Locklin on science

Andreessen-Horowitz craps on “AI” startups from a great height

Posted in investments by Scott Locklin on February 21, 2020

Andreessen-Horowitz has always been the most levelheaded of the major current year VC firms. While other firms were levering up on “cleantech” and nonsensical biotech startups that violate physical law, they quietly continued to invest in sane companies (also hot garbage bugman products like soylent).  I assume they actually listen to people on the front lines, rather than what their VC pals are telling them. Maybe they’re just smarter than everyone else; definitely more independent minded. Their recent review on how “AI” differs from software company investments is absolutely brutal. I am pretty sure most people didn’t get the point, so I’ll quote it emphasizing the important bits.

https://a16z.com/2020/02/16/the-new-business-of-ai-and-how-its-different-from-traditional-software/

They use all the buzzwords (my personal bete-noir; the term “AI” when they mean “machine learning”), but they’ve finally publicly noticed certain things which are abundantly obvious to anyone who works in the field. For example, gross margins are low for deep learning startups that use “cloud” compute. Mostly because they use cloud compute.

 

Gross Margins, Part 1: Cloud infrastructure is a substantial – and sometimes hidden – cost for AI companies 🏭

In the old days of on-premise software, delivering a product meant stamping out and shipping physical media – the cost of running the software, whether on servers or desktops, was borne by the buyer. Today, with the dominance of SaaS, that cost has been pushed back to the vendor. Most software companies pay big AWS or Azure bills every month – the more demanding the software, the higher the bill.

AI, it turns out, is pretty demanding:

  • Training a single AI model can cost hundreds of thousands of dollars (or more) in compute resources. While it’s tempting to treat this as a one-time cost, retraining is increasingly recognized as an ongoing cost, since the data that feeds AI models tends to change over time (a phenomenon known as “data drift”).
  • Model inference (the process of generating predictions in production) is also more computationally complex than operating traditional software. Executing a long series of matrix multiplications just requires more math than, for example, reading from a database.
  • AI applications are more likely than traditional software to operate on rich media like images, audio, or video. These types of data consume higher than usual storage resources, are expensive to process, and often suffer from region of interest issues – an application may need to process a large file to find a small, relevant snippet.
  • We’ve had AI companies tell us that cloud operations can be more complex and costly than traditional approaches, particularly because there aren’t good tools to scale AI models globally. As a result, some AI companies have to routinely transfer trained models across cloud regions – racking up big ingress and egress costs – to improve reliability, latency, and compliance.

Taken together, these forces contribute to the 25% or more of revenue that AI companies often spend on cloud resources. In extreme cases, startups tackling particularly complex tasks have actually found manual data processing cheaper than executing a trained model.

This is something which is true of pretty much all machine learning with heavy compute and data problems. The pricing structure of “cloud” bullshit is designed to extract maximum blood from people with heavy data or compute requirements. Cloud companies would prefer to sell the time on a piece of hardware to 5 or 10 customers. If you’re lucky enough to have a startup that runs on a few million rows worth of data and a GBM or Random Forest, it’s probably not true at all, but precious few startups are so lucky. Those who use the latest DL woo on the huge data sets they require will have huge compute bills unless they buy their own hardware. For reasons that make no sense to me, most of them don’t buy hardware.

In many problem domains, exponentially more processing and data are needed to get incrementally more accuracy. This means – as we’ve noted before – that model complexity is growing at an incredible rate, and it’s unlikely processors will be able to keep up. Moore’s Law is not enough. (For example, the compute resources required to train state-of-the-art AI models has grown over 300,000x since 2012, while the transistor count of NVIDIA GPUs has grown only ~4x!) Distributed computing is a compelling solution to this problem, but it primarily addresses speed – not cost.

Beyond what they’re saying about the size of Deep Learning models which is doubtless true for interesting new results, admitting that the computational power of GPU chips hasn’t exactly been growing apace is something rarely heard (though more often lately). Everyone thinks Moore’s law will save us. NVIDIA actually does have obvious performance improvements that could be made, but the scale of things is such that the only way to grow significantly bigger models is by lining up more GPUs. Doing this in a “cloud” you’re renting from a profit making company is financial suicide.

 

Gross Margins, Part 2: Many AI applications rely on “humans in the loop” to function at a high level of accuracy 👷

Human-in-the-loop systems take two forms, both of which contribute to lower gross margins for many AI startups.

First: training most of today’s state-of-the-art AI models involves the manual cleaning and labeling of large datasets. This process is laborious, expensive, and among the biggest barriers to more widespread adoption of AI. Plus, as we discussed above, training doesn’t end once a model is deployed. To maintain accuracy, new training data needs to be continually captured, labeled, and fed back into the system. Although techniques like drift detection and active learning can reduce the burden, anecdotal data shows that many companies spend up to 10-15% of revenue on this process – usually not counting core engineering resources – and suggests ongoing development work exceeds typical bug fixes and feature additions.

Second: for many tasks, especially those requiring greater cognitive reasoning, humans are often plugged into AI systems in real time. Social media companies, for example, employ thousands of human reviewers to augment AI-based moderation systems. Many autonomous vehicle systems include remote human operators, and most AI-based medical devices interface with physicians as joint decision makers. More and more startups are adopting this approach as the capabilities of modern AI systems are becoming better understood. A number of AI companies that planned to sell pure software products are increasingly bringing a services capability in-house and booking the associated costs.

Everyone in the business knows about this. If you’re working with interesting models, even assuming the presence of infinite accurately labeled training data, the “human in the loop” problem doesn’t ever completely go away. A machine learning model is generally “man amplified.” If you need someone (or, more likely, several someone’s) making a half million bucks a year to keep your neural net producing reasonable results, you might reconsider your choices. If the thing makes human level decisions a few hundred times a year, it might be easier and cheaper for humans to make those decisions manually, using a better user interface. Better user interfaces are sorely underappreciated. Have a look at Labview, Delphi or Palantir’s offerings for examples of highly productive user interfaces.

 

 Since the range of possible input values is so large, each new customer deployment is likely to generate data that has never been seen before. Even customers that appear similar – two auto manufacturers doing defect detection, for example – may require substantially different training data, due to something as simple as the placement of video cameras on their assembly lines.

 

Software which solves a business problem generally scales to new customers. You do some database back end grunt work, plug it in, and you’re done.  Sometimes you have to adjust processes to fit the accepted uses of the software; or spend absurd amounts of labor adjusting the software to work with your business processes: SAP is notorious for this. Such cycles are hugely time and labor consuming. Obviously they must be worth it at least some of the time. But while SAP is notorious (to the point of causing bankruptcy in otherwise healthy companies), most people haven’t figured out that ML oriented processes almost never scale like a simpler application would. You will be confronted with the same problem as using SAP; there is a ton of work done up front; all of it custom. I’ll go out on a limb and assert that most of the up front data pipelining and organizational changes which allow for it are probably more valuable than the actual machine learning piece.

 

In the AI world, technical differentiation is harder to achieve. New model architectures are being developed mostly in open, academic settings. Reference implementations (pre-trained models) are available from open-source libraries, and model parameters can be optimized automatically. Data is the core of an AI system, but it’s often owned by customers, in the public domain, or over time becomes a commodity.

That’s right; that’s why a lone wolf like me, or a small team can do as good or better a job than some firm with 100x the head count and 100m in VC backing. I know what the strengths and weaknesses of the latest woo is. Worse than that: I know that, from a business perspective, something dumb like Naive Bayes or a linear model might solve the customer’s problem just as well as the latest gigawatt neural net atrocity. The VC backed startup might be betting on their “special tool” as its moaty IP. A few percent difference on a ROC curve won’t matter if the data is hand wavey and not really labeled properly, which describes most data you’ll encounter in the wild. ML is undeniably useful, but it is extremely rare that a startup have “special sauce” that works 10x or 100x better than somthing you could fork in a git repo. People won’t pay a premium over in-house ad-hoc data science solutions unless it represents truly game changing results. The technology could impress the shit out of everyone else, but if it’s only getting 5% better MAPE (or whatever); it’s irrelevant. A lot of “AI” doesn’t really work better than a histogram via “group by” query. Throwing complexity at it won’t make it better: sometimes there’s no data in your data.

 

Some good bullet points for would be “AI” technologists:

Eliminate model complexity as much as possible. We’ve seen a massive difference in COGS between startups that train a unique model per customer versus those that are able to share a single model (or set of models) among all customers….

Nice to be able to do, but super rare. If you’ve found a problem like this, you better hope you have a special, moaty solution, or a unique data set which makes it possible.

Choose problem domains carefully – and often narrowly – to reduce data complexityAutomating human labor is a fundamentally hard thing to do. Many companies are finding that the minimum viable task for AI models is narrower than they expected.  Rather than offering general text suggestions, for instance, some teams have found success offering short suggestions in email or job postings. Companies working in the CRM space have found highly valuable niches for AI based just around updating records. There is a large class of problems, like these, that are hard for humans to perform but relatively easy for AI. They tend to involve high-scale, low-complexity tasks, such as moderation, data entry/coding, transcription, etc.

This is a huge admission of “AI” failure. All the sugar plum fairy bullshit about “AI replacing jobs” evaporates in the puff of pixie dust it always was. Really, they’re talking about cheap overseas labor when lizard man fixers like Yang regurgitate the “AI coming for your jobs” meme; AI actually stands for “Alien (or) Immigrant” in this context. Yes they do hold out the possibility of ML being used in some limited domains; I agree, but the hockey stick required for VC backing, and the army of Ph.D.s required to make it work doesn’t really mix well with those limited domains, which have a limited market.

Embrace services. There are huge opportunities to meet the market where it stands. That may mean offering a full-stack translation service rather than translation software or running a taxi service rather than selling self-driving cars.

In other words; you probably can’t build a brain in a can that can solve all kinds of problems: you’re probably going to be a consulting and services company. In case you aren’t familiar with valuations math: services companies are worth something like 2x yearly revenue; where software and “technology” companies are worth 10-20x revenue. That’s why the wework weasel kept trying to position his pyramid scheme as a software company. The implications here are huge: “AI” raises done by A16z and people who think like them are going to be at much lower valuations. If it weren’t clear enough by now, they said it again:

To summarize: most AI systems today aren’t quite software, in the traditional sense. And AI businesses, as a result, don’t look exactly like software businesses. They involve ongoing human support and material variable costs. They often don’t scale quite as easily as we’d like. And strong defensibility – critical to the “build once / sell many times” software model – doesn’t seem to come for free.

These traits make AI feel, to an extent, like a services business. Put another way: you can replace the services firm, but you can’t (completely) replace the services.

I’ll say it again since they did: services companies are not valued like software businesses are. VCs love software businesses; work hard up front to solve a problem, print money forever. That’s why they get the 10-20x revenues valuations. Services companies? Why would you invest in a services company? Their growth is inherently constrained by labor costs and weird addressable market issues.

This isn’t exactly an announcement of a new “AI winter,” but it’s autumn and the winter is coming for startups who claim to be offering world beating “AI” solutions. The promise of “AI” has always been to replace human labor and increase human power over nature. People who actually think ML is “AI” think the machine will just teach itself somehow; no humans needed. Yet, that’s not the financial or physical reality. The reality is, there are interesting models which can be applied to business problems by armies of well trained DBAs, data engineers, statisticians and technicians. These sorts of things are often best grown inside a large existing company to increase productivity. If the company is sclerotic, it can hire outside consultants, just as they’ve always done. A16z’s portfolio reflects this. Putting aside their autonomous vehicle bets (which look like they don’t have a large “AI” component to them), and some health tech bets that have at least linear regression tier data science, I can only identify only two overtly data science related startup they’ve funded. They’re vastly more long crypto currency and blockchain than “AI.” Despite having said otherwise, their money says “AI” companies don’t look so hot.

My TLDR summary:

  1.  Deep learning costs a lot in compute, for marginal payoffs
  2. Machine learning startups generally have no moat or meaningful special sauce
  3. Machine learning startups are mostly services businesses, not software businesses
  4. Machine learning will be most productive inside large organizations that have data and process inefficiencies

 

 

“AI” and the human informational centipede

Posted in fraud, stats jackass of the month by Scott Locklin on September 2, 2017

Useful journalism about technology is virtually nonexistent in the present day. It is a fact little commented on, but easily understood. In some not too distant past, there were actually competent science and technology journalists who were paid to be good at their jobs. There are still science and technology journalists, but for the most part, there are no competent ones actually investigating things. The wretches we have now mostly assist with press releases. Everyone capable of doing such work well is either too busy, too well paid doing something else, or too cowardly to speak up and notice the emperor has no clothes.

Consider: there are now 5 PR people for every reporter in America.  Reporters are an endangered species. Even the most ethical and well intentioned PR people are supposed to put the happy face on the soap powder, but when they don’t understand a technology, outright deception is inevitable. Modern “reporters” mostly regurgitate what the PR person tells them without any quality control.

The lack of useful reporting is a difficulty presently confronting Western Civilization as a whole; the examples are obvious and not worth enumerating. Competent full time reporters who are capable of actually debunking fraudulent tech PR bullshit and a mandate to do so: I estimate that there are approximately zero of these existing in these United States at the moment.

What happens when marketing people at a company talk to some engineers? Even the most honest marketing people hear what they want to hear, and try to spin it in the best possible way to win the PR war, and make their execs happy.  Execs read the “news” which is basically marketing releases from their competitors. They think this is actual information, rather than someone else’s press release.  Hell, I’ve even seen executives ask engineers for capabilities they heard about from reading their own marketing press releases, and being confused as to why these capabilities were actually science fiction. So, when your read some cool article in tech crunch on the latest woo, you aren’t actually reading anything real or accurate. You’re reading the result of a human informational centipede where a CEO orders a marketing guy to publish bullshit which is then consumed by decision makers who pay for investments in technology which doesn’t do what they think it does.

centipede

 

Machine learning and its relatives are the statistics of the future: the way we learn about the way the world works. Of course, machines aren’t actually “learning” anything. They’re just doing statistics. Very beautiful, complex, and sometimes mysterious statistics, but it’s still statistics. Nobody really knows how people learn things and infer new things from abstract or practical knowledge. When someone starts talking about “AI,” based on some machine learning technique, the Berzerker rage comes upon me. There is no such thing as “AI” as a science or a technology. Anyone who uses that phrase is a dreamer, a liar or a fool.

You can tell when a nebulous buzzword like “AI” has reached peak “human information centipede;” when oligarchs start being afraid of it. You have the famous example of Bill Joy being deathly afraid of “nanotech,” a previously hyped “technology” which persists in not existing in the corporeal world. Charlatan thinktanks like the “center for responsible nanotechnology” popped up to relieve oligarchs of their easy money, and these responsible nanotech assclowns went on to … post nifty articles on things that don’t exist.

These days, we have Elon Musk petrified that a near relative of logistic regression is going to achieve sentience and render him unable to enjoy the usufructs of his toils. Charlatan “thinktanks” dedicated to “friendly AI” (and Harry Potter slashfic) have sprung up. Goofball non-profits designed to make “AI” more “safe” by making it available as open source (think about that for a minute) actually exist. Funded, of course, by the paranoid oligarchs who would be better off reading a book, adjusting their exercise program or having their doctor adjust their meds.

Chemists used nanotech hype to drum up funding for research they were interested in. I don’t know of anything useful or interesting which came out of it, but in our declining civilization, I have no real problem with chemists using such swindles to improve their funding. Since there are few to no actual “AI” researchers existing in the world, I suppose the “OpenAI” institute will use their ill gotten gainz to fund machine learning researchers of some kind; maybe even something potentially useful. But, like the chemists, they’re just using it to fund things which are presently popular. How did the popular things get popular? The human information centipede, which is now touting deep reinforcement networks as the latest hotness.

My copy of Sutton and Barto was published in 1998. It’s a tremendous and interesting bunch of techniques, and the TD-gammon solution to Backgammon is a beautiful result for the ages. It is also nothing like “artificial intelligence.” No reinforcement learning gizmo is going to achieve sentience any more than an Unscented Kalman filter is going to achieve sentience. Neural approaches to reinforcement learning are among the least interesting applications of RL, mostly because it’s been done for so long. Why not use RL on other kinds of models? Example, this guy used Nash Equilibrium equations to build a pokerbot using RL. There are also interesting problems where RL with neural nets could be used successfully, and where an open source version would be valuable: natural language, anomaly detection. RL frameworks would also help matters. There are numerous other online approaches which are not reinforcement learning, but potentially even more interesting. No, no, we need to use RL to teach a neural net to play freaking vidya games and call it “AI.” I vaguely recall in the 1980s, when you needed to put a quarter into a machine to play vidya on an 8-bit CPU, the machines had pretty good “AI” which was able to eventually beat even the best players. Great work guys. You’ve worked really hard to do something which was doable in the 1980s.

“The bot learned the game from scratch by self-play, and does not use imitation learning or tree search. This is a step towards building AI systems which accomplish well-defined goals in messy, complicated situations involving real humans.”

No, you’ve basically just reproduced TD-gammon on a stupid video game.  “AI systems which accomplish well-defined goals in messy … situations” need to have human-like judgment and use experience from unrelated tasks to do well at new tasks. This thing does nothing of the sort.  This is a pedestrian exercise in what reinforcement learning is designed to do. The fact that it comes with accompanying marketing video (one which probably cost as much as a half year grad student salary, where it would have been better spent) ought to indicate what manner of “achievement” this is.

Unironic use of the word “AI” is a sure tell of dopey credulity, but the stupid is everywhere, unchecked and rampaging like the ending of Tetsuo the Iron Man.

Imagine someone from smurftown took a data set relating spurious correlations in the periodic table of the elements to stock prices, ran k-means on it, and declared himself a hedge fund manager for beating the S&P by 10%. Would you be impressed? Would you you tout this in a public place? Well, somebody did, and it is the thing which finally caused me to chimp out. This is classic Price of Butter in Bangladesh stupid data mining tricks. Actually, price of butter in Bangladesh makes considerably more sense than this. At least butter prices are meaningful, unlike spurious periodic element correlations to stock returns.

This is so transparently absurd, I had thought it was a clever troll. So I looked around the rest of the website, and found a heart felt declaration that VC investments are not bets. Because VCs really caaaare, man. As if high rollers at the horse races never took an interest in the digestion of their favorite horses and superfluous flesh on their jockeys. Russians know what the phrase “VC” means (туалет). I suppose with this piece of information it still could be a clever Onionesque parody, but I have it on two degrees of Erdős and Kevin Bacon that the author of this piece is a real Venture Capitalist, and he’s not kidding. More recently how “Superintelligent AI will kick ass” and “please buy my stacked LSTMs because I said AI.” Further scrolling on the website reveals one of the organizers of OpenAI is also involved. So, I assume we’re supposed to take it seriously. I don’t; this website is unadulterated bullshit.

gartner

Gartner: they’re pretty good at spotting things which are +10 years away (aka probably never happen)

A winter is coming; another AI winter. Mostly because sharpers, incompetents and frauds are touting things which are not even vaguely true. This is tragic, as there has been some progress in machine learning and potentially lucrative and innovative companies based on it will never happen. As in the first AI winter, it’s because research is being driven by marketing departments and irresponsible people.

But hey, I’m just some bozo writing in his underpants, don’t listen to me, listen to some experts:

http://www.rogerschank.com/fraudulent-claims-made-by-IBM-about-Watson-and-AI

http://thinkingmachines.mit.edu/blog/unreasonable-reputation-neural-networks

https://medium.com/project-juno/how-to-avoid-another-ai-winter-d0915f1e4798#.uwo31nggc

https://www.researchgate.net/publication/3454567_Avoiding_Another_AI_Winter

Edit Add (Sept 5, 2017):

Congress is presently in hearings on “AI”. It’s worth remembering congress had hearings on “nanotech” in 2006.

http://www.nanotechproject.org/news/archive/congressional_hearing_on_nanotechnology/

“By 2014, it is estimated that there could be $2.6 trillion worth of products in the global marketplace which have incorporated nanotechnology. There is significant concern in industry, however, that the projected economic growth of nanotechnology could be undermined by either real environmental and safety risks of nanotechnology or the public’s perception that such risks exist.”

Edit Add (Sept 10, 2017) (Taken from Mark Ames):

My favorite photo of this wacky election

Posted in stats jackass of the month, Uncategorized by Scott Locklin on November 9, 2016

This dope got lucky in 2012, essentially using “take the mean” and was hailed as a prophet. He was wrong about virtually everything, and if someone were to make a table of his predictions over time and calculate Brier scores, I’m pretty sure he’ll get a higher score than Magic-8 ball (Brier scores, lower is better). Prediction is difficult, as the sage said, especially regarding the future. Claiming you can do prediction when you can’t is irresponsible and can actually be dangerous.
silver.jpg

While he richly deserves to find his proper station in life as an opinionated taxi driver, this clown is unfortunately likely to be with us for years to come, bringing shame on the profession of quantitative analysis of data. We’ll be watching, Nate.

 

 

Michael Lewis: shilling for the buyside

Posted in finance journalism, microstructure by Scott Locklin on April 4, 2014

Journalism, in the ideal world, is supposed to inform the citizenry of facts important to their well being. Modern journalism seems to involve issuing press releases from the oligarchical reptiles who are destroying Western Civilization. Maybe I am a naive fool, and it was always thus. Either way, Michael Lewis’s latest book lends credence to the view that he is a very modern journalist.

lewis_a

lookin a bit scaly, bub

Lewis’s book purports to be about high frequency trading. He manages to write several hundred pages of gobbledeygook without actually speaking to a High Frequency Trader (unless you count his incongruous encounter with poor Sergey Aleynikov ). The story Lewis actually tells is one of incompetent sell side traders who started an exchange which serves the interests of wealthy buy siders and shady brokers.

Brad Katsuyama is the hero of the book. Lewis’s recent books use the dreary trope: the band of clever and plucky outsider misfits who take on the establishment. Katsuyama’s misfittery is he’s an Asian who is good at sports, bad at math and computers; and even though he worked for a Wall Street bank, he went to a crappy Canadian school instead of Yale-vard. Among his misfit sidekicks are a potato-wog who is good at network ops, but who could never get a break on the street (I can relate). Also, a fat grouch from Brooklyn, a computer genius, a puzzle wizard and a few other guys who fade into the woodwork. They worked for RBC: a Canook bank which is supposedly the least Wall Streety place on the Street.

The plucky outsiders in this story are not portrayed in a particularly flattering way. In fact, they come off as  dimwitted incompetents. Katsuyama was an old school block shopping sell side trader. If you remember my previous pieces on  HFT nay sayers: Joe Saluzzi was also a sell side block shopper. Old fashioned sell side guys have obsolete jobs. Their jobs are to find liquidity for “buy side” customers buying into or liquidating a large position. Katsuyama’s anger at the idea that “the market is rigged” seems the simple rage of a man who has been assigned a task he is not qualified for. There are tales of he and his team wasting hundreds of thousands in RBC money executing bad trades to see what happens. They seemed shocked, shocked, that the market would move away from their ham-fisted dumpings of huge blocks of shares to someone else’s routing system.

fbi

Lewis keeps going on about how “nobody understood” any of this back in 2009, except for his plucky outsider heroes. If “nobody” understood it, how was I was able to write about it on my blog in 2009? Over 100,000 people read my various blogs on HFT that year.  If you were not among the elite group of more than 100,000 insiders who read blogs, any punter could have purchased the Larry Harris book “Trading and Exchanges” available on Amazon.com for $71.58 + tax. This is how I originally clued myself in (thanks FDAX-H). Larry’s book was published in 2002.  In early 2010 Barry Johnson published the book, “Algorithmic Trading and DMA” which explains the profession dedicated to getting a good fill on the modern electronic trading landscape. So, in 2010, there was not only a  job description, “algorithmic trader,” for getting a good buy-side fill, there was also a “how to” book on the subject.  Such people perform the function that used to be done by sell side people like Katsuyama and Joe Saluzzi. Lewis repeatedly states that this was a mysterious topic and nobody was talking. Actually, it is an extremely well understood topic;  library shelves groan with volumes dedicated to the subject.

No books were really needed; history and experience should suffice. Back in the days of pit traders, if you threw a huge order at the pit, you might get a fill on a couple of round lots. The rest of the pit is going to change their prices, because they figure anyone swinging 10,000 or 100,000 share orders around must be informed traders. If they’re informed traders, they need to pay for their immediacy. Informed traders may be criminal insider-trader creeps,  they may be people with really good trading strategies; it doesn’t matter -they’re informed somehow: they know stuff. If the market maker doesn’t adjust their prices in front of an informed trader, the market maker will go bankrupt. That’s market economics 101. As I previously described it in 2009 in the Three Stooges of the High Frequency Apocalypse;

What happens when you buy something? …  If you want it for cheap, you sit around and look at different markets (ebay, amazon, craigslist) until someone displays a price you find acceptable. If you want that “something” right now, you drive to a store and buy it. You’ll almost certainly pay a little more at the store, because they need to make enough money to pay employees to prevent barbarians from stealing everything, and to keep the lights on and other such things for your convenience. You can also generally return what you bought to the store much easier than to ebay or amazon. You’re paying for the immediacy (buy it now!) and liquidity (buy as many as you want!) provided by the store. This is a service which costs money.

Immediacy costs money. Markets have always moved prices away from large orders. Market participants  have always been able to cancel or move a limit order. That’s one of the features of the limit order. If  Katsuyama didn’t understand these simple facts, he had no business collecting a $2 million a year paycheck shopping blocks for his customers, because he didn’t understand the basics of his profession. It’s  possible that Lewis simply misunderstood something Katsuyama explained to him. It’s also  possible that Katsuyama is a shark who told Lewis a lot of bullshit to get good press for IEX.  This leaves only two possibilities: either Lewis is a credulous idiot who is not competent as a journalist, or Katsuyama is an idiot who was not competent as a trader. Take your pick.

 

CBOTpit1900-GB

They made their money trading flow as well

Where it gets interesting is where Lewis claims bigshot buysider crybabies like Loeb and Einhorn  never heard of any of this. They made it sound as if, back in the day when Loeb and Einhorn were paying 1/8 of a dollar spreads to knuckle-dragging pit orcs, no rock-ribbed he-man trader with 10lbs of undigested beef in his lower intestine would would dare move his price away from where Loeb and Einhorn wanted it. Why, moving the price away from a big order: that’s un-American!

So … these “plucky underdogs” helped Katsuyama form a new stock exchange, IEX. They claim that no sort of nefarious activity is possible on IEX, because, well, “trust us!”  Liquidnet’s average cross is 45,000 shares; over 100 times the vaunted liquidity figures provided by IEX.  If I traded stocks, why should I trust IEX over Liquidnet? Because Michael Lewis says they’re honest guys? If I believe the tales of Michael Lewis, the founders of  IEX are a collection of “traders” who do not know how to trade, and the market itself is owned by … buy side traders. He seems to give IEX sloppy wet kisses for honesty, yet sees nothing wrong with the fact that they’re owned by a bunch of buy side guys. They’re also owned by some unknown buy side guys, which does not inspire confidence. Buy side guys, if they’re good at their jobs are informed traders.  Nobody wants to trade against informed traders. Everyone wants to trade against noise traders.

cbot1929

IEX  has simple order types; limit, midpoint, fill or kill and market: I  approve of this. On the other hand:

“IEX follows a price-priority model first, then by displayed order second. Then comes broker priority, which means a broker will always trade with itself first, which Katsuyama described as “free internalization.” He explained that brokers do not pay IEX to trade should an order be matched against another order from that same broker. This, he added, offers brokers incentive to trade in IEX.”

Hey now, wait a minute. Internalization and broker priority is pretty much the same thing as dark crossing, which Lewis was trying to tell us was bad. Now it’s supposed to be OK when Goldman does it?  Later, Lewis actually quotes Katsuyama saying there were only a few brokers acting in their customer’s interests:

“Ten,” Katsuyama said. (IEX had dealings with 94.) The 10 included RBC, Bernstein and a bunch of even smaller outfits that seemed to be acting in the best interests of their investors. “Three are meaningful,” he added: Morgan Stanley, J. P. Morgan and Goldman Sachs.

I think this is the crux of this story: according to Michael Lewis and Katsuyama, we’re supposed to trust people like Einhorn who have been convicted of insider trading, people who are suspected of insider trading (buy side is by definition rife with this; particularly firms that do merger arb and special events), J.P. Morgan, Goldman and Morgan Stanley: we’re supposed to trust these guys more than we’re supposed to trust a bunch of tiny little market making firms who had been inconveniencing them by taking away some of their flow. Lewis tries to make this seem like a battle between the underdog “good guys” and the evil establishment. To believe this, you’d have to believe that Goldman Sachs and people like Einhorn are underdogs, rather than the actual establishment. To believe this, you’d have to believe the tiny industry of HFT traders actually rules the world and buys off congressmen and the SEC more than … J.P. Morgan and Goldman.

To give you a sense of scale: the largest HFT firm I know of, KCG, has operating cash flows of $140 million a year and a  modest market cap of $1.4 billion (betcha didn’t know it was a publicly traded company: Lewis certainly doesn’t mention it). JPM has operating cash flows of  $100 billion a year, almost a trillion on the balance sheets, and a quarter trillion or so in market cap. David Einhorn is personally worth $1.25 billion dollars. KCG’s entire market cap is only slightly more than that, and it employs 1200 people. Yet, somehow the HFT firms are the evil establishment, and JPM and Einhorn are … the plucky underdogs standing up for truth, justice and market makers not changing their quotes when some reptilian oligarch dumps 200,000 shares of YoyoDyne on the market.

Yeah, I might believe that. I might believe that if I were a dribbling retard.

they-live-16

Put down Lewis’s book; read one by Bernays

Doing a bit of investigation into who owns IEX: we have the $13.2 billion  activist shareholder fund Pershing Square, owned by Bill Ackman, another “underdog” worth $1.2 billion. We have the $6.7 billion Senator Investment Group. Scoggin Capital is only worth $1.8 billion; they do distressed debt and mergers, and have managed to only have one down year in 25. Another investor is venture capitalist  Jim Clark, net worth $1.4 billion. He is particularly noteworthy as being a pal of Michael Lewis, and almost certainly the guy who made the introductions to the “flash boys” at IEX. Brandes Investment Partners is an old $29 billion AUM politically influential money management firm doing value investments, and is run by another billionaire. Third point, a hedge fund with $15 billion, also working in special situations aka “distressed debt and mergers,” run by  Danny Loeb (who also miraculously has only one down year). Another investor in IEX is a little place called  Capital Group Companies, one of the biggest buy side investors in the world, with $1.15 trillion AUM. Capital Group has been more or less scientifically proven to be one of the most powerful and influential corporations in the world.

You get the idea: IEX is not owned by plucky underdogs. It is owned by very rich and powerful “buy side” people. People who find the present system of liquidity provision inconvenient.  Buy side has always found liquidity providers inconvenient; they had to pay old school “sell side” traders like Katusyama to work the trades for them at the very least. There wasn’t much they could do about it until now. Now that they own  almost everything, they can open their own damn stock exchange and buy some cheap brokerage flow. That and unleash Michael Lewis, the FBI and New York Attourney General on the peasants who make them pay for liquidity.

TheyLive1

IEX … little investor… blah blah blah

I  don’t think IEX and their investors represent the interests of “the little guy” at all. The actual little guy (aka people like me) does pretty well making small orders with the present system. If you  believe Lewis’s book, the thing we’re supposed to be worried about is telegraphing a big buy or sell by routing  your order to several different exchanges. The thing is, “the little guy” doesn’t make large buy or sell orders, and unless he does, what Lewis describes is impossible. The people IEX benefits are exclusively preposterously wealthy buy side people. That and the brokerages who get to trade against the pieces of their flow that they want. Pardon me if I notice that such people aren’t exactly tribunes of the people. What’s actually going on here is the brokers are, as usual, taking the flow. They’re giving up some of the leftovers to the buy side guys, who also pocket the exchange fees. If you’re worried about flow or think the present system of liquidity provision is somehow predatory: this is a buzzard and a hyena sharing a carcass.

I have no dog in this race: I’m not a HFT, I have never taken a dime from any exchange, and I haven’t so much as executed a stock trade in 4 years.  Everything I’ve read, and all the traders (buy side and otherwise) I’ve spoken with seem to think that HFT market makers have improved things from the pre-decimalization bad old days of pit traders who got their jobs because they went to the right New York City high schools.  I know for a fact that HFT market making as a business is nowhere near as profitable as it was even a few years ago. This is exactly what you would expect when you have lots of smart people competing in a not-so profitable business. I don’t think the use of computers makes markets any more inherently dangerous, any more than the use of computers in automobiles makes them  more inherently dangerous. If you asked me what I thought the worst thing about the present system was, it would be the profusion of weird order types. Something that IEX, to their credit, gets right.  There are actual frauds in HFT, just like there are in any other business involving money, from the Avon lady on up the food chain. The worst HFT tort I can think of is the practice of “quote stuffing.”  Lewis (of course) never mentions this, and I have read nothing which indicates IEX is ready for it.

I know a few HFT type people. One of ’em might be even be as rich as Michael Lewis.  So far, all the ones I have met are clever and decent people, and I figure whatever they’ve managed to earn by the sweat of their brows, they deserve it. I’m not real pleased with the idea of a small group of decently paid, politically helpless nerds being the fall guys for a bunch of crooked oligarchs who don’t want to pay for their liquidity.

Speaking of which: FREE SERGEY

 

sergey-aleynikov-1.png

despite the awesomeness of his pantaloons, this man’s legal problems are a bad sign for America

 

Edit Add:
This review by a trader lists 15 more technical inaccuracies in the book. He also noticed that broker priority is shady business if we’re talking about helping “the little guy” here.

Edit Add2:

This trader gives a really great review.