PortfolioPython

July 23, 2020

Introduction to portfolio construction and analysis with Python

https://www.coursera.org/learn/introduction-portfolio-construction-python/
Compiled by ruud waij

1 Week 1 Section 1 - Fundamentals of risk and returns

1.1 Video: Fundamentals of returns
1.1.1 Compute the return
The return R (= fraction of profit) from time t to time t + 1 for price P is given by:

Pt+1 − Pt Pt+1
Rt,t+1 = = −1
Pt Pt
If dividends D have been paid out at t + 1, the formula above is for the price return. the total
return is given by:
Pt+1 + Dt,t+1 − Pt Pt+1 + Dt,t+1
Rt,t+1 = = −1
Pt Pt

1.1.2 Multi period returns

Compounded returns
Rt,t+2 = (1 + Rt,t+1 )(1 + Rt+1,t+2 ) − 1

1.1.3 Annualizing returns

Annualizing a 1% per month return:

((1 + 0.01)12 − 1)

1.2 Video: Measures of risk and reward

1.2.1 Variance
Variance is the average of the square of the deviations from the mean:

1 X
N
2
σR = (Ri − R̄)2
N
i=1

where R̄ is the arithmetic mean of the returns.

1
1.2.2 Standard deviation
Standard deviation is the square root of the variance:
v
u
u1 X N
σR = t (Ri − R̄)2
N
i=1

1.2.3 Annualizing volatility

√
σann = σp p
where p is the number of periods in a year, σp is the volatility per period and σann is the annualized
volatility.

1.2.4 Why is variance proportional to time?

I wondered why variance is proportional to time. Three answers:
• Why Is Volatility Proportional to the Square Root of Time?
• Why Variances Add
• Why does the variance of the Random walk increase?

1.2.5 Sharpe ratio

The Sharpe ratio measures the return Rp of a portfolio P compared to the return Rf of a risk-free
asset, after adjusting for its risk (volatility σp ).

Rp − Rf
SharpeRatio(P) =
σp

1.3 Video: Measuring max drawdown

1.3.1 Max drawdown
The Max Drawdown is the maximum theoretical loss from the previous high to a subsequent low:
the worst possible return you could have seen if you bought high, sold low.
Investopedia: “A maximum drawdown (MDD) is the maximum observed loss from a peak to a
trough of a portfolio, before a new peak is attained. Maximum drawdown is an indicator of
downside risk over a specified time period.”

1.3.2 Calmar ratio

The Calmar Ratio is the ratio of the annualized return over the trailing 36 months to the maximum
drawdown over those trailing 36 months.

2
2 Week 1 Section 2 - Beyond the Gaussian case: extreme risk
estimates
2.1 Video: Deviations from normality
2.1.1 Skewness
Skewness is a measure of asymmetry of the distribution. If distribution has a negative skewness,
then the probability of getting an outcome below the mean is higher than the probability of an
outcome above the mean. The opposite is true for a positively skewed distribution.

E[R − E(R))3 ] E[(R − E(R))3 ]

S(R) = 3 == 3
[V ar(R)] 2 σR

2.1.2 Kurtosis
Kurtosis is a quantitative measure of the thickness of the fatness of the tail of in relation to that
of the Gaussian distribution. The Gaussian distribution has a kurtosis of 3, if the value is higher
it has a fat tail distribution.

E[R − E(R))4 ] E[(R − E(R))4 ]

K(R) = == 4
[V ar(R)]2 σR

2.1.3 Jarque-Bera test

There are many tests that tell if a distribution is normal and perhaps one of the most commonly
used is called the Jarque-Bera test.

n 2 (K − 3)2
JB = (S + )
6 4

2.2 Video: Downside risk measures

2.2.1 Semi-deviation
Semi-deviation is the volatility of the sub-sample of below-average or below-zero returns.
v
u
u1 X
σsemi = t (Rt − R̄)2
N
Rt ≤R̄

N is the number of returns that fall below the mean.

2.2.2 VaR - Value at Risk

VaR represents the maximum expected loss over a given time period at a specified confidence level.
The loss is represented by a positive number. See also Investopedia.

3
 2.2.3 CVaR - Conditional VaR
CVaR is the expected loss (as a positive number) beyond the VaR. See also Investopedia.
R V aR
x· fR (x)d(x)
CV aR = −E(R|R ≤ −V aR) = −∞
FR (−V aR)
fR is the return density function and FR is the cumulative probability distribution. CVar is nothing
more than the average of all those returns that are worse than the VaR.
Question: Consider the following sequence of monthly returns on a portfolio: -4%, +5%, +2%,
-7%, +1%, +0.5%, -2%, -1%, -2%, +5%. What is the 80% monthly CVaR for the portfolio?
Answer: Out of 10 monthly returns, the worst outcome is -7% and the second worst outcome is
-4%, so 8 out of 10 outcomes are better than -4%. The worst return after excluding the 20% worst
return, that is after excluding the two worst return, is therefore -2%. VaR is typically expressed as
a positive number so 2% is the 80% VaR. The CVaR is given by the average loss beyond the VaR,
that is $ -(-7%-4%)/2=5.5% $.

2.3 Video: Estimating VaR

Four methods for calculating VaR.

2.3.1 Method 1: Historical (non parametric)

Calculation of VaR based on the distribution of historical changes in the value of the current
portfolio under market prices over the specified historical observation window.

2.3.2 Method 2: Parametric Gaussian methodology

Calculation of VaR based on portfolio volatility, i.e. on volatilities and correlations of components.
You assume a Gaussian distribution and estimate its mean µ and volatility σ.

Pr(R ≤ µp + zα σp ) = α

Zα is the one-sided α-quantile of the standard normal distribution of the return, for example for
Zα =-1.65, the probability is ~5% to get below that α-value.

VaRα = −(µ + zα σ)

Question: Consider a portfolio with a +.5% monthly expected return and 4% monthly volatility.
What is the 95% monthly Gaussian VaR for the portfolio?
Answer: VaR = −(.5% − 1.65x4%) = 6.1%. (−6.1 or less is the absolute value range that will
occur in 5% of the cases for a normal distribution with µ = 0.5 and σ = 4.)

[1]: import warnings

warnings.simplefilter(action='ignore', category=FutureWarning)

%load_ext autoreload
%autoreload 2

4
 %matplotlib inline

import numpy as np
import seaborn as sns
import matplotlib as plt
from scipy import stats
import matplotlib.pyplot as plt

mu= 0.5
sigma=4
x5= -6.1

extend= 3
xx = np.arange(mu-extend*sigma, mu+extend*sigma, 0.1)
yy = stats.norm.pdf(xx,loc=mu,scale=sigma)
delta=0.000000001
xxx= xx[xx<= x5]
yyy= yy[xx<= x5]

plt.fill_between(xxx, yyy, alpha=0.1, color='blue')

sns.lineplot(x=xx,y=yy).set_title('95% monthly Gaussian VaR=6.1 example')

sns.lineplot(x= [mu,mu+delta], y= [0,yy.max()],markers='.')
sns.lineplot(x= [x5,x5+delta], y= [0,stats.norm.pdf(x5,loc=mu,scale=sigma)])

[1]: <matplotlib.axes._subplots.AxesSubplot at 0x140aad62bc8>

5
 The Gaussian distribution has thin tails and may underestimate risk. It is not a good model for
VaR.

Method 3: Parametric non-Gaussian VaR There are many parametric non-Gaussian distri-
butions:
• Define
• Pareto
• Student
• Loggamma
• Frechet

2.3.3 Method 4: Cornish-Fisher

The Cornish-Fisher expansion is a semi-parametric approach for calculation of value at risk esti-
mates in a non-Gaussian setting. Cornish-Fisher does not force you to assume any particular return
distribution.

1 1 1
z̃α = zα + (zα2 − 1)S + (zα3 − 3zα )(K − 3) − (2zα3 − 5zα )S 2
6 24 36

VaRmod (1 − α) = −(µ + z̃α σ)

• z̃a : alpha quantile of the non-Gaussian distribution.
• za : alpha quantile of the Gaussian distribution
• S: skewness
• K − 3: excess kurtosis

3 Week 2 Section 1 - Introduction to optimization and the efficient

frontier
3.1 Video: The only free lunch in finance
3.1.1 2-asset combinations
σ 2 (wa , wb ) = σA
2 2 2 2
wA + σ B wB + 2wA wB σA σB ρAB
• wa : weight of asset A
• σa : volatility of asset A
• ρA,B = correlation coefficient of A and B

[7]: import pandas as pd

import edhec_risk_kit as erk

er= pd.Series([2.0,1.0]) # expected return

cov= pd.DataFrame({0:[2,1],1:[1,2]}) # covariance
erk.plot_ef2(21, er, cov, style='.-',figsize=(10,6))

6
[7]: <matplotlib.axes._subplots.AxesSubplot at 0x211f31beac8>

3.2 Video: Markowitz optimization and the efficient frontier

3.2.1 The efficient frontier
The efficient frontier represents: The portfolios that offer the lowest volatility for a given level of
return. They also represent the portfolio of highest expected return for a given level of volatility.
The efficient frontier is simply the space of portfolios that we would want to hold because there is
no way of getting a higher return for the same level of volatility or for you to get a portfolio with
equivalent return, but lower volatility.

3.2.2 Convex optimization to draw the efficient frontier

The return of the portfolio Rp is the weighted wi sum of the returns Ri of the assets in the portfolio.

X
k
Rp = wi Ri
i=1

The expression for the portfolio variance σ 2 is a generalization of the formula we’ve already seen
in the 2-asset case:
X k X k
2
σp = wi wj σi σj ρij
i=1 j=1

Covariance(i, j) = σij = σi σj ρij

7
 X
k X
k
σp2 = wi wj σij
i=1 j=1

3.2.3 Matrix form expression for the portfolio return

Rp = wT R
• w is the k ∗ 1 vector of weights
• R is the k ∗ 1 vector of asset returns

3.2.4 The covariance matrix

P
The covariance matrix Σ (not to be confused with sum hi lo ) is a symmetric square k ∗ k matrix
where the i and j element is σi j, the covariance between i and j, which is σi σj ρij . Each diagonal
element of Σ is σii = σi σi ρii .

σp2 = wT Σw

3.2.5 Plotting the efficient frontier

A portfolio is on the efficient frontier if it is the one that has the minimum volatility Rp = wT Σw for
a certain level of return. Minimize 21 wT Σw in a quadratic optimizer with the following constraints:
• the portfolio return must be at the desired level
• all asset weights have to be ≥ 0
• the asset weights add up to one

4 Week 2 Section 2 - Implementing Markowitz

4.1 Video: Fund separation theorem and the capital market line
4.1.1 Capital market line (CML)
The efficient frontier dramatically changes shape when a risk-free asset is introduced.
[8]: import pandas as pd
import edhec_risk_kit as erk

er= pd.Series([0.2,0.1]) # expected return

cov= pd.DataFrame({0:[2,-1.5],1:[-1.5,2]}) # covariance
erk.plot_ef2(21, er, cov, style='.-',figsize=(10,6))
plt.plot([0, 1], [0.12, 0.191])

[8]: [<matplotlib.lines.Line2D at 0x211f30594c8>]

8
The maximum Sharpe ratio portfolio (MSR) is at the point where the orange line that starts at the
risk-free rate (here generously set at 12%) touches the efficient frontier. The slope of the orange
line is the Sharpe ratio of the portfolio.
The MSR portfolio contains no exposure to specific risk. This type of risk is unrewarded because
it can be diversified away.
The tangency portfolio is the portfolio that maximizes the Sharpe ratio.

PN
µp − rf i=1 wi µi − rf
SRp = = qP
σp N
ww σ
i,j=1 i j ij

• SRp : Sharpe ratio

• µp : portfolio return
• rf : risk-free rate
• σp : portfolio volatility
• wi : individual asset weights
• µi : expected return per asset
• σij : covariance of return of assets i, j

4.2 Video: Lack of robustness of Markowitz analysis

4.2.1 Parameter uncertainty
It is difficult to estimate model parameters in portfolio optimization, especially the one for expected
return.

9
 4.2.2 Global minimum variance (GMV) portfolio
The GMV is the portfolio that is least sensitive to errors in parameter estimates. Since it requires
no expected return estimates, it is only sensitive to errors in risk parameter estimates. The asset
weights of the maximum Sharpe ratio portfolio vary a lot if the expected returns of the assets
change.
[9]: import pandas as pd
import edhec_risk_kit as erk

er= pd.Series([2,1]) # expected return

cov= pd.DataFrame({0:[2,1],1:[1,2]}) # covariance
erk.plot_ef2(11, er, cov, style='.-',figsize=(5,3))
plt.plot([1.224744871391589],[1.5],marker='o')
plt.title('global minimum variance portfolio')

[9]: Text(0.5, 1.0, 'global minimum variance portfolio')

The equal weight portfolio is the naive diversification portfolio.

5 Week 3 Section 1 -
5.1 Video: Limits of diversification
Diversification allows you to most efficiently harvest the highest possible reward per unit of risk,
by eliminating unrewarded risks. diversification cannot help you deal with systematic risk: the risk
factor that impacts all the assets simultaneously.

10
5.1.1 Hedging
Hedging (i.e. avoiding risk taking) is the only effective way to obtain downside protection. Hedging
is symmetric, you get protection for downside risk, but give up on possible returns.

5.1.2 Hedging versus insurance

Insurance provides downside protection while allowing for upside potential. Insurance is dynamic
hedging.

5.2 Video: An introduction to CPPI - Part 1

5.2.1 CPPI
CPPI=Constant Proportion Portfolio Insurance is a procedure that dynamically allocates total
assets to a risky asset and a safe asset.

I explain the diagram in a different way than the Coursera/EDHEC video does:
• T is the total amount of assets that you have
• Protection Floor F is the level of assets that you want to protect against downside risk
• The difference between total assets T and the desired protection floor F is called cushion
C =T −F

11
 • The amount of risky asset E you take on is E = M ∗ C, where M is a (yet undetermined)
multiplier
• The rest is put in a riskless asset B = T − E
If the risky asset E loses value, the cushion C becomes smaller and the risky asset must be replaced
with riskless asset. E = M ∗ C must hold. If the cushion becomes zero, only riskless asset remains.
Example:
• M =3
• F = 80%
• T = 100$
What is the value of E? C = T − F = 100 − 80 = 20. The initial investment in risky asset E is
E = M ∗ C = 3 ∗ 20 = 60.

5.2.2 Downside protection

Gap risk occurs if you do not trade continuously. If you trade intermittently, the value of your
assets may fall below the protection floor inbetween trading sessions. Gap risk materializes if and
only if the loss on the risky portfolio E relative to the the safe portfolio B exceeds 1/M within the
trading interval.
Question: Assume a drop of 20% of the risky asset. Which multiplier value would lead to a
violation of the floor.
Answer: ∆E = 0.2 ∆E > 1/M =⇒ M > 5

5.3 Video: An introduction to CPPI - Part 2

5.3.1 Introducing maximum drawdown constraints
Maximum drawdown constraint: Vt > αMt where:
• Vt is the value of the portfolio at time t
• Mt is the maxium value of the portfolio between time 0 and t (a running max process)
• 1−α is the maximum acceptable drawdown The maximum allowable drawdown is a percentage
of the highest reached portfolio value.
Question: Consider the following sequence of portfolio value over time: P(0)=100, P(1)=110,
P(2)=105, P(3)=108, P(4)=111. What is the value M(3) of the max process at date t=3?
Answer: 110, it is the value of the running max process at dates 1, 2 and 3, before it turns 111 at
date 4.

5.3.2 CPPI with performance cap

The return is also capped at the top.
Protect downside:
Ft ≤ At ≤ Tt ⇒ Et = m(At − Ft )
Protect upside:
Tt ≤ At ≤ Ct ⇒ Et = m(Ct − At )

12
 • F : floor (a minimum amount of wealth to protect)
• C: cap (a maximum (!) amount of wealth to protect)
• A: asset value
• E: the amount of risky asset that you take on
• m: CPPI multiplier
• T : threshold level where downside protection switches to upside protection, determined by F
and C.
Question: What is the initial allocation to the risky asset for a portfolio starting at $100, with a
multiplier equal to 4, a wealth preservation floor set at 90% and a cap set at 105%?
Answer: A = 100 M = 4 F = 90% C = 105% The portfolio initial value is closer to the cap
than it is to the floor. In this case, the investment in risky asset is 4 ∗ (105% − 100%) = 20%.
Threshold level of the asset value A = T where downside proection switches to upside protection
is determined by:
Ft + Ct
m(Tt − Ft ) = m(Ct − Tt ) → Tt =
2

Question: Consider a portfolio starting at $100, with a multiplier equal to 4, a wealth preservation
floor set at 90% and a cap set at 105%. What is the threshold value beyond which the allocation
to risky asset is computed with respect to the cap, and not to the floor?
Answer: A = 100 M =4 F = 90% C = 105%
100+95
T = 2 = 97.5%

5.4 Video: Simulating asset returns with random walks

5.4.1 A stochastic model for asset returns
There are two assets:
• St : a risky asset
• Bt : a riskfree asset
Return process modelled as
dSt
= (r + σλ)dt + σdWt
St
where r + σλ = µ.
dBt
= rdt ⇔ Bt = B0 ert
Bt
• dSt is the return: the change in value of the stock index between time t and time t + dt
• dS
St : the return as a fraction of the asset value
t

• the return consists of two components:

– (r + σλ): the return trend, the expected return
– σdWt : the stocastic component
• the expected return µ consists of two components:
– r: the riskfree rate
– σλ is the risk premium: the volatility σ times the Sharpe ratio λ.
1
• dt is the period over which we calculate the return; for one month dt = 12 if the expected
return has been calculated per year.

13
 • σdWt the stocastic component: volatility σ times dWt , the Brownian motion process.
• dB
Bt = rdt: is the return fraction of the riskfree asset. It is the riskfree rate r times the period
t

dt.
• B0 : the value of B at time zero(?).
Question: Assume that a stock index has a Sharpe ratio of 0.5 and a volatility of 20%? If the
risk-free rate is 2%, what is the drift of the stock index return process?
Answer: λ = 0.5, σ = 0.2 and r = 0.02. dSt
St = 0.02 + 0.2 ∗ 0.5 = 0.12

5.4.2 Brownian motion: a discrete time formulation

The standard Brownian motion process is equivalent to a random walk in continuous-time.

St+dt − St √
= (r + σλ)dt + σ dtξt
St
S −S
• t+dtSt
t
= dS
St : the fractional change of return.
t

• (r + σλ)dt is the expected return

– r riskfree rate
– σ volatility
√– λ Sharpe ratio
• σ dtξt the contribution of the Brownian motion to the return
• ξt is the normally distributed Brownian motion noise with mean zero and unit variance.

dSt Et [ dS
St ]
t

Et [ ] = µdt ⇒ =µ
St dt

dSt Vart [ dS
St ]
t

Vart [ ] = σ 2 dt ⇒ = σ2
St dt
• µ: annualized expected return
• σ 2 : annualized variance

5.4.3 Brownian motion

Brownian motion follows a normal distribution with zero mean and variance dt. At any moment
its value is equally likely to go up and go down.
Question: If you estimate volatility of a portfolio return to be 2% over the monthly horizon, what
would be your estimate for annual volatility?
Answer: The√ annualized 2
√ variance Var is 12 times the monthly variance 2 . The annualized
volatility σ = Var = 12 ∗ 22 = 6.9%

14
6 Week 3 Section 2 -
6.1 Video: Monte Carlo simulation
6.1.1 A model with a stochastic opportunity set
All elements of the equation below depend on time. For example, the riskfree rate is not a constant
but is a stochastic process that can change over time: rt .

dSt p p
= (rt + Vt λSt )dt + Vt dWtS
St
dSt
• St : the return as a fraction of
the asset value
• r√t : the time dependent riskfree rate
• Vt = σt : the estimate of the time dependent volatility (sqrt of variance Vt ).
• λS : the estimate of the time dependent Sharpe ratio
√t √
• Vt dWtS the stocastic component: volatility σt = Vt times dWtS . The time dependent
Brownian motion process is represented by dWtS .
The change in riskfree rate drt and volatility dVt are modeled as stochastic processes.

drt = a(b − rt )dt + σr dWtr

• a: the speed of reversion to the mean: the (constant) speed at which the riskfree rate changes
and drives the riskfree rate rt back to the mean.
• b: the mean of the riskfree rate
• rt : the current value of the riskfree rate
• dt: the time period
If the current value of the riskfree rate rt is lower than the average rate b, the change in riskfree rate
will be positive. The riskfree rate will revert back to the mean with the speed of mean reversion a.
These processes are called mean reverting processes.
The Variance Vt can be modelled in a similar way. Since the variance is σ 2 , it cannot be negative.
Because
√ a square root√cannot be negative either, the component that contains volatility is written
as σV Vt . (Since σV Vt equals Vt , why not simply write Vt ?)

p
dVt = α(V̄ − Vt )dt + σV Vt dWtV
• dVt : the change in variance
• α: the mean reverting speed of Vt
• V̄ : the mean of the variance
• Vt : √the current value of the variance
• σV Vt dWtr : the Brownian motion (≥ 0) for changes in variance
Question: Assuming a mean-reverting model for interest rates with long-term mean level equal
to 2%. If the current level of interest rate is 3% and if the speed of mean reversion is 0.5, what is
the average annual expected change in interest rate?
Answer: drt = a(b − rt )dt with b = 0.02, rt = 0.03, a = 0.5, dt = 1 ⇒ 0.5 ∗ (0.02 − 0.03) = −0.5%

15
Take a look at the Wikipedia page about Geometric Brownian Motion.

6.2 Video: Analyzing CPPI strategies

This is a demonstration of how CPPI parameters affect outcome.

6.3 Video: Designing and calibrating CPPI strategies

This is also a demonstration of how CPPI parameters affect outcome.

7 Week 4 Section 1 -
7.1 Video: From Asset Management to Asset-Liability Management
7.1.1 Funding ratio
In asset-liability management we measure the ratio between asset values and liability values.
Funding ratio measures the assets relative to the liabilities. If it is equal to 100%, it means that
assets are sufficient to cover liabilities.

At
Ft =
Lt
• Ft : funding ratio
• At : asset values
• Lt : liability values

7.1.2 Surplus
The surplus St is the difference between assets At and liability values Lt .

St = At − Lt

Question: Consider a pension fund that has a 50%-50% stock/bond allocation. Assume that
equity markets go down by 5%, bond markets go down by 10%, and further assume that the
liability value goes down by 8%. Is the funding ratio of the pension plan increasing or decreasing?
Answer: Change in return: (0.5 ∗ −5% + 0.5 ∗ −10%) = −7.5%. Change in liability value: −8%.
The funding ratio increases.

7.2 Video: Lab Session - Present Values, liabilities and funding ratio
This lab session video introduces some equations.

7.2.1 The present value of future liabilities

We need the present value of future liabilities to compute the funding ratio (see Section 7.1.1).
In order to compute the present value, we need to discount the amount of the liability based on the
relevant interest rate derived from the yield curve. For simplicity, we’ll assume that the yield curve

16
is flat, and so the interest rate is the same for all horizons. The present value of a set of liabilities
L is given by:

X
k
P V (L) = B(ti )Li
i=1

• PV(L): present value of a set of liabilities L

• k: the number of liabilities
• B(ti ): the price of a pure discount bond that pays 1 dollar at time ti
• Li : liability that is due at time ti
If we assume the yield curve is flat and the annual rate of interest is r then B(t) is given by

1
B(t) =
(1 + r)t

This means that after t years the bond (that now costs Bt ) will pay 1 dollar:

B(t)(1 + r)t = 1

7.3 Video: Liability hedging portfolios

7.3.1 Hedging = cash-flow matching
Liability-hedging portfolios (LHP) or Goal-hedging portfolios (GHP) are portfolios with payoffs
(cash-flows) matching the date and nominal amount of liability/goal payments.
Standard bonds are not suited for retirement pay outs, because the value is payed out in one lump
sum.
Question: What is the safe liability-hedging asset for an investor that has a liability defined as a
single inflation-indexed payment to be made in 10 years from now?
• 3 month T-Bills?
• 10 year T-Bond?
• Inflation-linked pure discount (zero coupon) bond with a 10 year duration
Answer:
• T-bills: This should not be selected. If interest rates go down, the value of short-term bonds
will increase moderately, while the value of the long-dated liabilities will increase significantly.
• T-bond: This should not be selected. The duration of a coupon-paying 10 year bond will be
strictly lower than 10. As a result, the bond value will increase less than the liabilities in case
of a decrease in interest rates. Besides, the 10-year T-Bond offers no protection with respect
to increases in inflation.
• Inflation linked bond: Correct. For this investor, the liability-hedging portfolio is a 10Y
inflation-linked pure discount bond.

17
7.3.2 Factor exposure matching
We would like bonds with payouts that match our liability payments. That would give perfect
cash-flow matching. This is not practical and we can use factor exposure matching instead.
Question: What is the safe liability-hedging asset for a 45Y old investor preparing for retirement
at age 65.
• 3 month T-Bills?
• 20-year inflation-linked pure discount bond?
• Deferred inflation-linked annuity with a 20 years deferral period?
Answer:
• 3 month T-Bills: This should not be selected If interest rates go down, the value of short-
term bonds will increase moderately, while the value of the long-dated liabilities, given by
replacement income, will increase significantly.
• 20-year inflation-linked pure discount bond: This should not be selected The 20-year pure
discount bond will pay a cash-flow when the investor reaches retirement at age 65, but the
investor needs to receive replacement income cash-flows throughout retirement, not at retire-
ment date.
• Deferred inflation-linked annuity with a 20 years deferral period?: A deferred inflation-linked
annuity with a 20 year deferral period will start paying cash flows when the investor reaches
age 65, and will pay replacement income cash-flow as long as the investor is alive. This make
it the perfect safe retirement asset.
These questions cannot be answered using the information learned in the lectures!

7.4 Video: Lab Session-CIR Model and cash vs ZC bonds

7.4.1 CIR Model
The Cox Ingersoll Ross Model (CIR) is used to simulate changes in interest rate. It is a mean
reverting model.

√
drt = a(b − rt ) dt + σ rt dWt
• drt : small change in interest rate
• a: the speed of reversion to the mean
• b: long term mean of the interest rate
• rt : current interest rate
• dt : a small amount of time
√
• σ rt dWt : random component
– σ: the volatility, which acts as a scaling factor
√
– rt : prevents negative interest rates
– dWt : a random normally distributed number (‘the shock’)

7.4.2 Short rate

The short rate is the instantaneous interest rate that you get for a very small amount of time. To
annualize this rate, we need to compound it.

18
If we compound every 1/N part of a year we get:

r N
(1 + )
N

As N → ∞ we get the generalization:

1 + rannual = erinst
rannual = erinst − 1
rinst = ln(1 + rannual )
• rannual : annualized interest rate
• rinst : instantaneous (or short) rate
• erinst : e to the power of the short rate

7.4.3 Generating the random price evolution of a Zero-Coupon Bond

(This has been copied from anaconda notebooks_and_codem01_v02/nb/lab_125.ipynb)
The CIR model can also be used to generate the movement of bond prices for a zero coupon bond
that are implied by the generated interest rate, using the following equations:

P (t, T ) = A(t, T )e−B(t,T )rt

where

!2ab/σ2
2he(a+h)τ /2
A(t, T ) =
2h + (a + h)(eτ h − 1)

and

2(eτ h − 1)
B(t, T ) =
2h + (a + h)(eτ h − 1)

and

p
h= a2 + 2σ 2

and

τ =T −t
• P (t, T ): price at time t of a bond that matures at time T
So this is the evolution in price of a zero-coupon bond, as interest rates go up, the price of the bond
comes down. As interest rates falls, the price of the bond goes up.

19
7.5 Video: Liability-driven investing (LDI)
7.5.1 Greed and Fear
• Performance generation through optimal exposure to rewarded risk factors
• Hedging against unexpected shocks
Performance-seeking portfolio (PSP): focus on diversified efficient access to risk premia.
Liability-hedging portfolio (LHP): focus on hedging impact of risk factors in liabilities.

7.5.2 Formal LDI model

AT λP SP P SP 1
maxw E[u( )] ⇒ w∗ = w + βL,LHP (1 − )wLHP
LT γσP SP γ
• λP SP : the PSP Sharpe ratio: λ = 0 ⇒ no investment in PSP.
• βLHP : the beta (interest rate??) of liabilities for LHP: β = 0 ⇒ no investment in LHP.
• σP SP is the PSP volatility: σ = ∞ ⇒ no investment in PSP.
• γ is the risk aversion: γ = ∞ ⇒ no investment in PSP.
Question: Assume that the best liability-hedging portfolio a pension fund can find has a zero
correlation with the liabilities and a 50% Sharpe ratio, and that the highest Sharpe ratio available
has a Sharpe ratio of 100%, what should be in this case the optimal allocation to the liability
hedging portfolio?
• 0%
• 50%
• 100%
Answer:
• 0%: Correct If the liability-hedging portfolio has no hedging ability, then the pension fund
should not hold any of this portfolio since hedging liabilities was the only reason why this
portfolio was useful in the first place.
• 50%: This should not be selected. If the liability-hedging portfolio has no hedging ability,
then the pension fund should not hold any of this portfolio since hedging liabilities was the
only reason why this portfolio was useful in the first place.
• 100%: This should not be selected. 100% would be in this case the allocation to the
performance-seeking portfolio under the simplifying assumption that cash is not used.

7.6 Video: Lab Session-Liability driven investing

7.6.1 Macaulay duration
The Macaulay duration is the weighted average term to maturity of the cash flows from a bond. The
weight of each cash flow is determined by dividing the present value of the cash flow by the price.
Macaulay duration is frequently used by portfolio managers who use an immunization strategy.

Pn t×C n×M
t=1 ( (1+y)t + (1+y)n
Macaulay duration =
Current bond price
• t: respective time period
• C: periodic coupon payment

20
 • y: periodic yield
• n: total number of periods
• M : maturity value
• Current bond price: present value of cash flows

8 Week 4 Section 2 -
8.1 Video: Choosing the policy portfolio
8.1.1 Optimal asset mix
What is the optimal asset mix for liability driven investment (Section 7.6) when investing in the
two building blocks PSP and LHP (Section 7.5.1)?
The risk aversion parameter γ is the fraction of allocation to LHP/PSP.
In practice, the allocation to PSP is increased until the risk budget is exhausted. The risk budget
is set by the stakeholders.

8.1.2 Conflicts
There is a conflict between short-term and long-term perspectives. For the long term perspective
you want to reach your funding ratio. For the short perspective you want to look at the volatility
and the max drawdown (Section 1.3.1).
If the risk budget is small, there will not be much upside potential. This requires a large investment
to meet the liability.

8.1.3 Solving the dilemma

• Hide the problem
– Use a higher discount rate
– use a higher risk premia value
• Attempt to solve the problem
– Request a higher contribution
– Request a higher risk budget
– Improve the performance seeking portfolio (PSP)
Question: What impact do you expect on the shape of the distribution of the final funding ratio
when increasing the allocation to performance-seeking assets?
• A decrease in the median value and a decrease in the dispersion of the funding ratio distri-
bution.
• An increase in the median value and a decrease in the dispersion of the funding ratio distri-
bution.
• A decrease in the median value and an increase in the dispersion of the funding ratio distri-
bution.
• An increase in the median value and an increase in the dispersion of the funding ratio distri-
bution.
Answer: Increasing the allocation to risk assets increases expected performance and also increases
risk.

21
Question: At a meeting of the board of trustees of a pension fund, it is decided that the acceptable
volatility of the funding ratio is increased from 8% to 10? What consequence should follow?
• An increase in the allocation to the PSP?
• A 10% increase in funding ratio?
• A 10% decrease in funding ratio?
• An increase in the allocation to the LHP?
Answer: An increase in the allocation to the PSP. Increasing the risk budget allows investors to
take on more risk.

8.2 Video: Beyond LDI

8.2.1 Fund separation theorem
The LDI paradigm (Section 7.5) uses two building blocks as described by the Fund separation
theory for performance seeking and liability hedging.

8.2.2 Fund interaction ‘theorem’

The Fund interaction ‘theorem’ tries to combine a high funding ratio with a low level of uncer-
tainty. This maximizes investor welfare. Investor welfare consists of: - pure performance contri-
bution (high sharpe ratio) - pure hedging contribution (meeting the liability) - cross-contribution
performance/hedging (maximizing the correlation between PSP and the liabilities)
Question: What is the contribution of the performance-seeking portfolio to investor welfare if
risk-aversion is infinite?
• 0%
• 100%
• We need to know what is the Sharpe ratio of the PSP to answer this question?
Answer: An investor with infinite risk-aversion would optimally hold 0% of the PSP even if the
PSP has an attractive Sharpe ratio. As a result the PSP will not contribute to the welfare of this
investor.

8.2.3 Fund separation/interaction in practice

Improve liability friendliness of PSP → - one can allocate more money to PSP for the same risk
budget. - a lower performance may be generated by PSP.
Question: Would the adoption of a more defensive equity benchmark lead to:
• An increase in investor welfare?
• A decrease in investor welfare?
• An impact on investor welfare that can be positive or negative depending on parameter values?
Answer: Adopting a more defensive equity benchmark would lead to a decrease in funding ratio
volatility, which in turns would allow for an increase in equity allocation for the same risk budget,
but it would also lead to a decrease in performance. The impact on investor welfare will be negative
if the decrease in performance will more than compensate for the increase in equity exposure. in
performance. The impact on investor welfare will be positive if the increase in equity exposure will
more than compensate for the decrease in performance.

22
8.3 Video: Liability-friendly equity portfolios
8.3.1 Liability friendliness
• cash-flow matching focus: choose stocks with high and stable dividends
• factor matching focus: choose stocks that may have a low tracking error with respect to the
liabilities: low volatility stocks
• isovariance portfolios: fill the risk budget upto the the allowed level with low volatility stocks.

23