Overview
Quantitative Methods is 5-10% of the CFA Level 2 exam and steps up from Level 1's descriptive statistics into building and diagnosing statistical models. It has three areas. Multiple regression and model building cover fitting and interpreting a regression with several independent variables, the coefficients, significance tests, the F-test, R-squared and adjusted R-squared, and using models to forecast. Model misspecification covers the assumptions of regression and what happens when they fail: heteroskedasticity (non-constant error variance), serial correlation (autocorrelated errors) and multicollinearity (correlated independent variables), how to detect them and their consequences. Time-series analysis and machine learning cover trend and autoregressive models, unit roots and stationarity, and an introduction to machine-learning methods used in finance.
The exam rewards interpreting regression output within a vignette, recognizing and diagnosing the assumption violations, and understanding time-series and machine-learning concepts. Though a smaller weight, it contains technically difficult material tested in dense item sets.
How it's tested
CFA Level 2 Quantitative Methods questions present regression output or a time-series model in a vignette and ask you to interpret it or diagnose a problem. CFA Level 2 questions sit in vignettes: a case gives facts, numbers and exhibits, and each question requires pulling the right input and applying the right model, so read the case, identify the model, and compute carefully.
-
Interpret the regression
Regression output is given, and you interpret a coefficient, significance, or R-squared.
-
Diagnose a violation
A regression problem is described, and you identify heteroskedasticity, serial correlation or multicollinearity and its effect.
-
Time series and ML
A time-series or machine-learning situation is described, and you apply the concept.
A regression's independent variables are highly correlated with each other, and the exam notes that the overall model is significant (high R-squared, significant F-test) but individual coefficients are not significant. This pattern, a significant model with insignificant individual coefficients and inflated standard errors, is the classic signature of multicollinearity, correlated independent variables, which is the problem the exam expects, distinct from heteroskedasticity or serial correlation.
Topic areas 3
Official topic areas from the CFA Institute outline, in outline order.
2.1 Multiple regression and model buildingFitting, interpreting and testing a multiple regression.
- What it means
- This covers multiple linear regression with several independent variables: estimating and interpreting the slope coefficients and intercept, testing individual coefficients (t-tests) and the overall model (F-test), R-squared and adjusted R-squared, confidence intervals, and using the model to make predictions.
- What to understand
- Each slope coefficient measures the effect of one variable holding the others constant. The t-test assesses whether a coefficient is significant, and the F-test whether the model as a whole is significant. Adjusted R-squared penalizes adding variables, unlike R-squared. Interpreting these outputs correctly within a case is the core skill, and it underlies factor models elsewhere.
- How it shows up
- Regression output is given, and you must interpret a coefficient, a significance test, or R-squared.
- Common confusion
- Confusing R-squared with adjusted R-squared, or misinterpreting a coefficient's meaning (holding others constant).
2.2 Model misspecification, heteroskedasticity, and multicollinearityRegression assumptions and their violations.
- What it means
- This covers the assumptions underlying regression and the problems when they fail: heteroskedasticity (error variance not constant), serial correlation (errors correlated across observations) and multicollinearity (independent variables correlated with each other), how each is detected, its consequences for the estimates and tests, and how it is addressed.
- What to understand
- Each violation has a signature and a consequence: heteroskedasticity and serial correlation distort the standard errors (affecting significance tests), while multicollinearity inflates standard errors so individual coefficients look insignificant even when the model fits well. Recognizing the symptom, naming the problem, and knowing its effect is heavily tested.
- How it shows up
- A regression problem is described, and you must identify the violation and its consequence.
- Common confusion
- Confusing the three violations and their distinct symptoms and effects.
2.3 Time-series analysis and machine learningTrend and autoregressive models, stationarity, and ML.
- What it means
- This covers time-series analysis: linear and log-linear trend models, autoregressive (AR) models, testing for stationarity and unit roots, seasonality and the random walk, and an introduction to machine-learning methods (supervised and unsupervised learning, overfitting) and big data in investment analysis.
- What to understand
- Time-series models forecast a variable from its own past. A key requirement is stationarity (stable mean and variance); a unit root (non-stationarity) invalidates standard inference and must be addressed (for example, by differencing). Machine learning adds flexible pattern-finding methods, with overfitting as a key risk. Knowing the time-series requirements and the machine-learning concepts is tested.
- How it shows up
- A time-series or machine-learning situation is described, and you must apply the concept or diagnose the issue.
- Common confusion
- Overlooking non-stationarity (a unit root) in a time series, or confusing supervised with unsupervised learning.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Multiple regression
Several independent variables.
- Remember
- Coefficients hold others constant.
- Tested as
- Regression questions.
- Trap
- Misinterpreting a coefficient.
-
t-test vs. F-test
Individual coefficient versus whole model.
- Remember
- t for one, F for all.
- Tested as
- Regression questions.
- Trap
- Confusing the tests.
-
R-squared vs. adjusted R-squared
Fit versus penalized fit.
- Remember
- Adjusted penalizes extra variables.
- Tested as
- Regression questions.
- Trap
- Confusing the two.
-
Heteroskedasticity
Non-constant error variance.
- Remember
- Distorts standard errors.
- Tested as
- Diagnostic questions.
- Trap
- Confusing with multicollinearity.
-
Serial correlation
Autocorrelated errors.
- Remember
- Distorts standard errors.
- Tested as
- Diagnostic questions.
- Trap
- Confusing with heteroskedasticity.
-
Multicollinearity
Correlated independent variables.
- Remember
- Inflates standard errors; model significant, coefficients not.
- Tested as
- Diagnostic questions.
- Trap
- Confusing with the others.
-
Stationarity
Stable mean and variance.
- Remember
- Required for time-series inference.
- Tested as
- Time-series questions.
- Trap
- Ignoring non-stationarity.
-
Unit root
A non-stationary series.
- Remember
- Invalidates standard inference.
- Tested as
- Time-series questions.
- Trap
- Missing it.
-
Autoregressive model
Forecasts from own past values.
- Remember
- AR models.
- Tested as
- Time-series questions.
- Trap
- Misapplying it.
-
Overfitting
Model fits noise, not signal.
- Remember
- A machine-learning risk.
- Tested as
- ML questions.
- Trap
- Ignoring it.
-
Supervised vs. unsupervised
Labeled versus unlabeled learning.
- Remember
- Different ML approaches.
- Tested as
- ML questions.
- Trap
- Confusing the two.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Heteroskedasticity vs. serial correlation vs. multicollinearity
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Heteroskedasticity | Non-constant error variance | Distorts standard errors | Confusing with the others |
| Serial correlation | Correlated errors over time | Distorts standard errors | Confusing with heteroskedasticity |
| Multicollinearity | Correlated independent variables | Model significant, coefficients not | Confusing with the others |
R-squared vs. adjusted R-squared
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| R-squared | Variation explained | Always rises with more variables | Using it to compare models |
| Adjusted R-squared | Penalized for variables | Can fall when adding weak variables | Confusing with R-squared |
Stationary vs. non-stationary series
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Stationary | Stable mean and variance | Valid for inference | Assuming all series are stationary |
| Non-stationary (unit root) | Trending or unstable | Needs differencing | Ignoring it |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Significant model, significant coefficients
- What candidates pick
- Assume a significant model (high R-squared, significant F) means all coefficients are significant.
- Why it looks right
- The model fits well.
- Why it's wrong
- Multicollinearity can make a model significant overall while inflating standard errors so individual coefficients appear insignificant.
Clue to watch for The model is significant but individual coefficients are not.
-
R-squared compares models
- What candidates pick
- Use R-squared to decide whether adding a variable improved the model.
- Why it looks right
- Higher R-squared seems better.
- Why it's wrong
- R-squared always rises when variables are added; adjusted R-squared penalizes extra variables and is used to compare models.
Clue to watch for The question compares models with different numbers of variables.
-
Heteroskedasticity is multicollinearity
- What candidates pick
- Call non-constant error variance a multicollinearity problem.
- Why it looks right
- Both are regression issues.
- Why it's wrong
- Heteroskedasticity is non-constant error variance; multicollinearity is correlation among independent variables, with different symptoms.
Clue to watch for The issue is the error variance versus correlated predictors.
-
Ignore the unit root
- What candidates pick
- Run standard inference on a non-stationary time series.
- Why it looks right
- The model estimates.
- Why it's wrong
- A unit root (non-stationarity) invalidates standard inference; the series must be made stationary, for example by differencing.
Clue to watch for The time series trends or has a unit root.
-
Overfitting is good fit
- What candidates pick
- Treat a model that fits the training data almost perfectly as excellent.
- Why it looks right
- High fit looks strong.
- Why it's wrong
- A model that fits the sample too closely may be overfitting noise and generalize poorly out of sample.
Clue to watch for The model fits in-sample data extremely well.
Answer framework
For any Quantitative Methods item set, use these steps to find the answer the exam expects.
- For regression output, interpret each coefficient as the effect holding others constant, and use the t-test for coefficients, the F-test for the model, and adjusted R-squared to compare models.
- Diagnose assumption violations: heteroskedasticity and serial correlation distort standard errors, while multicollinearity makes a significant model have insignificant coefficients.
- For time series, check stationarity, address any unit root (for example, by differencing), and apply trend or autoregressive models.
- For machine learning, distinguish supervised from unsupervised methods and watch for overfitting.
- Pull the right figures from the vignette's exhibits and apply the correct model or diagnosis. Where two options seem to fit, choose the one that matches the specific statistical symptom or model the vignette describes.
Practice question
Want more like this? The CFA Level 2 practice test has questions for every domain.
Checklist
0 of 10
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Approach Level 2 Quantitative Methods as technically demanding despite its smaller weight, since the item-set format punishes shaky understanding. Start with multiple regression: interpreting coefficients (holding others constant), the t-test and F-test, and adjusted versus plain R-squared, practicing on regression-output exhibits like those in vignettes.
Then master the three misspecification problems, which are heavily tested: keep heteroskedasticity, serial correlation and multicollinearity distinct by their symptoms and effects, especially multicollinearity's signature of a significant model with insignificant coefficients. Learn time-series requirements (stationarity, unit roots) and the machine-learning concepts (supervised versus unsupervised, overfitting).
Because this is applied within cases, drill item sets that present regression output and ask you to diagnose or interpret. Because Level 2 is tested through vignettes, practice full item sets under timed conditions (about 12 minutes per set), pulling the right figure from the right exhibit, and review misses by checking both input selection and calculation. Afterwards, move to Economics.
FAQ
How much of CFA Level 2 is Quantitative Methods?
Quantitative Methods is 5-10% of the exam, a smaller weight, but it contains technically difficult regression, time-series and machine-learning material tested in dense vignettes.
What is multicollinearity and how is it recognized?
Multicollinearity is high correlation among the independent variables in a regression. Its classic signature is a model that is significant overall (high R-squared, significant F-test) but with individual coefficients that appear insignificant because their standard errors are inflated.
What is the difference between R-squared and adjusted R-squared?
R-squared measures the proportion of variation explained and always rises when variables are added. Adjusted R-squared penalizes the addition of variables, so it can fall when a weak variable is added, making it the better measure for comparing models with different numbers of variables.
What are the main regression assumption violations?
Heteroskedasticity (non-constant error variance), serial correlation (errors correlated across observations), and multicollinearity (correlated independent variables). The first two mainly distort the standard errors and significance tests, while multicollinearity inflates coefficient standard errors.
Why does stationarity matter in time series?
Standard time-series inference assumes the series is stationary (stable mean and variance). A non-stationary series with a unit root invalidates that inference and can produce spurious results, so the series must be made stationary, often by differencing, before modeling.
What is overfitting in machine learning?
Overfitting occurs when a model fits the training data too closely, capturing noise rather than the underlying signal, so it performs poorly on new, out-of-sample data. Guarding against overfitting is a central concern in applying machine learning.
How does Quantitative Methods connect to the rest of the Level 2 exam?
Its regression tools underlie the multifactor models in portfolio management, its forecasting supports economics and valuation, and the general skill of interpreting model output within a case applies throughout the vignette-based exam.
Other Finance certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 2 knowledge
You have covered multiple regression, misspecification, time series and machine learning. Test whether you can interpret and diagnose models under exam conditions.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the CFA Level 2 AI tutor Get any concept explained your way