Guiding seminar 1 (2020)

Question 1

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What are the two reasons why consumers rely less on expert advice and turn to fellow customers instead?

Answer 1

i) emergence and proliferation of social media
platforms;
ii) creation and consumption of user-generated content
-> consumers turning to fellow customers when choosing among products instead of relying on expert advice.

Question 2

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What were the two research questions the authors researched?

Answer 2

1) Do peer opinions actually transmit value-relevant information (hasn’t been incorporated into prices yet)? (or are they just “random chatter”, and we rather leave the task to professionals?) -> is information incorporated into prices?
2) Are some users attempting to intentionally spread false information to mislead “fellow customers”? -> or is it false information to manipulate the prices?

Question 3

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What are the two channels to voice one’s opinion on seekingalpha.com?

Answer 3

There are two channels through which investors can voice their opinions:

1) Opinion articles (reviewed by a panel, subject to editorial changes)
2) Commentaries written in response to articles, other users sharing their views

Sample: 2005-2012, articles (single-ticker) and commentaries written by around 6,500 and 180,000 different users, respectively, covering more than 7,000 firms.

Question 4

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What methodology was used by the authors?

Answer 4

Textual analysis: the frequency of negative words (as a fraction of the total word-count) used in an article/commentary captures its tone (eg., bad, overvalued..)

The finance-specific negative word list compiled by Loughran and McDonald (2011) has been used
NegSAi,t – average fraction of negative words across all single-ticker articles published on SA about company i on day t -> if articles have an impact -> article sentiment will explain significant amount of abnormal returns
NegSA-Commenti,t – average fraction of negative words across all SA comments posted in response to single-ticker articles about company i on day t -> fraction of negative words that represents the segment

Question 5

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What are the two reasons for the relationship between negative words in an article and negative abnormal returns for the stock?

Answer 5

1. Predictability channel- the articles contain value-relevant information, which is not yet reflected in the price of the stock –> market participants learn the information and adjust the prices accordingly (ex., Apple stock in China)
2. Clout channel- SA views reflect false and spurious information–> SA readers are naive and trade in the direction suggested, which is unlikely due to
   • > (i) no return reversal and
   • > (ii) capital constraints of SA followers

Question 6

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What is an earnings surprise? And how are they connected with the authors findings?

Answer 6

Earnings surprise: difference between the reported quarterly Net income and the average Net income forecasts across all equity analysts following the company

If there is information about earnings that means that the earning surprises must be influenced. If we assume that naive investors drive the price, they can impact earnings but not the earning surprises of a single company. Nevertheless, the new regression showed that the comments have an impact on the earning surprises.

Question 7

Wisdom of Crowds: The Value of Stock Opinions
Transmitted Through Social Media

What incentives informed market participants have to share their value-relevant insights?

Answer 7

• Striving to become celebrities: Users derive significant utility from the attention and recognition they receive from others when their opinions are confirmed by the stock market. Occasionally SA contributors are even referred to in such prominent outlets as Forbes, WSJ-Marketwatch, and Morningstar
• Money: Each SA contributor earns $10 per 1,000 page views that his/her article receives, and authors with good track record attract more followers -> checked by number of pages viewed and number of times the article is read to the end -> this way followers can differentiate between authors and their “popularity”
• Feedback system: other users can intervene and correct bad articles, which further discourages attempts of misinformation -> checked by author/follower disagreement -> disagree with the author more if the articles have been inconsistent
• Convergence to fundamental value: if SA users can move stock prices, authors may want to incentivize the convergence of market prices to what they perceive
to be the fair fundamental values -> if people can drive prices prices will go up to the authors perceived fundamental value

Question 8

The U.S. Equity Return Premium: Past, Present, and
Future

What is an equity premium puzzle?

Answer 8

Definition: for more than a century, diversified long-horizon investments in America’s stock market has consistently received higher returns (almost by 6% p.a. on average) than those in bonds with no more risk (bearing the same risk).

In the short run equity offers more in 65% of the time, while in the 20 year horizon - 91% of the time. -> maybe it could be the liquidity -> but then the puzzle should be less pronounce for long term bonds with lower liquidity -> but it is more pronounce (only in 2% of the cases bonds outperformed)

Question 9

The U.S. Equity Return Premium: Past, Present, and
Future

Why do more people invest in T-bills or bonds rather than equities if they outperform bonds?

Answer 9

1. Utility theory (marginal utility of wealth differs for gains and losses) -> one dollar loss would be 170 times more painful as 1 dollar gain.
   
   • > Expected utility theory: agents should be risk averse on all bets that do not involve their overall wealth -> stock returns do not covariate with current consumption and lifetime wealth
2. Risk aversion (rather make certain gains as opposed to uncertain ones).People on average dislike risk. -> High risk aversion should lead to high risk free-rate- but in reality risk free rate very low –> risk-free rate puzzle
3. Loss aversion (losing money brings more pain compared to the happiness of gaining money –> prospect theory). -> people have biases ->
   
   3. 1 Myopia (difficult to fins discount rate (pay more attention to short term rates).
      - > people know that they have biases -> explanation should also account for inability to deal with the biases
4. Transaction costs and investor heterogeneity (most people do not trade at all) -> risk bearing capacity is constrained (borrowing constrains);
5. Uninsurable idiosyncratic income shocks correlated with the market (e.g. (2) equity premium is included into human capital -> unwilling to increase equity exposure (being fired during a recession)
6. Unknown true lower-tail risk -> not representative sample: (1) catastrophic events(mortgage crisis); (2) black swans.
7. Learning about the return distribution- investors and regulators misread the riskiness of equities in the early 20th century, then they adjusted their expectations–> prices rose even more and not they overstated the equity premium –> we can expect the equity premium to decrease as market participants learn about the return distribution of equities

Question 10

The U.S. Equity Return Premium: Past, Present, and
Future

What is the future of the equity premium?

Answer 10

• Many Wall Street observers agree that a substantial equity premium remains as of today and will persist in the future
• Equity premium forecasts: financial economists: 6-7% over the next 10-30 years; CFOs : 3.2%; authors’ estimates: 2.55-4.33%
• It is reasonable to believe that equity premium will likely continue, though at a lower rate than historically – perhaps at around 4% p.a. instead of 6% (institutional changes (ERISA), interest in profiting from equity premium, faded memory from the Great Depression)

Question 11

Two Pillars of Asset Pricing

What are market efficiency tests/ Joint hypothesis problem (JHP)? What are the three forms of market efficiency?

Answer 11

3 forms of market efficiency:
▪ Weak: prices incorporate only the information from past price movements (invalidates technical
analysis)
▪ Semi-strong: all publicly-available information is incorporated into prices (past and current) -> seeking alpha
▪ Strong: prices incorporate all available information (public and non-public) -> may lead to insider trading

Market efficiency tests: comparing how asset prices should behave to the way they actually behave. Model how they should behave with an asset pricing model. If your tests reject market efficiency:
▪ Either the financial market in question is inefficient
▪ Or your asset pricing model is no good
This issue is called the joint hypothesis problem (JHP) and to date remains an unresolved
conundrum in asset pricing

Question 12

Two Pillars of Asset Pricing

What does Fama find about market efficiency through event studies?

Answer 12

Event studies: In an efficient market, stock prices adjust promptly and accurately to new information (no further meandering or reversals) -> stock splits signals to the market but the market already knows -> prices do not move after the stock split
-> stock splits -> good performance -> becomes more liquid
▪ Fama finds that all stock-split related (positive) information is incorporated into prices months
before the split, corroborating market efficiency
▪ With short event windows, the JPH is rendered relatively unimportant. Over long-term horizons, it’s back in the spotlight -> changes in the interest rates are predicting changes in inflation-> stock return reversal -> inflation increases -> either good performance or dividends are worth noting -> model or the market?

Question 13

Two Pillars of Asset Pricing

What does Fama find about market efficiency through predictive regressions?

Answer 13

Can expected inflation determine interest rates? –> bond and real estate prices already incorporate the best possible forecast for inflation
Expected inflation is negatively related to stock returns.

Question 14

Two Pillars of Asset Pricing

What does Fama find about market efficiency through time-varying expected stock returns?

Answer 14

Time-varying expected stock returns. Investors’ capacity and willingness to bear risk as well as the risk itself are not constant over time. This leads to time-varying expected returns and explains (according to rationalists) a lot of volatility in stock prices, which many behavioralists (prominently Shiller) attribute to investor irrationality

Question 15

Two Pillars of Asset Pricing

What does Fama find about market efficiency through bubbles?

Answer 15

Bubbles: irrational price increases that lead to predictable declines. Fama doesn’t believe in bubbles. Fama claims that:
▪ No price declines are ever predictable (50/50 guessing game)
▪ Price declines are not irrational because they are driven by slowdowns in real economic activity Fama denounces behavioural finance for not offering a viable alternative but merely criticising the existing models. Fama: “Which leg of a “bubble” is irrational: the up or the down?” (i.e. irrational optimism vs. irrational pessimism)
Effient market hypothesis –> cannot predict price decreases and cannot anticipate bubbles

Question 16

Two Pillars of Asset Pricing

What are the two types of asset pricing models?

Answer 16

Asset Pricing Models:
• Standard models work forward from assumptions about investor tastes and investment opportunities (CAPM and its variations)
• Empirical models work backwards – they take as given the patterns in average returns, and propose other variables that capture them (FF factor models, APT)

Question 17

Behavioral Economics: Past, Present, and Future

What is the main problem with the current theory?

Answer 17

One theory, two tasks: relying on one theory to both characterise optimal behaviour and predict actual behaviour (supposedly irrelevant factors, or SIFs, determining behaviour) –> one theory cannot do both -> A.Smith: “ Certain factors (myopia) might have an impact which can cause your behaviour to deviate from the optimal”

Question 18

Behavioral Economics: Past, Present, and Future

What are the characteristics of a Homo economicus (rational behavior)?

Answer 18

Idealized model of Homo economicus:
• Agents have well-defined preferences and unbiased beliefs and expectations (infinite cognitive
abilities)
• They make optimal choices based on these beliefs and preferences (infinite willpower)
• Their primary motivation is self-interest

-> models just show how to optimise but it does not mean that the agents will do it

Question 19

Behavioral Economics: Past, Present, and Future

What are explain-away-tions?

Answer 19

“Explain-away-tions” (“explain and go away”):
Models of rational behavior became standard because they were easier to solve (not meant as a
put-down: “one begins learning physics by studying objects in vacuum; atmosphere can be added
later, but its importance was never denied by physicists”)
Models of rational behavior are good in some cases, but they miserably fail in others (tic-tac-toe vs.
chess – agents are expected to act as if they understood the complicated model)

Question 20

Behavioral Economics: Past, Present, and Future

Refute the claim: We should judge theories based on not whether they describe but whether they predict
behavior – expert billiard player example

Answer 20

Refute: But what about non-experts? Moreover, even experts are unable to optimize when the
problems are difficult (e.g. chess vs tic tac to)

Question 21

Behavioral Economics: Past, Present, and Future

Refute the claim: The errors of humans are randomly distributed with the mean of zero, cancelling out when
individual observations are aggregated; non-experts just have more noise (higher variance)

Answer 21

Refute: humans make judgements that are systematically biased, which could be predicted using a theory of human cognition; by utilising clever framing, a majority of subjects can be induced to select a dominated pair of options (prospect theory: decision making under uncertainty – A)
riskless payoff vs. B) risky payoff tests) -> think of airlines(priming effect form past experience); or drugs (90% success vs 10% loss)

Question 22

Behavioral Economics: Past, Present, and Future

Refute the claim(incentives and learning): agents should have higher stakes and pay more attention or they can choose a few times and learn from previous mistakes (big vs. small decisions – not both)

Answer 22

Refute 1: no learning in the Heads-Tails experiment due to the lack of feedback and false reinforcement; milk buying
Refute 2: “preference reversal” – of those who preferred the p-bet, a majority reported a higher selling price for the $-bet, implying that they valued it more (mug vs book example)

Question 23

Behavioral Economics: Past, Present, and Future

Refute the claim the “invisible handwave” – when agents interact in a market environment, any tendencies to misbehave will be vanquished (errors vanish)

Answer 23

Refute: how is the market going to help? For businesses it is easier and more profitable to cater to biases than to eradicate them ($70 unlimited monthly fee vs. 10 visits for $100, house mortgages and the recent financial crisis, retirement savings)

Question 24

Behavioral Economics: Past, Present, and Future

What is the efficient market hypothesis? And what are the two components of EMH?

Answer 24

The activities of “smart money” arbitrageurs can assure that markets behave “as if” everyone is smart

Two EMH components:
“No free lunch”: not possible to beat the market on a risk-adjusted basis (the active mutual fund
industry does not beat the market on average)
“Price is right”: difficult to test as the security fundamental value is not known – use the law of one price (CUBA closed-end mutual fund trading at a huge premium relative to its net asset value)

Question 25

Behavioral Economics: Past, Present, and Future

What are the two issues of rationalist theories?

Answer 25

Policymakers shouldn’t assume that EMH holds to avoid bubbles
There isn’t much confidence that prices in other markets with no easy short selling opportunities are good measures of value (e.g. labor power – can workers become significantly less productive by changing industries)

Question 26

Behavioral Economics: Past, Present, and Future

What is prospect theory?

Answer 26

Prospect theory:
• Utility is derived from changes in wealth relative to some reference point rather than levels of wealth
• The “value function” has a kink at the origin with losses weighing more heavily than gains (“loss
aversion”)
• Decision weights are a function of probabilities

Hence, people are not rational and should include bahavioral theory in the models.

Question 27

Behavioral Economics: Past, Present, and Future

What is intertemporal choice?

Answer 27

Intertemporal choice (“myopia”) – the discount rate between “now” and “later” is much higher than between “later” and “even later” (i.e. people are impatient)

Question 28

Behavioral Economics: Past, Present, and Future

What are the alternative models to the rational models?

Answer 28

1. Intertemporal choice (“myopia”) - people are impatient
   • Two-self, or “two-system” approach, which helps to overcome the paradox of long-sighted
   planner and “myopic” doer (elements of both- rational and myopic behavior)
   • “Beta-delta” model, where delta is the standard exponential discount rate and beta measures
   short-term impatience (portable extension to existing models – standard model is a special case
   with a beta of 1)
2. Prospect theory (losses weigh more than gains)
3. “Other-regarding preferences” models: humans are not completely selfish, even to strangers (e.g.
   one-shot prisoner’s dilemma, public goods games)

Question 29

… and the Cross-Section of Expected Returns

What do the authors research?

Answer 29

There have been hundreds of papers trying to explain almost the same cross- section of expected returns since one of the first CAPM tests forty years ago.
Thus the usual t-statistics cut-off is insufficient to compensate for the extensive data mining.Our paper introduces a new multiple testing framework and provides historical cutoffs from the first empirical tests in 1967 to today

RQ:

(1) Is there still a valid relation between average return and β? (Does “CAPM” still” apply)
(2) Are other risk factors elated to average returns?

Question 30

… and the Cross-Section of Expected Returns

Three reasons for why criteria today should be tougher?

Answer 30

1. The rate of discovering a true factor has likely decreased
2. There is a limited amount of data
3. The cost of data mining has dramatically decreased

Question 31

… and the Cross-Section of Expected Returns

What is Multiple testing?

Answer 31

Multiple testing: “In Statistics, multiple testing refers to the potential increase in Type I error that occurs when statistical tests are used repeatedly, for example while doing multiple comparisons to test null hypotheses stating that the averages of several disjoint populations are equal to each other (homogeneous).

Question 32

… and the Cross-Section of Expected Returns

What is Type I error rate and Type II error rate?

Answer 32

Type I error rate: ‘the probability of finding a factor to be significant when it is not; a value α is used to control type I error rate”.
In a multiple testing framework, restricting each individual test’s type I error rate at α is not enough to control the overall probability of false discoveries. Thus, we need measures of the type I error that help us simultaneously evaluate the outcomes of many individual tests.

Type II error rate: the probability of missing true factors —> one needs to set p- value to balance type I and type II error rates

Question 33

… and the Cross-Section of Expected Returns

What is Family-wise error rate and False discovery rate

Answer 33

Family-wise error rate: the probability of at least one type I error; measures the probability of even a single false discovery, regardless of the total number of tests
FWER=Pr(N0|r ≥1).

False discovery rate: The false discovery proportion (FDP) is the proportion of type I errors.
FDR = E[FDP] —> measures the expected proportion of false discoveries among all discoveries

Question 34

… and the Cross-Section of Expected Returns

What is the sample?

Answer 34

Sample: 313 articles that propose and test new factors (250 published, 316 factors in total). The sample is likely to underestimate the factor population —> use multiple testing framework

Question 35

… and the Cross-Section of Expected Returns

Which five factors survive the test?

Answer 35

The 5 factors from a sample of most prominent factors surviving all three adjustments are:
1. HML - book-to-market
2. MOM - momentum
3. DCG - durable consumption goods
4. SRV - short-run volatility
5. MRT - market beta
—> we need a much higher t-statistic threshold when multiple testing is taken into account

Question 36

… and the Cross-Section of Expected Returns

What is the conclusion?

Answer 36

The authors argue that a newly discovered factor today should have a t- statistic that exceeds 3.0 (a p-value of 0.27%)
Of the 296 published significant factors, 158 would be considered false discoveries under Bonferonni, 142 under Holm, 132 under BHY (1%), and 80 under BHY (5%). In addition, the idea that there are so many factors is inconsistent with the principal component analysis, where, perhaps there are five “statistical” common factors driving time-series variation in equity returns
‘A case can be made that a factor developed from first principles should have a lower threshold t-statistic than a factor that is discovered as a purely empirical exercise. Nevertheless, a t-statistic of 2.0 is no longer appropriate— even for factors that are derived from theory.’

Question 37

Why many of the historically discovered factors would be deemed “significant” by chance”?

Answer 37

Type I error —> false positive —> declaring that a factor is significant when in fact it’s not
To control for Type I, set a p-value, e.g. 5%. Then one could say that 5% is the chance that the observed variation is due to chance. When a new factor is proposed, hypothesis testing is performed. Most frequently, the factor testing is performed on more-or less the same data set of returns and the test represents a separate hypothesis testing. However, given how many published factors are proposed and how many more have been tried but not made public, the number of tests can be estimated as quite large (the authors use a sample of around 300 - all these are from established, reputable journals). Now let’s imagine we apply the same p-value of 5% ubiquitously. If there are 100 separate factor tests, one could expect that 5 factors are considered significant by chance. If there are 1000 separate factor tests conducted, the expected ‘false positive’ factors rise to 50. Thus, a bottomline is that the more factors are tested over approximately the same data set, and the p- value is left unadjusted, the higher is the number of factors deemed “significant” by chance.