Doug Kass: Not Even The Algo Creators Know What Is Going On

Authored by Seabreeze Partners Management's Doug Kass,

If you haven't read my opening missive, please do, as it's important.

This column extends the conversation about the problem with advanced market technologies and strategies a step further than most retail and institutional investors understand.

Most people think of artificial intelligence and algos as simply executing logical rules programmed into them by humans -- the same rules that the programming humans would follow if they were presented with the same data and data analysis. The algos and AIs are doing it in the same way humans have always done and would do, but at a much slower speed or perhaps not at all because of the very weak and distant relationship of some data items to other data items.

The general belief is that algos and AIs are just "faster humans able to do a lot more calculations in a meaningful time frame". That may NOT be a correct characterization of some of the more powerful AIs that may be working in the markets. Of course, we don't know what AIs are working because there are no regulations requiring that machine decision-making accounts disclose and register as such ... a very, very big gap in regulation.

True, AI and the related "machine learning" developments at the leading edge of such technology do NOT simply duplicate human rules and logic. Instead, while they may perform simple repetitive correlations initially on data as humans currently formulate that data, the more advanced machines go on to program themselves at successive layers, where the data being analyzed and correlated is no longer what we think of as data. Rather, it is often data artifacts created by the first layers in a form that no human would ever consider or has ever seen. To put in a more street-level way, the first level creates ghosts and apparitions and shadows that the second layer treats as real data on which it assesses correlation and predictability in the service of some decision asked of it. AND ... a third and fourth and on and on are doing the same thing with output from each layer below it.

The result of this procedure is striking and terrifying when the the leading experts in AI and machine learning are interviewed. They admit that they have no way of determining what rules AI and machine- learning powered machines are following in making their decisions AND we cannot even know what inputs are being used in making those decisions.

Think about that. The creators have no knowledge of what their creations are thinking or what kind of inputs the machines are thinking about and how decisions about that are being made. The machines are inscrutable and, most terrifyingly important, UNPREDICTABLE.

We are not telling these AIs how to make decisions. The machines are figuring out how to decide to "make a profit" on their own and subject to no enforceable constraint.

The resulting risk of "flash crashes" -- to lump all sudden and unexpected behaviors into a catchphrase -- is unknowable but probably much greater than anyone even dreams. The machines have no fear of flash crashes or any other kind of crash. Such crashes might even serve their purpose of "making a profit."

Be forewarned as last Friday's Nasdaq schmeissing may be a walk in the park compared to what may happen in the future..

The risks are much greater than most imagine.

People need to understand this threat much better than they do.

It needs to be better factored into the investment process. in the column entitled "risks."

Listen, Luddites, It's A Thing About the Machines

"Let's consider that statement. In the last 20 years the VIX closed lower than 10 on a total of 11 days, and 7 of those days were in the past month. Think about that - over the past 2 decades, was the last month the most benign macro environment? (e.g. last week: Comey testimony, UK elections, ECB, geopolitical uncertainty, Qatar, FANG flash crash, etc.)."

-- Marko Kalonovic, JPMorgan's head quant

I vividly remember episode 40 of Twilight Zone entitled "A Thing About Machines." (Here is a clip of the episode)

Originally aired on Oct. 28, 1960, it's the story of a repairman who pays a house call to Bartlett Finchley, who is having trouble with his television's reception. Finchley is an ill-tempered and lonely gourmet magazine critic. He abuses machines in his home and he is as inept with human beings as he with the machines, which he concludes are conspiring against him.  

Though Finchley is seen as paranoid by many, eventually every machine in the house (including his car) turns against him: 

  • His typewriter types out the message, "Get out of here Finchley."
  • A woman on the television speaks the same message.
  • His electric razor rises menacingly in the air and lunges at him.
  • An unplugged telephone has a voice that speaks the same words as the typewriter. 

Finchley drinks a bottle of liquor and passes out. When he awakes, the machines in the house tell him to get out and the razor pursues him. He runs out of the house and is chased by his driverless car, which winds up pushing him into his pool. 

Sinking to the bottom, Finchley drowns. When the police pull him out, they can't explain how he could sink to the bottom when not being weighed down, as normally a body would float, nor could they explain the car near the pool. 

Listen Luddites, for the stock market, too, it's a thing about the machines. 

Throw away your fundamental analysis, your price charts, interest rates and economic growth forecasts, as the market has lost its moorings. 

It is no longer a pyramid of fundamental and technical analysis nor is it a response to changing investor sentiment. 

The ongoing multiyear changes in the market structure and dominant investor strategies in which quants, algos and other passive strategies (e.g., ETFs) have replaced active managers raise the same risks that Finchley faced 57 years ago. 

And the overwhelming impact of central bankers' largesse is the cherry on the market's non-fundamentally influenced sundae. 

As I have written: 

"The combination of central bankers' unprecedented largesse (and liquidity) when combined with mindless quant strategies and the enormous popularity of ETFs will, as night follows day, become a toxic cocktail for the equity markets. While we live in an imperfect world, we face (with valuations at a 95% decile on a number of metrics) a stock market that views the world almost perfectly." 

Back to JPMorgan's Marko Kalonovic, who is quoted at the top of this piece and again here: 

"... some striking facts: to understand this market transformation, note that Passive and Quantitative investors now account for ~60% of equity assets (vs. less than 30% a decade ago). We estimate that only ~10% of trading volumes originates from fundamental discretionary traders. This means that while fundamental narratives explaining the price action abound, the majority of equity investors today don't buy or sell stocks based on stock-specific fundamentals.


The next and perhaps just-as-important driver is, of course, central banks, based on Kalonvic's analysis: "With ~$2T asset inflows per year, central bank liquidity creates strong interest rate and policy sensitivity for sectors and styles. Low rates also invite investors to sell volatility."

Everyone should read this important note from JPMorgan's head quant (hat tip to Zero Hedge) in order to understand how risk parity, volatility trending, stat arb and other quant strategies that are agnostic to balance sheets, income statements and private market value artificially are impacting the capital markets and, temporarily at least, are checking volatility. 

Last Friday's market schmeissing was the first shot across the bow. There will be many more of those Fridays. 

This is not my Grandma Koufax's stock market.


Overleveraged_… Wed, 06/14/2017 - 17:16 Permalink

I don't need an "Algo" to make money. All I need to do is what I've been doing since late last year: Buy loads of 3x S&P 500 Leveraged Long. It's a perfect strategy that does nothing but win.I still see people around thinking that there is a hint of reality to these markets or that "The crash must be coming soon!" No, just.. No.. Shhhh. The whole "crash is coming" theory is just to get people to go short, so the big money players can skim them by pumping the markets back up to all time highs.People wake up. Donald Trump is all in on the stock market. When the President of the USA associates himself with the stock market it is going to go up to new all time highs. He used to hate Yellen? NOW HE LOVES YELLEN. Trump while appearing to be a President of the people is still just a Wall Street/Military Industrial Complex president. And guess what? He's going to take us to new all time highs.I plan on quitting my job this year with all the gains I am making.

. . . _ _ _ . . . Wed, 06/14/2017 - 17:18 Permalink

If you owned the company which developed the original AI, the one that when paired with a quantum computer could take over all the world's internet, military, stock markets, electric grids... would you A) keep everyone around and make everything free for all, or B) kill everyone except a bunch of mechanics and a harem?AI will be the death of us all.

vegas Wed, 06/14/2017 - 17:19 Permalink

BTFD you fucking idiot ... just buy the fucking dip. I too watched the anime movie yesterday on ZH, and of course it works ... up until it doesn't. But oh well, I digress; just buy the fucking dip cuz central banks won't let this stuff drop. Some day, they might not have a choice, but for now ... just buy the fucking dip you fucking idiot.

TheSilentMajority Wed, 06/14/2017 - 17:21 Permalink

The problem for the algos(and every other non-cb investor) is that the SNB, BOJ, and ECB are now very aggresively printing unlimited digital cash and using it to buy USA equities and bonds, in an efffort to stabilize/depreciate their currencies.

This time is truly different.

They are the marginal buyers of everything, no matter if the fundamentals are good or bad.

The challenge is trying to figure out when/if those central banks will stop their print & buy programs???

Uranium Mountain Wed, 06/14/2017 - 17:19 Permalink

In photography, when you save a jpeg then copy and save it again then copy and save it again you get something called jpeg artifacts. Empty spaces in the photo and loss of more and more of the original. Same thing as copying a photocopy and then copying the next photo copy and copying that copy.. Each time there is loss. . It makes for a very shitty photo.

Abbie Normal Uranium Mountain Wed, 06/14/2017 - 18:08 Permalink

That's only true if you use lossy compression -- each generation gets progressively worse.  If you continue to save the raw format, then there is no loss even after many generations.What's happening with the algos and AI is like computers playing a game of chess.  They can think dozens of moves in advance and if the end result is to win, they are willing to sacrifice anything to do so.  Translated to the world economy, the AI would be willing to destroy the world to make itself the winner.  Shades of Skynet....

In reply to by Uranium Mountain

Cordeezy Wed, 06/14/2017 - 17:24 Permalink

I doubt the creators simply don't know what their algorithims are doing, they had to set parameters in the creation, they simply don't just go off on a tangent.  AI and machine learning isn't that far along yet......  

. . . _ _ _ . . . Cordeezy Wed, 06/14/2017 - 17:56 Permalink

 Do you remember the Bilderberg meeting in Dresden last year. Guess who was there.Bengio, Yoshua (CAN), Professor in Computer Science and Operations Research, University of MontrealLook him up. I think they might be onto something. "Yoshua Bengio is a Canadian computer scientist, most noted for his work on artificial neural networks and deep learning."Also in attendance were:Hassabis, Demis (GBR), Co-Founder and CEO, DeepMindSchmidt, Eric E. (USA), Executive Chairman, Alphabet Inc.Karp, Alex (USA), CEO, Palantir Technologies One of the key topics for discussion this year is technological innovation.

In reply to by Cordeezy

BadLibertarian Cordeezy Wed, 06/14/2017 - 18:45 Permalink

My understanding is that for advanced neural networks this is precisely the case. The learning that is going on isn't easily broken down into an algorithm that can be described or logically tested. Instead, inputs are connected to outputs in the neural net with different weighs assigned to the connections between nodes, more or less randomly to start.Then as the network is fed a series of inputs, you check to see if the output it is producing is useful or expected, and if it is, then you 'breed' it with another parent network which also produced good results. Over many generations of evolutionary development, what you end up with is an expert system which is good at answering the problems you give it to solve - most of the time, but which you have zero understanding of the inner workings of.

In reply to by Cordeezy

. . . _ _ _ . . . BadLibertarian Wed, 06/14/2017 - 19:37 Permalink

Well stated.Also, simulations can be run and tested and checked before implementation in many cases, so it isn't quite like releasing a dragon into the world either... though it could be if checks are overlooked for the benefit of speed. But my feeling is that nobody wants to lose it all in a nanosecond because the algo screwed up. There must be a check-sum in there somewhere.

In reply to by BadLibertarian

Apeon BadLibertarian Wed, 06/14/2017 - 22:51 Permalink

Makes me think of the muskets which sere better than spears an arrows------and then someone developed rifling, and the ones with plain muskets lost out, and then along came cartridges, and revolving barrels and auto feed, and........................................................................................................................

In reply to by BadLibertarian

Radical Marijuana Wed, 06/14/2017 - 18:14 Permalink

Fundamentals: money is measurement backed by murder. Advanced market technologies became automated frauds.Globalized Neolithic Civilization developed the abilities to back up legalized lies with legalized violence to astronomically amplified sizes. Meanwhile, being able to back up lies with violence never stops those lies from still being fundamentally false.But nevertheless, in each short to medium term increment, about exponentially advancing technological powers and capabilities are enabling sociopolitical systems based upon public governments enforcing frauds by private banks to become about exponentially more fraudulent.The longer term consequences of globalized electronic monkey money frauds, backed by the threat of force from apes with atomic weapons, are going further and further beyond the imagination of ordinary flesh and blood human beings. Electronic frauds, backed by the force of weapons of mass destruction, have allowed the development of an overall magnitude of collective fraudulence which has far exceeded the flesh and blood individuals' previously presumed frames of references.The old-fashioned systems of metal money backed by swords, which segued into paper money backed by gunpowder, were still based on chemical energy, which could be roughly compared to and comprehended by individual human beings who were also living through harnessing chemical energy. However, the development of electronic money, backed by atomic weapons, has left individual human beings so far behind that it is no longer possible for the vast majority of those human beings to understand the degree to which that has become so ...IF, IF, IF, the "fundamentals" were to become more intelligible, then those would have to made more consistent with quantum mechanics and the special theory of relativity, etc. ... However, what is actually happening is as described in the article above, except that it follows the general pattern of the rest of the content published on Zero Hedge by grossly understating the situation.Theoretically, it is becoming more and more imperative for series of intellectual scientific revolutions and profound paradigm shifts to happen in political science, and especially in political economy. Theoretically, doing so is possible. However, doing so continues to be politically impossible, due to the intensifying paradoxes that were driven by the ways that natural selection pressures drove the development of artificial selection systems to become as dishonest as possible, while about exponentially advancing technologies are endeavouring to enable exponentially advancing dishonesties.Globalized Neolithic Civilization was always based on being able to back up lies with violence, while advancing physical science enabled those sociopolitical systems to back up bigger lies with more violence. The results are understated in the article above, because it continues to presume upon fundamentally false views of what the "fundamentals" were.The vast majority of human beings continue to rely upon various old-fashioned religions and ideologies with respect to attempting to understand themselves and their Civilization. However, technologies such as AI exist due to series of profound paradigm shifts already achieved through mathematical physics. Those are tending towards creative syntheses between post-modernizing science with ancient mysticism.The ways that most people continue to think about everything are based on taking for granted the dualities of false fundamental dichotomies and the related impossible ideals. However, the progress achieved in physical sciences more and more became based on developing unitary mechanisms to better understand general energy systems.Theoretically, it is possible to better understand human beings and Globalized Neolithic Civilization as manifestations of general energy systems. However, doing so requires systematic paradigm shifts in the ways that people think about time and space, and therefore, think about entropy. It is barely possible to exaggerate the degree to which "we" are living inside Wonderland Matrix Bizarro Worlds, where everything is publicly presented in the most absurdly backward ways possible.The so-called previously presumed "fundamentals" regarding political economy were due to the history of the biggest bullies' bullshit becoming the banksters' bullshit. Almost everyone continued to take for granted thinking about time and space in absurdly backward ways, and therefore, everything else which was achieved through the advancements of thermodynamics and information theory were similarly based on misunderstanding everything in the most absurdly backward ways.At the present time, and for the foreseeable future, it will continue to be the case that governments are the biggest forms of organized crime, dominated by the best organized gangsters, i.e., public governments will continue to enforce frauds by private banks, while those banks, and the big corporations that have grown up around those banks, will be able to continue to publicly present themselves as doing the opposite of what they are really doing.Enforced frauds are symbolic robberies, which are growing at about an exponential rate, due to about exponential progress in technology. The results are accumulating apparent anomalies, which are becoming more and painfully obvious to those who bother to examine those anomalies, such outlined in the article above. However, such articles tend to collapse back to presumptions that there was some previous validity in the old-fashioned notions regarding the "fundamentals," despite that those notions were fundamentally WRONG, and as absurdly backwards as humanly possible to make and maintain, due to thousands of years of being able to build Civilization based on enforcing frauds.

withglee Wed, 06/14/2017 - 17:35 Permalink

Of course, we don't know what AIs are working because there are no regulations requiring that machine decision-making accounts disclose and register as such ... a very, very big gap in regulation.Written as if universally corruptible, and thus universally corrupt, regulation ever works.

medium giraffe Wed, 06/14/2017 - 17:35 Permalink

"no way of determining what rules AI and machine- learning powered machines are following in making their decisions AND we cannot even know what inputs are being used in making those decisions."What, the best minds in AI don't know how a fucking log file works?  Bullshit.

logicalman medium giraffe Wed, 06/14/2017 - 17:42 Permalink

Once you give away control over your life to another entity, the outcome is unlikely to benefit you over the other entity.Pandora's box regarding physics was opened with Trinity.AI isn't quite on the top of the shot tower yet, but I think the results, when that happens, will be about as positive as the invention of nuclear weapons has been.Human minds are no different from the minds of the hunter gatherers that were the first humans.You can't destroy the planet with spears, bows anbd arrows or clubs, but now the 'spears' have thermonuclear tips.Why would any thinking human try to make something that, if it goes awry, will be the end of themselves?Homo should drop the 'sapiens' bit.

In reply to by medium giraffe

logicalman Wed, 06/14/2017 - 17:34 Permalink

I don't think the author understands the true idea of the 'Luddites'It's a term almost as ill-understood as 'anarchist'Interestingly, both concepts scare the shit out of those who wish to control others for their own benefit, hence the corruption of the meanings of the words.

meditate_vigorously Wed, 06/14/2017 - 17:35 Permalink

"They admit that they have no way of determining what rules AI and machine- learning powered machines are following in making their decisions AND we cannot even know what inputs are being used in making those decisions."

That is complete and utter bullshit. I stopped reading there. The author should be shot for his stupidity. I can give him the benefit of the doubt and say he oversimplified, but the ultimate analysis would not change.

withglee Wed, 06/14/2017 - 17:38 Permalink

Hey ... they replaced the arm on the slot machine with a button. Why not replace the gamble that is the stock market casino with a computer program. 

adr Wed, 06/14/2017 - 17:40 Permalink

Algos buy because some stock hit a percentage of some arbitrary line and sell because they rose by a percentage above some other arbitrary price. Nothing means anything. An algo can't understand real business conditions or a business cycle. But the market makers don't care either. Algos make them money, and that's what they are playing the game for. Our economy is 100% made of a computer's imagination. We can all be dead and algos will still be trading stocks to fibbonacci lines. 

Hikikomori Wed, 06/14/2017 - 18:21 Permalink

If I was designing an algo - and I'm not - I wouldn't want to program it for fear.  Irrational fear and irrational exhuberance are the downfall of human traders.

Stormtrooper Wed, 06/14/2017 - 18:31 Permalink

Someone should invent a stock market with human traders who trade by candlelight.  I might take part in that as an investor.  Otherwise, best to stick with PMs to weather the armageddon.

Is-Be Wed, 06/14/2017 - 18:54 Permalink

In her book "Thinking in Systems" Donnella Meadows (rip) shows us that in order for a control loop to be efficient (keep the output smooth) the negative feedback loop must be timely and powerful.The algos are timely (fast) and the FED's algos feed them digital money, so they are powerful.So algos are fast and powerful.Therefore the VIX, which is a measure of volatility, goes to Zero.I think that the whinge here is that the Ape can't compete against the machine. Welcome to Future Shock, Ape.