Popular in Professional

Culture and Society

The City as a Growth Machine: Toward a Political Economy of Place

New York University Harvey Moltoch

Education

The Role of Education in Reducing Racial Inequality

University of Southern California Pedro A. Noguera

Law

Diversity and Social Mobility in the UK legal profession

Law Society I. Stephanie Boyce

Internet Law

The Law of AI for Good

University of San Diego Orly Lobel

Pharmacy

Cytotoxic Activity and Anti-Migratory Effect of Berberine-Phytantriol Liquid Crystalline on Non-Small-Cell Lung Cancer

University of Technology Sydney Kamal Dua

Human Rights Law

Transforming Restorative Justice for Transitional Settings

University of Nottingham Kerry Clamp

Education

Doing Organizational Identity: Earnings Surprises and the Performative Atypicality Premium

View more insights in Professional - Business - Organisational Behaviour

View more figures from Stanford University

View the page of Amir Goldberg

Read the Study

How do organizations reconcile the cross-pressures of conformity and differentiation? Existing research predominantly conceptualizes identity as something an organization has by virtue of the products or services it offers. Amir Goldberg argues that organizational members’ interactions with external audiences also dynamically produce identity.

Image courtesy of interviewee. November 28, 2023

Generating Workspace

Export as PDF

This video video discusses the complexities of understanding organizations and how information about them is presented, particularly in the context of quarterly earnings calls. It highlights that while organizations are critical for production and investment, there's a lack of transparency regarding their internal workings. The discussion draws upon Elon Musk's acquisition of Twitter as an example of exposing organizational processes, though not necessarily representative of typical operations.

The analysis focuses on how investors and society infer and categorize organizational performance based on the information released during earnings calls. It emphasizes the distinction between categorical and performative conceptualizations, akin to gender studies, where perceptions and expectations about behaviors influence evaluations of organizations.

The approach includes a linguistic analysis using natural language processing to explore how executives communicate and how their language aligns with their industry peers. The findings suggest that atypical communication styles, like those of unconventional leaders such as Musk, can lead to overvaluation by the market, as analysts may misinterpret their unique styles as indicators of innovative strategies. This video video phenomenon can lead to negative surprises when performance does not meet expectations.

Overall, the text underscores the importance of developing better methods to discern reliable information from misleading signals in the context of organizational performance. The aim is for this research approach to be applicable beyond firms to other entities, addressing the challenges posed by information overload in contemporary society.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.235 --> 00:04.485
Organizations, constitute the most important process

00:04.865 --> 00:07.605
by which, things are being produced in the world.

00:08.225 --> 00:12.165
Uh, but, there's so much

00:12.395 --> 00:14.245
that we don't understand about them

00:14.525 --> 00:15.685
relative to what we do understand.

00:15.785 --> 00:18.805
We see what, we see how they're presented on the

00:18.805 --> 00:20.565
outside, but we don't really know

00:20.825 --> 00:22.605
how the sausage is made, so to speak.

00:22.605 --> 00:26.485
You know, think about like Elon Musk's takeover,

00:27.275 --> 00:30.565
over Twitter, that kind of exposed a little bit of

00:30.565 --> 00:31.805
how the sausage is made.

00:31.905 --> 00:35.965
But, not all of his antics are, are good representations

00:35.965 --> 00:38.285
of what is going on inside the organization.

00:39.185 --> 00:43.965
And given that so much, capital is invested,

00:44.545 --> 00:45.685
in these entities,

00:46.105 --> 00:48.685
and, there's so little information about

00:48.685 --> 00:49.725
what is going on in them.

00:49.845 --> 00:51.405
I think it's really important for us to understand

00:51.545 --> 00:55.005
how investors make inferences about,

00:55.215 --> 00:56.645
about organizations,

00:56.705 --> 00:59.445
but more broadly, how society at large understands

00:59.445 --> 01:02.845
what happens inside them, how it reaches conclusions,

01:03.505 --> 01:06.085
about, which,

01:06.635 --> 01:09.845
organizations are performing well, which aren't,

01:09.915 --> 01:11.525
whether it is from a financial,

01:11.525 --> 01:12.725
a purely financial perspective,

01:13.345 --> 01:15.365
but also I think, you know, our study

01:15.385 --> 01:17.725
as we'll discuss in a moment, I think has ramifications

01:17.725 --> 01:19.245
beyond the financials per se.

01:19.825 --> 01:21.845
We need to understand what happens in organizations

01:21.845 --> 01:25.245
because, they have broader societal implications.

01:25.465 --> 01:27.685
Uh, and in light of the fact that we understand so little,

01:27.745 --> 01:30.125
the main motivation question is what can we learn?

01:30.625 --> 01:31.725
And, and when,

01:31.955 --> 01:35.085
when organizational leaders talk about their organizations,

01:36.185 --> 01:40.485
is the information that they divulge, informative?

01:41.105 --> 01:44.365
And do we make, and do people make the correct inferences on

01:44.365 --> 01:45.605
the basis of that information?

01:45.825 --> 01:48.805
Uh, our analysis focuses on quarterly earning calls,

01:48.805 --> 01:50.325
which is, publicly traded.

01:50.685 --> 01:52.125
Companies are not required,

01:52.465 --> 01:55.005
but they normally, meet on a quarterly basis

01:55.035 --> 01:57.925
with analysts and, and, and provide divulge information.

01:57.945 --> 01:59.885
Now, as you would imagine, this is a very scripted

02:00.545 --> 02:02.405
and very choreographed experience.

02:03.065 --> 02:04.525
Um,

02:04.905 --> 02:07.285
and the general literature that we speak to is,

02:07.345 --> 02:09.245
is twofold.

02:09.385 --> 02:10.805
One is, is the literature

02:10.925 --> 02:13.485
that has focused on these quarterly earning calls

02:14.145 --> 02:18.805
and generally the kind of, formal, mechanisms

02:18.805 --> 02:21.685
through which organizations provide information to investors

02:21.825 --> 02:25.805
and how reliable that information is, how to investors,

02:26.225 --> 02:29.805
and analysts and, act on this information, et cetera.

02:29.825 --> 02:33.445
So that's one, body of of literature.

02:33.825 --> 02:34.965
Uh, a second body

02:34.965 --> 02:39.245
of literature is a more broader social scientific literature

02:39.245 --> 02:41.765
ranging from psychology to sociology

02:41.785 --> 02:45.845
and economics, on how people make inferences,

02:46.985 --> 02:51.605
on other entities, be them other humans,

02:52.015 --> 02:53.405
other organizations.

02:53.955 --> 02:56.445
There's a broad literature that's specifically focused

02:57.025 --> 02:58.525
on organizations per se.

02:59.105 --> 03:02.725
Uh, there's a broad literature, in organizations known

03:02.985 --> 03:05.125
as quote unquote, the categories literature,

03:05.415 --> 03:09.645
which is focused on, how do people

03:10.555 --> 03:12.525
basically enforce some sort

03:12.525 --> 03:15.005
of cognitive order on this broad space

03:15.945 --> 03:18.205
of the organizational landscape.

03:18.785 --> 03:22.925
Um, and, and we know that, all of us humans,

03:23.225 --> 03:26.725
we have this amazing, a machinery,

03:26.725 --> 03:30.645
called our cognition, but that uses a lot of heuristics.

03:31.225 --> 03:34.765
Um, and we have these conceptual maps in our

03:34.765 --> 03:35.805
heads that we hold.

03:35.905 --> 03:39.485
And when we are encountered with new information, we try

03:39.485 --> 03:43.165
to understand it with respect to these maps,

03:43.165 --> 03:45.285
or you can think about them as cognitive filters.

03:45.465 --> 03:46.845
Uh, we see something in the world.

03:46.865 --> 03:48.245
We say, okay, this is a tree.

03:48.245 --> 03:50.525
This is a human, this is a car.

03:51.065 --> 03:55.405
Uh, and we have a lot of conceptual associations around

03:55.635 --> 03:56.725
that focal concept,

03:57.425 --> 04:00.045
and that guides us in a very automatic way in

04:00.045 --> 04:01.485
how we, act.

04:02.065 --> 04:04.485
And there's been a lot of research in, in demonstrating

04:04.485 --> 04:05.925
that people also have these

04:06.435 --> 04:08.085
categorical filters when they try

04:08.085 --> 04:09.405
to understand organizations.

04:09.465 --> 04:14.285
So, for example, if you go into a store, you have a concept

04:14.425 --> 04:16.045
of what a store is,

04:16.865 --> 04:21.125
and that very much kind of structures your expectations

04:21.385 --> 04:23.485
and how you will behave,

04:23.505 --> 04:26.525
and a store that, doesn't,

04:26.915 --> 04:29.405
look like a a traditional store confuses you.

04:30.105 --> 04:32.405
The literature, you know, we make the distinction,

04:32.405 --> 04:35.565
and I think there's an, a novel distinction that we make in,

04:35.585 --> 04:38.525
in, in this particular paper, is that we make kind of the,

04:38.945 --> 04:41.325
we distinguish categorical

04:41.385 --> 04:44.965
and performative, conceptualizations.

04:45.545 --> 04:49.525
Um, and actually, I think the best way

04:49.525 --> 04:52.925
to understand it is, is the metaphor that we, we kind

04:52.925 --> 04:54.605
of got into this,

04:55.315 --> 04:57.365
terminology was from research on gender.

04:58.145 --> 05:02.685
Um, so 30 years ago already rasman and Faust

05:02.745 --> 05:04.285
and introduced the idea of doing gender.

05:04.305 --> 05:05.525
And, and the idea was the following.

05:05.525 --> 05:09.685
So in a very kind of traditional binary sense of gender

05:09.685 --> 05:13.805
where there are women and men,

05:13.905 --> 05:15.845
and it is one of the most fundamental categories

05:15.845 --> 05:17.245
for which we perceive other humans

05:17.345 --> 05:19.325
as we first associate them with gender.

05:20.105 --> 05:21.645
Um, what Vasser men

05:21.645 --> 05:23.565
and FA said was that it's not just

05:23.565 --> 05:26.245
that people are given a categorical identity, is

05:26.245 --> 05:28.045
that they need to continuously perform it.

05:28.145 --> 05:32.445
So, there are ways

05:32.545 --> 05:35.845
by which we kind of make distinctions, between men and

05:35.865 --> 05:38.285
and women on, on the basis of facial feature

05:38.345 --> 05:39.605
and other physical features.

05:40.305 --> 05:44.325
Uh, but you know, the most, dominant features,

05:44.325 --> 05:46.485
are usually in, in modern society are

05:46.485 --> 05:47.645
hidden from, from view.

05:48.265 --> 05:52.365
Um, and as a consequence, we also expect people

05:52.425 --> 05:55.005
to perform their identities,

05:55.225 --> 05:57.005
to wear their hair in a certain way,

05:57.185 --> 05:59.725
but they're also feminine and masculine ways of behavior.

06:00.425 --> 06:01.565
Um, there are feminine

06:01.745 --> 06:04.525
and masculine, ways of kind of carrying your body.

06:04.985 --> 06:08.045
And when people fail to perform in a way that's consistent

06:08.045 --> 06:11.205
with our expectation, that research has shown

06:11.205 --> 06:14.205
that people get confused and dismayed,

06:14.945 --> 06:18.645
and they're kind of start questioning, is this a real male

06:18.705 --> 06:20.445
or a real female, so to speak?

06:20.945 --> 06:24.085
Um, so the distinction here is that we first kind of

06:25.205 --> 06:28.325
identify the other and put them into one of these buckets.

06:28.625 --> 06:30.445
Is this a male or is this a female?

06:30.785 --> 06:34.205
But then we continuously evaluate whether this

06:35.415 --> 06:37.645
human is, is behaving

06:37.945 --> 06:40.205
and carrying themselves in a way that's consistent

06:40.205 --> 06:41.805
with our behavioral expectations.

06:42.225 --> 06:44.365
So that is the distinction between the categorical

06:44.365 --> 06:46.685
and the performative, is that first we kind of

06:47.195 --> 06:50.525
designate an entity into a preconceived notion

06:51.105 --> 06:52.165
of a category,

06:53.045 --> 06:56.085
and then we evaluate whether it performs in a way

06:56.085 --> 06:59.285
that's consistent with our expectations of how a member of

06:59.285 --> 07:01.125
that entity should, should behave.

07:01.865 --> 07:05.685
Um, and the latter has really not been discussed much in,

07:05.685 --> 07:07.085
in the literature on organizations.

07:07.425 --> 07:10.005
It has in the literature on gender,

07:10.025 --> 07:11.765
and other, in, in, in sociology.

07:12.175 --> 07:16.085
We've analyzed, a large swath of,

07:16.155 --> 07:18.725
quarterly earning calls as I, as I described them earlier,

07:19.225 --> 07:22.005
that executives, traditionally these are the,

07:22.005 --> 07:23.965
the chief financial officer or the CEO,

07:23.965 --> 07:26.725
but often other executives as well, conduct with,

07:26.915 --> 07:27.925
with analysts.

07:28.705 --> 07:31.685
Um, and we basically employ a,

07:32.085 --> 07:34.485
a natural language processing model known

07:34.485 --> 07:38.565
as word embedding models, word embedding models,

07:38.665 --> 07:41.085
are a family of relatively, at this stage

07:41.085 --> 07:44.925
of our lives relative to Chad GPT, relatively primitive

07:45.145 --> 07:47.805
or fundamental, models.

07:48.405 --> 07:50.485
Actually, Chad, GPT has many, many millions

07:50.485 --> 07:51.485
of word embedding models.

07:52.025 --> 07:55.765
The idea is, is is quite, intuitive.

07:55.765 --> 07:58.565
However, what the model does is that it,

07:58.625 --> 08:01.645
it reads a sentence and there's a mask word.

08:02.105 --> 08:04.445
And on the basis of usually the three proceeding

08:04.445 --> 08:07.085
and following words, it kind of makes a guess on

08:07.085 --> 08:09.765
what is most likely to be the masked word.

08:10.145 --> 08:13.125
Uh, and the idea is that, the meanings

08:13.125 --> 08:14.565
of words are a function of their context.

08:14.705 --> 08:18.045
So think for example about the following sentence I took my

08:18.605 --> 08:19.965
hmm, to the veterinarian,

08:20.585 --> 08:22.405
hmm would be the masked word, right?

08:22.465 --> 08:25.805
The veterinarian kind of suggests that, it's more likely

08:25.825 --> 08:28.765
to be dog or cat and not grandfather, right?

08:29.305 --> 08:34.005
Um, and what, a model, a word embedding model, kind

08:34.005 --> 08:35.965
of the, the machinery is not that important,

08:36.145 --> 08:39.965
but if it's trained on, on a sufficiently large, corpus

08:39.965 --> 08:43.725
of data, it learns the meanings of words

08:43.945 --> 08:46.485
by understanding their context of enunciation,

08:46.865 --> 08:49.725
and it ends up representing them in a hyperdimensional

08:49.725 --> 08:52.325
space, usually a 300 dimensional coordinate.

08:52.425 --> 08:54.405
So think about a space that's

08:54.405 --> 08:57.125
Multidimensional, and it turns out that the distance

08:57.125 --> 08:59.925
between words in these spaces is consistent with

08:59.925 --> 09:02.725
how people understand the semantic meanings of these words.

09:03.665 --> 09:06.405
And with that, we can now start kind

09:06.405 --> 09:08.765
of looking at whether a performance

09:08.765 --> 09:12.365
of a particular executive is typical to the category

09:12.365 --> 09:14.805
to which it belongs, by looking at whether the words

09:14.835 --> 09:18.885
that they use in their presentations are far

09:19.425 --> 09:21.165
or close in embedding space,

09:21.225 --> 09:24.365
and hence in semantic space to the words used

09:24.425 --> 09:26.885
by similar executives in com in competing firms.

09:27.505 --> 09:29.845
Uh, so to go back to the Elon Musk example, right?

09:29.845 --> 09:33.885
If you run, let's think about him in the head of the CEO

09:33.985 --> 09:38.445
or of of Tesla, if you run a, an automotive,

09:38.785 --> 09:43.605
company, is Elon Musk, when he talks to,

09:43.745 --> 09:47.445
investors are the meanings of the words that he uses,

09:47.445 --> 09:50.005
similar to the meanings of the words that his competitors,

09:50.665 --> 09:52.365
the CEOs of Ford and,

09:52.825 --> 09:56.285
and Daimler bands, et cetera, use,

09:56.825 --> 09:59.765
as you would imagine, our expectation would be that

09:59.765 --> 10:02.365
for Musk in particular, they would be different,

10:02.365 --> 10:07.165
because he's an atypical, CEO that's,

10:07.165 --> 10:09.485
that's, without getting into the Adidas of how

10:09.485 --> 10:11.285
that's effectively done,

10:11.875 --> 10:13.845
intu intuitively, that's what we're inferring.

10:14.145 --> 10:16.125
Uh, I'll just proceed this by saying there's been a lot

10:16.125 --> 10:18.445
of literature, looking at, at firms

10:18.445 --> 10:20.685
that are categorically different, from,

10:20.685 --> 10:23.085
from their competitors who sell, say, products,

10:23.545 --> 10:25.045
that are unusual.

10:25.185 --> 10:28.845
So imagine a, a university that would also manufacture cars,

10:28.875 --> 10:30.805
that would be a very weird entity.

10:31.145 --> 10:33.405
Uh, so we know that these are discounted in the market

10:33.405 --> 10:35.285
because they're confusing and,

10:35.425 --> 10:37.125
and people don't know how to make sense of them.

10:37.745 --> 10:40.005
Um, but what we found was

10:40.005 --> 10:43.045
that conditional on your categorical identity,

10:43.425 --> 10:47.685
if you are performatively atypical, if you're an Elon Musk,

10:47.825 --> 10:51.845
so to speak, that is actually overvalued by the market.

10:52.545 --> 10:57.005
So if you are an automotive ma manufacturer,

10:57.005 --> 10:59.845
but your CEO speaks in quarterly earning calls in ways

10:59.845 --> 11:03.805
that are very weird and atypical, what tends to happen is

11:03.805 --> 11:08.445
that analysts as a, overestimate,

11:08.445 --> 11:12.285
your company's future performance, which ironically leads

11:12.285 --> 11:15.205
to a bad outcome, which is a negative earnings surprise.

11:15.865 --> 11:18.005
Uh, so they tend to overestimate

11:18.025 --> 11:19.365
how much your performance will be in the future.

11:20.105 --> 11:23.485
Uh, but because it's an overestimation, they get,

11:23.585 --> 11:24.845
negatively surprised.

11:25.265 --> 11:29.245
Um, and what we assume is happening is the following,

11:29.545 --> 11:31.845
people, because as we started, right,

11:31.865 --> 11:34.445
people know very little about what is going on

11:34.665 --> 11:37.205
inside an organization, how the sausage is made, so

11:37.205 --> 11:39.365
to speak, and they make inferences.

11:40.025 --> 11:43.765
And if they see an Elon Musk type being kind

11:43.765 --> 11:47.405
of brazenly atypical, I think that

11:47.405 --> 11:48.965
what is happening in their heads is

11:48.965 --> 11:51.685
that they're inferring this as a signal about the fact

11:51.995 --> 11:54.465
that this company has a unique strategy.

11:55.205 --> 11:59.105
It has the audacity and, and,

11:59.685 --> 12:03.265
and the capacity, to be non-conformist.

12:03.565 --> 12:06.385
And they infer that is information about,

12:06.385 --> 12:08.865
something unusual about, its, its future performance

12:08.865 --> 12:11.785
that tends to be, an incorrect inference.

12:12.085 --> 12:16.825
Uh, so they read a signal that is in fact, bss, you know,

12:16.855 --> 12:18.555
it's hype,

12:18.735 --> 12:20.355
but they infer it to be,

12:21.275 --> 12:23.875
a reliable signal about the company's performance.

12:24.135 --> 12:27.515
Uh, what we found was that, this is especially the case,

12:28.495 --> 12:33.435
when your atypicality is, emulating,

12:34.135 --> 12:37.155
the performances in these kinds of,

12:37.265 --> 12:40.395
quarterly earning calls of companies that are kind

12:40.395 --> 12:41.835
of celebrated innovators.

12:42.575 --> 12:47.215
So if you say, I don't know, run a retail, chain,

12:48.395 --> 12:53.295
and, the kind of meanings that you express in your,

12:53.295 --> 12:55.215
quarterly earning calls are very different

12:55.725 --> 12:59.615
from your competitors, but similar to the Elon Musk

12:59.755 --> 13:01.055
and Googles of,

13:01.315 --> 13:05.395
and Nike's of the world, that's when,

13:06.175 --> 13:10.555
analysts really are kind of go astray and, and,

13:10.855 --> 13:15.035
and fall prey to your hype as if it were a signal

13:15.095 --> 13:16.675
of real, performance.

13:17.375 --> 13:20.035
Um, I like to think about it as kind

13:20.035 --> 13:21.315
of we capture the cool factor

13:21.935 --> 13:24.155
and nobody knows how to really quantify cool.

13:24.825 --> 13:26.955
Some people are cool, and,

13:27.095 --> 13:30.355
and when people emulate coolness,

13:31.465 --> 13:33.795
they attempt to project quality,

13:35.125 --> 13:37.305
but it's a vacuous signal at the end of the day.

13:38.015 --> 13:40.185
What this really reminds me is the story of WeWork.

13:40.445 --> 13:43.465
You know, WeWork, which never went public.

13:44.245 --> 13:46.425
Uh, that's when it failed, when it, when it filed

13:46.445 --> 13:47.985
for an IPO, but WeWork was,

13:48.565 --> 13:50.585
was really running an atypical way.

13:51.165 --> 13:52.745
Um, and it's,

13:53.045 --> 13:57.385
former CEO Adam Neuman was a very extravagant and, and,

13:57.445 --> 14:01.305
and kind of, atypical guy who would walk around barefoot

14:01.485 --> 14:04.905
and, and have these very, weird visions of the world.

14:05.525 --> 14:08.705
Um, and, what happened was that I think

14:08.775 --> 14:11.785
that was misinterpreted by very, very seasoned investors

14:11.925 --> 14:15.305
as information about some sort of innovative quality,

14:15.415 --> 14:17.905
that really, and that balloon the air out of

14:17.905 --> 14:20.225
that balloon came out quickly when,

14:20.605 --> 14:22.985
due diligence was happening during the IPO phase.

14:23.725 --> 14:26.025
And I think that's precisely representative of the,

14:26.365 --> 14:29.505
of the qualities that we were able to capture with our,

14:29.635 --> 14:30.785
linguistic approach.

14:31.125 --> 14:33.265
One potential implication is that you can think about

14:33.285 --> 14:34.545
as an investment strategy.

14:34.765 --> 14:36.385
I'm not sure that, I would do this,

14:36.445 --> 14:40.425
but, this would probably be a signal

14:40.615 --> 14:44.465
that the markets are going to be over optimistic,

14:46.005 --> 14:48.985
unless they all correct using our algorithm, right?

14:49.005 --> 14:50.545
So that would be a, an immediate

14:50.545 --> 14:51.545
Implication. But I think

14:51.545 --> 14:53.205
the broader implications are

14:54.235 --> 14:57.285
that when you're trying to estimate something so complex

14:57.465 --> 14:59.845
as the performance of a firm,

15:00.185 --> 15:02.445
and it's you don't have really reliable data,

15:02.665 --> 15:05.165
you're not there, even in the employees in the firm,

15:05.165 --> 15:07.325
don't really fully understand what's going on in the firm,

15:07.915 --> 15:09.125
what often

15:09.625 --> 15:13.205
nor does the, the CEO, so let alone, outside observers

15:14.325 --> 15:19.075
is, that we, we follow these

15:20.085 --> 15:21.275
misleading signals,

15:22.175 --> 15:26.355
and I think we all knew that we follow misleading signals.

15:26.435 --> 15:29.715
I mean, the term hype is precisely intended

15:29.815 --> 15:33.755
to describe information that people kind of latch onto,

15:33.945 --> 15:36.995
even though it is an unreliable signal of real quality.

15:37.855 --> 15:42.515
Um, I hope that, our methods will be,

15:42.695 --> 15:45.075
applicable in other context so

15:45.075 --> 15:47.755
that we can have better kind of understanding

15:48.575 --> 15:50.555
of when information is hype

15:51.375 --> 15:55.915
and when is it reliable information about the true quality

15:55.935 --> 15:57.395
of the entity being observed.

15:57.395 --> 15:58.875
These need not be just firms.

15:59.135 --> 16:01.915
You can imagine, you know, the, the quality of governments,

16:01.975 --> 16:04.515
the quality of, of other types of actors.

16:05.455 --> 16:09.875
We live in a, in an era of complex information overload.

16:10.715 --> 16:15.375
And hopefully we can harness machine learning and, and,

16:15.515 --> 16:18.775
and AI in ways that would help us reign in some

16:18.775 --> 16:20.615
of these complex informations.

00:00 - Organizations, constitute the most important process by which, things are being produced in the world. but, there's so much that we don't understand about them relative to what we do understand. We see what, we see how they're presented on the

00:18 - outside, but we don't really know how the sausage is made, so to speak. You know, think about like Elon Musk's takeover, over Twitter, that kind of exposed a little bit of how the sausage is made. But, not all of his antics are, are good representations of what is going on inside the organization. And given that so much, capital is invested, in these entities, and, there's so little information about

00:48 - what is going on in them. I think it's really important for us to understand how investors make inferences about, about organizations, but more broadly, how society at large understands what happens inside them, how it reaches conclusions, about, which, organizations are performing well, which aren't, whether it is from a financial, a purely financial perspective,

01:13 - but also I think, you know, our study as we'll discuss in a moment, I think has ramifications beyond the financials per se. We need to understand what happens in organizations because, they have broader societal implications. and in light of the fact that we understand so little, the main motivation question is what can we learn? And, and when, when organizational leaders talk about their organizations, is the information that they divulge, informative?

01:41 - And do we make, and do people make the correct inferences on the basis of that information? our analysis focuses on quarterly earning calls, which is, publicly traded. Companies are not required, but they normally, meet on a quarterly basis with analysts and, and, and provide divulge information. Now, as you would imagine, this is a very scripted and very choreographed experience.

02:04 - and the general literature that we speak to is, is twofold. One is, is the literature that has focused on these quarterly earning calls and generally the kind of, formal, mechanisms through which organizations provide information to investors and how reliable that information is, how to investors, and analysts and, act on this information, et cetera. So that's one, body of of literature. a second body

02:34 - of literature is a more broader social scientific literature ranging from psychology to sociology and economics, on how people make inferences, on other entities, be them other humans, other organizations. There's a broad literature that's specifically focused on organizations per se. there's a broad literature, in organizations known as quote unquote, the categories literature, which is focused on, how do people

03:10 - basically enforce some sort of cognitive order on this broad space of the organizational landscape. and, and we know that, all of us humans, we have this amazing, a machinery, called our cognition, but that uses a lot of heuristics. and we have these conceptual maps in our heads that we hold. And when we are encountered with new information, we try to understand it with respect to these maps,

03:43 - or you can think about them as cognitive filters. we see something in the world. We say, okay, this is a tree. This is a human, this is a car. and we have a lot of conceptual associations around that focal concept, and that guides us in a very automatic way in how we, act. And there's been a lot of research in, in demonstrating that people also have these

04:06 - categorical filters when they try to understand organizations. So, for example, if you go into a store, you have a concept of what a store is, and that very much kind of structures your expectations and how you will behave, and a store that, doesn't, look like a a traditional store confuses you. The literature, you know, we make the distinction, and I think there's an, a novel distinction that we make in,

04:35 - in, in this particular paper, is that we make kind of the, we distinguish categorical and performative, conceptualizations. and actually, I think the best way to understand it is, is the metaphor that we, we kind of got into this, terminology was from research on gender. so 30 years ago already rasman and Faust and introduced the idea of doing gender. And, and the idea was the following.

05:05 - So in a very kind of traditional binary sense of gender where there are women and men, and it is one of the most fundamental categories for which we perceive other humans as we first associate them with gender. what Vasser men and FA said was that it's not just that people are given a categorical identity, is that they need to continuously perform it. So, there are ways

05:32 - by which we kind of make distinctions, between men and and women on, on the basis of facial feature and other physical features. but you know, the most, dominant features, are usually in, in modern society are hidden from, from view. and as a consequence, we also expect people to perform their identities, to wear their hair in a certain way, but they're also feminine and masculine ways of behavior.

06:00 - there are feminine and masculine, ways of kind of carrying your body. And when people fail to perform in a way that's consistent with our expectation, that research has shown that people get confused and dismayed, and they're kind of start questioning, is this a real male or a real female, so to speak? so the distinction here is that we first kind of identify the other and put them into one of these buckets. Is this a male or is this a female?

06:30 - But then we continuously evaluate whether this human is, is behaving and carrying themselves in a way that's consistent with our behavioral expectations. So that is the distinction between the categorical and the performative, is that first we kind of designate an entity into a preconceived notion of a category, and then we evaluate whether it performs in a way that's consistent with our expectations of how a member of

06:59 - that entity should, should behave. and the latter has really not been discussed much in, in the literature on organizations. It has in the literature on gender, and other, in, in, in sociology. We've analyzed, a large swath of, quarterly earning calls as I, as I described them earlier, that executives, traditionally these are the, the chief financial officer or the CEO, but often other executives as well, conduct with,

07:26 - with analysts. and we basically employ a, a natural language processing model known as word embedding models, word embedding models, are a family of relatively, at this stage of our lives relative to Chad GPT, relatively primitive or fundamental, models. Actually, Chad, GPT has many, many millions of word embedding models. The idea is, is is quite, intuitive.

07:55 - However, what the model does is that it, it reads a sentence and there's a mask word. And on the basis of usually the three proceeding and following words, it kind of makes a guess on what is most likely to be the masked word. and the idea is that, the meanings of words are a function of their context. So think for example about the following sentence I took my hmm, to the veterinarian, hmm would be the masked word, right?

08:22 - The veterinarian kind of suggests that, it's more likely to be dog or cat and not grandfather, right? and what, a model, a word embedding model, kind of the, the machinery is not that important, but if it's trained on, on a sufficiently large, corpus of data, it learns the meanings of words by understanding their context of enunciation, and it ends up representing them in a hyperdimensional space, usually a 300 dimensional coordinate. So think about a space that's

08:54 - Multidimensional, and it turns out that the distance between words in these spaces is consistent with how people understand the semantic meanings of these words. And with that, we can now start kind of looking at whether a performance of a particular executive is typical to the category to which it belongs, by looking at whether the words that they use in their presentations are far or close in embedding space, and hence in semantic space to the words used

09:24 - by similar executives in com in competing firms. so to go back to the Elon Musk example, right? If you run, let's think about him in the head of the CEO or of of Tesla, if you run a, an automotive, company, is Elon Musk, when he talks to, investors are the meanings of the words that he uses, similar to the meanings of the words that his competitors, the CEOs of Ford and, and Daimler bands, et cetera, use, as you would imagine, our expectation would be that

09:59 - for Musk in particular, they would be different, because he's an atypical, CEO that's, that's, without getting into the Adidas of how that's effectively done, intu intuitively, that's what we're inferring. I'll just proceed this by saying there's been a lot of literature, looking at, at firms that are categorically different, from, from their competitors who sell, say, products, that are unusual.

10:25 - So imagine a, a university that would also manufacture cars, that would be a very weird entity. so we know that these are discounted in the market because they're confusing and, and people don't know how to make sense of them. but what we found was that conditional on your categorical identity, if you are performatively atypical, if you're an Elon Musk, so to speak, that is actually overvalued by the market. So if you are an automotive ma manufacturer,

10:57 - but your CEO speaks in quarterly earning calls in ways that are very weird and atypical, what tends to happen is that analysts as a, overestimate, your company's future performance, which ironically leads to a bad outcome, which is a negative earnings surprise. so they tend to overestimate how much your performance will be in the future. but because it's an overestimation, they get, negatively surprised. and what we assume is happening is the following,

11:29 - people, because as we started, right, people know very little about what is going on inside an organization, how the sausage is made, so to speak, and they make inferences. And if they see an Elon Musk type being kind of brazenly atypical, I think that what is happening in their heads is that they're inferring this as a signal about the fact that this company has a unique strategy. It has the audacity and, and,

11:59 - and the capacity, to be non-conformist. And they infer that is information about, something unusual about, its, its future performance that tends to be, an incorrect inference. so they read a signal that is in fact, bss, you know, it's hype, but they infer it to be, a reliable signal about the company's performance. what we found was that, this is especially the case, when your atypicality is, emulating,

12:34 - the performances in these kinds of, quarterly earning calls of companies that are kind of celebrated innovators. So if you say, I don't know, run a retail, chain, and, the kind of meanings that you express in your, quarterly earning calls are very different from your competitors, but similar to the Elon Musk and Googles of, and Nike's of the world, that's when, analysts really are kind of go astray and, and,

13:10 - and fall prey to your hype as if it were a signal of real, performance. I like to think about it as kind of we capture the cool factor and nobody knows how to really quantify cool. Some people are cool, and, and when people emulate coolness, they attempt to project quality, but it's a vacuous signal at the end of the day. What this really reminds me is the story of WeWork.

13:40 - You know, WeWork, which never went public. that's when it failed, when it, when it filed for an IPO, but WeWork was, was really running an atypical way. and it's, former CEO Adam Neuman was a very extravagant and, and, and kind of, atypical guy who would walk around barefoot and, and have these very, weird visions of the world. and, what happened was that I think that was misinterpreted by very, very seasoned investors

14:11 - as information about some sort of innovative quality, that really, and that balloon the air out of that balloon came out quickly when, due diligence was happening during the IPO phase. And I think that's precisely representative of the, of the qualities that we were able to capture with our, linguistic approach. One potential implication is that you can think about as an investment strategy. I'm not sure that, I would do this,

14:36 - but, this would probably be a signal that the markets are going to be over optimistic, unless they all correct using our algorithm, right? So that would be a, an immediate Implication. But I think the broader implications are that when you're trying to estimate something so complex as the performance of a firm, and it's you don't have really reliable data, you're not there, even in the employees in the firm,

15:05 - don't really fully understand what's going on in the firm, what often nor does the, the CEO, so let alone, outside observers is, that we, we follow these misleading signals, and I think we all knew that we follow misleading signals. I mean, the term hype is precisely intended to describe information that people kind of latch onto, even though it is an unreliable signal of real quality. I hope that, our methods will be,

15:42 - applicable in other context so that we can have better kind of understanding of when information is hype and when is it reliable information about the true quality of the entity being observed. These need not be just firms. You can imagine, you know, the, the quality of governments, the quality of, of other types of actors. We live in a, in an era of complex information overload. And hopefully we can harness machine learning and, and,

16:15 - and AI in ways that would help us reign in some of these complex informations.

Faculti AI can make mistakes