Fundamentals of Time Series Cross Section Analyses

Lorena Barberia, University of São Paulo, Andrew Philips, University of Colorado at Boulder and Guy Whitten, Texas A & M University


Time series variables (e.g. presidential approval, public mood liberalism, GDP, inflation, education level) are extremely common in the social sciences. However, due to certain properties, these series cannot always be handled using standard regression approaches.  
This five-day course is an applied intermediate-level course focusing on the techniques for testing theories with time series data. Each day the course will involve about three hours of lecture time with breaks, then lunch, then three to four hours of hands on instruction in analysis that takes place in smaller groups using STATA.


This course runs January 21-25, 2019.

TEACHING FELLOW Rodrigo Nakahara, University of São Paulo


Day 1: On the first day, we will make the jump from the previous week---which covered univariate time series models---to discussing regression models more common in applied analysis (i.e. a dependent variable and multiple independent variables). We will discuss how autocorrelation is commonly treated in these models, lag structures, and interpreting the output from multivariate regression models.

Day 2: On the second day, we will cover an important topic, cointegration, which means that two or more non-stationary series to have a stable long-run relationship. Often, this is estimated using a model known as an error correction model. There has been a lot of recent discussions about cointegration and error correction, so we will spend some time talking about this debate.

Day 3: On the third day, we will relax assumptions that our independent variables are exogenous; in other words, what if our dependent variable affects our independent variables? We will cover two models, vector autoregressive models (VAR) and the cointegrating equivalent, vector error correction models (VECM).

Day 4: On the fourth day, we extend the strict stationary-non-stationary cut-points to talk about a more nuanced phenomena, fractional integration. We will discuss how to identify, test for, and model fractionally integrated series.

Day 5: On the last day, we will cover compositional time series data. Much of time series data is compositional (e.g. the level of public support for parties over time, budget allocations). We will discuss a number of recent approaches to model and visualize compositional time series data.


A full-semester graduate-level course in multiple regression analysis and Modeling Dynamics (offered in the IPSA-USP 2017 Summer School) or the equivalent background in time series analysis.