Washu Econ 4002

Econ 4002 provides a comprehensive introduction to econometrics, covering essential topics from simple linear regression to time series analysis. The course emphasizes understanding OLS assumptions, hypothesis testing, and the application of econometric techniques to real-world economic data, while addressing common pitfalls in empirical research.

Note 1: Introduction to Econometrics and the Nature of Economic Data Econometrics is the application of statistical methods to economic data to give empirical content to economic relationships. It combines economic theory, mathematics, and statistical inference to quantify and test economic hypotheses. Unlike purely mathematical economics, econometrics is fundamentally concerned with measurement and validation against real-world data.

The nature of economic data is diverse. We encounter data collected through experiments (rare in economics compared to natural sciences), observational studies (most common), and surveys. Data can be cross-sectional (observations at a single point in time), time series (observations over time for a single entity), or panel data (observations over time for multiple entities). Understanding the data's source and structure is crucial for choosing appropriate econometric techniques. For example, time series data often exhibit serial correlation, requiring specialized models. A common misconception is treating all data as equally informative, ignoring potential biases from sampling methods or measurement errors. Recognizing data limitations and potential biases is fundamental to sound econometric analysis.

Note 2: Simple Linear Regression Model The simple linear regression model posits a linear relationship between a dependent variable (y) and a single independent variable (x), represented as y = β₀ + β₁x + u, where β₀ is the intercept, β₁ is the slope, and u is the error term. The error term captures all other factors affecting y besides x. The goal is to estimate β₀ and β₁ using observed data.

The Ordinary Least Squares (OLS) method is the most common technique for estimating these parameters. OLS minimizes the sum of squared residuals (the difference between observed and predicted values). The resulting estimators, denoted as b₀ and b₁, are functions of the sample data. Under certain assumptions (zero conditional mean of the error term, independence of the error term and x, homoskedasticity, and no serial correlation), OLS estimators are unbiased and efficient (Best Linear Unbiased Estimator - BLUE). A common misconception is assuming OLS is always the best method; violation of these assumptions requires alternative estimation techniques.

Note 3: OLS Assumptions and Properties The OLS assumptions are critical for the validity of OLS estimates. The zero conditional mean assumption (E[u|x] = 0) is paramount; it implies that the independent variable is uncorrelated with the error term. Violation of this assumption leads to omitted variable bias. Homoskedasticity requires that the variance of the error term is constant across all values of x (Var[u|x] = σ²). Heteroskedasticity, where the variance is not constant, does not bias the OLS estimators, but it does invalidate the standard errors, leading to incorrect hypothesis testing.

The property of unbiasedness (E[b₁] = β₁) ensures that, on average, the estimated coefficient is equal to the true population coefficient. Efficiency means that the OLS estimator has the smallest variance among all linear unbiased estimators. Understanding these assumptions and properties is essential for interpreting regression results correctly. A frequent mistake is ignoring the assumptions and drawing conclusions based solely on the R-squared value, which only indicates the proportion of variance in y explained by x.

Note 4: Hypothesis Testing and Confidence Intervals Hypothesis testing allows us to assess the statistical significance of our estimated coefficients. The most common hypothesis is testing whether a coefficient is equal to zero (H₀: β₁ = 0), indicating no relationship between x and y. We construct a t-statistic (t = (b₁ - 0)/SE(b₁)), where SE(b₁) is the standard error of the estimator. We compare the t-statistic to a critical value from the t-distribution with appropriate degrees of freedom (n - k - 1, where n is the sample size and k is the number of independent variables).

A confidence interval provides a range of values within which we are confident the true population parameter lies. A 95% confidence interval is constructed as b₁ ± (critical value * SE(b₁)). If the hypothesized value (e.g., 0) falls within the confidence interval, we fail to reject the null hypothesis. Common mistakes include confusing statistical significance with economic significance and misinterpreting the meaning of the p-value (the probability of observing a test statistic as extreme as, or more extreme than, the one observed, given that the null hypothesis is true).

Note 5: Multiple Linear Regression Model The multiple linear regression model extends the simple linear regression model to include multiple independent variables: y = β₀ + β₁x₁ + β₂x₂ + ... + βₖxₖ + u. This allows us to control for other factors that may affect the dependent variable and isolate the effect of a specific independent variable of interest. The interpretation of the coefficients changes; β₁ now represents the change in y for a one-unit change in x₁, holding all other variables constant (ceteris paribus).

The OLS estimation procedure remains the same, minimizing the sum of squared residuals. However, the OLS assumptions need to be extended to account for multiple variables. Multicollinearity, a high correlation between independent variables, can inflate the standard errors and make it difficult to obtain precise estimates. The adjusted R-squared is used to compare models with different numbers of independent variables, penalizing the inclusion of irrelevant variables. A common misconception is that adding more variables always improves the model; irrelevant variables can decrease the adjusted R-squared and lead to less precise estimates.

Note 6: Dummy Variables and Interaction Terms Dummy variables are binary variables (0 or 1) used to represent categorical variables in regression models. For example, a dummy variable could represent gender (1 for female, 0 for male) or region (1 for urban, 0 for rural). Including dummy variables allows us to estimate the effect of the category on the dependent variable.

Interaction terms are created by multiplying two or more independent variables together (e.g., x₁ * x₂). Interaction terms allow the effect of one independent variable on the dependent variable to vary depending on the value of another independent variable. For example, the effect of education on wages may differ for men and women; an interaction term between education and a gender dummy would capture this difference. Using dummy variables and interaction terms correctly requires careful consideration of the research question. A common mistake is including too many interaction terms without a strong theoretical justification, leading to overfitting and difficult-to-interpret results.

Note 7: Heteroskedasticity and Robust Standard Errors Heteroskedasticity, where the variance of the error term is not constant, violates one of the OLS assumptions. While it does not bias the OLS estimators, it invalidates the standard errors, leading to incorrect hypothesis testing. We can detect heteroskedasticity using tests like the Breusch-Pagan test or the White test.

A common solution to heteroskedasticity is to use robust standard errors (also known as heteroskedasticity-consistent standard errors). These standard errors are valid even in the presence of heteroskedasticity. Most statistical software packages provide options for calculating robust standard errors. Another approach is to use weighted least squares (WLS), which gives more weight to observations with smaller error variances. While robust standard errors correct for the issue, understanding the source of the heteroskedasticity is important. Ignoring the underlying cause can lead to misinterpretations. Common mistake is just applying robust standard errors without exploring if there's a better model specification available.

Note 8: Time Series Data and Autocorrelation Time series data consists of observations over time for a single entity (e.g., GDP, stock prices). A key characteristic of time series data is autocorrelation (or serial correlation), where the error term in one period is correlated with the error term in another period. Autocorrelation violates the OLS assumption of uncorrelated errors, leading to inefficient estimators and invalid standard errors.

We can detect autocorrelation using tests like the Durbin-Watson test or by examining the autocorrelation function (ACF) and partial autocorrelation function (PACF). To address autocorrelation, we can use techniques like Generalized Least Squares (GLS) or include lagged dependent variables as regressors in the model. Models such as AR (Autoregressive) and MA (Moving Average) models are specifically designed for time series data. A common misconception is treating time series data like cross-sectional data and applying standard OLS without accounting for autocorrelation, leading to spurious regression results.

Browse public study notes