Investors Beware: #Robo-Blindness Ahead

I’m on the fence when it comes to the “robo-advisory” craze. Clearly, given the gush of venture money and subsequent marketing buzz behind automated trading and investing methods for the masses, lots of very smart folks think that “robots” managing your money (on an highly automated or assisted basis) are here to stay. Note that in 2016 Betterment and Personal Capital bagged rounds of $100 million and $75 million, respectively.

Here’s the first shoe: Quantitative methods for retail investors were always inevitable once a sufficient level of maturity in the underlying mechanisms had been achieved. In an earlier generation of quantitative trading strategy development (ie – late 1990’s), we used say, “You can either fish or sell bait.” Meaning, simply, that you can either use your trading signals on a proprietary basis or sell your modeling output for a fee. For those who were good at it, the profitability in the beginning was – or at least could be – far too juicy to sell for a fixed fee. And, frankly, these methods were in a far too formative stage back then to be applicable to mom and pop investors, for legal and/or moral reasons.

Now, here we are. Welcome to the bandwagon. Exhibit 1 from investorjunkie.com (Dec 2016) lists 21 robo-advisory offerings, both stand-alone and as a component of broader wealth management services. And though this list is fairly comprehensive, it is still not complete. A Goldman Sachs-backed team, Kensho Technologies is the most notable that comes to mind (even though they don’t specifically promote themselves as robo-advisors).

Why now? A convergence of interconnected drivers have spawned an explosion of automated wealth advisory services (and other financial solutions based on “artificial intelligence” or “machine learning”): 1) Born in the late 1980’s by the most adventuresome quantitative pioneers, automated strategies today are (by my measure) in their 4th generation and therefore are perceived to be sufficiently stable and mature to be reconfigured for mass adoption; 2) Successfully deploying highly automated trading strategies on a proprietary basis  – aka HFT, et al – has become increasingly competitive and expensive relative to its current theoretical capacity (IOW – expected scale of opportunity and level of profitability); 3) An increasing quant and data science talent pool needs new outlets (beyond the limited and exclusive proprietary trading arena), 4) Net performance of more traditional, full-service advisory offerings relative to their fees have motivated investors to pursue lower-fee DIY (do-it-yourself) solutions, much like the boom in ETFs or the attractiveness of discount brokers of years past; and, finally 5) There is exponentially more clean and streaming data than ever before – a gross understatement.

With these combined drivers set to be persistent, increasing automation of workflows will persist as well. This is a common theme in FinTech circles as well as the overarching digital revolution. As a result, robo-advisory services should be expected to levitate along with the broader rising tide of process automation for more sophisticated corners of financial services.

Now, here’s the other shoe:  In the post-GFC market regime, central bank interventionism – led by the Fed – has poisoned market data, the amniotic fluid of nascent robo-advisory services (and other quantitative investment methods). So, the logic is that if the Fed changes its behavior – such as, becoming less interventionist or less manipulative of markets – then automated investment methods, like robo-advisors, would experience a sort of “blindness” because their models are not tuned to data from any period outside the post-GFC period when central banks have been so active. The impact of this blindness would show up in the results fairly quickly, and the list of players (above) would consolidate to those with the most adaptive models.

Truth be told, I’ve been saying this for the past  5 or  6 years – so take this cautionary tale with a grain of salt. If the Fed never changes it’s behavior, this point could be moot. But, just sayin’…

And, here’s some of the (excerpted) details and back story from September 2012 (Tabbforum):

“Central bank intervention is poisoning the data. From direct impacts on just about all market-related datasets to indirect impacts on economic and other fundamental datasets, the Age of Intervention will serve to confuse and disorient a broad spectrum of automated trading methods today and for years to come.

Now, this topic is ripe for a good old-fashioned expletive-laced rant which I will try to water down for the younger members of our audience. For starters, and as a prime example of the heights of duplicity we have here, an unprecedented transformation fueled by regulatory onslaught is currently underway in the global OTC derivative markets precisely for the purpose of fostering greater transparency. One pillar of these transparency goals includes pre-trade price discovery for certain markets and trade structures, like interest rate swaps and credit derivatives, that have been privately-negotiated (or bilateral) since they were first conceived in the early 1980’s and which as of the end of 2011 represent over $700 trillion of notional outstanding exposures (before trade compression activities). As the logic goes, new multi-lateral market mechanisms – modeled largely after those found in equity and listed derivatives – will yield better, truer prices for swaps.

Funny, this. Since when is there true price discovery in US Treasury markets, equity markets, real estate markets – just about any market anymore – when the Ultimate Omniscient Insiders are using infinite money-creation capabilities and unrestrained access to private information to influence asset prices all over the world? At any moment in time, one could argue that there simply cannot be true price discovery in any market where intervention occurs – which is most of them.

Moreover, with hints of such activity now becoming downright conspicuous, market demographics are being tragically altered. Though obfuscated under the guises of Euro-instability, impending fiscal cliffs, and all other manner of global economic abnormalities, ongoing central bank intervention is the actual underlying force that spooks traditional segments to the sidelines (i.e. – cash or the “mattress fund”) or into non-traditional investments, like precious metals, or crowds them into perceived safe havens, like bond funds. Increasingly, the demographics that remain include only those that are stuck there by internal mandate, like an indexed fund, or by limitation of design, like an automated market-making or latency arbitrage program. As a result of this exodus, the machinations of central banks are becoming a proportionately greater force in capital markets.

This is where the quant angle and disorientation come back into the picture: Consider that algorithmic trading is little more than pattern recognition. Patterns are simply manifestations of consistent behaviors. The behaviors of market actors left behind in the data – like so many clues – represent the patterns that automated methods are designed to home in on. Furthermore, overlaying logic or intuition – if not, actual confirming evidence – about the nature of investment, trading or execution strategies improves the efficiency of algo development by minimizing the prevalence of false positives (or positive identification of patterns that don’t actually exist); a major drawback of pattern recognition studies.

We know tons of behaviors today. We know, in advance, how a VWAP, TWAP or Iceberg algo works – and so, faster algos can pick these off. We know that mutual funds and others perform window dressing at the end of each quarter. We know that certain economic releases and corporate earnings have a high probability to elevate volatility. We know that biasing an order book with excessive quotes can cause other, slower, and less-sophisticated algos to flinch with regularity. We know that order handling rules in one liquidity pool versus those in another can cause temporary price discrepancies. We know that resting stops left behind by unsuspecting day traders are like sitting ducks to be goaded into wrong-way, momentum-inducing reactions. We know that most futures markets on Sunday nights and during holiday lulls are deliciously easy to push around – the after effects of which can last for days or weeks or months. And on and on and on. Literally, hundreds of persistent patterns that can be harvested, given a lot of speed and a little intelligence – or sometimes with little more than a tiny sliver of a gargantuan balance sheet in a burst of shock and awe. Converting theoretical, yet (hopefully) persistent, patterns (or “theoretical alpha”) into harvested or actual alpha is the name of the game.

What we don’t know (yet) is the behavior patterns of central banks – other than the fact that they seem to refuse to allow markets to fall very far or very fast (since 2008 and with few exceptions), enter and exit markets asymmetrically, and have no logical or intuitive mandate other than the aforementioned hypotheses. We don’t know how forceful (or desperate) they will behave going forward, nor how coordinated their efforts globally, nor which markets they will seek to influence at any given time, nor when they will stop. However, one thing’s for sure, the record of their activities is preserved in the data; increasingly, their fingerprints are on more and more of the data.

Of course, it is an occupational hazard of quantitative researchers to figure out new patterns; a skill that is a requirement for longevity. Perhaps, there are persistent patterns to be detected that can be pinned confidently on interventionist activities, thereby forging an acceptable level of predictability and sustainable profits. And, also, the impacts vary by strategy. Though the highest turnover strategies are typically simpatico with volatility (some of which being caused by intervention), the indirect impacts of the current receding tide of liquidity is playing havoc with the speedier players; tales about which seem to indicate that some type of disorientation has plagued their profit-making abilities for months.

As for the slower or more traditional strategies, all I can say is good luck. We are in uncharted territory in terms of market demographics and global interconnectedness. While I have been an advocate for applying automated methods to lower turnover strategies – and still am – it is clear that the signals embedded in the data – both market and economic data – are potentially polluted by these alien influences. What’s worse, when we use data from this period in history to provide guidance for the future, these alien influences will still be there. They are inextricably intertwined in the record for all of eternity.

Going forward, keen awareness of changing demographics, methods of strategy deployment, and cross-market forces are among the skill requirements for ongoing success with automated methods. And, above all, never forget that the Fed uses algos too.”

By | 2017-01-12T09:43:26+00:00 January 10th, 2017|Alphacution Feed|

About the Author:

Paul Rowady is the Director of Research for Alphacution Research Conservatory, the first digitally-oriented research and strategic advisory business model focused on providing data, analytics and technical infrastructure intelligence within the financial services industry. He has 28 years of senior-level research, risk, technology, capital markets and proprietary trading experience. Contact: paul@alphacution.com; Follow: @alphacution.

Leave A Comment