It is common wisdom in the industry that not all alpha factors make good risk factors. There is, however, plenty of evidence in the literature that some of the style risk factors commonly found in fundamental multi-factor models accurately capture factor premia that have outperformed broad capitalisation-weighted benchmarks over long periods of time.
Some of these factors include growth, momentum, quality/profitability, value, and the low volatility anomaly.
In response to investors' demand for these factor premia, asset managers have released families of Smart Beta products aimed at capturing one or more of them. Axioma has written at length about these attempts and in this note we take a closer look at the compromises involved in constructing a viable smart beta product.
The key portfolio construction issue we are focusing on here is how to balance our desire for target-factor purity with our goal of achieving a high exposure to the target factor. Can we have both?
The poster child for this goal is the utopian ideal of a factor-mimicking portfolio (FMP). These FMPs are long-short dollar neutral portfolio constructed over the risk model's estimation universe and used by model builders to represent pure factor returns. These portfolios are often not investible and incur a lot of turnover across periods. They are therefore not suitable for a long-only manager wishing to construct a low-cost smart beta product.
The first portfolio construction strategy ('LO-FMP') we will use will simply tries to 'replicate' a reconstructed FMP portfolio subject to a long-only constraint by using the S&P/ASX 200 as investment universe. We want to see how close to the long-short FMP we can get and what is the implicit cost of the long-only constraint, both in terms of target factor exposure obtained as well as the return delivered by this optimised portfolio versus the FMP return we seek to capture.
We fully accept that under this strategy, unlike the attribution of a pure FMP, the total return of the portfolio will come from multiple sources (i.e. other styles, industry exposures, and specific risk as we are not imposing any constraints on the optimisation).