the GMV portfolio about pairs HOT 4 OPEN

also here for implementation:

http://enricoschumann.net/R/minvar.htm

from pairs.

because its a regression
https://machinelearningmastery.com/penalized-regression-in-r/

from pairs.

and a lasso benchmark
https://github.com/YaohuiZeng/biglasso

this would work for 2K stocks

from pairs.

these produce the minimum variance portfolio:
a1<-solve(cov(x), numeric(ncol(x)) + 1) %>% {total_gross*./sum(.)}

require(lars)
a2<-lars(x[ , 1L] - x[ , 2:ncol(x)],x[,1,drop=FALSE] , type="lasso",max.steps = 10)
a3<-tail(coef(a2),1)[1,]%>%
{total_gross*c(1 - sum(.),.)}

from pairs.