Emotions Affect Markets in Predictable Ways: Behavioral Finance and Sentiment Analysis

H2O ConsultingCristian Bissattini, MBA

H2O Consulting

2

H2O Consulting

Lugano (Switzerland)www.h2oconsulting.ch

Inn

ovati

on

fro

m S

wis

s t

rad

itio

n

The Prize in Economic Sciences 2013

3

There is no way to predict the price of stocks and bonds over the next few days or weeks. But it is quite possible to foresee the broad course of these prices over longer periods,

such as the next three to five years.

Eugene F. FamaUniversity of Chicago

Robert J. ShillerYale University, New Haven

Lars Peter HansenUniversity of Chicago

The Prize in Economic Sciences 2013

4

5

Neoclassical Finance Model

CA

PM

Por

tfolio

Prin

cipl

es

Opt

ion

pric

ing

Arb

itrag

e pr

inci

ples

Modigliani & Miller

Markowitz

Sharpe, Lintner, Black Black,

Sholes, Merton

Neoclassical Finance Model

All investors are rational, well-informed

and hope for maximizing profit

Market prices immeditely refllect all available information

6

All private information

All public information

Information in past stock

prices

Efficient Market Hypothesis

Weak form

Semi-strong form

Strong form

7

Neoclassical Finance Model

Markets are normally distributed with daily stock return lying under their theoretical bell shaped curve

Stock prices reflect the discounted value of expected cash-flows

It is not possible to beat the market over time without taking excess risk

Sentiment does not play a role in this classic framework.

Neoclassical Finance Model

8

The value of a stock is the present value of all of the expected future dividends.

Gordon Dividend Discount Model (DDM)

Zero growth Constant growth Differential growth

Neoclassical Finance Model

9

$3 dividend per share

a year from today

10% dividend expected growth rate per year

(foreseeable future)

15% required return

(iPear’s risk)

$3

0.15 - 0.10= $60

Constant Growth Scenario

Share Price

An investor is considering the purchase of a share of the iPear Inc.

Neoclassical Finance Model

Oct

ob

er

19

87

19

90

s

19

29

Oct

ob

er

20

08Financial

Turmoil Internet bubble

Black Monday Crash Great

Crash

10

Neoclassical Financial Model is unable to explain extreme cases of bubbles and crashes

It seems timely to define a human sentiment function in stochastic discount factor (SDF)

Neoclassical Finance Model

11

Neoclassical Finance Model

12

Adherents of geometric Brownian motion or log normally distributed stock returns (one of the foundation blocks of modern finance) must ever after face a disturbing fact: assuming the hypothesis that stock index returns are log normally distributed with about a 20% annualized volatility, the probability that the stock market could fall 29% (the decline in S&P futures on October 19th, 1987) in a single day is 10-160. So improbable is such an event that it would not be anticipated to occur even if the stock market were to last for 20 billion years, the upper end of the currently estimated duration of the universe. Indeed, such an event should not occur even if the stock market were to enjoy a rebirth for 20 billion years in each of 20 billion big bangs.

 Mark Rubinstein

in Comments on the 1987 Stock Market Crash

Behavioral Finance Theory

13

Psychology

Economics

People in standard finance are rational. People in behavioral

finance are normal

If you don’t know who you are, the stock market is an expensive

place to find out

Meir Statman Professor of Finance (Leavey School of Business, Santa Clara University)

Adam SmithScottish moral philosopher and a pioneer of political economy

Behavioral Finance Theory

14

Brain’s biological and physiological

limits

Simplification of reality

Approximation of information

(heuristics and cognitive filters)

Errors and biases

Behavioral Finance Theory

15

Germany Austria

Donate organs Don’t donate

12% 1%

12% Donors 99% Donors

Opt-in system vs Opt-out system

The Role of Inertia

Behavioral Finance Theory

16

Gain

Uncertainty

Risk

Fear

Behavioral Finance Theory

17

In the Mind of the Market: Theory of Mind Biases Value Computation during Financial Bubbles (Benedetto De Martino)

Keynesian Beauty Contest Theory

18

1

2 3

Robert J. Shiller (1981)

19

20

Behavioral Finance Theory

Risk Perception

Prospect Theory

Cognitive Errors

Behavioral Finance Theory

Source: based on the risk formula by risk researcher Dr. Peter M. Sandman.

21

Risk Aversion vs Seeking Aversion

22

Option BOffer: €125’000

Option A50% Win €250’000

50% Win €0

Expected Value of option A is: 0.5 · €250’000 = €125’000The two options are equivalent in term of expected value

78% of people choose option B Risk Aversion!

Source: Psicologia e Investimenti FinanziariPaolo Legrenzi (2006)

Risk Aversion vs Seeking Aversion

For the same expected value, are investors always risk-averse?

23

Risk Aversion vs Seeking Aversion

24

Option BCertain Loss: €125’000

Option A50% Loss €250’000

50% Loss €0

Expected Value of option A is: 0.5 · -€250’000 = -€125’000The two options are equivalent in term of expected value

61% of people choose option A! Risk Seeking!

Source: Psicologia e Investimenti FinanziariPaolo Legrenzi (2006)

Risk Aversion vs Seeking Aversion

For the same expected value, are investors always risk-averse?

When we are faced with a sure gain

we tend to be risk averse

but

When we are faced with a certain loss

we tend to be risk seeking

25

Prospect Theory

26

The combination of risk-aversion with risk-seeking is represented by the value function

- 10+20

- 20+10

27

Cognitive Errors

Overconfidence Anchoring Representativeness Loss aversion

Regret minimizing Frame dependence Defense mechanisms

Behavioral Finance Theory

28

Sources: www.forrester.com/findresearchBlackRock

29

Our Strategic Concept

Modern Portfolio Theory (MPT)

Markowitz (1952)

Behavioral Portfolio Theory (BPT)

Shefrin, Statman (2000)

H2O Consulting launches RiskAdvisor® platform that combines Modern Portfolio Theory with Behavioral Portfolio Theory

30

Sentiment AnalysisH2O Consulting

Università della Svizzera italiana

Sentiment Analysis

31

Trust Calculation

Sentiment

Individual Recommendation

AggregationSocial Media

Online News

Message Board

Sources Social Intelligence

Web Crawling Technology

Data Processing

Semantic Analysis

Classification

Algorithm

Sentiment Analysis

32

Our dataset consist of 447’393 messages, on the 30 Dow Jones Index (DJIA) stocks,

posted on the Yahoo! Finance message board in the period August 2012 to May 2013,

of which 55’217 with sentiment tag and 5’967 distinct authors.

Trust Calculation

33

Period from August 28, 2012 to October 23, 2013, on the 30 Dow Jones Index (DJIA) stocks

****t_

suck

z

****o

rking

man

****b

uco2

012

****_

refu

nd

****le

rs_n

ightm

are

****te

lkins

@at

t.net

****p

s4_b

est

****_

a_do

g

****b

ucco

2112

****h

lion2

001

0.70

0.75

0.80

0.85

0.90

0.95

Microsoft Corp (MSFT)

****t_suckz (MSFT)

****tmimi (BAC)

****ab33 (INTC)

****joiner (BAC)

****_refund (MSFT)

0.78 0.80 0.82 0.84 0.86 0.88 0.90 0.92 0.94

0.916

0.887

0.876

0.876

0.876

0.875

0.875

0.844

0.832

0.828

Top 10 (DJIA)

A novel way to generate sentiment based on author’s credibility

calculated on accuracy of his past messages

Empirical Validation

From August 28, 2012 to May 16, 2013

on the 30 DJIA stocks

*** p-value < 0.001 ** p-value < 0.01 * p-value < 0.05Coefficients are reported in basis points (0.01%)

 3-scale index model

(Weighted)5-scale index model

(Weighted)

StockN° Observations(Trading Days)

AdjR-Square

AdjR-Square

MMM 34 -2.5 0.69 3.9 0.73

AA 152 24.1 0.40 11.2 0.38

AXP 30 -10.9 0.33 -0.99 0.35

T 162 37.3*** 0.40 23.7*** 0.41

BAC 174 131.9*** 0.46 55.7*** 0.46

BA 172 48.2** 0.21 18.3* 0.19

CAT 168 37.3* 0.50 23.2* 0.51

CVX 110 3.9 0.55 4.3 0.57

CSCO 153 22.9 0.12 11.8 0.14

DD 80 17.2 0.37 12.3 0.39

XOM 147 1.3 0.75 2.7 0.65

GE 90 20.0 0.24 4.6 0.27

HPQ 174 119.8** 0.14 56.7** 0.16

HD 97 3.2 0.23 -2.7 0.24

INTC 174 90.1*** 0.38 40.3*** 0.35

IBM 139 6.0 0.17 6.2 0.19

JNJ 104 -11.1 0.35 -6.1 0.36

JPM 155 27.7** 0.62 13.4** 0.62

MCD 113 18.6 0.37 7.8 0.35

MRK 89 19.0 0.05 4.3 0.05

MSFT 174 116.4*** 0.52 53.9*** 0.52

PFE 155 38.5*** 0.35 20.9*** 0.38

PG 66 0.9 0.31 5.0 0.35

KO 110 9.5 0.29 8.1 0.28

TRV 12 N/A N/A N/A N/A

UTX 62 20.6 0.50 13.0* 0.50

UNH 32 2.3 0.31 3.8 0.40

VZ 127 6.8 0.26 6.4 0.27

WMT 170 52.6*** 0.23 27.8*** 0.24

DIS 82 -1.9 0.23 -2.6 0.23

34

Empirical Validation

35

 3-scale index model (Weighted) 5-scale index model (Weighted)

Stock N° Observations(Trading Days)

AdjR-Square

AdjR-Square

T 162 37.3*** 0.40 23.7*** 0.41

BAC 174 131.9*** 0.46 55.7*** 0.46

BA 172 48.2** 0.21 18.3* 0.19

CAT 168 37.3* 0.50 23.2* 0.51

HPQ 174 119.8** 0.14 56.7** 0.16

INTC 174 90.1*** 0.38 40.3*** 0.35

JPM 155 27.7** 0.62 13.4** 0.62

MSFT 174 116.4*** 0.52 53.9*** 0.52

PFE 155 38.5*** 0.35 20.9*** 0.38

WMT 170 52.6*** 0.23 27.8*** 0.24

Empirical Validation

36

3-scale index model (Weighted)

5-scale index model(Weighted)

Stock N° Obs.(Trading Days) N° posts

BAC 174 10’090 131.9*** -99.0** 55.7*** -44.9**

HPQ 174 5’146 119.8** -93.6* 56.7** -49.8*

INTC 174 6’545 90.1*** -53.6** 40.3*** -19.3

MSFT 174 8’337 116.4*** -31.6* 53.9*** -13.6*

Empirical Validation

37

Bull Market

Investors are

optimistic

Sentiment index raises

Human behavior (prospect theory)

Lock in gains

Market peak

Empirical Validation

38

3-scale index model

(Weighted)5-scale index model

(Weighted)

StockN° Obs.(Trading

Days)N° posts

BAC 174 10’090 131.9*** -99.0** 55.7*** -44.9**HPQ 174 5’146 119.8** -93.6* 56.7** -49.8*INTC 174 6’545 90.1*** -53.6** 40.3*** -19.3MSFT 174 8’337 116.4*** -31.6* 53.9*** -13.6*

1)

2)

Empirical Validation

39

3-scale index model 5-scale index model

Stock N° Observations(Trading Days)

T 162 37.3*** 25.0 23.7*** 12.6

BAC 174 131.9*** 60.4 55.7*** 33.2

BA 172 48.2** 16.5 18.3* 7.9

CAT 168 37.3* 20.6 23.2* 16.2

HPQ 174 119.8** 102 56.7** 62.0

INTC 174 90.1*** 82.5* 40.3*** 46.8*

JPM 155 27.7** 26.0* 13.4** 16.4**

MSFT 174 116.4*** 101.4** 53.9*** 57.4**

PFE 155 38.5*** 32.6** 20.9*** 21.7**

WMT 170 52.6*** 0.12 27.8*** -0.4

Sentiment Trading Strategy

40

If Sentiment at trading day t is

greater than Upper Limit

If Sentiment at trading day t is

lower than Lower Limit

BUY

SELL

  3-scale index model 5-scale index model

Upper Limit 0.97 1.00

Lower Limit -0.83 -1.70

Upper and lower limits have been estimated through a best-fitting process on time series, with proprietary genetic algorithms.

August 28, 2012 May 16, 2013

$1 million

Sentiment Trading Strategy

41

From August 28, 2012 to May 16, 2013 (Initial Investment $1 million)

Sentiment Trading Strategy

42

From August 28, 2012 to May 16, 2013 (Initial Investment $1 million)

Sentiment Trading Strategy

Portfolio Expected Return (CAPM): 24.1% ($241K)

From August 28, 2012 to May 16, 2013 (Initial Investment $1 million)

S&P500: 17.1%Risk-free: 0%

Beta (portfolio): 1.41

Can we build an active investment strategy, using our sentiment trading rule and source of information,

in order to generate greater risk-adjusted returns than a passive, naïve, yet achievable, investment strategy?

Yes. We can!43

Publications / About us

44

http://ssrn.com/abstract=2309375

H2O Sentiment Analysis

45

Instantly capture human emotion in financial markets as it happens.

Sentiment Analysis

46

Track Real-Time Sentiment Analysis On Your

Mobile Device

H2O Consulting © 2013 All Rights Reserved

Thank You for Your Attention