Portfolio Choice

Diversification, Risk and return,

The two-fund separation theorem

Mehran Ebrahimian
Investment Management (BE452)
Spring 2025
Learning Goals

Basics of portfolio choice:

▶ The relationship between risk and return
▶ The two-asset example
▶ Diversification
▶ Mean-variance frontier with multiple assets
▶ Sharpe Ratio, Tangency portfolio
▶ Two-fund separation theorem
▶ Practical investment example

2 / 38
Prelude—Return Rates, Mean and Variance
The historical tradeoff

▶ The required rate of return should depend on the risk because we

are “risk-averse”
▶ Is risk represented by variance?
3 / 38
Risk = variance?

▶ We generally view risk as dispersion around expected rates of

return
▶ Suppose investments A and B have the same expected return.
▶ If the return of A is less variable than B, then investors
generally prefer investment A.

▶ But this is not the whole story! Indeed, Risk=Variance is

▶ generally true for broad (well-diversified) portfolios...
▶ ...false for individual stocks.

▶ Keep in mind: risk is about covariance. not variance.

▶ covariance with the state of economy...
▶ for large/diversified portfolios, cov=var
▶ let’s see.

4 / 38
Relationship Between Mean and Volatility

For “broad” portfolios there seems to be a general increasing

relationship between historical volatility (=standard deviation) and
average return, i.e., one-to-one relationship:

5 / 38
Relationship Between Mean and Volatility - Individual
Stocks

No relationship between volatility and average return for individual

stocks:

6 / 38
Relationship Between Mean and Volatility - Individual
Stocks

▶ Return on an individual stock in the S&P500 is much more

volatile than the S&P500 itself...
▶ ...for about the same expected return

Investment Risk Premium Variability

Stock Market 5% 20%
Typical Individual Stock 5% 30%-40%

▶ Is this a paradox?
▶ No. Individual stocks, in general, are not “well diversified”...
▶ The component of risk that can be “diversified away” is not
“priced” by the market

7 / 38
Diversification: the one “free lunch” in finance!

▶ Benefits of diversification come without having to give up returns -

the one “free lunch” in finance!
▶ Because it’s easy to diversify :)
▶ although, retail investors usually do not use this free lunch :/

(Example: A naive investor with undiversified portfolio on Webull)

8 / 38
Risk and Return of Portfolios
Key insights

▶ By combining individual securities in a portfolio you can

reduce the volatility as the stock returns do not move in
lockstep → some ups and downs cancel out
▶ The more “common risks” that affect all stocks — returns
move together? — the lower the risk reduction
▶ When correlation is +1 [just common risk!], no diversification benefits

▶ Diversification helps with independent variations, or when

weak correlation...
▶ Also, negative correlation of an asset with the rest of the
portfolio insures against bad outcomes → Hedging example: Gold

9 / 38
Gains from Diversification, historical evidence
▶ good news: Most of the diversification benefits come from
adding the first 15-30 securities (chosen at random)

”Triumph of the Optimists: 101 Years of Global Investment Returns”, by Elroy Dimson, Paul Marsh, Mike Staunton

10 / 38
Limits to Diversification

▶ Total portfolio risk will typically not go down to zero even

when you hold a very large number of securities:
▶ Diversify away firm-specific, non-systematic, idiosyncratic,
unique risk
▶ Left with exposure to non-diversiftable, systematic,
“market” risk
▶ common risk in a single asset: the covariance term (joint
movement of an asset return and the pool of assets)
▶ total variance = common variation part (covariance) +
idiosyncratic variation, (diversifiable)
▶ “risk is about covariance, not variance...”

Example: Equally-weighted portfolios

Example: Factor model return correlations

11 / 38
Efficient portfolio: a first glance

▶ How to mix all assets in an optimal way? equally weighted?

value weighted? etc?
▶ Efficient (risky) portfolio:
The one mix of individual risky securities, that gets the most
out of diversification/hedging feature of assets
▶ Eliminate most of idiosyncratic risks, and leaves us with
minimal common/systemic risk
▶ Let’s do a generic formulization...

12 / 38
Modern Portfolio Theory (MPT)

Markowitz’s Modern Portfolio Theory (published 1952, Nobel

prize 1990)
▶ Start with individual securities whose investment properties
are summarized by:
▶ Expected return, eg estimated as the historical average return
▶ Risk or volatility, eg estimated as the standard deviation of
historical returns
∗ ↑ the input list
▶ We can find out the risk and return of any given portfolio of
securities from this input list statistics formula

▶ Optimize portfolio weights!

13 / 38
MPT, Assumptions

▶ Assume that investors are risk averse:

▶ for the same expected return, prefer less risk (=standard
deviation)
▶ for the same risk, prefer more return
▶
▶ Assume that only mean and variance matters...
1. Investors do not care about higher moments (skewness, etc)
2. Security returns are normally distributed (all moments are fully
described by mean and std)
▶ Assume that markets are frictionless:
▶ Securities may be traded at any price/quantity without
transaction costs or other restrictions
▶ Examples for such restrictions?
▶ Simplifies analysis, may or may not have an important effect

14 / 38
MPT, Optimal Portfolio Choice

How should investors choose an optimal portfolio?

1. find out the mean-variance frontier analytical representation

2. find the “efficient portfolio”

3. mix the efficient portfolio with the riskfree asset

let’s see this in pictures...

15 / 38
Example: Portfolios of 2 Risky Securities
▶ Basic Information
Stock Expected Return Volatility Correlation
Intel Coca Cola
Intel 26% 50% 1 0
Coca Cola 6% 25% 0 1

Example: Suppose you have $20,000 in cash to invest. 1) You decide

to invest $10,000 in Intel and 10,000 in Coca-Cola. 2) You decide to
short sell $10,000 worth of Coca-Cola stock and invest the proceeds from
your short sale, plus your $20,000, in Intel. What is the expected return
and volatility of your portfolio in either scenario of 1) and 2)? How about
if correlation was 0.5 or -0.5? Try alternative weights in 1), e.g.
$5,000:$15,000 or $15,000:$5,000?

Recall the 2-assets formula: E (rP ) = w1 E (r1 ) + w2 E (r2 ) and

σP2 = w12 σ12 + w22 σ22 + 2w1 w2 ρ1,2 σ1 σ2 , for w1 and w2 be portfolio shares
(w1 + w2 = 1)
16 / 38
Portfolios of 2 Risky Securities —
A portfolio that invests in Intel and Coca-Cola (assumed correlation = 0):

Note the efficient and the inefficient parts...

17 / 38
Portfolios of 2 Risky Securities — Correlation matters!
A portfolio that invests in Intel and Coca-Cola, with different assumed
correlations:

See, the STD is additive only when ρ = 1

18 / 38
Portfolios of 2 Risky Securities — Short-selling helps!
A portfolio that invests in Intel and Coca-Cola (assumed correlation =
0), while allowing for short-selling:

19 / 38
The Efficient Frontier

▶ Simple with two assets.

▶ With many assets:
▶ Many more possibilities for means and variances.
▶ Many portfolios will deliver the same expected return.
▶ Simplify the mean-variance optimization by identifying the
Efficient Frontier:
▶ Portfolios with the lowest variance for a given E [˜r]
▶ Dominates all other portfolios with the same E [˜ r]
▶ Portfolios on the frontier are efficient portfolios
▶ Let’s see on the picture...

20 / 38
Portfolios of 3 Risky Securities

Starting from pairwise combinations of 3 stocks:

21 / 38
Portfolios of 3 Risky Securities — Efficient frontier

All possible combinations of 3 stocks:

22 / 38
Large Portfolios of Risky Securities

If you continue to add risky securities:

▶ efficient frontier improves...

23 / 38
Multiple Securities: Implications

▶ Adding securities shifts the frontier to the left

▶ ......but at a decreasing rate! As we add more and more
securities, these shifts become smaller and smaller: decreasing
marginal benefits of diversification...
▶ Individual securities are in general dominated by the
frontier! (regardless)

24 / 38
Adding a Riskfree Asset
▶ Any portfolio p (on the efficient frontier, or in general) can be
combined with the riskfree asset that has:
▶ return rf
▶ zero variance and zero covariance with all risky assets
▶ example: T-bills
▶ Expected return of portfolio Q:

r¯Q = (1 − wp )rf + wp r¯P

▶ Variance of portfolio Q:
q
σQ = (1 − wp )2 × var (rf ) + wp2 × var (rP ) + 2(1 − wp )wp cov (rf , rP )
q
σQ = wp2 × var (rP ) = wp σp since var (rf ) = cov (rf , rP ) = 0

σQ = w p σp

25 / 38
Adding a Riskfree Asset → Capital Allocation Line

▶ The expected return and variance of the combined position

are:
r¯Q = (1 − wp )rf + wp r¯P and σQ = wP σP
σQ
This implies: wP = σP
▶ Combine the two equations:
 
σQ σQ rP − rf )
(¯
r¯Q = 1 − rf + r¯P = rf + σQ
σP σP σp
▶ The possible risk-return combinations are given by a straight
line between the risk-free asset and the portfolio P, known as
the Capital Allocation Line

26 / 38
“Complete Portfolio”: Combining risky and risk-free asset
Investing in P and the riskfree asset (here x = wP ):

27 / 38
Sharpe Ratio

rP −rf )
(¯
▶ The Capital Allocation Line is: r¯Q = rf + σp σQ
▶ The slope of any Capital Allocation Line is given by:

Portfolio Mean “Excess” return E (rp ) − rf

=
Portfolio Volatility σp
▶ This is known as the Sharpe ratio and gives the
reward-to-risk ratio of portfolio P (and all investments along
the line)

28 / 38
Expanding the Mean-Variance Frontier
▶ Investors can choose the portfolio that (combined with the
risk-free asset) has the CAL with the highest Sharpe Ratio
→ the tangency or mean-variance efficient (MVE) portfolio

29 / 38
Review of efficient frontier

▶ Gives the set of efficient portfolios

▶ The portfolios with minimum variance given expected return.
▶ With the riskfree asset: the CAL with the highest Sharpe Ratio
▶ How do we construct our ideal portfolio?
▶ Analytical formulas
▶ Use a short-cut property called two-fund separation

30 / 38
Two-fund Separation

▶ Tobin (two-fund) separation theorem: identifying the

optimal portfolio for a risk-averse investor can be broken down
into two steps:
1. Find the optimal portfolio of risky securities - the tangency
portfolio
note: this does not depend on risk preferences (same for all
investors)
2. Based on the specific risk preferences of the investor,
determine the appropriate amount to invest in the two optimal
portfolios (tangency and risk-free).

31 / 38
Two-fund Separation
Tobin step 1: how do we identify the tangency portfolio?
▶ The risky portfolio that maximizes the Sharpe ratio:

▶ Analytical:
PN
maxw E (rσP )−r
P
f
, such that: E (rP ) = i=1 wi E (ri )
N N N
σP2 = i=1 j=1 wi wj σi,j and
P P P
i=1 wi = 1

32 / 38
Two-fund Separation
Tobin step 2: for a given investor, how do we identify how
much to invest in the tangency portfolio versus riskfree?

▶ NOTE: Step 1 does not tell us which efficient (complete)

portfolio to choose
▶ That depends on risk aversion...
▶ Eg very risk-averse investors will choose the minimum
variance portfolio (the risk-free)...
▶ That’s why we need Tobin step 2
▶ Two approaches:
1. Based on risk preferences (utility functions) - lower risk
aversion, invest more in the risky fund
2. Specify a target portfolio expected return (or risk)

33 / 38
Example
Let’s say you want a target expected return of 15%:

1. identify the efficient portfolio [E (rT ) = 20%, σT = 14%]

2. invest 2/3 in the efficient, 1/3 in the riskfree.
▶ check that E (rP ) = 15%. Note that σP = 2/3 ∗ .14 = 9.33%

34 / 38
MPT in Practice, Issues

▶ Issues with inputs:

▶ how to estimate the inputs for large N (answer: use factor
models)
▶ expected returns, variances, correlations all estimated with
error
▶ expected returns, variances, and correlations vary a lot over
time
▶ Issues with outputs:
▶ Portfolio weights very sensitive to small changes in inputs
▶ Unconstrained optimization often leads to portfolios with large
short positions, levered positions, etc
▶ Typically deal with this by imposing constraints on portfolio
weights
▶ Other solutions (Google if interested): robust statistical
estimators, resampling, economic models, etc

35 / 38
Two-fund Separation
Tobin step 1: revisited

▶ We may start with tradable risky asset classes in the market

→ build an efficient risky portfo by mixing them...
▶ Example: Equity/Debt funds, 60:40 weights: analytical derivations

36 / 38
Two-fund separation in practice
Example from a Private banking company:

37 / 38
Two-fund separation in practice

ABN-AMRO MeesPierson Private Banking asset allocation profiles:

Quiz: does this portfolio advise follow the two-funds approach?

38 / 38
Where are we?

“Basics of Portfolio Choice”

Today’s Lecture:
▶ Markowitz’s Modern Portfolio Theory
▶ Diversification
▶ Mean-variance frontier
▶ Sharpe Ratio, Tangency/Efficient portfolio
▶ Two-fund separation

Next Lecture:
▶ CAPM — Passive strategies, Index Funds, ETFs
Key Terms

▶ Diversification
▶ Idiosyncratic risk
▶ Systematic risk
▶ Minimum variance/efficient frontier
▶ Capital allocation line
▶ Sharpe ratio
▶ Tangency portfolio
▶ Two-funds separation
APPENDIX
Diversification in the Limit
Example: Variance of an Equally Weighted Portfolio

▶ Variance of a portfolio of N stocks [the general formula]

n X
X n n
X XX
σp2 = wi wj σij = wi2 σi2 + wi wj σij
i=1 j=1 i=1 i=1 j̸=i

▶ If we set wi = wj = 1/n we have∗ :

1
Var (rP ) =
(Average Variance of the Individual Stocks)
n
 
1
+ 1− (Average Covariance between the Stocks)
n
*Note that there are n variances each with 1/n2 weight. There are n2 − n
covariance terms each with weight 1/n2 .

go back
Example: Factor model return correlations
Example: Factor model return correlations
▶ Consider the following model of returns for securities:

R1 = α1 + β1 Rm + ε1

R2 = α2 + β2 Rm + ε2
..
.

▶ in which αs are constants and ε are noise with E(ε) = 0

▶ Stocks’ returns depend on a common (random) factor Rm
▶ Assume that the errors are uncorrelated with each other and
with the common factor Rm , i.e.,
Cov (εi , εj ) = 0, Cov (εi , Rm ) = 0, Cov (εj , Rm ) = 0
▶ From this, it follows that Cov (Ri , Rj ) = βi βj Var (Rm )
▶ What is the variance of the equally weighted portfolio
rp = n1 (r1 + r2 + ... + rn ) in the limit n → ∞?
go back
Example: Factor model return correlations
The case with βs = 1

0.4

0.35

0.3

0.25

0.2

5 10 15 20 25 30 35 40

go back
Mean and Variance of a Portfolio of Assets
Mean and Variance of a Portfolio of Assets

▶
n
X
E [rp ] = wi E [ri ]
i=1

▶
n X
X n
σp2 = wi wj Cov (ri , rj )
i=1 j=1

▶ wi s are portfolio weights

go back
Tobin step 1: 2 Asset case
Tobin step 1: 2 Asset case

▶ To compute the mean-variance efficient (MVE), or tangency

portfolio, maximize the Sharpe ratio:
E (rp )−rf
max σp
w
where E (rp ) = wE (rA ) + (1 − w)E (rB )
1/2
σp = w2 σA2 + (1 − w)2 σB 2 + 2w(1 − w)ρ

A,B σA σB

▶ The solution to this is ugly

go back
Solution of the MVE problem

▶ The tangency portfolio weight for A:

2 − E (R ) ρ
E (RA ) σB σ σ
wT =
2

2
 B
A,B A B

E RA σB + E RB σA − E RA + E RB ρA,B σA σB

▶ Where R denotes the excess return

E (Ri ) = E (ri ) − rf i = A, B

▶ (sometimes we show it also with the notation rie )

go back
Numerical Example

Asset Expected Return Volatility Correlation

Asset A Asset B Risk-free Asset
Asset A 10% 20% 1 0.5 0
Asset B 15% 30% 0.5 1 0
Risk-free Asset 3% 0% 0 0 1
Minimum variance 10.75% 19.64% wA = 0.85
(only risky assets) we already saw how to derive this portfolio.

go back
Numerical Example

▶ Tangency portfolio weight on asset A:

▶ How much you should put in B?

go back
The tangency portfolio

Given the weights, we can compute the risk and return of the
tangency portfolio:

E (rT ) = wT E (rA ) + (1 − wT ) E (rB )

= 0.5 · 0.10 + 0.5 · 0.15 = 0.125 = 12.5%
1/2
σT = w σA + (1 − w)2 σB
 2 2 2
+ 2w(1 − w)ρA,B σA σB = 21.79%

go back
Maximal Sharpe Ratio?

▶ This portfolio should have the maximal Sharpe ratio

▶ Does it? Compute Sharpe Ratios

go back
The Sharpe Ratio

▶ Sharpe ratios of the stocks A and B

E(rA )−rf 0.10−0.03
SRA = σA = 0.20 = 0.35
E(rB )−rf 0.15−0.03
SRB = σB = 0.30 = 0.40

▶ Sharpe ratio of the tangency portfolio

E (rT ) − rf 0.125 − 0.03

SRT = = = 0.4359
σT 0.2179

go back
Example: Hedging with gold!
2 Risky Assets: Gold and Market — Hedging

▶ Basic Information

Asset Expected Return Volatility Correlation

S&P500 Gold
S&P500 12.80% 18.30% 1 -0.4
Gold 8.80% 20.80% -0.4 1

▶ Gold has a lower return and higher volatility than the S&P 500
▶ Is Gold a bad investment?

go back
Risk and Return: Portfolios of 2 Risky Assets

▶ Gold does not look good on its own.

▶ Lower return than the stock market (8.8% vs. 12.8%)
▶ Higher volatility than the stock market (20.8% vs. 18.3%)
▶ BUT, adding it to a portfolio can reduce total variance
▶ It acts as insurance against some forms of risk (inflation risk,
. . . etc.)
▶ Technically speaking, the low correlation coefficient (here
negative indeed!) with the S&P 500 makes gold a reasonably
good hedging instrument

go back
Portfolio Theory: A Two Asset Portfolio.

Investment strategy:
a fraction of wGold % in gold and (100 − wGold )% in the S&P500.

w Gold Return Variability

100% 8.80% 20.80% Hold only gold
80.0% 9.60% 15.54%
60.0% 10.40% 11.67%
45.40% 10.98% 10.65% Minimum variance portfolio
40.0% 11.20% 10.80%
20.0% 12.00% 13.52%
0.0% 12.80% 18.30% Hold only S&P500

go back
Return & Risk of Stock+Gold Portfolios
Average return is linear in wGold , Std is not!

Expected value is additive, but Standard deviation is not!

“Minimized” value at w ≃ 45%
go back
Efficient Frontier: Analytical Representation
Efficient Frontier
▶ i = 1, 2, ..., n individual securities
▶ inputs: expected returns E [ri ], variance-covariance Cov (ri , rj )
▶ wi s are portfolio weights, construct portfolio p: i wi = 1
P

▶ Mean, vectorized format:

n
X
E [rp ] = wi E [ri ] = w T E [r ]
i=1

▶ Variance, matrix format:

n X
X n
σp2 = wi wj Cov (ri , rj ) = w T Σw
i=1 j=1

▶ Optimization, portfolio choice:

min σp2
{wi }

s.t. E [rp ] ≥ r¯ , target rate

go back