linking psychometric risk tolerance with choice behaviour fur conference – july 2008 peter brooks,...
TRANSCRIPT
LINKING PSYCHOMETRIC RISK TOLERANCE WITH CHOICE BEHAVIOUR
FUR Conference – July 2008
Peter Brooks, Greg B. Davies and Daniel P. Egan
2
Presentation Aims
To introduce the Barclays Wealth Risk Tolerance Scale To introduce the effects of an exponential utility function on
asset allocation To describe an experiment that provides a link between risk
tolerance scores and risk parameters. Are different risk profiles characterised by different risk/utility
parameters in choices?
3
Pre-experiment Analysis Overview
Examine risk and utility measures using simulated portfolios involving equities and bonds
Mix the simulated portfolios with different proportions of cash Holding cash is assumed to be a riskless alternative Calculate the optimal portfolio for different values of the risk
parameter
4
Example Utility Measure
5 Year Bond/Equity Mixes
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
0.05
0.07
0.09
0.11
0.13
0.15
0.17
0.19
0.21
0.23
0.25
0.27
0.29
0.31
0.33
0.35
0.37
0.39
100B
20E-80B
40E-60B
50E-50B
60E-40B
80E-20B
100E
Expect
ed U
tilit
y
Low values of imply risk tolerant behaviour – Optimal portfolio is 100% equities Higher values of imply
risk averse behaviour – Optimal portfolio is now a mix of equities and bonds
xeEEU1
5
Optimal Portfolio Mixes with Cash
We have modelled a range of values of the risk parameter for 5 year returns
For low – optimal portfolio is 100% Equities
For between 0.08 and 0.16, the optimal portfolio is a mix of equities and bonds
For greater than 0.17, the optimal asset allocation includes cash.
0
10
20
30
40
50
60
70
80
90
100
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
Ass
et
Allo
cati
on %
Equities
Bonds
Cash
5 Year Investment Horizon
6
Pre-experiment Analysis Overview 2
The analysis suggests that the optimal portfolio is sensitive to the value of a risk parameter.
Assuming utility maximisation individual choices between portfolios make it possible to calibrate a risk parameter.
Choices constrain a risk parameter to a range of values where the portfolio would be preferred by a utility maximising individual.
Analysing a number of choices makes it possible to find a “best” value of the risk parameter for each individual.
7
Barclays Wealth Risk Tolerance Scale
8 question psychometric questionnaire
Responses given on a 5-point Likert scale
Produces a score between 8 and 40
Higher scores signal higher risk tolerance
Scores bucketed into 5 risk profiles from low up to high.
Risk Tolerance Scale Distribution
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Risk Tolerance Score
Fre
qu
ency
Higher=More Risk Tolerant
8
Experiment Aims
To test various risk measures and utility functions using actual choices
To estimate risk/utility parameters for individual respondents.
To provide a link between the risk tolerance scores and risk parameters.
Are different risk profiles characterised by different risk/utility parameters in choices?
9
Experimental Design
Create stylised distributions of the final values of an investment. It is difficult to use distributions based upon real data. Increases
in volatility cause the tails of the distribution to become long. Long tailed distributions are difficult to display accurately to
survey respondents. Take log-normal distribution and set the mean and standard
deviation. Generate 120 equally spaced observations across the
distribution. Round each of these observations to the nearest integer. Plot the frequency table of the observations to create the
distributions for the experiment.
10
Experimental Design
Expected utility is increasing in the mean of the distribution. Expected utility is decreasing in the “risk” of the distribution. Create a preference order between two distributions by
compensating for an increase in “risk” by increasing the mean. The most risk averse will prefer lower mean and lower “risk”
distributions. The least risk averse will prefer higher mean and higher “risk”
distributions.
11
Example Distribution
Mean = £103,000
12
Example Distribution 2
Mean = £105,000
13
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Distribution Comparisons – Example Using Exponential Risk Measures
Utility Parameter ()
Expect
ed
Uti
lity
Mean = 106
Mean = 105
Mean = 104
Mean = 103Mean = 102
14
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Utility Parameter ()
Expect
ed
Uti
lity
Mean = 106
Mean = 105
Mean = 104Mean = 103
Mean = 102
Distribution Comparisons – Example Using Exponential Risk Measures
15
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Utility Parameter ()
Expect
ed
Uti
lity
Mean = 106
Mean = 105
Mean = 104
Mean = 103Mean = 102
Distribution Comparisons – Example Using Exponential Risk Measures
16
Experiment Procedures
Participants recruited through iPoints
Participants paid in iPoints All participants reported either
gross annual income above £50k or investable wealth above £100k
Delivered through a non-branded external website
Respondents had participated in previous surveys but had not participated within the past 6 months
Over-sampling of the extreme risk profiles
6 section experiment1. Demographics
2. Psychometric Risk Tolerance
3. Training stage
4. 9 Pairwise choice tasks between distributions
5. Filler Task – maze
6. 9 Pairwise choice tasks between distributions
17
Experimental Results
108 Participants completed all parts of the survey
1 participant removed for inconsistent responses
Over-sampling of the end points successful
Individuals in higher risk profiles tend to choose higher variance distributions more often
Use MLE to estimate the utility risk parameter for individuals - grouped by risk tolerance score
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Low Med-Low Medium Med-High High
Fre
qu
ency
18
MLE Fit Results
-0.5
0
0.5
1
1.5
2
5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00
Mean Risk Tolerance Score
Es
tim
ate
d U
tilit
y P
ara
me
ter
19
Conclusions and Extensions
Our psychometric risk tolerance measure is consistent with risky choice
There is potential for a behavioural calibration of a risk measure for portfolio optimisation
Separate work on whether utility measures are better than variance, VaR or CVaR as risk measures for portfolio optimisation
Geographical calibration exercise – current ongoing work