The trading floor is long gone. So are Sammy's Exchange bar and a way of life. As recently as the mid-1990s, stock trading on Bay Street was still done on the Toronto Stock Exchange floor by hundreds of men (there were hardly any women) wearing loud blazers—one team colour or garish pattern per brokerage firm. The acid yellow and black McLeod tartan was renowned. Telephone clerks at desks on the side yelled into receivers; runners raced out with client orders; traders at posts manically hand-signalled prices and then haggled face to face. Paper covered the floor, and not that long before, cigarette smoke clouded the air.
Boozing and frat-boy pranks were almost as important as the trading itself. In 1980, an enterprising Palestinian-born former insurance salesman named Sammy Salhia opened the first Sammy's Exchange—with a then snazzy new scrolling LED ticker-tape display on the wall—in the underground shopping concourse right next to the old art deco Toronto Stock Exchange (TSE) on Bay Street (now a prettified event venue, the Design Exchange).
There were rival bars nearby, but Sammy's quickly became a hub. Mark Grimes, who started as a trader in 1979 and is now a Toronto city councillor, recalls that he and some pals "got into the martinis at lunch" one day and, just for a lark, came back to the floor with a 52-piece marching band. "It was a different time," Grimes says. When the TSE moved to a larger new trading floor around the corner in 1983, Salhia opened another Sammy's next to that. (That space is quieter now—it's a luggage store.)
Greg Mills remembers the old days too, but rolls his eyes about the bars and the shenanigans. Like a lot of floor traders, he was no finance specialist when he started. He had an undergrad degree in geology and worked on the floor in 1985 as a phone clerk for Burns Fry, wearing the firm's yellow jacket.
Compared with today's rapid-fire, down-to-the-nanosecond online stock trading, the floor was primitive. Cutting-edge technology, Mills says, was "how fast did I flash a hand signal and get it right?" Using a swift beckoning motion meant "buy," pushing away with your hands meant "sell." Traders also used yells, "almost like a bark or a scream," to catch one another's attention, Mills says.
Since then, the 56-year-old has climbed a long way in a business in which technology has raced ahead, almost literally at the speed of light. The TSE closed the floor and went fully electronic in 1997. By the early 2000s, U.S. and Canadian regulators had modernized rules that unleashed a tsunami of electronic trading. Dozens of new online markets opened, and a flurry of novel pricing models and more complex order types soon followed.
Mills is now head of the global equities division at RBC Capital Markets, the biggest investment bank on Bay Street. From his glassed-in office, he overlooks rows of desks and screens on the high-ceilinged trading floor for stocks, bonds, currencies and other assets at Royal Bank's Toronto headquarters.
There are still a few middle-aged shirt-and-tie guys like Mills among the several dozen online equity traders on RBC's trading floor. But in the back row just outside his office are a half-dozen newbies in their 20s who are part of a team of 20 young techies—all of them engineers or other computer specialists, and many of them women. They are building an electronic trading platform that will vault RBC beyond the next frontier: artificial intelligence (AI). In theory, when provided with reams of Big Data from within the bank and public markets, the algorithms will trade and alter themselves without human input.
The newbies certainly don't look or sound like old-school TSE traders. On a recent lunch hour, Cynthia Huang, 24, a quiet and petite statistics grad, was poring over an article in an online mathematical journal. Until she and her other team members started working at RBC, many were only vaguely aware that trading used to be done by over-aged frat boys in colourful jackets. "We're nerdier," says her colleague Fiona Hu, also 24.
The machines are taking over, and for good reason: Over the past two decades, electronic trading has slashed costs for individual investors and institutions, allowed them to get their orders filled in seconds and narrowed market price spreads down to pennies or less. Industry numbers are hard to come by in Canada and the United States, but Goldman Sachs, the second-largest investment bank on Wall Street, last year disclosed that it has thinned the number of equity traders it employs from 600 in 2000 to only a few today. Instead, trading at Goldman is now guided by hundreds of engineers.
One look at the newest staff on RBC's trading floor and it's clear the vocation is never going to be the same. The era of flamboyant star traders—such as Gordon Capital founder Jimmy ("The Piranha") Connacher in the 1980s and GMP's Michael Wekerle in the 1990s and early 2000s—is over. And now that the machines can learn and think for themselves, the era of human traders, period, may be drawing to a close.
Quants, techies and software engineers may be the kings and queens of today's virtual trading room, but John Christofilos, now chief trading officer at Toronto-based asset management giant AGF Investments Inc., remembers when floor traders, some with just a high school education, ruled the roost. He traces the start of his career to another Bay Street bar called the Cork Room, where his father was a bartender. In the early 1980s, the teenage Christofilos sat at the end of the counter and listened to traders brag and moan about what went down that day. "I got the itch," he says.
That itch turned into a summer job as an input operator at a post on the TSE floor in 1982. Traders yelled stock prices to him, and he frantically typed them into a computer. The prices would then load onto electronic price boards above individual trading posts, each handling a couple of dozen stocks. "You got to look in people's eyes and get a sense of the market," says Christofilos.
In those days, trading was centralized in a venue where all member dealers could compete, and prices were transparent to those in the room. But floor trading was restricted and costly. Stocks only traded on the exchanges on which they were listed, and firms had to pay to be members. The smallest price increments for major stocks were fractions of a dollar—one-quarter (25 cents) and one-eighth (12.5 cents). Commissions for brokers were fat.
After Christofilos finished university in 1987, he became a floor trader for Hector M. Chisholm & Co. Seniority ruled. It was a lot like a basketball court, he says, where all-star veterans get favourable calls from officials and rookies do not. There were lots of pranks, too. Under their brokers' jackets—Christofilos's was navy—traders had to wear a shirt and tie. One day a senior trader jumped in front of him and cut his tie in half with a pair of scissors. "That's a $25 fine, son. You can't have a tie that's not full-length," he said. Then, the trader stapled it back together and waived the fine. "I'm a kid, I just paid $7 for this tie," Christofilos says. "It's an expensive tie."
The booze also flowed. Doug Clark started as a retail trader with Toronto-Dominion in 1993. As he tells it, rookies had to take an oral exam for their trading licence, and the exchange's chief regulator conducted the exams at Sammy's. They weren't tough exams to pass. "The answer to every question was 'you call surveillance,'" Clark says. He is now managing director of research for the Canadian branch of Investment Technology Group, Inc. (ITG), the giant New York–based online trading firm for institutional investors.
At the trading posts, pros, known as registered traders (RTs), were in control. Each stock had at least one RT, designated by the exchange, who managed how orders were filled and tried to maintain an orderly, two-sided market. These pros not only bought and sold stocks on their firm's account and attempted to earn a profit, but they also stepped in to ensure orders could be filled. The RTs had all the info—who was buying, who was selling—and a pretty good idea of where the price was about to go.
By 1993, a technological revolution was brewing. Large money managers started to enter their orders directly into the exchange's computers, bypassing the floor traders. This did not sit well with the major brokerage firms, which made millions in commissions as the middlemen for institutional clients. "I think the domestic market didn't take that element—electronic trading—all that seriously," says industry veteran Nick Thadaney, a former CEO of ITG Canada and the TMX Group's global equity capital markets division. "The theory was always that Canada is different—it's not going to happen here."
It did. In 1997, the TSE closed the floor for good, moving trading from the raucous posts to the silent glow of thousands of screens. The job, once primal, had been confined to a desk.
In many ways, the electronic trading revolution was similar to the AI revolution that's just ramping up now. Some in the industry buried their heads in the sand and others warily prepared for it, while still others took advantage of the disruption to play the system and make some easy cash.
Those opportunistic others included a new breed of super-fast electronic trading platforms called high-frequency traders, or HFTs. They were the product of both the new technology and a decision by regulators to allow purely online trading venues to compete with established exchanges, such as the Toronto Stock Exchange (which soon demutualized, went public and rebranded itself as the TSX).
Founded by engineers and other computer whizzes, the HFTs made money by digesting data and trading faster than anyone else. Thanks to the market digitization, the spreads were getting thinner, but by quickly executing large numbers of small trades, the HFTs racked up big profits. "The answer was to automate," says Alex Bevziouk, a Ukrainian-born robotics specialist who was chief technology officer at Infinium Capital, a Toronto-based HFT firm founded in 2003. Algorithms he and competitors developed could fire in 500 small orders a second (today, 20,000 is easy) and, Bevziouk says, in the early days, even simple trading strategies used on the floor—such as arbitraging price differences between Toronto and New York—could be lucrative.
Speed was crucial. Almost all major North American exchanges located their computer servers in data centres in suburbs—New Jersey for the NYSE, Nasdaq and other major U.S. exchanges; and Markham, north of Toronto, for the TSX. Investment dealers and trading firms started renting servers just a few feet away, cutting their transmission time for orders to less than a millisecond.
By 2009, Infinium had 70-odd quants and traders working in its Toronto offices, and it generated more orders for some large TSX companies than the bank-owned dealers did. Slowly, however, the Big Five banks and other traditional investment dealers began to adopt strategies for coping with the tidal waves of high-frequency trading and speculation. RBC, in particular, made a name for itself when it introduced an online order routing system called Thor in 2010. The project was spearheaded by a team led by Brad Katsuyama, a whip-smart young finance graduate from Markham, who had been appointed head of RBC's U.S. trading operations in New York (he later became head of global electronic sales and trading). Thanks to a bestselling book by Michael Lewis called Flash Boys, Katsuyama has become one of the most famous stock traders ever.
Lewis's books are so appealing because of the writer's penchant for underdog heroes, and Katsuyama is a gem. The 39-year-old still exudes a geeky, down-to-Earth Canadian enthusiasm. He and his wife, Ashley, and their three young children now live on Connecticut's tony Gold Coast, but he takes a train and New York's rough-and-ready subway to and from work. "It gives me about two hours to work by myself every day," he says.
As Lewis chronicles in his book, Katsuyama originally developed Thor largely in response to a growing threat from HFTs. At the time, like many big brokers, RBC was struggling to fill orders from deep-pocketed institutional clients to buy or sell large blocks of stock. If RBC got an order to, say, purchase 100,000 shares, the bank would slice it into portions as small as a few hundred shares to disguise its eagerness to buy. The bank would then route the orders to several venues at once.
But many of the offers to sell posted on those venues were teasers placed by HFTs. By the time RBC's orders reached most of the exchanges, the quoted prices had disappeared. Super-fast HFTs, detecting the big orders, jumped ahead of RBC and bought, bidding up the market price. Thor ensured RBC's orders reached all the venues at the same time, thwarting the HFTs' early-warning system. The platform won RBC kudos on Wall Street, but Katsuyama felt much more had to be done to foil speculators and make markets fairer. So in 2012, he left the bank—amicably—to launch the Investors Exchange (IEX).
He and his staff of about 80 are now housed in a 78-storey glass skyscraper near one of the two black square waterfall pools at the 9/11 memorial. Usually, Katsuyama sits in the middle of a row of terminals. He also has a small windowed office on the side, which doubles as a meeting room. It contains a couple of shelves of personal memorabilia, including a photo of him with Stephen Harper, a photo with Justin Trudeau, and the wall-mounted silver bell IEX rang on its first day of trading in October 2013.
Wearing jeans and a Patagonia vest, and constantly folding and refolding a small piece of yellow paper into an airplane as he talks, Katsuyama quickly gets wound up. Roughly two-thirds of mom-and-pop investors participate in the market through mutual funds and pension funds, he says. But those institutions continue to struggle to buy and sell stocks at fair prices because markets still cater to HFTs, he argues. "Winners and losers in the market should be dictated by their own investment strategy," Katsuyama says. Instead, the moms and pops are, in effect, getting "baited and switched."
To trip up the high-speed crowd, IEX installed a so-called speed bump of 350 microseconds on orders. The bump is actually physical—a 61-kilometre-long coil of fibre optic cable into IEX's trading server in New Jersey. A more recent IEX innovation is the so-called "crumbling quote indicator," which predicts when a bid or ask price may be about to change. Fairness is proving to be a slow sell, alas. The exchange's market share of total U.S. trading volume is still tiny—climbing from about 0.1% in early 2014 to around 2.5% lately. Katsuyama says some innovations, like the crumbling quote indicator, initially reduce volume on the IEX and make his job harder. "We then have to provide the data to show that quality has gone up," he says.
The problem is, the rest of the trading world has also adapted since the early 2000s. Other trading systems have installed speed bumps too, and most banks have either adopted the strategies used by HFTs or figured out how to thwart them (or both). As the banks got better and faster, competitive pressures mounted on upstart HFT firms. By 2010, Bevziouk says, "the easy money was gone," and Infinium soon faded and disappeared.
High-frequency traders still account for roughly half of all U.S. stock trading, according to estimates by New York–based research firm Tabb Group LLC. But it's a much tougher business than it used to be. Tabb estimates that HFT industry revenues sank below $1 billion (U.S.) in 2017 for the first time since the financial crisis, down from $7.2 billion (U.S.) in 2009, and many large firms have merged to reduce costs.
In many ways, electronic traders and their super-fast technology are victims of their own success. The arms race in both hardware and software has been so intense, no one really possesses a decisive advantage anymore. That doesn't mean the speculators and HFTs didn't change anything, though—quite the opposite. Trading is now faster and more liquid, and much cheaper than ever before. What used to be a relatively low-volume business with big commissions is now a high-volume business with razor-thin margins. And it's one that needs fewer and fewer humans.
"Oh, fuck, it's game time," says Anthony George from behind his six-screen desk in Toronto. "Okay, kids, let's go."
It's less than a minute before the Canadian stock market opens at 9:30 on a Thursday morning in January. George is a managing director with AltaCorp Capital Inc., a boutique Calgary investment bank allied with provincially owned ATB Financial, where he's a registered trader assigned to two-dozen stocks by the TSX. One on his roster is sure to get massive action this morning: Canopy Growth Corp.
Canopy is Canada's largest legal pot producer, and volatility in the sector has been wild over the past couple of weeks. Today's opening will be huge because the night before, Canopy announced a $175-million bought deal with BMO Capital Markets to issue new shares at $34.60 apiece. (By day's end, Canopy will have raised about $200 million.) It's the first time a dealer owned by a Big Six bank has deigned to co-lead a stock sale for a marijuana company.
At age 53, George is a rumpled, laid-back 30-year trading veteran, his voice rarely rising above a monotone. He's wearing jeans and a Roots sweater, with his grey hair tied back in a short ponytail. He has a tin of chewing tobacco on the left side of his desk and a Tim Hortons coffee on the right. He points to two small boxes on his bottom middle screen—one will show current bid and offer prices for Canopy in the market, and on the other, a list of filled orders will scroll by.
At 9:30, the market opens and the boxes indeed go haywire. The prices flash and pop, and the filled orders race by. By 9:33, volume in Canopy hits one million shares. But George is sitting back in his chair and not hitting his mouse. Trading bots have taken over the market.
Compared with the old TSE trading floor, where George worked when he started with Midland Walwyn in 1986, the speed and quantities are astonishing. But investment dealers have tightened their purse strings, and there's not much risk tolerance left for traders who deploy their firms' capital. George says individual traders at AltaCorp can only throw 20% of the capital they are assigned into a single stock, and if they get in a $50,000 hole, that position is wound down.
By 9:42, bid-offer spreads on Canopy have widened to 20 cents—$36.40 to $36.60—which would have been normal in the 1980s but is a large chasm today. "This is absurd," says George. "Honestly, it should be $36.42 to $36.43." But there's often chaos on the frontier where humans and machines meet, and George has to be careful. He's trying to play a mix of defence and offence—not only watching for prices to settle and spreads to narrow, but also trying to anticipate big moves.
The trouble is that the algorithms, and the privacy that technology affords them, can greatly exaggerate both those trends in an instant. In the old floor-trading days, RTs could literally survey the floor to see who was playing, but now that trading has gone virtual, you don't really know who's out there. "I can't see where the price is gonna be," George complains. "I can't see, visually, where potential buyers and sellers are."
After the opening fireworks on Canopy, the bid-offer spread narrows to two cents shortly after 10:00. And the stock ends up trading just 6.5 million shares for the day—less than half the total on some active days recently.
Today's trading technology "is what everyone wanted 10 years ago," George says, with a hint of resignation, but it hasn't been so great for guys like him. He doesn't think it's been that great for investors, either. "When you're an RT on a stock, you feel like it is part of your family. It's like your kid. You're the pro, and you want the thing to trade well." The bots will put up 100-share orders here and there to test the market and try to nudge prices, he says, but they are "never going to offer liquidity the same way a person would."
He knows the machines are here to stay, though, and with the advent of artificial intelligence systems, there will be less and less need for the higher-level trading functions provided by registered traders like him. In the '80s, we were like celebrities, George says. But he thinks most of the RTs who remain will be replaced by machines in two years, tops.
In comparison to the old TSE trading floor, there isn't much horseplay along the back row outside Greg Mills's office at RBC. The quants like to have fun, but in their own way. They show off a small collection of bottles of hot sauce on the windowsill. There's a doctored photo of Shary Mudassir, RBC's director of global algorithmic trading, standing at a microphone in front of Katy Perry. In brief written biographies, team members list hobbies such as playing violin or piano, and taking backpacking holidays.
They are tackling a complex project. In pushing further into AI, Mills wants to corral orders from retail and institutional clients across all the divisions of the bank—plus the trades on its own account—onto one platform. Ideally, RBC wants to develop a system that allows it to match as many of the buy or sell orders placed through the bank as possible.
The initiative is called Project Unity, and if it works, it will transform trading on Bay Street yet again. The benefit, RBC says, is that its customers get better pricing, the bank's trading costs are lower, and there's no need to worry about competing traders interfering. It has caused quite a stir in the industry. The idea is to match more client orders in the marketplace through a unified trading platform and optimized routing practices. If an RBC client has been selling shares of Telus, for example, and another RBC client wants to buy shares of Telus, the odds of these two clients meeting in the marketplace are higher under the new routing logic. Some investors and rival dealers complain that, as a result, they don't have a fair crack at accessing those orders, and it's harder for them to see what the real supply and demand is. Last year, Canadian regulators said they were taking a closer look at the sudden rise in same-broker stock trading.
Project Unity is the result of years of in-house development at Canada's largest bank. RBC began building and deploying trading algorithms in the early 2000s. But algorithms are static, and as trading got faster and more competitive in the 2010s, the bank wanted to develop AI systems so the software would adapt to changes in the market and alter itself. Thirty-nine-year-old Mudassir, who joined the bank in 2009, travelled to Silicon Valley in 2015 to try to buy some AI solutions. The trouble was, "I couldn't explain trading to them," he says.
So RBC is building its own AI. Mudassir and one intern started in 2015, and since then the team has grown to almost 20. They will move into their own office beside RBC's trading floor in May, and it will be more like Silicon Valley than Bay Street—all modern couches and whiteboards. Mudassir says there has been a complete flip in new trading hires at RBC. When he started, about nine out of 10 were finance specialists—MBAs and CFAs. Now, 90% have a background in STEM: science, technology, engineering and math.
There are still some imposing technological challenges to tackle before trading can be completely automated. Team members say developing an algorithm that can trade like a human isn't as simple as taking a formula from a journal article and converting it to code. "We're not solving for x," says Boston Walker, a 23-year-old University of Toronto engineering science student whose spare-time activities include snowboarding and playing classical piano. Often the best test is to deploy $100,000 or even $1 million in the market. "There's no substitute for real trading," he says.
Walker says he still learns a lot from watching the bank's veteran traders. "How do you automate that gut feeling?" he asks. Algorithms and AI are good at recognizing and responding to historical patterns, but news events such as big mergers, or bouts of plain old panic or euphoria, often knock markets for a loop. Sometimes Walker takes a break and just watches the traders. He wants to understand how they think, how they react to different kinds of markets, the strategies they use. Then he turns back to his computer to continue writing the algorithms that will, one day soon, replace them.