Last Revised: 05/12/2023

2024 Level 1 - Quantitative Methods

Readings Page

Rates and Returns 2

The Time Value of Money in Finance 8

Statistical Measures of Asset Returns 17

Probability Trees and Conditional Expectations 23

Portfolio Mathematics 27

Simulation Methods 31

Estimation and Inference 36

Hypothesis Testing 41

Parametric and Non-Parametric Tests of Independence 45

Simple Linear Regression 49

Introduction to Big Data Techniques 58

This document should be used in conjunction with the corresponding readings in the 2024 Level 1 CFA® Program curriculum.
Some of the graphs, charts, tables, examples, and figures are copyright 2023, CFA Institute. Reproduced and republished with
permission from CFA Institute. All rights reserved.

Required disclaimer: CFA Institute does not endorse, promote, or warrant accuracy or quality of the products or services
offered by MarkMeldrum.com. CFA Institute, CFA®, and Chartered Financial Analyst® are trademarks owned by CFA
Institute.

© 2533695 Ontario Limited d/b/a MarkMeldrum.com. All rights reserved.

1
 Last Revised: 05/12/2023

Rates and Returns

a. interpret interest rates as required rates of return, discount rates, or opportunity

costs and explain an interest rate as the sum of a real risk-free rate and
premiums that compensate investors for bearing distinct types of risk

b. calculate and interpret different approaches to return measurement over time

and describe their appropriate uses

c. compare the money-weighted and time-weighed rates of return and evaluate the
performance of portfolios based on these measures

d. calculate and interpret annualized return measures and continuously

compounded returns, and describe their appropriate uses

e. calculate and interpret major return measures and describe their appropriate
uses

2
 Last Revised: 05/12/2023

Rates and Returns

Page 1
Interest rates (r) - can be thought of in 3 ways
1/ required rates of return ➞ determining the FV of a PV
2/ discount rate ➞ determining the PV of a FV
3/ opportunity cost ➞ value forgone (current consumption vs. saving)
Determinants of Interest Rates
𝐫= 𝐫𝐟 ➞ real default risk-free rate (single period)
vary over
+ inflation premium ➞ expected inflation over a period of time
time
and + default risk premium ➞ compensates for credit risk
continuously
+ liquidity premium ➞ risk of loss vs. fair value if an investment
change
needs to be converted to cash quickly
+ maturity premium ➞ compensation for greater price sensitivity
from changes in rates

Page 2

nominal risk-free rate:

(1 + r) = (1 + 𝐫𝐟 ) (1 + 𝛑𝐞 ) or r = 𝐫𝐟 + 𝛑𝐞
Note: all rates are quoted on an annual basis
e.g. 3-mos. T-Bill @ 4% is . 𝟎𝟒$𝟒 = 1% over 3 months

3
 Last Revised: 05/12/2023

Page 3

Rates of Return/
1/ HPR - holding period return e.g./
𝐏𝐭 - ending price 105
R = (𝐏𝐭 − 𝐏𝟎 ) + 𝐈
𝐏𝟎 𝐏𝟎 - beginning price 100
𝐈 - all income ∅
HPR = (𝟏𝟎𝟓 − 𝟏𝟎𝟎)
= 5%
- multi-period HPR 𝟏𝟎𝟎

R = [(1 + HPR1)(1 + HPR2)(1 + HPR3)] - 1

𝐓
2/ Arithmetic or Mean Return 𝟏
(𝐢 =
𝐑 - 𝐑 𝐢𝐭
𝐓
𝐭.𝟏

3/ Geometric Mean Return 𝐑 ( 𝐆 = [(𝟏 + 𝐑 𝐢𝟏 )(𝟏 + 𝐑 𝐢𝟐 ) … (𝟏 + 𝐑 𝐢𝐓 )]𝟏0𝐓 − 𝟏

𝐢
b e.g./ 𝐑 𝐢𝟏 = -50%
𝟏
𝐑 𝐢𝟐 = 35% [(. 𝟓)(𝟏. 𝟑𝟓)(𝟏. 𝟐𝟕)] )𝟑 − 𝟏 = −𝟓% - the growth rate or
𝐑 𝐢𝟑 = 27% compounded return on
an investment

Page 4
Rates of Return/
𝐑𝐆 ≤ 𝐑𝐀 - unless all observations are equal, then 𝐑 𝐆 = 𝐑 𝐀
- as the variability in the data increases, difference
growth of $1 avg. between 𝐑 𝐆 and 𝐑 𝐀 increases
return
PV(1 + RG)N = FV
over time PV(1 + RA)N ≠ FV
- use to estimate
- use to estimate E(R) over one period
E(R) over multiple
periods

4/ Harmonic Mean
𝐧 - arithmetic mean - all obs. have equal weight
4𝐇 =
𝐗
𝟏
∑7 8𝐗9 obs. weight inversely proportional to its magnitude
∴ reduces the effect of outliers
b - most often used with ratios (amount/unit)
e.g./ P/E 45, 15, 15
= 𝟑8 ̇ = 𝟏𝟗. 𝟐𝟖
𝟑
4𝐇 =
𝐗
𝐑 𝐀 = 25 ∑,𝟏)𝟒𝟓 0 𝟏)𝟏𝟓 0 𝟏)𝟏𝟓1 . 𝟏𝟓

4
 Last Revised: 05/12/2023

Page 5
Rates of Return/
4/ Harmonic Mean
applied e.g. - dollar-cost averaging (typical DC strategy)
1,000/month for 2 months in a stock P0 = 10 P1 = 15
&𝐇 =
𝐗
𝟐
= 𝟐$ ̇ = 𝟏𝟐/𝐬𝐡. Proof: 𝟏𝟎𝟎𝟎1 = 100 sh.
𝟏𝟎
𝟏𝟎𝟎𝟎1 = 66.67 sh.
𝟏𝟓
#𝟏&𝟏𝟎 ( 𝟏&𝟏𝟓* . 𝟏𝟔
𝟐𝟎𝟎𝟎1
𝟏𝟔𝟔. 𝟔𝟕 𝐬𝐡. = 12/sh.
Relationship: 𝐑 𝐀 × 𝐗
( 𝐇 = 𝐑𝟐𝐆

5/ Others
a) trimmed mean (common with CPI)
oves - remove a %’age from both the largest and smallest
rem ers
i e.g. 100 obs., 8% trimmed = 84 obs. (8 highest, 8 lowest)
outl
bo th
(on ides) b) winsorized mean - replacing values at both end with
b s
the cutoff value
e.g. 100 obs. obs. 1-8 replaced all = obs. 9
obs. 93-100 replaced all = obs. 92

Page 6
Money-weighted Return (IRR, YTM)
- accounts for the timing and magnitude of investments
10
5 470
committed more
200 225
CF0 = -200 CF1 = -220 CF2 = 480 CPT IRR = 9.39% money to a poor
performance year
𝟐𝟓 + 𝟓 𝟐𝟎 + 𝟏𝟎
𝐇𝐏𝐑 = = 𝟏𝟓% 𝐇𝐏𝐑 = = 𝟔. 𝟔𝟕% 𝐑𝐀 = 𝟏𝟎. 𝟖𝟒% (money weighted)
𝟐𝟎𝟎 𝟒𝟓𝟎
- not comparable across investors/investments
- mwrr represents what ‘your’ money earned, not what $1 could earn

Time-weighted returns (= RG)

- the growth of $1 over a given time period
- comparable across investments
c
➞ break investment period into holding periods (determined by any
significant cash in/out-flows)
➞ calculate each HPR, then compound the HPRs, express annually

5
 Last Revised: 05/12/2023

Page 7
Time-weighted returns (= RG)
- previous example from mwrr
𝟏)
=(𝟏. 𝟏𝟓)7𝟏. 𝟎𝟔𝟔̇9@ 𝟐 − 𝟏 = 𝟏𝟎. 𝟕𝟓𝟒𝟗𝟖%
- large funds, HPR = 1 day ➞ (1 + HPR1)(1 + HPR2) + ... + (1 + HPR365) - 1

for liquid underlyings with

market prices

Page 8

Annualized Return/ - all rates/returns are quoted annually

𝟑𝟔𝟓0
𝐑 𝐚𝐧𝐧𝐮𝐚𝐥 = 7𝟏 + 𝐑 𝐩𝐞𝐫𝐢𝐨𝐝 8 𝐩𝐞𝐫𝐢𝐨𝐝 ➞ daily (𝟏 + 𝐑)𝟑𝟔𝟓 − 𝟏 7d/wk. 365d/yr.
weekly (𝟏 + 𝐑)𝟓𝟐 − 𝟏 52 wk./yr.
period = 43 days monthly (𝟏 + 𝐑)𝟏𝟐 − 𝟏 12 mos./yr.
𝟑𝟔𝟓
𝐑 𝐚𝐧𝐧𝐮𝐚𝐥 = (𝟏 + 𝐑 𝟒𝟑 ) )𝟒𝟑 −𝟏 quarterly 𝟒
(𝟏 + 𝐑) − 𝟏

period = 540 days 250 trading days

𝟑𝟔𝟓
𝐑 𝐚𝐧𝐧𝐮𝐚𝐥 = (𝟏 + 𝐑 𝟓𝟒𝟎 ) )𝟓𝟒𝟎 −𝟏 5 d/wk., 20 d/m., 250 d/yr.

- annualizing returns can be misleading - assumes that returns can be

repeated
Continuously Compounded Returns
d e.g./
𝐏
𝐈𝐧 : 𝐭;𝐏 < 𝐫𝐜 = 𝐈𝐧 C𝟏𝟎𝟓8𝟏𝟎𝟎D = 𝟒. 𝟖𝟕𝟗
𝟎 𝐞𝐫𝐜 − 𝟏 = 𝐫
or 𝐈𝐧 (𝟏 + 𝐫) 𝐫𝐜 = 𝐈𝐧 (𝟏. 𝟎𝟓) 𝐞.𝟎𝟒𝟖𝟕𝟗 − 𝟏 = 𝟓%

6
 Last Revised: 05/12/2023

Page 9
Gross and Net Return/
gross - return before deductions for mgmt. exp., custodial fees, taxes, etc.
but after trading expenses (what the fund earns)
- appropriate measure for evaluating and comparing the
investment skill of managers
net - what the investor earns

Pre-tax and After-tax Nominal Return/

- default is to report/state pre-tax return
- each investor’s marginal tax rate may differ 2% div.
- components of return may be reported e.g. 11% 1% int.
4% realized
Real Returns/ gains
𝟏 + 𝐧𝐨𝐦𝐢𝐧𝐚𝐥 𝐫𝐞𝐭𝐮𝐫𝐧
e 𝟏 + 𝐧𝐨𝐦𝐢𝐧𝐚𝐥 𝐫𝐞𝐭𝐮𝐫𝐧 = 𝟏 + 𝐫𝐢𝐬𝐤 𝐩𝐫𝐞𝐦𝐢𝐮𝐦
𝟏 + 𝐫𝐞𝐚𝐥 𝐫𝐞𝐭𝐮𝐫𝐧 = 𝟏 + 𝐫𝐢𝐬𝐤 − 𝐟𝐫𝐞𝐞 𝐫𝐚𝐭𝐞
𝟏 + 𝐢𝐧𝐟𝐥𝐚𝐭𝐢𝐨𝐧 𝐫𝐚𝐭𝐞
(typically p-o-p CPI)

Page 10
Real Returns/
After-tax real return ➞ investor measure of growth in
purchasing power of portfolio
Leveraged Returns/
- leverage can be obtained through margin loans, derivatives,
or collateralized loans (repos)
- if 𝐑 𝐏 > 𝐫𝐝 , leverage enhances return
Value borrowed

𝐑 𝐋 = 𝐑 𝐏; = 𝐑 𝐏 × (𝐕𝐄 + 𝐕𝐁 ) − (𝐕𝐁 × 𝐫𝐛 ) = 𝐑 + 𝐕
𝐏𝐄 𝐕𝐄 𝐏 𝐁
(𝐑 − 𝐫𝐝 )
leveraged 𝐕𝐄 𝐏
return portfolio
equity 𝐑 𝐏 𝐕𝐄 𝐑 𝐏 𝐕𝐁 𝐕𝐁 𝐫𝐝 𝐕𝐁 (𝐑 𝐏 − 𝐫𝐝 )
? + − = 𝐑𝐏 + @
𝐕𝐄 𝐕𝐄 𝐕𝐄 𝐕𝐄
e.g./ 10m equity portfolio
e
𝐑 𝐏 = 8%
𝐑 𝐋 = 𝟖% + 𝟑𝐌8𝟕𝐌 (𝟖% − 𝟓%) = 𝟗. 𝟐𝟖𝟓𝟕%
30% debt financed, 𝐫𝐝 = 5%

7
 Last Revised: 05/12/2023

The Time Value of Money in Finance

a. calculate and interpret the present value (PV) of fixed-income and equity
instruments based on expected future cash flows

b. calculate and interpret the implied return of fixed-income instruments and

required return and implied growth of equity instruments given the present
value (PV) and cash flows

c. explain the cash flow additivity principle, its importance for the no-arbitrage
condition, and its use in calculating implied forward interest rates, forward
exchange rates, and option values

8
 Last Revised: 05/12/2023

The Time Value of Money in Finance

Page 1
3 rules of money
larger CFs are worth more
less risky CFs are worth more (lower discount rate)
CFs sooner are worth more (time value of money)
risky safe risky
176 213 306
which one is best?
t=0 t=1 t=2 t=3

- calculate PV ➞ value today - requires discounting at a rate that

depends on the timing and type of CF
recall: 𝐫𝐟 + 𝛑𝐞 - gov’t. bonds
+ default
corporate/private debt
+ liquidity
+ maturity - longer-term debt
+ equity - equity over debt

Page 2

FV = PV(𝟏 + 𝐫)𝐭 or FV = 𝐏𝐕𝐞𝐫𝐓

𝐅𝐕
PV = = FV(𝟏 + 𝐫)A𝐭 PV = 𝐅𝐕𝐞A𝐫𝐓
(𝟏 + 𝐫)𝐭

single cash flow.

Fixed Income/ debt instruments (bonds, loans, mortgages, etc.)
ZCB - zero coupon bonds (i.e. T-Bills ➞ up to 1 yr. mat.)
- sold at a discount, mature at par - single CF at maturity
Coupon bonds (Notes, Bonds)
- investor receives a number of interest payments over
time and par at maturity
Fully ammortizing bonds (mortgage, auto loan)
- investor receives level payments of both interest
and principal

9
 Last Revised: 05/12/2023

Page 3
ZCB/zero-coupon bond: PV = 𝐅𝐕8(𝟏 + 𝐫)𝐓
r = discount rate, IRR, or YTM
e.g./
20-yr. ZCB, YTM = 6.7% ➞ PV = 𝟏𝟎𝟎$(𝟏. 𝟎𝟔𝟕)𝟐𝟎 = 27.33453
(TVM keys FV = 100 𝐈#𝐘 = 6.7 PMT = 0 N = 20 CPT PV)
price in 3 years if YTM is unchanged? PV = -27.33453
N=3
a) FV3 = PV(𝟏 + 𝐫)𝟑 = 27.33453(1.067)3 = 33.20510591 𝐈1 = 6.7
𝐘
or b) PV3 = 𝐅𝐕#(𝟏 + 𝐫)𝟏𝟕 = 𝟏𝟎𝟎#(𝟏. 𝟎𝟔𝟕)𝟏𝟕 = 33.20510591 PMT = 0
FV = 0 CPT FV
PV = 22.68224 ➞ YTM = ? 𝐈1 = 6.7%
𝐘
22.68224 = 𝟏𝟎𝟎#(𝟏 + 𝐫)𝟐𝟎 ➞ PV = -22.68224 PMT = 0
N = 20 N = 17 CPT PV
𝟏
(𝟏𝟎𝟎/𝟐𝟐. 𝟔𝟖𝟐𝟐𝟒) )𝟐𝟎 -1 = r PMT = 0
FV = 100 CPT 𝐈'𝐘 = 7.6999 ~ 7.7%
r = - .05% , 10 yr. ZCB
FV = 100 N = 10 𝐈1 = - .05 PMT = 0
PV = 𝟏𝟎𝟎#(. 𝟗𝟗𝟗𝟓)𝟏𝟎 = 100.50137 𝐘
CPT PV
6 yrs. later P0 = 95.72, YTM = ? 95.72 = 𝟏𝟎𝟎'(𝟏 + 𝐫)𝟒 ➞ (𝟏𝟎𝟎/𝟗𝟓. 𝟕𝟐) %𝟒 - 1 = 1.09957%
𝟏

PV = - 95.72 FV = 100 N=4 PMT = 0 CPT 𝐈1𝐘

Page 4

Coupon bond FV = 100

𝐏𝐌𝐓𝟏 𝐏𝐌𝐓𝟐 𝐅𝐕 + 𝐏𝐌𝐓𝐍 N = yrs. × period
PV = + + ⋯+
(𝟏 + 𝐫) (𝟏 + 𝐫)𝟐 (𝟏 + 𝐫)𝐍 PMT = coupon/period
𝐈# = 𝐘𝐓𝐌#
𝐘 𝐩𝐞𝐫𝐢𝐨𝐝
7 yr., 2% annual bond issued at YTM = 2%
par bond FV = 100 PMT = 2 𝐈#𝐘 = 2 N=7 CPT PV

one year later, P = 93.091 , YTM = ?

FV = 100 N=6 PMT = 2 PV = -93.091 CPT 𝐈#𝐘 = 3.287566

20 yr., 6.7% semi issued at YTM = 6.70% par bond

7.7% ? FV = 100 N = 20 × 2 = 40 PMT = 𝟔. 𝟕𝟎1𝟐 = 3.35 𝐈1 = 𝟕. 𝟕𝟎1 = 3.85
𝐘 𝟐
CPT PV = 89.8788

20 yr. ZCB assuming YTM = 6.7% semi

FV = 100 N = 40 𝐈# = 3.35 PMT = 0 CPT PV = 26.7658
𝐘
vs. 27.33 annually

10
 Last Revised: 05/12/2023

Page 5
Perpetuity (some bonds, preferred shares)
PV = 𝐏𝐌𝐓 e.g./ 3.3% qtly. coupon, P = 97.03
𝐫 YTM . 𝟖𝟐𝟓
97.03 = ➞ 𝐫𝟒 = .8502%
𝐫𝟒
Annuity (mortgage, car loans) 𝐫 = 3.401%

A= 𝐫(𝐏𝐕) or/ TVM keys ➞ CPT PMT

pmt
𝟏 − (𝟏 + 𝐫)A𝐭
N = 30 × 12 = 360
e.g./ 800k, 30 yr. - FRM @ 5.25% FV = 0 (fully amortizing)
PV = -800,000
𝐈' = 5.25/12 = .4375 CPT PMT
Amortization Table: 𝐘
4,417.63
PMT 𝒊 P B
Month 1 4417.63 3500 917.63 799,082.37
2 4417.63 3495.99 921.64 798,160.73

PMT - 𝐢 𝐁𝟎 − 𝐏 = 𝐁𝟏
B × 𝐫1𝟏𝟐

Page 6

Equity - pref. shares, common shares

constant dividend - perpetuity
growing dividend (constant rate) - growing perpetuity
growing dividend (non-constant rate)
Constant dividend/ - many REITs
D = 1.50 r = 15%
- perpetuity
PV = 𝐃;𝐫 PV = 𝟏. 𝟓𝟎#. 𝟏𝟓 = 10

Constant growth dividend - commercial real estate (to calculate

PV = 𝐃𝟎 (𝟏 + 𝐠) property values)
𝐫−𝐠 for CRE: Prop. Value = 𝐍𝐎𝐈𝟏
𝐫 − 𝐠 ➞ cap rate
Variable growth dividend - growth moving to value
terminal
- 2 stage model 𝐧 value
𝐃𝟎 (𝟏 + 𝐠 𝐬 ) )𝐧 (𝟏
𝐃𝟎 (𝟏 + 𝐠 𝐬 + 𝐠𝐋)
𝐏𝐕 = K + = perpetuity
(𝟏 + 𝐫) 𝐧 𝐫−𝐠
𝐢C𝟏
explicit discount (𝟏 + 𝐫)𝐧
period

11
 Last Revised: 05/12/2023

Page 7

e.g./ D0 = 1.50 g = 6% - growing perpetuity (typical of value stocks)

r = 15%
PV = 𝟏. 𝟓𝟎(𝟏. 𝟎𝟔) = 𝟏. 𝟓𝟗$
. 𝟏𝟓 − . 𝟎𝟔 . 𝟎𝟗 = 𝟏𝟕. 𝟔𝟕

➞ now assume 𝐠 𝐬 = 6% for 3 yrs., 𝐠 𝐋 = 2% thereafter

𝟐 𝟑 𝟏. 𝟓𝟎(𝟏. 𝟎𝟔)𝟑 (𝟏. 𝟎𝟐)
PV = 𝟏. 𝟓𝟎(𝟏. 𝟎𝟔) + 𝟏. 𝟓𝟎(𝟏. 𝟎𝟔) + 𝟏. 𝟓𝟎(𝟏. 𝟎𝟔) + . 𝟏𝟓 − . 𝟎𝟐
𝟏. 𝟏𝟓 (𝟏. 𝟏𝟓)𝟐 (𝟏. 𝟎𝟏𝟓)𝟑
(𝟏. 𝟏𝟓)𝟑
= 1.3826 + 1.2744 + 1.1747 + 𝟏𝟒. 𝟎𝟏𝟕𝟑𝟒
(𝟏. 𝟏𝟓)𝟑
= 13.04833

Page 8

recall: PV = 𝐅𝐕 ➞ PV(1 + r)T = FV

(𝟏 + 𝐫)𝐓
(1 + r)T = 𝐅𝐕;𝐏𝐕
𝟏
1 + r = 7𝐅𝐕8 9 )𝐓
𝐏𝐕
𝟏)
e.g./ ZCB/ r = 7𝐅𝐕8 9 𝐓
−𝟏
100 𝐏𝐕

t=0 t=6 t=10 𝟏)

r = C𝟗𝟓. 𝟕𝟐8𝟏𝟎𝟎D
𝟔
100.50 95.72 − 𝟏 = - .8088%
implied return? YTM? 𝟏&
r = N𝟏𝟎𝟎$𝟗𝟓. 𝟕𝟐P
𝟒
− 𝟏 = 1.10%

Coupon bond/
2 2 FV = ? = 2(1.02) + 2 + 93.091
100 93.091 = 97.131
implied ret. 𝟏&
r = Q𝟗𝟕. 𝟏𝟑𝟏$𝟏𝟎𝟎S 𝟐
− 𝟏 = -1.445%

12
 Last Revised: 05/12/2023

Page 9
Equity/ PV = 𝐃𝟎 (𝟏 + 𝐠) ➞ PV(r - g) = D0(1 + g)
𝐫−𝐠
r - g = 𝐃𝟎 (𝟏 + 𝐠)
e.g./ P0 = 63 D1 = 1.76 g = 4% 𝐏𝐕
r = 𝟏. 𝟕𝟔 r = 𝟎 + 𝐠)
𝐃 (𝟏 GGM
+ 𝟒% = 𝟔. 𝟕𝟗% + 𝐠
𝐏𝐕
𝟔𝟑 growth
- what if r = 7% ➞ g? div. yield

g = 7% - 𝟏. 𝟕𝟔#𝟔𝟑 = 4.21% g = r - 𝐃𝟎 (𝟏 + 𝐠)
𝐏𝐕
PV = 𝐃𝟎 (𝟏 + 𝐠) ➞ divide both sides by E
𝐫−𝐠
dividend payout ratio - DPR
𝐃𝟎8 (
𝐏𝐕 𝐄 𝟏 + 𝐠) - given a 𝐏#𝐄 and DPR - we can solve for
=
𝐄 𝐫−𝐠 𝐫 or 𝐠 given the other
PE ratio - compare the implied 𝐫 or 𝐠 to the required 𝐫 or
expected/estimated 𝐠

Page 10

forward P/E ➞ 𝐏𝐕8 𝐃𝐭(𝟏

𝐄𝐭0𝟏 = $𝐄
𝐭(𝟏 - if the forward div. is expected
𝐫−𝐠 to ↑ or 𝐠 is expected to ↑,
stock/index will trade at a
e.g./
1/ forward PE = 28 E(DPR) = 70% higher forward multiple
g = 4% r= ?
- a higher required return leads
28 = 𝟕𝟎% ➞ 28r - 1.12 = .7 to a lower multiple
𝐫 − . 𝟎𝟒 28r = 1.82
r = 1.82/28 = 6.5%

2/ forward PE = 19 E(DPR) = 60% 3/ forward PE = 28 E(DPR) = 70%

r = 8% g= ? r = 9% g = 4.5% buy/sell stock?

PE = .𝟕
19 = . 𝟔𝟎
➞ 1.52 - 19g = .6 = 15.5 ×
. 𝟎𝟗 − . 𝟎𝟒𝟓
. 𝟎𝟖 − 𝐠
.92 = 19g - should be trading at 15.5×
g = . 𝟗𝟐'𝟏𝟗 = 4.84% forward earnings, not 28 ∴ sell

13
 Last Revised: 05/12/2023

Page 11

Cash flow additivity ➞ principle of no arbitrage - 2 economically

equivalent strategies should have the same
price
r = 6% 45 45 45
which one? A
- calculate PV of both, select the
60 40 32.5 higher one
B or/
take the difference in CFs (A - B)
-15 5 12.50
A-B if PV > 0, choose A, else B
PV = 0 ➞ both equivalent
Proof/
A 100 - B is clearly superior
r = 6% 120 PV(A) = 94.339
B
- 20
PV(B) = 113.2075
A-B - 18.8679
−𝟐𝟎1
𝟏. 𝟎𝟔 = -18.8679

Page 12
r = 2.5% invest for 2 yrs.
𝐟𝟏,𝟏 buy a 2 yr. ZCB
t=0 t=1 t=2 buy a 1 yr. ZCB, buy another in one year
r = 3.5% spot 𝐟𝟏,𝟏
(1.035) = (1.025)(1 + 𝐟𝟏,𝟏 )
2
(implied forward rate)
(𝟏. 𝟎𝟑𝟓) 𝟐
𝐟𝟏,𝟏 = - 1 = 4.5097%
(𝟏. 𝟎𝟐𝟓)

- if 𝐟𝟏,𝟏 = 5% ➞ lock-in rate with FRA ➞ (1.025)(1.05) = 1.07625

vs.
- would provide an arbitrage profit (1.035)2 = 1.071225
of .5025 per 100 of par .005025
e.g./ 1 yr. 2 yr. ∆𝐟𝟏,𝟏 = ?
May 31 98.028 95.109 (𝟏𝟎𝟎/𝟗𝟖. 𝟎𝟐𝟖) − 𝟏 = 𝟐. 𝟎𝟏𝟏% 𝐟𝟏,𝟏 = (𝟏. 𝟎𝟐𝟓𝟑𝟗)𝟐
−𝟏
June 15 97.402 93.937 (𝟏𝟎𝟎/𝟗𝟓. 𝟏𝟎𝟗)
𝟏%
𝟐 − 𝟏 = 𝟐. 𝟓𝟑𝟗%
(𝟏. 𝟎𝟐𝟎𝟏𝟏)
= 3.069%

14
 Last Revised: 05/12/2023

Page 13

e.g./ 1 yr. 2 yr. ∆𝐟𝟏,𝟏 = ?

May 31 98.028 95.109 𝐟𝟏,𝟏 = 3.069%
June 15 97.402 93.937 ➞ 𝟏𝟎𝟎 (𝟏. 𝟎𝟑𝟏𝟕𝟔𝟕)𝟐
E #𝟗𝟕. 𝟒𝟎𝟐F − 𝟏 = 𝟐. 𝟔𝟔𝟕𝟑% −𝟏
𝟏& (𝟏. 𝟎𝟐𝟔𝟔𝟕𝟑)
E𝟏𝟎𝟎#𝟗𝟑. 𝟗𝟑𝟕F 𝟐
− 𝟏 = 𝟑. 𝟏𝟕𝟔𝟕%
= 3.6886
∆𝐟𝟏,𝟏 = 3.6886% - 3.069% = 61.964 bps

or/ 98.028/95.109 = 3.069%

3.688% - 3.069%
97.402/93.937 = 3.688%
forex - forward 1000 USD
exchange rate FV 𝟏𝟎𝟎𝟎𝐞𝐫𝐝𝐓 𝟏𝟎𝟎𝟎𝐒𝐟& 𝐞𝐫𝐟𝐓
𝐝
𝐅𝐟)
𝐝
𝟏𝟎𝟎𝟎𝐞𝐫𝐝 𝐓 𝐅𝐟) = 𝟏𝟎𝟎𝟎𝐒𝐟) 𝐞𝐫𝐟 𝐓
𝐝 𝐝
𝐝𝐚𝐲𝐬'
𝐅𝐟) = 𝐒𝐟) 𝐞𝐫𝐟 𝐓 ➞ 𝐒𝐟) 7
𝟏 + 𝐫8 𝟑𝟔𝟎>
?
𝐝 𝐝 𝐞𝐫𝐝 𝐓 𝐝 𝐝𝐚𝐲𝐬'
𝟏 + 𝐫8 𝟑𝟔𝟎>

Page 14

Forex/ e.g./ 𝐒𝐉𝐏𝐘) = 134.40 𝐫𝐔𝐒𝐃 = 2% 𝐫𝐉𝐏𝐘 = .05% 6 mos. 𝐅𝐉𝐏𝐘) =?

𝐔𝐒𝐃 𝐔𝐒𝐃
𝐞𝐫𝐉𝐏𝐘𝐓 𝐞.𝟎𝟎𝟎𝟓(.𝟓)
𝐅𝐉𝐏𝐘) = 𝐒𝐉𝐏𝐘) = 𝟏𝟑𝟒. 𝟒𝟎 S T = 𝟏𝟑𝟑. 𝟎𝟗𝟓𝟗
𝐔𝐒𝐃 𝐔𝐒𝐃 𝐞𝐫𝐔𝐒𝐃 𝐓 𝐞.𝟎𝟐(.𝟓)

e.g./ 𝐫𝐟 𝐫𝐝 𝐞.𝟎𝟐𝟏𝟐
May 31 2.012% 𝐅
1.291% ➞ 𝐟)𝐝 = 𝟏. 𝟐𝟔𝟎𝟐 = 𝟏. 𝟐𝟔𝟗𝟑𝟏𝟗
𝐞.𝟎𝟏𝟐𝟗𝟏
June 15 2.667% 1.562% 𝐞.𝟎𝟐𝟔𝟔𝟕
𝐅𝐟) = 𝟏. 𝟐𝟔𝟎𝟐 .𝟎𝟏𝟓𝟔𝟐 = 𝟏. 𝟐𝟕𝟒𝟐𝟎𝟐
𝐒𝐟0 = 1.2602 1 yr. ➞ ∆ 𝐅𝐟0 = ? 𝐝 𝐞
𝐝 𝐝
∆𝐅𝐟0 = ↑ .004883
𝐝
or 48.83 pips
Option pricing/
hS+ hS+ - c+ = hS- - c-
select
c+ hS+ - hS- = c+ - c-
hS h
-c such that h(S+ - S-) = c+ - c-
both outcomes ( 5
hS- h = 𝐜 −𝐜
c- are equal 𝐒( − 𝐒5

15
 Last Revised: 05/12/2023

Page 15
Option pricing/
u = 1.4 S+ = 56 h = Z 𝟔 − 𝟎 Z = 𝟔$ = . 𝟐𝟓
𝟓𝟔 − 𝟑𝟐 𝟐𝟒
c+ = 6
40 hS+ - c+ = .25(56) - 6 = 8
-c identical
- hS- - c- = .25(32) - 0 = 8
(X = 50) d = .8 S = 32 ∴ 0.25 S - c is
c- = 0 a risk-free portfolio
r = 5%
hS0 - c0 = 𝐡𝐒0 − 𝐜 0 .25(40) - c0 = . 𝟐𝟓(𝟓𝟔) − 𝟔
𝟏+𝐫 𝟏. 𝟎𝟓
c0 = 10 - 𝟖#𝟏. 𝟎𝟓 = 2.38095
Put
56
h = 𝟎 − 𝟏𝟖
0 Z Z = . 𝟕𝟓
𝟓𝟔 − 𝟑𝟐
40
+p .75(40) + p0 = . 𝟕𝟓(𝟑𝟐) + 𝟏𝟖
32 hS0 + p0 = 𝐡𝐒 + 𝐩 5 5 𝟏. 𝟎𝟓
18 𝟏+𝐫 𝟒𝟐
p0 = #𝟏. 𝟎𝟓 - 30
(X = 50)
r = 5% p0 = 10

16
 Last Revised: 05/12/2023

Statistical Measures of Asset Returns

a. calculate, interpret, and evaluate measures of central tendency and location to

address an investment problem

b. calculate, interpret, and evaluate measures of dispersion to address an

investment problem

c. interpret and evaluate measures of skewness and kurtosis to address an

investment problem

d. interpret correlation between two variables to address an investment problem

17
 Last Revised: 05/12/2023

Statistical Measures of Asset Returns

Page 1

Measures of central tendency/

- where the data are centered i.e. the expected value
∑𝐧𝐢7𝟏 𝐗 𝐢
Arithmetic mean &=
𝐗 - describes a representative
𝐧
possible outcome
- sensitive to outliers (extreme
values)
Median - middle value (𝐧 + 𝟏)
𝟐 odd # of obs.
not affected by
outliers 𝐧 𝐧+𝟐
N P+N P even # of obs. - average the
𝟐 𝟐
𝟐 middle 2 obs.

mode - most frequently occurring value in a dataset

- a dataset can have more than one mode, or no mode
- single mode ➞ unimodal all obs. are
- two modes ➞ bimodal different

Page 2

count % ∑count ∑%

Outliers/ do nothing - if values are legitimate and correct

- eliminates judgment
Delete ➞ trimmed mean
Replace ➞ winsorized mean
- compare mean with and without outliers

18
 Last Revised: 05/12/2023

Page 3
Measures of location - quantiles IQR - interquartile range
quartiles 25, 50, 75
quintiles 20, 40, 60, 80
deciles 10, 20 ... 90
Box and whisker plot percentiles 1, 2 ... 99

upper = (1.5 x IQR)

fence + upper
Box & whisker plot
bound
lower fence = lower bound - (1.5 x IQR)
uses/ rank performance of portfolios and investment managers
in terms of percentile/quartile in which they fall
investment research ➞ bottom return decile ➞ short long/short
➞ top return decile ➞ long HF

Page 4

Dispersion - variability around the central tendency

- a measure of risk or uncertainty

range: max. value - min. value

- no information about the shape of the distribution
- sensitive to outliers
mean absolute deviation 5|
MAD = ∑|𝐗 𝐢 − 𝐗
𝐧
sample variance 5)𝟐
𝐒 𝟐 = ∑𝐧𝐢C𝟏(𝐗 𝐢 − 𝐗 𝐧
pop: 𝛔𝟐 = ∑𝐢C𝟏(𝐗 − 𝛍)𝟐
in units squared 𝐧−𝟏
𝐧
- can be added/subtracted
degrees of freedom
(n - 1 independent obs.)

sample standard deviation &) 𝟐

S = √𝐒𝟐 = \∑𝐧𝐢7𝟏(𝐗 𝐢 − 𝐗 pop 𝛔 = √𝛔𝟐
- in the same units as the 𝐧
data itself

19
 Last Revised: 05/12/2023

Page 5

Downside deviation
e.g. target semideviation: 𝐧
𝐒𝐭𝐚𝐫𝐠𝐞𝐭 = !∑𝐢@𝟏(𝐗 𝐢 − 𝐁)
𝟐
∀𝐗 𝐢 ≤ 𝐁
𝐧−𝟏
some minimum
full sample n level
as B ↑ 𝐒𝐭𝐚𝐫𝐠𝐞𝐭 ↑

coefficient of variation CV = 𝐒84 - a measure of relative

𝐗
dispersion
e.g. - for returns, CV measures the risk per unit of return

- lower = better (less uncertainty, tighter distribution)

Page 6

Normal distribution - most common

symmetrical completely described by

2 parameters: 𝛍 , 𝛔𝟐

X ~ 𝐍(𝛍, 𝛔𝟐 )
mean = median = mode

Skew

Skew = 0
Skew > 0 Skew < 0

mean = median = mode mean > median > mode mean < median < mode

20
 Last Revised: 05/12/2023

Page 7
𝟏 ∑𝐧𝐢C𝟏(𝐗 𝐢 − 5)𝟑
𝐗
𝐒𝐤𝐞𝐰 ≈ for 𝐧 > 𝟏𝟎𝟎
𝐧 𝐒𝟑
5)𝟒
𝟏 ∑𝐧𝐢C𝟏(𝐗 𝐢 − 𝐗 - measures the combined weight of
Kurtosis 𝐊𝐄 = ? @−𝟑
𝐧 𝐒𝟒 the tails relative to the rest of the
distribution

leptokurtic (K > 3) mesokurtic (K = 3)

more
less weight
weight
platykurtic
(K < 3)
more weight less weight

Lepto ➞ 𝐊 𝐞 > 𝟎 - ow head, uw shoulders, ow tails

Meso ➞ 𝐊 𝐞 = 𝟎 - normal dist.
Platy ➞ 𝐊 𝐞 < 𝟎 - uw head, ow shoulders, uw tails

Page 8
Scatter Plot
- used to visualize the joint variation
in 2 numerical values
identify
outliers - may be no relationship, a linear or non-linear
relationship
- scatter plot matrix
- assess for pairwise association
among many variables

Covariance ➞ the joint variability of 2 random variables

➞ expressed in the same units as the
variables
()(𝐘𝐢 − 𝐘
∑𝐧𝐢.𝟏(𝐗 𝐢 − 𝐗 () 𝐒𝐗𝐘 > 0 when they covary together
𝐒𝐗𝐘 = 5) > 0 when (𝐘𝐢 − 𝐘
(𝐗 𝐢 − 𝐗 5) > 0
𝐧−𝟏
5) < 0 when (𝐘𝐢 − 𝐘
and (𝐗 𝐢 − 𝐗 5) < 0

21
 Last Revised: 05/12/2023

Page 9

Correlation ➞ measures the linear association between 2 variables

𝐒𝐗𝐘 Properties:
𝐫𝐗𝐘 =
𝐒𝐗 𝐒𝐘 1/ -1 ≤ 𝐫 ≤ 1
2/ 𝐫 = 0 implies no linear relationship ➞ maximum
𝐂𝐨𝐯(𝐗𝐘) diversification
determines the 3/ 𝐫 = 1 ➞ perfect positive correlation ➞ perfect
replication
sign of 𝐫𝐗𝐘 4/ 𝐫 = -1 ➞ perfect negative correlation ➞ perfect hedge

Limitations/ linear association only

unreliable when outliers are present
correlation does not imply causation
spurious correlation ➞ chance relationship
- a third variable

calculation 𝐚 ➞ 𝐗
𝐗$ 𝐫𝐗𝐘 𝐘$ 𝐫𝐗𝐘
𝐚 𝐚 𝐚 ➞ 𝐘

Page 10

for all
4
𝐧 = 11
&=9
𝐗
&
𝐘 = 7.5
𝐒𝐗 = 3.32
𝐒𝐘 = 2.03
𝐫𝐗𝐘 = .82

∴ visual inspection important

22
 Last Revised: 05/12/2023

Probability Trees and Conditional Expectations

a. calculate expected values, variances, and standard deviations and demonstrate

their application to investment problems

b. formulate an investment problem as a probability tree and explain the use of

conditional expectations in investment application

c. calculate and interpret an updated probability in an investment setting using

Bayes’ formula

23
 Last Revised: 05/12/2023

Probability Trees and Conditional Expectations

Page 1

Key point - investment decisions are made under uncertainty

- the expected value of a random variable ➞ E(X) ➞ is a

probability-weighted average of the possible outcomes

E(X) forecast of future value

estimate of the ‘true’ population mean based on a sample
𝐧

E(X) = 𝐏(𝐗 𝟏 )𝐗 𝟏 + 𝐏(𝐗 𝟐 )𝐗 𝟐 + … + 𝐏(𝐗 𝐧 )𝐗 𝐧 = H 𝐏(𝐗 𝐢 )𝐗 𝐢

𝐢(𝟏
one outcome only

Variance of a random variable

𝐧
𝟐 (𝐗) ]𝟐 ]𝟐
𝛔 = 𝐏(𝐗 𝟏 )[𝐗 𝟏 − 𝐄(𝐗) + … + 𝐏(𝐗 𝐧 )[𝐗 𝐧 − 𝐄(𝐗) = H 𝐏(𝐗 𝐢 )[𝐗 𝐢 − 𝐄(𝐗)]𝟐
𝐢(𝟏
𝛔(𝐗) = S𝛔𝟐 (𝐗)

Page 2

e.g./ P EPS (X)

.15 2.60 E(X) = .15(2.60) + .45(2.45) + .24(2.20) + .16(2.00)
.45 2.45 = 2.3405
.24 2.20
.16 2.00 .15 2.60
.45
2.45
.24
.16 2.20
2.00
𝛔𝟐 (𝐗) = . 𝟏𝟓(𝟐. 𝟔𝟎 − 𝟐. 𝟑𝟒𝟎𝟓)𝟐 + . 𝟒𝟓(𝟐. 𝟒𝟓 − 𝟐. 𝟑𝟒𝟎𝟓)𝟐 + . 𝟐𝟒(𝟐. 𝟐𝟎 − 𝟐. 𝟑𝟒𝟎𝟓)𝟐
+ . 𝟏𝟔(𝟐. 𝟎𝟎 − 𝟐. 𝟑𝟒𝟎𝟓)𝟐
= . 𝟎𝟑𝟖𝟕𝟖𝟓

𝛔(𝐗) = √. 𝟎𝟑𝟖𝟕𝟖𝟓 = . 𝟏𝟗𝟔𝟗𝟑𝟗

24
 Last Revised: 05/12/2023

Page 3

= 𝐏(𝐗 𝟏 |𝐒)𝐗 𝟏 + 𝐏(𝐗 𝟐 |𝐒)𝐗 𝟐 + 𝐏(𝐗 𝟑 |𝐒)𝐗 𝟑

mutually exclusive conditional
𝐗𝟏 probability
and exhaustive

unconditional 𝐒 𝐄(𝐗|𝐒) 𝐗𝟐 mutually exclusive

expectation and exhaustive
𝐄(𝐗) conditional 𝐗𝟑
expectation 𝐏(𝐗 𝟏 ) = 𝐏(𝐗 𝟏 |𝐒)𝐏(𝐒) + 𝐏(𝐗 𝟏 |𝐒𝐜 )𝐏(𝐒𝐜 )

𝐒𝐜 𝐗𝟏 𝐏(𝐗 𝟐 ) = 𝐏(𝐗 𝟐 |𝐒)𝐏(𝐒) + 𝐏(𝐗 𝟐 |𝐒𝐜 )𝐏(𝐒𝐜 )

𝐄(𝐗|𝐒𝐜 )
𝐗𝟐 𝐏(𝐗 𝟑 ) = 𝐏(𝐗 𝟑 |𝐒)𝐏(𝐒) + 𝐏(𝐗 𝟑 |𝐒𝐜 )𝐏(𝐒𝐜 )
= 𝐏(𝐒)𝐄(𝐗|𝐒) 𝐗𝟑 unconditional
+ 𝐏(𝐒 𝐜 )𝐄(𝐗|𝐒 𝐜 ) probability
= 𝐏(𝐗 𝟏 |𝐒 𝐜 )𝐗 𝟏 + 𝐏(𝐗 𝟐 |𝐒 𝐜 )𝐗 𝟐 + 𝐏(𝐗 𝟑 |𝐒 𝐜 )𝐗 𝟑

Page 4
𝐏(𝟐. 𝟔𝟎|𝐝𝐞𝐜𝐥𝐢𝐧𝐢𝐧𝐠 𝐫𝐚𝐭𝐞𝐬)

declining 2.60 .25 × .60 = .15

.25
rates
∑ = 100%
.60
.75 2.45 .75 × .60 = .45
𝐄(𝐄𝐏𝐒)
.60 2.20 .60 × .40 = .24
.40
∑ = 100%
stable rates
.40
2.00 .40 × .40 = .16
∑ = 100%
∑ = 100%
𝐏(𝟐. 𝟎𝟎|𝐬𝐭𝐚𝐛𝐥𝐞 𝐫𝐚𝐭𝐞𝐬)
𝐄(𝐄𝐏𝐒) = 𝐏(𝐗 𝟏 |𝐒)𝐏(𝐒)𝐗 𝟏 + 𝐏(𝐗 𝟐 |𝐒)𝐏(𝐒)𝐗 𝟐 + 𝐏(𝐗 𝟑 |𝐒 𝐜 )𝐏(𝐒 𝐜 )𝐗 𝟑 + 𝐏(𝐗 𝟒 |𝐒 𝐜 )𝐏(𝐒 𝐜 )𝐗 𝟒
(.25)(.60)2.60 + (.75)(.60)2.45 + (.60)(.40)2.20 + (.40)(.40)(2.00)

25
 Last Revised: 05/12/2023

Page 5

Bayes’ Formula: a method for updating prior probabilities

based on new information
Recall: Total Probability Rule
𝐏(𝐄) = 𝐏(𝐄|𝐒𝟏 )𝐏(𝐒𝟏 ) + 𝐏(𝐄|𝐒𝟐 )𝐏(𝐒𝟐 ) + … + 𝐏(𝐄|𝐒𝐧 )𝐏(𝐒𝐧 )
Q: given that we observe 𝐄, what is 𝐏(𝐒𝐧 )? ➞ 𝐏(𝐒𝐧 |𝐄)
𝐏(𝐒𝐧 |𝐄) = 𝐏(𝐄|𝐒𝐧 )𝐏(𝐒) 𝐏(𝐄|𝐒𝐧 )
= 𝐏(𝐒)
𝐏(𝐄) 𝐏(𝐄)
prior Exp.
.75 𝐏(𝐄) =
e.g./
.25 .45(.75) (.3375) What is 𝐏(𝐞𝐱𝐜𝐞𝐞𝐝𝐬|𝐞𝐱𝐩𝐚𝐧𝐝𝐬)
𝐄 = .45
+ (. 𝟕𝟓)(. 𝟒𝟓) . 𝟕𝟓
.20 = = . 𝟒𝟓 = . 𝟖𝟐𝟑𝟏
m = .30 .30(.20) (.06)
.80 . 𝟒𝟏 . 𝟒𝟏
+
.05
𝐅𝐒 = .25 .25(.05) (.0125) = 𝐏(𝐄𝐱𝐩. |𝐄𝐱𝐜. )𝐏(𝐄𝐱𝐜. ) = 𝐏(𝐄𝐱𝐩. |𝐄𝐱𝐜. ) 𝐏(𝐄𝐱𝐜. )
.95 𝐏(𝐄𝐱𝐩. ) 𝐏(𝐄𝐱𝐩. )
41%
no Exp.

26
 Last Revised: 05/12/2023

Portfolio Mathematics

a. calculate and interpret the expected value, variances, standard deviation,

covariances, and correlations of portfolio returns

b. calculate and interpret the covariance and correlation of portfolio returns using
the joint probability function for returns

c. define shortfall risk, calculate the safety-first ratio, and identify an optimal
portfolio using Roy’s safety-first criterion

27
 Last Revised: 05/12/2023

Portfolio Mathematics
Page 1

calculate and interpret the expected value,

variance, standard deviation, covariances,
and correlations of portfolio returns

⟹ Portfolio Returns ( 𝐄(𝐑 𝐏 ) , 𝛔𝐑 𝐏 , 𝐂𝐨𝐯𝐢𝐣 , 𝛒𝐢𝐣 )

1/ 𝐄(𝐑 𝐏 ) = 𝐄(𝐖𝟏 𝐑 𝟏 + 𝐖𝟐 𝐑 𝟐 + ⋯ + 𝐖𝐧 𝐑 𝐧 )

also a random variable possible value of 𝐑 𝟏

𝐄(𝐑 𝟏 ) = 𝐏(𝐑 𝟏𝟏 )𝐑 𝟏𝟏 + 𝐏(𝐑 𝟏𝟐 )𝐑 𝟏𝟐 + ⋯ + 𝐏(𝐑 𝟏𝐧 )𝐑 𝟏𝐧
probability

Page 2

e.g./ W 𝐄(𝐑 𝐢 )
SnP500 .50 13%
Corp. bonds .25 6%
MSCI EAFE .25 15% 𝐄(𝐑 𝐏 ) = .5(13%) + .25(6%) + .25(15%) = 11.75%

measure of expected reward

𝐧 𝐧
2/ 𝛔𝟐 (𝐑 𝐏 ) = ^ ^ 𝐖 𝐖 𝐂𝐨𝐯7𝐑 𝐑 8 𝐧
𝐢 𝐣 𝐢 𝐣
𝐧
𝐄(𝐑 𝐢 ) = ∑ 𝐏(𝐑 𝐢 )𝐑 𝐢
𝐢7𝟏 𝐣7𝟏
𝐂𝐨𝐯7𝐑 𝐢 𝐑 𝐣 8 = H (𝐑 𝐢𝐭 − 𝐑
( 𝐢 )7𝐑 𝐣𝐭 − 𝐑( 𝐉8
to calculate 𝐭(𝟏
portfolio variance, need: 𝐧−𝟏
1/ all 𝐄(𝐑 𝐢 ) assume 3 assets ➞ 𝐑 𝟏 , 𝐑 𝟐 , 𝐑 𝟑
2/ all 𝐂𝐨𝐯7𝐑 𝐢 , 𝐑 𝐣 8
𝛔𝟐 (𝐑 𝐏 ) = 𝐖𝟏𝟐 𝛔𝟐 (𝐑 𝟏 ) + 𝐖𝟐𝟐 𝛔𝟐 (𝐑 𝟐 ) + 𝐖𝟑𝟐 𝛔𝟐 (𝐑 𝟑 )
(Exhibit #11) + 𝟐𝐖𝟏 𝐖𝟐 𝐂𝐨𝐯(𝐑 𝟏 𝐑 𝟐 ) + 𝟐𝐖𝟏 𝐖𝟑 𝐂𝐨𝐯(𝐑 𝟏 𝐑 𝟑 )
+ 𝟐𝐖𝟐 𝐖𝟑 𝐂𝐨𝐯(𝐑 𝟐 𝐑 𝟑 )

28
 Last Revised: 05/12/2023

Page 3
𝛔𝟐 (𝐑 𝐏 ) = f(variances, covariances)
can be
always > 0
< 0 or > 0
- major point - by selecting assets with zero or negative covariance,
portfolio risk is lowered
n= 5
- for 𝐧 securities (or asset classes) ➞ 𝐧 variances 5 vars.
𝐧𝟐 − 𝐧 covariances 25 - 5 = 20 Covars.
(𝐧𝟐 − 𝐧)/𝟐 distinct covariances 𝟐𝟎1𝟐 = 10
3/ Correlation ➞ 𝛒𝐢𝐣 = 𝐂𝐨𝐯7𝐑 𝐢 𝐑 𝐣 8 unique
➞ 𝐂𝐨𝐯7𝐑 𝐢 𝐑 𝐣 8 = 𝛒𝐢𝐣 𝛔𝐑 𝐢 𝛔𝐑 𝐣 Covars.
𝛔𝐑 𝐢 𝛔𝐑 𝐣
𝟏𝟖𝟗
𝟏 𝟏&
𝟒𝟎𝟎 &𝟐 𝟒𝟒𝟏 𝟐
Ex #13
𝟏𝟖𝟗
= = 𝟎. 𝟒𝟓
𝟐𝟎. 𝟐𝟏

Page 4
Recall: calculate and interpret the covariance
4 𝐀 97𝐑 𝐁 − 𝐑
4 𝐁9 and correlation of portfolio returns using
𝐂𝐨𝐯(𝐑 𝐀 𝐑 𝐁 ) = ∑7𝐑 𝐀𝐢 − 𝐑 𝐢
a joint probability function for returns
𝐧−𝟏
𝟏 𝟏 𝟏
= D 𝐀 EB𝐑 𝐁 − 𝐑
B𝐑 − 𝐑 D 𝐁E + D 𝐀 EB𝐑 𝐁 − 𝐑
B𝐑 − 𝐑 D 𝐁E + ⋯ + D 𝐀 EB𝐑 𝐁 − 𝐑
B𝐑 − 𝐑 D 𝐁E
𝐧 − 𝟏 𝐀𝟏 𝟏
𝐧 − 𝟏 𝐀𝟐 𝟐
𝐧 − 𝟏 𝐀𝐧 𝐧

weights
probabilities?
- the concept of joint probability
𝐧 𝐧
𝐂𝐨𝐯(𝐑 𝐀 𝐑 𝐁 ) = K K 𝐏 C𝐑 𝐑 D 7𝐑 − 𝐑
4 𝐀 9 C𝐑 𝐁 − 𝐑
4 𝐁D where 𝐢 & 𝐣 = 1 to 𝐧
𝐀 𝐢 𝐁𝐣 𝐀𝐢 𝐣
𝐢C𝟏 𝐣C𝟏 are scenarios
probability value of cross product

if returns are independent:

➞ since independence is a stronger property than
➞ 𝐏(𝐑 𝐀 𝐑 𝐁 ) = 𝐏(𝐑 𝐀 )𝐏(𝐑 𝐁 ) uncorrelatedness, this property holds for
uncorrelated random variables Exhibit #6/7

29
 Last Revised: 05/12/2023

Page 5
Safety first rules focus on shortfall risk - the risk a
portfolio value (or return) will fall below some
minimum acceptable level over some time horizon

Let 𝐑 𝐋 = minimum acceptable level of return

objective is to maximize this ratio
𝐒𝐅𝐑𝐚𝐭𝐢𝐨 = 𝐄(𝐑 𝐏 ) − 𝐑 𝐋
z-value 𝛔𝐏 - optimal portfolio minimizes
N(-SFRatio)

e.g./ 𝐑 𝐋 = 2% 𝟏𝟐 − 𝟐
𝐒𝐅𝐑𝐚𝐭𝐢𝐨𝟏 = = 𝟎. 𝟔𝟔̇ = NORM.S.DIST(-.667,1)
Portfolio 𝐑 𝐏 𝛔 𝟏𝟓 = 0.2525
𝟏𝟒 − 𝟐
1 12% 15% 𝐒𝐅𝐑𝐚𝐭𝐢𝐨𝟐 = = 𝟎. 𝟕𝟓 = NORM.S.DIST(-.75,1)
𝟏𝟔
2 14% 16% = 0.227

Note: if 𝐑 𝐋 = 𝐑 𝐟 ➞ SFRatio = Sharpe Ratio

Example 3

30
 Last Revised: 05/12/2023

Simulation Methods

a. explain the relationship between normal and lognormal distributions and why
the lognormal distribution is used to model asset prices when using
continuously compounded asset returns

b. describe Monte Carlo simulation and explain how it can be used in investment
applications

c. describe the use of bootstrap resampling in conducting a simulation based on

observed data in investment applications

31
 Last Revised: 05/12/2023

Common Probability Distributions

Page 1
𝟐U
𝝁 = 𝐞R𝝁 0 .𝟓𝛔
- commonly used to model
the probability distribution
of asset prices
right
skewed
- a variable Y follows a lognormal
0 ∞ distribution if LN(Y) is normally
distributed
bounded below by 0

- completely described by 2 parameters ➞ the 𝝁 and 𝛔𝟐 of its associated

normal distribution

𝐒𝐓
;𝐒 = 1 + 𝐑 𝐇 where 𝐑 𝐇 = holding period return
𝟎

i.e./ 𝐒𝐓 − 𝐒𝟎
= 𝐑𝐇
𝐒𝟎
𝐒𝐓 𝐒 𝐒𝐓 𝐒
8𝐒 − 𝟎8𝐒 = 𝐑 𝐇 ➞ 8𝐒 − 𝟏 = 𝐑 𝐇 ➞ 𝐓8𝐒 = 1 + 𝐑 𝐇
𝟎 𝟎 𝟎 𝟎

Page 2

e.g. 𝐒𝐓 = 34.50 𝐒𝟎 = 30
𝐒𝐓
;𝐒 = 𝟑𝟒. 𝟓𝟎;𝟑𝟎 = 𝟏. 𝟏𝟓
𝟎
= 1 + 𝐑𝐇 ∴ 𝐑 𝐇 = 15%

𝐒
𝐥𝐧 : 𝐓;𝐒 < = 𝐫 where 𝐫 = continuously compounded return
𝟎

𝐫 = 𝐥𝐧:𝟑𝟒. 𝟓𝟎;𝟑𝟎< = 𝐥𝐧(𝟏. 𝟏𝟓) = 𝟎. 𝟏𝟑𝟗𝟕𝟔 and 𝐫 ~ 𝐍(𝛍𝐓, 𝛔𝟐 𝐓)

∴ 𝟑𝟒. 𝟓𝟎 = 𝟑𝟎𝐞.𝟏𝟑𝟗𝟕𝟔
𝐒 𝐒
more generally: 𝐒𝐓 = 𝐒𝟎 𝐞𝐫 : 𝐓;𝐒 = 𝐞𝐫 𝐚𝐧𝐝 𝐥𝐧 : 𝐓;𝐒 < = 𝐫<
𝟎 𝟎

- to assume returns are normally distributed, we assume returns

are 1/ independent
2/ identically distributed (𝛍 and 𝛔𝟐 do not change from
period to period)
- so, while 𝐒𝐓 = 𝐒𝟎 (𝟏 + 𝐑 𝐇 )𝐓
with cont. comp: 𝐒𝐓 = 𝐒𝟎 𝐞𝐫𝐓

32
 Last Revised: 05/12/2023

Page 3
Volatility ➞ annualized sd of the continuously compounded
daily returns of the underlying asset

- since 𝐫 ~ 𝐍(𝛍𝐓, 𝛔𝟐 𝐓) , sd = 𝛔√𝐓

- so both the mean and variance of 𝐫 scale linearly

with time, but the s.d. scales linearly with the square
root of time

e.g. if daily vol. = .01, annualized vol. = . 𝟎𝟏√𝟐𝟓𝟎 = 𝟏𝟓. 𝟖𝟏% example #1

Page 4
Monte Carlo Simulation/
Step 1: Specify the quantity of interest
e.g. 𝐌𝐕𝐩 in 10 years

Step 2: Specify a time grid ➞ 𝐊 sub-periods with ∆𝐭 increment

for the full time horizon
e.g. 20 sub-periods , ∆𝐭 = 6 months

Step 3: Specify distributional assumptions for the key risk factors

e.g. 𝐄(𝐑 𝐏 ) = 6.4% 𝛔 = 12%
𝐌𝐕𝐭 = 𝐌𝐕𝐭A𝟏 7𝟏 + 𝐑 𝐏𝟏 8

Step 4: Draw standard normal random numbers for each key risk
factor over each 𝐊 sub-periods.
- random number generator ➞ produces a distribution of
random numbers from 0 to 1, all equally likely

33
 Last Revised: 05/12/2023

Page 5
Monte Carlo Simulation/
Step #4 distribution of random #’s
#1 ➞ 0.32 1000 runs
#2 ➞ 0.64

normal cdf 0 1 #20

100% Step #5:
50%
#1 ➞ 0.31561 on cdf
#2 ➞ 0.7673 on cdf
50%

#20
0 Step #6:
#1 ➞ -0.48 z-value
𝐄(𝐑 𝐏 ) = 𝟔. 𝟒% − 𝟎. 𝟒𝟖(𝟏𝟐%) = 𝟎. 𝟔𝟒
𝐌𝐕𝟏 = 𝐌𝐕𝟎 (𝟏. 𝟎𝟎𝟑𝟐)
#2 ➞ 0.73 z-value
𝐄(𝐑 𝐏 ) = 𝟔. 𝟒% + . 𝟕𝟑(𝟏𝟐) = 𝟏𝟓. 𝟏𝟔%
𝐌𝐕𝟐 = 𝐌𝐕𝟏 (𝟏. 𝟎𝟕𝟓𝟖)
#20

Page 6
Monte Carlo Simulation/

distribution of possible 𝐌𝐕𝐏𝟏𝟎

𝐌𝐕𝐏𝟏𝟎 = 𝐌𝐕𝐏𝟎 (𝟏 + 𝐄(𝐑 𝐏 )𝟏𝟎

Objective: 𝐌𝐕𝐏𝟏𝟎 = $2M with 95% PoS.

𝟐𝐌
𝐌𝐕𝐏𝟎 d 𝟏𝟎 = Beginning Capital
7𝟏 + 𝐑 𝐏𝟗𝟓 9
𝐑 𝐏𝟗𝟓

10 yrs.

Monte Carlo Simulation ➞ provides only statistical estimates, not

exact results
➞ does not support cause and
effect conclusions

34
 Last Revised: 05/12/2023

Page 7
Resampling ➞ repeatedly draw samples from an
original data sample in order to estimate
population parameters

1/ Bootstrap method/ uses computer simulation

𝐁𝟏
𝐁𝟏
Population all with - draw 1 observation,
sample 𝐁𝟐
- unknown 𝐧 = 30 record, replace
𝐧 = 30
distribution 𝐁𝟑 - draw another obs.,
- all we have
record, replace
is a sample 𝐁𝟏𝟎𝟎𝟎
𝐧 times
- rather than estimate the distribution, ( 8𝟐
o𝐢 − 𝛉
m∑𝐁𝐢.𝟏7𝛉
this method creates the distribution 𝐒𝐗N =
𝐁−𝟏
- can also find SE of an estimator even when
no analytical formula is available standard
(e.g. median) error

35
 Last Revised: 05/12/2023

Estimation and Inference

a. compare and contrast simple random, stratified random, cluster, convenience,

and judgmental sampling and their implications for sampling error in an
investment problem

b. explain the central limit theorem and its importance for the distribution and
standard error of the sample mean

c. describe the use of resampling (bootstrap, jackknife) to estimate the sampling

distribution of a statistic

36
 Last Revised: 05/12/2023

Sampling and Estimation

Page 1

sample ➞ a method of obtaining information

about a population’s parameters (𝛍 & 𝛔)
population
3 & 𝐒)
through sample statistics (𝐗

A/ Probability sampling ➞ every member of a population has an equal

chance of being selected
∴ samples will be more representative of the population
1/ Simple Random Sampling ➞ a subset of a larger population such that
each element has an equal probability of being selected
e.g.
population random number
sample
𝐧 = 500 generator selects
size
50 numbers between
𝐧 = 50
1 and 500
useful when data are homogeneous

Page 2

2/ Systematic Sampling ➞ when the population is too

large to code
- select every 𝐊 𝐭𝐡 element until the desired
sample size is reached
3/ Stratified Random Sampling - population is sub-divided into sub-populations
based on one or more classifications
- simple random samples are then drawn from each sub-pop.
- each sample is then pooled to form the main sample

- each sub-sample is proportionate to the

sample
size of its sub-population
- guarantees that population sub-divisions
are represented in the sample
- statistics will be more precise

37
 Last Revised: 05/12/2023

Page 3
- sample statistics are estimates of population parameters
- not exact, subject to error
sampling error ➞ difference between observed values of
a statistic and population parameters as a result
of using just a subset of the population
&𝟏
𝐗
&𝟐
sampling distribution of
𝐗
&𝟑
𝐗 the sample means
&𝐧
𝐗 Q𝟑 𝐗𝐧 𝐗
𝐗 Q𝟏 Q𝟐
𝐗

all equal 𝐧 𝐬𝐝𝐗N = 𝐒; example #1

√𝐧
4/ Cluster sampling - pop. is divided into clusters each of which is
a mini representation of the population
- certain clusters are then selected as a whole using
simple random sampling ➞ one-stage cluster sampling

Page 4

4/ Cluster sampling - if sub-samples are selected from

each cluster ➞ two-stage cluster sampling
usually results in lowest precision since a cluster
may not be representative of the population
- is both cost and time efficient however

38
 Last Revised: 05/12/2023

Page 5
B/ Non-probability sampling - depends on factors such
as judgment or convenience (in terms of
access to data)
- risk that samples may be non-representative

5/ Convenience sampling - observations are selected that are easy to

obtain or are accessible
not necessarily representative, but low cost

6/ Judgmental sampling ➞ select observations based on experience

and knowledge
➞ useful when there is a time constraint and/or
the specialty of the researcher would result
in better representation

example 2, 3, 4

Page 6

population with
𝐒𝟏 𝐒𝟐 𝐒𝟑 𝐒𝐧
any distribution
with 𝝁 and finite 𝛔𝟐
sample
size 𝐧 = 30
4𝟏
𝐗 4𝟐
𝐗 4𝟑
𝐗 4𝐧
𝐗
sample size 𝐒𝟏 𝐒𝟐 𝐒𝟑 𝐒𝐧 Standard Error
𝐧 = 100 𝐒8 or 𝛔8 if we
sample size √𝐧 √𝐧
know 𝛔
𝐧 = 200 ➞ the sampling
Note: sd ≠ SE
- as 𝐧 ↑, sampling distribution of
error decreases the sampling sd = dispersion from
means the mean
(data description)

5 𝐗V ➞ best estimate of 𝛍
𝐗 SE = sampling error
𝟐 (data inference)
𝐒𝐗𝟐V = 𝐒 8𝐧
𝐒𝐄 𝐒𝐗V = 𝐒8
√𝐧

39
 Last Revised: 05/12/2023

Page 7
Resampling ➞ repeatedly draw samples from an
original data sample in order to estimate
population parameters

1/ Bootstrap method/ uses computer simulation

𝐁𝟏
𝐁𝟏
Population all with - draw 1 observation,
sample 𝐁𝟐
- unknown 𝐧 = 30 record, replace
𝐧 = 30
distribution 𝐁𝟑 - draw another obs.,
- all we have
record, replace
is a sample 𝐁𝟏𝟎𝟎𝟎
𝐧 times
- rather than estimate the distribution, ( 8𝟐
o𝐢 − 𝛉
m∑𝐁𝐢.𝟏7𝛉
this method creates the distribution 𝐒𝐗N =
𝐁−𝟏
- can also find SE of an estimator even when
no analytical formula is available standard
(e.g. median) error

Page 8
2/ Jackknife method/ - omit one observation from
a sample, one at a time

e.g./ 𝐧 = 30 𝐉𝟏 𝐧 = 29, omit 𝐗 𝟏 will produce similar

𝐉𝟐 𝐧 = 29, omit 𝐗 𝟐 results from sample to
sample (bootstrap may
𝐉𝟑𝟎 𝐧 = 29, omit 𝐗 𝟑𝟎 not)

40
 Last Revised: 05/12/2023

Hypothesis Testing

a. explain hypothesis testing and its components, including statistical significance,

Type I and Type II errors, and the power of a test.

b. construct hypothesis tests and determine their statistical significance, the

associated Type I and Type II errors, and power of the test given a significance
level

c. compare and contrast parametric and nonparametric tests, and describe

situations where each is the more appropriate type of test

41
 Last Revised: 05/12/2023

Hypothesis Testing
Page 1
Statistical Inference ➞ the process of making judgments
about a larger group (pop.) based on a
smaller group (sample)
e.g./ hypothesis testing - test to see whether a sample
statistic is likely to come from a population with
the hypothesized value of the population parameter
i.e. Does 3 = 𝛍𝟎 ?
𝐗
Hypothesis ➞ a statement about one or more populations that are
tested using sample statistics
Process: Step 1: State the hypothesis
2: Identify the appropriate test statistic
3: Specify the level of significance
4: State the decision rule
5: Collect data and calculate the test statistic
6: Make a decision

Page 2
Step #1: State the hypothesis
null ➞ 𝐇𝟎 ➞ assumed to be true unless
alternative ➞ 𝐇𝐚 we can reject
- typically want to reject 𝐇𝟎

Two-sided (two-tailed) test

e.g. 𝐇𝟎 : 𝛍 = 𝟔% could be fail to
< 𝟔% could
vs. 𝐇𝐚 : 𝛍 ≠ 𝟔% reject 𝐇𝟎
be > 𝟔%
reject 𝐇𝟎 𝛍 = 6%
reject 𝐇𝟎

One-sided (left or right tailed) test

e.g. 𝐇𝟎 : 𝛍 ≤ 𝟔% 𝐇𝟎 : 𝛍 ≥ 𝟔% left-tailed
or/
𝐇𝐚 : 𝛍 > 𝟔% 𝐇𝐚 : 𝛍 < 𝟔% do not
< 𝟔% reject 𝐇𝟎
right-tailed left-tailed
Reject 𝛍 = 6%
𝐇𝟎

42
 Last Revised: 05/12/2023

Page 3
- the null (𝐇𝟎 ) always contains the equality sign
( = 𝛍𝟎
𝐇𝟎 : 𝐗 ( ≤ 𝛍𝟎
𝐇𝟎 : 𝐗 ( ≥ 𝛍𝟎
𝐇𝟎 : 𝐗
- testing 𝐇𝟎 is always done at equality

Test Statistic:
(Step #2)
pop. 𝛔𝟐 is known
4 − 𝛍𝟎
𝐗
𝐙= 𝛔
8 𝐧
distributed √
normally
pop. 𝛔𝟐 is unknown
4 − 𝛍𝟎
𝐗
𝐭=
𝐒8
t-distributed
√𝐧

Page 4
Step 3: Specify the Level of Significance
- level of sig. depends on the seriousness of making
a mistake
𝐇𝟎 = true 𝐇𝟎 = false
fail to reject Correct Type II error
(𝟏 − ∝) 𝜷 as ∝ ↓ , 𝛃 ↑
confidence level
reject Type I error Correct only way to
∝ (𝟏 − 𝜷) decrease both is
level of sig. Power of a test to increase 𝐧
( − 𝛍𝟎
𝐗
𝐭=
𝐒/√𝐧 ➞ as 𝐧 ↑,
e.g.: 𝐇𝟎 : not pregnant denom. ↓,
𝐇𝐚 : pregnant t-stat ↑

43
 Last Revised: 05/12/2023

Page 5
Step #4: State the Decision Rule

2-tail: Reject 𝐇𝟎 when |𝐭𝐞𝐬𝐭 − 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜| > |𝐜𝐫𝐢𝐭𝐢𝐜𝐚𝐥 𝐯𝐚𝐥𝐮𝐞|

𝐭 or 𝐳 𝐭 ∝) or 𝐳∝)
𝟐 𝟐
right tail: Reject 𝐇𝟎 when test-statistic > critical value
(𝐭 ∝ or 𝐳∝ )
left tail : Reject 𝐇𝟎 when test-statistic < critical value
∝ = 5% ∝ = 5%
2.5% fail to 2.5% fail to
5% reject 𝐇
reject 𝐇𝟎 reject reject 𝐇𝟎 𝟎
reject
𝐇𝟎
𝐇𝟎 - 𝐳 or 𝐭 + 𝐳 or 𝐭 + 𝐳 or 𝐭

or T.INV (p,df)

Page 6
Parametric Testing Non-parametric testing

sample stats distributional no no

to test pop. assumptions parameters distributional
parameters tested assumptions
& or 𝐒𝟐 for 𝛍 or 𝛔𝟐 )
(𝐗
1/ when data do not meet distributional
assumptions
i.e. 𝐧 < 30, pop. is non-normal
2/ when there are outliers
- test of median instead of mean
3/ when data are given in ranks or use
an ordinal scale ordered
NO IR
categorical
4/ hypothesis do not concern a parameter
e.g. Is a sample random?

44
 Last Revised: 05/12/2023

Parametric and Non-Parametric Tests of Independence

a. explain parametric and nonparametric tests of the hypothesis that the

population correlation coefficient equals zero, and determine whether the
hypothesis is rejected at a given level of significance

b. explain tests of independence based on contingency table data

45
 Last Revised: 05/12/2023

Hypothesis Testing
Page 1
Tests of Correlation
1/ Parametric test left right
2-sided 𝐇𝟎 : 𝐩 = 𝟎 one-sided 𝐇𝟎 : 𝐩 ≥ 𝟎 𝐇𝟎 : 𝐩 ≤ 𝟎
𝐇𝐚 : 𝐩 ≠ 𝟎 𝐇𝐚 : 𝐩 < 𝟎 𝐇𝐚 : 𝐩 > 𝟎

- recall 𝐫𝐱𝐲 = 𝐂𝐨𝐯(𝐱, 𝐲)

➞ called a Pearson correlation
𝐒𝐱 𝐒𝐲
or a Bivariate correlation
test-statistic: 𝐭 𝐧D𝟐 = 𝐫√𝐧 − 𝟐
√𝟏 − 𝐫 𝟐
- in testing 𝐫, as 𝐧 ↑ , 𝐇𝟎 rejected for even small correlations
- big data sets, almost any 𝐫 will be significant
e.g./ 𝐫 = .02 . 𝟎𝟐√𝟗, 𝟗𝟗𝟖
𝐭 𝟗,𝟗𝟗𝟖 = ~𝟐 vs. critical 𝐭 = 1.96
𝐧 = 10,000 √𝟏 −. 𝟎𝟐𝟐
ex. #1

Page 2
Tests of Correlation
2/ Non-parametric test
- if normality assumption for 𝐗 or 𝐘 violated, or
outliers are present
Spearman rank correlation coefficient
basically a correlation, but calculated on rank values
and not the values of the observations of 𝐗 or 𝐘

1/ Rank all 𝐗 from largest to smallest

- assign a rank: 1 = largest … 𝐧 = smallest
- for a tie: assign all the same ➞ average rank
e.g. 3 tied for 6th (6 + 7 + 8)/3 = 7
- each get a rank of 7
- repeat for 𝐘

46
 Last Revised: 05/12/2023

Page 3
Tests of Correlation/
2/ Non-parametric test
2/ On original data set (pre-ranked):
calculate 𝐝𝟐𝐢 = (𝐫𝐚𝐧𝐤 𝐗 𝐢 − 𝐫𝐚𝐧𝐤 𝐘𝐢 )𝟐 - where 𝐗 and 𝐘
are original
𝐧 𝟐
3/ 𝐫𝐒 = 𝟏 − ?𝟔7∑𝐢.𝟏 𝐝𝐢 8@ paired obs.
𝐧(𝐧𝟐 − 𝟏)
𝐫𝐒 √𝐧 − 𝟐 white text example
𝐭 𝐧A𝟐 =
Example #3
test 𝐫𝐒 , if 𝐧 > 𝟑𝟎 m𝟏 − 𝐫𝐒𝟐

Tests of Independence/
- test if classification types are independent
e.g./
Are growth stocks equally likely to be any size or
are they more likely to be large-cap stocks?

Page 4
Tests of Independence/ Contingency Table (2-way)
observed
non-parametric test of indep.
𝐦 𝟐
𝟐
Q𝐎𝐢𝐣 − 𝐄𝐢𝐣 S df = (r -1)(c -1)
𝛘 =^
𝐄𝐢𝐣 (right-tailed)
𝐢7𝟏

m = # of cells (3 x 3 = 9)
𝐎𝐢𝐣 = observed value in each cell
𝐄𝐢𝐣 = expected value in each cell
𝐄𝐒𝐕
𝐄𝐌𝐆 𝐄𝐢𝐣 = (𝐫𝐨𝐰𝐢 𝐭𝐨𝐭𝐚𝐥) × Q𝐜𝐨𝐥𝐮𝐦𝐧𝐣 𝐭𝐨𝐭𝐚𝐥S
𝐎𝐯𝐞𝐫𝐚𝐥𝐥 𝐭𝐨𝐭𝐚𝐥

𝐄𝐒𝐕 = (𝟓𝟎𝟑 × 𝟏𝟒𝟖)/𝟏𝟓𝟗𝟒 = 𝟒𝟔. 𝟕𝟎𝟑

𝐄𝐌𝐆 = (𝟑𝟔𝟔 × 𝟑𝟖𝟏)/𝟏𝟓𝟗𝟒 = 𝟖𝟕. 𝟒𝟖𝟐

∴ E𝐎
𝟐
𝐢𝐣 − 𝐄𝐢𝐣 F (𝟓𝟎 − 𝟒𝟔. 𝟕𝟎𝟑)𝟐
𝐒𝐕 = = . 𝟐𝟑𝟐𝟖
𝐄𝐢𝐣 𝟒𝟔. 𝟕𝟎𝟑

47
 Last Revised: 05/12/2023

Page 5
Tests of Independence/
𝐇𝟎 : size and type are independent
𝐇𝐚 : size and type are not independent
𝐦
𝟐
H 7𝐎𝐢𝐣 − 𝐄𝐢𝐣 9
= 𝟑𝟐. 𝟎𝟖𝟎𝟐𝟓
𝐢(𝟏 𝐄𝐢𝐣
= CHISQ.INV(0.95,4) = 9.4877
∴ reject 𝐇𝟎
standardized residuals = 𝐎𝐢𝐣 − 𝐄𝐢𝐣 ➞ > 𝟎 means more obs. than
B𝐄𝐢𝐣 expected if categories were
e.g./ independent
𝟓𝟎 − 𝟒𝟔. 𝟕𝟎𝟑
𝐒𝐑 𝐒𝐕 = = 𝟎. 𝟒𝟖 < 𝟎 - opposite (i.e. fewer obs.)
√𝟒𝟔. 𝟕𝟎𝟑
more than
𝟏𝟐𝟐 − 𝟖𝟕. 𝟒𝟖𝟐 expected if
𝐒𝐑 𝐌𝐆 = = 𝟑. 𝟔𝟗 example #4
√𝟖𝟕. 𝟒𝟖𝟐 independent

48
 Last Revised: 05/12/2023

Simple Linear Regression

a. describe a simple linear regression model, how the least squares criterion is
used to estimate regression coefficients, and the interpretation of these
coefficients

b. explain the assumptions underlying the simple linear regression model, and
describe how residuals and residual plots indicate if these assumptions may
have been violated

c. calculate and interpret measures of fit and formulate and evaluate tests of fit
and of regression coefficients in a simple linear regression

d. describe the use of analysis of variance (ANOVA) in regression analysis,

interpret ANOVA results, and calculate and interpret the standard error of
estimate in a simple linear regression

e. calculate and interpret the predicted value for the dependent variable, and a
prediction interval for it, given an estimated linear regression model and a
value for the independent variable

f. describe different functional forms of simple linear regressions

49
 Last Revised: 05/12/2023

Simple Linear Regression

LOS a (2.5p) Simple Linear Regression - describe

LOS b (8.5p) Estimating Parameters - describe

LOS c (7p) Assumptions of SLR - explain, describe

LOS d, e (5.5p) Analysis of Variance - calculate, interpret, describe

LOS f (8p) Hypothesis Testing ➞ Coefficients - formulate, determine

LOS g (4p) Prediction & Prediction Intervals - calculate, interpret

LOS h (7p) Functional Forms of LR - describe

Page 1
- Simple Linear Regression (LR) ➞ one IV LOS a
- describe
DV - dependent variable -𝐘 - the variable we
IV - independent variable - 𝐗 are seeking to explain
the explanatory variable
LR assumes a linear relationship between the DV and the IV
𝐧

Variation of 𝐘 = K(𝐘𝐢 − 𝐘
4)𝟐 ➞ SST or total sum of squares
𝐢C𝟏
3, thus if 𝐗 gives a more accurate
- best guess for 𝐘 is 𝐘
estimate of 𝐘 than 𝐘3 , we say 𝐗 helps explain 𝐘
LOS b
𝐘𝐢 = 𝐛𝟎 + 𝐛𝟏 𝐗 𝐢 + 𝛆𝐢 𝛆 ➞ error term (residual) - describe

intercept slope coefficient - the portion of the DV that

cannot be explained by the IV
regression coefficients
(𝐘 is regressed on 𝐗)

50
 Last Revised: 05/12/2023

Page 2
regression compute a line of best fit that LOS b
𝐘
residuals minimizes the sum of the squared - describe

deviations between the observed values of

𝐘 and the predicted values (the regression
𝐧 𝐧 line)
𝐗 𝟐
… 𝟎−𝐛
i.e. min. -7𝐘𝐢 − 𝐛 … 𝟏 𝐗 𝐢 8 = - 𝛆𝟐𝐢 ➞ SSE
(𝐗
(, 𝐘
( ) lies on the
𝐢.𝟏 𝐢.𝟏 - sum of
regression line DV predicted values the squares
…𝟎=𝐘 … 𝟏𝐗
(−𝐛 ( o)
of DV (𝐘
𝐛 error
- Note: 𝛆 = 𝐘𝐢 − 𝐘o implies the (a.k.a. residual
residual is in the same units of sum of squares)

𝐂𝐨𝐯(𝐗, 𝐘)
measurement as the DV (𝐘)
()(𝐘𝐢 − 𝐘
∑𝐧𝐢.𝟏(𝐗 𝐢 − 𝐗 ()
( 𝐄(𝛆) = 𝟎
∑𝛆𝐢 = 𝟎)
…𝟏=
𝐛 = ➞ denominator can never be
𝛔𝟐𝐱 ( )𝟐
∑𝐧𝐢.𝟏(𝐗 𝐢 − 𝐗
… 𝟏 is
negative, ∴ sign of 𝐛
determined solely by 𝐂𝐨𝐯(𝐗, 𝐘)
q 𝟏 > 𝟎, 𝐫𝐗𝐘 > 𝟎
- if 𝐛

Page 3
D 𝟎 and 𝐛
D𝟏 LOS b
Interpreting 𝐛
- describe
o = 𝐛𝟎
𝐘 if 𝐗 𝐢 = 𝟎 ➞ only makes sense if
the IV has meaning at 𝐗 = 𝟎
… 𝟏 ➞ the change in 𝐘 for a one unit change in 𝐗
𝐛

e.g.
ROA(%) = 4.875% + 1.25 CAPEX(%) ➞ ROA = 4.875% if CAPEX = 0
➞ if CAPEX ↑ 1 unit (i.e. 1%), then
ROA ↑ 1.25%

Data/ cross-sectional - many observations on 𝐗 & 𝐘 for the same

time period
time-series - many observations on 𝐘 (and sometimes 𝐗) from
different time periods
example #2/3

51
 Last Revised: 05/12/2023

Page 4
Assumptions/ LOS c
- explain
1/ Linearity ➞ the relationship between 𝐗 & 𝐘 is linear in - describe
the parameters 𝐛𝟎 and 𝐛𝟏 ➞ neither is multiplied
or divided by another regression parameter
➞ implies the IV must not be random - if so, there
would be no linear relation between 𝐗 & 𝐘
2/ Homoskedasticity ➞ 𝐕𝐚𝐫(𝛆) is the same for all observations
(vs. heteroskedastic) - a violation indicates the data series
may come from 2 different populations (CS) or
regimes (TS)

3/ Independence ➞ the pairs (𝐗, 𝐘) are independent of each other

∴ 𝛆 is uncorrelated across observations
(no serial correlation)
- needed to correctly estimate the variances
of 𝐛𝟎 and 𝐛𝟏

Page 5
Assumptions/ LOS c
- explain
4/ Normality ➞ 𝛆 is normally distributed - describe
- required to conduct valid tests of the
example #4 values of the regression coefficients
LOS d
Analysis of Variance/ - calculate
Total sum of squares (SST) - interpret
𝐧
total
4)𝟐
K(𝐘𝐢 − 𝐘 variance
𝐢C𝟏

sum of squared errors (SSE) Regression sum of squares (SSR)

𝐧
𝟐
unexplained 𝐧
𝟐
explained
K7𝐘𝐢 − 𝐘o9 variance K7𝐘 o − 4
𝐘9 variance
𝐢C𝟏 𝐢C𝟏

∴ SST = SSE + SSR

or/ total SS = unexplained SS + explained SS

52
 Last Revised: 05/12/2023

Page 6
Analysis of Variance/ LOS d
- calculate
Coefficient of Determination - measures the
- interpret
fraction of the total variation in the
DV that is explained by the IV (goodness of fit measure)
- if only 1 IV, square the correlation between IV and DV
𝐧

- if multiple IVs: total variation in 𝐘 = ^(𝐘𝐢 − 𝐘

& )𝟐
𝐢7𝟏
𝐧
explained variation in 𝐘 = ^Q𝐘f − 𝐘&S
𝟐

𝐢7𝟏
𝟐
𝐒𝐒𝐑 𝐞𝐱𝐩𝐥𝐚𝐢𝐧𝐞𝐝 𝐯𝐚𝐫. ∑𝐧 7𝐘 o−𝐘 (8 measures fit but is
𝐑𝟐 = = = 𝐧𝐢.𝟏
𝐒𝐒𝐓 𝐭𝐨𝐭𝐚𝐥 𝐯𝐚𝐫. ()𝟐
∑𝐢.𝟏(𝐘𝐢 − 𝐘 not a statistical test
Coeff. of
Determination
multiple IVs
∴ statistical test = F-test 𝐇𝟎 : 𝐛𝟏 = 𝟎 𝐇𝟎 : 𝐛𝟏 = 𝐛𝟐 = ⋯ = 𝐛𝐤 = 𝟎
𝐇𝐚 : 𝐛𝟏 ≠ 𝟎 𝐇𝐚 : 𝐚𝐭 𝐥𝐞𝐚𝐬𝐭 𝐨𝐧𝐞 𝐛𝐤 ≠ 𝟎

Page 7
Analysis of Variance/ LOS d
- calculate
𝐒𝐒𝐑; 𝐒𝐒𝐑;
𝐌𝐒𝐑 𝐝𝐟 𝐤 𝐤 = slope - interpret
𝐅= = =
𝐌𝐒𝐄 𝐒𝐒𝐄;𝐝𝐟 𝐒𝐒𝐄;𝐧 − (𝐤 + 𝟏) coefficients
df1 = 𝐤
mean 𝐤 + 𝟏 = regression coefficients
df2 = 𝐧 − 𝐤 − 𝟏

ANOVA table/ LOS e

- describe
- calculate
- interpret

= √𝐌𝐒𝐄 = 𝐒𝐄𝐄

53
 Last Revised: 05/12/2023

Page 8
Standard Error of the Estimate (SEE) - a measure of LOS e
- describe
the s.d. of 𝛆
'𝐢
𝟏0 - calculate
𝟏0
∑𝐧𝐢.𝟏7𝐘𝐢
−𝐘o 8𝟐 𝟐
∑𝐧𝐢.𝟏 𝛆𝟐𝐢 𝟐 - interpret
𝐒𝐄𝐄 = √𝐌𝐒𝐄 = † ‡ =ˆ ‰
𝐧−𝟐 𝐧−𝟐
the smaller the SEE, the more accurate the regression
(a.k.a. the standard error of the regression or the root mean square
error)
e.g./

= F.INV(.95,1,4)
= 7.71
𝐒𝐄𝐄 = √𝟏𝟏. 𝟗𝟔𝟖𝟕𝟓
Example #5

Page 9
LOS f
…𝟏:
1/ Hypothesis Tests of 𝐛
- formulate
… 𝟏 − 𝐛𝟏 hypothesized value
test statistic: 𝐭 = 𝐛 - determine
𝐒𝐄𝐄
df = 𝐧 − (𝐤 + 𝟏) 𝐒𝐛X𝟏 ➞ standard error of 𝐛
…𝟏 =
&) 𝟐
g∑𝐧𝐢7𝟏(𝐗 𝐢 − 𝐗
𝐧

e.g./ 𝐛𝟏 = 𝟏. 𝟐𝟓 &)𝟐 = 𝟏𝟐𝟐. 𝟔𝟒

^(𝐗 𝐢 − 𝐗
𝐢7𝟏
𝐇𝟎 : 𝐛𝟏 = 𝟎
at ∝ = 𝟓%
𝐇𝐚 : 𝐛𝟏 = ∅
= T.INV(.05,4) = 2.776

𝐧 − (𝐤 + 𝟏)
𝐒𝐛X𝟏 = √𝟏𝟏. 𝟗𝟔𝟕𝟖 = 𝟎. 𝟑𝟏𝟐𝟑𝟗𝟖 𝟏. 𝟐𝟓 − 𝟎
𝐭= = 𝟒. 𝟎𝟎𝟏𝟑𝟏 Reject 𝐇𝟎
√𝟏𝟐𝟐. 𝟔𝟒 . 𝟑𝟏𝟐𝟑𝟗𝟖
(𝐭 = 𝐅, 𝟒. 𝟎𝟎𝟏𝟑𝟏𝟐 = 𝟏𝟔. 𝟎𝟏𝟎𝟒)
𝟐

Note: 𝐇𝟎 : 𝐩 = 𝟎
𝐫√𝐧 − 𝟐 𝟎. 𝟖𝟗𝟒𝟓√𝟒 SLR
𝐇𝐚 : 𝐩 ≠ 𝟎 𝐭= = 𝟒. 𝟎𝟎𝟏𝟑𝟏 Reject 𝐇𝟎
√𝟏 − 𝐫 𝟐 √𝟏 − . 𝟖𝟗𝟒𝟓𝟐 only
df = 𝐧 − 𝟐

54
 Last Revised: 05/12/2023

Page 10
…𝟎:
2/ Hypothesis Tests of 𝐛 LOS f
… 𝟎 − 𝐁𝟎 - formulate
𝐭= 𝐛 5𝟐
and 𝐒𝐛X𝟎 =e𝟏 + 𝐗 - determine
𝐒𝐛X𝟎 𝐧 5)𝟐
∑𝐧𝐢C𝟏(𝐗 𝐢 − 𝐗
df = 𝐧 − (𝐤 + 𝟏)
… 𝟏 if IV is an indicator variable (dummy var.)
3/ Hypothesis Tests of 𝐛

f𝟏 takes on a value of 0 or 1
- same process as a test of 𝐛
𝐘 = 𝐛𝟎 + 𝐛𝟏 𝐈𝐍𝐃 - if 𝐈𝐍𝐃 = 𝟎 , 𝐘 = 𝐛𝟎 𝐛𝟎 = avg. of all 0 obs.
- if 𝐈𝐍𝐃 = 𝟏 , 𝐘 = 𝐛𝟎 + 𝐛𝟏 𝐛𝟎 + 𝐛𝟏 = avg. of all 1 obs.

𝐒𝟏 all 0 obs. ( 𝟎 = 𝐛𝟎
𝐗 difference in means
∴(𝐗(𝟏 − 𝐗(𝟎 ) = (𝐛𝟎 + 𝐛𝟏) − 𝐛𝟎
𝐒𝟐 all 1 obs. ( 𝟏 = 𝐛𝟎 + 𝐛𝟏
𝐗 = 𝐛𝟎 − 𝐛𝟎 + 𝐛𝟏
= 𝐛𝟏

∴ test-stat. for 𝐛𝟏 = test-stat. of differences in means

Page 11
LOS f
Level of Significance and p-values/
- formulate
- most software output ➞ ∝ = 𝟓% , 𝐇𝟎 : parameter = 0 - determine
recall: = (1 - T.DIST(+ 𝐭 , df, 1)) × 2 example #6

J: LOS g
Prediction interval (or CI) for 𝐘
- calculate
o=𝐛
𝐘 … 𝟎+𝐛… 𝟏𝐗 - interpret
estimated with error 2 sources of
o8 = 𝛆 ➞ estimated with error
7𝐘𝐢 − 𝐘 error
J +⁄− 𝐭 𝐜 𝐒𝐘G
- recall for CI : 𝐘
- since there are 2 sources of error and not just one ➞ adjust 𝐒𝐘G
forecast value of the IV
𝟏 3)𝟐
(𝐗 𝐟 − 𝐗
adjusted 𝐒𝐘G : 𝐒𝐟𝟐 = 𝐒𝐞𝟐 N𝟏 + + P and 𝐒𝐟 = M𝐒𝐟𝟐
𝐧 (𝐧 − 𝟏)𝐒𝐱𝟐
𝐧
SEE
&) 𝟐
^(𝐗 𝐢 − 𝐗
𝐢7𝟏

55
 Last Revised: 05/12/2023

Page 12
J:
Prediction interval (or CI) for 𝐘 LOS g
3 - calculate
3 )𝟐
𝟏 (𝐗 𝐟 − 𝐗
𝟐 𝟐 - interpret
𝐒𝐟 = 𝐒𝐞 N𝟏 + + P
𝐧 (𝐧 − 𝟏)𝐒𝐱𝟐
1
2
1. the better the fit of the regression model ➞ lower 𝐒𝐞𝟐 ➞ lower 𝐒𝐟𝟐
2. larger 𝐧 = smaller 𝐒𝐟𝟐
3 ➞ smaller 𝐒𝐟𝟐
3. close 𝐗 𝐟 is to 𝐗
Steps/ Determine o
𝐘
Select ∝
Determine 𝐭𝐜
Determine 𝐒𝐟
Determine o𝐟 +⁄− 𝐭 𝐜 𝐒𝐟
𝐘

example #7

Page 13
1/ Log-lin model LOS h
- describe
𝐘 = 𝐞𝐛𝟎 J𝐛𝟏 𝐗𝐢 Revenues
- take the log of
both sides
growth rate
𝐥𝐧 𝐘𝐢 = 𝐛𝟎 + 𝐛𝟏 𝐗 𝐢 of revenue

relative
change in 𝐘 for
absolute change in 𝐗
2/ Lin-Log model
𝐘 = 𝐛𝟎 + 𝐛𝟏 𝐥𝐧(𝐗 𝐢 )
absolute change in 𝐘 for relative change in 𝐗
- when 𝐘 and 𝐗 are significantly different in scale exh. #35/36
e.g./ 𝐘 = percent 𝐗 = billions of $ in Revenue
(transform 𝐗 with 𝐥𝐧(𝐗))

56
 Last Revised: 05/12/2023

Page 14
3/ Log-Log model 𝐥𝐧 𝐘𝐢 = 𝐛𝟎 + 𝐛𝟏 𝐥𝐧(𝐗 𝐢 ) LOS h
- describe
the relative change in 𝐘𝐢 for a
relative change in 𝐗 𝐢
- exh. #37/38

selecting a model depends on goodness of fit

𝐑𝟐
F-stat
SEE (𝐒𝐜 )
- a plot of the residuals should show randomness and the
distribution should be normal
- if not, consider transforming the DV, IV, or both.

57
 Last Revised: 05/12/2023

Introduction to Big Data Techniques

a. describe aspects of “fintech” that are directly relevant for the gathering and
analyzing of financial data.

b. describe Big Data, artificial intelligence, and machine learning

c. describe applications of Big Data and Data Science to investment management

58
 Last Revised: 05/12/2023

Intro. to Big Data Techniques

Page 1
Fintech - technological innovation in the design and
delivery of financial services and products
analysis of large datasets
- traditional data
- alternative data - social media, sensor networks
analytical tools - for extremely large datasets ➞ AI ➞ very
useful for complex non-linear relationships
Big Data/ data generated from traditional/alternative sources
Characteristics
Volume - millions to billions of data points
quality Velocity - speed of recording/transmitting/generating
vs. has accelerated (real time/near real-time)
quantity Variety - text, image, speech, video
Veracity - credibility/reliability of different sources

Page 2

structured - can be organized in tables, database entries

semi unstructured - cannot be represented in tabular format (voice, images)
(all data must be structured before use)
Sources/
financial markets - price/volume
businesses - filings, press releases
governments - economic data
individuals - web footprints, transactional data
sensors - images, traffic volume
Internet of Things - smart buildings, appliances
3 main sources of alternative data:
Individuals (text, photos, audio, website clicks/engagement)
Business processes (sales, credit card data, corporate exhaust)
supply chain info.
Sensor data (smartphones, RFID chips)
PoS scanner data
➞ IoT - array of physical devices
etc...

59
 Last Revised: 05/12/2023

Page 3
Ethical/Legal issues ➞ use of personal info./data scraped from
web data
Challenges quality selection bias outliers
volume missing data data cleansing/organizing
appropriateness
AI - artificial intelligence - systems that exhibit cognitive and
decision-making ability comparable or superior to humans
- from expert systems to neural networks
(if - then rules) (based on how brains learn and
process info.)
Machine learning - seek to extract knowledge from large amounts
of data without assumptions of the data’s underlying
probability distribution
- learn from known examples to determine underlying
structure in the data - generate structure without
help from a human

Page 4

requires large training datasets

training validation test

data data data test ability to predict
on new data

labelled data inputs model validated

outputs and tuned
supervised learning
over/under-fit - will not perform well out-of-sample
Unsupervised learning - only inputs, no outputs

Deep learning - multi-stage, non-linear data processing

(supervised or unsupervised)

Data Science - comp. science + statistics to extract info. from

big data

60
 Last Revised: 05/12/2023

Page 5
Data processing methods/
Capture - low latency systems (very little delay) vs. high latency
Curation - ensuring data quality/accuracy
- detect errors and make adjustments for missing data
Storage - record/archive/access in databases
Search - how data are queried
Transfer - movement of data from storage to analytical
application
Data Visualization/
- how data is formatted/displayed/summarized in
graphical format
(e.g. heat maps, tree diagrams, network graphs)
- tag clouds (text data) - words are sized/displayed
based on frequency

Page 6

Text analytics - use large unstructured voice or text datasets

Natural language processing (NLP) - comp. science + AI + linguistics

- analyze/interpret human language
(e.g. translation, speech recognition)

61