 Industry Classifications and Return ComovementLouis K.C. Chan, Josef Lakonishok and Bhaskaran Swaminathan Financial Analysts Journal, 2007, vol. 63, issue 6, 56-70 Abstract: A company’s industry affiliation is commonly used to construct homogeneous stock groupings for portfolio risk management, relative valuation, and peer-group comparisons. A variety of industry classification systems have been adopted, however, creating disagreements as to companies’ industry assignments. This analysis of the Global Industry Classification System (GICS) and Fama–French system indicates that common movement in returns and operating performance resulting from industry effects is stronger for stocks of large companies than for those of small companies. Also, increasingly fine levels of disaggregation improve discrimination up to six-digit GICS codes, after which the benefits tail off. Stock groupings based on industry exhibit stronger out-of-sample homogeneity than groups formed from statistical cluster analysis.Investors and researchers commonly use a company’s industry affiliation to construct homogeneous stock groupings for the purposes of portfolio risk management, relative valuation, and peer-group comparisons. A variety of industry classification systems have been adopted, however, creating disagreements as to companies’ industry assignments. Academic researchers tend to use industry groups developed by Fama and French, which are initially based on Standard Industrial Classification codes. Investment practitioners favor the Global Industry Classification System (GICS). Little evidence is available on the relative performance of alternative classification procedures in terms of their ability to produce groupings of stocks that share coincident price movements. Moreover, the literature provides scant guidance on choosing the fineness of industry disaggregation.We compare the Fama–French (FF) and GICS industry classification schemes in terms of their capacity to isolate common return movements of stocks within an industry relative to covariation with returns on stocks outside the industry. In addition, we document the gains from successively finer levels of industry partitioning. Finally, we verify that industry groups correspond to collections of economically similar companies in terms of sharing common movements in operating performance (as measured by sales growth) and exhibit stronger out-of-sample return covariation than statistical clusters formed without regard to industry affiliation.Large-cap stocks that belong to the same industry as defined by two-digit GICS codes share a simple mean correlation of 0.38, compared with a mean correlation of 0.26 for stocks drawn from different industries. Measured net of an equally weighted market index, when two-digit codes are used, return correlations for large-cap stocks in an industry average 0.17. The magnitudes of within-industry commonality in movements of raw and excess returns highlight the potential benefit of using industry affiliation as one dimension for managing portfolio risk and tracking error in the case of large companies.For smaller companies, however, the comovement of returns associated with commonality in industry membership is much less pronounced. In part, small-cap stocks’ responses to industry effects may be dominated by their higher idiosyncratic volatility. One implication of these results is that constraints on portfolio weights to limit exposures to industries plays a more important role for large-cap stocks than for small-cap stocks.The FF categories fare as well as four-digit GICS groups in terms of the magnitude of mean within-industry correlations but at a cost in terms of parsimony (48 FF groups are needed versus 24 four-digit GICS groups). Finer levels of disaggregation improve discrimination between within-industry and outside-industry correlations except that the benefits generally tail off beyond six-digit GICS codes. For the large-cap sample, correlations between industry members differ from correlations between nonmember companies by, on average, 0.13 at the two-digit GICS level, 0.14 at the four-digit level, 0.17 at the six-digit level, and 0.18 at the eight-digit level.Common industry affiliation also translates into heightened covariation in sales growth. The mean correlation of sales growth for large companies within four-digit GICS industries is 0.22, as opposed to 0.10 for the average outside-industry correlation.Pseudo-industry groups formed from statistical cluster analysis of stock returns do not match the performance of industry classifications on an out-of-sample basis. Both the FF and GICS schemes produce classifications with larger within-group correlations and sharper discrimination over outside-group correlations than grouping by statistical cluster analysis produces. Date: 2007 Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:taf:ufajxx:v:63:y:2007:i:6:p:56-70 Ordering information: This journal article can be ordered from Access Statistics for this article Financial Analysts Journal is currently edited by Maryann Dupes More articles in Financial Analysts Journal from Taylor & Francis Journals |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.