Finance · Benchmark note
The certificate is evaluated on four collections of portfolio-selection problems, all built from real market data.
The problem underneath every one of them is the same one described on this site: hold a limited number of names out of a universe, in chosen amounts, at the best trade-off between return and risk we can prove. What changes from one collection to the next is where the numbers come from, and how large a universe they let us test.
Public benchmark · OR-Library
The standard, most-cited test set for cardinality-constrained portfolio selection: real covariances built from the Hang Seng, DAX, FTSE 100, S&P 100 and Nikkei 225. Small by institutional standards, but the set every published result in this area gets compared against.
Public benchmark · Frangioni & Gentile
A mean-variance benchmark distributed with a known upper and lower bound at a fixed cardinality. Its problems add minimum buy-in sizes to the cardinality limit.
Real market data · Cesarone et al.
Sample covariances built from real historical prices, not generated — from the Dow Jones up to a 1,203-name NASDAQ composite. Two of the six are rank-deficient, because there are fewer weeks of history than assets, which is typical of real covariances at that size.
Our own pull · a real market at institutional scale
Built because every collection above stops well short of institutional size. Current NASDAQ and NYSE common stock, filtered to names with a complete five-year weekly history and a real, adjusted price — no forward-filled gaps. It carries the survivorship bias of any backfilled universe.