Mitigating equity market risk with investor sentiment

In conjunction with

Andrey Karpov, Hiroki Shimada, Kenneth Tan

December 2010

Research Objectives

Filling gaps left unexplained by the Market

Equilibrium Theory

Assumptions:
Sentiment and Market Efficiency do not always
contradict each other
Investor sentiment is over-optimism/pessimism
about future fundamentals
Role of investor sentiment varies market by
market
Investor sentiment is a non-linear phenomenon

Developing a statistical model, based on

investor sentiment, capable of estimating the
probability of extremely negative returns in the
US equity market.

Oxford Sad Business School

Limitations of Research

The sample period of 19 years includes 225

months and only 21 month with extremely
negative returns

Low number of fat tail observations

imposes limitations on the accuracy of
estimation

The scope of our research and the nature of

selected indicators require a trade-off
between the sample size and the combination
of particular variables

The model incorrectly classifies certain non fat

tail events, in particular months with excess
positive returns, and therefore limits the upside
potential of investments

Oxford Sad Business School

Multivariate Regression Analysis

Benchmarking against the S&P500

-

from April 1991 to Dec 2009 (In-sample)

from Jan 2010 to Sep 2010 (Out-of-sample)

From Jan 2007 to Sep 2010 (Out-of sample) w/ 3 variables

Measures
(a) Average Log Returns
(b) Standard Deviation
(c) Sharpe Ratio
(d) Downside semi standard deviation
(e) Sortino Ratio
(f) Kurtosis
(g) Skewness

Asset allocation rules

Allocate 100% of our portfolio to the JP Morgan developed government bond
index, or to the S&P500

Oxford Sad Business School

Multivariate Regression Analysis

Asset allocation rules

cut-off for risk on/off at 9.3%, allocate 100% of our portfolio to the JP
Morgan developed government bond index, or to the S&P500
70.000%

3500

60.000%

3000

50.000%

2500

40.000%

2000

Average

Probability
30.000%

1500

20.000%

1000

S&P 500
Price

10.000%

500

Model

0
01/04/1991
01/12/1991
01/08/1992
01/04/1993
01/12/1993
01/08/1994
01/04/1995
01/12/1995
01/08/1996
01/04/1997
01/12/1997
01/08/1998
01/04/1999
01/12/1999
01/08/2000
01/04/2001
01/12/2001
01/08/2002
01/04/2003
01/12/2003
01/08/2004
01/04/2005
01/12/2005
01/08/2006
01/04/2007
01/12/2007
01/08/2008
01/04/2009
01/12/2009

0.000%

1991-2009 In-sample results

Oxford Sad Business School

Multivariate Regression Analysis

In-sample(1991-2009) Risk Profile

Out-of-sample(2010) Risk Profile

1991-2009

Average Log Returns

St. Dev
Sharpe Ratio
Downside SemiStDev
Sortino Ratio
Skewness
Kurtosis
Jan 2010 - Sep 2010

Average Log Returns

St. Dev
Sharpe Ratio
Downside SemiStDev
Sortino Ratio
Skewness
Kurtosis

Oxford Sad Business School

JPM govt
S&P 500
bond
Model
0.058
0.063
0.150
0.032
0.386
1.955
0.097
0.014
0.595
4.566
-0.908
-0.187
2.077
0.315

0.113
0.087
1.292
0.040
2.813
0.276
1.042

JPM govt
S&P 500
bond
Model
0.029
0.086
0.044
0.211
0.024
0.116
0.136
3.604
0.380
0.112
0.024
0.116
0.255
3.604
0.380
-0.080
0.332
-0.448
-1.684
0.623
0.408

Multivariate Regression Analysis

Distribution Graphs for In-sample 1991-2009

S&P500 Distribution graph of Returns

Oxford Sad Business School

Model Distribution Graph of Returns

Multivariate Regression Analysis

2010 Out-of-sample results

1991-2009

1250

S&P 500

1200
1150
1100

JPM govt
bond
Model

Average Log Returns

0.058

0.063

0.113

St. Dev

0.150

0.032

0.087

Sharpe Ratio

0.386

1.955

1.292

1050

S&P

Downside SemiStDev

0.097

0.014

0.040

1000

Model

Sortino Ratio

0.595

4.566

2.813

-0.908

-0.187

0.276

2.077

0.315

1.042

Skewness

950

Kurtosis
Jan 2010 - Sep 2010

S&P 500
Date

S&P 500 Price S&P 500 log Returns z

1/29/2010
1073.87
-3.768%
2/26/2010
1104.49
2.811%
3/31/2010
1169.43
5.713%
4/30/2010
1186.69
1.465%
5/31/2010
1089.41
-8.553%
6/30/2010
1030.71
-5.539%
7/30/2010
1101.6
6.652%
8/31/2010
1049.33
-4.861%
9/30/2010
1139.42
8.24%

Oxford Sad Business School

p
-1.599
-1.683
-1.677
-1.221
-1.264
-1.997
-0.833
-0.956
-1.224

5.491%
4.617%
4.675%
11.109%
10.303%
2.289%
20.236%
16.958%
11.045%

JPM govt
bond
Model

Average Log Returns

0.029

0.086

0.044

St. Dev

0.211

0.024

0.116

Sharpe Ratio

0.136

3.604

0.380

Downside SemiStDev

0.112

0.024

0.116

Sortino Ratio

0.255

3.604

0.380

Skewness

-0.080

0.332

-0.448

Kurtosis

-1.684

0.623

0.408

Multivariate Regression Analysis

Out of Sample Testing (2007-10) using 3 variables

Oxford Sad Business School

Multivariate Regression Analysis

Out of Sample Testing (2007-10) using 3 variables

1991-2006

1800
1600

S&P 500

1400

JPM govt
bond

Model

1200

Average Log Returns

0.084

0.066

0.110

1000

St. Dev

0.137

0.031

0.088

Sharpe Ratio

0.617

2.077

1.252

800
600

S&P

Downside SemiStDev

0.082

0.013

0.040

400

Model

Sortino Ratio

1.024

4.872

2.719

-0.669

-0.267

0.393

1.498

0.425

1.511

Skewness

200

Kurtosis
01/01/2007
01/04/2007
01/07/2007
01/10/2007
01/01/2008
01/04/2008
01/07/2008
01/10/2008
01/01/2009
01/04/2009
01/07/2009
01/10/2009
01/01/2010
01/04/2010
01/07/2010

Jan 2007 - Sep 2010

S&P 500
Average Log Returns

Model

-0.058

0.059

0.039

0.204

0.034

0.052

-0.286

1.744

0.748

0.142

0.034

0.052

Sortino Ratio

-0.411

1.744

0.748

Skewness

-0.760

0.052

-1.004

0.674

0.317

5.089

St. Dev

Sharpe Ratio
Downside SemiStDev

Kurtosis

Oxford Sad Business School

JPM govt
bond

Multivariate Regression Analysis

Out of Sample Testing (2007-10) using 3 variables

Model

Oxford Sad Business School

S&P

Multivariate conclusions
- Out-of-sample testing of 3 years show consistency and
robustness of model
- Downside risk is mitigated with the model

Oxford Sad Business School