ECOM216 - Applied Econometrics with R

ECOM216 - Applied Econometrics with R

• This module covers applied econometric methods.
• Students are equipped with essential skills to analyze data, make informed decisions, and prepare for advanced studies in econometrics and finance.
• Hands-on experience using econometric tools in R is emphasized to solve real-world problems.
• The module introduces key applied econometric methods (micro and macro).
• It covers methods for identifying causal impacts of policies, including randomized controlled trials, causal regressions, and differences-in-differences.
• Time series modeling, prediction, and the interpretation of econometric models are also included.
• The module utilizes R for practical applications, emphasizing implementation skills instead of just theoretical understanding.
• This practical learning approach involves experimentation with methods and coding within R.
• Prior R knowledge is not assumed, and learning requires active participation and effort.
• All lectures have associated code books.
• Reproducible examples and real-world data are used for better understanding.
• Lectures focus on theory and methods with live coding examples.
• Tutorials focus on coding aspects, including clarifying and expanding lecture material and introducing new tools.

Part I - General Concepts

• This section covers fundamental concepts, spanning two lectures.
• It introduces R, core R functions, the tidyverse library, and essential data manipulation techniques.
• Data loading, manipulation, and regression analysis are explored, along with interpretations from models such as OLS.
• Key properties of Ordinary Least Squares (OLS) regression and concepts of exogeneity and causality are covered.

Part II - Causality

• This section, comprising four lectures, focuses on causal interpretations of econometric models.
• It explains how to establish causal effects from regressions.
• The Rubin causal model, randomization ideals, and implementation using regression models are discussed.
• Control regressions and conditional models are highlighted.
• The module emphasizes moving beyond simple controls to methods like fixed effects and matching for more rigorous causal inference.
• It covers differences-in-differences, staggered treatments, and continuous treatment variables.

Part III - Forecasting & Time Series Models

• This part covers forecasting and time series models, spanning four lectures.
• It includes introductory concepts in time series, stationarity, autocorrelation functions, and partial autocorrelation functions.
• The module delves into autoregressive (AR) and moving average (MA) processes.
• Discusses Dickey-Fuller unit root tests for assessing stationarity.
• It explores Autoregressive Integrated Moving Average (ARIMA) models and seasonal terms.
• The module provides a practical approach with external variables, with a focus on model selection using Box-Jenkins methodology.

Assessment

• Assessment is 80% exam and 20% online quiz through QMplus.
• The online quiz timing is either in week 8 or week 9, with the exact date announced by the office.
• Coding is examinable, with reasoned constraints.
• Questions may involve interpreting code, outcomes, and next steps, and suggesting strategies for particular problems or questions.
• Live coding will not be assessed.

Feedback and Support

• Office hours, typically held on Tuesdays at 4 p.m., with additional sessions available upon request.
• Support sessions occur via Zoom and require prior booking.
• Additional timeslots may be scheduled to accommodate busy periods, and online feedback is welcomed.

What is R?

• R is a free software environment for statistical computing and graphics.
• It is open-source, user-supplied, with a strong and active community.
• There are over 20,000 packages, some with their own unique syntaxes.

Understanding Society

• This dataset is a nationally representative, longitudinal survey of individuals.
• Data are collected over many years, including demographic, household, well-being, and life satisfaction, occupation, employment, politics, and environmental data.
• The course uses a subset of data including waves 1-9 (up to 2019).
• Data definitions are harmonized, and any distortion from COVID-19 is removed.
• Students can access the Understanding Society teaching dataset from the UK Data Service website using index code SN 8715.

Possible Interpretations of an Econometric Model

• Econometric models can be categorized as predictive (forecasting) or causal.
• Predictive models use existing data to predict future outcomes.
• Causal models assess the effect of policies or actions on individuals, households, etc.
• External validity assesses results' applicability to other settings.
• Models' interpretations must consider confounders, selection biases, unobserved variables, and other complexities to yield valid conclusions.

Use Case - Predictive Models

• Models use satellite data and night-lights to enhance official GDP statistical data for more precise forecasting, by identifying economic activity indirectly by linking to physical activity.
• This is an area of active research frequently updating and expanding the scope of applications.

Use Case - Causal Models

• Causal models can analyze the effect of events or interventions on economic outcomes, e.g., the 1992 mortgage reform in Denmark.
• This helps identify the relationship between financial frictions, interest rates, labor markets, and other outcomes of interest on economic data collected from surveys and observations.

We think counterfactual when we say causal

• Causal estimates require considering alternative scenarios (e.g., what would have happened if a policy wasn't implemented).
• Isolating policy effects from other influences (e.g., confounders, selection issues, unobserved variables) is crucial but challenging.
• Time-dependent factors, constantly changing reality, create complications to policy prescriptions based on causal interpretations of data.

Counterfactual, Identification, Selection

• Complexities arise in understanding health outcomes when considering the interaction between hospitalization and health status, which might not lead to causality as inferred.
• Comparisons should consider potential underlying reasons causing a difference in observation outcome.
• Comparisons/analyses need to consider individuals' starting conditions (e.g. pre-existing health factors).

Towards a Causal Model

• Causal models draw upon past work.
• Potential outcomes and methods for estimating causal effects from observed data are introduced, along with various challenges.
• The fundamental problem states that outcomes under different treatment scenarios are not directly observed; potential outcomes are inferred to quantify a potential causal relationship.

ATE, ATT, ATU and not too LATE

• This section covers various treatment effects, conditional on specified variables for better evaluation.
• Concepts of Average Treatment Effects (ATE), Average Treatment Effects on the Treated (ATT), and Average Treatment Effects on the Untreated (ATU) are introduced along with their conditions of use.
• These effects quantify the impact of specific treatments applied to different groups.

Selection Bias

• Selection bias occurs when individuals' choices to participate in a treatment influence analyses.
• It highlights the difficulties in evaluating treatment effects and the challenges in drawing causal assessments from observed data.
• It also considers how this bias arises when an intervention/treatment is chosen by individuals who's circumstances are fundamentally different from those not chosen, and this difference influences the outcome assessment.

Conclusion

• A summary of the concepts presented, with core R, tidyverse basics, data loading, running regressions and interpreting outputs covered in the lectures and this introduction section.

This quiz focuses on applied econometric methods using R, essential for analyzing data and making informed decisions in economics and finance. It covers causal analysis, time series modeling, and real-world problem-solving, providing students with hands-on experience in coding and implementation. Active participation is encouraged as students learn crucial econometric tools and techniques.