There is a level of material involving NDAs so I need to be somewhat vague beyond the lingo. The most efficient skew model is based upon spread premium; index cs/ps. It’s normed and is w/o peer in a modeling environment. Matrices run >10x faster (prob much more than using vols but not granular on time).
I have tick data skew modeling reflecing microstructure (<5% actionable) and persistent skew conditions which can be arbitraged. Once I had the skews it became possible lock skews/switch values (D1/D2 (duration)) that added buying power under PM and TIMS RBH. IOW, the haircut required to hold say, a 300-lot SPX position, can ADD as much as $10K in buying power. Almost always a lower initial req than a box.
Modeling skew in annualized vol-fig is 90’s era.
