Home / Wiki / P / Physical ETF Replication

Physical ETF Replication

Investment and Capital MarketsUpdated 17 Nov 2026

Physical ETF replication is the approach in which an exchange-traded fund tracks its benchmark by buying and holding the securities that make up the index. A physically replicated equity ETF owns shares in the index companies, and a physically replicated bond ETF owns the bonds, held for the fund by a custodian bank. It is the most common way ETFs are built and contrasts with synthetic replication, in which the fund receives the index return from a swap counterparty.

Physical replication comes in two main forms. Full replication holds every index constituent in its index weight. Sampling, also called optimised or representative sampling, holds a subset of constituents chosen so that the portfolio's characteristics match the index closely. The choice depends on how many securities the index contains, how liquid they are and how expensive they are to trade. Most large equity ETFs in the United States use physical replication, and it has become the dominant method in Europe too.

01

Full replication

Under full replication, the fund owns each security in proportion to its weight in the index. An ETF tracking the S&P 500 would hold all of the index's roughly 500 companies, adjusting its holdings whenever the index changes membership or weights. Full replication gives the closest tracking where the securities are liquid and the index has a manageable number of members. It is transparent, because the fund's holdings mirror the published index.

The method becomes costly for indices with thousands of constituents, many small or illiquid members, or securities in markets with high trading costs or restrictions on foreign ownership. In those cases buying every security can cost more than it gains in accuracy.

02

Sampling and optimisation

Sampling selects a subset of the index that reproduces its main risk characteristics. In stratified sampling, the manager divides the index into cells, for example by sector, country and company size for equities, or by maturity, credit rating and issuer type for bonds, and holds securities in each cell so the fund's exposures match the index. Optimised sampling uses a risk model to choose the holdings and weights that minimise expected tracking error subject to limits on cost and the number of positions.

Sampling is standard for broad bond ETFs, because bond indices can contain thousands of issues, many of which trade rarely. It is also used for broad global and emerging market equity indices. The trade-off is a higher risk of short-term deviation from the index in return for lower trading costs.

03

Sources of tracking difference

Even a fully replicated fund does not match its index exactly. The main reasons are:

These effects together make up the fund's tracking difference, the gap between fund and index returns over a period, while tracking error measures how much that gap varies.

04

Securities lending in physical ETFs

Many physically replicated ETFs lend part of their holdings to brokers and other borrowers in return for a fee and collateral. The practice can earn meaningful income, especially on stocks in high demand from short sellers, and it is one reason some funds track their index more closely than their fees alone would suggest. It introduces a limited counterparty risk if a borrower defaults and the collateral falls short. ETF securities lending policies, collateral rules and revenue splits are disclosed in fund documents.

05

Physical versus synthetic replication

Physical replication's main advantages are transparency and ownership of real assets, which limits counterparty exposure to securities lending. Its disadvantages are trading costs and difficulty tracking markets that are hard or expensive to access. Synthetic ETFs can track such markets more precisely but take on swap counterparty risk. After regulators raised concerns about synthetic ETFs around 2011, European investors shifted strongly towards physical funds, and the larger providers converted many products to physical replication.

Last updated 17 Nov 2026