suppose that there are
- observations of stocks return for firm i in 1..N
- date of observations t in 1..T
The error structure is such that cov(epsilon_it,epsilon_jt) are far from zero. An important statistical issue is that firm returns are cross-sectionally correlated at the same time, because there are macroeconomic news that impact the whole economy simultaneously which push up all stock prices or push down all stock prices in the same day. In such a setup:
- Is it still valid to use Adaptive-LASSO? I understand from Medeiros & Mendes Journal of Econometrics paper that adaptive-lasso can handle autocorrelations and retain oracle property, but they are silent about cross-sectional correlation
- What is the valid LASSO-like method to use in this application? Is it Zhentao&Philips Econometrica 2016, but it doesn't seem that relevant as I'm not interested group classification, just interested in variable selection.