Halts, Delistings and Gaps¶
Prices go missing: before a listing, during a halt, after a delisting, on a vendor's bad day. qex treats every gap the same way, on the signal side and in P&L.
Returns across a gap¶
Every return in qex follows one rule (price_returns):
- a return is the ratio of two observed prices; a day with no price has no return, not 0%;
- a missing start price is replaced by the last price before it, so the day a halted stock trades again carries the whole move since its last price;
- a price more than 10 rows old (
MAX_PRICE_GAP) no longer counts, so a stock back after a long suspension gets no return for that day rather than one enormous one.
So a halted or delisted stock drops out of rank and the group operators instead of counting as a 0% day, and
volatility and ts_beta don't see fake quiet days. The rule gives the same answer on pandas 2 and 3.
Positions across a gap¶
A stock with no price can't be traded. Backtest holds the position from the stock's last priced day
(hold_untradable), marked at its last price, so P&L is 0 during a halt and the move across it is booked when the
stock trades again. A position is held for at most 10 days with no price; after that (a delisting, or a long
suspension) it is dropped with no P&L for the missing move.
Delisting returns are not modelled. A stock that delists after bad news usually loses part of its last price on the way out; qex books nothing for it. That flatters a long book in distressed stocks. If your data has delisting returns, apply them before running a backtest.
Held weights are not rescaled, so on a day a halted stock is held, the book's gross can be a little above or below 1.
build_book itself still returns the exact target book. See Books.