Tweet
'Humility is essential for programmers like me. You must always assume something is your own fault'
We meet at the Royal Exchange Grand Cafe just off Bank, at the heart of the City of London. It's around 6pm, there's a light drizzle and outside on the street the heels of business people's shoes go tick-tack on the pavement, like tap dancers, each moving to their own metronome. The computer programmer is in his mid-30s and dressed in jeans and a shirt. As we are sitting down he says he doesn't like the area around Bank: "There's no greenery here, if you need to go out for a walk to think." He orders a cola.
"I am an IT guy with a strong background in maths. I have co-written the program for our high-frequency trading systems – 'the engine', as we call it. I come into the office around 7am, well before the stock market opens. My team and I need to check and double-check all our systems prior to trading. With so many trades taking place each day, it's important to ensure everything is accounted for correctly before and after trading. During the day our team monitors the engine as it buys and sells, thousands of times. Lunch is 10 minutes, the time required to run across the street, grab a sandwich and get back.
"Compare the movements of shares on the stock market with waves. Our company is like a surfer trying to spot a wave, ride it for a tiny moment and then get out again before it breaks. On any given day, our computers buy and sell shares tens of thousands of times, holding them for very short periods of time, sometimes even less than a minute. No human being, or collection of human beings, could do the volume of trades computers are doing at the moment at stock exchanges across the world. This is not stuff that was once done by humans and now taken over by computers. This is something new altogether.
"We constantly check for bugs, disruptions or indications of incorrect activity. Even if the engine misbehaved for just a second, the number of trades it could do in that time is enormous. Thus it is important to monitor it as closely as possible. The mark of a good program is not just how it performs during normal operations, but also how it reacts to unexpected events. It is important to ensure that there are several layers of fail-safes built into the engine itself.
"Another natural safety valve is your collateral. When you buy stock on the exchange you need to have capital, obviously, to pay with. So suppose our computer goes nuts and embarks on a buying spree, it will automatically stop when our collateral runs out.
"The day gets slightly more tense as we draw nearer to 4.30pm, when the market closes. Once closed, any mistakes can't be corrected until the following trading day, so could end up being expensive.
"Real stress occurs when the machine does something unusual and you can't figure out whether it's an internal bug or something in the market. Humility is essential for programmers like me. You must always assume something is your own fault. If you are arrogant and you tend to blame a bug on the outside world, it is likely you'll miss a bug of your own.
"Success in the business of high-frequency trading? One way is brute force. The faster your computer, the faster your program can act. We are talking about milliseconds here, so even the speed of light is significant. The other driver is the quality of the program itself. Part of my work is constantly trying to improve it, make it work faster, more efficiently.
"Improvements in the program's logic can see the biggest gains. Even if you use a bigger computer and speed up the code by 100%, it will never be as fast as if you didn't have to run that bit of code in the first place. This type of solution is the most satisfying.
"And then there's the computer program itself, the black box that decides what and when to buy and sell. The key term here is correlations. We take all data from a subset of shares at the London Stock Exchange, the shares we have decided to trade in. This dataset is about 3GB, and it consists of each and every movement of the share that day. We call these movements 'ticks'. We analyse the patterns of these ticks for our subset of shares and seek out correlations. For example, when Vodafone goes up, Deutsche Telekom probably will too, because they are in the same industry. That is a very crude correlation. Our model has hundreds of variables, and every day we look for new patterns.
"It is highly complicated stuff and it has nothing to do with analysing a particular company's value or strength. It is about historical patterns of trading which we project on to the future.
"There are peculiar sides to this. One is a so-called black swan event, something entirely new, unforeseen and unprecedented that you couldn't have modelled for it – it's new so it could not logically have shown up in historical patterns.
"Another peculiar aspect is that there are more companies like ours, so when we analyse market movements we need to take into account the activities of those other companies and include these in our model. Meanwhile the other companies are doing the same so you get into a 'they know that we know that they know' situation.
"I have always loved maths, the precision and beauty of it. An answer is either right or wrong. So it's really very ironic for me to have ended up in this one area of maths that is all about correlations and approximations. No exactness here. But I don't care. I expect to make enough money to be out of this business in a few years. I think I would like to go back to university. I have become very interested in the humanities and philosophy.
"I just feel incredibly lucky to be living now. What would I have been doing with my maths skills 100 years ago? Or 100 years from now? This is exactly the right time in history to have these skills. And I have them."
Joris Luyendijk
guardian.co.uk © 2011 Guardian News and Media Limited or its affiliated companies. All rights reserved. | Use of this content is subject to our Terms & Conditions | More Feeds
More...
