Portfolio Optimization: Stephen Boyd EE103 Stanford University

Portfolio Optimization
Stephen Boyd
EE103
Stanford University
December 8, 2017
Outline
Return and risk
Portfolio investment
Portfolio optimization
Return and risk 2

Return of an asset over one period
I asset can be stock, bond, real estate, commodity, . . .

I invest in a single asset over period (quarter, week, day, . . . )
I buy q shares at price p (at beginning of investment period)
I h = pq is dollar value of holdings
I sell q shares at new price p+ (at end of period)
p+ −p
I profit is qp+ − qp = q(p+ − p) = p h
p+ −p profit
I define return r = p = investment
I profit = rh
I example: invest h = $1000 over period, r = +0.03: profit = $30
Return and risk 3

Short positions
I basic idea: holdings h and share quantities q are negative

I called shorting or taking a short position on the asset
(h or q positive is called a long position)
I how it works:
– you borrow q shares at the beginning of the period and sell them at
price p
– at the end of the period, you have to buy q shares at price p+ to
return them to the lender
I all formulas still hold, e.g., profit = rh
I example: invest h = −$1000, r = −0.05: profit = +$50
I no limit to how much you can lose when you short assets
I normal people (and mutual funds) don’t do this; hedge funds do
Return and risk 4

Examples
prices of BP (BP) and Coca-Cola (KO) for last 10 years

70
KO
BP
60
50
40
Prices
30
20
10
0
0 500 1000 1500 2000 2500
Days
Return and risk 5

Examples
zoomed in to 10 weeks
70
KO
BP
60
50
40
Prices
30
20
10
0
1600 1605 1610 1615 1620 1625 1630 1635 1640 1645 1650
Days
Return and risk 6

Examples
returns over the same period

0.2
KO
BP
0.15
0.1
0.05
Returns
−0.05
−0.1
−0.15
−0.2
1600 1605 1610 1615 1620 1625 1630 1635 1640 1645 1650
Days
Return and risk 7

Return and risk
I suppose r is time series (vector) of returns

I average return or just return is avg(r)
I risk is std(r)
I these are the per-period return and risk
Return and risk 8

Annualized return and risk
I mean return and risk are often expressed in annualized form

(i.e., per year)
I if there are P trading periods per year
– annualized return √
= P avg(r)
– annualized risk = P std(r)
(the squareroot in risk annualization comes from the assumption
that the fluctuations in return around the mean are independent)
I if returns are daily, with 250 trading days in a year
– annualized return √
= 250 avg(r)
– annualized risk = 250 std(r)
Return and risk 9

Risk-return plot
I annualized risk versus annualized return of various assets

I up (high return) and left (low risk) is good
25
SBUX
20
GS
Annualized Return
15
BRCM
MMM
10
USDOLLAR
0
0 10 20 30 40 50 60
Annualized Risk
Return and risk 10

Outline
Return and risk
Portfolio investment 11
Portfolio of assets
I n assets
I n-vector ht is dollar value holdings of the assets
I total portfolio value: Vt = 1T ht (we assume positive)
I wt = (1/1T ht )ht gives portfolio weights or allocation

(fraction of total portfolio value)
I 1T wt = 1
Examples
I (h3 )5 = −1000 means you short asset 5 in investment period 3 by

$1,000
I (w2 )4 = 0.20 means 20% of total portfolio value in period 2 is
invested in asset 4
I wt = (1/n, . . . , 1/n), t = 1, . . . , T means total portfolio value is
equally allocated across assets in all investment periods
Portfolio return and risk
I asset returns in period t given by n-vector r̃t

I dollar profit (increase in value) over period t is r̃tT ht = Vt r̃tT wt
I portfolio return (fractional increase) over period t is
Vt+1 − Vt Vt (1 + r̃tT wt ) − Vt
= = r̃tT wt
Vt Vt
I rt = r̃tT wt is called portfolio return in period t
I r is T -vector of portfolio returns
I avg(r) is portfolio return (over periods t = 1, . . . , T )
I std(r) is portfolio risk (over periods t = 1, . . . , T )
Compounding and re-investment
I VT +1 = V1 (1 + r1 )(1 + r2 ) · · · (1 + rT )
I product here is called compounding
I for |rt | small (say, ≤ 0.01) and T not too big,
VT +1 ≈ V1 (1 + r1 + · · · + rT ) = V1 (1 + T avg(r))
I so high average return corresponds to high final portfolio value

I Vt ≤ 0 (or some small value like 0.1V1 ) called going bust or ruin
Constant weight portfolio
I constant weight vector w, i.e., wt = w for t = 1, . . . , T

I requires rebalancing to weight w after each period
I define T × n asset returns matrix R with rows r̃tT
I so Rtj is return of asset j in period t
I then r = Rw
Cumulative value plot
I assets are Coca-Cola (KO) and Microsoft (MSFT)

I constant weight portfolio with w = (0.5, 0.5)
I V1 = $10000 (by tradition)
4
x 10
3
uniform portfolio
individual assets
2.5
2
Value
1.5
0.5
0
0 500 1000 1500 2000 2500
Days
Cumulative value plot
I w = (−3, 4)
I portfolio goes bust (drops to 10% of starting value)
4
x 10
3
leveraged portfolio
individual assets
2.5
2
Value
1.5
0.5
0
0 500 1000 1500 2000 2500
Days
Outline
Return and risk
Portfolio optimization 19
I how should we choose the portfolio weight vector w?

I we want high (mean) portfolio return, low portfolio risk
I we know past realized asset returns but not future ones

I we will choose w that would have worked well on past returns
I . . . and hope it will work well going forward (just like data fitting)
minimize std(Rw)2 = (1/T )kRw − ρ1k2

subject to 1T w = 1, avg(Rw) = ρ
I w is the weight vector we seek

I R is the returns matrix for past returns
I Rw is the (past) portfolio return time series
I require mean (past) return ρ
I we minimize risk for specified value of return
I we are really asking what would have been the best constant
allocation, had we known future returns
Portfolio optimization via least squares
minimize kRw − ρ1k2

T
1 1
subject to w=
µT ρ
I µ = RT 1/T is n-vector of (past) asset returns

I ρ is required (past) portfolio return
I equality constrained least squares problem, with solution
−1 
2RT R
   
w 1 µ 2ρT µ
 z1  =  1T 0 0   1 
z2 µT 0 0 ρ
Examples
I optimal w for annual return 1% (last asset is risk-less with 1%

return)
w = (0.0000, 0.0000, 0.0000, . . . , 0.0000, 0.0000, 1.0000)
I optimal w for annual return 13%
w = (0.0250, −0.0715, −0.0454, . . . , −0.0351, 0.0633, 0.5595)
I optimal w for annual return 25%
w = (0.0500, −0.1430, −0.0907, . . . , −0.0703, 0.1265, 0.1191)

I asking for higher annual return yields
– more invested in risky, but high return assets
– larger short positions (‘leveraging’)
Cumulative value plots for optimal portfolios
cumulative value plot for optimal portfolios and some individual assets
optimal portfolio, rho=0.20/250

optimal portfolio, rho=0.25/250
individual assets
5
10
Value
4
10
0 500 1000 1500 2000 2500

Days
Optimal risk-return curve
red curve obtained by solving problem for various values of ρ
25
20
Annualized Return
15
10
0
0 10 20 30 40 50 60
Annualized Risk
Optimal portfolios
I perform significantly better than individual assets

I risk-return curve forms a straight line
– one end of the line is the risk-free asset
I two-fund theorem: optimal portfolio w is an affine function in ρ
−1  T 
2RT R
  
w 1 µ R 1
 z1  =  1T 0 0   1 
z2 µT 0 0 ρT
The big assumption
I now we make the big assumption (BA):

future returns will look something like past ones
– you are warned this is false, every time you invest
– it is often reasonably true
– in periods of ‘market shift’ it’s much less true
I if BA holds (even approximately), then a good weight vector for past

(realized) returns should be good for future (unknown) returns
I for example:
– choose w based on last 2 years of returns
– then use w for next 6 months
I trained on 900 days (red), tested on the next 200 days (blue)
I here BA held reasonably well
Train
25 Test
20
Annualized Return
15
10
0
0 2 4 6 8 10 12 14 16
Annualized Risk
I corresponding train and test periods

4
x 10
5
4.5
4 Train Test
3.5
2.5
1.5
0.5
0
0 500 1000 1500 2000 2500
I and here BA didn’t hold so well

I (can you guess when this was?)
25 Train
Test
20
15
10
Annualized Return
−5
−10
−15
−20
0 2 4 6 8 10 12 14 16
Annualized Risk
I corresponding train and test periods

4
x 10
5
4.5
4 Train Test
3.5
2.5
1.5
0.5
0
0 500 1000 1500 2000 2500
Rolling portfolio optimization
for each period t, find weight wt using L past returns
rt−1 , . . . , rt−L
variations:
I update w every K periods (say, monthly or quarterly)
I add cost term κkwt − wt−1 k2 to objective to discourage turnover,
reduce transaction cost
I add logic to detect when the future is likely to not look like the past
I add ‘signals’ that predict future returns of assets
(. . . and pretty soon you have a quantitative hedge fund)
Rolling portfolio optimization example
I cumulative value plot for different target returns

I update w daily, using L = 400 past returns
4
x 10
1.3
rho=0.05/250
rho=0.1/250
rho=0.15/250
1.25
1.2
1.15
Value
1.1
1.05
0.95
1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
Days
Rolling portfolio optimization example
I same as previous example, but update w every quarter (60 periods)

4
x 10
1.3
rho=0.05/250
rho=0.1/250
rho=0.15/250
1.25
1.2
1.15
Value
1.1
1.05
0.95
1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
Days

Portfolio Optimization: Stephen Boyd EE103 Stanford University

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Portfolio Optimization: Stephen Boyd EE103 Stanford University

Uploaded by

Copyright:

Available Formats

Portfolio Optimization

Return and risk

Return and risk 2

I asset can be stock, bond, real estate, commodity, . . .

Return and risk 3

I basic idea: holdings h and share quantities q are negative

Return and risk 4

prices of BP (BP) and Coca-Cola (KO) for last 10 years

Return and risk 5

Return and risk 6

returns over the same period

Return and risk 7

I suppose r is time series (vector) of returns

Return and risk 8

I mean return and risk are often expressed in annualized form

Return and risk 9

I annualized risk versus annualized return of various assets

Return and risk 10

Return and risk

I n-vector ht is dollar value holdings of the assets

I total portfolio value: Vt = 1T ht (we assume positive)

I wt = (1/1T ht )ht gives portfolio weights or allocation

I (h3 )5 = −1000 means you short asset 5 in investment period 3 by

I asset returns in period t given by n-vector r̃t

I so high average return corresponds to high final portfolio value

I constant weight vector w, i.e., wt = w for t = 1, . . . , T

I assets are Coca-Cola (KO) and Microsoft (MSFT)

Return and risk

I how should we choose the portfolio weight vector w?

I we know past realized asset returns but not future ones

minimize std(Rw)2 = (1/T )kRw − ρ1k2

I w is the weight vector we seek

minimize kRw − ρ1k2

I µ = RT 1/T is n-vector of (past) asset returns

I optimal w for annual return 1% (last asset is risk-less with 1%

w = (0.0000, 0.0000, 0.0000, . . . , 0.0000, 0.0000, 1.0000)

I optimal w for annual return 13%

w = (0.0250, −0.0715, −0.0454, . . . , −0.0351, 0.0633, 0.5595)

I optimal w for annual return 25%

w = (0.0500, −0.1430, −0.0907, . . . , −0.0703, 0.1265, 0.1191)

optimal portfolio, rho=0.20/250

0 500 1000 1500 2000 2500

red curve obtained by solving problem for various values of ρ

I perform significantly better than individual assets

I now we make the big assumption (BA):

I if BA holds (even approximately), then a good weight vector for past

I corresponding train and test periods

I and here BA didn’t hold so well

I corresponding train and test periods

for each period t, find weight wt using L past returns

I cumulative value plot for different target returns

I same as previous example, but update w every quarter (60 periods)

You might also like