The thing about useful stuff is

This New York Times story reports on the increasing interest by capital markets firms in utilising neural networks to detect previously undetectable patterns in data such as news feeds, from which trading strategies may be constructed and implemented.

Neural networks permit computers to create new rules and automaticallychange underlying assumptions by experimenting with thousands of random sequences and processes.

“Five years ago it would have taken $500,000 and 12 people to do what
today takes a few computers and co-workers,” said Louis Morgan,
managing director of HG Trading, a three-person hedge fund in
Wisconsin. “I’m executing 1,500 to 2,000 trades a day and monitoring
1,500 pairs of stocks. My software can automatically execute a trade
within 20 milliseconds — five times faster than it would take for my
finger to hit the buy button.”

........

The downside with these systems is their black box-ness,” Mr. Williams said. “Traders have intuitive senses of how the world works. But withthese systems you pour in a bunch of numbers, and something comes out the other end, and it’s not always intuitive or clear why the black box latched onto certain data or relationships.”

........

“There are some pretty substantial misconceptions about what these things can and cannot do,” he said. “As with any black box, if you don’t know why it works, you won’t realize when it’s stopped working. Even a broken watch is right twice a day.”

Several key things are pertinent to this article

• The desire by firms to see "order" in "chaos" and thereby extract relationships from which they can identify trading opportunities that others may not see


• The importance of speed - its no go being right about an opportunity if you are not fast enough to implement it before its gone; such opportunities may last less than a second!


• The dangers of blindly adopting such technologies without having a clear understanding of how they work and how they draw "conclusions"

The first of these relies upon the past being a guide to the future. Human behaviour is shaped by our past experiences - putting your hand in a fire will hurt! Yet we also appreciate that environments change whilst cause and effect can be incorrectly inferred e.g. markets go up on sunny days. Firstly we have to find such relationships which the computers can do - harder is for them to demonstrate causality. Yet on top of that, we have to assume that others infer the same relationships and will react in the same way.

Many people that work outside of capital markets fail to appreciate that investing is actually about speculating on the reactions of others and assessing to what extent those expectations are already factored into the current price.

A recent example was the legislation in the US that outlawed internet gambling. The share price of many leading UK online gambling companies was devasted by that change in the law. It seemed reasonable to assume that if the legislation passed, everyone would believe that these businesses would be materially affected and put a lower value on them. So, "the profitable bet" was that the legislation passed and so the price would materially drop.

Had the introduction of this legislation already been factored into the price, there would have been no significant drop and hence no opportunity to profit. That it hadn't, was evidence in the price drop. Sometimes, when a profits warning is issued, the share price goes up and the layman is baffled as it seems contrary to common sense. The reality is that the profits warning had been factored into the price already and the upturn may be because the warning is not as bad as had been expected or simply other things are influencing the price on the day i.e. new positive information.

To continue the example, it was easy to see a connection between the two things; legislation and impact on the gambling companies. Yet some of the relationships that the neural networks uncover will be "new discoveries" which the wider market is ignorant of. Had investors not linked the legislation and gambling companies share price, the price would not have moved in immediately, only feeding through in subsequent periods as profits diminished - a longer term bet.

Where a relationship/link is widely acknowledged, it then becomes a speed game of who can react fastest to changes - e.g. have you noticed that when there is a supermarket tannoy announcement that bread has been reduced then what you had perceived to be infirm little old ladies, become superhuman athletes with sharp elbows that beat you to the shelf!

This speed advantage may be in the detection of the change or in being able to execute your order fastest before the price moves. Firms invest staggering amounts in both - the NYT article is focussed on the former, but having fastest comms is just as vital and this is operating at the level of miliseconds. I was chatting with a friend who is a trader in the week about how they've just spent £300k to have their comms moved closer to the Stock Exchange machines - an extra 10 feet inside a machine room. Why bother for 10 feet? Because the laws of physics means that 10 feet closer gives them a speed improvement that they think they can translate into profits!

However, its the last of these three things that is scaring most firms and regulators. Such is the complexity of the "computers", that very few people have any idea how they work. Moreover, if the "machines" stop working, would anyone realise.

The worry is that Firms are commiting considerable capital to backing the "ideas" of the machines and if the answers are "wrong", not only could those firms could bust, this could also affect the stability of the markets.

As an aside, if you think about it, every day we put life/death reliance on things that we don't understand how they work e.g. getting into a lift to travel down from the 20th floor [it could of course accelerate to high speed and smash into the bottom!].

About Me

Name: John Wilson

Location: London, United Kingdom

View my complete profile