Join institutional investors, analysts and strategists from the world's largest banks: Subscribe today for as little as $7/month
Already have an account? log in
Amid The Volatility, Here’s The Latest From Marko Kolanovic, Goldman On Systematic Flows And Liquidity Provision
Back on December 29, in "Trump, Machines And Markets: A Willing Suspension Of Disbelief", I talked at length about the extent to which some in the quant community (i.e., some who employ quantitative strategies) were engaged in an increasingly desperate-sounding effort to convince the investing public that systematic flows don't have a meaningful impact on markets.
That effort, I argued, stems at least in part from self-preservation concerns and an effort to stave off the inevitable in terms of
I first read your analysis on Seeking Alpha…..figured a guy that used Breaking Bad as a pseudonym might have a bit different take on the Markets than the gang trying to sell you their expertise.
It has been an education……I am just a plumbing contractor who wanted to take a more hands on approach in investing my retirement savings. I’m also a total political junky who enjoys good writing and a sense of humor…..which you have in spades.
I’m the guy who would have been told……”you just won’t understand” a year ago…..and would have agreed. After reading you for over a year……I’ve gotten quite the education…..and can understand far more than before.
Thanks for that, for being an amazing news feed, for providing a laugh a day, for brilliant political and economic analysis.
I really appreciate your hard work…….not sure why you do work this hard, but it is appreciated.
thanks for the kind words.
Agree. Best $5 I spend each month
Machine-learning algos are trained on historical data that was generated by human (emotional monkeys) carbon-based traders. They, therefore, trade like their teachers did, only faster. This makes it look like the machines are wildly emotional. At some point, however, the training data is going to contain some threshold of machine-generated trading data that will produce a new type of result…machine creativity; new moves that have never been seen by human.
AI
Your last conclusion is very interesting and a point I have not seen publicly up to now. This is the way machines will gradually wield power, whether we like it or not. We will never know what part of their learning is based on what they do for themselves and so we will be powerless to stop it. We want the tool and this is what we will get as a Secret Santa present along with it. Ain’t we got fun?
In professional chess-playing, any creative move that has never been seen before will disqualify you because it implies AI was used. Why a machine-learning algo does what it does is a black box. Even the guy that programs the neural network doesn’t know why is does what it does. You can’t know the patterns it has seen.
I can see a time when instead of betting on stocks or ETFs, we are betting on specific algos that in turn bet on something we have no idea of. That has already started, in a clumsy way. Impossible to know what it all means for the future. All we can be sure of is that it WILL happen.
Look up AlphaGo.
Exactly, well said
Thank you. As always, thoughtful heisenberg report.
I second the comments of Anonymous above. I tend to fall into the rut of reading news and opinion from the same sources all the time. And that is rarely a good idea. I absolutely love hearing a divergent voice from someone who is looking at facts I may not be thinking about, and who lays out all the information supporting his opinion. Thank you.
As to all the quant trading, as a small investor, I can’t do much about it. But I need to understand what is happening and where we are going. A couple weeks ago, there was a great article in The NY Times about the progress that computers made in chess, go and other games: https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html
The computers started with human instructions on how to play the game. But over the years, they switched to a mode where they just figured it out on their own. That works in a zero sum game, where, no matter how complicated, there are finite variables and finite possibilities. When I saw that article, I thought about the use of computers in trading. In trading, there are infinite variables and infinite possibilities. Chess and go are easy by comparison. Still, they should be able to process much more than any human. And what seems like volatility to a human may be much less so to a machine processing information basically at the speed of light. In the coming years, I have to believe we will see tremendous progress on the use of machines in making decisions where possibilities are infinite, and trading would have to be one of the profitable immediate uses.
No offense to second-hand Kolanovic/Gandalf (thanks, love it), but now waiting on Powell&Co. to signal a drop in FFR after dragging Yellen and Bernanke into the fire pit yesterday (i.e., explicitly acknowledge policy mistake).
Kind of like explaining away anthropogenic climate change; because the climate has changed throughout the history of the planet. A deception strategy that works for the groups that need to say it, and the people who want to hear/believe it.
2017 I think showed the impact of passive and quants/momo. Markets do not act like that (even with the data/policy etc) when humans are involved. Humans de-risk, look for mean reversion, look for what can go wrong etc.
AAPL in 2018 is another good example. A real company, real earnings, big market cap. Did fundamentals really change enough for investors to rally it (almost straight up) 50% from March to Sept (the smart money was not long a lot on the last of the move) then realizing over charging for the same thing they sell it off 40+% from the highs (did the services business really change $300-450bn or did the hardware business?).
Yes humans get caught up in fear and greed but computers do much better in straight up and straight down momo trading.
The last couple of years show to me tge impact of the computers and how much trend following they are relative to professional investors/traders in the past.
Imagine robot systems being fed knowledge about each other, and capital market microstructure liquidity vulnerabilities. Then imagine a human with high frequency technology that is ready for all that. ETF liquidation scenarios can also be fed in to all this. Eventually you might want to avoid using margin, especially if you are piling on to a crowded trade like we just saw with short 10 yr.
There will be “winners” in the early algo days (now) due to differing quality / performance, but will there be when there are 1×10^9 comprehensive yet different algo platforms using AI/ML trading? I think not… it will be a stabilizing, mean reversion exercise due to the balance created by pulls in different directions as algos search for infinitesimal advantage over one another. Rest easy my friends.