Ji bai zi

Recent Reading

A few of the books picked up in the last month or so:

Garry Wills, Henry Adams and the Making of America (a book without a subtitle!). Garry Wills is a fine writer, certainly in the same stylistic rank as Adams. The title seems at bit a first curious. After all, Henry Adams was born in 1838, and had little to do with the "Making of America" normally presented in elementary school pageants or multi-part PBS epics. This book next seems to be an appreciation of Adams' 2700-page (Library of America edition) history of the period from 1800 to 1817. It can be viewed also as a precis of the history of the United States from 1800 to 1817, a period that is usually represented, rather than explained, by a few episodes: the Louisiana Purchase, Lewis and Clarke, the Star Spangled Banner, the Battle of New Orleans; but a period that is much more easily understood, as presented in this book, in light of the Napoleonic wars. The book can be seen as an intellectual history of Adams, showing how his ideas about 1800-1817 were based on his experience during the Civil War; and it can also be seen as a factual corrective to Adams' elegant but impressionistic Education. But in retrospect, the book is really about Henry Adams and the making of "America": the development and subsequent interpretation of Adams' ideas; the development of "history" as a discipline; and (ultimately) the arguments that continue today about original intent and the political significance of history in the United States.

Nandan Nilekani, Imagining India: Ideas for the New Century (I read the Indian edition, rather than the as-yet unpublished American edition referenced – the subtitle for the American edition is "The Idea of a Renewed Nation"). Nilekani's book reads like an extended Thomas Friedman essay on the development and future of India. In some ways, it's reportage – many quotes from the elite of India, especially the IT-inspired group who hope to pull India out of the 19^th century into the 21^st. In other ways, it's advocacy. For someone who's not Indian, much of the discussion lacked context. The voice seems a bit odd; unlike Friedman, Nilekani is not a reporter, and the mixture of reportage ("When [Goplalswami] tells me of his technology-aided solutions to stop box-stuffing, I cannot miss the twinkle in his eye") and advocacy jars. The combination is not always appealing to this reader, and the book feels over-long; still, anyone who's a reader, who does business in India, should read this book. (The American edition is 528 pages, the Indian edition 532 plus front matter; this, plus the different subtitle leads one to believe that the book will be somewhat different in the new edition.)

Jeremy Grantham (more on standard deviation?)

Despite the suggestive title, this is a follow up to a post from the beginning of the week, in response to John Auther's January 29 column On Diversification from ft.com.

According to Grantham Mayo Otterloo, the MSCI World index of developed world stocks fell 49 per cent during the sell-off, the same as the cap-weighted version of the index.

The odds were that the main index would suffer a fall like this about once every 100 years; it would happen to the equal-weighted index once every 34,000 years [as it's a 4 sigma event].

In passing I commented:

It's hard to look much deeper, as the report doesn't seem to be public; if time permits, I'll come back with the results of some searching.)

I was led to an interview with Jeremy Grantham, chairman of Grantham, Mayo, van Otterloo Company.

Grantham uses standard deviation terminology when talking about the housing bubble. Here's an interesting quote, with emphasis added. (He's been asked what one question he would ask Alan Greenspan, and then this follow-up.)

What would you ask Ben Bernanke?

How did you possibly miss the housing bubble? How could you say at the peak of this three-Sigma event--a one in 1,000 year event at the top of 2006--how could you say the U.S. housing market merely reflects a strong U.S. economy? Surrounded as you are by all this statistical help, and with your experience with the Great Depression, how could you miss it? I simply don't get it.

We can conclude from this response (apart from the fact that Grantham shares 20/20 hindsight with the rest of us) that Grantham shares the use of sigma-notation as shorthand and a quantifier for big excursions; but also uses a model that somehow lets you translate sigma into time.

Many people have written about Value-at-Risk analysis, which lends itself to this kind of thinking. There's a decent Wikipedia article on the topic; and there's pithy and detailed criticism from Nassim Taleb (for example in this debate), with sophisticated discussions of heteroskedasticity (!) and covariance matrices. However, no matter what one thinks of "Value-at-Risk" theory, it's clear that pseudo-statistical theory has slopped over into discussion of pricing trends in general. Clearly this is a mistake. Calling the asset bubble of 2006 a "1,000-year event" is grossly misleading. Prices are not like the weather, because the market is not like the atmosphere. You can't apply insurance theory or extreme-event theory from natural systems to economics, because there is no stable underlying mechanical system – instead, the market is an outgrowth of the wealth, knowledge, emotions and experiences of the individual self-aware actors. A crash in the market, and a violent storm are similar in that they arise from (and are really indistinguishable from) the collective actions of much smaller entities; and they can be approximated without simulating the behavior of each individual component. But they differ in that the particles in a storm have no memory of their prior experiences, and no view of the future; in a given set of external conditions, they'll always act the same. In contrast the individuals who make up a market may very well, given the same external conditions, decide much differently, based on accumulated experience, views of the future, and express substantially different preferences.

John Authers on Diversification

I greatly enjoy John Authers' Short View on ft.com. However, his 1-29 column on Diversification cites some dubious analysis from Grantham Mayo Otterloo.

Diversification is one of those rare concepts that accords both with academic theory and common sense: "Don't put all your eggs in one basket."

But last year's sell-off was so extreme that diversification would not have helped one whit. ….

According to Grantham Mayo Otterloo, the MSCI World index of developed world stocks fell 49 per cent during the sell-off, the same as the cap-weighted version of the index.

The odds were that the main index would suffer a fall like this about once every 100 years; it would happen to the equal-weighted index once every 34,000 years."

(In the video report, the conclusion about the equal-weighted index is based on a calculation that this represents a "4 standard deviation event".)

This is a very suspicious claim. (It's hard to look much deeper, as the report doesn't seem to be public; if time permits, I'll come back with the results of some searching.)

Standard deviations were invented to characterize experimental and measurement error, where each measurement is an approximation to the "real value". They are calculated based on an assumption that small errors are more likely than large errors (and that you know nothing else about the errors).

The calculation procedure can be applied to any data series, but the underlying assumptions about what "standard deviation" means cannot.

In particularly, it's clearly wrong to apply it in this way to the stock moves after the bubble burst. Stock markets are not measuring some abstract value -- they are the value themselves. Furthermore, they are inherently correlated because they are set by human actors.

This would be a big (34,000 year) fluctuation if we thought that the stock prices were not correlated with each other. But they obviously are correlated with each other based on past, current, and forecasted economic performance. Very clearly, major exogenous events that would cause market havoc (wars, epidemics, running out of key resources) happen more often than every 34,000 years