The behaviour of style anomalies in worldwide sector indices : a univariate and multivariate analysis

Master Thesis


Permanent link to this Item
Journal Title
Link to Journal
Journal ISSN
Volume Title

University of Cape Town

The aim of this thesis is to explain the cross-section of International Classification Benchmark (ICB) level 4 (sector) index returns. A worldwide study of 48 developed and emerging countries is conducted, considering up to 38 sector indices per country. In cluster and factor analyses of the sector returns all the developed markets are found to cluster together, as are the emerging markets, suggesting diversificationary benefits from investing across the two. The one-month-ahead return forecasting power of 35 sector-specific attributes is investigated over an in-sample period from 31 January 1995 to 31 December 2001 and an out-sample period from 31 January 2002 to 31 December 2005. The data is adjusted for look-ahead bias, outliers, influential observations and non-uniformity across markets. Monthly sector returns are cross-sectionally regressed on the attributes in a similar fashion to Fama and MacBeth (1973). Sector returns are considered both before and after risk adjustment with the Capital Asset Pricing Model (CAPM), the Arbitrage Pricing Theory (APT) model and Solnik's (2000) version of the International CAPM (ICAPM). The ICAPM is found to be the best performing model but, in general, the evidence does not support covariance-based models of asset pricing. Nine attributes are found to be significant and robust over the two sample periods namely cash earnings per share to price (CP), dividend yield (DY), cash earnings to book value (CB), 6 and 12-month growth in cash earnings, to price (C-6P & C-12P), 12 and 24-month growth in dividends, to price (D-12P & D-24P), the payout ratio (PO) and 12-month prior return (MOM-12). All the significant attributes from the univariate regression tests are found to payoff consistently in the positive direction when tested with the nonparametric Sign Test. Nine of the significant attributes namely book value per share to price (BP), dividend yield (DY), earnings yield (EY), 6-month growth in cash earnings, to price (C-6P), cash earnings to book value (CB), 24-month growth in dividends, to price (D-24P), 24-month growth in earnings, to price (E-24P), 12-month and 18-month prior return (MOM-12 & MOM-18) are also found to have significantly low frequencies of changes in payoff direction when assessed with the nonparametric Runs Test. Seven style timing models are developed, all of which produce significantly accurate payoff direction forecasts for most of the significant attributes. The timing models are however generally inaccurate in forecasting the magnitude of the payoffs. Very little seasonality is observed in the payoffs to the significant attributes. Two sets of seven 'stepwise optimal' and 'control' multivariate models are constructed from the significant univariate in-sample attributes in order to forecast the payoffs to the factors in a controlled multifactor setting. The stepwise optimal models are derived from a stepwise procedure, whilst the 'control' models comprise all the attributes which are found to be significant in one or more of the 'optimal' models. The forecasting power of the all the models is found to be below an exploitable level; of the 'control' models the single exponential smoothing model is the most accurate outsample performer. Weighted Least Squares (WLS) models are used to allow for the possibility of heteroskedasticity, which may exist in the cross-section of worldwide sector returns. The WLS models are ineffective in improving forecasting power when the inverse of the 12-month rolling standard deviation of the residuals is used as the weight series.

Includes bibliographical references.