Applied Econometrics - Introduction Hosein Joshaghani Sharif University of Technology February 3, 2018 1 / 23 Economic Analysis In general, research in economics has the following steps: 1. Start from real world, 2. Data mining, to come up with a questions, puzzle, regularity, or even irregularity, etc. 3. Create some models to explain facts and observations 4. more empirical analysis to test models, 5. expanding economic models to better explain important facts, anomalies, ... 6. more empirical analysis ... 7. expanding economic models ... 2 / 23 Empirical Analysis Two types of empirical analysis: 1. Reduced form estimation: I I I finding correlation between outcome variables identifying cause and effects very few structure on the data 2. Structural Estimation I I I Imposing much more structure to the data Estimating parameters of the model Counterfactual analysis Complete economic research incorporates both reduced form and structural estimation. 3 / 23 Reduced form estimation I Ultimate goal in reduced form estimation is to identify causal effects. But, it is usually very hard! I Randomized Control Trials (RCT) is one solution. I RCT is not always feasible: Monterey policy, banking, ... I Natural experiment is another solution I Endogeneity is the main culprit. 4 / 23 Reduced form estimation: questions I What is the effect of pollution on productivity of firms? I What is the effect of one more year of education on earning of workers? I Does reducing class size cause an improvement in student performance? I Does lowering business property tax rate cause an increase in city economic activity? I Does increasing oil prices reduce private transportation and increase public transportation? I Does expansionary monetary policy decrease poverty in rural areas? I Does free media lead to fewer corruption? I Do good institutions cause higher economic growth? 5 / 23 Reduced form estimation: Methods Main identification tools are: 1. Regression 2. Instrumental Variables (IV) 3. Diff in Diff or even Diff in Diff in Diff 4. Regression Discontinuty Design (RDD) 5. Propensity Score Matching (PSM) 6. Panel Data and Fixed effects Very nice resource for this topic: I Angrist and Pischke (2014). Mastering’metrics I Angrist and Pischke (2008). Mostly harmless econometrics 6 / 23 Reduced form estimation: FAQs Four questions to be asked in any reduced form emprircal research 1. What is the causal relationship of interest? 2. What is the ideal experiment that could be used to capture the causal effect of interest. 3. What is your identification strategy? 4. What is your mode of statistical inference? Required reading: MHE, chapter 1 and 2. 7 / 23 Reduced form estimation This topic will be covered in the second half of the course! 8 / 23 Structural Estimation I Start from theory, impose a sort of structure that you think will match the desired and important facts, I Solve the model for equilibrium, steady state, balanced growth path or etc. I Try to adjust parameters of the model to match some futures of the data. I Use the estimated parameters to study counterfactual. 9 / 23 Structural versus Reduced Form Estimation I A statistical model describes the relation between two or more random variables. Example: Y = X 0 β + I An economic model starts with assumptions about agents’ preferences, constraints, firms’ production functions, some notion of equilibrium, etc.; then it makes predictions about the relation between observable, often endogenous variables. I Structural estimation is an attempt to estimate an economic model’s parameters and assess model fit. Parameters to estimate often include I I I Preference parameters: risk aversion coefficient Technology parameters: production function’s curvature Other time-invariant institutional features: agents’ bargaining power, financing frictions 10 / 23 what does reduced form mean? I For many other researchers reduced form means: I I What is the (causal) effect of X on Y? While structural means: I I I I I Why does X affect Y? What are the magnitudes of the parameters? How well does theory line up with the data? How would the world look if one of the parameters (counterfactually) changed? What would happen if you (counterfactually) shocked the system? 11 / 23 ”Structural” versus ”Design-Based” I the fact that there are advantages and disadvantages makes them complements rather than substitutes I Avoid structural estimation if there is designed based solution for your question I In practice, good research uses both modeling I There are very very few (if any?) structural paper about Iran being explicit about identification and estimation! 12 / 23 ”Structural” versus ”Design-Based” Structural Design-Based Better on External Validity Map from parameters to implications clearer Formalizes conditions for external validity Forces one to think about where data comes from Easier to interpret what parameters mean Better on Internal Validity Map from data to parameters more transparent Requires fewer assumptions Might come from somewhere else Estimates more credible 13 / 23 14 / 23 14 / 23 ”Calibration” versus ”Structural Estimation” I Calibration: I I I I I Take many parameter values from other papers Usually have more parameters than moments → model isn’t identified → can’t put standard errors on parameters Mainly a theoretical exercise Never, ever, say that in front of James Heckman! Structural estimation: I I I Infer parameter values from the data Get standard errors for parameters An empirical exercise Chris Taber: To me, calibration is structural estimation without identification and standard error! 15 / 23 Structural versus Reduced Form Estimation: Questions I Reduced Form: What is the (causal) effect of X on Y? I Structural Estimation: I I I I Typically assumes X causes Y What are the parameters’ magnitudes? How well does theory fit the data? How would the world look if one of the parameters (counterfactually) changed? 16 / 23 Structural versus Reduced Form Estimation: Tools I Reduced Form: I I I Estimators: OLS, IV, DiD, RD Design Software: Stata Structural Estimation: I I Estimators: GMM, SMM, ML, SML Software: Python, Julia, MATLAB, C++, Fortran, etc. 17 / 23 Structural versus Reduced Form Estimation: Benefits Stractural Estimation has at least three benefits: 1. Estimates of interesting economic primitives 2. ”Deep” tests of theory: I I I Testing quantitative, not just directional predictions ”Seeing where models fail opens doors to future research” Example: Mehra and Prescott (1985), equity premium puzzle 3. Can answer interesting counterfactual questions 18 / 23 Structural Estimation and Lucas Critique I Reduced-form papers can also ask counterfactual questions, by changing a regressors from its actual value to a counterfactual value. I But less convincing, since harder to believe ”all else equal”. i.e., Lucas critique is more severe for reduced-form counterfactuals. I Also, impossible to shock primitives in reduced-form. 19 / 23 Disadvantages of Structural Estimation I More assumptions I Harder to do. (honestly much harder!) I Harder to convince general audience WARNING If structural and reduced-form will both get the job done, then go reduced-form!! 20 / 23 Road-map of the course I Part 0: Introduction to the course I Part 1: Estimation methods: OLS, ML, MM and GMM I Part 2: Introduction to Structural Estimation I Part 3: Introduction to Reduced Form Identification Methods 21 / 23 Evaluation I Homework (10-15 psets) 20% I Presentation in one of the reading groups 15% (Extra) I Midterm 40% I Final 40% 22 / 23 Teaching Assistants I Sajad Ghorbani I Ebrahim Aliabadi 23 / 23