Mar

18

The Chair advises us to use simple statistics for market analysis. They are good enough for sure. But I wonder every once in a while: why eschew the sophisticated stuff? It must have been developed for a reason, mustn't it?

Or must it?

Little by little, I am starting to see the light in keeping it simple. Today yet again, I got a glimpse of it while reading Unit Roots, Cointegration, and Structural Change by  G. S. Maddala and In-Moo Kim.

This is a book about complicated and modern stuff that is not being used in the tests posted on DailySpec. I am only half through, but I think I can already comment on it. This is an excellent book. It is very clearly written. Usually, when reading this type of book, I am left with the impression that the author is confused. Not here. Maddala and Kim are clear-thinkers who obviously understand their subject matter, to the point that they are able to write about it in an articulate and insightful way.

They understand it so much that they can distance themselves, warning that these tools are mostly ineffective — not to say bull. What I really appreciated upon reading their conclusions about unit root tests (which is just another word for random walk tests), is that they state without any passing thought for political correctness, that the ones everybody uses, like Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP), should not be used, because of dismal power in small samples. This is based on Monte Carlo simulations from other researchers, whose results they provide in the book. These results are ominous and devastating for these oft-run tests. But they don't stop there. They add that they spent two chapters on answering the question: "Which Unit Root Test?". Whereas a better question should have been "Why Unit Root Tests?". To which the answer would frequently be: "They is no reason to perform unit root tests, they are useless for most purposes". I am exaggerating a bit, their statements are mellower, but that's the gist of it. And I like it a lot. A book about unit root tests saying that they are frequently useless (frequently, but not always, to be fair).

In a similar vein, there is a part on panel data unit roots where they mention that a Fisher meta-analysis test from the 1930s, is as good or better than some clever and modern stuff from the 1990s.

So let's Keep It Simple and Stupid!

For 90% of our needs, a grassroots OLS is just as good and more robust than all this rocket-science 20th Century mumbo.

A final word: this book also strikes a perfect balance in maths. They go deeper than usual textbooks, inasmuch as they don't only provide the formulas for the tests, but also the mathematical intuition behind them. But they don't do the full demonstrations, for which they provide an ample bibliography. This is still graduate to upper graduate maths and econometrics so please don't buy this book if you are looking for an introductory text.

Alan Millhone adds:

The game of Checkers when played scientifically follows the KISS method. Checkers is a game of utility. If you have a win on the board you execute that win in as few moves as possible. I remember a time when a novice (at times I think I am still a novice!) and my opponent had a King in each opposing double corner and I had three Kings and did not know how to consumate the win. With the help of a beginner's Checker book that win on the board is now 'old hat'. Knowledge is power in about anything.

Chair admonishes the Market trader maintain a hand written manuscript. I have a Checker manuscript and record my games for later reference. Any 'tool' that makes you more effective is valuable. 


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search