The credit one caught my eye. 73% of the bond returns in the first five days of the month is the same shape I see at trade level: most of the value shows up early, and if you're late you're holding a different trade. My question for the paper would be costs. Does it survive actually being positioned for those five days every month? Corporate bonds are not cheap to trade.
Read together, three of these papers describe one failure: a hard binary sitting where everyone assumes a smooth curve. In sample gating, a classifier flips from calm to turbulent and swaps the whole estimation window overnight. In oil and nickel, a delivery constraint flips traders from "can stay" to "must leave" and swaps the whole pool of counterparties. Hydari's fix, weighting by probability instead of committing, hints at a market analogue: exits that phase in by degree rather than all at once should produce smaller cliffs than an 87% or 93% imbalance. The bond result comes at it from the other side. If the premium arrives on a calendar, the regime worth conditioning on may simply be the date, a variable no classifier is watching.
Striking that roughly 70% of the monthly credit return lands in the first five trading days. It mirrors the turn-of-month effect in equities. Does the paper put it down to month-start inflows and index rebalancing, or something structural in dealer balance sheets? I'd also want to know whether it held up after 2008, when dealer capacity changed.
The credit one caught my eye. 73% of the bond returns in the first five days of the month is the same shape I see at trade level: most of the value shows up early, and if you're late you're holding a different trade. My question for the paper would be costs. Does it survive actually being positioned for those five days every month? Corporate bonds are not cheap to trade.
Read together, three of these papers describe one failure: a hard binary sitting where everyone assumes a smooth curve. In sample gating, a classifier flips from calm to turbulent and swaps the whole estimation window overnight. In oil and nickel, a delivery constraint flips traders from "can stay" to "must leave" and swaps the whole pool of counterparties. Hydari's fix, weighting by probability instead of committing, hints at a market analogue: exits that phase in by degree rather than all at once should produce smaller cliffs than an 87% or 93% imbalance. The bond result comes at it from the other side. If the premium arrives on a calendar, the regime worth conditioning on may simply be the date, a variable no classifier is watching.
Striking that roughly 70% of the monthly credit return lands in the first five trading days. It mirrors the turn-of-month effect in equities. Does the paper put it down to month-start inflows and index rebalancing, or something structural in dealer balance sheets? I'd also want to know whether it held up after 2008, when dealer capacity changed.