Shanghai Jiao Tong University · Summer School 2026
Syllabus
The course teaches the econometrics of finance through models and real data, using Stock & Watson (4th ed.) as the main text. The course asks how to estimate a relationship, when it is causal rather than merely predictive, and how to model and forecast financial time series. Every method is taught twice, in Stata and in Python.
Course overview
Econometrics turns data into evidence. The course builds from the ground up: the probability and statistics behind every estimate; the linear regression model and how to read its output; inference with the right standard errors; the omitted-variable problem that separates a correlation from a causal effect; and the tools of panel data and time series that finance relies on.
The second half is financial time series: how to forecast a series and why stock returns resist it (the efficient-market hypothesis), how a shock propagates through the economy (dynamic causal effects), and how to model risk itself with ARCH/GARCH, several series at once with a VAR, and the long-run links between non-stationary series with cointegration. Throughout, two questions are asked of every estimate: is it precise, and is it causal?
Learning outcomes
- read and interpret regression output: coefficients, standard errors, t-statistics, p-values, confidence intervals and R²;
- distinguish prediction from causation, and state the exogeneity assumption E(u | X) = 0 that a causal claim requires;
- diagnose and sign omitted-variable bias, and choose the right standard error — heteroskedasticity-robust, clustered (panel), or HAC (time series);
- estimate and test multiple regressions, including joint hypotheses with the F-test;
- analyse panel data with entity and time fixed effects and clustered standard errors;
- model and forecast financial time series: autoregressions, the random walk and the efficient-market hypothesis, dynamic causal effects, and volatility (ARCH/GARCH), VAR and cointegration;
- work fluently in both Stata and Python, reproducing every analysis on real data.
Teaching & assessment
Teaching. A 2-hour-40-minute class each day, Monday to Friday, for three weeks; in addition, the teaching assistants run roughly 3 hours 20 minutes of lab and discussion sessions per week. Each lecture is built in three blocks — theory, then a Stata lab, then a Python lab — and ships a folder of real data, a .do file and a Colab/Kaggle notebook, so every result can be reproduced.
Assessment. The final grade combines attendance (12%), three quizzes (18%), a midterm (20%), a group project and presentation (20%), and a final examination (30%). Quizzes are short and held at the start of class; Quiz 1 (Lectures 1–3) on Thursday 2 July, Quiz 2 (Lectures 4–6) on Wednesday 8 July, and Quiz 3 (Lectures 7–8) on Tuesday 14 July. The midterm (Lectures 1–6, Stock & Watson Ch 1–7) is on Friday 10 July; the group presentations on Thursday 16 July; and the final (Lectures 7–10, Ch 10, 14–16) on Friday 17 July. Dates, coverage and papers are on the Exams page.
Grading scale
Weekly schedule
| Day | Session | |
|---|---|---|
| Week 1: Foundations & the regression model | ||
| Mon 29 Jun | L1 · Economic Questions and Probability | |
| Tue 30 Jun | L2 · Review of Statistics | |
| Wed 1 Jul | L3 · Linear Regression with One Regressor | |
| Thu 2 Jul | Quiz 1 (L1–3) | |
| Fri 3 Jul | L4 · Hypothesis Tests and Confidence Intervals | |
| Week 2: Inference, panel data & into time series | ||
| Mon 6 Jul | L5 · Multiple Regression | |
| Tue 7 Jul | L6 · Hypothesis Tests in Multiple Regression | |
| Wed 8 Jul | Quiz 2 (L4–6) | |
| Thu 9 Jul | L7 · Panel Data and Fixed Effects | |
| Fri 10 Jul | Midterm Examination (L1–6) | |
| Week 3: Financial time series | ||
| Mon 13 Jul | L8 · Time Series and Forecasting | |
| Tue 14 Jul | Quiz 3 (L7–8) + Presentation Information | |
| Wed 15 Jul | L9 · Dynamic Causal Effects · L10 · Volatility, VAR and Cointegration | |
| Thu 16 Jul | Group Project Presentations | |
| Fri 17 Jul | Final Examination (L7–10) | |
Lecture overview
Economic Questions and Probability
What econometrics answers and why finance relies on observational data; random variables, expectation and variance as return and risk; conditional expectation, covariance and correlation; the normal, the t, and the central limit theorem.
Key idea: regression estimates the conditional mean E(Y | X); correlation is not causation.
Review of Statistics
Estimation (unbiasedness, consistency, efficiency, BLUE); hypothesis testing (the t-statistic, the p-value, Type I and II errors, size and power); confidence intervals; comparing two means.
Key idea: a p-value is the probability of the data under the null, not the probability the null is true.
Linear Regression with One Regressor
The population regression line and the OLS estimator; fitted values and residuals; R² and the standard error of the regression; the three least-squares assumptions; the sampling distribution of OLS.
Key idea: E(u | X) = 0 is the assumption that turns a slope into a causal effect.
Hypothesis Tests and Confidence Intervals
Testing and building confidence intervals for the slope; the binary regressor as a difference in means; heteroskedasticity and why robust standard errors are the default; the Gauss–Markov theorem.
Key idea: the coefficient is the same; only the standard error — and so the inference — changes when errors are heteroskedastic.
Multiple Regression
Omitted-variable bias, its two conditions and its sign; the ceteris paribus interpretation; R² versus adjusted R²; the OLS assumptions; multicollinearity and control variables.
Key idea: bias is the change in a slope when a confounder is added; a control need not itself be causal.
Hypothesis Tests in Multiple Regression
Tests on one coefficient and joint tests; the F-statistic and why two t-tests are not enough; the overall F; restrictions across coefficients; specification and confidence sets.
Key idea: correlated regressors can be individually insignificant yet jointly significant — only an F-test gets it right.
Panel Data and Fixed Effects
Between and within variation; entity and time fixed effects (dummies, demeaning, first differences); what fixed effects can and cannot remove; clustered standard errors and how they change inference.
Key idea: fixed effects absorb every time-invariant confounder; clustering then corrects the standard errors — and can flip a result.
Time Series and Forecasting
Lags, differences and growth rates; autocorrelation and stationarity; AR models and persistence; forecasting and its limits; the random walk, unit roots and the efficient-market hypothesis; the yield curve as a leading indicator.
Key idea: inflation is forecastable, stock returns are not — competition arbitrages predictable returns away.
Dynamic Causal Effects
Distributed lags and the impulse response; dynamic, cumulative and long-run multipliers; HAC (Newey–West) standard errors for serially correlated errors; exogeneity, and why it cannot be tested.
Key idea: an oil shock passes into prices fast and into output slowly; HAC and lincom give honest inference, but exogeneity is a judgement.
Volatility, VAR and Cointegration
Volatility clustering and ARCH/GARCH; persistence, volatility forecasting, Value-at-Risk and the leverage effect; vector autoregressions and Granger causality; unit roots, cointegration and error correction.
Key idea: returns are unforecastable but their risk is not; GARCH turns risk into a time-varying number, and cointegration ties wandering series together.
Textbook, reading & data
The course is taught with data and software; every analysis is reproduced in both Stata and Python, on real financial and macro series, and students are encouraged to re-run the labs and extend them to their own questions.
- Required text. Stock, James H., and Mark W. Watson. Introduction to Econometrics. 4th ed. (Global Edition). Pearson, 2020. ISBN 9781292264455.
- Brooks, Chris. Introductory Econometrics for Finance. 4th ed. Cambridge University Press, 2019. Supplementary — the finance applications.
- Wooldridge, Jeffrey M. Introductory Econometrics: A Modern Approach. 7th ed. Cengage, 2019. Supplementary.
- Regression & inference: the least-squares assumptions; heteroskedasticity-robust standard errors (White, 1980); omitted-variable bias; the F-test.
- Panel & causality: fixed effects and clustered standard errors; the distinction between prediction and causation (Granger, 1969).
- Time series & finance: the random walk and the efficient-market hypothesis (Fama, 1970); unit roots (Dickey & Fuller, 1979); HAC standard errors (Newey & West, 1987); ARCH (Engle, 1982) and GARCH (Bollerslev, 1986); cointegration and error correction (Engle & Granger, 1987).
- Software: Stata for the lab
.dofiles; Google Colab / Kaggle for the Python notebooks (pandas,statsmodels,arch). - Data sources: the Fama–French data library; FRED, Federal Reserve Bank of St. Louis (daily S&P 500, Treasury rates, CPI, industrial production, GDP); and the course's own bundled lab datasets.
