# Belief Overreaction and Stock Market Puzzles - Andrei Shleifer

1K views|5 months ago

ðŸ’« Short Summary

The video explores the relationship between stock prices, expectations, and beliefs in financial economics, emphasizing the impact of changing beliefs on stock price volatility. It challenges traditional rational expectations and highlights the role of overreaction of beliefs in driving pricing anomalies. The importance of long-term growth rate expectations in predicting returns is discussed, with optimism leading to disappointment and lower returns. The model predicts that overreactions in long-term growth expectations result in negative predictability in errors and returns. The significance of incorporating expectations in financial modeling and understanding investor sentiment is emphasized.

âœ¨ Highlights

ðŸ“Š Transcript

âœ¦

Professor Andre Schleier presents a new paper on addressing excess stock price volatility through expectations and beliefs in financial economics.

02:59Schleier discusses the present value formula and rational expectations operator in relation to stock market valuation.

Robert Schiller's 1981 test revealed significant volatility discrepancies in the basic valuation equation, indicating the complexity of stock market valuation.

âœ¦

Relationship between stock prices, future dividends, and discount rates.

07:04High valuations may not align with low expectations of returns.

Changing beliefs impact stock price volatility.

Research agenda explores various models and theories.

Direct measurement of changing risk attitudes is challenging.

âœ¦

Key Highlights:

12:39The importance of using survey data for expectations in understanding financial markets.

Changes in required returns or risk aversion are not necessary to explain financial market evidence.

Overreaction of beliefs to fundamentals is a driving force behind pricing anomalies and return predictability.

Strong earnings growth can lead to excess optimism, affecting price dividend ratios and predictably low returns in the stock market.

âœ¦

Calculating long-term growth rates of earnings for individual stocks and the market.

16:03Aggregation of earnings forecasts for up to five fiscal years and dividends per share.

Importance of observed forecasts and utilizing standard sources like CFO surveys and Schiller's website.

Asset pricing models such as Campo Schiller decomposition and calibration using average constants.

Emphasis on the predictability of returns based on earnings indexes and volatility of price indexes.

âœ¦

Importance of Long-Term Growth Rate Expectations in Predicting Returns.

21:56Short-term growth rates are not as relevant for predicting returns compared to long-term growth rates.

Analysts' one and two-year ahead earnings growth expectations do not significantly impact return predictability.

Overly optimistic long-term earnings growth forecasts can lead to low index returns due to overreactions to information.

Measured expectations are key in understanding market dynamics and providing return predictability and access volatility.

âœ¦

Relationship between LTG and return predictability.

27:27LTG's predictive power is attributed to capturing variations in required returns or non-rational market expectations.

Results show that LTG consistently predicts long-run returns on the stock market negatively.

Risk measures are generally insignificant in return predictability.

The analysis seeks to understand changes in LTG and their implications.

âœ¦

The impact of LTG expectations on prices and returns in financial markets.

32:30The model examines forecast errors and abnormal returns based on changes in expectations and past growth rates.

Overreactions in LTG expectations result in negative predictability in errors and returns.

Understanding expectation shocks is crucial for assessing volatility and predictability in financial markets.

âœ¦

The impact of forecast errors on long-term earnings growth predictions.

34:04Revising earnings growth forecasts upwards can lead to lower returns than anticipated.

The significance of the three columns in predicting forecast errors and future returns is highlighted, demonstrating a pattern of increased optimism leading to disappointment.

The model suggests that expectations of long-term growth tend to overreact, ultimately resulting in increased disappointment.

Excess volatility in prices and negative impacts on returns are discussed in relation to optimism in forecasting.

âœ¦

Importance of Optimism and Rational Expectations in Future Returns.

42:25Firm-level regressions and fixed effects are used to analyze data patterns and predict future returns.

Traditional finance models are challenged by introducing flexibility and discipline in expectation modeling using empirical data.

Cross-sectional evidence in the last 30 years shows systematic movement in returns explained by exposure to risk factors.

âœ¦

Importance of Systematic Irrationality in Financial Markets.

43:30Emphasis on the significance of aggregate LTG in capturing investor sentiment and volatility.

Systematic movements in sentiment, reflected in future returns, are highlighted.

Exposure to expectation factors is more relevant than traditional risk factors.

Impact of changes in expectations of long-term earnings growth on stock returns and profitability is discussed.

âœ¦

Importance of focusing on long-term expectations in financial markets.

51:13Using expectations can provide valuable insights and empirical evidence for building models with different investor perspectives.

This approach aims to address issues in a more nuanced way and move beyond standard methods of analysis.

Speaker emphasizes the effectiveness of incorporating expectations in financial modeling.

Potential benefits of considering diverse investor perspectives are highlighted.

âœ¦

Key Highlights from the Video Segment

52:54The research emphasizes understanding investor expectations, beliefs, and preferences in market equilibrium.

Different types of investors' beliefs should be taken into account in future models.

The discussion includes analysis of dividends and cash flow, as well as separate evaluations of earnings and dividends.

Combining earnings and dividends in more elaborate formulations is discussed for its implications in analysis.

âœ¦

AI revolutionizing measurement of expectations in finance.

01:00:14Exploring new ways to understand returns by leveraging more data.

Emphasizing the importance of moving away from time-varying discount rates towards realistic financial analysis.

Audience thanked for attending presentation aimed at sharing research with universities without access to academic experts like Professor Scher.

âœ¦

Gratitude for collaboration on future seminars and the FMA website.

01:01:35The speaker thanks the Professor for participating in the collaboration.

The experience is described as wonderful.

00:00okay so good morning afternoon or

00:03evening

00:04depending where on where you are

00:06attending this this seminar it is very

00:09much my pleasure to welcome you all to

00:11another research seminary in finance

00:13economics and banking a joint

00:16organization of the finance and

00:18economics and banking research Network a

00:21network of several universities

00:23worldwide and FMA International since

00:27January 2023 we have been very pleased

00:30to have FMA International as our partner

00:33my name is SAR Matos and I will moderate

00:36the today's seminar from alberg

00:38University Business School in Denmark if

00:41you have a question or comment to our

00:44speaker please use the chat Q&A box

00:48which is located Below on your screen I

00:52will collect the questions throughout

00:54the session and put them to our speaker

00:57as time allows okay so let's go ahead

00:59ahead and get started it is very much my

01:03pleasure to introduce our speaker today

01:06Professor Andre schlier Professor

01:08schlier is John L professor of Economics

01:11at Harvard University he has worked in

01:15the areas of comparative corporate

01:17governance low low and finance

01:20behavioral Finance as well as

01:23institutional economics he has published

01:26seven books as well as over a 100 uh

01:30articles he is an editor of the

01:32quarterly Journal of economics and the

01:35faure of the econometric society the

01:38American Academy of Arts and Science and

01:42the American Finance Association in 1999

01:46Professor schlier won the John Bates

01:48Clark medal of the American economic

01:52Association according to repac Professor

01:55SCH is the most seated Economist in the

01:59world and has an age index of

02:02170 according to google schooler without

02:06further Ado Professor schlier thank you

02:08very much for accepting our invitation

02:11and we are looking forward to your

02:13presentation the V floor is

02:16yours um uh thank you very very much

02:20it's a great honor for me to uh give

02:23this talk to give this

02:25presentation um it's a relatively new

02:28paper with uh

02:30my usual set of collaborators Pedro

02:33bordalo nicoa joli and Rafael

02:36laporta them collaborators for for for

02:3930 years uh and it deals with the

02:44central problem of uh Financial

02:48economics which is the exess volatility

02:51of stock prices and it tries to um

02:57address this problem by focusing on

02:59expect

03:00expectations um and uh

03:03beliefs uh so just to set up the problem

03:07let me start with uh what is probably

03:11the

03:12single most

03:15basic uh formula in financial economics

03:19which is the present value formula uh in

03:24According to which the value of a

03:26security or a collection of securities

03:30uh uh p is given to the expected present

03:36value of its future cash flows or

03:39dividends uh so here uh the expectation

03:43operator very importantly is the

03:46rational expectations operator so the

03:50rational value of of a security in a

03:53dividend discount model is this present

03:56discounted value now in this simplest

03:59form the discount rate that we use to

04:02compute the expected present value of

04:05future dividends is a constant discount

04:07rate and so that

04:10is the most basic equation for the

04:14valuation of a given stock or the

04:17valuation of the stock market as a

04:21whole so uh in

04:251981 uh Robert Schiller tested this

04:30equation and showed that it fails

04:34dramatically in the data what he showed

04:37in particular is that if you look at the

04:40volatility of the left hand side of this

04:42equation which is the volatility of the

04:44stock market

04:46index and you compare it to a

04:49calculation of the right hand side of

04:52this equation uh which is the present

04:55value of future dividends where he

04:56actually looked at the realizations of

04:59dividends

05:00uh that the F left side is extremely

05:04volatile and the right side is uh not

05:08volatile at all and so the basic

05:11prediction of this theory that if the

05:13two things are the same their volatility

05:15should be the

05:17same is rejected um in fact later

05:21Campbell and Schiller showed that

05:24another basic prediction of this Theory

05:28uh which concerns the price dividend

05:31ratio uh also fails which is to say that

05:34the price dividend ratio for the market

05:35as a whole has weak correlation with

05:38future growth in dividends earnings but

05:41is strongly negatively correlated with

05:44uh future returns so this uh equation or

05:52Schiller's uh findings basically set the

05:56the tone or set the agenda for financial

06:00economics with respect to stock prices

06:04uh for the next 40 plus years as people

06:08struggle to try to figure out uh how to

06:12reconcile uh the uh present value

06:17formula with the

06:18evidence uh now the way in which Finance

06:22uh approach this problem or the uh is by

06:27basically if you go back to the equation

06:30by keeping the rational expectations

06:33assumption uh but uh playing with the

06:37denominator of the present value of the

06:40right hand side which is to say that

06:42rather than um uh challenge rational

06:47expectations which is where Shiller was

06:49going what people said is maybe there is

06:53variation in discount rates over time

06:56and that variation in discount rates

06:58account for the volatility uh of prices

07:03so uh this is uh there are a number of

07:06papers looking at this from many

07:09different perspective looking at

07:11consumption based model looking at long

07:14run risk models um and so

07:19forth so there is some this has been

07:22very productive research agenda but it

07:26has some issues uh the most important

07:30issue probably is if you look at survey

07:34expectations the central prediction of

07:37all the theories of volatility of the

07:41stock market and predictability of

07:43returns that deal with changes and

07:45discount rates is that when valuations

07:50are high it is when expectations of

07:53returns are low that is to say in an

07:55equilibrium model uh High valuation

07:59ations and low expectations go together

08:02um in fact if you look at the data when

08:05stock prices are high expected returns

08:08if as measured by survey expectations

08:12are are high rather than low so that

08:14goes exactly in the opposite direction

08:17of all the risk models and perhaps a

08:20deeper

08:21issue uh which we often sweep under the

08:24rug is that there is no direct

08:27measurement of changing risk attitudes

08:29or changing risk it's uh kind of an

08:33unobservable variable in the standard

08:36models that uh that uh uh people rely on

08:42uh without any direct uh imperical

08:46support so what we do in this paper is

08:51to go in a very different direction in

08:55this equation which is we keep uh the

08:59the returns or the expected

09:02returns at discount rates constant but

09:05relax the Assumption of rational

09:09expectations uh that is to say we keep

09:12uh R in the discount rates constant but

09:16instead of assuming rational

09:18expectations we use survey expectations

09:21of short and long-term growth of

09:26earnings uh to uh

09:29compute if you like the present value

09:33formula um and if you want to summarize

09:36this paper in one sentence what I will

09:40try to show you is the reason that stock

09:42prices are volatile is because beliefs

09:46are volatile the reason that stock

09:49prices exhibit excess volatility is

09:52because beliefs exhibit excess

09:55volatility there is to say that there is

09:57overreaction of belief to news so I'm

10:01going to present to you three main

10:03findings uh that expectations and what I

10:06want to stress is unlike in models with

10:09risk it's measured expectations it's

10:12expectations that we uh take from survey

10:16data rather than uh treat as an

10:19unobservable expectations jointly

10:21explain excess volatility and

10:23predictability of aggregate stock

10:26returns uh but also returns The Firm

10:29level aggre optimism is linked to

10:33cross-sectional return spread and the

10:35same mechanism of LTG overreaction

10:38overreaction of the survey estimate of

10:41long-term earnings growth explains

10:43market and uh cross-sectional puzzles uh

10:48so what I will try to argue is that in

10:51fact we don't really

10:53need

10:56uh changes in required returns or

10:59changes in uh risk aversion in discount

11:03rates to understand the evidence in

11:06financial market so this is kind of the

11:09one uh figure summary of the

11:12results so in this figure um I think

11:16okay I don't have the the map I think

11:19the green line is uh the I'm sorry the

11:23market price is the red line that is the

11:26source of the shell puzzle that explains

11:28the

11:29volatility uh of the stock

11:32market uh the blue line is schill's

11:37right hand side which is the present

11:38value of future dividends and the

11:41Shiller puzzle is that the red line is

11:45much much more volatile than the blue

11:48line and the uh green line is the

11:52present value with the expected present

11:54value when you use survey expectations

11:57of future divid

11:59growth and as you can see if you use

12:02survey expectations even with constant

12:04discount rate you can explain

12:08volatility uh extremely extremely well

12:12uh right without without resorting to

12:15changes in disc

12:18discoveries let me skip this so I can

12:20move along uh so as I said we'll try to

12:24offer unified assessment of these

12:26puzzles we'll identify overreaction of

12:30beliefs to fundamentals as the drive of

12:33pricing

12:34anomalies we'll try to link systematic

12:37forecast errors to the predictability of

12:39returns and the basic narrative or the

12:42basic model I'll show you a bit of a

12:44formal model is that when you see strong

12:47earnings growth that leads to excess

12:50optimism for expectations of future

12:53long-term earnings growth that gives a

12:56rise to the price dividend

12:59ratio uh that leads to systematic

13:02disappointment and to predictably low

13:05returns and I want to stress the last

13:08point which is you know the way since

13:10farma in the 1960s for 60 years in we uh

13:15one of the ways in which we assess the

13:18rationality of the market is whether you

13:21could have predictably whether you could

13:23predict future returns uh on financial

13:27securities and what I'll show you is

13:29that in fact you have predictably low

13:32returns uh in the uh in the in the stock

13:36aggregate stock market so I'll talk

13:39about data I'll talk about expectations

13:41of the puzzles I'll present a small

13:43model and then I'll show you some uh

13:46some

13:47data so let me go through this quickly

13:50you can read the paper um uh if you're

13:54interested in details or in

13:57replication uh so so what we use is

14:00again uh this V variable what I call the

14:03magic variable that Rafael laort

14:07actually introduced in his uh Harvard

14:11PhD thesis which I supervised which he

14:14finished in 1994 30 years ago but which

14:18kind of uh was not used very much uh in

14:24the following you know 25 years which is

14:27this variable

14:29called LTG or analyst estimates of

14:33long-term growth rate of earnings this

14:36variable is available for individual

14:38stocks but then you can Aggregate and

14:41create a measure of

14:43LTG uh long-term growth uh estimates for

14:47the market uh we look at earnings per

14:49share forecast to up to uh five fisal

14:53years look at dividends per share and uh

14:57uh again this are observed forecasts

15:00rather than uh uh rational expectations

15:04forecasts so as I mentioned we aggregate

15:07long-term growth rates of earnings by

15:10valuating firm level

15:13forecasts uh there is some something

15:16similar that was done by Stefan Nagel

15:18and Je uh in 2020 there are some small

15:22differences they don't

15:24matter

15:27um we

15:29do things likewise with uh aggregate

15:34earnings forecasts uh we take care of

15:37things like stock splits and uh

15:40disappearance of FBS and so this is the

15:42basic strategy of constructing the

15:45numbers uh we also look at CFO surveys

15:49comput crisp uh Schiller's website those

15:53are very standard uh sources of

15:57information

15:59okay let me start by just doing

16:03something very standard in asset pricing

16:06which is the Campo Schiller

16:11decomposition which uh

16:14represents uh the log price index uh

16:18based on earnings expectations data and

16:22so this is all in logs so the price is

16:26the log of the uh price of the stock

16:29market small e is the log of today's

16:32earnings per

16:34share uh that and then you have terms

16:38that look at earnings

16:41growth expected earnings growth next

16:44year two years from now and then the

16:47fifth term here is uh long run earnings

16:51growth now as always in these equations

16:54you need to make some assumptions about

16:56the terminal uh long run uh terminal

17:00growth rate and I'll show you what we do

17:02about it now the way in which we

17:05calibrate this model is we use pretty

17:07standard assumptions for the constants

17:11that appear in this

17:12formula uh you know using the average

17:16log price dividend ratio average log

17:19payout ratio uh average discount rates

17:23and average growth rate of earnings to

17:26to match the level uh or of of stock

17:29prices which is very s very standard

17:32strategy uh in this

17:36literature we also as a starting point

17:39compute the rational price following

17:42Schiller which is you know we look at

17:44the terminal price assuming constant

17:47growth we look at the same discount

17:50rates to sample average as a shill did

17:53and we use the rational price uh using

17:57actual dividends

18:01so this is in some sense I'm going to

18:03start with a punchline and then I'll

18:04spend most of my time looking at the

18:06predictability of

18:09returns uh so this is the earnings based

18:12index of the

18:14valuation so there are three columns in

18:16this table the first one is a measure of

18:20volatility of prices around

18:2314.8% in our

18:25sample uh which uh is uh uh very high uh

18:32the second uh column is the

18:35volatility of the Schiller

18:38index which is only 7% That's the puzzle

18:43the expected present value is not

18:45volatile at all if you use constant

18:47discount rate and then the third column

18:50is uh as I've shown you in the picture

18:54is the volatility of the price index

18:56using survey expectation of earnings

18:59growth and long-term earnings growth uh

19:03instead of assuming uh rational

19:07expectations and what you see the

19:09numbers in the third column are

19:12shockingly close to the numbers in the

19:14First Column which is to say taking into

19:17account the volatility of

19:19expectations uh accounts for the

19:21volatility of the uh uh S&P index so we

19:26don't need any volatility in discount

19:29rates uh and you could do the same

19:31calculation using the dividend price

19:34ratio uh or a co-integrated model or

19:38whatever you would

19:41prefer so

19:44uh the question is you know how do we

19:47think about this

19:49index so this is as I said this is the

19:52picture I already showed you that in

19:55fact the fit between the market market

19:58price which is uh uh the red line and

20:05the market the fair price we compute

20:08using uh the price we compute using

20:10expectations which is the green line is

20:13is a very very close

20:15match now let me turn to sort of the the

20:19the more exciting and challenging part

20:22of the of the of the project or of the

20:25study which is return predictive ability

20:29uh expectations help account for the

20:32price path but does that does not

20:35guarantee that they predict

20:37returns

20:39uh expectations and predicted variation

20:42could be close to rational predictable

20:45returns could uh reflect time bearing

20:47discount rates so can we J can we take

20:51the analysis to the next step and try to

20:55understand where expect expectations

20:58variables help us understand returns so

21:01this is what we do next and this is uh

21:05these are the results so I just want to

21:08spend quite a bit of time on this table

21:10and then I will explain it and I will

21:13um then elaborate and show some

21:16mathematically or some econometrically

21:18more sophisticated results in a minute

21:21so the three columns here going

21:25forward are realized in index returns

21:29one year ahead 3 years ahead and 5 years

21:34ahead um the three panels here uh

21:39looking at these things horizontally is

21:42using these forecast of long-term

21:45earnings growth as an explanatory

21:48variable uh this is LTG this is to 3 to

21:52five year Horizon expectations of

21:56earnings growth the second panel is

21:59again from analyst forecast data next

22:03year forecast of earnings growth so

22:06that's a short run uh forecast of

22:10earnings growth and the third panel is

22:13uh the forecast of earnings growth

22:15between year a year from now and two

22:18years from now so what you find is

22:21something very important and this is

22:23actually whatever you think about our

22:26whole Enterprise this is something I

22:28think that's clearly and obviously true

22:30in the data which is that what matters

22:33for predicting returns particularly uh

22:37long run returns uh is uh LTG which is

22:42to say expectations of long run earnings

22:45growth uh that uh analysts have the one

22:50year ahead earnings growth expectations

22:52and two year ahead earnings growth

22:55expectations basically have no

22:57explanatory power uh that is to say what

23:02matters for uh return predictability is

23:06the same thing as what matters for

23:08return uh uh for understanding

23:12volatility which is to

23:14say uh the long run reversion if you

23:19like of valuations when people are ex

23:22extremely optimistic about long run

23:26earnings growth over Horizons to 3 to 5

23:30years Returns on the aggregate index are

23:32going to be extremely low people

23:35overreact to information and stock

23:39prices uh

23:42revert so what this tells you is that at

23:45least in the data measured expectations

23:47offer a parsimonious account of access

23:50volatility and return

23:52predictability uh LTG the long-term uh

23:55growth rate plays a key role uh the

24:00short-term growth rate of earnings from

24:03forecast that has been stressed by the

24:06and

24:07Myers uh is it matters for the variation

24:10of price dividend ratio but not for the

24:14price level and not for understanding

24:17return uh predictability so all the

24:20action comes from

24:22LTG uh so what do we do with this well

24:26there are two potential explanations

24:28there are two

24:30possibilities one is that LTG sply

24:33captures variation and required

24:36returns uh which is to say that you know

24:40it is

24:41uh just a coincidence or something that

24:46uh is how these expectations are formed

24:50that is correlated uh with the variation

24:53and required

24:55returns or alternatively which is

24:57preferred explanation uh is that LTG

25:01captures non-rational Market

25:03expectations which account for valuation

25:06and return volatility and return puzzle

25:08so we need to rule out the first

25:11explanation uh to really move forward

25:13and that's what I'm going to try to be

25:14doing over the next uh uh several slides

25:19so let me try to talk about the about

25:21whether LTG is piously capture variation

25:24in uh required returns so so one thing

25:29that we can do one method we can use is

25:32we can control for proxies of required

25:35returns and predictability regressions

25:38that are common in the literature so we

25:40below I'll show you results on Surplus

25:43consumption from Campbell and C Cochran

25:46from K from letau and

25:49Lon for volatility measure from uh Ian

25:54Martin and several others so that's why

25:58way to test uh the hypothesis so whether

26:02the predictive power of LTG is

26:06spous uh another way to look at it is to

26:09ask our LTG revisions driven by past or

26:12expected returns so we're going to look

26:15at the RO of fundamental news and

26:17surprises relative to cyclically

26:19adjusted

26:21earnings to look at that

26:25prediction so this is a again a very

26:28basic

26:30regression where on the left hand side

26:32is the fiveyear return on the stock

26:34market

26:36index and on the right hand side we put

26:41various variables and there is several

26:43more in the paper uh of uh have been

26:48that have been used

26:51to uh look at return predictability the

26:54first one is LTG the log to growth rate

26:56of earnings

26:58and the second one which we say is X for

27:01different axes that I describ in the

27:05uh uh in uh the last row of the table is

27:11the various axes which is the various

27:13measures of risk uh and so you know one

27:17of them is Surplus consumption one is K

27:20and one is the measure of volatility so

27:23you can see something uh pretty obvious

27:27in this paper that LTG consistently

27:30predicts long run Returns on the stock

27:32market negatively again which is to say

27:35when people are very optimistic about uh

27:38earnings

27:39growth uh the uh returns Al low uh going

27:44forward but the measures of risk with

27:47the exception of Ian Martin's measure of

27:50volatility uh are insignificant they're

27:54consistently insignificant regardless of

27:58uh

27:59specifications uh um and again if you

28:02think that uh the in Martin variable is

28:05the right measure of risk risk does have

28:08a role to play in return predictability

28:10in addition in addition to um earnings

28:16forecasts uh but otherwise it looks like

28:19Risk to not have much of a role to

28:24play now what we want to do then is to

28:27do something slightly more elaborate

28:31which is to try to understand uh What uh

28:36explains uh changes in uh

28:40LTG uh and so what the data tell us is

28:45what you see in this table and this is

28:47again to try to test the hypothesis that

28:50LTG uh is just inert for prices or

28:54reflect something else is that uh LTG is

29:00predicted changes in LTG are predicted

29:03by past

29:05LTG uh By changes in uh uh

29:09earnings but that past returns and

29:13expectations of returns just like Risk

29:17measures are not that important for

29:20understanding uh changes in

29:23expectations uh so mean reversion and

29:26tangible news Drive about oneir of the

29:31uh of uh of the

29:35revisions okay so let me so what I think

29:38this table showed you the last two

29:40tables is kind of the basic step but it

29:43doesn't look like changes in

29:46expectations which Drive returns um are

29:50particularly related to changes in

29:53prices does it look like they're

29:55reflective of uh risk estimates so the

29:59question is can we do more with

30:02understanding how uh they are related uh

30:07to uh uh how they're related to access

30:11volatility and predictability and for

30:13that we need a little bit of a model

30:15I'll go through the model relatively

30:17quickly because my time is limited but I

30:20think it might be helpful for for to

30:22think about

30:26it uh

30:28so let's start by looking at C will

30:32Shiller decomposition with the con

30:35require return so you know you have the

30:38price uh dividend ratio logs is a

30:41constant plus the sum of future growth

30:44rates um of dividends uh let's suppose

30:48that the dividend growth this is very

30:50standard uh follows an ar1 uh

30:55process uh let let

30:58also uh postulate that the average firm

31:03uh which is the market has the ar1

31:06dividend growth

31:08process um and let's just for for for

31:11concreteness to have a specific model

31:15let's postulate that the market

31:17expectations about future growth May

31:20deviate from rational

31:23expectations uh uh by this uh

31:28distributed

31:29lag uh function of expectation shock uh

31:34Epsilon uh

31:37t uh let's finally assume that the

31:40expectation shocks are persistent these

31:43epsilons are

31:45persistent uh so if this parameter row

31:50uh is less than uh a than expectations

31:55mean revert

31:58let's finally assume and this is kind of

32:01the overreaction or under reaction

32:03assumption that the expectation shock is

32:06proportional to rational belief which is

32:08to say that people really don't have a

32:12random way of forming their beliefs but

32:15they either over or under react

32:18depending of the

32:20magnitude or the sign of this uh

32:22parameter Theta okay I'm sorry about all

32:25this algebra but I'll show you

32:27where it takes

32:30us so this model gives us uh several

32:36predictions that we can bring to the

32:38data that we have the first one is that

32:41forecast errors are

32:43predictable and if you run a regression

32:48with surprises in growth rates CH

32:51differences between the realized and the

32:53expected in the day in the expected by

32:56analyst growth rates that that is going

32:59to be a function of changes and

33:02expectations as well as of past

33:05expectations of growth rates and second

33:08that abnormal returns going forward

33:11depend negatively on expectation shock

33:14which is something very close to what we

33:17found uh in the

33:20data uh so to put it differently if LTG

33:24expectations overreact this is the key

33:27prediction of the model both forecast

33:29errors and returns are negatively

33:32predictable from the revision of LTG and

33:34from the lagged level of LTG okay so the

33:37model

33:39delivers uh these very sharp predictions

33:42which is to say when long-term earnings

33:44growth forecasts are revised

33:46up uh you uh uh that predicts negative

33:51returns and uh that uh when the levels

33:55of LTG

33:57period before a high you get likewise uh

34:02lower

34:04returns um so this is probably the most

34:08important table in the paper so let me

34:12uh go slowly so because I want to make

34:15sure I'm clear and not confusing about

34:18it so here is what the three column show

34:22the First Column is try to predict the

34:25forecast error remember that the most

34:28basic prediction of the rational

34:30expectations

34:32hypothesis is that forecast errors um

34:36are not uh predictable that is the

34:39definition of rational expectations so

34:42the first column on the left hand side

34:45of the regression you have the

34:47realization of aggregate earnings

34:50growth minus the forecast LTG forecast

34:54of aggregate earnings growth so that is

34:56the forec c

34:57eror um in the First Column uh in uh so

35:03let me stick with the First Column the

35:05two explanatory variables here are uh

35:10change in LTG as the model predicts and

35:12lag LTG and what you find is this kind

35:16of basic negative coefficient on change

35:19in LTG that is to say what it says is if

35:22I am revising

35:25upwards my forecast of earnings

35:29growth my I am likely to be

35:33disappointed when I'm becoming more

35:36optimistic realized earnings are going

35:39to be lower than the

35:42forecast expectations are

35:47predictable expectation errors excuse me

35:49are predictable that is to say as I sit

35:53here at time T if I become more

35:56optimistic I'm going to be too

35:58optimistic the second column does the

36:01same thing with returns fiveyear returns

36:06and it says that when I'm becoming more

36:09optimistic or when I used to be

36:11optimistic

36:12before returns five years forward are

36:16going to be um uh are going to be low

36:21that is to say uh changes and levels

36:25past levels of LT G predict fiveyear

36:29returns but what is the most important

36:32here is the third

36:34column because here what I'm trying to

36:37see to do is to do the following I've

36:39triying to predict returns going

36:42forward based on my

36:46prediction of the forecast eror which is

36:49to say I take the first colum and I use

36:52it as a first stage in an IV regression

36:56where I

36:57predicting how how wrong I'm going to be

37:02in terms of my earnings growth forecast

37:05and what the third column says is that

37:07my prediction of forecast

37:10error uh also helps me uh account uh for

37:15returns going forward that is to say I'm

37:17standing here at time T I have the level

37:21of LTG before and the

37:25change suppose I've become more

37:28optimistic or the analysts have become

37:30more

37:32optimistic what that tells me is that

37:34the first of all they're going to be too

37:36optimistic they're going to be

37:37disappointed and then what the third

37:40column says is the predicted

37:42disappointment predicted not the actual

37:45disappointment the predicted

37:48disappointment uh helps me account for

37:50returns going

37:51forward uh which is exactly what the

37:55model would predict

37:59so the bottom line here is that LTG

38:03expectations

38:04overreact increased optimism predicts

38:08disappointment uh this is in line with

38:10excess volatility of

38:12prices uh returns are negatively

38:15predicted by optimism and even by

38:18predicted disappointment and that

38:21finally that the magnitudes are large

38:23one standard deviation higher division

38:25leads to point 4 standard deviation

38:28lower returns roughly 40% lower returns

38:32over the last five

38:37years uh could the results be driven by

38:40outlier episode such as the.com bubble

38:45uh you know in order to address these

38:47you know we have 40 Years of data uh

38:52there are unquestionably if you go back

38:54to the Schiller picture three or for

38:57episodes uh which are absolutely

38:59essential for understanding the results

39:02you know this is the dotom bubble is

39:04being one of them which is to say when

39:06people were getting more and more

39:08optimistic and very optimistic returns

39:11going forward turned out to be very low

39:14so one of the things we can try to do is

39:17to run firm level regressions uh using

39:21time dummies to control for common

39:23shocks so that will take care of things

39:25like

39:26uh the com Bobble uh that we have in the

39:30sample uh we also can look at for fixed

39:33effects to absorb differences in average

39:38returns now obviously since we are

39:41looking at the firm level you have

39:45vastly more observations you go from a

39:48few hundred observations to 400,000

39:51observations but what is

39:53absolutely

39:55striking is is that even with uh fixed

40:01effects and with

40:05uh uh even with fixed effects uh you

40:08have extremely similar patterns in the

40:12cross section as you get in the time

40:14series which is to

40:16say that LT changes in LTG and La LTG in

40:21a

40:22cross-section uh uh negatively predict

40:25returns and more moreover even in the

40:27crosssection the strategy for

40:30forecasting uh prediction eror and then

40:34correlating it with future returns uh is

40:38a strategy that

40:41um uh accounts uh for future

40:46returns so again expectations of

40:49long-term earnings growth overreact

40:51they're too optimistic after periods of

40:52good news uh periods of excess optimism

40:56predict lower future returns departures

40:59from rational expectations are

41:02disciplined by expectations data uh

41:06account for a sizable chunk of aggregat

41:09predictability now I want to stress the

41:10last point because I guess this is

41:13something where I feel Finance has gone

41:16a little bit wrong or very wrong which

41:18is that you know I'm adding a degree of

41:22Freedom here which is rather than

41:25assuming Ral

41:27expectations um I

41:31um provide more flexibility in the model

41:34of expectations but I'm also

41:37disciplining this this this more General

41:40model with the actual data on

41:44expectations so you know it's if it's

41:47anything there's a lot more discipline

41:49here than there is an irrational

41:51expectations model with changes in

41:54discount rates because that model

41:57actually has no direct

42:00empirical uh content it just assumes

42:04that there are changes in required

42:07returns um that account for the

42:11evidence let me uh I probably have about

42:15four or five minutes so uh let me just

42:19say very few words about the

42:22cross-sectional evidence and then I'll

42:25conclude um in the crosssection the uh

42:30research in the last 30 years has been

42:33dominated by the work of uh farm and

42:37French um who argued that systematic

42:40movement and returns in the cross

42:42section is explained by

42:46exposure uh to risk factors John Cochran

42:50famously wrote that systematic Co

42:53movement uh of portfolios of Securities

42:58has to be risk because it cannot be the

43:01syncratic irrational beliefs now again

43:04there is a slip here which is that

43:08irrational beliefs and this is all the

43:10evidence I've shown you all the evidence

43:13I've showed you do not need to be

43:15idiosyncratic in fact Larry Summers and

43:18I published a paper in

43:211990 uh 35 years ago almost in which we

43:25explained exped that the uh

43:28irrationality in financial markets has

43:30to be uh systematic uh rather than

43:35idiosyncratic and the evidence I've

43:37shown you about the importance of

43:40aggregate LTG in capturing investor

43:43sentiment and accounting for

43:45volatility or returns that shill has

43:49documented is to a very large extent a

43:53systematic uh phenomenon that is to say

43:56phenomenon of systematic aggregate

44:00movements in sentiment as captured by

44:04expectations uh reflected in uh in in

44:07future returns uh so uh and likewise

44:12just because

44:14portfolios uh of Securities can Cove

44:18there is no reason to jump to the

44:21conclusion uh that this has to reflect

44:23risk that is uh there is not a logically

44:28uh valid

44:29conclusion so we'll show next that uh in

44:33fact

44:34empirically uh as uh as I would suggest

44:38this uh Cochran conclusion is just

44:41inaccurate uh fluctuations in aggregate

44:43optim over optimism measured by the same

44:46LTG index uh captur to a large extent

44:50systematic movement uh of portfolios of

44:54Securities which is another way of

44:57putting this is

45:00that factors like value and uh uh uh

45:06that are described as risk factors uh

45:10seem to a large extent uh reflect

45:13instead movement and beliefs and

45:15expectations I'm not sure how much of it

45:17I can do uh so let me start with just

45:21show you laort forms in the top LTG

45:24desile have systematically low future

45:26returns then perms in the bottom LTG

45:30desile like again this is a 30-y old

45:32finding so the question is does

45:34overreaction and aggregate LTG predict

45:37the returns differential or different

45:40portfolios of Securities over

45:43time uh so this is I'm just going to

45:45show this to you

45:47for again this is five year

45:51returns uh and uh we here look at uh low

45:56LTG portfolios High LTG portfolios and

46:00the

46:01spread and what you find is pretty clear

46:05evidence that first of all risk factors

46:07traditional risk factors don't matter

46:09and that the spreads are

46:12explained uh by exposure by the same

46:16expectation factors as I showed you in

46:19the time

46:20series uh and put defitely High LTG

46:24stocks have much lower returns after

46:26periods of aggregate over optimism which

46:29accounts for something like 70% of the

46:33spread I have to move forward I'm sorry

46:35we can extend the model to

46:40um uh to the cross-section of returns

46:44look at the uh farmer French

46:48portfolios and what you will see is uh

46:52again I'm very sorry I've run out of

46:54time is that even the Pharm of French

46:58portfolios and the patterns of returns

47:00in the farm of French portfolio are

47:03substantially accounted for uh By

47:07changes in expectations of long-term

47:10earnings

47:13growth so evidence is consistent with

47:16the idea that sub stocks Low Book to

47:19Market low profitability aggressively

47:22investing are more strongly overreacting

47:25to Agate news these stocks are

47:28especially overvalued after good

47:29aggregate news and disappoint in the

47:31cross-section going forward both in

47:33returns and earnings growth so the same

47:37expectations explanation helps account

47:40for the

47:42crosssectional and uh as well as time

47:45serious uh patterns over

47:50returns I think this is okay now I

47:53screwed up uh let me me see if I have a

47:56concluding slide uh that I could click

48:02on

48:04okay okay that is my final slide which

48:08is overreaction of expectations of

48:10future fundamentals accounts for sizable

48:13type series of cross sexual return

48:16predictability without requiring

48:18variation in discount rates or Price

48:21extrapolation uh this is very clearly

48:25shown the evidence with forecast errors

48:28uh overreaction of this LTG variable is

48:31consistent with much survey evidence or

48:33the expectations of investors firm

48:35managers professional forecasters and so

48:38on and so belief over reaction holds

48:40promise to parsimoniously explain

48:42long-standing macro Financial puzzles I

48:45should say that we have a followup paper

48:47in case somebody's interested in which

48:50we show that the same methodology can

48:53actually account for a lot of Mac

48:55macroeconomic findings um including uh

48:59volatility of

49:01investment uh volatility of credit

49:04spreads um and uh other phenomena re

49:08related to the business cycle and the

49:12credit cycle um so thank you very much

49:15uh let me stop here and see if uh I can

49:19answer uh some questions thank you very

49:23much professor schlier and um for being

49:25with us yes there is um some questions

49:28that I'll I'll pass to to you but before

49:31that I think that we have a small

49:33problem in the beginning in the first

49:35couple of minutes so my apologies for

49:36the attendees because they were able to

49:39see the slides but not to see Professor

49:41schlier but it was sort out I trust in

49:44after a couple of minutes so my

49:46apologies for that situation we have

49:49some some questions if we are unable to

49:51to go through all of them then by the

49:54end of the session I'll pass them

49:56through to to Professor schlier all the

49:59questions that are still in the chat so

50:01the the first one is coming from Ben

50:04Jensen and basically what he ask is are

50:08analyst forecast

50:10sufficient uh to reflect the average

50:13investor expectation and the complement

50:16saying CFOs used as robustness but both

50:21analysts and CFOs are likely more

50:24financially located than the typical

50:27Market participant and then he he makes

50:30a comment saying I came across a Wall

50:31Street bats post this morning that

50:34reflects the gap between Financial

50:36professionals and at least some Market

50:38participants so the question is

50:39basically if we can rely in full in the

50:43analyst forecast or if it's enough or

50:45should be complement with CFOs as

50:48robustness but then they tend to be more

50:51informal financially educated than than

50:53the average investor okay thanks this is

50:56obviously a fantastic question uh again

50:59let me let me start by just kind of

51:02placing this paper in context okay what

51:04this paper says you could call it proof

51:06of concept if you like is uh that you

51:13get very far by focusing on expectations

51:17instead of risk uh here we use the

51:20measure of expectations particular

51:23longterm expectations

51:26uh that is pretty widely

51:28available and it works extremely well

51:34empirically uh you are absolutely 100%

51:37correct that once we start writing down

51:41models or if we want to understand

51:44evidence in a more nuanced way uh rather

51:49than just

51:50reject uh the standard approach uh to

51:55the these problems we want to build

51:57models with heterogeneous investors

52:00perhaps different investors holding

52:02different

52:03expectations uh this raises a whole host

52:06of issues and I think there is a lot of

52:08new research there is kind of an

52:10explosion of new research uh in this

52:13area of you know what types of investors

52:17hold what expectations why do investors

52:20disagree who trades with whom how is the

52:24equilibrium Market determined when

52:26investors with different beliefs and

52:28expectations or maybe also preferences

52:31uh trade with each other so I think that

52:33is an

52:34extremely uh constructive and helpful

52:38Direction but the first thing we need to

52:41do is to get over the hurdle of risk and

52:46once we are you know on this uh new and

52:49Greener pasture of thinking about

52:52beliefs and expectations you are

52:54absolutely

52:56correct that uh the the the the future

52:59models will need to take into account

53:03the beliefs of various types of uh

53:06investors thank you Professor schli um

53:10another more relate with the the

53:12dividends and the discount cash flow um

53:14that unap Park uh asks dividends that

53:17become less important as um as a pay

53:21payout meod during the last the past

53:24decades does this

53:26change uh or affect the the analysis uh

53:31that you

53:32present uh so we have done in this paper

53:37uh something very straightforward and

53:40naive uh which is to just do the

53:47analysis separately with earnings and

53:49with dividends uh without

53:53thinking uh too

53:56much uh about you know combining the two

54:00in one framework uh as you recall in the

54:03in the

54:05formulation that I have produced there

54:08is a constant payout rate or constant

54:11payout ratio rather than variation in

54:14the payout ratio so I don't know but I

54:17would love to know uh what is what are

54:21the implications of having kind of more

54:24elaborate formulations in which you

54:27think uh simultaneously about uh

54:31earnings and dividends but here we've

54:34just checked for robustness rather than

54:37do anything like that which would be

54:38very

54:40helpful thank you uh two short questions

54:44that can go together even though one can

54:47be of long answer never know uh do the

54:50results hold in out of sample analysis

54:55and the other one says or ask so we

54:59don't need database U past information

55:03only expectations

55:06matter so it's two different one that

55:09that ask if the results hold in out of

55:11sample yeah and the other more a

55:14statement to say okay we should not need

55:18past information what matters his

55:20expectations in terms of the

55:24analysis well let me answer them uh

55:27sequentially so on uh on on the first

55:33question there is one kind of analysis

55:36that we could perform out of sample if

55:38you like but it's out of sample in a

55:40funny way which is to go from time

55:45series to cross-section so the puzzles

55:48the model the the the theory were

55:52developed as motivated by shill

55:56and uh motivated by these

55:5940-year-old debates about the aggregate

56:02stock price volatility it's not at all

56:05obvious it wasn't at all obvious to us

56:08that once we uh turn to the

56:11cross-section uh that the same

56:14specifications are going to work uh but

56:18they did so it's not out of sample in

56:20terms of looking for one period of time

56:23and then you know remember we don't have

56:25that much data because these

56:27expectations variables uh don't go that

56:30far actually as a small digression but I

56:34think it may be interesting

56:37digression uh now I hope I don't turn it

56:39into an all long digression um there are

56:43now uh people who are trying to use AI

56:48methods and text

56:51analysis uh that uh produce

56:55measures of

56:58expectations not directly from

57:01surveys but use the Train the model

57:05based on the data which is

57:08contemporaneous and then go back in

57:11history uh for decades maybe even a

57:14century to try to construct measures of

57:19expectations uh in this artificial way

57:22assuming that there is some stability in

57:24in the model so AI has completely

57:28revolutionized uh this measurement and

57:31so there is a there is a gentleman

57:33called Leland BBY who is on the market

57:36from from Yom who done that uh uh for

57:41for for some of the related analysis and

57:45has in fact shown some level of

57:47robustness where you go back in history

57:50where you do not have direct measures of

57:52survey um expectations so those are the

57:55kinds of things that one can do um I

57:59don't think we've done to gone to other

58:01countries uh which is in principle

58:03possible and certainly we haven't gone

58:05forward uh because you know the future

58:08doesn't we don't know the future yet uh

58:11so I think you could do many many things

58:13and I think AI is very uh very promising

58:17on the second question I you know again

58:21um I think we're at the very beginning

58:24of this exploration and I want to repeat

58:28what I said at the you know at the

58:30beginning I mean I'm going to I'm going

58:33to since this is the end I'm going to be

58:35a little radical you know Finance has

58:38been in the desert for 40 years you know

58:41uh We've uh Schiller posed the most

58:45fundamental

58:46challenge uh for our field and then in

58:50response to that challenge we somehow

58:53made the wrong turn

58:55and uh decided that everything is time

58:58varying discount rates and so on and

59:01we've been on that path in the wrong

59:03direction uh for a very very long time

59:07uh what I think this paper shows as well

59:11as a lot of work by other people not

59:14just us by D and Myers by Stefan Nagle

59:18uh by I think there is you know I think

59:21the QJ has received three new papers on

59:24uh expectations and financial markets in

59:27the last week which is uh quite a bit uh

59:31what I think this another papers show is

59:35uh you know maybe there is another turn

59:37we should have taken uh some years ago

59:41and trying to explore this uh this very

59:45different way of understanding uh

59:47returns so I think that uh ironically I

59:51think we need need more data rather than

59:53less data

59:55because these data have been uh to a

59:59very substantial extent uh neglected so

01:00:03I'm extremely

01:00:04optimistic um but I feel that uh you

01:00:08know what we need is a turnout

01:00:10turnaround first uh to get ourselves

01:00:14away from uh discount factors and uh

01:00:18back to

01:00:20reality thank you very much um we are at

01:00:234:00 4 P.M in central eastern time and

01:00:2710:00 a.m. in in the US East Coast uh so

01:00:31we need to to finish there is still some

01:00:34questions uh that I will share uh then

01:00:36later with Professor slier uh and what I

01:00:40want to to say is thank you very much

01:00:43for being with us today with this vast

01:00:46audience uh we had um

01:00:49260 colleagues online uh attending this

01:00:53this presentation

01:00:55so thank you very much for your

01:00:57availability our h of This research

01:01:00network of universities is to

01:01:02disseminate uh research to Regions areas

01:01:06universities that for obvious reasons

01:01:08they might not be able to

01:01:10invite uh academics like Professor Scher

01:01:13so we really appreciate you spend the

01:01:16time with us for the attendees thank you

01:01:19very much for being with us today you

01:01:21can follow us in LinkedIn regarding uh

01:01:24further activities of the research

01:01:26Network and also uh the the the the

01:01:30seminars in the future also the FMA

01:01:32website that we are very pleased to work

01:01:35together with these seminars thank you

01:01:37very much Professor for being with us

01:01:39today thank you this was really

01:01:42wonderful thank you thank you very much

01:01:44thank

01:01:53you

ðŸŽ¥ Related Videos

ðŸ”¥ Recently Summarized Examples