AQR Capital Management: The Ominous Shapes of Strategy

“The cave you fear to enter holds the treasure you seek.” – Joseph Campbell

 

For 12 straight years beginning Q4 2001, AQR Capital Management, LLC (AQR) – one of the great and legendary quant hedge funds of the current era – grew equity positions until peaking at 2,346 (long equity) positions by Q4 2013. Since that time, AQR’s long US equity book has found an ominously consistent plateau averaging 2,140 positions. Here, in what would normally seem to be a benign factoid, lie the seeds of the story for why AQR has been suffering performance challenges of late; and, apparently, performance challenges for the foreseeable future according to co-founder and front-man, Cliff Asness.

We start that story with the exhibit, below, where Alphacution presents the full 72-quarter record of total 13F (long) positions for the lineage of AQR Capital Management entities beginning Q4 2001 and ending Q3 2019. With these first shapes, we want to highlight that stocks are the dominant product class, thereby implying that there is little application of other product classes, including ETFs. And, there are rarely any option positions. Translation: The dominant product class is typically the dominant engine of performance.

Moreover, we also want to highlight that AQR appears to have discovered and is maintaining a minimum liquidity threshold currently marked at 2,000 – 2,100 names on the long side of the book (which, by default in an absolute return strategy, would be similar to the short side) for the 5+ years since about Q4 2014 after apparently running afoul of that line during the long pre-Q4 2014 AUM ramp-up. This threshold is significant, as will be supported below…

For the longest time, I have been eager to model the data on AQR to see if it fits into our ecosystem map the way I have been expecting it would. The question that we have been eager to answer, and continue to re-examine going forward, is: What is the maximum amount of capital that can be deployed by an automated, quantitatively-based strategy platform? Another way to think of this question is: What is the highest capacity of the spectrum of relative-value, absolute-return strategies? Our thinking has been that AQR is among those few that can lead us to these answers…

Of course, with technical innovation, an expanding library of data sources and shifting market dynamics, the answers to these questions are likely to change through time, and not necessarily in a linear fashion. Even if our hypothesis of finite alpha capacity at any point in time is true, that finite capacity is also elastic, as well. So, at somewhere in the vicinity of $185 billion in AUM, AQR is one of a short list of hedge fund leaders – along with the likes of Bridgewater Associates – who is applying quantitative methods in a way that is redefining the outer boundary of AUM capacity for the predominantly relative-value strategies found in the hedge fund section of our asset management ecosystem map; a segment that is known in Alphacution’s vernacular as the active management zone. Or, at least that was the mythology that I had allowed myself to believe…

There is really no doubt that AQR is among those who are defining and redefining the capacity boundaries of quantitatively-driven strategies. The better question is whether it can expand capacity boundaries while delivering superior risk-adjusted returns (that continue to justify the common hedge fund fee structure). Otherwise, what we actually have in AQR is an alpha player who is expanding into the beta realm where the ETF behemoths reside…

Now, taking the first exhibit (and its tracking of historical 13F position counts) back into consideration, Alphacution presents the next puzzle piece: In the exhibit, below, we show the same 72 quarters beginning Q4 2001 and ending Q3 2019, except this time with 13F gross notional long market value (GNLMV) by product class.

Here, more supporting evidence and more color is added to the case, which are primarily these: Yes, stock exposures are dominant from this vantage, but even if liquidity thresholds eventually limit the universe of applicable stocks to their suite of factor-based absolute return strategies, AQR appears to continue to pile more capital into that limited list of names long after it should have been clear that the applicable universe was, in fact, limited. And, secondly, one might conclude that AQR overshot its capacity limits – with aggregate 13F portfolio AUM peaking at ~$100 billion by Q3 2018 – which tended to stifle performance to an extent that eventually led to a cascade of cost reduction responses, namely two annual cycles of headcount trimming in early 2019 and early 2020; and, in fact, may have been a catalyst for listing a prized Miami penthouse for sale by the co-founder.

Now, none of this critique will stick if AQR has decided to move the goal posts. Meaning: If AQR has decided to leverage its “factor investing” brand and jump on the “smart beta” train in order to become an “asset maximizer / management fee maximizer” – like many of the large, traditional asset management platforms have become – and thus, gradually abandon the “asset / performance optimizer” category occupied by most hedge funds, then its perfectly fine to engage in strategy evolution.  The problems lie in saying you are one thing, and then going and doing a different thing…

For instance, in the exhibit, below, Alphacution presents a comparison of average 13F stock position by value for AQR and one of our recent case study subjects, Two Sigma, for the overlapping 68-quarter period beginning Q4 2002. The point here is to showcase key differences among a subset of core active strategies deployed by mature managers. Namely, the difference in implied native holding period (or, implied native turnover frequency) expressed by the difference in average stock position values. In other words, different alpha factors need to be extracted at different turnover frequencies, and therefore, managers need to (attempt to) align trading frequencies with how quickly those factors are expected to decay in the market.

Our assumption is that AQR is attempting to harvest alpha factors that decay more slowly, in this case, than those of Two Sigma, and therefore, it can afford to run larger average stock positions without incurring excess slippage costs. (Recall: Theoretical strategy capacity is tethered to portfolio turnover frequency, where lower turnover yields higher capacity and vice versa.) Alternatively, Two Sigma limits average stock position size (relative to AQR) because its turnover frequency is faster, and larger positions would risk excess slippage costs given expected liquidity in their available universe.

What we don’t know, from the chart above, is whether AQR has – purposely or inadvertently – scaled average stock position values too far given the liquidity of the securities universe relevant to its core strategy’s performance expectations. If on purpose, AQR would have had to calculate that the incremental management fees on the additional assets would have exceeded the incremental performance fees from not testing strategy capacity constraints any further with those additional assets. Of course, the actual calculus tends to be far less academic in reality…

Now, we can make the previous comparison more vivid by expanding it to a broader spectrum of notable players: In the exhibit, below, Alphacution presents a comparative ranking of average stock position values over the maximum available period (beginning Q4 1998) for a list of legendary hedge fund managers and proprietary trading firms. Note: The legend (on the right) corresponds to the ranking as of the most recent disclosure date.

Forgiving that this is a messy chart for the moment, what we are trying to illustrate here is a relationship between various alpha factors, implied native turnover frequency (necessary to successfully harvest those factors), and average position value; a discovery we first proposed in the Feed post “Ranking Strategy Speed for Top Quants, Market Makers.” And, since an explicit roadmap for the actual alpha factors and their actual rates of decay for each of these strategies does not exist, Alphacution is interpreting the average position value rankings for the strategies above, all of which are believed to be operating at or near capacity, as being indicative of their relative rates of turnover (which, in turn, is indicative of their targeted alpha factors, and also which – by the way – break down into two main categories: structural and asymmetric).

With AQR at the top of this average position value list, their data is telegraphing that they believe that their alpha factors are decaying more slowly than those of these peers, which further implies that they believe their strategy has more capacity than the others. However, one critical risk of that posture is that those “peers” may be indirectly consuming some or all of AQR’s expected alpha because they are turning over their positions (and portfolios) faster in the exact same universe of names…

Now, even if this is not the case – that AQR is still pursuing statistically unique sources of alpha in the same universe of securities – their optimal performance may still remain “under-harvested” as a result of their portfolio construction parameters. In the exhibit, below, Alphacution presents AQR’s largest position (in orange) and ten largest positions (in blue) as a percentage of 13F portfolio values for the 72-quarter period beginning Q4 2001 and ending Q3 2019. Here, it is easy to see that, with the exception of the formative period Q4 2001 – Q4 2004 (when the core strategy portfolio was still relatively small – and the ecosystem far less hostile to industrialized quantitative methods) and the most recent period Q3 2017 – Q3 2019 (when it appears AQR is tweaking certain strategy parameters), the intervening position concentrations were held within rigorous thresholds. In the case of the top ten positions, it seems clear that this threshold was 10% of the (long) portfolio value.

There is also evidence of strategy experimentations and enhancements along the way, which are almost always the result of weakened confidence in performance expectations. Both temporary and persistent use of additional product classes is one category of experimentation or strategy enhancements that Alphacution always looks for in the available data. For example, in the exhibit, below, Alphacution presents average ETF position by value over the 72-quarter period beginning Q4 2001 and ending Q3 2019.

Here, the lack of ETF positions speaks as loudly as those periods with ETF positions. The historically normal operating state of AQR’s strategy does not appear to require ETF’s. However, in one post-GFC period (Q1 2009 – Q4 2009), ETFs appear to be used as a portfolio hedge. More recently, a persistent use of ETFs (never exceeding 10 long positions, and including a subset of ETF options?!) emerges. The role of this latest ETF usage remains to be seen, although we can’t help but notice that their timing corresponds with the imposition of the implied minimum liquidity threshold for stocks within the 13F securities universe.

Taking the previous position concentration and ETF usage findings into consideration, in the exhibit, below, Alphacution presents four (13F) portfolios at 5-year intervals with all positions ranked by value (in log scale) for the quarterly periods Q3 2004 (“AQR 4930”), Q3 2009 (“AQR 9930”), Q3 2014 (“AQR 14930”), and Q3 2019 (“AQR 19930”). The objective in this visualization is to allow the shapes to tell the story, given the previously outlined descriptions of observable strategy shifts.

A brief “re-summarization” of an interpretation of this chart might go like this (albeit, admittedly, with some benefit of hindsight, as even we are learning how to interpret these shapes): As of Q3 2004, a relatively small and concentrated portfolio gave rise to strong performance that attracted capital and allowed AQR to expand its trading universe; causing a positive feedback-loop of performance and additional capital. With the exception of limited (ETF) hedging overlays in the post-GFC period, this strategy continued to grow in size and scope beyond Q3 2014 when the applicable universe in US stocks appears to reach its capacity. Though solid performance and additional capital accumulation continues through 2016, performance appears to become more challenged thereafter, and therefore, the applicable universe began to be trimmed and the portfolio was allowed to become slightly more concentrated in the top positions in an attempt to make up for recent performance challenges.

This last point on position concentration – the “tilt” in the shape of the strategy – is where we will bring this one to a close for now. It is clear that AQR’s recent performance challenges for 2018 and 2019 is a result of a mismatch between strategy and market – as certain factors, like momentum and value, have not been responding as they normally would. The reason(s) for that mismatch is what remains open for debate, although Alphacution believes that the pervasiveness of automated mean-reversion strategies and the relative weakness of discretionary / value-oriented managers are underappreciated phenomena that bear a material portion of the explanation for the market’s “unexplainable” predicament.

In the interim, we can’t help but wonder if more position concentration might be an effective combative, as is exemplified by Fidelity…

And, as for shifts in mythology: I now see AQR with its roots in the alpha realm and its aspirations in the beta realm, and I can’t decide at this hour which side of that line is more ominous.

Until next time…

By | 2020-02-10T11:27:23+00:00 February 9th, 2020|Alphacution Feed|

About the Author:

Paul Rowady is the Director of Research for Alphacution Research Conservatory, the first digitally-oriented research and strategic advisory platform uniquely focused on modeling and benchmarking the impacts of technology on global financial markets and the businesses of trading, asset management and banking. He is a 30-year veteran of the proprietary, quantitative and derivatives trading arenas with specific expertise in strategy research, risk management, and techno-operational development. Contact: feedback@alphacution.com; Follow: @alphacution.