In the last 12 months we’ve witnessed a profound uptick in the adoption and implementation of multi-factor equity programs – by clients that run the gamut from the smallest pension plans to the largest sovereign wealth funds in the world. Naturally, the types of questions these clients are asking, has correspondingly evolved and much interest is now seen in the nuances and vagaries of program design. While these questions are varied and, often, quite complex, we find many warrant essentially the same answer. To our clients we, of course, sound like a broken record but what has emerged is recognition of what we believe to be the single most important principle of multi-factor program design: factor efficiency and stability.
Specifically, these five questions are increasingly common. Paraphrasing actual client communications:
- “I’ve read a lot about so-called ‘top down’ versus ‘bottoms up’ approaches to building multi-factor programs. Recently a paper by Clarke, De Silva and Thorley (2016) suggests the ‘bottoms up’ approach produces superior performance. Is this true?”
- “When putting together a ‘top-down’ multi-factor program does it make much difference which single-factor vehicles I use? Isn’t the point to diversify factor exposures anyway?”
- “I’m evaluating a few different factor-based strategies. I’m not sure I really understand the differences and I’m tempted to choose the one with the best track record. Is this the right way to go?”
- “In the past I’ve always diversified my fundamental active managers. We’re moving to factors and it seems like I should diversify here too, e.g., have more than one value manager, more than one quality manager, etc. Any thoughts on this?”
- “I’m implementing value and quality sleeves in my ‘top down’ factor program. Provider ABC’s value and quality products have a correlation of -0.40 while provider XYZ’s value and quality products have a correlation of 0.03. It seems provider ABC’s products offer better diversification because they are more negatively correlated. Right?”
Short answers: 1) Not necessarily; 2) Yes, it makes a tremendous difference; 3) Absolutely not; 4) No, not if you do things right; 5) Not necessarily
Somewhat longer answer: all of these questions depend on the efficiency and stability of factor exposure…
…but first, perhaps we should define a few terms:
Top-down vs. Bottoms-up: Conceptually I can construct a multi-factor program in two ways. First, I might implement separate, independent factor sleeves where each factor is captured individually, e.g., a small size sleeve that is separate from a value sleeve, etc. Each sleeve could, in principle, be fulfilled by a distinct manager or implemented with a distinct index vehicle. This approach has been termed the ‘top down’ approach by some authors and in past webinars and blogs we’ve referred to it as the ‘sleeved’ approach.
Two ways to construct a multi-factor program: Top-down vs. Bottoms-up. First, implement separate, independent factor sleeves where each factor is captured individually. Second, construct a single vehicle that targets several factors simultaneously.
Second, I might construct a single vehicle that targets several factors simultaneously, i.e., one strategy that gives me exposure to size, value, quality, etc. The objective here is to find stocks that are at the intersection of the desired factors or, in other words, rank highly on all factor criteria. While this approach has been deemed the ‘bottom up’ approach, in previous webinars and blogs we’ve also referred to it as the ‘intersection’ approach.
Factor Efficiency: When tilting toward any factor there is a strong tendency to take unintended and potentially uncompensated or even negatively compensated biases to certain sectors, regions, other style factors and idiosyncratic risk. For example, many low volatility strategies have very significant exposures to the utilities and consumer staples sectors, a large capitalization bias and a negative value bias, i.e., low volatility stocks are expensive. Hunstad and Dekhayser (2015 & 2016) show these unintended bets add materially to risk but don’t add to, or potentially even detract from, return. Factor strategies that don’t eliminate these unintended bets in their design process are taking more risk than necessary and are thus deemed ‘inefficient’. The authors analyze a cross-section of popular factor-based equity strategies and show that efficiency is strongly related to risk-adjusted returns. Unfortunately, the authors also find most of the strategies analyzed were remarkably inefficient with, on average, 83% of their risk being unintended and uncompensated. As a result, their performance was dominated not by the intended factor exposure but by the 83% of risk that was extraneous.
Factor Stability: Even if a strategy is successful at capturing an intended factor exposure this exposure, and thus the potency of the strategy, can change through time and is deemed ‘unstable’. In a previous blog we suggested this is an all-too-common occurrence and found a proximate cause for instability in low factor efficiency and poor product design. In other words, low-efficiency factor-based strategies tend to have unstable intended factor exposures.
Q #1: Is ‘bottom-up’ approach’ produces superior results compared to ‘top-down approach’? : Let’s use a simple example of targeting just two factors, say, quality and value. In a ‘top down’ approach we would implement this multi-factor program using separate vehicles for each factor. The problem is that if factor vehicles are not efficient there is a very strong tendency for performance dilution due to impurity. For example, high value stocks tend to be of lower quality and, likewise, high quality stocks tend to be of lower value. In capturing the quality and value factors these low quality and low value biases are clearly unintended and negatively compensated and thus reflect impurities, i.e., a drag on efficiency. It’s somewhat ironic but in our attempt to target the high quality and high value factors, if our vehicles are inefficient we may end up owning a large number of low quality and low value names. This result generalizes beyond value and quality to virtually any combination of factors.
The ‘bottoms up’ approach, in contrast, attempts to target stocks that are simultaneously high quality and high value and, thus, potentially increases efficiency and reduces performance dilution. In other words, the joint factor construction makes the introduction of unintended biases more difficult.
Clarke, De Silva and Thorley (2016) study these two approaches and conclude the ‘bottoms up’ design of multi-factor programs has better performance than ‘top down’ and is therefore superior. While we don’t necessarily disagree with their analysis, we must note the authors failed to recognize the performance gap between ‘top down’ and ‘bottom up’ approaches were due to dilution, i.e., the direct result of their choice of low-efficiency single factor vehicles. Had the authors selected higher efficiency factor sleeves the performance gap between ‘top down’ and ‘bottom up’ approaches would have narrowed significantly. In fact, if very efficient factor sleeves are utilized there should be no difference in the performance of ‘top down’ and ‘bottom up’ approaches.
Answer #1: Both ‘top down’ and ‘bottom up’ approaches have merit. The key is to choose high-efficiency vehicles.
Q #2: In ‘top-down’ multi-factor program does it make much difference which single-factor vehicles we use?: A common retort to Answer #1 is that within ‘top down’ approaches the individual factor sleeves need not be individually efficient because their unintended biases might somehow cancel each other. In fact, this is the position taken by at least one major factor index provider. In other words, it doesn’t really matter how you construct factor vehicles as long as they each have some intended factor content.
In a previous blog we showed that low-efficiency factor vehicles also have highly unstable biases: both intended and unintended. As a result, any perceived diversification benefit of using impure factors and hoping for cancellation of uncompensated and negatively compensated exposures is likewise unstable. By leaving it to chance we are taking a big gamble on the ability of factor vehicles to somehow offset each other’s impurities.
Answer #2: In this gamble you’ll only lose. The key is to choose high-efficiency vehicles.
Q #3: Is track record the best way to choose the right ‘factor-strategy’?: Yikes! This all-too-common question illustrates why a thorough understanding of the efficiency concept is essential. Above we noted a study by Hunstad and Dekhayser (2015 & 2016) which found 83% of active risk taken by popular factor-based equity products was unintended. This suggests the bulk of the performance variation across factor-based products is due to unintended, often random bets rather than the contribution of intended factor content. In other words, some products may have good track records simply because their unintended bets paid off. A better method of choosing factor products is to evaluate the stability and purity of intended exposures.
Answer #3: Again, in this gamble you’ll eventually loose. The key is to choose high-efficiency vehicles.
Q #4: Is diversification of factor-based strategies the right way to manage uncompensated risk?: This is similar to Question #2. While we have a natural inclination to diversify managers we must ask ourselves what it is we are hoping to diversify. For traditional active managers the answer is typically – to use our terminology – uncompensated bets. In their pursuit of idiosyncratic returns fundamental managers often cannot avoid large biases to certain sectors, regions, negatively compensated style factors, etc. While we could make the same case for factor-based strategies I would simply ask: why are these strategies taking uncompensated bets? Well-designed factor products should be highly efficient have little or no uncompensated bets to diversify.
Answer #4: Highly efficient strategies deliver pure exposure to intended risks and require no diversification within individual factors. The key is to choose high-efficiency vehicles.
Q #5: Is correlation the right criteria to choose a company for investment?: Inefficiency can manifest itself in negative correlations between factors, i.e., negative correlation can be arbitrarily manufactured by increasing inefficiency. Using the quality and value example, one can easily drive the correlations of quality and value sleeves toward -1 by lowering the quality content of the value sleeve and/or lowering the value content of the quality sleeve. At the extreme I can achieve perfectly offsetting value and quality positions between the two vehicles simply by introducing inefficiency. This result also generalizes to any combination of factors.
Answer #5: Don’t be fooled by correlations. The key is to choose high-efficiency vehicles.
References
- Clarke, Roger, Harindra De Silva, and Steven Thorley. "Fundamentals of efficient factor investing." Financial Analysts Journal 72.6 (2016): 9-26.
- Hunstad, Michael, and Jordan Dekhayser. "Evaluating the Efficiency of “Smart Beta” Indexes." The Journal of Index Investing 6.1 (2015): 111-121.
Hunstad, Michael, and Jordan Dekhayser. "Investors Care about the Purity of Factor Indexes: A Reply." The Journal of Index Investing 7.1 (2016): 14-16.Grinold, Richard C. "The fundamental law of active management." The Journal of Portfolio Management 15.3 (1989): 30-37