THE PORTFOLIO FLOWS: The Behavior of Portfolio Flows 2

As stated in the introduction, it is natural to think of international trades as embodying a regional component, due to shocks to regional information, preferences, or wealth, and a local-market component, associated with local-market information or liquidity. In what follows, we attempt to decompose flows into these components, and to examine the properties of each component separately. In particular, we assume that the flows of each country can be decomposed as:
where \x( is a country-specific mean flow, ftR a regional factor, (3j is country f s loading on the regional factor, and Eit is the country-specific portion of time-/ flows. The simplest and most natural null hypothesis is that flows are cross-sectionally and serially uncorrelated—i.e., that the regional component ftR is zero and the local-market component Sit is uncorrelated over time. This null is implied if the data are drawn from a random selection of buyers and sellers in each country and over time. It is also implied if local information shocks drive portfolio flows. Since local information shocks can be thought of as orthogonal to regional shocks, then the flows which result from local shocks should be uncorrelated as well.

To effect this decomposition, we first estimate fR using factor analysis (see, Johnson and Wichern (1992) for an example). Specifically, we assume that the correlation matrix of country flows illustrated in Figure 1 can be written as
where ft is the vector of country flows at time /.
Figure 4 provides an example of this regional factor. It shows cumulated inflows into all emerging market countries from August 1994 to May 1998. The factor accounts for net purchases equal to about 1.4% of emerging-market capitalization over the sample. The usefulness of the regional factor measure can be gauged by contrasting its behavior with that of equal- and market capitalization-weighted averages of the flows, which are also shown in Figure 4. For example, very large inflows into Brazil are recorded in the data over a short period in July 1997. These are reflected in a large jump in both the equal- and market capitalization-weighted averages. By contrast, the regional factor exhibits a much smaller jump. The reason is that inflows into Brazil are relatively uncorrelated with other inflows into emerging markets. Accordingly, Brazil receives a relatively small weight in the linear combination fR, and the surge of assets into Brazil has a relatively small impact on the curve.