Ensemble Methods For Causal Effects in Panel Data Settings

1 Intro

Combinations of prediction methods (“ensembles”) perform better than individual methods.

2 A Simple Example

Two independent variables that are independent of each other and one dependent variable. There are multiple ways to combine them.

3 Methods for Synthetic Control Problems

Predict the missing value.

There are three methods to do it. Synthetic control methods (VT)

Unconfoundedness or Horizontal Regression (HZ)

Matrix Completion (MC) Methods

4 Ensemble Methods

Methods to combine these three synthetic control methods together:

Vertical Crossvalidation (VC)

Horizontal Crossvalidation (HC)

5 An Application

Three different outcomes: state GDP, log GDP and GDP growth rates. N = 51 and T = {10, 25, 100, 270}. Compute average root-mean-squared-error for in-sample predictions. Ensemble methods do well compared to individual methods separately.