generalized modern portfolio theory · portfolio optimization in an uncertain world marielle de...
TRANSCRIPT
Portfolio optimization in an uncertain world
Marielle de Jong Head of Fixed-Income Quant Research at Amundi
Stowe, March 2017
Presented at the 29th Annual Research Conference of Northfield in Stowe, Vermont
Generalized Modern Portfolio Theory
Generalized MPT - de Jong
In this speech we
1. generalize the Markowitz optimization problem
2. solve it
3. link it to risk parity and similar methods
2
March 2017
Modern Portfolio Theory
Generalized MPT - de Jong * 3
March 2017
Generalize Modern Portfolio Theory We relax the certainty hypothesis,
asset returns don’t strictly obey to predefined laws accord the definition of utility,
investors seek to minimize loss due to unfavourable price shocks, which can be
foreseen – within the sphere defined by the Hessian unforeseen – unframed and accord the optimization objective
minimize loss due to
foreseeable shocks – via variance minimization unforeseeable shocks – via diversification maximization
Generalized MPT - de Jong * 4
March 2017
Diversification
Generalized MPT - de Jong * 5
March 2017
Generalized Modern Portfolio Theory
Generalized MPT - de Jong * 6
March 2017
Solution to the generalized problem
Generalized MPT - de Jong Proof by Roncalli (2013) 7
diversification
variance
March 2017
Risk parity is optimal
in an uncertain world By admitting uncertainty one acknowledges the need to protect against the
unknown.
Portfolio is optimal, i.e. loss is minimized, when portfolio variance and diversification optimized.
Risk parity when variance-over-diversification at maximum.
Risk parity solution not unique.
frontier of solutions depending on θ entropy (disorder/uncertainty) several measures of diversification
Generalized MPT - de Jong * 8
March 2017
Optimal solutions depending on θ variance-diversification frontier
High entropy – optimal portfolio 1/N
Intermediate – optimal portfolio 1/ risk contribution
Low entropy – optimal portfolio Markowitz Generalized MPT - de Jong 9
diversification
variance
ϑ → ∞
ϑ = 0
March 2017
Choice of diversification measure
Generalized MPT - de Jong * 10
March 2017
Choice of diversification measure
Generalized MPT - de Jong * 11
March 2017
Bayesian optimization is optimal
Generalized MPT - de Jong * 12
March 2017
Diversification measure based on risk estimates
Generalized MPT - de Jong * 13
portfolio volatility
asset volatility
March 2017
Recap
Generalized MPT - de Jong 14
March 2017
Investment example - equity-bond allocation problem
Generalized MPT - de Jong Example put forward by Qian (2011) 15
March 2017
Strategies coherent with generalized MPT
1. Top-down approach* (not full optimization) As covariance more predictable within asset classes than between,
Markowitz optimization is deployed within- and risk budgeting between.
2. Bridgewater® All Weather Funds Four plausible economic scenarios are weighted by their respective volatility levels. 3. Global Macro Asset class weights set inversely proportional to volatility levels
Generalized MPT - de Jong * Discussed by Scherer (2007) Portfolio Construction & Risk Budgeting
16
March 2017
A bit of philosophy
Generalized MPT - de Jong 17
March 2017
Risk model that defines order
Whether to adopt a risk factor, assess if it more damaging to ignore it – so that risk efficiency will be foregone to adopt it – so that diversification will be foregone
Statistical significance may not be an effective criterion. We explore two models
Model 1. one notch down from darkness only volatility levels predictable
Model 2. two notches down from darkness: Capital Asset Pricing Model
Generalized MPT - de Jong 18
March 2017
Risk model that defines order
Generalized MPT - de Jong * Clarke et al. (2013) derive the near closed-form solution 19
March 2017
Back-test Model 2bis on corporate bonds
Testbed – Region: Eurozone - Universe: Barclays Euro Corporate – Period: 2007 to 2016 - Monthly data frequency – Industries: Barclays level 3 sectors aggregated to ten groups
Test setup
– Portfolio rebalanced at regular intervals (no foresight) – Ex post performance over test period compared with standard benchmark Generalized MPT - de Jong 20
Consumer Discretionary
Consumer Staples
Utility & Energy
Finance_SENIOR
Finance_SUB
Insurance
Basic Industry
Captial Goods
Communication
IT
March 2017
Parity portfolio
Ten sectors contribute equally to overall portfolio risk (DTS) Within sectors firms (bond issuers) contribute equally
Generalized MPT - de Jong As of June 2016 21
0%
2%
4%
6%
8%
10%
March 2017
80
100
120
140
160
01/0
5/20
0703
/09/
2007
01/0
1/20
0801
/05/
2008
01/0
9/20
0802
/01/
2009
01/0
5/20
0901
/09/
2009
04/0
1/20
1003
/05/
2010
01/0
9/20
1003
/01/
2011
02/0
5/20
1101
/09/
2011
02/0
1/20
1201
/05/
2012
03/0
9/20
1202
/01/
2013
01/0
5/20
1302
/09/
2013
02/0
1/20
1401
/05/
2014
01/0
9/20
1402
/01/
2015
04/0
5/20
1501
/09/
2015
05/0
1/20
1602
/05/
2016
Euro Corporate Bond Benchmark
Parity portfolio
Back-test: return
Generalized MPT - de Jong Performance calculations based on monthly total returns data provided by Barclays.
22
Amundi Quant Research
Parity portfolio in line with benchmark and resilient during crises.
Underlying figures on next page
?
March 2017
Key figures
Generalized MPT - de Jong All figures are annual if not specified 23
benchmark Parity index Parity portfolio return volatility Sharpe ratio max drawdown TE # bonds # issuers turnover per month # weight
4.8% 4.1% 1.2 -6.3% - 1843 400 30 0.3%
5.4% 3.4% 1.6 -4.3% 1.64% 1843 400 30 -
5.4% 3.4% 1.6 -4.1% 0.11% (from target) < 233 < 127 10 1.6%
March 2017
Convex risk profile
Risk parity amortizes drawdowns
Generalized MPT - de Jong As of June 2016 24
-3%
-1%
1%
3%
-4% -2% 0% 2% 4%
QQ plot market vs parity
March 2017
Induced quality bias The parity index has a structural bias towards low-indebted firms As measured by two debt ratios. Debt to assets Debt to equity
Generalized MPT - de Jong 25
0,00
0,05
0,10
0,15
0,20
01/0
5/20
07
01/0
9/20
07
01/0
1/20
08
01/0
5/20
08
01/0
9/20
08
01/0
1/20
09
01/0
5/20
09
01/0
9/20
09
01/0
1/20
10
01/0
5/20
10
01/0
9/20
10
01/0
1/20
11
01/0
5/20
11
01/0
9/20
11
01/0
1/20
12
01/0
5/20
12
01/0
9/20
12
01/0
1/20
13
01/0
5/20
13
01/0
9/20
13
01/0
1/20
14
01/0
5/20
14
01/0
9/20
14
01/0
1/20
15
Market indexParity index 0,0
0,5
1,0
1,5
2,0
2,5
3,0
01/0
5/20
07
01/0
9/20
07
01/0
1/20
08
01/0
5/20
08
01/0
9/20
08
01/0
1/20
09
01/0
5/20
09
01/0
9/20
09
01/0
1/20
10
01/0
5/20
10
01/0
9/20
10
01/0
1/20
11
01/0
5/20
11
01/0
9/20
11
01/0
1/20
12
01/0
5/20
12
01/0
9/20
12
01/0
1/20
13
01/0
5/20
13
01/0
9/20
13
01/0
1/20
14
01/0
5/20
14
01/0
9/20
14
01/0
1/20
15
Market index
Parity index
March 2017
Induced size bias The parity index has a structural bias towards small firms as measured by two firm fundamentals, sales revenues and book value Sales Assets
Generalized MPT - de Jong 26
0
10000
20000
30000
40000
01/0
5/20
07
01/0
9/20
07
01/0
1/20
08
01/0
5/20
08
01/0
9/20
08
01/0
1/20
09
01/0
5/20
09
01/0
9/20
09
01/0
1/20
10
01/0
5/20
10
01/0
9/20
10
01/0
1/20
11
01/0
5/20
11
01/0
9/20
11
01/0
1/20
12
01/0
5/20
12
01/0
9/20
12
01/0
1/20
13
01/0
5/20
13
01/0
9/20
13
01/0
1/20
14
01/0
5/20
14
01/0
9/20
14
01/0
1/20
15
Market indexParity index
0
100000
200000
300000
400000
01/0
5/20
07
01/0
9/20
07
01/0
1/20
08
01/0
5/20
08
01/0
9/20
08
01/0
1/20
09
01/0
5/20
09
01/0
9/20
09
01/0
1/20
10
01/0
5/20
10
01/0
9/20
10
01/0
1/20
11
01/0
5/20
11
01/0
9/20
11
01/0
1/20
12
01/0
5/20
12
01/0
9/20
12
01/0
1/20
13
01/0
5/20
13
01/0
9/20
13
01/0
1/20
14
01/0
5/20
14
01/0
9/20
14
01/0
1/20
15
Marketindex
March 2017
Conclusion
Incorporating uncertainty and consequently generalizing the Markowitz optimization problem makes sense. It corresponds to investment practice, with a list a portfolio construction methods with alternative philosophies, e.g. regret theory Incorporating uncertainty doesn’t suit active high-conviction investment.
Generalized MPT - de Jong 27
March 2017