Blog Post number 2
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of loremipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Posts by Tags
category1
category2
Blog Post number 2
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of loremipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
causal infernce
Synthetic Controls in Dense Settings: the $\ell_2$ relaxer
Published:
Plenty of posts have been done in the last decade on the synthetic control method and related approaches; folks from Microsoft, Databricks, Uber, Amazon, Netflix, Gainwell Technologies, and elsewhere have gone over it, detailing different aspects of the method. Many (not of course not all) of these go over the standard SCM. Broadly, the original SCM tends to favor, under certain technical conditions, a sparse set of control units being the underlying weights that reconstruct the factor loadings/observed values of the treated unit, pre-intervention. And while sparsity has plenty of appealing properties, such as our ability to interpret the synthetic control, sometimes this advice simply breaks down because in some cases the sparsity notion is wrong. In other words, most of the coefficeints being 0 is a notion that cannot be defended. So in this post, I demonstrate the $\ell_2$ panel data approach, an econometric methodology very recently developed by Zhentao Shi and Yishu Wang which accommodates a dense data generation processes, or when the true vector of coefficient is mostly not zero. The Python code for these results may be found here.
cool posts
Blog Post number 2
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of loremipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
synthetic controls
Synthetic Controls in Dense Settings: the $\ell_2$ relaxer
Published:
Plenty of posts have been done in the last decade on the synthetic control method and related approaches; folks from Microsoft, Databricks, Uber, Amazon, Netflix, Gainwell Technologies, and elsewhere have gone over it, detailing different aspects of the method. Many (not of course not all) of these go over the standard SCM. Broadly, the original SCM tends to favor, under certain technical conditions, a sparse set of control units being the underlying weights that reconstruct the factor loadings/observed values of the treated unit, pre-intervention. And while sparsity has plenty of appealing properties, such as our ability to interpret the synthetic control, sometimes this advice simply breaks down because in some cases the sparsity notion is wrong. In other words, most of the coefficeints being 0 is a notion that cannot be defended. So in this post, I demonstrate the $\ell_2$ panel data approach, an econometric methodology very recently developed by Zhentao Shi and Yishu Wang which accommodates a dense data generation processes, or when the true vector of coefficient is mostly not zero. The Python code for these results may be found here.