Before the recent Wall Street crisis took hold, my knowledge of the stock market came from Trading Places. Dan Akroyd and Eddie Murphy stick it to the man by gaming frozen concentrated orange juice futures using pit traders on the stock exchange floor.
My idea of short selling was taking my roommate’s last can of chili because I knew that he would later drink half my beer without asking.
The days of traders selling stock by “open outcry” on the market floor are fading fast. In the early 1970s, the Depository Trust Company was created to handle post-transaction functions, including money transfers. This computerized approach to transaction recording and money transfers paved the way for our modern system.
So what? Well, last year, what is now the Depository Trust & Clearing Corporation handled $1.8 quadrillion worth of transactions. That’s $1,800,000,000,000,000. That’s 14 friggin’ zeros.
The advent of computerized securities and exchange gave rise to automated trading, usually breaking up and distributing one big buy over many smaller purchases to hide activity from competitors.
By the 1980s, software was set up to initiate trades as soon as a certain profit was reached, It was also used as “portfolio insurance” to protect against assets whose value gets too low. Many blame the crash of ’87 on automated systems of the time.
But stocks were still being traded on the floor, with electronic orders generating paper tickets that poured into the pit on conveyer belts. Fully electronic trading appeared in the early ’90s, and the old belts eventually sputtered to stop. Trades were executed in seconds, and as the efficiency and scale of trading increased, the costs went down.
The speed advantage of making trades before others could was very lucrative, so massive amounts of resources went into developing new, more complex computing systems.
A whole new industry based around algorithmic trading blossomed.
At their simplest, algorithmic trading systems (or Algos) take a set of variables (like stock price over time, etc.) and produce a recommended action. Many Algos are set up to automatically take an action, like buying or selling stock.
After the first algorithmic trading systems appeared, other programmers came up with algorithm-busting algorithms. They reverse-engineered the patterns shown by the trades of other algorithms, and came up with a pre-emptive way of gaming their advantage.
It’s programs hacking programs. Clever, eh?
But there is a danger with so much automated trading going on: the manipulation of the information that these programs use.
The market gets affected by a lot more than lists of numbers. Natural disasters, politics, legal issues, or even a psychotic stockbroker who caught his girlfriend banging the DJ last night, all play a part.
What do you call that stuff? The news.
Last year, Thomson Reuters announced it had come up with a system that analyzes news stories for positive or negative tone. The result is an XML-like data feed that not only summarizes news, but includes a numerical value of whether it was positive or negative, and to what degree.
So you get something like an RSS feed that gives a headline, and tells you whether it was good or bad for the company. It actually parses the language and quantifies the emotive content of words like “gloomy.”
OK, so a numerical feed of the emotional state of the world doesn’t sound like such a big deal. But who do you think likes numbers a lot?
Algorithmic trading systems.
There are systems in place that currently trade a stock automatically based on the news from these feeds. A program reads another program’s interpretation of the news and trades stock, without an actual person involved.
If that pushes the price down far enough, it’s feasible that another news story would be automatically generated and then added to a new data feed, which the other Algos pick up and act on, continuing downward.
Even though news-only trading is very uncommon right now (usually the data is included along with many other data sets that feed these algorithms), the implications are staggering.