Part 1 is here.
As they left the pub after a few too many pints, Tom’s mind was already racing with everything he needed to do.
For the rest of the week he started setting up a home-office in his garage. Dan had agreed to turn up on Saturday morning to discuss the finer points of the code architecture.
The idea was beautifully simple.
All he needed to do was come up with a way of processing information that was already available and feed that into their own program which would analyze the input and ultimately produce a prediction of the world’s financial markets.
Tom had based the idea on genetic algorithms. Back in COMP203A he learned how GAs are a collection of algorithms that form a program which uses the same model as evolutionary biology. They utilize concepts such as inheritance and mutation, and evolve following a survival of the fittest model, where the strongest algorithms survive and reproduce.
In his case he would need to model them such that the strongest algorithms are those which most closely predict the stock market.
As he explained to Dan, genetic algorithms, like biological individuals, comprise a set of ‘chromosomes’, but each chromosome is essentially an input variable ultimately influencing the algorithm’s output. In their case, the chromosomes would represent factors likely to affect the financial markets, such as inflation, tax-rates, major corporate shareholdings, or even more abstract events like the weather.
Tom knew that GAs were hot technology, all IT magazines du jour were salivating over their tremendous potential in modeling dynamic systems. They were even hailed as the tool that could find a cure for cancer.
The problem of course is, as Tom soon found out, they are limited by the processing capability available.
Writing the framework proved to be relatively simple, and it had taken Dan less than a month to produce a working prototype.
However, feeding the pool of genetic algorithms enough data that was both current and relevant, took considerable more effort.
“It’s impossible”, Dan stated exasperatedly. It had been four months, and Dan was starting to get grumpy. Weekend after weekend he cycled over to Tom’s garage, for beer, pizza, and marathon bouts of coding. But no matter how much he optimised his program, he couldn’t sustain the population of GAs for more than 100 generations. He had tired of Tom’s project, and just wanted a solid weekend of lounging and doing nothing.
“All I’m getting is a bunch of random predictions. I get a soup of chaos and then they all die off”.
“We need more input”, Tom suggested. It had become his generic answer to everything.
“Tom, we already have over 100 inputs. I’ve got references to NZ weather, all of the corporate shareholdings, political approval ratings, sporting events, major natural disasters, and everything else.”
“Maybe it’s not being parsed properly?”.
“The problem isn’t with my code, Tom”, Dan grumbled. “The parsing is fine. You’re right we need more input. But it will take a computer the size of Google’s data bank to process it all”.
“Hm, that sounds expensive. I wonder if I can get a loan?”, Tom pondered.
“Who knows, but I’m done. I’m going home to watch an episode of The IT Crowd. Call me when you find a better computer”.
Tom knew Dan was right, he needed faster hardware.
Of course the banks were incredulous towards his business model. They demanded to see proof-of-income, business plans, missions statements and all sorts of other bureaucratic material that seemed familiar, but which he had never paid attention to in 101 economics.
He was at a loss. Tom realized if he wanted to see this through, he could either enlist his ex-girlfriend, Mandy, to help him get a loan. She was a shrewd penny-pinching accountant, who took full custody of their pet dog, Bunty, when they split up. Neither Dan, nor any other of Tom’s friends liked her, but she did know business.
Or, he could design his own hardware.
Designing his own suddenly seemed preferable. He had already laid the groundwork from his post-grad research.
With a bit of help from his ties at Auckland University, he began fashioning a revolutionary microcontroller specifically designed for processing GAs.
For the next several months he spent all his free time in his garage feverishly working on the chip.
It wasn’t until he received a call from Dan, that he realized he now had the hardware that Dan had been asking for.
“Tom, You haven’t been at Galbraiths for months. You’re not still trying to predict the stock market are you?”
“Hey Dan, you need to come over. I’ve been working on a new type of microcontroller. It uses field programmable gate arrays. It’s going to fix your program”.
“There’s nothing wrong with my program”, Dan stated defensively.
“Sorry, I mean you’ll actually be able to use it with more input. I think this time it’s really going to work”.
“Yeah?”, Dan sounded doubtful. But Tom’s contagious enthusiasm soon convinced him to give it another go.
The microcontroller lived up to both their expectations. It finally allowed Dan to feed the algorithms input from over 1000 separate information sources.
Their stock-market predictions, though still not very accurate, started becoming more consistent.
The micro-controller was Tom’s first break-through.
He reasoned, if anything, it alone could finance a very nice party in Fiji, with free pina coladas for anyone wearing a bikini, or nylons. But he was confident it would only take a few more months before they’d hit the jackpot. Then they’d make millions, buy mahogany power boats, and impress girls with their flashy phones.
Or such was the plan.
Chapter 2 is here.