Panel parametric, semiparametric, and nonparametric construction of counterfactuals

Abstract

We consider panel parametric, semiparametric and nonparametric methods of constructing counterfactuals. We show through extensive simulations that no method is able to dominate other methods in all circumstances, since the true data-generating process is typically unknown. We therefore also suggest a model-averaging method as a robust method to generate counterfactuals. As an illustration of the sensitivity of counterfactual construction, we reexamine the impact of California's Tobacco Control Program on per capita cigarette consumption and election day registration (EDR) laws on voters' turnout by different methods.

Can a time-varying structure provide a more robust panel construction of counterfactuals-straitjacket or straitjackets?

“…When the parameters are constant (D1-D3), CPM generally outperforms TVM. In line with the results in Hsiao and Wan (2014) and Hsiao and Zhou (2019), both RMSE and bias based on averaging are smaller than those of PDA. When the parameters evolve over time, the RMSE and bias of CPM are slightly smaller than those of TVM meaning that the estimation error outweighs the need to model the time-varying pattern.…”

“…There are many panel data methods to construct counterfactuals -parametric (e.g., Xu 2017), semi-parametric (Hsiao and Zhou 2019) or nonparametric (e.g., Abadie et. al 2010;Hsiao et.…”

“…We shall use the remaining units as controls to predictȳ t using CPM, TVM and PDA. Since Hsiao and Zhou (2019) found that using simple average of the forecasts forȳ t is more robust in terms of predicting the counterfactual than using the control units selected by a specific model selection criterion, and the performance of time-varying estimation is likely to be deteriorated when the model size increases, we restrict the comparison to models generated by a single predictive variable. Our choice for predictive variable is the…”

“…Using annual data between 1980 and 2000 (T 21) of N 38 U.S. states Hsiao and Zhou (2019), studied the impact of California tobacco control program since 1989 (T 1 9) on per-capita health expenditure. The data also contains per-capita income yielding a total of 76 control variables.…”

Using Panel Data for Macroeconomic Policy Evaluation: A Survey

“…They show that (2) are generically biased, derive support conditions for SCM and use Monte Carlo to compare the methods, Wan et al (2018) compare the panel time series approach and the synthetic control method. Hsiao & Zhou (2019) compare parametric, semi-parametric and non-parametric methods. SCM & PDA are non-parametric methods.…”