Weigh full replication against stratified sampling and optimisation as ways to build an index-tracking equity portfolio

Full replication holds every constituent at its index weight; it gives the lowest tracking error and suits indexes with a modest number of liquid stocks, but costs rise when there are many small or illiquid names. Stratified sampling divides the index into cells, for example sector by size, and holds a sample of stocks in each cell so that cell weights match the index; it suits broad indexes and smaller portfolios. Optimisation uses a factor risk model to choose holdings that minimise expected tracking error; it copes with large universes but relies on historical covariances and needs regular re-optimisation. The trap is assuming replication is always best; with many illiquid constituents its trading costs can outweigh sampling error.