Behind the Market Swoon: The Herdlike Behavior of Computerized Trading
wsj.com
wsj.com
Yes, there is herdlike behaviour. But why?
Here's a little story about my investment career.
I once hired a guy for a fund I was partner in. He was a proper old school equity investor. He'd fly around the world to different countries and visit businesses. He'd think about each country's prospects, each industry, and each company. He'd meet withe the CEOs and look them in the eye, and ask them whether his company was gonna make money. I shit you not. He's come back to the office and share his thoughts on why one country was good, why this industry, why this business. It was a different story each time.
Another guy I worked more closely with on a fixed income desk. We'd think about different countries, their bonds, their IR rates, swaptions, etc. There's be some global stories inevitably, but a lot of local ones too and we'd chat about what to do in our trading book.
Then there was quantitative FX trading, another one of my projects. We'd look at things that affect currencies, patterns in the prices, anything you might imagine arrives as a live feed. And we built the infrastructure to trade the regularities that we found. Quite a lot of research and execution infrastructure. But at the end of the day, the computers are doing what? Looking at a variety of information and judging what will happen. Just like my two colleagues.
Fast forward to the last few years. Pretty much every conversation I have with anyone in the investment business talks about one thing: QE and zero rates. There's only one thing that matters for everything now. When interest rates are really low, what happens? Everything is worth buying. Buy houses, buy stocks. Buy them in the developed countries, buy them in EM. Sell options.
What used to be a conversation containing rates as an ingredient has become a conversation only containing rates.
People, and computers which learn what worked for people in the recent past, have gone from a rich market conversation about many things to one about just QE.
It will be interesting to see what happens in the near future. I can't think of a lot of bubbles that were deflated in a controlled manner. And I also expect more differentiation in equity debates. Company does this, industry does that.
> People, and computers which learn what worked for people in the recent past, have gone from a rich market conversation about many things to one about just QE.
So are you saying that in the past N years since QE and 0 rates, the parameters that define which assets/resources to buy/trade have gone from wide-ranging (e.g., "We'd think about different countries, their bonds, their IR rates, swaptions, etc"), to singular (e.g., "...has become a conversation only containing rates.") ?
How does this tie back into the "herd-like behavior" comment? Are you saying that every algo trader is only focusing on rates because that (more so than anything else), is determining where to allocate capital?
But they can herd into factors specific to their markets that are nonetheless still the same thing, due to rates causing behaviour everywhere.
Well, the Fed seems to ending that. Certainly it's signaling an end to this. How would this change the conversion?
"OMG SELL"
I assume because financing is less than inflation?
Wouldn't you still have the same relative tradeoffs? Bonds lower yielding, but more liquid. Stocks riskier etc.
>the computers are doing what? Looking at a variety of information and judging what will happen
Probably but not necessarily. That code which is explicitly understood and written explicitly to weigh information to choose outcomes certainly is.
But were complex neural networks are extensively involved, they become a black box. In markets where gains could be maximized by cooperation, those systems that find ways to cooperate with other systems outside their organization will eventually be chosen and become the dominant. Chosen purely from their ability to get better results.
Even if completely isolated, signaling is not impossible. A straight forward technique would be to use transactions to communicate in the same way bridge players do.
So potentially software could outperform humans for at least two reasons: better and faster analysis of data. But also doing things humans are unable.
But many traders will claim their use because they are obtuse and impressive sounding.
One PM once described it as "machine learning in the streets, linear regression in the sheets".
When passive funds as a group have net outflows, all their holdings must be reduced in roughly the same proportion. But passive funds as a group cannot reduce their holdings by selling stocks to each other! It's impossible to take water out of a boat by scooping water from one spot and pouring it back into the same boat in some other spot!
Therefore, to reduce holdings, passive funds (or their broker-dealers) necessarily must find other -- i.e., non-passive -- buyers willing to take those stocks at some market-clearing price. Alas, everyone else, in the aggregate, also owns the same stocks in the same proportion.
Potentially, this can produce temporary imbalances between passive net sellers and non-passive net buyers that get corrected via declines in price. Passive net sellers, being price agnostic, do not care; they're on automatic.
Right now, the pool of people putting money into the market has been steadily increasing as Millenials enter the workforce. Boomers are retiring, but not really in large numbers yet, so the overall number of folks saving for retirement has been continually increasing since Boomers started entering their prime earning years in the early 1980s.
However, the peak of the Boomer years will soon start entering retirement soon (~2020) while the early Boomers are starting to die off. And the generation after Millenials - who will start comprising the workforce in 2020 - is much smaller than the Millenial generation. There are fundamental issues with demographics that can't be papered over by financial engineering: a smaller working-age population supporting a larger dependent population (absent massive technological advancement in care) = lower standard of living for everyone.
Demographically, this would play out as a generation-long bear market, but markets tend to correct as soon as everyone adjusts their expectations for the future. That implies a sudden and massive crisis at some point with stock market levels correcting to the yields and prices of the 1970s, adjusted for inflation. That would imply an S&P 500 of about 370 (down ~95%) and interest rates in the 10% range.
On average, assuming nothing else changes. In reality, we have the levers of immigration and workforce utilization (e.g. making it easier for ex-felons or non-violent drug convicts to re-enter the workforce) to juice labour inputs. And we have education and R&D to juice productivity. On top of that are random factors like natural resource utilization, falling energy and material intensities of GDP, et cetera.
> That would imply an S&P 500 of about 370 (down ~95%) and interest rates in the 10% range
One cannot so neatly connect demographics to interest rates, let alone the S&P 500. Demographics map to certain consumption sectors (e.g. durables) very well. They map to others (e.g. luxury or entertainment products) quite badly.
To illustrate one of the many confounding variables between demographics and stock prices, consider what a small shift in asset allocations in a single generation from real estate to equities would do to prices.
However, I agree with JumpCrisscross (https://news.ycombinator.com/item?id=18763124) that such a horrific scenario is unlikely.
What would reverse this, internationally, is if the high-youth countries of Latin America, the Middle East, sub-Saharan Africa, and rural India could be rapidly integrated into the global economy. There are large political, cultural, and educational barriers to this, though.
Boomers were born 1945-1960, meaning in 2020 they'll be 60-75. They started retiring long ago.
This describes Japan for the past decade or so, right?
Also, are you saying the standard of living would regress back to the 90's/80's/etc., or that growth in the standard of living would be curtailed significantly?
However, sectors where productivity hasn't risen that fast - like health care, elder care, mental health services, natural resources, etc. - will encounter sharply rising prices as there are too few workers to provide services to everyone who needs them. The market is a way of allocating resources to those who are willing to pay the most - so those with resources will still be able to afford them, but will end up liquidating a lot of their investments for it, while those without assets will just have to go without. In the process, the exodus of cash from financial markets to health & elder care will drag down asset prices in general. It'll be a good time to put money into the market, but a bad time to have money already in it.
You're saying that the rate of growth in the standard of living would be curtailed, and that the degree of curtailment would be sector dependent? If so, then I definitely agree.
Realize it doesn't meet the true technical definition, but more the idea of passing the buck (so to speak) and someone getting hit pretty hard, in an inevitable / unavoidable sorta way. Also, because this also (to me) feels less like investing and more like a timing game.
But you can balance a boat and keep it from tipping over that way.
THIS. This is how it works. They also have similar basis prices for their positions, and similar pain thresholds. It's not a big surprise that when Institution XYZ reaches its' pain threshold and stop loss orders are used, a few more dozen Institution ABC, DEF, and GHI hit theirs also! This looks like a "huge selloff" on a chart, but it's just the same event being experienced by multiple institutions near the same point in time.
tldr: it looks like a herd because it is a herd.
Passive funds have no pain thresholds which force them to sell. Investors in them may. But that’s a difference in how individual investors’ risk tolerances are abstracted to broad market pricing, not a change in those risk expressions themselves.
edit: Sometimes I'm blinded by the part of "market" I operate within, which is the commodity futures market. I could be(and likely am) completely wrong when you apply this to the securities market, which I am less familiar with.
Human behavior (even human-programmed behavior) is pro-cyclic. Everyone wants the same stuff and makes the same decisions with the same input. I don't see anything notable about this market cycle as compared with previous ones at all, only the jargon is changing.
Twenty years ago was 1998. Ten years before that, Black Monday was faster and more vicious than anything we've seen since. It remains "the largest one-day percentage decline in the DJIA" [1].
Pretty sure that was exactly my point. I'm sure someone finds the behavior "interesting", but if the same thing is happening for the same reason, I'm arguing those details that everyone finds "interesting" aren't "important".
That's voodoo thinking. "OK, all those other times everyone was wrong about this. But this time we have new jargon, so we can figure it out!"
Nothing in the linked article (or anywhere else) gives you the magic you want. This is just a new way of explaining long-settled ideas.
This is like trying to debug a linked list bug via disassembly. "OK, fine, it was a bug before on x86, but look, now it's compiled for RISC-V and uses entirely different register schemes, so this time it will work!"
That's not how any of this works. Every share bought is a share sold by some counterparty. Buyers and sellers are always in equilibrium.
To explain it a step further... at this moment in time, every private and institutional investors stopped selling APPL...I can still buy a share, likely thousands of them... from the market participants that are always there: market makers. When you make a "bad call"(like selling into a rally), a market maker is likely on the other end of your trade, and they will profit from your "bad call". Now the inverse also applies, often times a market maker is taking the other end of your trade that is a good(profitable) trade for you. The market maker isn't losing though, they are just playing the odds. They are convicted that for every losing trade they take out of obligation(as a market maker), they are going to take 2 or more winning trades. They also operate with trade costs much lower than you or I(ie retail investors) have access to.
That was more reply than I originally intended to write...but you have to understand this(or fail at profitable trading). There is no equilibrium, and there are parties(market makers) ensuring that there never will be. That is their job, to create a state of constant liquidity, even if buyers and/or sellers individually are unwilling to play.
Your understanding is exactly backwards. Market makers provide the "magical equilibrium" by bridging supply and demand across time.
> and if there was, there would be no profit.
Wrong again. Even with infinite shares on the bid/ask, there's still a spread for market makers to collect.
More generally, what exactly do you think your comment proves? If a retail investor buys the sole 100 shares at top-of-book, the price (mid) moves up, but there's one buyer and one seller. Where's the "delta"? And if a big hedge fund buys 100,000 shares from 10 market makers... 100,000 shares demanded and 100,000 shares supplied. Magic!
You're arguing against an accounting identity. I know what you're trying to say: what moves prices is relative eagerness of buyers and sellers. You're just too inexperienced to be able to explain it.
So contrary to the article, most of these funds don’t take broad bets for or against the market. What they are really doing is correcting the relative valuations of each individual company by shifting bits of capital away from overpriced companies and towards underpriced ones. That doesn’t have the effect of raising or lowering broad market indices.
Now there are of course other investors who take broad directional bets on the market but those are probably more likely to be discretionary bets rather than systematic.
Delta neutral. And this is very difficult (and expensive) to attain in real life. Most funds are long biased because it's cheaper to be long than short.
Why do you think its so difficult and expensive to attain? What do you think institutional borrow costs are? And I'm not talking about Interactive Brokers... I'd wager that most funds are long biased because its more difficult to generate short alpha - not because of the costs of portfolio construction.
For the vast majority of symbols, stock borrow is cheap at institutional size.
Stock borrow costs vary from symbol to symbol. The ones you want tend to be the ones you pay for. Moreover, cheap doesn't mean free. Shorting half of one's portfolio is rancidity expensive compared to buying some puts. Even that is pretty expensive for total hedging purposes.
At the end of the day, most long-short funds are not market neutral. The ones that are operate on minuscule time horizons, and are better categorized as shadow market makers than funds. (Both in their operation and fundraising mechanics, the short-term traders tending to be more likely to deploy permanent--versus limited--capital.)
Imagine you think Apple is better than any other technology company and you want to bet it will outperform other technology companies. If you simply buy Apple stock, you might lose money even if it outperforms all other tech companies, in the case that the whole market is going down. Because of the above, you need to devise a strategy that will earn you money on the difference between Apple stock and the overall tech sector stocks. Those strategies usually involve combinations of stock and derivatives and you can optimize your returns vs exposure with simple models.
Algorithmic, low latency trading reduces bid-ask spread under normal circumstances, but increases the risk of extreme events.
Popular Delusions And the Madness of Algos !
October 24 - 11% drop, followed by some recovery. October 28 - 13% drop October 29 - 12% drop
The lowest point in 1929 was reached on November 13. The bottom of the crash was on July 8, 1932.