Optimization vs. Full Replication

As ETFs evolve into newer and narrower investment categories, tracking error-or the difference between an ETF's returns and those of its benchmark index-has become increasingly important. One key factor impacting tracking error is index construction.

ETFs tracking highly liquid indices and assets are often constructed via full replication, holding a portfolio of securities that exactly matches the benchmark in composition and weight. By its very nature, a full-replication ETF should show minimal tracking error (although discrepancies sometimes arise, due to management and transaction expenses).

Sometimes, however, full replication is infeasible, due to the illiquidity of the underlying securities, the sheer size of the index or country-specific tax laws for foreign holdings. In some cases, fund managers may also be limited by diversification regulations, which generally prohibit mutual funds (including ETFs) from concentrating more than 25% of their total assets in any one security. This makes full replication impossible for ETFs tracking narrow sectors or industries, such as telecom, which are dominated by one particular company or a small handful of companies. Full replication can also be highly problematic where an index is itself massive and consists of securities with varying liquidity-a category that includes most major broad-market bond indexes.

The alternative to full replication is to construct ETFs using index optimization, relying on mathematical models that attempt to build a smaller representative portfolio that mimics the performance characteristics of the broader benchmark. The most common method is simple representative sampling, wherein fund managers select a subset of securities that best embody the investment characteristics, fundamentals and liquidity of the index as a whole.

While optimization can be a cost-effective approximation, and can improve the trading characteristics of the ETF itself, it also inherently increases the chances of tracking error. Consider two ETFs tracking the MSCI Emerging Markets Index: the Vanguard Emerging Markets ETF (NYSE Arca: VWO) and the iShares MSCI Emerging Markets ETF (NYSE Arca: EEM).

Both track the same index, but while VWO holds 95% of the names in the benchmark, EEM relies on a representative sampling methodology that selects the 329 largest and most liquid emerging market stocks from the index's 741 securities (as of June 30, 2009).

Over the past several years, the more fully replicated VWO has done a much better job of tracking the MSCI Emerging Markets Index.

Historical Annual Returns Data courtesy of Morningstar, MSCI Barra

Of course, tracking error cuts both ways: During the market crisis of 2008, EEM outperformed both VWO and its benchmark index.

Optimization is neither good nor bad in a vacuum. Rather, it is a choice made by an investment manager on how a portfolio will be run-one that has surprisingly large impacts on investor experience, despite the fact many investors are unaware of the very concept. With the simple choice of optimization strategy driving a greater-than 5% variance in our emerging markets example, picking the right implementation can be almost more important than picking the right index.