Why Quantitative Analysis Matters
Quantitative Analysis accounts for approximately 20% of FRM Part I. Combined with Valuation and Risk Models (another 30%), quantitative reasoning underlies around half the exam. Candidates who struggle with the quant domain typically find that the weakness bleeds into Valuation and Risk Models as well, since VaR calculations and option pricing require fluent statistical and mathematical thinking.
The Quantitative Analysis domain is also the one that most often divides candidates by background. Finance professionals who studied economics, accounting, law, or non-quantitative social sciences may not have encountered probability distributions, regression analysis, or Monte Carlo methods in their undergraduate or postgraduate education. The FRM is not a mathematics exam, but the quantitative domain expects genuine familiarity with statistical concepts applied to financial risk data.
This post covers the specific topics within Quantitative Analysis where candidates most commonly lose marks, and what effective preparation for each looks like.
Probability Distributions and Their Properties
Probability distributions are foundational to almost everything else in the quant domain. The exam tests several specific distributions and their properties, including the normal distribution, the lognormal distribution, the Student's t-distribution, and the binomial distribution.
The normal distribution appears throughout FRM at both Part I and Part II. Candidates need to know its properties (symmetry, the relationship between mean and median, the 68-95-99.7 rule), how to standardise a normally distributed variable using z-scores, and how to use the standard normal table.
The lognormal distribution is particularly important for asset price modelling. The key property that trips candidates up: if the natural log of a variable is normally distributed, the variable itself is lognormally distributed. Asset prices are often assumed to be lognormally distributed (so that returns, which are log differences, are normally distributed). Questions test whether candidates understand this relationship and can apply it to forward price calculations and option pricing.
The t-distribution becomes relevant for small samples and for hypothesis testing where the population variance is unknown. Candidates need to understand how the t-distribution differs from the normal as sample size changes (thicker tails with small samples, converging to normal as sample size grows) and when each is appropriate.
Where candidates lose marks: Confusing which distribution applies to a given scenario, or knowing a distribution's properties in the abstract but being unable to apply them to a stated financial risk problem.
Regression Analysis
Linear regression is tested in several forms. Candidates need to understand ordinary least squares (OLS) regression, interpret regression output (coefficients, standard errors, R-squared, F-statistic), and identify violations of OLS assumptions.
The exam tests the Gauss-Markov conditions and what happens to OLS estimates when they are violated. Heteroskedasticity (non-constant variance of residuals) does not bias coefficient estimates but makes standard errors unreliable, which invalidates hypothesis tests. Autocorrelation (correlated residuals) has a similar effect. Multicollinearity (high correlation among independent variables) inflates standard errors, making it difficult to identify the individual contribution of correlated predictors.
Logistic regression appears in the curriculum for credit risk contexts, particularly probability of default modelling. The exam tests the interpretation of logistic regression output rather than the underlying maximum likelihood mechanics.
Where candidates lose marks: Knowing the names of OLS violations without understanding their consequences. An exam question that asks "what is the effect of heteroskedasticity on OLS coefficient estimates?" expects the answer that estimates remain unbiased but inference is unreliable. Candidates who confuse effect on estimates with effect on standard errors score this wrong.
Hypothesis Testing and Statistical Inference
Hypothesis testing questions require candidates to interpret test statistics, p-values, and confidence intervals, and to make correct decisions about null hypotheses given stated significance levels.
The standard framework is familiar: set up a null hypothesis and alternative, calculate a test statistic, compare to a critical value (or compare the p-value to the significance level), and decide whether to reject. The FRM exam applies this framework to financial contexts: testing whether a trading strategy's returns are statistically different from zero, whether a portfolio's beta is significantly different from one, whether two means are statistically equal.
Type I and Type II errors (false positives and false negatives) are tested directly. Candidates need to know that the significance level is the probability of a Type I error, that reducing Type I error increases Type II error for a fixed sample size, and that the power of a test is one minus the probability of a Type II error.
Where candidates lose marks: Mechanically applying the test without reading what the question is asking. Some questions ask for the test statistic, others ask for the decision, others ask what happens if the significance level changes. Reading carefully before calculating is as important as knowing the formula.
Estimating Volatility
Volatility estimation is one of the bridges between the Quantitative Analysis domain and Valuation and Risk Models. The exam tests several approaches.
Equally weighted historical volatility is the simplest: compute the daily return variance as the average squared deviation from the mean (in practice, returns are often assumed to have a zero mean over short periods, simplifying the calculation) and annualise by multiplying by the square root of 252 (trading days in a year).
EWMA (Exponentially Weighted Moving Average) gives more weight to recent observations. The GARP curriculum introduces the RiskMetrics EWMA model with decay factor lambda (λ). A higher lambda gives more weight to distant observations and produces smoother volatility estimates; a lower lambda reacts more quickly to recent price moves. The exam commonly tests the EWMA recursion formula and what changes when lambda changes.
GARCH(1,1) extends EWMA by adding a long-run average variance component, allowing volatility to mean-revert. The GARCH(1,1) model has three parameters: the weight on the long-run variance (omega), the weight on the last period's squared return (alpha), and the weight on the last period's variance estimate (beta). The constraint that omega + alpha + beta = 1 (in the EWMA case omega = 0) is regularly tested.
Where candidates lose marks: Confusing EWMA and GARCH(1,1), or misinterpreting the effect of parameter values on volatility forecasts. The exam also tests whether candidates can calculate a one-day volatility estimate from a stated EWMA recursion, which requires numerical comfort with the formula.
Simulation Methods
Monte Carlo simulation and historical simulation both appear in the quantitative domain. For Monte Carlo, candidates need to understand how random draws from distributions generate scenarios, what determines the quality of a simulation estimate (number of draws, the accuracy of the distributional assumption), and how simulation is applied to VaR estimation.
Historical simulation uses actual historical returns to construct the distribution of potential portfolio losses. Its advantage is that it does not require a distributional assumption; its limitation is that it is backward-looking and depends on whether the historical period captures the risk being modelled.
Where candidates lose marks: Questions about simulation often ask about the limitations of each approach. Candidates who can describe what the method does but cannot identify its failure modes — particularly what each approach gets wrong in a regime-change environment — find these questions difficult.
Preparing for Quantitative Analysis
The most effective preparation for this domain is structured question practice that begins early. Reading about probability distributions and regression does not build the ability to apply them under exam conditions; answering questions and reviewing wrong answers does.
Part I Quantitative Analysis practice questions are available on passfrm.com. Working through questions by subtopic (probability, then regression, then hypothesis testing, then volatility, then simulation) allows you to identify which specific areas need the most reinforcement before consolidating with mixed-topic practice.
Candidates from non-quantitative backgrounds should budget more time for this domain than the 20% exam weighting implies. Given that weak quant performance also affects performance on Valuation and Risk Models, the hours invested here have leverage across approximately 50% of the Part I exam.