Advanced econometrics

Chapter 5 : The statistics of the linear regression model, small samples

By Lund University

The statistics of the linear regression model, small samples

The linear regression model

We begin by looking at what we mean by a random sample. Given a random sample, we formulate a statistical model or a linear statistical model by modeling the conditional expectation of y given x. We then consider b as an estimator of the unknown parameter beta. From the conditional expectations, we can introduce the error terms which will lead us to the fundamental regression model.

Random sample, statistical model and linear statistical model

Estimating the parameters of a statistical model

Conditional expectations in matrix form

Error terms and the regression model

Misspecified models

Properties of the OLS estimator

We have now created the setup that we need to analyze the properties of the OLS estimator. We begin by looking at the OLS statistical formula which related the OLS estimator directly to the error terms. Using this formula, we can show that the OLS estimator is unbiased under the assumption of exogeneity. In order to anything more about the OLS estimator, we introduce the concept heteroscedasticity (constant variance) and the Gauss-Markov assumptions. We will be able to find the variance of the OLS estimator these new assumptions. We also look at the Gauss-Markov theorem which tells us that that the OLS estimator is the Best Linear Unbiased Estimator under the GM assumptions. We end this section by looking at how to estimate the variance of the OLS estimator.

The OLS statistical formula

Unbiasedness of the OLS estimator

Homoscedasticity and Gauss-Markov assumptions

The variance of the OLS estimator

The Gauss Markov theorem

Estimating the variance of the error terms

Estimating the variance of the OLS estimator:

Inference in the linear regression model

Next, we look at hypothesis testing and confidence intervals in the linear regression model. We then need the distribution of the OLS estimator which requires additional assumptions on the error terms. In this section, we will assume that the error terms follow IID normal distributions. Using this assumption, we can derive the distribution of the OLS estimator. Once we have this distribution, we can do hypothesis testing. We begin with simplest ones, testing if an individual beta-parameter is zero. We build on this and learn how to test general linear restrictions using t, F and chi-square tests. This section also has a page on confidence intervals.

The distribution of the OLS estimator

More on the distribution of the OLS estimator, the t-distribution

Hypothesis testing: simple t-test

Simple t-test: extensions

Confidence interval for

Testing a single linear restriction

Testing several linear restrictions jointly

Problems

Problems

Prove the OLS statistical formula

Proof that the OLS estimator is unbiased

A transpose problem

The variance of the OLS estimator

Proof of the Gauss Markov theorem

A t-distribution

A t-distribution

An F-distribution