Correlation coefficients are statistics that provide a measure of the relationship between two sets of data. Correlation coefficients can be calculated for two different asset classes, or two stock markets, sectors or stocks. They are widely used in studying the risk characteristics of a portfolio, in attempting to understand the behaviour of a portfolio or in setting asset allocation. Typically, a risk software package would make use of correlation coefficients amongst stocks or markets, in matrix form, to model portfolio behaviour. In some cases, historical correlation coefficients can be over-written with estimated coefficients, in an attempt to improve predictive capabilities. Knowledge of correlations can also be used in developing an investment process best suited to a client mandate: for example, in determining whether to use a top-down country approach to research, a global sector approach or a sectors-within-regions approach.

Correlation coefficients, however, are rarely constant. Generally lowly correlated markets can suddenly move in lock step for years at a time, or may fall together during a sharp market correction. Bonds and equities in many markets have traditionally been positively correlated, but this pattern does not always hold. For example, in the U.S. since late 1999, bonds and equities have been inversely correlated based on daily price changes. Using historical correlation coefficients to structure a global portfolio can lead to unexpected consequences--witness the interaction of bonds and equities. This makes the asset allocator's job difficult.

Not all sectors can be treated on a global basis. Several sectors, producing standardized products sold into markets around the world (e.g. oil, commodities) often behave as one block. This can be illustrated with charts of major oil stocks over time, using common currencies for improved comparison. Many of the oil majors move in tandem. Looking at other sectors such as banks, there are some similarities between the price bahaviour of major banks from different regions, suggesting a global sector effect here. But studying a chart of the largest retailers in Japan, the U.S. and the U.K., for example, reveals no sign of a pattern at all. Some sectors are multi-local, not global.

Comparing a stock to its local market, or to a stock in a foreign market, the same issue of correlations changing over time is observed. As well as the vagaries of market sentiment, there are specific reasons why the relationship between a stock and its market, or peers, may change. Corporate strategy may change at any time, new products may be developed or older products retired. Restructuring may lead to the disposal of one division or the acquisition of another. All of these change the nature of a company, and lead to considerable variation in correlation coefficients over time. By using weighting structures that emphasize most recent data, obsolescence is reduced, but not eliminated.

THEMATIC INVESTMENT
Of interest to global portfolio managers lately has been thematic investment, where one seeks to exploit themes such as globalization, the rise of outsourcing of services or improvements in communications infrastructure.

Related to this technique (which makes use of linkages between markets for much of its validity) has been the question "Do investment themes start in the U.S.?" In other words, can a manager study trends in the U.S., and then position a global portfolio to benefit from the spread of these trends around the world?

Many leading companies and new technologies originate in the U.S., but many trends actually start elsewhere. For example, third-generation mobile telecommunications licences (an important influence on telecom sector performance recently) originated as a theme outside the U.S. Note that country effects often dominate theme effects--witness the recent Turkish devaluation, or Argentinean political influences, which have proven to be key drivers of these stock markets this year.