Robust Synthetic Control

Authors: Muhammad Amjad, Devavrat Shah, Dennis Shen

JMLR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments, using both synthetic and real-world datasets, demonstrate that our robust generalization yields an improvement over the classical synthetic control method.
Researcher Affiliation Academia Muhammad Amjad EMAIL Operations Research Center Massachusetts Institute of Technology Cambridge, MA 02139, USA; Devavrat Shah EMAIL Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology Cambridge, MA 02139, USA; Dennis Shen EMAIL Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology Cambridge, MA 02139, USA. All affiliations point to Massachusetts Institute of Technology, an academic institution.
Pseudocode Yes Algorithm 1 Robust synthetic control; Algorithm 2 Bayesian robust synthetic control
Open Source Code No The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes We conduct two sets of experiments: (a) on existing case studies from real world datasets referenced in Abadie et al. (2010, 2011); Abadie and Gardeazabal (2003), and (b) on synthetically generated data. For Basque Country: We only use as data the per-capita GDP (outcome variable) of 17 Spanish regions from 1955-1997. Referenced Abadie and Gardeazabal (2003). For California Anti-tobacco Legislation: we use the annual per-capita cigarette consumption at the state-level for all 50 states in the United States, as well as the District of Columbia, from 1970-2015. Referenced Abadie et al. (2010).
Dataset Splits Yes Without loss of generality, let the first unit represent the treatment unit exposed to the intervention of interest at time t = T0 + 1. The remaining donor units, 2 i N, are unaffected by the intervention for the entire time period [T] = {1, . . . , T}. Let T0 as the number of pre-intervention periods with 1 T0 < T, rendering T T0 as the length of the post-intervention stage. For synthetic simulations: N = 100, T = 2000, while assuming the treatment was performed at t = 1600.
Hardware Specification No The paper mentions "Spark (through alternative least squares) and Tensor-Flow come with built-in SVD implementations" and "computational infrastructure", but no specific GPU/CPU models, memory details, or other hardware specifications are provided.
Software Dependencies No The paper mentions "Spark (through alternative least squares) and Tensor-Flow" as tools that can perform SVD, but it does not specify any version numbers for these or other software dependencies.
Experiment Setup Yes The algorithm utilizes two hyperparameters: (1) a thresholding hyperparameter µ 0... and (2) a regularization hyperameter η 0. We will employ three different learning procedures as described in the robust synthetic control algorithm: (1) linear regression (η = 0), (2) ridge regression (η > 0), and (3) LASSO (ζ > 0). In order to choose an appropriate choice of the prior parameter α, we first use forward-chaining for the ridge regression setting to find the optimal regularization hyperparameter η. ... we choose α = η/ˆσ2 where η is the value obtained via forward chaining.