Smart Beta versus Dumb Alpha
In “An Aussie sense of style”, Axioma’s latest paper on smart beta products, we take a look at the inherent compromise between delivering target factor purity versus maximizing factor exposure. The decision has to be made at the portfolio construction stage and constraints are the weapon of choice in this battle for investment compliance. Armed with this scalpel of sorts, managers can decide to surgically remove all non-target factors. Alternatively, they may opt to put down tools and let the optimizer find the target exposure wherever it may lie; even if it means acquiring some non-target factor exposures in the process.
More emphasis (certainly more marketing dollars) is usually given to the objective function behind smart beta portfolios, but constraints play a leading role (ultimately exaggerated into a central one), in their performance and predictability. In this paper we construct four variants of each of our smart beta portfolios, as well as two variants of a Low Volatility portfolio. We use the Growth, Momentum, Profitability, and Value factors from our medium horizon Australian equity risk model (AXAU4-MH) as our target smart betas respectively. Our goal is to identify the tradeoffs that come with varying sets of constraints in terms of risk decomposition, target factor exposure and purity, and (where applicable) sector allocation.
The first of these variants simply aimed at quantifying the cost of the long-only constraint on smart beta portfolios. We first constructed a long/short factor-mimicking portfolio on our investment universe (S&P/ASX200) then optimized against it with a long-only constraint to construct our LO-FMP variant. In all four portfolios constructed in this way, the cost of the long-only constraint came in the form of a very low target-factor exposures. So, while this construction methodology delivered portfolio returns that were very similar to the original FMPs, the attribution showed that those were delivered via non-target factor exposures.
The other three variants imposed constraints on both non-target styles and industry factors, or non-target styles alone (i.e. no industry constraints), or none at all (i.e. an unconstrained portfolios). The paper discussed in detail the individual results for each variant on our four style portfolios. One clear consequence of striving for target factor purity via constraints on other systematic sources of risk, is that stock specific risk (i.e. dumb alpha since there was no stock-level research inputs), becomes more and more influential in the performance of these portfolios. Given the randomness of specific returns, the future performance of highly constrained smart beta portfolios becomes hard to predict, even when strong target-factor views exist.
In all four style factors we tested, letting the optimizer do what it does best gave us the strongest results in terms of target-factor exposure and (with the exception of the Value factor) portfolio returns that were the most aligned with their respective style factor returns. Constraints took the portfolio performance decision out of the hands of the systematic beta we targeted and into the whimsical hands of stock specific risk. To paraphrase the Rolling Stones, with constraints, you can’t always get what you want, just so long as you don’t get what you deserve.