How to study for an econometrics exam

The most effective way to study for an econometrics exam is to work past problem sets and practice exams under timed conditions, organized around four pillars: estimator properties, the Gauss-Markov assumptions and what happens when they fail, hypothesis testing mechanics, and the interpretation of regression coefficients — especially log specifications. Re-reading notes feels productive but builds little. Econometrics exams reward students who can both derive an estimator's properties and read a regression table quickly, and each of those is a skill built only through practice.

Why econometrics is unlike any other economics exam

Most economics exams test one mode of thinking. A microeconomics paper asks you to reproduce diagrams and marginal reasoning; a macro paper asks you to trace a policy shock through a model. Econometrics asks for two modes at once.

On the theory side, you may be asked to prove that OLS is unbiased, or to derive the estimator

$\hat{\beta} = (X'X)^{-1}X'y$

from the normal equations. On the applied side, you will be handed a regression output table and asked what the coefficient on $\ln(\text{income})$ means, whether it is statistically significant, and how the estimates change if a relevant variable was omitted.

These are genuinely different skills, and students who prepare for only one of them get caught. Proofs reward careful, line-by-line algebra. Applied questions reward speed and judgment. Your study plan needs deliberate time on both.

Triage your material around four pillars

A semester of econometrics can run to hundreds of slides, but nearly every exam draws from the same four areas. Sort your notes accordingly before you do anything else.

1. Estimator properties

Know what unbiasedness, consistency, and efficiency mean formally. $E[\hat{\beta}] = \beta$ is a statement about repeated sampling, not about your one estimate — examiners check whether you understand that distinction. Then learn the standard proof of unbiasedness until you can reproduce it cold: substitute $y = X\beta + u$ into the OLS formula to get

$\hat{\beta} = \beta + (X'X)^{-1}X'u,$

then take expectations under the exogeneity assumption $E[u \mid X] = 0$. The derivation looks intimidating in lecture notes, but it uses the same few moves every time. If your course also proves consistency, know exactly where the law of large numbers enters the argument.

2. The Gauss-Markov assumptions — and what breaks when they fail

Listing the assumptions earns almost no marks. The real exam question is always: which assumption fails in this scenario, and what is the consequence?

Assumption	Typical violation	Consequence for OLS
Exogeneity, $E[u \mid X] = 0$	Omitted variables, simultaneity	Biased and inconsistent estimates
Homoskedasticity	Error variance depends on regressors	Coefficients still unbiased; standard errors invalid
No perfect collinearity	Dummy variable trap	Estimator not defined
No autocorrelation	Serial correlation in time series	Coefficients still unbiased; standard errors invalid

Notice the pattern examiners love: some violations leave $\hat{\beta}$ unbiased but wreck inference, while others bias the estimates themselves. If you can sort any scenario into the right column, you are ahead of most of the room.

3. Hypothesis testing mechanics

The t-statistic

$t = \frac{\hat{\beta}_j - \beta_{j,0}}{\operatorname{se}(\hat{\beta}_j)}$

should be automatic, along with the F-test for joint significance and a precise, one-sentence definition of a p-value. These are mechanical points — the cheapest marks on the paper — and losing them to hesitation is the most avoidable mistake in the subject.

4. Interpretation, especially log specifications

This is where applied questions live or die. Memorize this table until you can reproduce it from a blank page:

Specification	Model	Interpretation of $\beta_1$
Level-level	$y = \beta_0 + \beta_1 x$	One-unit increase in $x$ changes $y$ by $\beta_1$ units
Log-level	$\ln y = \beta_0 + \beta_1 x$	One-unit increase in $x$ changes $y$ by about $100 \cdot \beta_1$ percent
Level-log	$y = \beta_0 + \beta_1 \ln x$	One percent increase in $x$ changes $y$ by about $\beta_1 / 100$ units
Log-log	$\ln y = \beta_0 + \beta_1 \ln x$	Elasticity: one percent increase in $x$ changes $y$ by $\beta_1$ percent

The log-log case is the one to internalize deeply: $\beta_1$ is the elasticity, which is why economists reach for that specification so often. Know also what $R^2 = 1 - \frac{\text{SSR}}{\text{SST}}$ measures — the share of variation in $y$ explained by the regressors — and that a high $R^2$ says nothing about causality or correct specification.

Work problems, do not re-read

The research here is unambiguous. In a major review of ten study techniques, Dunlosky and colleagues (2013) rated practice testing as one of only two methods with high utility, while re-reading and highlighting — the default habits of most students — rated low. Roediger and Karpicke (2006) found that students who tested themselves retained substantially more a week later than students who spent the same time repeatedly studying the material.

For econometrics, that means: collect every past problem set and past paper your course provides, and redo them from scratch — not by reading your old solutions, but by writing new ones with the answer hidden. When you run out of problems, generate more. You can upload your lecture notes and problem sets to ExamTeX and get a fresh, LaTeX-typeset practice exam with an answer key — useful in this subject in particular, because $(X'X)^{-1}$, hats, and Greek letters render as proper mathematics rather than mangled plain text.

Read regression output like an examiner

Applied exam questions usually hand you software output and a stack of sub-questions. Under time pressure, you need a fixed scan order, not a fresh strategy each time:

1. Functional form first. Which variables are logged? Which are dummies? This determines every interpretation that follows.
2. The coefficient of interest. Sign, magnitude, units. Does the sign match economic intuition? If not, suspect omitted variable bias and say so.
3. Standard errors and significance. Form the t-statistic, compare to the critical value, state the conclusion in words.
4. $R^2$ and sample size last. They contextualize the result; they rarely drive the answer.

Then write your interpretation as one disciplined sentence: "Holding the other regressors fixed, a one percent increase in price is associated with a $\beta_1$ percent decrease in quantity demanded." Practice this scan on real output until a full table takes you under two minutes. That speed is what frees up time for the proof questions.

The failure modes that show up every semester

A few mistakes account for a remarkable share of lost marks:

• Memorizing formulas without knowing when assumptions break. Quoting the Gauss-Markov theorem earns nothing if you cannot spot heteroskedasticity in a scenario about income and spending.
• Confusing statistical and economic significance. A tiny coefficient can be highly significant in a large sample; say which kind of significance the question is asking about.
• Unit slips in log models. Forgetting the factor of 100 in a log-level interpretation is the single most common applied error.
• Dummy variable confusion. Always state the reference category; a coefficient on a dummy is a comparison, not a level.
• Hand-waving omitted variable bias. The direction of the bias depends on the sign of the omitted variable's effect on $y$ and its correlation with the included regressor. Work the sign out explicitly; do not guess.

Definitions, assumptions, and interpretation rules are classic retrieval-practice material. Putting them on spaced-repetition flashcards in the weeks before the exam keeps them available under pressure, and the underlying statistical machinery — distributions, tests, p-values — overlaps heavily with our statistics flashcard sets. For the broader evidence on why self-testing outperforms review, see our guide to practice testing.

The week before the exam

Keep it simple. Alternate days between theory and application: one day re-deriving the unbiasedness proof and the assumption table from memory, the next day working timed applied questions on regression output. Two or three days out, sit one full past paper under exam conditions — timed, closed book, no pausing. The point is not the score; it is finding out where you hesitate while there is still time to fix it.

The bottom line

Econometrics exams test two skills: deriving estimator properties and interpreting regression output, both under time pressure. Triage your material around estimator properties, the Gauss-Markov assumptions, hypothesis testing, and log interpretations. Then spend your hours working problems from a blank page rather than re-reading notes — the evidence on practice testing is decisive, and in this subject the gap between recognizing a derivation and reproducing one is exactly where exams are won and lost.