34:14change in leverage on the left-hand side
- 34:18as a function of change in the tax rate,
- 34:20that's the T variable, set of
- 34:23firm-specific variables that denote you
- 34:26with with X and I's.
- 34:29Uh and and uh industry-specific
- 34:32variables as well.
- 34:34So, you're trying to control for all
- 34:35these other possible determinants of of
- 34:38leverage or CEO shock to tax rates at
- 34:42point in time,
- 34:44do you see firms respond
- 34:46uh with changes in leverage? All right.
- 34:48Now,
- 34:50as with most natural experiments, as I
- 34:52mentioned before, they're not pure
- 34:54natural experiments. So, there's there's
- 34:56still a little bit of endogeneity that
- 34:59you worry about in this. So,
- 35:01you might worry that somehow there's
- 35:03still some systematic differences that
- 35:05you can't control for between firms,
- 35:08say, within states that had a tax rate
- 35:10change versus those that didn't. So,
- 35:13you'll see clever clever ways in to try
- 35:16to get at that by looking right around
- 35:18the borders between states, look at
- 35:20counties that are right along the
- 35:21border, with the idea being that the
- 35:23economic conditions of of companies on
- 35:26either side of the border have got to be
- 35:28roughly
- 35:29identical, or borders of countries in
- 35:32the case of of of Argentina, issue.
- 35:35And so, you you So, you do your best
- 35:38to hold constant all the other possible
- 35:41economic determinants of this leverage,
- 35:43but one group has the shock tax rates,
- 35:46the other group did not have that shock
- 35:47tax rates, do we see a difference
- 35:50uh in leverage as a result of that of
- 35:53that shock to leverage. So, if done
- 35:57in in a
- 35:59in a correct manner, this type of study
- 36:01is really useful for getting at this
- 36:03causal connection between this factor,
- 36:06taxes,
- 36:07and capital structure change.
- 36:11What it can do,
- 36:13right, is tell you something about
- 36:14whether taxes are really a first-order
- 36:16consideration or not, because
- 36:19you by definition, if you've done study
- 36:21right, you've held all these other
- 36:23factors constant.
- 36:25And you've isolated on on tax rate. Now,
- 36:28and that's not a criticism of this type
- 36:30of study.
- 36:31I it's just
- 36:33a point that
- 36:35in doing this sort of study, you got to
- 36:36be clear on what it is your goal is.
- 36:39Like if you're going in this study
- 36:41saying, "Well, my question is I want to
- 36:43identify what determines capital
- 36:45structure." This isn't really doing
- 36:47that.
- 36:48All right.
- 36:50If your question though is, "Do taxes
- 36:53have an impact on capital structure at
- 36:55the margin?" Yeah, this is the way you
- 36:57want to do it. You you're able to
- 36:59isolate in on the causal impact of taxes
- 37:03on leverage decisions.
- 37:05It's very useful in that
- 37:07But it's a subset of the information
- 37:09that we're trying to put together in
- 37:11order to get a bigger picture
- 37:13of what's really happening with
- 37:16corporate capital structure decisions.
- 37:18All right, so that that's one
- 37:20one type of study.
- 37:22All right, so the other one of the other
- 37:23types of studies I talked about is is
- 37:25descriptive data analysis. Now,
- 37:28in some respects, this is
- 37:31a tricky way to go about doing empirical
- 37:35research because as I said before,
- 37:37you're not necessarily starting with
- 37:39very firm hypotheses, but rather it's a
- 37:42little bit more exploratory in nature.
- 37:44So, the goal ultimately in a study like
- 37:48this is to provide enough information to
- 37:51the reader that they that they somehow
- 37:53will think differently about capital
- 37:55structure than they did before. All
- 37:57right, if they don't feel that way,
- 38:00then you shouldn't have anything. You
- 38:01just described a bunch of data that
- 38:03everyone shrugs their shoulders and
- 38:05says, "Well, so what?" Okay.
- 38:08Um
- 38:09so, let me I'm going to give you a
- 38:11couple of examples of this type of work,
- 38:13one of which is by Donald Stiefel and
- 38:15Ian Dew-Becker in the RFS a couple of
- 38:16years ago.
- 38:18And so, here the the approach we took in
- 38:20this in this study
- 38:21is is to say, "Oh, let's let's get away
- 38:24from thinking about it in in sort of a
- 38:27standard regression framework where
- 38:28we're looking at a bunch of right-hand
- 38:30side variables determining what's on the
- 38:32left-hand side, which is leverage."
- 38:34Let's Let's instead let's start with the
- 38:37left-hand side.
- 38:39All right. Well, so let's try to observe
- 38:42some major discontinuity in the
- 38:44financing behavior of the firm and then
- 38:46back out what seems to be driving that
- 38:50that change in financing behavior. So,
- 38:52the way we went about doing this is to
- 38:54say, "Okay, well, let's let's first
- 38:55identify firms that have what we call
- 38:58proactive change in leverage." And by
- 39:01proactive, we simply mean that they
- 39:03actually issued debt in order to change
- 39:06their leverage as opposed to they bought
- 39:08back shares or had some other action
- 39:10that just ended up causing their their
- 39:12leverage to be different. We want them
- 39:14to be proactively issuing debt in a way
- 39:17that changes their leverage ratio. And
- 39:19the second aspect of that was that we
- 39:22were requiring that their resulting
- 39:24leverage ratio
- 39:26was at least 10% above some estimate of
- 39:29their target, where the target has been
- 39:31estimated using sort of standard
- 39:34empirical models of of of
- 39:36capital structure.
- 39:38All right. So, we're putting in this
- 39:40second requirement because
- 39:43we thought, "Well, this is this is sort
- 39:44of a a unique and challenging situation
- 39:47for the literature in the sense that
- 39:49we've now got firms that are
- 39:51deliberately taking an action that
- 39:53pushes them well away from what we think
- 39:56of as their target leverage ratio."
- 39:59All right. And so, we're just asking two
- 40:01very simple questions subsequently. Why
- 40:04did they do it?
- 40:05All right. What looks like is the
- 40:07underlying motivation
- 40:09for undertaking this leverage change?
- 40:11And then secondly, how do we see the
- 40:13leverage ratio evolve subsequently?
- 40:17Does that tell us something about
- 40:18whether they they seem to be targeting a
- 40:21particular
- 40:23leverage ratio. All right, so
- 40:25So, what do we find? So, well, first of
- 40:27all, what we tend to observe is that
- 40:29they seem to be doing this
- 40:32uh for reasons of of financing specific
- 40:36investment needs. Sometimes it's capital
- 40:38expenditures or R&D. Sometimes it's it's
- 40:41more working capital considerations,
- 40:43right? But, these don't appear
- 40:45that they're borrowing
- 40:47in order to say, for example, buy back
- 40:49shares.
- 40:50And it also doesn't appear that they're
- 40:53borrowing so that they can move to some
- 40:54new target leverage ratio.
- 40:57All right, but also the second main
- 40:59result is that when you look at what
- 41:01happens after this initial jump in
- 41:03leverage,
- 41:05like it it looks like they're starting
- 41:07to move back towards what we think is
- 41:09the target leverage ratio,
- 41:11but they're in no hurry at all to do it.
- 41:14All right, it's a really slow adjustment
- 41:16back uh
- 41:18uh towards the target, and they even
- 41:21though they have opportunities to be a
- 41:23little bit more proactive about it, and
- 41:25they could have moved back to the target
- 41:27quicker in some cases, they don't.
- 41:30And you you sort of see it drifting back
- 41:33down. And what's kind of interesting too
- 41:34is that when we see subsequent
- 41:36situations where they seem like they
- 41:38have a need for funds, they go out and
- 41:40borrow again. Even though they're still
- 41:42well above their target, they get even
- 41:44further above their target. They don't
- 41:46know.
- 41:47All right, so
- 41:49what does all that mean? How is how is
- 41:51this then a useful study in capital
- 41:54cycle? Some of you might think it isn't
- 41:56useful study in the capital cycle. Why
- 41:58do we think it's it's it's a useful uh
- 42:01study to do? Well, well, first of all,
- 42:03if you if you think about it,
- 42:05what this seems to imply that
- 42:09this managing towards some sort of
- 42:11stationary or static target leverage
- 42:13ratio doesn't seem to be a first order
- 42:15concern for managers. They don't seem to
- 42:17be paying that much
- 42:20Well, that's kind of interesting in the
- 42:22sense of
- 42:23what we think of in terms of capital
- 42:25structure models. Many of them are sort
- 42:27of stationary static target sorts of
- 42:30models. And this is kind of saying maybe
- 42:33we shouldn't be thinking along those
- 42:35lines in terms of theoretical capital
- 42:37structure. And the second aspect of it
- 42:40that it seems to be pointing
- 42:43towards financing and investment needs
- 42:45and then the subsequent evolution of
- 42:47cash flows
- 42:49as being more first order drivers of the
- 42:51dynamics of leverage through time.
- 42:55All right. Now, let's be clear that, you
- 42:57know, when you do a study like this,
- 43:01we in no way can claim some causation in
- 43:04any of the anything that I just
- 43:06described to you. You know, and we have
- 43:07to be really careful as we're as we're
- 43:09laying out these results, that we're not
- 43:12at all making any such claims
- 43:14about causation. All right. But why
- 43:17might it still be useful? What in a
- 43:19sense what we're trying to do is we're
- 43:20sort of
- 43:22reducing the the scope of the models
- 43:24that are really plausible models for how
- 43:27firms really behave with respect to
- 43:29capital structure. And if we can do
- 43:31that,
- 43:32and if we can do that sufficiently,
- 43:35uh then we feel like, well, that's a
- 43:37contribution that that is useful to the
- 43:40capital structure literature.
- 43:41Fortunately for us, a referee agreed
- 43:44that
- 43:45uh it was it was useful enough.
- 43:47All right. Now, a similar sort of
- 43:50example
- 43:51uh is this recent paper by DeAngelo and
- 43:53Roll that um I don't think it's out yet
- 43:56in the in Journal of Finance. That's
- 43:57that's forthcoming
- 43:59in Journal of Finance. So,
- 44:01what they do at a very basic level is is
- 44:04is fairly simple. They're really just
- 44:06describing
- 44:08the variation in leverage ratios
- 44:11within individual firms through time.
- 44:14Now,
- 44:15why should you do that? I mean, why
- 44:17would you start a study like that?
- 44:21If you think you're going to contribute
- 44:23to the capital structure literature.
- 44:24Well, the the idea
- 44:26is that most of the models and most of
- 44:29our most of the thinking, I think,
- 44:31within the profession through time has
- 44:33been that capital structure within firms
- 44:36is reasonably stable
- 44:38through time.
- 44:40Right? And where you really get a lot of
- 44:41variation is in the cross-section.
- 44:44Right? And that's kind of reflected in
- 44:46the type of work that you see in the
- 44:48capital structure literature. That most
- 44:50of the models and most empirical work
- 44:52are are designed to try to explain that
- 44:54cross-section.
- 44:56Why the cross-sectional variation? Well,
- 44:58that's not necessarily a good approach
- 45:01if what's really going on is that a lot
- 45:04of variation is happening within the
- 45:05firm over time. If that's just as
- 45:07important as the cross-section, then we
- 45:09sort of got to rethink
- 45:11what our models are really saying and
- 45:13how we're approaching the empirical
- 45:14works. That's sort of where this where
- 45:17the study is headed from the outset.
- 45:19Again, the idea is it's descriptive data
- 45:22analysis that's trying to describe the
- 45:24data in a way that's going to cause
- 45:26people to rethink
- 45:28how they're approaching capital
- 45:30structure and if they're finding
- 45:31something that's unique enough
- 45:33that does cause people to do this
- 45:35rethinking of
- 45:37how capital structure is determined,
- 45:39this is a useful way to to approach the
- 45:42problem. Now, what did they find? Well,
- 45:43first of all, they do find
- 45:45uh substantial
- 45:47instability in the leverage ratios of of
- 45:50individual firms
- 45:52through time.
- 45:54Right? A lot of variability. And
- 45:56surprisingly, I think, to a to a lot of
- 45:58people, a ton of variability in
- 45:59individual firms. They do find episodes
- 46:03in which firms look like they have
- 46:05relatively stable leverage ratios, but
- 46:08those are pretty limited
- 46:10to time periods in which leverage level
- 46:13is really quite low, often times zero.
- 46:16So, you see low stability where some
- 46:18firms have no debt and they keep no debt
- 46:20for a while or really low debt, they
- 46:22keep that for a while.
- 46:24Right? And that's when you tend to see
- 46:25stability. When leverage is higher, it
- 46:27tends to be bouncing all over the place.
- 46:29Now, why does it bounce all over the
- 46:30place? Well, they find like like we did
- 46:33in a paper of Steve McKeon and I, they
- 46:35find it seems to be connected with the
- 46:38episodes in which there are strong
- 46:40investment needs. Right? And this is
- 46:42where they sort of will they they do a
- 46:44really nice job
- 46:46of combining some clinical or case-based
- 46:48evidence to complement what they're
- 46:50doing on on a large-scale basis in in
- 46:53terms of documenting these stability.
- 46:55Right? They they dig into the data and
- 46:58identify well, what seems to be the
- 47:01underlying cause for these individual
- 47:03firms for
- 47:04for increasing their leverage when they
- 47:06see an increase in or decrease in when
- 47:08they see it go down. And and they find
- 47:10pretty strong evidence that it seems to
- 47:12be associated first with the company
- 47:15expansions
- 47:16uh during the changing economic
- 47:18circumstances for those firms and then
- 47:20subsequently
- 47:22uh reduction in leverage in those time
- 47:24periods in which it it it's actually
- 47:28uh either excess cash flows or they're
- 47:30producing other assets that they
- 47:31transfer.
- 47:33Right? So, again, I think this is very
- 47:35useful type of evidence in this approach
- 47:37even though it looks like a very simple
- 47:40data description approach at the at the
- 47:42outset, it's carefully done in a way
- 47:45that ultimately when you're done with
- 47:46it, you're thinking, "Well, I got to
- 47:48rethink the way I'm thinking about
- 47:50capital structure as a result of this."
- 47:52So, it's very informative to subsequent
- 47:54theory because now any theory that comes
- 47:57along that purports to be modeling how
- 48:01capital structure is actually chosen has
- 48:03to deal with with this sort of facts
- 48:05that are out there, these stylized facts
- 48:07that that say that the firm's action
- 48:15Okay, that's sort of that's sort of what
- 48:17one is summarizing here. There's
- 48:18incredible theories of capital structure
- 48:21that are going to have to deal with this
- 48:22time series variation and not just focus
- 48:25on some construction anymore.
- 48:27It also is implying that
- 48:30sort of contrary to probably what most
- 48:32of us teach in in the classroom with
- 48:34respect to capital structure that the
- 48:37leverage per se is really kind of a
- 48:39second order importance
- 48:41to valuation.
- 48:42And the firm's behavior is that they
- 48:44don't really care that much what their
- 48:46leverage ratio is at any point in time.
- 48:49You know,
- 48:50if you're like me, you're teaching in
- 48:52the classroom you're oftentimes teaching
- 48:54a sort of a
- 48:55you know, static tradeoff sort of model
- 48:56where you're optimizing at this point
- 48:59that
- 49:00is the capital structure that maximizes
- 49:02the value. This kind of implies that
- 49:04either that that nice
- 49:07inverted U-shaped curve that we draw up
- 49:09on on the whiteboard for capital
- 49:12structure is either really flat
- 49:15or it's total nonsense. One of one one
- 49:17of the three.
- 49:20And third, then obviously then what
- 49:22seems to be true then is that the main
- 49:24determinants of leverage ratios
- 49:27must be factors that are a little
- 49:29different than what we traditionally
- 49:30think of as standard measures of taxes
- 49:33and distress costs that are their
- 49:35tradeoff.
- 49:38Okay, so let's move on to another
- 49:40another different type of study. So
- 49:42those last two are kind of descriptive
- 49:45data analyses that
- 49:47you know, hopefully I've convinced you
- 49:48that they they shed some useful light
- 49:51on the capital structure puzzle. Another
- 49:53way to go about this
- 49:55uh is to do a longitudinal type of
- 49:58study. And again, the
- 50:00the potential merits of a longitudinal
- 50:02study is that you've got this long
- 50:04period of time where lots of things
- 50:06might be changing. Lots of the
- 50:08underlying factors that could determine
- 50:10capital structure can change over a long
- 50:13period of time, and we can see whether
- 50:15capital structures tend to change in the
- 50:17direction that we think they should
- 50:19based on those changes and and
- 50:22underlying fundamentals. So, this study
- 50:24by Graham, Leary, and Roberts that's
- 50:26coming out
- 50:27uh in the JFE is a good example of that.
- 50:30I mean, they're really looking at
- 50:31leverage ratios over nearly a century
- 50:35uh within the US. So, you've got a
- 50:37period of time that's spanning
- 50:39uh World War II, uh the
- 50:43uh macroeconomic expansion that takes
- 50:45place after World War II in the US, high
- 50:48inflation period in the '70s in the US,
- 50:51and most recently the financial crisis.
- 50:53All this gets spanned
- 50:56uh in their study, and they can ask then
- 50:58some fair fairly simple questions. You
- 51:00know, first, have capital structures
- 51:02changed at all on average over that
- 51:04period of time?
- 51:05Um secondly, can our existing models
- 51:09account for those changes? Are they
- 51:11consistent with those those changes?
- 51:14And then third, if not, and the answer
- 51:16does seem to be no in this case,
- 51:19what underlying forces seem to be
- 51:21driving this this variation? So, so
- 51:23again, when you when you approach the
- 51:26question in this way, the the result
- 51:29that you're going to get
- 51:31um you know, unlike maybe you find that
- 51:33everything's consistent with what we
- 51:34thought. You usually don't find that
- 51:36kind of thing in reality. So, if you
- 51:38don't, all right, your your end result
- 51:41is to present a set of facts that are a
- 51:44challenge to existing theory. If they're
- 51:46going to inform theory, it's going to be
- 51:48useful in that sense.
- 51:50All right, and it's also going to narrow
- 51:52the set of of theories that are truly
- 51:55credible.
- 51:56Any any theory that purports to explain
- 51:58capital structure is going to have to
- 52:00account for this this time series
- 52:02variation in leverage ratios that
- 52:04doesn't seem to be accounted for by our
- 52:07our standard factors.
- 52:09So, what do they find? All right, they
- 52:11What they find is first of all that
- 52:12there's been
- 52:14a fairly large increase in leverage over
- 52:16time. If you go back to sort of the
- 52:18post-World
- 52:20War I period and look at average
- 52:22leverage leverage ratios back then,
- 52:25uh I think
- 52:26I can't remember the magnitudes, but
- 52:28they're nearly double now what they were
- 52:30back then. All right?
- 52:32Coincident with that, we see a fairly
- 52:35large decline in cash holdings.
- 52:39Pronounced as they suffered an increase
- 52:40in leverage. Now, that may be surprising
- 52:42to some because a lot of what you put
- 52:45you read in the press and even in some
- 52:47our studies in corporate finance in
- 52:49recent years talk a lot about these huge
- 52:51cash balances that that US firms have
- 52:54now.
- 52:55That's a more recent phenomenon. And
- 52:57these cash balances actually were
- 52:59enormous back in the early part of the
- 53:02of the last century, and they declined
- 53:05quite a bit over the period of time as
- 53:07these leverage ratios have
- 53:09have gone up. It's only recently that
- 53:11they their sort of cash balances have
- 53:14turned back up.
- 53:15All right.
- 53:16What's interesting now is that firm
- 53:18characteristics that we normally
- 53:20associate
- 53:21with capital structure on average don't
- 53:24really change significantly in a way
- 53:27that's consistent with these firms
- 53:29having greater debt capacity.
- 53:32So, think of our traditional
- 53:34determinants of leverage. They don't
- 53:36account for the fact that leverage
- 53:37ratios are so much higher now than they
- 53:40were, you know, back in 1920 or so.
- 53:44All right. What's also interesting, I
- 53:46think, is that they do find a negative
- 53:49association between corporate borrowing
- 53:53and government borrowing
- 53:55in the US. And what sort of gives some
- 53:57hints at the possibility that there's
- 54:00there's there's some crowding out that's
- 54:02taking place. Government is borrowing a
- 54:04lot. Uh that's crowding out the
- 54:07corporate sector from from borrowing at
- 54:09at reasonable rates. Um
- 54:11So, again, once once they're done with
- 54:14this, have they established any
- 54:16causation between variables and and
- 54:18leverage? No.
- 54:20They're not claiming any such causation.
- 54:22Even with their last result of the
- 54:24government crowding out, the most
- 54:26they're saying is that, you know,
- 54:28there's some hints of that in the data.
- 54:30All right. Which I think is useful in
- 54:32terms of subsequent studies that that
- 54:34that could potentially examine that
- 54:36question in in more detail. So, you
- 54:39know, these aren't This is again not a
- 54:41study that has
- 54:43great identification of anything.
- 54:45All right. But, I would argue it's very
- 54:47useful to us as as scholars that who are
- 54:50interested in in capital structure cuz
- 54:53it's it's again narrowing the range of
- 54:55plausible explanations for for what's
- 54:57going on. You finish this paper, you
- 55:00learn something that you didn't already
- 55:02know
- 55:03about capital structure. That's the
- 55:05definition of a contribution.
- 55:09All right. Now, the last thing I'll talk
- 55:11about is again one of my own and I
- 55:13apologize in advance for for
- 55:16talking about another study of mine.
- 55:18Again, I'm not representing these as the
- 55:20best studies. I'm just
- 55:22representing numbers as representative
- 55:24studies of a particular approach. So,
- 55:26this is This is another longitudinal
- 55:29type of study that I've been working on
- 55:31with co-authors Lance Barrett and and
- 55:33and Ken Lane.
- 55:35In which we're studying capital
- 55:36structure decisions of US firms over a
- 55:39period of time that actually predates
- 55:41the Graham and Harvey and Roberts paper.
- 55:43Our our period of time is 1905
- 55:46to 1924. And I'll apply why I want to
- 55:49study capital structure back in
- 55:52the dark ages like that, especially if
- 55:55if you've ever looked at data back then.
- 55:56It's it's a little challenging to
- 55:59to look at financial data for a period
- 56:02that predates the SEC. It's It's
- 56:04standard They standardization of data is
- 56:06not particularly good
- 56:09back at that time. But,
- 56:11we thought, "Well, this is a very useful
- 56:13time period to take a look at for
- 56:15capital structure theories for two
- 56:17reasons." And one is that it spans the
- 56:19period of time
- 56:21in which the US first introduced
- 56:23corporate and personal income tax rates.
- 56:25So, prior to 1909, I believe it was,
- 56:29there were no corporate income taxes.
- 56:31There were no personal income taxes,
- 56:33either, in the US. Great days, but
- 56:37not anything any of us have experienced.
- 56:40So, if taxes are really important, then
- 56:42we ought to see it when taxes are
- 56:44introduced. Was sort of the logic of of
- 56:46this. At the same time, this spans this
- 56:50period of time in which the US enters
- 56:53World War I,
- 56:55which we show in the paper is associated
- 56:57with very large but short-term shock to
- 57:00investment opportunities for for the US
- 57:03firms.
- 57:04Right? And what's what's neat, we think,
- 57:06from
- 57:07a
- 57:08theoretical standpoint is that that
- 57:10shock to investment opportunities is
- 57:12plausibly quite exogenous. No one's
- 57:14really anticipating the outbreak of
- 57:17World War I.
- 57:19Right? And so, we can see how firms
- 57:21react
- 57:23to the shock to investment opportunities
- 57:25at the same time as they're we're
- 57:28observing how they react to fairly large
- 57:31changes in the personal and corporate
- 57:34income tax rates because one of the
- 57:35other things I won't get into the
- 57:37details, but one other things that you
- 57:39see when the war breaks out is that the
- 57:42US imposes very large excise taxes on on
- 57:46the firms, particularly the firms that
- 57:48would potentially benefit from from the
- 57:50outbreak of of the war. So, there's this
- 57:52very large change in marginal tax rates
- 57:55that actually gives a strong incentive
- 57:57to equity financing during the war. And
- 58:00so, we're able in some sense to have a
- 58:02bit of a horse race going on between tax
- 58:05effects and
- 58:06dynamic investment effects. All right.
- 58:08So, what we find
- 58:10in the paper is that there's really very
- 58:12little evidence that shocks to tax rates
- 58:15have much of an impact at all
- 58:17on the leverage ratios.
- 58:19You see these corporate
- 58:21income taxes initially imposed,
- 58:24basically nothing happens to the average
- 58:26leverage ratio.
- 58:28All right. But, what we do see, which is
- 58:30quite similar actually to the DeAngelo
- 58:32Roll paper, is that even though average
- 58:35leverage ratios aren't changing very
- 58:37much,
- 58:38uh individual firm leverage ratios are
- 58:41bouncing all over the place during this
- 58:44period of time. And we're able to link
- 58:46it pretty tightly
- 58:47with the evolution of investment
- 58:49opportunities and and cash flows in
- 58:52these firms over this over this period
- 58:54of time.
- 58:55All right. So,
- 58:56so our interpretation is kind of similar
- 58:59to the DeAngelo Roll argument that, you
- 59:02know, it looks like the dynamics and
- 59:04investment opportunities and cash flows
- 59:06are what's really driving the leverage
- 59:08ratios rather than some of the
- 59:11traditional determinants of of of
- 59:14leverage. All right. But,
- 59:16you know, in trying to interpret this,
- 59:19there are challenges, some of which are
- 59:21closely related to the identification
- 59:23challenges that that we've been talking
- 59:25about. One is that when you look at when
- 59:28you're doing a historical longitudinal
- 59:29study, if you're going to focus in on
- 59:32specific shocks that interest you, you
- 59:34have to be aware of the fact that
- 59:36they're not the only shocks that are
- 59:37going on
- 59:38during this period of time. So, in our
- 59:41case, there's sort of a
- 59:43a market panic that takes place in in
- 59:451907.
- 59:47The Federal Reserve is created in 1913.
- 59:52And after the the end of World War I,
- 59:55there's actually a pretty strong
- 59:57recession that worries on on the on the
- 1:00:00depression
- 1:00:02during that period of time. All of which
- 1:00:05could potentially have some impact on
- 1:00:07observed leverage ratios. So, even
- 1:00:09though we can find some evidence that's
- 1:00:11consistent
- 1:00:13uh with this investment shock associated
- 1:00:16with World War I being a causal
- 1:00:18determinant of of the leverage changes,
- 1:00:21we can't necessarily point to that
- 1:00:23unless we do something more. I mean, so
- 1:00:25so where we are in the study now is sort
- 1:00:27of trying to supplement what we've got
- 1:00:30with a little bit more of a clinical
- 1:00:32analysis of individual companies. That's
- 1:00:35involved actually collecting the annual
- 1:00:38reports from around 1917 for these 60 or
- 1:00:42so firms that that comprise our sample.
- 1:00:45So, as you might expect, they're they're
- 1:00:47not in most libraries. So, you know, you
- 1:00:50got to go find in our case I go to
- 1:00:52Columbia University Library, get these
- 1:00:55all copied. You know, it's a huge stack
- 1:00:58of annual reports that that we've got to
- 1:01:00go through. But it's a very useful way
- 1:01:03to try to tease out what's really
- 1:01:05driving these changes in leverage. Cuz
- 1:01:07now when you read these annual reports
- 1:01:09and the narrative that's in the annual
- 1:01:11reports, like you can actually see that
- 1:01:14the firms are saying specifically, you
- 1:01:17know, we're borrowing in order to do
- 1:01:19this.
- 1:01:20All right? And that way, you are getting
- 1:01:22directly at the issue of both causation
- 1:01:26and the data by by using a supplementary
- 1:01:29clinical type of of evidence. You don't
- 1:01:31have a clean natural experiment. You
- 1:01:33don't have instrumental variables. I but
- 1:01:36looking in a clinical sense, you are
- 1:01:38getting at some some of the same issues.
- 1:01:40And in fact, I I would encourage you to
- 1:01:42think about
- 1:01:44sort of clinical sort of analysis in in
- 1:01:47in the same way that you would think of
- 1:01:48as a natural experiment. And both of
- 1:01:51them are identifying sort of a narrow
- 1:01:54situation in which you can fairly
- 1:01:57cleanly identify what's really going on.
- 1:02:00The natural experiment does it one way.
- 1:02:03The clinical study does it it does it
- 1:02:05another way. A structural model, for
- 1:02:06that matter, does it yet another way.
- 1:02:08It's a very specific set of conditions,
- 1:02:10this what this is what happens.
- 1:02:12All these are different ways of
- 1:02:15obtaining the same piece
- 1:02:17in in essence.
- 1:02:19All right, so let me let me just kind of
- 1:02:20conclude at this point and then open up
- 1:02:22to to questions. So,
- 1:02:24the points I want to make are first of
- 1:02:26all that
- 1:02:27these identification problems aren't
- 1:02:29going away.
- 1:02:30They're pervasive in corporate finance
- 1:02:33research. They're
- 1:02:35much more so in corporate finance
- 1:02:37research than in some other aspects of
- 1:02:40of finance research. They're there.
- 1:02:42They're real problems. They're things we
- 1:02:44have to deal with as as scholars in
- 1:02:46order to be able to tease out some sort
- 1:02:49of believable conclusion. Econometrics
- 1:02:52can help and has
- 1:02:53really helped. I think we've got great
- 1:02:55advances
- 1:02:57in our understanding of econometrics,
- 1:02:59like I said, over the last 10 to 15
- 1:03:01years. And they really help us narrow
- 1:03:03down the scope of this problem. They can
- 1:03:05never solve the problem. Econometrics
- 1:03:07doesn't make the identification problem
- 1:03:09go away.
- 1:03:11It just sort of narrows the the scope.
- 1:03:14Natural experiments are in some ways the
- 1:03:16most useful econometric solution.
- 1:03:19Um, but but I worry that an
- 1:03:21over-reliance on these natural
- 1:03:24experiments is it can be harmful
- 1:03:27in our pursuit of knowledge in the sense
- 1:03:29I described before, that it really kind
- 1:03:31of
- 1:03:32it it if you think you have to have a
- 1:03:33natural experiment, you are constraining
- 1:03:36yourself in the types of questions that
- 1:03:38you can really ask. You know, if the
- 1:03:39profession as a whole were to do that,
- 1:03:42then we're constraining the body of
- 1:03:43knowledge that that we're producing
- 1:03:46about questions in in uh in
- 1:03:48in corporate finance. So,
- 1:03:51I'd like to encourage you to think about
- 1:03:52these alternative methodologies that
- 1:03:54that we went through as as equally
- 1:03:56useful useful ways of of expanding that
- 1:04:00uh body of knowledge, even if they're
- 1:04:03not able to really provide complete
- 1:04:05identification. And they frequently
- 1:04:07don't. Uh
- 1:04:09but I would argue that's okay. As long
- 1:04:11as you understand and can explain the
- 1:04:14limitations of of what you're doing, you
- 1:04:16can still produce papers that are are
- 1:04:19very useful in terms of our
- 1:04:21understanding of the problems.
- 1:04:23All right.
- 1:04:23So, let me stop there and
- 1:04:25open up to any questions.
- 1:04:29You correct that he's using the
- 1:04:31simultaneous model, but that doesn't
- 1:04:33mean
- 1:04:34it it there's some other omitted
- 1:04:36variable he was trying to specify.
- 1:04:39Thank you.
- 1:04:45Anything else?
- 1:04:50Okay, then. Well, thanks for listening
- 1:04:52and uh yeah.
About this transcript
This page contains the full transcript of YouTube transcript (DKWVDvk1QYo) , generated from the public captions YouTube serves with the video. The transcript has 10,446 words across 1,734 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.