Advertisements
algos hft

Goldman Asks: ‘Will The Machines Amplify the Next Downturn?’

Expect this to get plenty of attention. "HFTs know the price of everything and the value of nothing."

Here’s the thing about folks who harbor delusions of grandeur: they have a demonstrable (and highly amusing) tendency to couch everything in terms of things they said at one time or another. As the President would put it, “I’m consulting with myself, because I’ve said a lot things.”

I speak from extensive personal experience when it comes to nurturing an overriding sense of self-importance. In fact, I’d consider myself something of an authority on the matter, and I can tell you definitively that if left unchecked, it eats away at one’s ability to objectively process incoming information. Eventually, it renders people incapable of conceiving of a reality that’s independent of themselves.

You see this all the time in market analysis. It’s a form of confirmation bias, but it’s amplified by an almost pathological need to convince everyone else to join in on the fantasy that you knew it before everyone else. “I told you portfolio insurance was a bad idea, remember?” “I told you housing was a bubble, remember?” “I told you Canadian asset-backed commercial paper was a liquidity mismatched nightmare, remember?” “I told you China’s shadow banking complex was bound to catalyze a global meltdown, remember?” “I told you the rebalance risk in those inverse VIX ETPs was going be realized one day, remember?” And on, and on.

And look, sometimes individual people make prescient calls that scarcely anyone else was willing to make and deserve to get credit for it. But guess what? They usually do – get credit for it, that is. As it turns out, stories about individual people who called the next “big one” sell books and those books make for great feature films. It’s the natural way of things.

But more common than the story of Michael Burry and the housing meltdown is a story that has played out time and again, cycle after cycle, over and over, throughout history. Something comes along that reshapes the structure of markets in a fundamental way  and although it’s initially celebrated as an “innovation”, over time a growing chorus of people suggest there are potential pitfalls or that what was originally heralded as a miracle of market modernity is in fact downright dangerous. Obviously, CDOs fall into that category and were at the heart of the Burry legend, but I would argue that unlike some of the debates that are going on currently (e.g., ETFs, passive investing more generally, and HFT) CDOs were more obscure at least when it comes to popular discourse.

There are several of those stories playing out right now, in real time. There’s a vociferous debate around the epochal active-to-passive shift, for instance. A related storyline revolves around what that shift means for price discovery and still another offshoot of the same debate focuses on the potential for the liquidity mismatch inherent in some fixed income ETFs to eventually manifest itself in a fire sale for relatively illiquid high yield or emerging market debt. The thesis that ETFs and passive investing could be dangerous isn’t the “property” of any one skeptic or cynic – that thesis has been adopted and adapted and otherwise discussed by everyone from Howard Marks to Carl Icahn to every sellside shop on the Street to financial journalists.

The HFT debate and the related discussion of liquidity provision in modern markets is another example. It’s been the subject of vociferous debate for years and as you’re undoubtedly aware, Michael Lewis wrote a bestseller documenting the backstory. Apparently, that’s now going to be a Netflix original movie (really, it’s weird that it took this long for Flash Boys to be adapted into a feature length film).

HFT’s detractors (and those who bemoan the presence of non-carbon-based lifeforms in markets more generally) argue that the proliferation of “the machines” (to employ the overused Terminator analogy) invariably makes markets more fragile. This fragility shows up in times of stress and manifests itself in flash crashes and liquidity vacuums.

While this is often framed in needlessly opaque terms and while the term “algorithmic” implies a certain degree of complexity, conceptually it is not difficult to understand. I’ll just grab a random quote from the most recent Marko Kolanovic note:

We have noted in the past that a combination of computerized sellers, and computerized market makers poses a threat to equity price stability. As volatility increases, systematic investors have to sell, and at the same time market depth as provided by electronic market makers quickly disappears. For instance, S&P 500 futures market depth dropped over 90% during the February selloff. What is the reason for such a dramatic drop in liquidity? The most important driver is likely the increase of volatility (e.g. VIX), given that many market making algos (as well as business models) were calibrated during the years of low volatility. As these programs don’t have an obligation to make markets and are optimized for profits, they likely adjust quotes and reduce size in order to maximize their own Sharpe ratio.

In short, the events that transpired on Monday, February 5, laid bare the risks of modern market structure. The rebalance risk inherent in the structure of levered and inverse VIX ETPs was realized and amid the turmoil, liquidity seemed to be sparse. The concurrent VIX spike (the largest in history), triggered a cascade of de-risking from the same systematic strats that some folks have been warning about for years. By the time that was over, some $200 billion in equity exposure was forcibly unwound.

And that was of course just the latest example. There have been probably a dozen other acute episodes since 2008 and God only knows how many fleeting instances of “anomalous” price action. All of those events lend credence to the notion that something isn’t quite right, and that the proximate cause of these anomalies (which, by virtue – or maybe, by Virtu is better – of their increasing frequency aren’t really “anomalies” anymore) is likely to be the presence of machines.

The more of these episodes that transpire, the more concerned people get and naturally, that draws more attention to the “problem” (and I put “problem” in scare quotes out of respect for our quant friends who don’t enjoy being ceaselessly maligned).

Ok, so coming back to the points made here at the outset, there are a handful of folks out there (and we don’t mean analysts) who are predisposed to explicitly suggesting that the HFT debate is somehow “new” to everyone but them, which is what one does when one comes to believe the world revolves around oneself.

Outwardly, they’ll suggest it’s a good thing that the HFT issue is being openly discussed, but behind the feigned sense of fulfillment is a certain bitterness that the discussion is being carried on “without” them by reputable institutions and people who are taken seriously.

That’s exceedingly unfortunate because when people like, say for instance, Goldman, come along with a fantastic new note entitled “Will The Machines Amplify The Next Downturn?”…

GS1

… constructive discussion of the points raised in that note is subjugated to an effort on the part of some HFT critics to see how many different ways they can say “I told you so”.

So how about this: let’s not do that here. And do you know what else let’s not do? Let’s not take Goldman’s note out of context where that means framing every single passage in the most dire terms possible as though we pulled up Thesaurus.com, typed “apocalypse” into the search bar and then paired every single synonym with an excerpt from Goldman’s note in an absurd effort to chase after a sense of vindication that’s been eluding us for a decade.

To be clear, Goldman does indeed paint a rather ominous picture. Basically, the piece is an expanded version of a shorter note out back in March (more on that here) which asked if “markets themselves” are the biggest risk to, well, to markets.

The bank’s Charles Himmelberg begins his Tuesday note by observing what multiple other folks have mentioned in the course of searching for coal mine canaries: namely that while leverage isn’t as big of an issue as it was pre-crisis, the hunt for yield has likely forced investors into a situation where they aren’t getting compensated enough for taking liquidity risk. That’s another discussion entirely, but suffice to say Himmelberg “sees far less scope for systemic risks arising from the post-crisis search for yield” (compared to pre-Lehman leverage in structured credit).

With that out of the way, he launches into the discussion at hand, starting with this:

The past 10 years have seen fairly dramatic changes in the regulatory environment, industry composition, and the trading technologies by which liquidity is supplied to markets. The resulting market evolution is one in which algorithms are replacing people, and speed is replacing capital.  Exhibits 1 and 2 show how the volume share of algorithmic traders has grown by asset class and across products in futures markets.

GSHFT

Next, Goldman asks the same question a lot of other folks have asked. Namely this (paraphrased): How resilient will this setup ultimately prove to be when stress tested?

Well, recall what we noted above about there having been “God only knows how many fleeting instances of ‘anomalous’ events that lend credence to the notion that something isn’t quite right”. Perhaps the most worrisome thing about some of those events is that they unfolded in ostensibly deep markets.

“The fact that even some of the biggest, most heavily traded markets appear vulnerable to flash crashes provides plenty of ex-ante reason to worry that these small cracks in the foundation may betray deeper structural issues that have simply not yet been exposed,” Himmelberg writes, referencing the following set of charts:

GSHFT3

He then goes on to reference February 5 again, as the latest example of something being “not quite right with the current state of trading liquidity.”

Himmelberg eschews much of the well-worn territory when it comes to HFT critiques in favor of a more nuanced discussion, which he summarizes in the executive summary by quipping that “HFTs know the price of everything and the value of nothing.”

Implicit there is the notion that in a scenario where headline-scanning machines are unable (by virtue of being machines) to comprehend the nuances of, say, a policy announcement, they’ll find themselves at an informational disadvantage or, more simply, liquidity takers will have an edge and in order to avoid getting fucked, HFTs will pull back. Here’s Goldman:

The idea is that, because trading algorithms are more limited than human traders in their ability to process “complex” information flow (e.g. unconventional monetary announcements), they are at greater risk of being adversely selected when such information is known to be in the market and thus may withdraw liquidity. This raises the prospect that complex negative news events could cause outsized price declines due to the withdrawal of liquidity supply.

Well intuitively, that has the potential to become self-fulfilling. That is, they’ll all start chasing each other down the rabbit hole without realizing what’s happened to them (and yes, that’s funny). Here’s Goldman again:

Adverse selection is the enemy of liquidity supply. When shocks of unknown origin cause sudden price declines, HFTs may have reason to assume that the shock is being driven by fundamental news (e.g., if the price decline follows a complex macro surprise or dramatic policy announcement). Under these circumstances, HFTs are at higher risk of being adversely selected by more fundamentally informed traders, so their optimal response is to withdraw liquidity by widening their quotes or by withdrawing them altogether. As selling continues, a feedback loop can arise where the resulting lack of liquidity causes bigger price declines, which then causes HFTs to supply even less liquidity, in some case even switching to strategies that aggressively demand liquidity rather than supplying it. A particularly clear discussion of this logic can be found in a recent academic paper by coauthors Vikas Raman, Michel A. Robe, and Pradeep K. Yadav.

Unfortunately (and hilariously), this kind of chain reaction could be kicked off by a simple “fat finger” fuck up.

So is there any evidence of this? Well, again unfortunately, the answer is “yes.” To wit:

A recent report on the behavior of HFTs in the Eurex Bund Futures market around high-impact macroeconomic news announcements suggests precisely this—that HFTs systematically withdraw liquidity when “complex” (non-routine) information is known to be in the market.  Exhibit 4 is borrowed from this report and shows how HFTs behave during the 60-minute interval around major news events. The left chart shows that the overall depth in the market declines sharply around the announcement time (around 70% on average). The right chart displays the participation rate by HFTs, which declines from around 50% 10 minutes before the announcement to 33% at the time of release. This is also a drop of around 70%, suggesting that the drop in market depth is due mainly to the retreat of HFTs.

GSHFT4

The flip side to an informational disadvantage during certain, specific situations is that HFTs have an obvious (and by definition insurmountable) advantage during “normal” conditions. That, Goldman suggests, has crowded out humanity (literally) raising questions about “who” will be there to provide liquidity when it’s needed most.

When you throw in the fact that, as Kolanovic has pointed out on any number of occasions including in the note cited above, “many market making algos as well as business models were calibrated during the years of low volatility,” you’re left with many more questions than answers.

You get the idea.

So is this all doom and gloom? Mostly. But not entirely. And to present excerpts from this note without also mentioning Goldman’s efforts to play devil’s advocate to their own thesis is to mistreat the material and misrepresent their analysis, something we’re not in the business of doing, so here is what the bank says on the positive side:

To be clear, we have only attempted to flag a tail risk. We think it is hard to deny that financial innovation and market evolution have improved market liquidity. By many metrics the liquidity improvements are dramatic, especially for small trade size. But a fixation on these benefits should not prevent careful scrutiny of the potential costs, such as trading fragility.

Right. At the same time, there’s no denying that the overall thrust of Goldman’s piece is foreboding. Indeed, in the conclusion Himmelberg cites a series of other reasons to be concerned in addition to scenarios where HFTs withdraw liquidity as a result of an informational disadvantage. For instance, Goldman says the following (the third point is particularly interesting):

Trading volumes may be backed by too-thin capital cushions and thus may be excessively exposed to operational risk. A second theory argues that fragility can result from increased market fragmentation due to a greater risk of breakdowns in cross-market arbitrage. Yet a third theory points out that the growth of algorithmic trading may be undermining the role of legal liability in constraining destabilizing trading behavior.

They conclude by stating the obvious which is that because this industry is evolving rapidly (and that’s only fitting – it is all about speed), it’s impossible to keep abreast of all the possible risk factors.

But to Goldman, the rest of the sellside and to any readers who might be concerned about this, we would say the following.

There’s no need to worry about trying to keep a running tally of everything that can go wrong. The doomsday crowd is all over that and especially as it relates to HFT. In fact, we’re just lucky the tinfoil hatters let the rest of us get a word in edgewise – if only so they can point to it and say “I told you so.”


Advertisements

4 comments on “Goldman Asks: ‘Will The Machines Amplify the Next Downturn?’

  1. Really, really good essay. I remember watching spellbound, the first famous flash-crash and equally dramatic rebound. I was not invested and thus just amused. It will be interesting to watch as AI advances perhaps with more ‘brains’ what with unlocking (?) the animal fear and greed. It seems odd to me that it’s even allowed. Would it not be more fair that transmission of an order was across the board equal? And that orders can’t reverse in nanosec but calibrated in humanly understood increments?

  2. Pingback: Need to Know: Goldman warns machines will ugly up the next selloff — but here’s your silver lining | Market Tamer

  3. Anonymous

    Thanks, H! A complex issue, and much appreciative of your time to frame the narrative so us Joe E*Trades can (kind of, sort of) see how it all fits together. Reading you every day is worth the time and definitely helps in upping my trading acumen.

  4. Pingback: Have You Heard The One About USA Today, Lifeguards And FANG Stocks? | Growth Investing Research

Speak On It

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Skip to toolbar