Wednesday, November 17, 2010

Causation, the beginning of a journey

We have a seemingly insatiable urge to find causes for things. From “The market sold off because…” to “smoking causes cancer” to (and I kid you not) "Fear of hell makes us richer, Fed says." If x causes y, we seem to be in the comfortable world of rationality. There are reasons why y occurred; it was not some random or inexplicable event.

We distrust correlations because they often appear to lack rational, defensible foundations. Perhaps worse, they often masquerade as causal relations. James Stock, a Harvard economist, offered an example that is perhaps even more bizarre than the “fear of hell” study. “He noted that U.S. national income has been growing significantly for at least the last 100 years, and at the same time Mars has been slowly but steadily getting closer to the Earth. Because of these two long-term trends, it is virtually guaranteed that a simple statistical correlation would support the hypothesis that U.S. national income is determined by the country’s proximity to Mars.” (Thomas Karier, Intellectual Capital, p. 269)

And yet some correlations intuitively seem more causally linked than others. Let’s assume that we were presented with two equity index trading systems, one based on the relative strength or weakness of the U.S. dollar and the other on the motion of the planets. Let’s assume further that back tested over ten years these systems had identical profiles. Which system would most people be likely to embrace? The former, I assume, since we can more or less explain the sometimes simple, at other times complex relationship between equities and currencies whereas most people would be hard pressed to make any sense of the relationship between equities and the planets.

There are two main paths we can follow in trying to sort all this out. One is to embrace a kind (and possibly kinds) of causality that is either weaker than or different from “mainstream” causality. The second is to redefine the kinds of effects we are expecting, from invariable or regular to probabilistic (and “probabilistic” covers a wide range of possibilities). Ideally, these exploratory paths will eventually converge.

I’ve decided to begin a series of posts, erratically spaced, to inquire into these issues. Quite frankly, I don’t know where we’ll end up. Perhaps back where we began. Perhaps with such a watered-down version of causality that it’s virtually indistinguishable from correlation. But with any luck the journey will be educational. Maybe it will even inspire some ideas for system development and testing.


  1. Excellent. I enjoy your posts and look forward to this series.

  2. Yes, excellent. I'll be referring back to this post and thinking about kinds of causality vs. expected effects, especially probabilistic effects.