CFA Level I
Quantitative Methods
Complete Summary — Readings 1 through 11
Explained simply, concept by concept
Reading 1: Interest Rates & Return Measurement
This reading answers: what is an interest rate, really, and how do we measure investment returns?
Interest Rates as Building Blocks
An interest rate is built from layers. Start with the real risk-free rate (the pure time value of money),
add an inflation premium (to maintain purchasing power), add a default risk premium (to compensate
for the chance the borrower doesn't pay back), add a liquidity premium (for investments that are hard
to sell quickly), and add a maturity premium (for locking up money for longer). Each layer
compensates you for a different type of risk.
Required return = Real risk-free rate + Inflation premium + Default risk +
Liquidity premium + Maturity premium
Measuring Returns
Holding Period Return (HPR): The simplest return measure. It's just (ending value / beginning
value) - 1. If you buy a stock at $100 and sell at $110, your HPR is 10%.
Arithmetic Mean Return: The simple average of a series of returns. Good for estimating the
expected return for a single period, but overstates growth over multiple periods.
Geometric Mean Return: Accounts for compounding. It gives you the actual growth rate that turns
your starting value into your ending value. Always less than or equal to the arithmetic mean. Use this
when measuring historical performance over multiple periods.
Harmonic Mean: Used specifically when averaging ratios like price-to-earnings. Gives less weight to
extreme values. Always the smallest of the three means.
Key Insight: Harmonic mean ≤ Geometric mean ≤ Arithmetic mean. They are only equal when
all values are identical.
Time-Weighted vs. Money-Weighted Returns
Time-weighted return (TWR): Measures the performance of the portfolio manager by linking sub-
period returns. It removes the effect of cash flows (deposits/withdrawals), so it purely reflects
investment skill. This is the industry standard for comparing managers.
Money-weighted return (MWR): The internal rate of return (IRR) of the portfolio, accounting for all
cash flows. It reflects the investor's actual experience. If you add money right before a big gain, your
MWR will be higher than the TWR. If you add money right before a loss, it will be lower.
Key Insight: TWR measures the manager's skill. MWR measures the investor's actual
experience. They differ when cash flows happen at different times.
Reading 2: Discounted Cash Flow Valuation
This reading is about the most fundamental idea in finance: a dollar today is worth more than a dollar
tomorrow.
Present Value and Future Value
Future Value asks: if I invest money today at a given rate, what will it grow to? Present Value asks
the reverse: if I'm going to receive money in the future, what is it worth today? They are two sides of
the same coin, connected by the discount rate and the number of periods.
FV = PV × (1 + r)^n PV = FV / (1 + r)^n
Annuities and Perpetuities
An annuity is a series of equal payments over a fixed period (like loan payments or coupon payments
on a bond). A perpetuity is an annuity that never ends — it pays forever. The present value of a
perpetuity is simply PMT / r, which is beautifully simple.
PV of perpetuity = PMT / r
Compounding Frequency
If a bank quotes you 12% per year compounded monthly, the actual rate you earn is higher than 12%
because interest earns interest within the year. The Effective Annual Rate (EAR) converts any
quoted rate into the true annual rate. More frequent compounding means a higher EAR.
EAR = (1 + APR/n)^n - 1
Cash Flow Additivity
A key principle: you can add or compare cash flows only if they are at the same point in time. To
combine cash flows from different dates, first discount them all to the same date (usually today), then
add them up. This is the foundation of all valuation in finance.
Reading 3: Statistical Measures
This reading gives you the toolkit for describing data: where is the center, how spread out is it, and
what shape does the distribution have?
Measures of Central Tendency
The mean is the arithmetic average — add everything up and divide by the number of observations.
The median is the middle value when data is sorted — it's resistant to extreme values (outliers). The
mode is the most frequent value. For a symmetric distribution, all three are equal. For skewed
distributions, they diverge.
Measures of Dispersion (Spread)
Range: The simplest measure — just max minus min. It's easy but tells you nothing about values in
between.
Variance and Standard Deviation: Variance is the average of squared deviations from the mean.
Standard deviation is the square root of variance, bringing us back to the original units. Standard
deviation is the most commonly used measure of risk in finance. For sample data, divide by (n-1)
instead of n, because you lose one degree of freedom when estimating the mean.
Coefficient of Variation (CV): The ratio of standard deviation to the mean (s/x̄ ). It measures relative
risk, allowing you to compare the riskiness of investments with different average returns.
CV = Standard Deviation / Mean
Shape: Skewness and Kurtosis
Skewness measures asymmetry. Positive skew means the right tail is longer (occasional large
gains). Negative skew means the left tail is longer (occasional large losses). For investors, negative
skew is dangerous because it means the risk of extreme losses is higher than a normal distribution
would suggest.
Kurtosis measures how fat the tails are. Excess kurtosis (kurtosis minus 3) tells you how much fatter
or thinner the tails are compared to a normal distribution. Positive excess kurtosis (leptokurtic)
means fatter tails — more extreme outcomes than expected. This matters enormously for risk
management.
Key Insight: For a positively skewed distribution: mean > median > mode. For a negatively
skewed distribution: mean < median < mode. The mean gets 'pulled' toward the longer tail.
Reading 4: Probability Trees & Conditional Expectations
This reading is about how to quantify uncertainty and update your beliefs when new information
arrives.
Expected Value
The expected value of a random variable is its probability-weighted average. If a stock has a 30%
chance of returning 20%, a 50% chance of returning 12%, and a 20% chance of returning 5%, the
expected return is (0.30 × 20%) + (0.50 × 12%) + (0.20 × 5%) = 13%. It's not a prediction of what will
happen — it's the long-run average if the scenario played out many times.
E(X) = Σ P(xi) × xi
Probability Trees
A probability tree visually maps out all possible scenarios and their probabilities. At each branch
point, you multiply probabilities to get the joint probability of any specific path. Then you can calculate
expected values by weighting the outcome at the end of each path by its joint probability. Probability
trees are especially useful when outcomes depend on multiple sequential events.
Bayes' Formula — Updating Beliefs
Bayes' formula is how you update probabilities when you receive new information. Imagine you
initially believe there's a 60% chance the economy will outperform. Then you observe that a stock
went up. Bayes' formula lets you ask: given that the stock went up, what's the updated probability
that the economy outperformed?
The formula works by comparing how likely the observed evidence is under each scenario. If stock
gains are much more common when the economy outperforms, then observing gains increases your
confidence that the economy did outperform.
Key Insight: Bayes' formula turns your 'prior' probability (before new info) into a 'posterior'
probability (after new info). It's the mathematical foundation for how rational analysts should
update their forecasts.
Reading 5: Portfolio Return and Risk
This reading answers the crucial question: when you combine investments, how do you calculate the
portfolio's expected return and risk?
Portfolio Expected Return
This is simply the weighted average of each asset's expected return. If 70% of your money is in
Stock A (expected return 10%) and 30% is in Stock B (expected return 6%), the portfolio's expected
return is (0.70 × 10%) + (0.30 × 6%) = 8.8%. No surprises here.
Portfolio Variance — Where Diversification Lives
Portfolio risk is NOT a simple weighted average. This is the most important insight in all of portfolio
theory. The formula for a two-asset portfolio includes an interaction term that captures how the two
assets move together (covariance). If they don't move perfectly in sync, the portfolio's risk will be less
than the weighted average of the individual risks.
Var(p) = wA² × σA² + wB² × σB² + 2 × wA × wB × Cov(A,B)
Covariance can also be expressed using correlation: Cov(A,B) = ρ × σA × σB. The lower the
correlation between assets, the greater the diversification benefit. At a correlation of +1.0, there is no
diversification benefit at all. At lower or negative correlations, portfolio risk drops significantly.
Key Insight: Diversification works because the interaction term in the portfolio variance formula
can reduce total risk. This is why 'don't put all your eggs in one basket' has mathematical
backing.
Safety-First Ratio and Shortfall Risk
Shortfall risk is the probability that your portfolio return falls below a minimum threshold. Roy's safety-
first criterion says: choose the portfolio that minimizes this shortfall probability. The safety-first ratio
(SFRatio) measures how many standard deviations your expected return sits above the threshold —
essentially a z-score. Higher is better.
SFRatio = (E(Rp) - RL) / σp
Reading 6: Simulation Methods
This reading is about what to do when the world is too complex for a single formula.
Lognormal Distribution
If returns are normally distributed (can be positive or negative), then prices are lognormally
distributed (always positive). This makes intuitive sense — a stock can drop to $0 but can never go
negative. The lognormal distribution is generated by the function e^x where x is normally distributed.
The key formula is Price(T) = Price(0) × e^r, where r is the continuously compounded return.
Monte Carlo Simulation
Monte Carlo simulation is like running thousands of 'what if' experiments on a computer. You specify
the probability distributions for your risk factors, the computer randomly generates scenarios, you
price the security under each scenario, and then you look at the distribution of all those values. The
average is your estimate of the security's value.
The big advantage: you can test extreme scenarios that have never actually occurred. The big
limitation: results are only as good as your assumptions. Garbage in, garbage out.
Bootstrap Resampling
Bootstrap is the alternative when you don't want to assume a specific distribution. Instead of
generating random values from an assumed distribution, you reshuffle your actual historical data by
drawing samples with replacement. The advantage is that you don't need to make distribution
assumptions. The limitation is that you're confined to what has actually happened — you can't test
truly extreme scenarios that never occurred historically.
Key Insight: Monte Carlo = assumed distributions, can test anything. Bootstrap = real data,
limited to historical range. Both are only as good as their inputs.
Reading 7: Estimation and Inference
This reading is about how to pick good data and what you can learn from it.
Sampling Methods
Simple random sampling gives every item an equal chance of selection. Stratified random sampling
divides the population into subgroups and samples proportionally from each — useful for bond
indexing. Cluster sampling randomly selects entire subgroups. Convenience and judgmental
sampling are cheaper but introduce more sampling error.
Key Insight: More randomness in selection = less sampling error. Stratified sampling guarantees
your sample reflects the population structure.
The Central Limit Theorem
This is arguably the most important theorem in statistics. It says: regardless of what the population
distribution looks like, the distribution of sample means will be approximately normal as long as the
sample size is large enough (n ≥ 30). This is why normal distribution tools work so broadly in
statistics — not because everything is normal, but because averages are.
Standard Error
The standard error of the sample mean tells you how precisely your sample mean estimates the true
population mean. It equals σ/√n (or s/√n when using the sample standard deviation). As sample size
increases, the standard error decreases — more data means more precision.
Standard Error = s / √n
Resampling: Jackknife and Bootstrap
The jackknife calculates multiple means by leaving out one observation at a time — simple and
computationally cheap. Bootstrap draws many complete samples with replacement from the original
data — more powerful but more demanding. Both improve estimates of sampling distributions.
Reading 8: Hypothesis Testing
This reading is about making yes/no decisions based on sample data — like a courtroom trial for
statistical claims.
The Framework
Every test has a null hypothesis (H₀) — the default claim you're trying to disprove — and an
alternative hypothesis (Hₐ) — what you actually believe. You calculate a test statistic that measures
how far your sample data is from what the null claims. If the test statistic exceeds the critical value,
you reject the null. If not, you 'fail to reject' it.
Test statistic = (Sample statistic - Hypothesized value) / Standard error
Type I and Type II Errors
Type I error: rejecting a true null (convicting an innocent person). Its probability is α, the significance
level. Type II error: failing to reject a false null (letting a guilty person go free). Its probability is β. The
power of the test (1 - β) is the probability of correctly rejecting a false null. Lowering α increases β
(and reduces power). The only way to reduce both errors is to increase sample size.
Which Test to Use
What You're Testing Which Test Key Detail
Single mean t-test (or z-test if large n) df = n - 1
Two means, independent Difference in means t-test df = n1 + n2 - 2
samples
Two means, dependent Paired comparisons t-test df = n - 1 (n = number of pairs)
samples
Single variance Chi-square (χ²) test df = n - 1; always positive
Two variances F-test Larger variance on top
Assumptions not met Nonparametric test No distribution assumptions
The p-value
The p-value is the smallest significance level at which you would still reject the null. A p-value of 0.03
means there's only a 3% chance of seeing these results if the null were true. If p-value < α, reject the
null. The p-value lets you see exactly how strong the evidence is, rather than just getting a binary
yes/no answer.
Reading 9: Tests of Independence
This reading asks: are two things related, or are they independent of each other?
Parametric Correlation Test
When you calculate a sample correlation (r) between two variables, the test of whether the true
population correlation (ρ) is zero uses a t-test with n - 2 degrees of freedom. A larger correlation and
a larger sample size both push the test statistic higher, making it easier to reject the null of no
correlation.
t = r × √(n - 2) / √(1 - r²)
Spearman Rank Correlation
The nonparametric alternative. Instead of working with actual values, you convert data into ranks and
measure how well the ranks agree. Useful when data isn't normally distributed or when you only
have ranked information. For samples over 30, the significance test uses the same t-test formula.
Contingency Table Chi-Square Test
Used to test whether two categorical characteristics are independent (e.g., is earnings growth related
to dividend yield?). You compare observed frequencies in each cell to the frequencies you would
expect if the characteristics were truly independent. The test statistic follows a chi-square distribution
with (r-1)(c-1) degrees of freedom.
Expected frequency = (Row total × Column total) / Grand total
Reading 10: Simple Linear Regression
This reading is about finding the mathematical line that best describes the relationship between two
variables.
The Basics
Simple linear regression finds the 'line of best fit' through a scatter plot of data. It explains variation in
a dependent variable (Y — the thing you're predicting) using an independent variable (X — the thing
you're predicting with). For example, using S&P 500 returns (X) to predict an individual stock's
returns (Y).
Y = b₀ + b₁X + ε
The intercept (b₀) is the predicted Y when X is zero. The slope (b₁) is the expected change in Y for a
one-unit change in X. For example, a slope of 0.64 means that for every 1% increase in S&P 500
excess returns, ABC stock's excess return is expected to increase by 0.64%.
Key Insight: The slope of a regression of a stock's excess returns on market excess returns is
the stock's beta — a measure of systematic risk.
How the Line Is Chosen: Ordinary Least Squares (OLS)
The regression line is the one that minimizes the sum of squared errors (SSE) — the squared
vertical distances between the actual Y-values and the predicted Y-values. Squaring ensures
positive and negative errors don't cancel out, and it penalizes large errors more heavily than small
ones.
Key Assumptions
The regression assumes a linear relationship, constant variance of residuals (homoskedasticity),
independence of residuals, and normally distributed residuals. Violations show up in residual plots: a
curve means nonlinearity, a fan shape means heteroskedasticity, and patterns over time suggest
dependence.
ANOVA and Goodness of Fit
ANOVA breaks total variation in Y into two components: the part explained by the regression (SSR)
and the unexplained part (SSE). The R² (coefficient of determination) is SSR/SST — the proportion
of total variation explained by the model. An R² of 0.63 means X explains 63% of the variation in Y.
For simple regression, R² is just the square of the correlation coefficient.
Total Variation (SST) = Explained Variation (SSR) + Unexplained Variation
(SSE)
R² = SSR / SST
Testing Significance
The F-test tests whether the slope is significantly different from zero (i.e., does X actually help
explain Y?). For simple regression, this is equivalent to testing whether the correlation between X
and Y is zero. You can also use a t-test directly on the slope coefficient: t = estimated slope /
standard error of the slope.
Functional Forms
When the relationship between X and Y is not linear, transformations can help. A log-lin model (ln Y
vs X) captures exponential growth. A lin-log model (Y vs ln X) captures diminishing returns. A log-log
model (ln Y vs ln X) captures power relationships. The correct form is chosen by examining the data
and comparing R² values.
Reading 11: Introduction to Big Data Techniques
This reading is conceptual — it introduces the technologies that are transforming how data is used in
finance.
Big Data: Volume, Velocity, Variety
Big Data refers to all potentially useful information generated in the economy, from traditional
sources (market data, financial reports) to nontraditional ones (social media, satellite images, sensor
data). It is characterized by three Vs: volume (sheer amount of data), velocity (how quickly it arrives
— low latency means real-time), and variety (structured data like spreadsheets, semistructured data
like web code, and unstructured data like video).
Artificial Intelligence and Machine Learning
AI refers to computer systems that simulate human cognition. Machine learning is a subset where
algorithms learn patterns from data without being explicitly programmed. In supervised learning, the
algorithm trains on labeled data (known inputs and outputs). In unsupervised learning, it finds
patterns in unlabeled data. Deep learning uses layers of neural networks to identify increasingly
complex patterns.
Two key pitfalls: overfitting (the model learns noise as if it were signal — too complex) and
underfitting (the model misses real patterns — too simple). Machine learning results can also be a
'black box,' producing useful outputs based on relationships that are difficult to explain.
Applications in Finance
Text analytics analyzes unstructured text like regulatory filings. Natural language processing
interprets human language for tasks like compliance monitoring and sentiment analysis. Algorithmic
trading uses predetermined rules for trade execution, including high-frequency trading. Risk
governance uses Big Data techniques for real-time risk monitoring and stress testing.
Master Overview: All Readings at a Glance
Reading Core Question Key Takeaway
1: Interest Rates What is a return and how do we Returns have multiple components; TWR
measure it? for managers, MWR for investors
2: TVM What is money worth at different A dollar today > a dollar tomorrow; discount
points in time? and compound to compare
3: Statistics How do we describe data? Center (mean), spread (std dev), shape
(skew, kurtosis)
4: Probability How do we quantify uncertainty? Expected values, probability trees, and
Bayes' updating
5: Portfolio Math What happens when we combine Portfolio risk < weighted average risk
assets? (diversification works)
6: Simulation What if formulas aren't enough? Monte Carlo (assumed dist.) vs Bootstrap
(real data)
7: Sampling How do we learn from limited data? CLT: sample means are normal for n ≥ 30,
regardless of population
8: Hypothesis How do we make statistical Null vs alternative, Type I/II errors, choose
Testing decisions? the right test
9: Independence Are two things related? Correlation test, Spearman ranks,
contingency tables
10: Regression What's the best line through the OLS minimizes SSE; R² measures
data? explanatory power
11: Big Data How is technology changing AI, machine learning, and data science
finance? applications
Key Insight: The entire Quantitative Methods section builds a logical progression: first you learn
how to measure returns (Readings 1-2), then how to describe and analyze data (Readings 3-4),
then how to combine assets and manage risk (Readings 5-6), then how to draw conclusions from
samples (Readings 7-9), then how to model relationships (Reading 10), and finally how
technology is changing all of this (Reading 11).