Level I Quantitative Methods

of the RATES AND RETURNS

CFA® Program
Learning Outcome Statements

LOS : 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.
LOS : Calculate and interpret different approaches to return
measurement over time and describe their appropriate uses.
LOS : Compare the money-weighted and time-weighted rates of return
and evaluate the performance of portfolios based on these measures.
LOS : Calculate and interpret annualized return measures and
continuously compounded returns, and describe their appropriate
uses.
LOS : Calculate and interpret major return measures and describe their
appropriate uses.
Basic Time Value of Money
Question: Which of the following options would you prefer?
• Receive $5,000, now or a year later?
• That’s time value of money.
• Time allows you the opportunity to postpone consumption and earn interest.

Rates of Return

Three Perspectives
Minimum rate of return Minimum return demanded to accept receipt of funds at a later date.
Discount rate A rate applied to the future cash flow to determine its present value.
Opportunity cost Cost of spending the money today instead of saving it for a certain
period and earning a return on it.

Components of Rates
Real risk-free rate Purely reflects individuals’ preferences for current v/s future consumption.
Inflation premium Compensation against expected loss in purchasing power over the term of a loan.

𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅: 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑅𝑅𝑅𝑅 + 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃

Default premium Compensates investors for the risk that the borrower might fail to make promised
payments in full in a timely manner.
Liquidity premium Compensates investors for any difficulty that they might face in converting their
holdings readily into cash at their fair value.
Maturity premium Compensates investors for the higher sensitivity of the market values of longer-
term debt instruments to changes in interest rates.
Rates of Return
Types of Rates of Return
Holding Period Return 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃1 − 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃0 + 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼1
𝐻𝐻𝐻𝐻𝐻𝐻 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 =
(Single period) 𝑃𝑃0

• Assumes the income is received at the end of each period.

• If the income is received any earlier and is reinvested, then the holding
period return can be higher.
Holding Period Return 𝐻𝐻𝐻𝐻𝐻𝐻 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 = [(1 + 𝐻𝐻𝐻𝐻𝐻𝐻1 )(1 + 𝐻𝐻𝐻𝐻𝑅𝑅2 ) … . (1 + 𝐻𝐻𝐻𝐻𝐻𝐻3 )] − 1
(Multiple periods)
Arithmetic Mean 𝐻𝐻𝐻𝐻𝐻𝐻1 + 𝐻𝐻𝐻𝐻𝐻𝐻2 + ⋯ + 𝐻𝐻𝐻𝐻𝐻𝐻𝑛𝑛
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴ℎ𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑀𝑀𝑀𝑀𝑎𝑎𝑎𝑎 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 =
Return 𝑛𝑛

• Assumes the amount invested at the start of each period is the same.
• Heavily influenced by outliers.
Geometric Mean 𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
1�
Return = [(1 + 𝐻𝐻𝐻𝐻𝐻𝐻1 )(1 + 𝐻𝐻𝐻𝐻𝑅𝑅2 ) … . (1 + 𝐻𝐻𝐻𝐻𝑅𝑅3 )] 𝑛𝑛 −1

• More accurate because it accounts for compounding across periods.

• It explains the ending value of a dollar invested at the start perfectly.
• Geometric mean will always be smaller than arithmetic mean,
o unless there is no variability in the data, where they will be equal.
o Higher the variability in the data, larger this difference.
• Geometric mean represents the growth rate per year, often referred to
as Compounded Annual Growth Rate (CAGR).
Rates of Return
Types of Rates of Return
Harmonic Mean 𝑛𝑛
𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 =
1 1 1
𝑥𝑥1 + 𝑥𝑥2 + ⋯ + 𝑥𝑥𝑛𝑛
Each X>0

• Used when ratios are applied to a fixed number.

• Example: Dollar cost averaging.
• The large outlier has a smaller impact on the harmonic mean than it
would have on the arithmetic mean.
• The harmonic mean will be less than the geometric mean which is, in
turn, less than the arithmetic mean . . . unless all observations are
equal.

Winsorised & Trimmed • Methods used to reduce impact from outliers.

Means • Trimmed mean removes a defined percentage of the largest and smallest
values from the data set.
• Winsorised replaces the outlier values with their nearest aboservations
before calculating the mean.

Example: A manager invests $5,000 annually in an ETF for four years. Below were the prices of the ETF
at the time of purchase.
Y1: $62, Y2: $76, Y3 = $84, Y4: $90.
The average cost price of the ETF for the manager across the four years equals – ?

Solution:
$76.48
Other Rates of Return used in Portfolio Management
Money-Weighted Rate of Return (Internal Rate of Return or IRR)
• Basics: The value of an asset can be considered as the present value of its future cash flows when
applying a relevant discount rate.
𝐶𝐶1 𝐶𝐶2 𝐶𝐶𝑛𝑛
𝑉𝑉0 = 1
+ 2
+ ⋯+
(1 + 𝑟𝑟) (1 + 𝑟𝑟) (1 + 𝑟𝑟)𝑛𝑛

• The rate that is being used here, to equate the future cash flows to the current value is called as
the Internal Rate of Return.
Calculation of Internal Rate of Return
Example: At t=0, an investor buys a share at the price of $100. After a year, she purchases another
share of the same stock at $120. After another year (t=2), she sells both the shares for a total price of
$270. She also received a dividend of $5 a share at the end of each year. What is the investor’s IRR?

Calculator method:
CF, 2nd CE|C
CF0 = -100, Enter, Down
C01 = -115, Enter, Down
F01 = 1, Enter, Down
C02 = 270, Enter, Down
F02 = 1, Enter
IRR, CPT
Ans: 16.5%
Other Rates of Return used in Portfolio Management
Time-Weighted Rate of Return
• TWR measures the compounded rate of growth of $1 invested over a stated period.
• It neutralized the impact of cash withdrawals and deposits into the portfolio.
Calculation of Time-weighted Return
Example: At t=0, an investor buys a share at the price of $100. After a year, she purchases another
share of the same stock at $120. After another year (t=2), she sells both the shares for a total price of
$250. She also received a dividend of $5 a share at the end of each year. What is the investor’s IRR?

Step 1: Splitting the holding period into two distinct years. Subperiods can of any duration and need
not be uniform.

Step 2: Calculate the returns for each subperiod.

Year 1 return:
(25%)

Year 2 return:
(12.5%)

Step 3: Calculate the holding period returns by linking the subperiod returns.

Holding Period Return:

(40.63%)

Step 4: Find Geometric mean return p.a.

(18.59%)

Conclusion for difference between TWR and MWR:

Other Rates of Return used in Portfolio Management

Annualized Returns
• Enables comparison across investment assets / portfolios with different maturities.
Stated vs Effective Annual Return (Discrete Compounding)
Example: An investor deposits money with a financial institution who provides her with three options.
Option 1: 5% per annum, compounded annually.
Option 2: 4.91% per annum, compounded quarterly.
Option 3: 4.88% per annum, compounded weekly.
Which option should the investor pick from?

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = (1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑠𝑠𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 )52 − 1

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = (1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑠𝑠𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 )365 − 1

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = (1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑠𝑠18 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 )365/18 − 1

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = (1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑠𝑠18 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 ℎ )2/3 − 1

Other Rates of Return used in Portfolio Management

Annualized Returns
Stated vs Effective Annual Return (Continuous Compounding)
Example: An investor is set to receive a return of 5% p.a., continuously compounded. What is the
effective annual return?

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑛𝑛𝑛𝑛𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = 𝑒𝑒 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟

−1

Ans: 5.13%
Calculator steps:

Example: An investor earns an effective 10% in a year in the stock market. What is the continuously
compounded rate?

𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑛𝑛𝑛𝑛𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = ln(1 + 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅)

Ans: 9.53%

Another Nuanced Example: An investor invests $1000 in the stock market, and it grows to $1,250 in 2
years. What is the continuously compounded rate?

Ans: 11.16%

𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑛𝑛𝑛𝑛ℎ𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 = ln(1 + 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅)

Other Rates of Return used in Portfolio Management
Gross and Net Return for a Fund
Gross Returns Net Returns
• Returns earned prior to the deduction of any • Actual returns investors earn after deduction
management expenses, custodian fees, taxes, of fees by the asset manager.
and other expenses that are not directly • Large funds spread admin costs across a larger
related to the generation of return. asset base to generate higher net return.
• Commissions, trading fees, and charges that • To stay competitive, smaller funds may waive
are related to the investment action itself are some admin fees to generate high net return.
deducted before arriving at the gross returns.
• Helpful in comparing the skills of various
managers.

Nominal and Real Returns

Nominal Returns Real Returns
Pre-tax Post-tax • Measures the actual growth in the purchasing
• Net returns • Net returns power for an investor.
calculated in the adjusted for taxes. • Adjusts for both, taxes, and inflation.
previous section. • Does not account • Useful in measuring returns across countries
• Do not account for for inflation. with different inflation levels.
inflation / taxes.

(1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 )(1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑝𝑝𝑝𝑝𝑒𝑒𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚)

(1 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 ) =
(1 + 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖)

Leveraged Return
• Investors can also used borrowed funds to invest in the market.
• Borrowed funds create leverage – an effect which magnifies profits and losses.
• More of this will be covered in equity and derivatives.

𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑒𝑒𝑑𝑑𝑑𝑑 𝑏𝑏𝑏𝑏
𝑟𝑟𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 = 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑛𝑛𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 + (𝑟𝑟 − 𝑟𝑟𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 )
𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑒𝑒𝑒𝑒𝑒𝑒𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝
 Level I Quantitative Methods
of the TIME VALUE OF MONEY IN FINANCE
CFA® Program
Learning Outcome Statements

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

LOS : 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 flow.

LOS : 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.
Fixed Income and Time Value of Money
Fixed Income Cash Flow Patterns
• Zero Coupon Bonds: The investor receives one payment at the bond maturity date. The value of
the bond is the present value of the final single payment.
• Periodic Interest / Coupon Bonds: Investors receive periodic coupon payments and the par value
at maturity.
• Level Payment Bonds / Amortizing Bonds: Investors receive uniform payments every period which
consists of both principal and interest amount.

A Few Nuances to practice TVM functionality.

Example 1: A 5% annual coupon paying bond has a par value of $100 and has 6 years to maturity.
If the required rate of return is 6%, what is the value of this bond today?

Example 2: A 10% coupon, semi-annual bond is currently priced at $105 and has 3 years to maturity.
What is the yield to maturity (required rate of return) on this bond?

Example 3: A borrower receives $100,000 from the bank as a part of a 10-year mortgage. The interest
rate charged is 8% p.a. What the is monthly installment required by the bank if the borrower needs to
pay the complete principal along with interest at the end of 10 years?

Example 4: A perpetual bond pays 4% quarterly coupon on a $100 par. What is the value of the bond
today if the required rate of return is 6%?
Equity and Time Value of Money
Dividend Patterns
• Constant $ Dividends: The investor expects to receive the same $ amount every period.
• Constant Dividend growth rate: Dividends are expected to grow at a constant rate forever.
• Changing Dividend growth rate: Dividends grow at different rates based on lifecycle phase.

Note: The rest of the module has been intentionally

left out at this stage as each of the concepts will get
covered in complete detail in the respective topics.
 Level I Quantitative Methods
of the STATISTICAL MEASURES OF ASSET RETURNS
CFA® Program
Learning Outcome Statements

LOS : Calculate, interpret, and evaluate measures of central tendency

and location to address an investment problem.
LOS : Calculate, interpret, and evaluate measures of dispersion to
address an investment problem.
LOS : Interpret and evaluate measures of skewness and kurtosis to
address an investment problem.
LOS : Interpret correlation between two variables to address an
investment problem.
Measures of Central Tendency
Arithmetic Mean
Formula

Key points • The mean is also unique; that is, each data set has only one mean.
• The arithmetic mean of a sample is the best estimate of the next observed value.
• The sum of the deviations from the arithmetic mean always equals 0.

Limitations • Sensitive to extreme outliers.

• Trimmed / Winsorised means can be used to overcome this limitation.

Geometric Mean
Formula
Key points • Smaller than Arithmetic mean, unless variability = 0, where it will be equal to AM.

Weighted Mean
Formula

Key points • Different observations often carry different weights in a portfolio.

• Weighted mean accounts for these differences in weights.

Median
Formula • The middle value of the data once arranged in ascending or descending order.
• The median for an odd number of variables is the observation in the position
identified by (n + 1)/2.
• The median for an even number of variables is the average of observations in the
positions identified by n/2 and (n + 2)/2.
Key points • Not sensitive to outliers and is a useful measure for asymmetric distributions.

Mode
Concept • The most frequently occurring value in a data set.
• For grouped data, the mode is the interval with the greatest number of observations.
• A data set may have no mode, one mode, or more than one mode.
Measures of Location
Quantiles
Divides data and groups it into parts
Quartile • Data divided into four parts.
Quintile • Data divided into five parts.
Decile • Data divided into ten parts.
Percentile • Data divided into one hundred parts.
Interquartile Range
Interquartile range is the difference between the top of the third quartile and the bottom of the
second quartile (i.e., top of the first quartile) or IQR = Q3 − Q1.
• Quantiles may be visualized by a box-and-
whisker plot.
• To better identify outliers, the boundaries
of the IQR +/- some percentage will be
applied to establish the whiskers.
• Outliers will appear outside the whiskers.
Measures of Location
Quantiles
Example: (2024_L1V1, Pg: 94, Example 2)

1. Identify the 10th and 90th percentile. (-0.876 & 0.991)

2. Identify first, second, third quintiles. (-0.432, -0.070, 0.173)

3. Identify the median. (0.044)

4. Calculate IQR. 0.460 – (-0.293) = 0.753

Measures of Location
Finding a Quantile from a data set
Step 1: • Find the positional location in an ascending array of the value below which y% of
observations lie –

Step 2: • When L is a whole number, it represents an actual observation.

• When L is not a whole number, it becomes necessary to use the linear interpolation
between the two observations around L.
Example: Calculate the first quartile of a distribution that consists of the following asset returns: 10%,
23%, 13%, 17%, 19%, 5%, 4%.

If we include one more return observation of 10% in our data set, what is the new value of the first
quartile?

What is the IQR of this new data?

Measures of Dispersion
Range
Concept • Simplest dispersion measure.
• Maximum value – Minimum Value.
Key points • Gives no information about distribution of data.
• If extreme outliers exist, the range is not representative of average dispersion.

Mean Absolute Deviation

Concept & • Average of the absolute deviations from the mean.
Formula

Key points • MAD, unlike range, uses all the observations in the data set.
• Difficult to manipulate mathematically.
Measures of Dispersion
Variance and Standard Deviation
Variance • Average of the squared deviations around the mean.
• However, when squaring, the unit of the original data set is not preserved.

2
∑𝑛𝑛𝑖𝑖=1(𝑥𝑥𝑖𝑖 − 𝜇𝜇)2 2
∑𝑛𝑛𝑖𝑖=1(𝑥𝑥𝑖𝑖 − X̄)2
𝜎𝜎 = 𝑠𝑠 =
𝑛𝑛 𝑛𝑛 − 1

Standard • Positive square root of variance.

Deviation • Original unit of the data set is preserved.
Key Points • Arithmetic mean is usually presented in the standard deviation calculations.
• If the series describes returns across time, geometric mean may be used.
• The relationship between geometric and arithmetic mean is –

• Mean Absolute Deviation will always be lower than standard deviation due to the
squaring of deviations.
Example: Calculate the variance and standard deviation for five golfers with the scores of 67, 71, 72,
75, 68. How do the numbers differ if they represent (i) a sample or (ii) the entire population.

Calculator Method:
Measures of Dispersion
Target Downside Deviation / Semi Deviation
Concept & • Measures the average deviation of the values below a target.
Formula • Focuses only on downside risk.

Example: Calculate the target downside deviation if the target is 71.

Data: 67, 71, 72, 75, 68

Relative Measures of Dispersion – Coefficient of Variation

Concept & • Measures dispersion relative to a reference / benchmark.
Formula • Coefficient of variation measures the level of risk per unit of return.

Key Points • When observations are returns, CV measures the amount of standard deviation per
unit of return.
• If the mean return is negative, then the statistic is economically useless.
Measures of Shape of a Distribution
Skewness
Concept • Measures the level of asymmetry in a data distribution.
Skewness = 0

Positively
Skewed,
Skewness > 0

• Frequent small losses, a few extremely large gains.

• Investors prefer this as the average return is greater than the median.
Negatively
Skewed,
Skewness < 0

• Frequent small gains, a few extremely large losses.

• Less preferred by investors as average return is lower than the median return.
Measures of Shape of a Distribution
Kurtosis
Concept • Measures the weight of the tails in a distribution compared to the rest of the
data distribution.
Different • Using a normal distribution as a benchmark,
Distributions

Graphical 2024_L1V1, CFA Curriculum, Pg: 112, Exhibit 18

Representation

• Most equity returns are leptokurtic.

• Risk managers must consider a higher probability of unexpectedly large
gains/losses in the portfolio.
• Return distibutions could exhibit kurtosis and skewness.
Correlation & Covariance
Linear Relationship between Two Variables
Covariance • Joint variability between two variables.

• A positive covariance indicates that both the variables can be expected to be above
(or below) their respective means simultaneously.
• A negative covariance indicates that when one variable is higher than its mean, the
other is expected to be lower than its respective mean and vice-versa.
• Variables could be of different scales and magnitudes. Covariance does little to
explain about the strength of the linear relationship.
Correlation • Standardized measure of linear relationship that describes not only the direction of
the linear relationship but also the strength.

• Importantly, correlation cannot be misconstrued to depict causation.

• Zero correlation doesn’t imply no relationship, it only indicates the absence of a linear
relationship between the variables.
Example: Calculate the Covariance and Correlation between A & B.
Asset A Asset B
Year 1 8% 3%
Year 2 10% 5%
Year 2 15% 8%
 Level I Quantitative Methods
of the PROBABILITY TREES AND CONDITIONAL
CFA® Program EXPECTATIONS
Learning Outcome Statements

LOS : Calculate expected values, variances, and standard deviations

and demonstrate their application to investment problems.
LOS : Formulate an investment problem as a probability tree and
explain the use of conditional expectations in investment
application.
LOS : Calculate and interpret an updated probability using Bayes’
formula.
Basics of Probability
Random variable One whose possible values or results are uncertain.
Outcome Observed value of a random variable.
Event Could be a single outcome or a set of outcomes.
Mutually exclusive events Are events that cannot happen simultaneous.
Exhaustive events Range of all possible outcomes of an event.

Unconditional probability • Unconditional or marginal probabilities estimate the probability of

(Standalone probability) an event irrespective of the occurrence of another event.
• Example: probability of passing the exam: P(pass)
Conditional probability • Conditional probabilities express the probability of an event
occurring given that another event has occurred.
• Example: probability of passing the exam given the exam was easy:
P(Pass/Easy)

Dependent event The occurrence of one is related to the occurrence of the other.
Conditional Probability ≠ Unconditional Probability
P (Pass/Easy) ≠ P (Pass)
Independent event The occurrence of one does has no bearing on the occurrence of the other.
Conditional Probability = Unconditional Probability
P(Pass/Rain) = P(Pass)
Rules of Probability
Rule What it computes? Formula
Multiplication • Probability of 2 key events Dependent events
occurring together (joint 𝐴𝐴
𝑃𝑃(𝐴𝐴𝐴𝐴) = 𝑃𝑃(𝐵𝐵) × 𝑃𝑃 � �
probability). 𝐵𝐵
• Key words: and, both, together and Independent events
combined. 𝑃𝑃(𝐴𝐴𝐴𝐴) = 𝑃𝑃(𝐵𝐵) × 𝑃𝑃(𝐴𝐴)
Addition • Probability of at least one of the P(A or B) = P(A) + P(B) − P(AB)
two events occur.
• Key words: or, either and at least. 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
P(AB) = 0
Total • Expresses the unconditional 𝑃𝑃(𝐴𝐴) = 𝑃𝑃(𝐴𝐴𝐴𝐴) + 𝑃𝑃(𝐴𝐴𝑆𝑆 𝑐𝑐 )
probability rule probability of an event in terms of
conditional probabilities for 𝐴𝐴 𝐴𝐴
𝑃𝑃(𝐴𝐴) = 𝑃𝑃(𝑆𝑆) × 𝑃𝑃 � � + 𝑃𝑃(𝑆𝑆 𝑐𝑐 ) × 𝑃𝑃 � 𝑐𝑐 �
mutually exclusive and exhaustive 𝑆𝑆 𝑆𝑆
events.
Where S and 𝑆𝑆 𝑐𝑐 represent 2 mutually
• Trick: the question give conditional
exclusive and exhaustive scenarios.
probabilities but asks to find
unconditional probabilities.
Deriving conditional probability from joint probability:
A P(AB) B P(BA)
P� � = or P � � = Note: P(AB) = P(BA)
B P(B) A P(A)
Probability Trees
Example: There is 30% probability that the exam shall be easy. There is 80% chance of passing if the
exam is easy and 40% chance of passing if the exam is hard. Given the probabilities find the following:
A. Probability that the exam is easy, and student shall pass the exam.
B. Probability that the student shall pass the exam.
C. Probability that student shall pass the exam, or the exam is easy.
Step for tree:
1. Start with independent probabilities.
2. Layout conditional probabilities in each scenario.
3. Calculate join probability.

Pass Pass & Easy

P(P/E) = 80% P(PE) = 30% x 80% = 24%
Easy
P(E) = 30%
Fail Fail & Easy
P(F/E) = 20% P(FE) = 30% x 20% = 6%
Exam
Pass Pass & Hard
P(P/H) = 40% P(PH) = 70% x 40% = 28%
Hard
P(H) = 70%
Fail Fail & Hard
P(F/H) = 60% P(FH) = 70% x 60% = 42%

A. Probability that the exam is easy, and student shall pass the exam.
P(PE) = P(E) x P(P/E) = 24%

B. Probability that the student shall pass the exam.

P(P) = P(PE) + P(PH) = 24% + 28% = 52%

C. Probability that student shall pass the exam, or the exam is easy.
P(P or E) = P(P) + P(E) – P(PE) = 52% + 30% - 24% = 58%
Probability Trees & Bayes’ Formula
Bayes’ formula uses the occurrence of the event to infer the probability of the scenario generating it.
• For this reason, Bayer’s formula is sometimes called an inverse (posterior) probability.
Hint: the question shall ask for inverse conditional probability compared to the one given in the
question.
Example: In the above question, what is the probability that the exam is easy given the student
passed the exam.
E P(EP) 24%
P� � = = = 46.15%
P P(P) 52%

Trick to identify Bayes question: watch the question asked for P(E/P) when in the question we were
given P(P/E).
Expected Value, Variance, Standard Deviation of a Random Variable
Expected The probability-weighted average of all possible outcomes for the random variable.
𝑛𝑛
value
𝐸𝐸(𝑋𝑋) = � 𝑃𝑃𝑖𝑖 𝑋𝑋𝑖𝑖
𝑖𝑖=1
• Expected values are forecasts of what might happen based on probabilities.
• The expected value represents our best guess – the average of all outcomes over
the long run.
Variance The probability weighted sum of the squared differences between each outcome and
the expected value.
𝑛𝑛
2
𝜎𝜎 = � 𝑃𝑃𝑖𝑖 [𝑋𝑋𝑖𝑖 − 𝐸𝐸(𝑋𝑋)]2
𝑖𝑖=1
• If variance equals zero, there is no dispersion in the distribution.
• In this case, the outcome is certain, and the variable, X, is not a random variable.
Standard Standard deviation is the positive square root of the variance.
deviation 𝜎𝜎 = �𝜎𝜎 2
The variance has no units while the standard deviation is expressed in the same unit as
the expected value and the random variable itself.
Example: Calculate the expected value, variance and standard deviation of coal prices given the
following possible outcomes and their associated probabilities:
Probability 8% 15% 40% 25% 12%
Coal Price $40 $50 $60 $70 $80

E (Coal price) = (0.08 × 40) + (0.15 × 50) + (0.40 × 60) + (0.25 × 70) + (0.12 × 80) = $61.80

σ2 = 0.08(40 − 61.8)2 + 0.15(50 − 61.8)2 + 0.40(60 − 61.8)2 + 0.25(70 − 61.8)2 + 0.12(80 − 61.8)2 = 116.76

σ = √116.76 = $10.81

Calculator Steps
Expected Value, Variance, Standard Deviation of a Random Variable
The total probability rule for expected value state (unconditional) expected values in terms of
conditional expected values.
𝑋𝑋 𝑋𝑋
𝐸𝐸(𝑋𝑋) = 𝑃𝑃(𝑆𝑆) × 𝐸𝐸 � � + 𝑃𝑃(𝑆𝑆 𝑐𝑐 ) × 𝐸𝐸 � 𝑐𝑐 �
𝑆𝑆 𝑆𝑆

Where S and 𝑆𝑆 𝑐𝑐 represent 2 mutually exclusive and exhaustive scenarios.

• Like expected value, variance also has a conditional counterpart to the unconditional concept.
• Conditional variance can be used to assess the risk of a particular scenario.
Example: The is 40% chance of a good economy and 60% chance of a poor economy.
• Good economy: 70% chance that EPS is $10 and 30% chance that EPS is $6.
• Poor economy: 20% chance that EPS is $6 and 80% chance that EPS is $3.
Given the above information, calculate the unconditional expected value [E(EPS)] and conditional
expected value [E(EPS/Good)] and [E(EPS/Bad)].

EPS = $10
Prob. = 70%
Good -> P(G) = 40%
E(EPS/Good) = 10 x 70% + 6 x 30% = $8.80
EPS = $6
Prob. - 30%
Economy
E(EPS)= 8.80 x 40% + 3.60 x 60% = $5.68
EPS = $6
Prob. = 20%
Bad -> P(B) = 60%
E(EPS/Bad) = 6 x 20% + 3 x 80% = $3.60
EPS = $3
Prob = 80%
 Level I Quantitative Methods
of the PORTFOLIO MATHEMATICS
CFA® Program
Learning Outcome Statements

LOS : Calculate and interpret the expected value, variance, standard

deviation, covariances, and correlations of portfolio returns.

LOS : Calculate and interpret the covariance and correlation of portfolio

returns using a joint probability function for returns.

LOS : Define shortfall risk, calculate the safety-first ratio, and identify an
optimal portfolio using Roy’s safety-first criterion.
Portfolio Risk and Return
Expected Return on a Portfolio
The expected return on a portfolio is a function of the returns on the individual assets and of their
respective weights.
𝑡𝑡

𝐸𝐸(𝑅𝑅𝑃𝑃 ) = � 𝑤𝑤𝑖𝑖 𝐸𝐸(𝑅𝑅𝑖𝑖 )

𝑖𝑖=1

𝑤𝑤𝑖𝑖 = market value of investment / market value of the portfolio

𝐸𝐸(𝑅𝑅𝑖𝑖 ) = expected return on individual assets.
Variance of a Portfolio
2 asset portfolios
𝜎𝜎 2 𝑃𝑃 = 𝑤𝑤 2𝐴𝐴 𝜎𝜎 2𝐴𝐴 + 𝑤𝑤 2 𝐵𝐵 𝜎𝜎 2 𝐵𝐵 + 2𝑤𝑤𝐴𝐴 𝑤𝑤𝐵𝐵 𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐵𝐵

We can replace 𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐵𝐵 = 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐵𝐵 𝜎𝜎𝐴𝐴 𝜎𝜎𝐵𝐵

𝜎𝜎 2 𝑃𝑃 = 𝑤𝑤 2𝐴𝐴 𝜎𝜎 2𝐴𝐴 + 𝑤𝑤 2 𝐵𝐵 𝜎𝜎 2 𝐵𝐵 + 2𝑤𝑤𝐴𝐴 𝑤𝑤𝐵𝐵 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐵𝐵 𝜎𝜎𝐴𝐴 𝜎𝜎𝐵𝐵

3 asset portfolios
𝜎𝜎 2 𝑃𝑃 = 𝑤𝑤 2𝐴𝐴 𝜎𝜎 2𝐴𝐴 + 𝑤𝑤 2 𝐵𝐵 𝜎𝜎 2 𝐵𝐵 + 𝑤𝑤 2 𝐶𝐶 𝜎𝜎 2 𝐶𝐶 + 2𝑤𝑤𝐴𝐴 𝑤𝑤𝐵𝐵 𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐵𝐵 + 2𝑤𝑤𝐴𝐴 𝑤𝑤𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝐴,𝐶𝐶 + 2𝑤𝑤𝐵𝐵 𝑤𝑤𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐵𝐵,𝐶𝐶

Similar to above we can replace covariance in the formula by correlation and product of individual
asset standard deviation.
Standard deviation
𝜎𝜎𝑃𝑃 = �𝜎𝜎 2 𝑃𝑃

Standard deviation can be directly calculated as the weighted average of individual asset standard
deviation when the correlation between assets is 1.

𝜎𝜎𝑃𝑃 = 𝑤𝑤𝐴𝐴 𝜎𝜎𝐴𝐴 + 𝑤𝑤𝐵𝐵 𝜎𝜎𝐵𝐵

Portfolio Risk and Return
Example: Calculate standard deviation of a portfolio with an investment of 60% in A and 40% in B.
A B
A 144 81
B 81 225

Covariance
Covariance between two random variables is the probability-weighted average of the cross-products
of each random variable’s deviation from its own expected value.
COVX,Y = P × [X i − E(X)] × [Yi − E(Y)]
Properties
• Covariance is symmetric, i.e., Cov(X,Y) = Cov(Y, X).
• Covariance can range from positive infinity to negative infinity.
• The covariance of X with itself, Cov(X,X), is equal to variance of X, Var(X).
• Covariance can be positive, negative and zero (no relationship).
Limitation
• Difficult to compare covariance across data sets that have different scales.
• Difficult to interpret covariance as it can take on extreme large values.
• Does not tell us anything about the strength of the relationship between the two variables.
Portfolio Risk and Return
Correlation
Correlation coefficient measures the strength and direction of the linear relationship between two
random variables.
COVX,Y
CORR X,Y =
𝜎𝜎𝑋𝑋 𝜎𝜎𝑌𝑌
Properties
• The correlation coefficient can have a maximum value of +1 and a minimum value of −1.
• A correlation coefficient greater than 0 means that when one variable increases (decreases) the
other tends to increase (decrease) as well.
• A correlation coefficient less than 0 means that when one variable increases (decreases) the other
tends to decrease (increase).
• A correlation coefficient of 0 indicates that no linear relation exists between the two variables.
Limitation
• It does not specify which factor or variable causes the linear relationship between the two
variables.
• Further, like covariance, the correlation coefficient is a measure of linear association.

Example: Calculate the covariance and correlation using the below joint probability distribution.
RA = 10% RA = 15% RA = 20%
RB = 15% 0.30 - -
RB = 20% - 0.60 -
RB = 25% - - 0.10

Example: Calculate correlation from the below covariance matrix.

A B
A 144 81
B 81 225
The Normal Distribution - Basics
A continuous random variable that follows the normal distribution has a bell-shaped probability
distribution.
Graphical depiction

Properties
• It is completely described by its mean (μ) and variance (σ2).
• Skewness is 0 (50% observation lie above the mean and 50% observations lie below the mean).
• Kurtosis equals 3 and excess kurtosis equals 0.
• A linear combination of normally distributed random variables is also normally distributed.
• The probability of the random variable lying in ranges further away from the mean gets smaller
and smaller, but never goes to zero.
Univariate vs Multivariate Distributions
Univariate • Describes the distribution of a single random variable.
Multivariate • Specify probabilities associated with a group of random variables considering
the interrelationships that may exist between them.
• A multivariate normal distribution for the return on a portfolio with n stocks is
completely defined by the following three parameters:
o n means
o n variances
o n(n-1)/2 pairwise correlation
• Example: Identify the parameters required for portfolio that includes 4 stocks?
o 4 means
o 4 variances
o 6 (4 x 3 / 2) pairwise correlation
The Normal Distribution – Confidence Interval
Confidence Interval
Represents the range of values within which a certain population parameter is expected to lie in a
specified percentage of the time.
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 = 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 ± 𝑍𝑍 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷
Approximate Z values Precise Z values
Degree of Confidence Z value Degree of Confidence Z value
50% 2/3 90% 1.65
68% 1.00 95% 1.96
95% 2.00 99% 2.58
99% 3.00
Graphical Representation

Example: The average annual return on a stock is 15% and the standard deviation of returns equals
5%. Given that the stock's returns are distributed normally, calculate the 90% confidence interval for
the return in any given year.
Solution
The 90% confidence interval is calculated as:
Upper limit = Sample Mean + Z value × Sample Standard Deviation
Upper limit = 15% + 1.65 × 5% = 23.25%

Lower limit = Sample Mean − Z value × Sample Standard Deviation

Lower limit = 15% − 1.65 × 5% = 6.75%

Interpretation: The probability that the return on the stock for a given year lies between 6.75% and
23.25% equals 0.90.
The Normal Distribution – Probability Applications
Calculating Probabilities with Normal Distribution Tables
• Standardization allows converting variables in different units into a single unit of measurement
(standard deviation).
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 − 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
𝑍𝑍 =
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
• This allows us to infer probabilities by referring to a single normal curve (standard normal
distribution curve with a mean = 0 and standard deviation = 1).

Key point: the z-value represents the number of standard deviations away from the population mean,
a given observation lies.

Steps for finding probability using the Z table:

• Step 1: calculate the z value
• Step 2: go the z table and find the probability corresponding to the z value. Remember the table is
designed to give us cumulative less than equal to probability [F(Z)].
• Step 3: if you need probability of more than some value, you have to subtract the less than
probability from step 2 from 1.

Key point:
• As the normal is a symmetrical distribution, we can say F(-Z) = 1 – F(Z)
• Remember for continuous random variables X > x is equal to X ≥ x.
The Normal Distribution – Probability Applications
Example: The points, X, scored by students in a class on their final exam are normally distributed with
a mean of 60 and standard deviation of 15.
Question 1: What is the probability that the points scored by a given student will be less than 80?

Question 2: What is the probability that the points scored by a given student will be more than 70?

Question 3: What is the probability that the points scored by a given student will be less than 40?

Question 4: What is the probability that the points scored by a given student will be between 30 and
90?
The Normal Distribution – Safety First & Shortfall Risk
Shortfall risk Refers to the probability that a portfolio's value or return, E(RP), will fall below a
particular target value or return (RT) over a given period.
Roy’s Safety- States that an optimal portfolio minimizes the probability that the actual portfolio
First Criteria return, RP, will fall below the target return, RT.

𝐸𝐸(𝑅𝑅𝑃𝑃 ) − 𝑅𝑅𝑇𝑇
𝑆𝑆𝑆𝑆 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 =
𝜎𝜎𝑃𝑃

Higher the better as its minimizes the probability of shortfall.

Probability of Calculating the probability of shortfall has 2 steps:
shortfall Step 1: find z value. Note that you Z = (-SFR)
Step 2: go the z table and find the probability corresponding to the z value
Example: An investor has $1,000 dollars to invest. His minimum acceptable portfolio value at the end
of the year is $1,050. He is considering two portfolios, A and B. Portfolio A has an expected return of
8% and a standard deviation of 10%, while Portfolio B has an expected return of 12% and a standard
deviation of 15%. Using Roy's safety-first criterion select the optimal portfolio and calculate the
probability that the portfolio's return will fall short of its target.
 Level I Quantitative Methods
of the SIMULATION METHODS
CFA® Program
Learning Outcome Statements

LOS : 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.
LOS : Describe Monte Carlo simulation and explain how it can be
used in investment applications.
LOS : Describe the use of bootstrap resampling in conducting a
simulation based on observed data in investment applications.
Lognormal Distribution

• A random variable, Y, follows the lognormal distribution if its natural logarithm, ln Y, is normally
distributed.
• The reverse is also true: If the natural logarithm of random variable Y, ln Y, is normally distributed,
then Y follows the lognormal distribution.
Key features
• It is bounded by zero on the lower end.
• The upper end of its range is unbounded.
• It is skewed to the right (positively skewed).
Application
• The lognormal distribution is frequently used
to model the probability distribution of asset
prices because it is bounded by zero on the
lower side.
• The normal distribution, on the other hand, Uses the ratio of ending value to beginning value
can be (and is frequently) used as an (V1/V0) as the random variable.
approximation for returns.
Key point: If a stock's continuously compounded return is normally distributed, then future stock
price must be lognormally distributed.

Scaling Returns and Volatility

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑛𝑛𝑠𝑠 ∗ 𝑁𝑁𝑁𝑁. 𝑜𝑜𝑜𝑜 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑖𝑖𝑖𝑖 𝑡𝑡ℎ𝑒𝑒 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦

𝐴𝐴𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 = 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 ∗ √𝑁𝑁𝑁𝑁. 𝑜𝑜𝑜𝑜 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑖𝑖𝑖𝑖 𝑡𝑡ℎ𝑒𝑒 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦
Monte Carlo Simulation
Concept • Generates random numbers and operator inputs to synthetically create
probability distributions for variables.
• It is used to calculate expected values and dispersion measures of random
variables, which are then used for statistical inference.
Steps Specifying the simulation
• Step 1: Specify the quantity of interest (revenue) along with the underlying
variables (e.g., price and quantity).
• Step 2: Specify a time period and break it into a number of subperiods.
• Step 3: Specify the distributional assumptions for the risk factors affecting the
underlying variables.
Running the simulation
• Step 4: Use a computer program or a spreadsheet function to draw random
values for each risk factor.
• Step 5: Based on the random values generated in Step 4, calculate the values for
the underlying variables (price and quantity).
• Step 6: Based on the average values of the underlying variables calculated in Step
5 calculate the revenue.
• Step 7: Iteratively go back to Step 4 until a sufficient number of trials has been
performed.
Investment • To experiment with a proposed policy before actually implementing it.
Applications • To provide a probability distribution that is used to estimate investment risk (Ex:
VAR).
• To provide expected values of investments that can be difficult to price.
• To test models and investment tools and strategies.
Limitations • Answers are only as good as the assumptions and model used.
• Does not provide cause-and-effect relationships.
Bootstrapping
Concept • Uses resampling – repeatedly draw samples from an original observed data.
• A special method of simulation where multiple samples are drawn from a sample
drawn from the original population.

2024_L1V1, Pg: 186, Exhibit 6

Steps: • Follows similar steps to Monte Carlo method, but the difference is that the
bootstrapping draws its samples from a sample from the original population,
whereas the Monte Carlo method generates random samples from the
underlying distribution.
Limitations • A risk factor that was not represented in historical data will not be considered in
the simulation.
• It does not facilitate “what if” analysis when the “if” factor has not occurred in
the past. Monte Carlo simulation can be used for “what if” analysis.
• It assumes that the future will be similar to the past.
• It does not provide any cause-and-effect relationship information.
 Level I Quantitative Methods
of the ESTIMATION AND INFERENCE
CFA® Program
Learning Outcome Statements

LOS : Compare and contrast simple random, stratified random,

cluster, convenience, and judgmental sampling and their
implications for sampling error in an investment problem.
LOS : Explain the central limit theorem and its importance for the
distribution and standard error of the sample mean.
LOS : Describe the use of resampling (bootstrap, jackknife) to
estimate the sampling distribution of a statistic.
Sampling Methods
Probability Sampling
• Gives every member of the population an
equal chance of being selected.
• It therefore creates a sample that is
representative of the population.
• Subtypes:
o Simple random sampling
o Systematic sampling
o Stratified sampling
o Cluster sampling
Simple • Each member of the population has the same probability or likelihood of being
random included in the sample.
sampling • Example: selecting a student from a class of 20 students where each student has an
equal likelihood of being selected.
• Useful with a homogenous population.
Systematic • Sampling, every kth member in the population list is selected until the desired
sampling sample size is reached.
• Example: every 3rd house in a row of 30 houses.
• Often used to generate approximately random samples.
Stratified • Stratification divides the group into mutually exclusive homogenous groups and
random undertakes simple random sampling from the groups in proportion to the
sampling population.
• Example: A class has 50 students, and we need to draw of sample of 10 students.
The class is first subdivided in 30 boys and 20 girls and 6 boys are selected from the
boys group and 4 girls are selected from girls group.
• Often improves the representativeness of the sample by reducing sampling error.
Cluster • Divides the population into clusters, where each cluster is a mini representation of
sampling the entire population and the selected certain (entire) clusters using simple random
sampling.
• In one stage, all members of the selected clusters are considered for sample and in 2
stage a subset of members of the selected members are considered in the sample.
• Example: Divide a city into different area cluster and then randomly select a cluster
to study consumption patterns.
• Time efficient and cost effective but may result in lower accuracy if cluster is not
representative of the population.
Sampling Methods
Non-Probability Sampling
• Rely on a researcher's sample selection capabilities (not on a fixed selection process).
• As a result, there is a significant risk of creating a non-representative sample.
Convenience • Selects elements based on ease of access or convenience, which limits sampling
sampling accuracy.
• Example: conducting survey in a mall.
• However, is allows quick data collection at a low cost in small scale pilot studies.
Judgmental • Handpicking elements from the population based on the researcher's knowledge
sampling and professional judgment.
• Example: archeologist choosing a place to dig given all options.
• Susceptible to biases and skewed results.
• Used when there is a time constraint and researcher skill is needed to select the
sample.

Sampling Error
Concept Is the error caused by observing a sample instead of the entire population to
draw conclusions relating to population parameters.
Formula Sampling error of the mean = Sample mean − Population mean = 𝑋𝑋� – 𝜇𝜇
The Sampling Distribution
A sampling distribution is the probability distribution of a given sample statistic under repeated
sampling of the population.
• Sampling distribution of the mean show the distribution of means obtained drawing same size
sample from the population and calculating the mean.
Example A population is made of 3 values: 2, 4 and 6. Given we can draw sample size of two
numbers with replacement, construct the sampling distribution of mean.

Sample no Sample value Sample Mean

1 2,2 2
2 2,4 3
3 2,6 4
4 4,2 3
5 4,4 4
6 4,6 5
7 6,2 4
8 6,4 5
9 6,6 6
Mean of Sample Mean 4

Key point: mean of sample mean is 4 which is equal to the population mean of 4 [(2 + 4 +
6) / 3 = 4
Central Limit Theorem
Make accurate statements about the population mean and variance using the sample mean and
variance regardless of the distribution of the population, as long as the sample size is adequate.
• An adequate sample size is defined as one that has more than 30 observations (n ≥ 30).
Key Properties
1 No matter what the distribution of the population, for a sample whose size is greater than or
equal to 30, the sample mean will be normally distributed.
2 The mean of the population (μ) and the mean of the distribution of sample means (𝑋𝑋�) are equal.
3 The variance of the distribution of sample means equals σ2/n, or population variance divided by
sample size.
Example: Given the population mean of 10% and population standard deviation of 20%, what shall be
the mean, variance, and standard deviation of sampling distribution with a sample size of 49.
Solution
• Mean of sampling distribution = population mean = 10%
• Variance of sampling distribution = σ2/n = 202/49 = 8.1632
• Standard deviation of sampling distribution = √8.1632 = 2.8571% or 20/√49 = 2.8571
Standard Error
The standard deviation of the distribution of sample means is known as the standard error of the
statistic. Standard error decreases as sample size increases.
Population variance is known Population variance is not known
𝜎𝜎 𝑠𝑠
𝜎𝜎𝑋𝑋� = 𝑠𝑠𝑋𝑋� =
√𝑛𝑛 √𝑛𝑛
Example: A manufacturer claims that the life of the batteries that it produces is normally distributed
with an average life of 30 hours, and a population standard deviation of 5 hours. For a random sample
of 40 batteries calculate the standard error.
𝜎𝜎 5
𝜎𝜎𝑋𝑋� = = = 0.79
√𝑛𝑛 √40
Interpretation: If we were to take all the possible random samples of 40 batteries each from the
entire population of batteries produced by this manufacturer, and prepare a distribution of sample
means, the distribution would have an average life of 30 hours, and a standard error (standard
deviation of the sampling distribution) of 0.79 hours.
Example: A sample containing the monthly returns for 50 U.S. stocks has a mean of 5% and a
standard deviation of 10%. Calculate the standard error of the sample mean.
𝑠𝑠 10
𝑠𝑠𝑋𝑋� = = = 1.41%
√𝑛𝑛 √50
Interpretation: If we were to take all possible samples of 50 stocks’ monthly returns from the
population of all stocks listed in the U.S., the sampling distribution would have a mean of 5% and a
standard error of 1.41%.
Standard deviation vs standard error
• If we want to draw conclusions about how spread out the data are, we use the standard
deviation.
• If we want to find out how precise the population parameter estimate (based on sampled data) is
relative to its true value, we use standard error.
Resampling
Resampling is the process of repeatedly drawing samples from the original observed data sample in
order to infer population parameters.
2 popular methods of resampling
Bootstrap • Process of using random sampling (with replacement) to construct the sampling
distribution of the sample mean.
• However, we have no knowledge of the population and thus the sample is treated as
the population.
• Powerful method for complicated estimators and does not rely on an analytical
formula to estimate the distribution of those estimators.
• Very accurate, and thus is used extensively in historical simulations in asset allocation
and in gauging an investment strategy's performance against a benchmark.
• The estimate of the standard error of the sample mean is calculated as follows:
𝑛𝑛
1 2
𝑠𝑠𝑋𝑋� = � ��θ�b − θ��
𝐵𝐵 − 1
𝑏𝑏=1

B denotes the number of resamples drawn from the original sample.

θ�b = b denotes the mean of a resample
θ� = denotes the mean across all the resample means
Jackknife • Also selects samples repeatedly but leaves out chosen observations from subsequent
resamples.
• Often used to reduce the bias of an estimator, and other applications include finding
the standard error and confidence interval of an estimator.
 Level I Quantitative Methods
of the HYPOTHESIS TESTING
CFA® Program
Learning Outcome Statements

LOS: Explain hypothesis testing and its components, including statistical

significance, Type I and Type II errors, and the power of a test.
LOS: 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.
LOS: Compare and contrast parametric and nonparametric tests, and
describe situations where each is the more appropriate type of test.
Hypothesis and Hypothesis Testing
Hypothesis • A hypothesis is a statement about the value of a population parameter
developed for the purpose of testing a theory.
• The population parameter could be mean, variance or correlation.
Hypothesis Testing • Hypothesis testing is the process of evaluating the accuracy of a
statement regarding a population parameter (e.g., the population mean)
given sample information (e.g., the sample mean).
Steps in Hypothesis Testing
Step 1: State the hypothesis
Null hypothesis: H0 Alternative hypothesis: Ha
• Statement we are interested in rejecting. • Statement we are trying to validate.
• Always includes equality (=, ≥, ≤). • Does not include equality (≠, >, <).
• Null testing = are two sided tests and null • Alternative with ≠ signs are two sided tests
with ≥, ≤ are one sided test. and alternative with >,< are one sided test.
Example: if analyst wants to test a belief that average rent in Emmar square is greater to AED
100,000, state the null and alternative.
• H0 : 𝜇𝜇 ≤ 100,000 / Ha : 𝜇𝜇 > 100,000

Step 2: Identify the appropriate test statistics

A test statistic is a value calculated on the basis of a sample that, when used in combination with a
decision rule, is the basis for deciding whether to reject the null hypothesis.
What we want to Test statistics Prob. distribution Degrees of Freedom
test of test statistics
Test of single mean �
𝑋𝑋 − 𝜇𝜇 z or t distributed n – 1 for t
𝑧𝑧 𝑜𝑜𝑜𝑜 𝑡𝑡 = 𝑠𝑠
� 𝑛𝑛
√
Test of differences (𝑋𝑋�1 − 𝑋𝑋�2 ) − (𝜇𝜇1 − 𝜇𝜇2 ) t distributed 𝑛𝑛1 + 𝑛𝑛2 – 2
𝑡𝑡 =
in mean 𝑠𝑠 2 𝑠𝑠 2
� 𝑃𝑃 + 𝑃𝑃
𝑛𝑛1 𝑛𝑛2
Test of mean ̅
𝑑𝑑 − 𝜇𝜇𝑑𝑑0 t distributed n–1
𝑡𝑡 =
differences 𝑠𝑠𝑑𝑑�
Test of single 2 Chi-square n–1
𝑠𝑠 (𝑛𝑛 − 1)
𝑋𝑋 2 =
variance 𝜎𝜎 2 0 distributed
Test of difference in 2 F test 𝑛𝑛1 − 1, 𝑛𝑛2 – 1
𝑆𝑆1
𝐹𝐹 = 2
variance 𝑆𝑆2
Test of correlation 𝑟𝑟√𝑛𝑛 − 2 t distributed n–2
𝑡𝑡 =
√1 − 𝑟𝑟 2
𝑚𝑚
Test of 2
�𝑂𝑂𝑖𝑖𝑖𝑖 − 𝐸𝐸𝑖𝑖𝑖𝑖 � Chi-square (r – 1) (c – 1)
2
independence 𝑋𝑋 = � distributed
𝐸𝐸𝑖𝑖𝑖𝑖
𝑖𝑖 = 1
Steps in Hypothesis Testing
Step 3: Specify a level of significance
• A hypothesis test is always conducted at a particular level of significance (α).
• The level of significance represents the chance that we are willing to take that the conclusion
from the test might be wrong.
• The most common level of significance are 10%, 5%, and 1%.

Step 4: State the decision rule

• The decision as to whether reject or fail to reject the null hypothesis is obtained by comparing the
calculated test statistic with the critical value.
• The critical value is based on the level of significance chosen and probability distribution of the
test statistics.
Identifying rejection region
2 sided 1 sided (Ha >) 1 sided (Ha <)

Reject if: Reject if: Reject if:

• Test statistics > upper • Test statistics > upper • Test statistics < upper
critical value or critical value critical value
• Test statistics < lower
critical value
Steps in Hypothesis Testing
Step 5: Collect the data and calculate the test statistics
• Collect the data through unbiased sampling procedure and cleanse it.
• Calculate the test statistics using formula mentioned in step 2 using the data from the sample.

Step 6: Make a decision

• Statical decision: reject or fail to reject the null based on the decision rule specified in step 4.
• Economic decision: consider a range of non-statistical considerations, such as the investor’s
tolerance for risk and financial position are considered before executing the statistical decision.
Example: A successfully reject the null and proves that an investment offers a return of 0.50%. The
investment has a transaction cost of 0.70% to execute. Explain the statistical and economic
significance.
Solution
As the researcher successfully rejects the null, the result is statistically significant however it is not
economically significant as after considering the cost of the strategy, the profits are negative (0.5% -
0.7% = -0.20%)
Key point: Even if we conclude that a strategy's results are economically significant, we should
examine whether there is a logical reason to explain the apparently significant returns offered by the
strategy before actually implementing it.
Testing Single Mean
Example: Suppose that the basketball player's average score in a sample of 49 games is 36 points with
a standard deviation of 9 points. Determine the accuracy of the statement that his career scoring
average is greater than 30 points. Use the 5% level of significance.
Step 1 State the hypothesis H0 : 𝜇𝜇 ≤ 30 This is a one-sided (tailed) test.
Ha : 𝜇𝜇 > 30
Step 2 Identify the appropriate This is test of single mean.
test statistics Given the sample size is large (≥ 30), we can apply the z statistics
given the central limit theorem.
Step 3 Specify the level of 5% (given in the question).
significance
Step 4 State the decision rule To find critical value, look at the corresponding z value to a 0.95
(1 – α) in the z table.
Critical value = 1.65.

Decision rule: Reject if the test statistics is more than 1.65.

Memorize: z value (one sided): 1% = 2.33, 5% = 1.65, 10% = 1.28

Step 5 Collect the data and 𝑋𝑋� − 𝜇𝜇 36 − 30
𝑧𝑧 = 𝑠𝑠 = = 4.67
calculate the test 9
statistics √𝑛𝑛 √49
Step 6 Make a decision Reject the null as the test statistic (4.67) > critical value (1.65).

Conclusion: as we reject the null, we can say that the career

scoring average is greater than 30.
Testing Single Mean
Example: Given that over a sample of 49 games, the player averaged 33 points with a standard
deviation of 9 points, test whether his career scoring average is different from 30 at the 5%
significance level.
Step 1 State the H0 : 𝜇𝜇 = 30 This is a two-sided (tailed) test.
hypothesis Ha : 𝜇𝜇 ≠ 30
Step 2 Identify the This is test of single mean.
appropriate test Given the sample size is large (≥ 30), we can apply the z statistics given
statistics the central limit theorem.
Step 3 Specify the level 5% (given in the question).
of significance
Step 4 State the decision To find critical value, look at the corresponding z value to a 0.975 (1 –
rule α/2) in the z table.
Critical value = ±1.96.

Decision rule: Reject if the test statistics is more than 1.96 or less than -
1.96.

Memorize: z value (two sided): 1% = 2.58, 5% = 1.96, 10% = 1.65

Step 5 Collect the data 𝑋𝑋� − 𝜇𝜇 33 − 30
𝑧𝑧 = 𝑠𝑠 = = 2.33
and calculate the 9
test statistics √𝑛𝑛 √49
Step 6 Make a decision Reject the null as the test statistic (2.33) > upper critical value (1.96).

Conclusion: as we reject the null, we can say that the career scoring
average is different from 30.
Testing Two Means
Population Relationship Assumptions regarding Type of Test
Distribution between Samples Population Variance
Normal Independent Equal t-test pooled variance
Normal Dependent Not applicable t-test paired comparison

Testing Two Means | Independent Samples | Difference b/w Means

Example: Given the below information if the difference between returns for FTSE was significant
across 2 decades at 10% level of significance.

Decade Number of months Mean monthly return (%) Standard deviation (%)
1950s 110 0.680 5.598
1960s 110 1.570 5.738
Step 1 State the hypothesis H0 : 𝜇𝜇1 − 𝜇𝜇2 = 0 This is a two-sided
Ha : 𝜇𝜇1 − 𝜇𝜇2 ≠ 0 (tailed) test.
Step 2 Identify the appropriate test This is test of two means and thus we shall use t test
statistics statistics.
Step 3 Specify the level of 10% (given in the question).
significance
Step 4 State the decision rule To find critical value, look at the at 10% column under
two tailed heading or 5% column under one tailed
heading at a degree of freedom of 218 (110 + 110 - 2)
Critical value = ±1.653.

Decision rule: Reject if the test statistics is more than

1.653 or less than -1.653.
Testing Two Means | Independent Samples | Difference b/w Means
Step 5 Collect the data and (𝑋𝑋�1 − 𝑋𝑋�2 ) − (𝜇𝜇1 − 𝜇𝜇2 )
𝑡𝑡 =
calculate the test statistics 𝑠𝑠 2 𝑠𝑠 2
� 𝑃𝑃 + 𝑃𝑃
𝑛𝑛1 𝑛𝑛2

(0.68 − 1.57 ) − 0
𝑡𝑡 = = −1.165
�32.1311 + 32.1311
110 110

Where
(𝑛𝑛1 − 1)𝑠𝑠 2 1 + (𝑛𝑛2 − 1)𝑠𝑠 2 2
𝑠𝑠 2 𝑃𝑃 =
𝑛𝑛1 + 𝑛𝑛2 − 2

(110 − 1) × 5.5982 + (110 − 1) × 5.7882

𝑠𝑠 2 𝑃𝑃 =
110 + 110 − 2

𝑠𝑠 2 𝑃𝑃 = 32.311

Step 6 Make a decision Fail to reject the null as the test statistic (-1.165) falls in-
between upper (1.653) and lower critical value (-1.653).

Conclusion: as we fail to reject the null, we can say that

the return between decades are not different.
Testing Two Means | Dependent Samples | Mean Difference
Example: A researcher is trying to ascertain whether there is a difference in the annual returns on
two investment strategies. Specifically, he is analyzing the return on a buy-and-hold strategy on the
FTSE versus the return on the top ten dividend yielding stocks on the FTSE over the last 40 years.
Determine whether the returns on the two strategies are different at the 1% level of significance.

Strategy Mean Return Standard Deviation

Buy and hold entire FTSE 17.66% 19.50%
Top 10 dividend yielding 14.71% 15.54%
stocks on FTSE
Difference 2.95% 6.71%
(sample standard deviation of differences)
Step 1 State the hypothesis H0 : 𝜇𝜇𝑑𝑑 = 0 This is a two-sided (tailed) test.
Ha : 𝜇𝜇𝑑𝑑 ≠ 0
Step 2 Identify the appropriate The paired comparisons test must be used because both the
test statistics populations that we are studying (various stocks listed on the
FTSE) are influenced by economic conditions in the UK, and are
therefore, not independent. The t test statistics shall be used.
Step 3 Specify the level of 1% (given in the question).
significance
Step 4 State the decision rule To find critical value, look at the at 1% column under two tailed
heading or 0.50% column under one tailed heading at a degree
of freedom of 39 (40 - 1)
Critical value = ±2.704.

Decision rule: Reject if the test statistics is more than 2.704 or

less than -2.704.
Step 5 Collect the data and 𝑑𝑑̅ − 𝜇𝜇𝑑𝑑0 2.95 − 0
𝑡𝑡 = = = 2.78054
calculate the test 𝑠𝑠𝑑𝑑� 6.71
statistics √40
Step 6 Make a decision Reject the null as the test statistic (2.78) > critical value (2.704).
Testing Single Variance
Example: Test the accuracy of a claim made by ZX Associates that the investment strategies they
follow result in a standard deviation of monthly returns of 5%. Use the 5% level of significance. ZX
performance data for the last 25 months has a standard deviation of 5.2%.
Step 1 State the hypothesis H0 : 𝜎𝜎 2 = 0.0025 This is a two-sided (tailed) test.
2
Ha : 𝜎𝜎 ≠ 0.0025
Step 2 Identify the This is test of single variance and thus chi-square statistics is
appropriate test appropriate.
statistics Three important features of the chi-square distribution are:
• It is asymmetrical.
• It is bounded by zero. Chi-square values cannot be negative.
• It approaches the normal distribution in shape as the degrees
of freedom increase.
Step 3 Specify the level of 5% (given in the question).
significance
Step 4 State the decision rule To find the critical look at 0.025 (α/2) and 0.975 (1 – α/2) column
with 24 (25 -1) degrees of freedom.
Critical value = lower 12.401 and upper 39.364.
Decision rule: Reject if the test statistics is more than 39.364 or
lower than 12.401.
Step 5 Collect the data and 𝑠𝑠 2 (𝑛𝑛 − 1) 0.0522 (25 − 1)
𝑋𝑋 2 = = = 25.96
calculate the test 𝜎𝜎 2 0 0.052
statistics
Step 6 Make a decision Fail to reject the null as the test statistic (25.96) fall between
upper critical value (39.364) and lower critical value (12.401).

Conclusion: as we fail to reject the null, we can say that the

strategies have a monthly standard deviation of 5%.
Testing Equality of Variances
Example: Susan examines two companies. She opines that Company A’s earnings are more volatile
than Company B’s. She obtains earnings data for the past 31 years for Company A, and for the past 41
years for Company B. The sample standard deviation of A's earnings is $4.40 and of B's earnings is
$3.90. Determine if A has a greater standard deviation than B at the 5% level of significance.
Step 1 State the H0 : 𝜎𝜎 2𝐴𝐴 ≤ 𝜎𝜎 2 𝐵𝐵 This is a one-sided (tailed) test.
hypothesis 2
Ha : 𝜎𝜎 𝐴𝐴 > 𝜎𝜎 𝐵𝐵2 Note:
In calculating the F-test statistic, we always put the
greater variance in the numerator so.
Step 2 Identify the This is test of equality of variance and thus F statistics is appropriate.
appropriate test Three important features of the F distribution are:
statistics • It is skewed to the right.
• It is bounded by zero on the left.
• It is defined by two separate degrees of freedom.
Step 3 Specify the level 5% (given in the question).
of significance
Step 4 State the decision To find the critical look 5% significance table and take degree of
rule freedom 40 (41 – 10) along the left most column and degree of freedom
30 (31 – 1) along the top row.
Critical value = 1.74
Decision rule: Reject if the test statistics is more than 1.74.
Note: The rejection region for an F-test, whether it be one-tailed or
two-tailed, always lies in the right tail.
Step 5 Collect the data 𝑆𝑆1 2 4.402
𝐹𝐹 = 2 = = 1.273
and calculate the 𝑆𝑆2 3.302
test statistics
Step 6 Make a decision Fail to reject the null as the test statistic (1.273) is below critical value
(1.74).

Conclusion: as we fail to reject the null, we cannot say that Company A

have a greater standard deviation than those of Company B.
Hypothesis Testing using Confidence Interval
Example: Construct a 95% confidence interval for the basketball player's career scoring average if,
over a sample of 49 games, he averaged 31 points with a standard deviation of 9 points. Use this
confidence interval to determine whether the player's career average is different from 33 points.
Step 1 State the hypothesis H0 : 𝜇𝜇 = 33 This is a two-sided (tailed) test.
Ha : 𝜇𝜇 ≠ 33
Step 2 Identify the appropriate This is test of single mean.
test statistics Given the sample size is large (≥ 30), we can apply the z statistics
given the central limit theorem.
Step 3 Specify the level of 95% (given in the question).
confidence
Step 4 Create a confidence 𝑠𝑠 𝑠𝑠
𝑋𝑋� − 𝑧𝑧𝛼𝛼� × ≤ 𝜇𝜇0 ≤ � + 𝑧𝑧𝛼𝛼� ×
𝑋𝑋
2 2
interval √𝑛𝑛 √𝑛𝑛
9 9
31 − 1.96 × ≤ 𝜇𝜇0 ≤ 31 + 1.96 ×
√49 √49
28.48 ≤ 𝜇𝜇0 ≤ 33.52
Step 5 Make a decision The hypothesized population mean (33) lies within this range.
Therefore, we fail to reject the null hypothesis at the 5% level of
significance.
Hypothesis Testing using p-values

• The p-value is the smallest level of significance at which the null hypothesis can be rejected.
• It represents the probability of obtaining a critical value that would lead to rejection of the null
hypothesis.
Decision rule
Reject p value < level of significance
Fail to reject p value > level of significance
Example: using a software, an analyst calculates the p value as 2%. At what level of significance from
those given below will the null be rejected?
• 5% level of significance: reject as p value is lower
• 1% level of significance: fail to reject as p value is higher
Type-I, Type-II Errors
Decision H0 is true H0 is false
Do not reject H0 Correct decision Incorrect decision

Type 2 error
Reject H0 Incorrect decision Correct decision

Type 1 error Power of test = 1 – P (Type 2 error)

Significance level = P (Type 1 error)

• The higher the power of the test, the better it is for purposes of hypothesis testing.
• Decreasing the significance level reduces the probability of Type I error but increase the
probability of Type 2 error and thus reduces the power of test.
• The Type 2 error can be decrease by increasing the significance level, but this would increase
probability of Type 1 error.
• The only way to decrease Type 2 error at a given level of significance is to increase the sample
size.
 Level I Quantitative Methods
of the PARAMETRIC AND NON-PARAMETRIC
CFA® Program TESTS, TEST OF INDEPENDENCE
Learning Outcome Statements

LOS: 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.
LOS: Explain tests of independence based on contingency table data.
Parametric vs Non-Parametric Tests
Parametric test Has at least one of the following two characteristics:
• It is concerned with parameters or defining features of a
distribution.
• It makes a definite set of assumptions.
Non-parametric test A non-parametric test is not concerned with a parameter and makes
only a minimal set of assumptions regarding the population.
Non-parametric tests are used when:
• The researcher is concerned about quantities other than the
parameters of the distribution (is the sample random?).
• The assumptions made by parametric tests cannot be supported.
• When the data available is ranked.
• When there are outliers.
Non-Parametric Parametric Non-parametric
Alternative to Parametric Single mean t or z test Wilcox signed ranked test
tests Difference between t test Mann-Whitney U test
means
Mean differences t test Wilcox signed ranked test
Sign test
Parametric Test of Correlation
Example: Suppose we determined that the correlation coefficient between money supply growth and
inflation during the period 1990 to 2010 for the six countries studied was 0.9573. Test the null
hypothesis that the true population correlation coefficient equals 0 at the 5% significant level.
Step 1 State the hypothesis H0 : 𝜌𝜌 = 0 This is a two-sided (tailed) test.
Ha : 𝜌𝜌 ≠ 0
Step 2 Identify the appropriate This is a test of correlation and thus t-test is appropriate.
test statistics
Step 3 Specify the level of 5% (given in the question).
significance
Step 4 State the decision rule To find critical value, look at the at 5% column under two tailed
heading or 2.5% column under one tailed heading at a degree of
freedom of 4 (6 - 2)
Critical value = ±2.776.
Decision rule: Reject if the test statistics is more than 2.776 or
less than -2.776.
Step 5 Collect the data and 𝑟𝑟√𝑛𝑛 − 2 0.9573√6 − 2
𝑡𝑡 = = = 6.623
calculate the test √1 − 𝑟𝑟 2 √1 − 0.95732
statistics
Step 6 Make a decision Reject the null as the test statistic (6.623) > upper critical value
(2.776).

Conclusion: as we reject the null, we can say that the correlation

is different from 0.
Non-Parametric Test of Correlation
When the population under consideration is not normally distributed, we can use a test based on the
Spearman rank correlation coefficient, rS.
• The Spearman rank correlation coefficient is essentially equivalent to the usual correlation
coefficient but is calculated on the ranks of the two variables within their respective samples.
Step to calculate rS
• Rank the observations on X from largest to smallest. Rank the observations on Y from largest to
smallest. If two values are tied assign the two the average of the ranks that they occupy.
• Calculate the difference, di, between the ranks for each pair of observations on X and Y, and then
calculate di2 (the squared difference in ranks).
• With a sample size, n, the Spearman rank correlation is calculated as:
6 ∑𝑛𝑛𝑖𝑖=1 𝑑𝑑𝑖𝑖 2
𝑟𝑟𝑠𝑠 = 1 −
𝑛𝑛(𝑛𝑛2 − 1)
Example: An analyst is examining the relationship between returns for two investment funds, A and
B, of similar risk over 30 years. A partial table is provided below:

Test the hypothesis that the correlation amongst the returns on the funds is different from zero at 5%
significance level.
Non-Parametric Test of Correlation
Step 1: state the hypothesis H0 : rS = 0
Ha : rS ≠ 0
Step 2: calculate the spearman 6 𝑥𝑥 1,982
𝑟𝑟𝑠𝑠 = 1 − = 0.6397
correlation 32(322 − 1)
Step 3: calculate t statistics. 𝑟𝑟√𝑛𝑛 − 2 0.6397√32 − 2
𝑡𝑡 = = = 4.522
(as n > 30) √1 − 𝑟𝑟 2 √1 − 0.63972
Step 4: find the critical value For a two-tailed test with a 5% significance level, the critical
values for n − 2 = 32 − 2 = 30 degrees of freedom are ±2.042.
Step 5: make the decision Since the test statistic (4.522) is greater than the upper critical
value (2.042) for this two-tailed test, we reject the null.

Conclusion: there is sufficient evidence to indicate that the

correlation between the returns of Fund A and Fund B is different
from zero.
Non-Parametric Test of Independence | Contingency Table
Example: An analyst compiles data on 1,594 exchange
traded funds (ETFs) and classifies them based on market
capitalization (size) and investment type. She wants to
test whether there is a relationship between the size
and investment at 5% significance level. Table X below is
a contingency table (or two-way) table showing the
frequency of these classifications in the ETFs.
Step 1 State the hypothesis H0 : 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑎𝑎𝑎𝑎𝑎𝑎 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑎𝑎 𝑛𝑛𝑛𝑛𝑛𝑛 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟
Ha : 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑎𝑎𝑎𝑎𝑎𝑎 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑎𝑎 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟
Step 2 Identify the When faced with categorical or discrete data, we must use the
appropriate test chi-square test to check whether there is a correlation between
statistics the variables.
Step 3 Specify the level of 5% (given in the question).
significance
Step 4 State the decision rule Degree of freedom = (r – 1) (c – 1) = (3 – 1) (3 – 1) = 2
With a 5% level of significance, the critical value is 9.4877.
Non-Parametric Test of Independence | Contingency Table
Example: An analyst compiles data on 1,594 exchange
traded funds (ETFs) and classifies them based on market
capitalization (size) and investment type. She wants to
test whether there is a relationship between the size
and investment at 5% significance level. Table X below is
a contingency table (or two-way) table showing the
frequency of these classifications in the ETFs.
Step 5 Collect the data and 𝑋𝑋 2 (𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑖𝑖 𝑥𝑥 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑗𝑗 )
𝑚𝑚 2 𝐸𝐸𝑖𝑖𝑖𝑖 =
calculate the test �𝑂𝑂𝑖𝑖𝑖𝑖 − 𝐸𝐸𝑖𝑖𝑖𝑖 � 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇
statistics =�
𝐸𝐸 𝑖𝑖𝑖𝑖
𝑖𝑖 = 1
Expected Frequency

Scaled Squared
Deviation

The total of all the

scaled squared
deviations in the
table above gives us
chi-square test
statistic, which is
32.079.
Step 6 Make a decision Since our computed chi-square test statistic (32.079) is greater
than critical value (9.4877), we reject the null hypothesis
Conclusion: ETF size and investment type are related (i.e., not
independent).
Sometimes standard residuals are also computed for each of the cells in the contingency table.
Oij − Eij
Standardized (pearson) residual =
�Eij
Standard residual for large-cap, growth ETFs = (202 − 244.536) / 244.5360.5 = −2.72
The (negative) standard residual for large-cap growth ETFs suggests that there are fewer such ETFs
than would be expected if size and investment type were independent.
 Level I Quantitative Methods
of the SIMPLE LINEAR REGRESSION
CFA® Program
Learning Outcome Statements

LOS: Describe a simple linear regression model, how least squares criterion is
used to estimate regression coefficients and interpretation of these coefficients.
LOS: Explain the assumptions underlying the simple linear regression model, and
describe how residuals and residual plots indicate if these assumptions may have
been violated.
LOS: Calculate and interpret the measures of fit and evaluate tests of fit and of
regression coefficients in a simple linear regression
LOS: 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.
LOS: 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.

LOS: Describe different functional forms of simple linear regressions.

Introduction to Linear Regression
Definition Linear regression is used to summarize the relationship between two variables
that are linearly related.
Applications • Make predictions about the dependent variable (scores in the exam) using
an independent variable (hours of study).
• Explain the variation (squared deviation from the mean) in the dependent
variable in terms of the variation in the independent variable.
• To test hypotheses regarding the relation between the two variables, and
to evaluate the strength of the relationship between them.

Simple Linear Regression

Variation in Data
• Suppose we want the examine the performance of students in a national level math Olympiad
and we have observed the following scores with respect to six candidates:
Student Scores (Y)
1 30
2 20
3 75
4 95
5 50
6 90

• In order to understand the different in scores, we have two choices:

1. Compare each pair of scores and analyze the reason for difference.
2. Understand why each score differs from the mean.

• The second method describes computing the variation from the mean and explaining the reasons
for deviation from the mean.
𝑛𝑛

�(𝑌𝑌𝑖𝑖 − 𝑌𝑌�)2
𝑖𝑖=1
• The variation of Y is often referred to as sum of squared totals (SST).
Example: Compute the variation of Y for the above data?
Solution
Mean

Variation
Simple Linear Regression
Introducing Independent Variable
• Let’s evaluate if it’s possible to explain the variation in scores using another variable.
• We believe that the number of hours of effort in studies impacts the score and thus we have now
extended the table to include the hours put in by thhe students in 6 months prior to the test:
Student Scores (Y) Hours (X)
1 30 140
2 20 110
3 75 220
4 95 350
5 50 170
6 90 300
• It’s possible to observe a positive relationship between the hours and scores.
• The scores are dependent variable and hours is independent variable.
Example: Compute the variation of X for the above data?
Solution
Mean

Variation

Scatter Plot
• To visually observe the relationship, we can plot the same on the scatter plot.

Scatter Plot
100
300, 90 350, 95
80
220, 75
60
Scores

170, 50
40
140, 30
20 110, 20
0
0 100 200 300 400
Hours

• Vertical axis is the dependent varibale and horizontal axis is the independent variable.
• Each on the scatter plot repesents a paired observation that consists of scores and hours.
Simple Linear Regression
Establishing a Linear Relationship
• A common method for relating the dependent and independent variables is through the
estimation of a linear relationship, which implies describing the relation between the two
variables as represented by a straight line.
• The following linear regression model describes the relation between the dependent and the
independent variables.
𝑌𝑌𝑖𝑖 = 𝑏𝑏0 + 𝑏𝑏1 𝑋𝑋𝑖𝑖 + 𝜀𝜀𝑖𝑖
𝑏𝑏0 & 𝑏𝑏1 are the regression coefficients.
𝑏𝑏1 is the slope coefficient.
𝑏𝑏0 is the intercept term.
𝜀𝜀𝑖𝑖 is the error term that represents the variation in the dependent variable that is not explained
by the independent variable.
Estimating a Regression Line
• We cannot observe the population parameter values 𝑏𝑏0 and 𝑏𝑏1 in a regression model and instead
we observe, 𝑏𝑏�𝑜𝑜 and 𝑏𝑏�1 , which are estimates of population parameter using sample.
• The regression process estimates the line of best fit from the data in the sample.
• The regression line equation with one independent variable takes the following form:
𝑌𝑌�𝑖𝑖 = 𝑏𝑏�𝑜𝑜 + 𝑏𝑏�1 𝑋𝑋𝑖𝑖
𝐶𝐶𝐶𝐶𝐶𝐶(𝑋𝑋, 𝑌𝑌) ∑𝑖𝑖=1(𝑌𝑌𝑖𝑖 − 𝑌𝑌�)(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)
𝑛𝑛
𝑏𝑏�1 = =
𝑉𝑉𝑉𝑉𝑉𝑉(𝑋𝑋) ∑𝑛𝑛𝑖𝑖=1(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)2

𝑏𝑏�𝑜𝑜 = 𝑌𝑌� − 𝑏𝑏�1 𝑋𝑋�

Line of Best Fit

120
y = 0.3228x - 9.3941
100 R² = 0.9292
80
Scores

60
40
20
0
0 100 200 300 400
Hours

• In simple linear regression, the estimated intercept, 𝑏𝑏�𝑜𝑜 and 𝑏𝑏�1 , are such that the sum of the
squared vertical distances from the observations to the fitted line is minimized.
𝑛𝑛 𝑛𝑛 𝑛𝑛
2 2
𝑆𝑆𝑆𝑆𝑆𝑆 = ��𝑌𝑌𝑖𝑖 − �𝑌𝑌�𝑖𝑖 �� = ��𝑌𝑌𝑖𝑖 − �𝑏𝑏�𝑜𝑜 + 𝑏𝑏�1 𝑋𝑋𝑖𝑖 �� = �[𝜀𝜀𝑖𝑖 ]2
𝑖𝑖=1 𝑖𝑖=1 𝑖𝑖=1
Simple Linear Regression
Simple Linear Regression
Example
Develop a linear regression equation using the below information provided:
Student Scores (Y) Hours (X)
1 30 140
2 20 110
3 75 220
4 95 350
5 50 170
6 90 300
Solution
Student Scores (Y) Hours (X) � )𝟐𝟐 (𝑿𝑿𝒊𝒊 − 𝑿𝑿
(𝒀𝒀𝒊𝒊 − 𝒀𝒀 � )(𝑿𝑿𝒊𝒊 − 𝑿𝑿
� )𝟐𝟐 (𝒀𝒀𝒊𝒊 − 𝒀𝒀 �)
1 30 140
2 20 110
3 75 220
4 95 350
5 50 170
6 90 300
Sum

Average

Y X
Variance

Std. Dev.

Covariance (X,Y)

Correlation (X,Y)
Simple Linear Regression
Example

Slope

Intercept

Regression Equation

Predicted value if a student

studies for 300 hours?
Simple Linear Regression
Example
Determine the slope coefficient and the intercept term of a simple linear regression using the below
data. The data is obtained from analysis of money supply growth as the independent variable and the
inflation rate as the dependent variable over 6 countries. Also state the regression line equation.

Regression line equation

Slope Coefficient

Intercept

Final relationship

Interpretation
Assumptions in Linear Regression
Linearity • The relationship between the dependent variable, Y, and the independent
variable, X, is linear.
Homoskedasticity • The variance of the regression residuals is the same for all observations.
Independence • The observations, pairs of Ys and Xs, are independent of one another. This
implies the regression residuals are uncorrelated across observations.
Normality • The regression residuals are normally distributed.

Assumptions in Linear Regression

n

𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐓𝐓 𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕𝐕 (𝐒𝐒𝐒𝐒𝐒𝐒) = �(Yi − �

Y)2
i=1
Explained variation Unexplained variation
Regression sum of squares (RSS) Sum of squared errors or residuals (SSE)
𝑛𝑛 + 𝑛𝑛
2 2
��𝑌𝑌� − 𝑌𝑌�� ��𝑌𝑌𝑖𝑖 − 𝑌𝑌��
𝑖𝑖=1 𝑖𝑖=1

Alterantive Graphical Presentation

Measures of Goodness of Fit – Standard Error of Estimate (SEE)
Concept • Also known as the standard error of the regression, is used to measure how well a
regression model captures the relationship between the two variables.
• SEE measures the standard deviation of the residual term in the regression.
• The smaller the standard deviation of the residual term (the smaller the standard
error of estimate), the more accurate the predictions based on the model.
Formula 2 1⁄2 1⁄2
∑𝑛𝑛𝑖𝑖=1�𝑌𝑌𝑖𝑖 − �𝑏𝑏�𝑜𝑜 + 𝑏𝑏�1 𝑋𝑋𝑖𝑖 �� ∑𝑛𝑛𝑖𝑖=1[𝜀𝜀𝑖𝑖 ]2 𝑆𝑆𝑆𝑆𝑆𝑆 1⁄2
𝑆𝑆𝑆𝑆𝑆𝑆 = � � =� � =� �
𝑛𝑛 − 2 𝑛𝑛 − 2 𝑛𝑛 − 2

Measures of Goodness of Fit – R-squared, Coefficient of Determination

Concept • The coefficient of determination (R2) tells us how well the independent variable
explains the variation in the dependent variable.
• It measures the fraction of the total variation in the dependent variable that is
explained by the independent variable.
Formula Method 1 𝑅𝑅 2 = 𝑟𝑟 2
Where r is correlation coefficient.
Method 2 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 𝑅𝑅𝑅𝑅𝑅𝑅 𝑆𝑆𝑆𝑆𝑆𝑆 − 𝑆𝑆𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆
𝑅𝑅 2 = = = =1 −
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 𝑆𝑆𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆
Analysis of Variance (ANOVA) table
Analysis of variance (ANOVA) is a statistical procedure that is used to determine the usefulness of the
independent variable(s) in explaining the variation in the dependent variable.
Source of Variation Degrees of Freedom Sum of Squares Mean Sum of Squares
Regression (explained) k RSS 𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑅𝑅𝑅𝑅
𝑀𝑀𝑀𝑀𝑀𝑀 = = = 𝑅𝑅𝑅𝑅𝑅𝑅
𝑘𝑘 1
Error (unexplained) n – (k + 1) SSE 𝑆𝑆𝑆𝑆𝑆𝑆
𝑀𝑀𝑀𝑀𝑀𝑀 =
𝑛𝑛 − 2
Total n–1 SST
k = number of slope coefficients
Measures of Goodness of Fit –F Test

• F-test, which tests whether all the slope coefficients in the regression are equal to zero.
• With one independent variable, the F-test basically tests a null hypothesis:
𝐻𝐻0 ∶ 𝑏𝑏1 = 0 / 𝐻𝐻𝑎𝑎 ∶ 𝑏𝑏1 ≠ 0
In order to the hypothesis, we need to compute the F stat and degree of freedom.
F statistics 𝑀𝑀𝑀𝑀𝑀𝑀
𝐹𝐹 − 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 =
𝑀𝑀𝑀𝑀𝑀𝑀
Degree of freedom Numerator = k = 1
Denominator = n – k – 1 = n – 1 - 1 = n - 2
Decision rule The F-test is a one-tailed test. The decision rule for the test is that we reject H0
if F-stat > Fcrit.
Example Application of F-test to determine whether the slope coefficient of the
regression equals 0 at the 5% significance level.
• Total variation in dependent variable (inflation) = 0.003094
• Sum of squared errors (unexplained variation) = 0.000259
• Number of observations = 6
• Number of independent variables = 1
• The F-critical value with 1 and 4 degrees of freedom and a 5% significance
level is 7.7086.

State the hypothesis:

Calculation of F stat:
Source Sum of squares Degree of Mean Square F-stat
Freedom
Regression

Error

Total

Decision:

Key point • For a one-independent variable regression, the F-stat is basically the same
as the square of the t-stat for the slope coefficient.
• Since it duplicates the t-test for the significance of the slope coefficient for
a one-independent variable regression, analysts only use the F-stat for
multiple-independent variable regressions.
Measures of Goodness of Fit –F Test

• F-test, which tests whether all the slope coefficients in the regression are equal to zero.
• With one independent variable, the F-test basically tests a null hypothesis:
𝐻𝐻0 ∶ 𝑏𝑏1 = 0 / 𝐻𝐻𝑎𝑎 ∶ 𝑏𝑏1 ≠ 0
In order to the hypothesis, we need to compute the F stat and degree of freedom.
F statistics 𝑀𝑀𝑀𝑀𝑀𝑀
𝐹𝐹 − 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 =
𝑀𝑀𝑀𝑀𝑀𝑀
Degree of freedom Numerator = k = 1
Denominator = n – k – 1 = n – 1 - 1 = n - 2
Decision rule The F-test is a one-tailed test. The decision rule for the test is that we reject H0
if F-stat > Fcrit.
Example Application of F-test to determine whether the slope coefficient of the
regression equals 0 at the 5% significance level.
• Total variation in dependent variable (inflation) = 0.003094
• Sum of squared errors (unexplained variation) = 0.000259
• Number of observations = 6
• Number of independent variables = 1
• The F-critical value with 1 and 4 degrees of freedom and a 5% significance
level is 7.7086.

State the hypothesis:

Calculation of F stat:
Source Sum of squares Degree of Mean Square F-stat
Freedom
Regression

Error

Total

Decision:

Key point • For a one-independent variable regression, the F-stat is basically the same
as the square of the t-stat for the slope coefficient.
• Since it duplicates the t-test for the significance of the slope coefficient for
a one-independent variable regression, analysts only use the F-stat for
multiple-independent variable regressions.
Hypothesis Test of Regression Parameters | Slope Coefficient
t-distributed test statistic is used to test hypotheses that a regression coefficient is different from a
specific value or whether the slope is positive.
t test statistics 𝑏𝑏�1 − 𝐵𝐵1
𝑡𝑡 =
𝑠𝑠𝑏𝑏�1
Where
𝑆𝑆𝑒𝑒
𝑠𝑠𝑏𝑏�1 =
�∑𝑛𝑛𝑖𝑖=1(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)2
𝑠𝑠𝑏𝑏�1 = standard error of the slope coefficient
𝑆𝑆𝑒𝑒 = standard error of the estimate
(X i − � X)2 = variation of the independent variable
Critical value n − k − 1 or n − 2 degrees of freedom
Example Use the t-test to determine whether the slope coefficient of the regression
equals 0 at the 5% significance level given the below information:
• Variation of the independent variable = 0.004921
• Standard error of estimate = 0.0085
• Estimated value of the slope coefficient = 0.7591
• n=6
• Critical value: With 4 degrees of freedom at the 5% significance level, the
critical t values are −2.776 and +2.776.

State the hypothesis

Calculation of t stat:
Calculation of standard error of
slope coefficient

Calculation of t stat

Decision:
Hypothesis Test of Regression Parameters | Intercept
t-distributed test statistic can also be used to test hypotheses as to whether the population intercept
is a specific value
t test statistics 𝑏𝑏�0 − 𝐵𝐵0
𝑡𝑡 =
𝑠𝑠𝑏𝑏�0
Where
1 𝑋𝑋� 2
𝑠𝑠𝑏𝑏�0 = � + 𝑛𝑛
𝑛𝑛 ∑𝑖𝑖=1(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)2
𝑠𝑠𝑏𝑏�1 = standard error of the slope coefficient
𝑆𝑆𝑒𝑒 = standard error of the estimate
(X i − � X)2 = variation of the independent variable
Critical value n − k − 1 or n − 2 degrees of freedom
Example Use the t-test to determine whether the intercept is less than 0 at the 5%
significance level. given the below information:
• Variation of the independent variable = 0.004921
• Mean value of independent variable = 0.1012
• Estimated value of the intercept = -0.005
• n=6
• Critical value: The t-critical value for this one-tailed test at the 5% level of
significance with 4 degrees of freedom is −2.132.
State the hypothesis

Calculation of t stat:
Calculation of standard error of
slope coefficient

Calculation of t stat

Decision:
Linear Regression using Indicator (Dummy) Variable

Indicator variables in regression models help analysts determine whether a particular qualitative
variable explains the variation in the model's dependent variable to a significant extent.
Example: A study is conducted to see if the return on a stock over a particular month is influenced
significantly by the company's quarterly earnings announcement.
𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖 = 𝑏𝑏0 + 𝑏𝑏1 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸1 + 𝜀𝜀𝑖𝑖
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸1 = value of 0 if there is no earnings announcement that month and 1 if there is an earnings
announcement.
The table shown here presents the results of a regression of 30 months of returns data with EARN,
the return during a month when there was an earnings announcement, as the indicator variable.
Regression Estimation Results
Source Estimated coefficients Standard error of Calculated t Statistics
coefficients
Intercept 0.5629 0.0560 10.0596
Earn 1.2098 0.1158 10.4435
Degree of freedom: 28 / Critical value = ±2.0484 (5% level of significance)

Analysis
• The intercept term indicates the average return for a non-earnings announcement month
(0.5629).
• The slope coefficient indicates the average difference in returns between an earnings
announcement month and a non-earnings announcement month (1.2098).
• The t-statistic (10.4435) enables us to reject the null hypothesis (at the 5% significance level) that
the slope coefficient on EARN is zero (since it is greater than 2.0484) and thus conclude that the
return during a month where earnings are announced is significantly different from the return
during a month with no earnings announcement.
• The t-statistic on the intercept (10.0596) allows us to reject the null hypothesis that the intercept
is zero at the 5% significance level.
Linear Regression using Indicator (Dummy) Variable

Regression results can also enable us to test the hypothesis that the return is the same across
earnings and non-earnings months.
𝐻𝐻0 : 𝜇𝜇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 = 𝜇𝜇𝑛𝑛𝑛𝑛𝑛𝑛 −𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 / 𝐻𝐻0 : 𝜇𝜇𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ≠ 𝜇𝜇𝑛𝑛𝑛𝑛𝑛𝑛 −𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
The table presented here shows the results of this hypothesis test.
Test of Differences in Means
Ret. – earning Ret. – non-earning Difference in mean
announcement month announcement month
Mean 1.7727 0.5629 1.2098
Variance 0.1052 0.0630
Observations 7 23
Pooled variance 0.07202
Calculated t statistics 10.4435
Degree of freedom: 28 / Critical value = ±2.0484 (5% level of significance)

Since the calculated t-statistic (10.4435) exceeds the critical value (2.0484), we reject the null
hypothesis that there is no difference in the mean RET for the earnings-announcement and non-
earnings-announcement months at the 5% level of significance.

Level of Significance and p-levels of the Hypothesis Tests

Level of • A lower level of significance increases the absolute critical value resulting in a
significance wider confidence interval and a lower likelihood of rejecting the null hypothesis.
• Increasing the significance level increases the probability of a Type I error but
decreases the probability of a Type II error.
p value • The p-value is the lowest level of significance at which the null hypothesis can be
rejected.
• A p-value of 0.007 tells us that we can reject the null hypothesis that the true
population parameter equals zero at the 0.7% level of significance (or with a
99.3% level of confidence).
Standard • The smaller the standard error of an estimated parameter, the stronger the
error results of the regression and the narrower the resulting confidence intervals.
Prediction and Prediction Intervals
Regression analysis is often used to make predictions or forecasts of the dependent variable based on
the regression equation.
• Typically, analysts construct confidence intervals around the regression forecasts.
Prediction interval
Y�f ± t α� × sf
2
�f = Predicted value
Y
sf = Standard error of forecast
There are two sources of uncertainty when we use a regression model to make a prediction regarding
the value of the dependent variable.
• The uncertainty inherent in the error term, ε.
• The uncertainty in the estimated parameters, b0 and b1.
Standard error of forecast
The estimated variance of the prediction error, of Y is calculated as:
2 2
1 �𝑋𝑋𝑓𝑓 − 𝑋𝑋�� 1 �𝑋𝑋𝑓𝑓 − 𝑋𝑋��
𝑠𝑠 2𝑓𝑓 2
= 𝑠𝑠 𝑒𝑒 �1 + + 2
� = 𝑠𝑠 𝑒𝑒 �1 + + 𝑛𝑛 �
𝑛𝑛 (𝑛𝑛 − 1)𝑠𝑠 2 𝑋𝑋� 𝑛𝑛 ∑𝑖𝑖=1(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)2
The standard error of the forecast is:
2
1 �𝑋𝑋𝑓𝑓 − 𝑋𝑋��
𝑠𝑠𝑓𝑓 = 𝑠𝑠𝑒𝑒 �1 + + 𝑛𝑛
𝑛𝑛 ∑𝑖𝑖=1(𝑋𝑋𝑖𝑖 − 𝑋𝑋�)2
Note the following:
• The better the fit of the regression model, the smaller the standard error of the estimate (𝑠𝑠𝑒𝑒 ) and,
therefore, the smaller standard error of the forecast.
• The larger the sample size (n) in the regression estimation, the smaller the standard error of the
forecast.
• The closer the forecasted independent variable (𝑋𝑋𝑓𝑓 ) is to the mean of the independent variable
(𝑋𝑋�) used in the regression estimation, the smaller the standard error of the forecast.
Prediction and Prediction Intervals
Example: Determine the 95% prediction interval for the inflation rate based on money supply growth,
given the below information:
• Money supply growth rate is 9%
• Intercept is -0.005 and slope is 0.7591.
• Mean value of money supply growth is 10.12%
• Standard deviation of money supply growth is 3.1373%
• Standard error of estimate as 0.008
• Number of observations is 6.
• The critical t-value for a 95% confidence interval with 4 degrees of freedom is 2.7764.
Standard error
of forecast

Predicted value

Confidence
interval
Different Functional Forms of Regression Models

• Sometimes the relationship between the independent variable and the dependent variable may
not be linear.
• In such cases, we can often transform one or both of these variables to convert this relation to a
linear form, which then allows the use of simple linear regression
Three commonly used functional forms
Log-lin model • The dependent variable is logarithmic, but the independent variable is linear.
• The slope coefficient in this model is the relative change in the dependent
variable for an absolute change in the independent variable.
Lin-log model • The dependent variable is linear, but the independent variable is logarithmic.
• The slope coefficient in this regression model provides the absolute change in
the dependent variable for a relative change in the independent variable.
Log-log model • Both the dependent and independent variables are logarithmic form.
In order to determine the appropriate functional form of a simple linear regression one must examine
the goodness of fit measures:
• Coefficient of determination (R2)—higher is better,
• F-statistic—higher is better, and
• Standard error of the estimate (se)—lower is better.
Further, the residuals should be examined. Ideally, the residuals should be random.
 Level I Quantitative Methods
of the INTRODUCTION TO BIG DATA TECHNIQUES
CFA® Program
Learning Outcome Statements

LOS : Describe aspects of “fintech” that are directly relevant for the
gathering and analyzing of financial data.
LOS : Describe Big Data, artificial intelligence, and machine learning.
LOS : Describe applications of Big Data and Data Science to investment
management.
What is Fintech?

• Technological innovation in the design and delivery of financial services and products.
• It encompasses advanced systems used to analyze information and make decisions based on
machine-learning logic.
• Using such systems has brought about high levels of efficiency that surpass human capabilities.
• Services, and applications of fintech relevant to the investment industry include:
o Analysis of large datasets: Includes traditional data like security prices, financial
statements etc. and alternative data obtained from non-traditional sources like social
media, satellite images, online mentions, sensor networks etc.
o Analytical tools: Artificial Intelligence (AI) identifies complex, non-linear relationships.
AI analyses humungous amounts of data.

What is Big Data?

• Volume: Data represents billions of data points. In the form of tables, files, records.
• Velocity: Accelerated speeds of data recording and transmission. Real-time data is the new norm.
• Variety: Data collected from different sources and formats – structured (ex: SQL tables), semi-
structured (ex: HTML code), unstructured (ex: video messages).

(CFA Curriculum, 2024_L1V1, Pg: 328, Exhibit 1)

Sources of Big Data

• Financial markets: Equity, swaps, futures, options, and other derivatives.

• Businesses: Financial statements, credit card purchases, and commercial transactions.
• Governments: Payroll, economic, trade, employment data, etc.
• Individuals: Product reviews, credit card purchases, social media posts, etc.
• Sensors: Shipping cargo information, traffic data, satellite imagery.
• The Internet of Things: Data generated by ‘smart ‘buildings through fittings such as CCTV
cameras, vehicles, home appliances, etc.

Sources of Alternative Data

Big Data not only draws from traditional sources but also from alternative data sources –
• Data generated by people: This is primarily generated through website clicks and page visits.
• Data produced by commercial operations: This is data left behind by business transactions and
activities including point-of-sale data, banking records from credit cards, corporate exhaust.
It is structured.
• Data produced by sensors: Typically, unstructured and is gathered through satellites,
smartphones, and webcams.

• Investment Professionals must beware of potential legal and regulatory issues. Ex: Scraping of
web data could lead to accessing personal information forbidden by regulators.

Challenges for Big Data

• Quality: Dataset used may have selection bias, missing data or has data outliers.
• Volume: The volume of the data set used may be insufficient.
• Appropriateness: The data may not be appropriate for the intended use.

Alternative data, is usually difficult to source, clean and organize before performing any analysis,
since it is mostly unstructured data.
Artificial intelligence and machine learning can be used to solve this problem.
Artificial Intelligence

• AI systems can perform tasks that traditionally demanded human intelligence.

• AI technology has allowed the development of computers with cognitive and decision-making
capabilities more advanced than humans.
• Examples of AI –
o Expert System: An early version where a computer program attempted to simulate
human knowledge and analytical capabilities using mainly “if-then” rules.
o Neural Networks: Programs that function based on how a human brain learns and
processes information. Used to detect abnormal changes in data, ex: used in credit card
fraud detection systems.
o Machine Learning: Uses computer techniques to extract information from big data
without any assumptions about the underlying probability distributions. Aims to
automate decision making by learning from examples and underlying structure in the
input data.

Machine Learning

• It’s the idea that when exposed to more data, machines can make changes on their own and
produce solutions to problems without reliance on human expertise, improving their
performance over time.
• Requires large amount of data for “training.”
• Computer algorithm is given a set of variables or datasets as inputs OR it may be given the target
data as output.
• The algorithm then learns from the data provided how best to model the inputs to outputs or
how to describe the data structures if no output is provided.
Machine Learning | Types
Supervised Learning
• Computer is given a set of inputs and labeled outputs.
• The computer seeks to describe the data or understand its structure.
• Ex: Categorizing given data into sectors / groups.

Unsupervised Learning
• Computer is given a set of inputs but no outputs.
• Using the provided data, the machine can learn relationships that link inputs to output.
• Based on the above learning, the computer predicts outcomes for new datasets.
• Ex: Identify technical signals, forecast stock returns, predict stock market performance etc.

Deep Learning
• Relates to neural networks.
• Neural networks have many hidden layers, use a lot of non-linear processing to identify patterns.
• Can use both supervised / unsupervised techniques.

A Few Points to Note

• ML requires a lot of human touch to ensure the data processing is free of bias.
• Before going live, machines go through distinct phases.
o Data is divided into 3 subsets: training subset, validation subset, and test subset.
o Training data is where the machine learns from.
o The machine learning predictions are validated using the validation set.
o The final test dataset is what determines whether the machine can make predictions on
new data.
• Overfitting: machine learns with too much precision and discovers unsubstantiated patterns.
o Too complex, the machine treats noise as true parameters.
• Underfitting is the opposite – machine learns too little, misses obvious patterns in data.
o Too simplistic, the machine treats valid relationships as noise.
Tackling Big Data with Data Science
Data Science
• Data science is an interdisciplinary field that uses developments in computer science, statistics,
and other fields to extract information from Big Data.
• Data scientists use various data processing techniques to handle big data.
Data Processing Methods
• Capture: How data is collected and transformed into a format used by the analytical process.
o Low-latency systems operate on networks that communicate high volumes of data with
minimal delay (latency). These are essential for automated trading applications that make
decisions based on real-time prices and market events.
• Curation: Process to clean data and ensure quality and accuracy.
o Involves reviewing all data to detect and uncover data errors—bad or inaccurate data—
and adjusting for missing data when appropriate.
• Storage: Addresses how the data will be recorded, archived, and accessed. Establishes the
optimal underlying database design.
o An important consideration for data storage is whether the data is structured or
unstructured and if analytical needs require low-latency solutions.
• Search: Means to query data. Big Data needs advanced applications to examine and review large
quantities of data to locate requested data content.
• Transfer: Take care of moving data from the underlying data source or storage location to the
underlying analytical tool. This could be through a direct data feed, such as a stock exchange’s
price feed.
Tackling Big Data with Data Science
Data Science
• Data science is an interdisciplinary field that uses developments in computer science, statistics,
and other fields to extract information from Big Data.
• Data scientists use various data processing techniques to handle big data.
Data Processing Methods
• Capture: How data is collected and transformed into a format used by the analytical process.
o Low-latency systems operate on networks that communicate high volumes of data with
minimal delay (latency). These are essential for automated trading applications that make
decisions based on real-time prices and market events.
• Curation: Process to clean data and ensure quality and accuracy.
o Involves reviewing all data to detect and uncover data errors—bad or inaccurate data—
and adjusting for missing data when appropriate.
• Storage: Addresses how the data will be recorded, archived, and accessed. Establishes the
optimal underlying database design.
o An important consideration for data storage is whether the data is structured or
unstructured and if analytical needs require low-latency solutions.
• Search: Means to query data. Big Data needs advanced applications to examine and review large
quantities of data to locate requested data content.
• Transfer: Take care of moving data from the underlying data source or storage location to the
underlying analytical tool. This could be through a direct data feed, such as a stock exchange’s
price feed.

Data Visualisation
• Refers to how data would be formatted, displayed, and summarized in graphical form.
• Traditional data is visualized in tables, charts, trend lines etc.
• Non-traditional data uses new techniques such as 3D graphics. Multidimensional data extending
beyond 3D would demand more data to accurately be visualized.
• Various visualization techniques are used based on underlying data. Ex: Heat maps, Tag (Word)
Clouds, Mind maps etc.
Tackling Big Data with Data Science
Text Analytics and Natural Language Processing (NLP)
• Text Analytics uses computer programs to derive meaning from large unstructured text and voice
data. Ex: From Company filings, earnings calls, emails, surveys etc.
o Lexical Analysis: Word frequency is used to detect patterns based on key words / phrases.
o Predictive Analysis: Identify indicators of future performance like consumer sentiment.
• NLP, within the larger scope of text analysis, is used for translation, speech recognition, text
mining, sentiment analysis, topic analysis etc.
o It uses computer science, AI, and linguistics to interpret human language.
o When aided by ML techniques, NLP is used to analyse annual reports, call transcripts,
news articles etc. with more scale and accuracy than what a human can achieve.
o NLP analyzes nuances to provide insights around trending topics of interest like interest
rate policy, aggregate output, inflation expectations.