The interview

 Well, the NSA didn't exactly come calling. That's the National Security Agency. It didn't exactly come calling. They had an operation in Princeton where they hired mathematicians to attack secret codes and stuff like that. And I knew that existed. And it had a very good policy because you could do half your time at your own mathematics and at least half your time working on their stuff. And they paid a lot. So that was an irresistible pull. So I went there. 

 Did you work as a code cracker before being fired from there?

 Well, I did get fired, yes. Well, how come I got fired? Because, well, the Vietnam War was on, and the boss of bosses in my organization was a big fan of the war and wrote a New York Times article, magazine section, cover story about how it was going to win in Vietnam. And I didn't like that war. I thought it was stupid. And I wrote a letter to the Times, which they published, saying not everyone who works for—it was Maxwell Taylor, if anyone remembers that name—works for General Taylor, agrees with his views. And I gave my own views, which were different from General Taylor's. But on the other hand, nobody said anything. But then I was 29 years old at this time, and some kid came around and said he was a stringer from Newsweek magazine and he wanted to interview me and ask about my views. And I told him, I said, 'I'm doing mostly mathematics now, and when the war is over, then I'll do mostly their stuff.' Then I did the only intelligent thing I'd done that day. I told my local boss that I gave that interview. He said, 'What did you say?' And I told him what I said. And then he said, 'I got to call Taylor.' He called Taylor. That took 10 minutes. I was fired five minutes after that. But it wasn't bad. 

 How did you get into trading?

 I was in my late 30s. I had a little money. I started trading, and it went very well. I made quite a lot of money. It was pure luck. I mean, I think it was pure luck. It certainly wasn't mathematical modeling. But in looking at the data, after a while, I realized, 'Hey, it looks like there's some structure here.' And I hired a few mathematicians, and we started trying to make some models. 

 Did you find any worthwhile patterns?

 Maybe a slight upward trend over that whole period of time. How on earth could you trade, look at that, and see something that wasn't just random? Well, it turns out, in the old days, and this is kind of a graph from the old days, commodities or currencies had a tendency to trend. Not necessarily the very light trend you see here, but trending in periods. And if you decided, 'Okay, I'm going to predict today by the average move in the past 20 days,' it is 20 days, maybe that would be a good prediction, and I'd make some money. And, in fact, years ago, such a system would work. Not beautifully, but it would work. So you'd make money. You'd lose money. You'd make money. But this is a year's worth of days. And you'd, you know, you'd make a little money during that period. But it's a very vestigial system. So you would test a bunch of different lengths of trends and time and see whether, for example, a 10-day trend or a 15-day trend was predictive of what happened next. Sure, sure. You would, you know, try all those things and see what worked best. But the trend following would have been great in the '60s, and it was sort of okay in the '70s. By the '80s, it wasn't such a wonderful thing because everyone could see that. 

 So how did you seperate yourself from the rest?

 Well, we stayed ahead of the pack by finding other approaches and shorter-term approaches, to some extent. But the real thing was to gather a tremendous amount of data. And we had to get it by hand in the early days. We went down to the Federal Reserve and copied interest rate histories and stuff like that because it didn't exist on computers. We got a lot of data and very smart people. And that was the key. I didn't really know how to hire people to do fundamental trading. I had hired a few. Some made money, some didn't make money. I couldn't make a business out of that. But I did know how to hire scientists because I have some taste in that department. And so that's what we did. And gradually, these models got better and better and better and better. 

 What role did machine learning play?

 Well, in a certain sense, what we did was machine learning. You look at a lot of data, and you try to simulate different predictive schemes until you get better and better at it. 

 How much data do you use daily?

 I have to say, whether it's annual reports, monthly, quarterly reports, the historic data itself, volumes, you name it, whatever there is. We take in terabytes of data a day and store it away and massage it and get it ready for analysis. And you're looking for anomalies. You're looking for, like you said, the efficient market hypothesis. But any one anomaly might be just a random thing. So is the secret here to just look at multiple strange anomalies and see when they align? Well, any one anomaly might be a random thing. However, if you have enough data, you can tell that it's not. So you can see an anomaly that's persisted for a sufficiently long time. The probability of it being random is not high. But these things fade after a while. Anomalies can get washed out. 

 So the money rolls after this operation?