Will The Real Factor Index Please Stand Up

Image Source: 
 

Investors face a choice of 3 main approaches when devising factor indexes—selection, tilting and optimization—each with their advantages and limitations.

 Every design element involves tradeoffs—what are your must- haves in a strategy and where do you have flexibility? It all depends on investor objectives.

There is no correct factor index, but some factor applications are more efficient than others. 

Might Just be the Next Best Thing: Factors
The proliferation of factor research has seen the classic Wall Street casino game transformed into a disciplined science.While investors are still free to speculate wildly, advanced mathematics can isolate long term drivers of excess return, what we call factor premia, such as quality, low volatility or size.  This scientific rigor notwithstanding, how factor scoring, and attribution models are blended into cohesive client strategies remains an artform. Confusion often arises over what is the true approach to constructing factor methodologies; here we review three leading strategies used in factor indexes, namely selection, tilting and optimisation.  

Selection: Keep it Simple
The most straightforward approach in designing a factor index is to use a selection or elimination methodology, which circumscribes the starting universe based on stock factor characteristics.This “in or out” operation is the defining feature of a selection strategy and can be thought analogous to the exclusion of alcohol, tobacco and other sin stocks due to social consideration. Selection strategies best succeed when pursuing specific mandates to which other portfolio concerns, such as turnover or tracking error, are clearly secondary—devising high dividend indexes is a prime example.The FTSE Dividend Growth with Quality Overlay Index family, for instance, implements a series of selection screens, including the selection of companies in the top quartile for forward dividend yield, and the elimination of equities in the bottom 10% for quality z-score. This successive slicing and dicing of the starting universe excels in either capturing or excluding tail-end factor risks; often allocators are more concerned with extreme results or one-sided skews than the entire distribution of factor loads.Factor Selection/EliminationThe challenge with selection, notwithstanding its transparency, is its lack of precision and efficiency.While an investor can readily discern a stock is excluded because it is in the top 15% for richness of valuation, for example, how much difference is there between the 15th and 16th percentile for valuation? Selection will overweight the latter and zero-weight the former, when there is slight difference in factor scores. Absent further rules, selection will treat the 15th and 1st percentiles for valuation equally, even though there are vast differences in factor loads.In other words, selection fails the commonsense proposition that likes should be treated alike, and non-likes should be treated disparately.

Tilting: unleveling the playing field
Moving from the most blunt to the most subtle of factor methodologies, tilting takes a starting universe (typically market cap) and reweights companies according to the intensity of their factor scores. There are many variations to tilt strategies, and indeed FTSE Russell operates two methodologies in this category. These tilt models include the legacy Fixed Tilt engine and the Target Exposure methodology, the latter of which dynamically reallocates based on factor intensity. Factor TiltThe most compelling feature of tilting is how the approach maintains fidelity to the benchmark, preserving the scope and purpose of the original asset class, while layering on factor premia. If in investor’s mandate includes minimising tracking error, the efficient use of active share will render tilt strategies quite attractive.Indeed, FTSE Russell’s tilt methodologies feature factor scores as a direct bolt-on to market cap weights, before being renormalised by the gaussian error or exponential functions.Best suited to low to moderate factor exposures, tilt indexes generally exhibit very high correlations to their starting universes (R stats of 0.95 or greater), and hence are popular in smart beta applications.The factor combinations and intensities can be endless customised—the Russell 1000 2Qual/Val Index for instance simultaneously tilts to the quality and value factors in a two to one ratio.Moreover, the proof is in the pudding; the fundamental portfolio characteristics shift in both directions—ROA increases while Price to Sales decreases.

Factor Indexes 101

  Selection Tilting Optimisation Process Subset portfolio by factor score Reweight by factor score Redefine opportunity set by factor score Advantages Highly intuitive, can pursue aggressive solutions Preserves benchmark fidelity Leverages correlations and volatility Disadvantages Inefficient use of active share Struggles going beyond moderate factor loads Non-intuitive, high dependence on past trading histories Tracking Error High Low Moderate Transparency Very High High Low Common Uses Dividends strategies, eliminating tail end results Single, multifactor models Multifactor models, concentrated baskets

The key limitation to tilt-based strategies is they struggle with very high factor loads.An allocator looking to make concentrated factor bets would look at the middle third of the index with roughly market cap weight as deadweight—why are large portions of my investment not expressing a factor view point? If a concentrated basket of 30 to 50 stocks is desired, then this is where optimization enters the equation.  

Optimisation: The 4-D Chess Approach
Whereas tilting begins with the benchmark and applies factor loads, optimization starts with the factor portfolio in mind and moves back towards the benchmark.The killer app to optimisation is its ability to leverage cross asset correlations and volatility measures to identify the mathematically ideal portfolio—it is a line of best fit on steroids.By unlocking these intra-asset efficiencies, optimisation can push the frontier of factor portfolios into truly creative territory, especially with aggressive multi-factor strategies.This potential does come at a considerable cost.Foremost, these complex mathematical models are effectively opaque to the uninitiated, and can deliver counter-intuitive results.Let’s suppose two stocks score very highly on the desired factor scores and are strongly correlated as well—but the first exhibits a better risk reward profile than the second.An optimiser will typically underweight the latter to free up allocation to heavily overweight the less volatile stock. This approach will see company pairs, such as Exxon and Chevron or Facebook and Google, receive disparate treatment even if their factor loads are similar. Alternatively, a low factor company can receive considerable weight if it is uncorrelated to the desired factor characteristics and provides diversification value, or if the stock helps replicate the benchmark.What is optimal is not always desirable, as optimisation is guided by constraint and is heavily reliant on past performance.When market conditions change, the highly calibrated model may be poorly positioned.This is not to say that optimisation cannot generate powerful results, but the process of implementation can be confounding.

I Guess There is a Factor in all of us
To unlock the power of factors, the ideal approach is to navigate the tradeoffs of each construction method until finding a suitable solution; it is not a matter of absolute best, but best fit for purpose.Whether selection, tilting, optimisation or even a unique hybrid strategy, the real factor is leveraging the modern tools of finance to empower client investment objectives.Calculating factors is a steadfast science, but their application remains a high concept artform.More By This Author: