Yesterday we had up part one of a Q&A up with Jack D. Schwager author of Hedge Fund Market Wizards. Today we have an excerpt from the book as well. The excerpt below represents part of a conversation Jack had with Ed Thorp. (Schwager’s questions in bold and Thorpe’s answers follow). For those who aren’t aware of Thorp’s long career, it encompasses as Schwager writes “an extraordinary number of first achievements.” Although Thorpe is no longer actively managing money I found his thoughts on trend-following to be of great interest the Abnormal Returns’ readers.
Do you have thoughts on futures trend-following as a strategy?
I believe there are versions of it that have a Sharpe ratio of about 1.0 or more, but that is risky enough so that it is hard to stay in a business because you can get shaken out.
I take it then that you believe there are trends inherent in the markets.
Yes. Ten years ago, I wouldn’t have believed it. But a few years ago, I spent a fair amount of time looking at the strategy. My conclusion was that it works, but that it was risky enough so that it was hard to stay with it.
Did you ever use trend-following as a strategy?
When did you trade futures?
We began the research project in 2006, and launched the trading program in late 2007. It was promising, and we were thinking of bringing it up in large-scale with institutional money. But in early 2010 my wife was diagnosed with brain cancer, and my heart wasn’t in it. Life is too short. I didn’t want to launch another major activity, so we gradually wound the program down.
So the program had worked well while you’re using it.
Reasonably well. It wasn’t as compelling as the Princeton Newport strategies or statistical arbitrage, but it would have been a good product, and it seemed to be better than most of the other trend-following programs out there that were managing a lot of money.
What kind of Sharpe ratio was your program running?
It was a little better than 1.0 annualized.
Since you are no longer using the strategy, can you talk about what modifications you made to improve a plain-vanilla trend-following approach? Was there something about your program that made it different from other trend-following approaches and that might explain why it did somewhat better?
We combined technical and fundamental information.
What kind of fundamental information?
The fundamental factors we took into account varied with the market sector. In metal and agricultural markets, the spread structure—whether a market is in backwardation or contango—can be important, as can the amount of storage relative to storage capacity. In markets like currencies, however, those types of factors are irrelevant.
Would it be accurate to describe your approach as combining technical trend-following rules with market specific fundamental filters that define favorable and unfavorable environments?
Were there other enhancements you made to the standard trend-following approach?
We had some risk-reducing approaches built into the system. We tracked a correlation matrix that was used to reduce exposures in correlated markets. If two markets were highly correlated, and the technical system went long one and short the other, that was great. But if it wanted to go long both or short both, we would take a smaller position in each.
Since correlations between markets change so radically over time, even changing sign, how long of a lookback period did you use?
We found that 60 days was about best. If you use too short of a window, you get a lot of noise; if you use too long of window, you get a lot of old information that isn’t relevant.
What other risk-reducing strategies did you incorporate into the system?
We also had a risk management process that worked a bit like the old portfolio insurance strategy. If we lost 5 percent, we would shrink our positions. If we lost another few percent, we would shrink our positions more. The program would therefore gradually shut itself down, as we got deeper in the hole, and then it had to earn its way out. We would wait for a threshold point between a 5 percent and 10 percent drawdown before beginning to reduce our positions, and then we would incrementally reduce our position with each additional 1 percent drawdown.
At what drawdown point would your position be reduced to zero?
How far down did you get?
Our maximum drawdown was about 14 percent to 15 percent, by which point we were trading only about one-third of our normal base position size.
How would you get restarted if the drawdown reached 20 percent?
You wouldn’t. You have to decide ahead of time how much of a drawdown would imply that the system is not as good as you thought it was, and therefore shouldn’t be traded.
In hindsight, do you think reducing your exposure on drawdowns was a good idea?
It all depends on how confident you are about your edge. If you have a really strong conviction about your edge, then the best thing to do is sit there and take your lumps. If, however, you believe there is a reasonable chance that you might not have an edge, then you better have a safety mechanism that constrains your losses on drawdowns. My view on trend-following was that I could never be sure that I had an edge, so I wanted a safety mechanism. Whereas for a strategy like convertible arbitrage, I had a high degree of confidence as to the payoff probabilities, so reducing exposure on drawdowns was unnecessary.
In a strategy like trend-following where you couldn’t accurately assess the probabilities, as you could in a strategy such as convertible arbitrage, what percent of the Kelly criterion was your bet size?
We didn’t use the Kelly criterion at all in trend-following because the bet size was such a small fraction of Kelly that it didn’t make any difference. I would guess that we were probably using something equivalent to 1/10 or 1/20 of Kelly.