Characteristic-sorted Portfolios Research Articles

Industry volatility spillover and aggregate stock returns

ABSTRACT We propose an industry volatility spillover index and find that this new predictor has an impressive in-sample and out-of-sample predictability of aggregate stock returns. A one-standard-deviation increase in this index leads to a 4.14% increase in the expected excess return over the next month. Furthermore, a mean-variance investor can realize sizable economic gains by using this spillover index in his/her asset allocation. The forecasting power of the spillover index remains significant after controlling for popular economic variables and newly proposed predictors. Due to the absence of learning, the spillover index shows increasing predictive ability over the recent period. The economic origins of the spillover index’s success stem from both the cash flow and discount rate channels, as well as significant association with investor sentiment and tail risk. The return predictability of this index is pervasive across characteristic-sorted portfolios and is particularly strong for large-cap or difficult-to-value stocks.

Factor seasonalities: International and further evidence

We study factor return seasonalities in international markets. Using up to 143 characteristic-sorted portfolios from 39 countries, we document a pervasive cross-sectional pattern: anomalies with a high average same-calendar month return outperform those with low average returns. The effect persists across individual markets and global samples and cannot be attributed to common risk factors. Neither factor momentum nor cross-sectional variation in unconditional premia explains the phenomenon. Instead, the effect originates from price seasonalities, which transmit to factor portfolios, engendering seasonality in their returns. Consequently—rather than manifesting an independent asset pricing phenomenon—factor seasonality merely reflects its stock-level equivalent.

Characteristic-sorted portfolios and macroeconomic risks—An orthogonal decomposition

We assess the relative importance of individual macroeconomic variables in explaining the time series variation of a broad range of characteristic-sorted portfolios. Employing an optimal orthogonalization approach, the explained variation in each characteristic-sorted portfolio is decomposed into components associated with individual macroeconomic variables. When examined unconditionally, the set of macroeconomic variables account for only limited explained variation. This low explained variation is partially resolved by allowing for dynamic and non-linear macroeconomic exposure. Finally, we highlight a degree of commonality in the explained macroeconomic variation associated with various characteristic-sorted portfolios, indicating that they proxy for particular macroeconomic characteristics simultaneously. • Assess links between characteristic-sorted portfolios and macroeconomic variables. • Focus on explained variation using an optimal variable orthogonalization approach. • Macroeconomic variables account for limited explained variation unconditionally. • Provide evidence for pronounced dynamics in macroeconomic exposure attributions. • Highlight a degree of commonality in the explained macroeconomic variation.

Optimal characteristic portfolios

Characteristic-sorted portfolios are the workhorses of modern empirical finance, deployed widely to evaluate anomalies and construct asset pricing models. We propose a new method for their estimation that is simple to compute, makes no ex-ante assumption on the nature of the relationship between the characteristic and returns, and does not require ad hoc selections of percentile breakpoints or portfolio weighting schemes. Characteristic portfolio weights are implied directly from data, through maximizing a Mean–Variance objective function with mean and variance estimated non-parametrically from the cross-section of assets. To illustrate the method, we evaluate the size, value and momentum anomalies and find overwhelming empirical evidence of the outperformance of our methodology compared to standard methods for constructing characteristic-sorted portfolios.

Machine Learning and Factor-Based Portfolio Optimization

We examine machine learning and factor-based portfolio optimization. We find that factors based on autoencoder neural networks exhibit a weaker relationship with commonly used characteristic-sorted portfolios than popular dimensionality reduction techniques. Machine learning methods also lead to covariance and portfolio weight structures that diverge from simpler estimators. Minimum-variance portfolios using latent factors derived from autoencoders and sparse methods outperform simpler benchmarks in terms of risk minimization. These effects are amplified for investors with an increased sensitivity to risk-adjusted returns, during high volatility periods or when accounting for tail risk.

Correlated idiosyncratic volatility shocks

The commonality in idiosyncratic volatility cannot be fully explained by time-varying volatility; correlated idiosyncratic volatility shocks are an important contributing factor. We empirically document this fact for various characteristic-sorted portfolios and industry portfolios. To capture the commonality in idiosyncratic volatility, we propose a novel multivariate generalized autoregressive conditional heteroscedasticity (GARCH) model called dynamic factor correlation (DFC). The DFC model has a closed-form likelihood function that allows computationally cheap estimation, even for a large number of assets. The DFC model improves statistical fit compared with existing multivariate GARCH models such as the dynamic conditional correlation model and the dynamic equicorrelation (DECO) model, achieving the lowest root mean square error and mean absolute error in simulations. Empirical tests also corroborate simulations in demonstrating the improved statistical fit of the DFC model. Mean--variance portfolio optimization using the DFC model outperforms alternative volatility models such as the historical covariance matrix and the DECO model. Out-of-sample mean--variance-efficient portfolios using the DFC model have the lowest volatility and the highest Sharpe ratios, thereby improving the investor's opportunity set. Under parametric restrictions, the DFC model reduces to the constant conditional correlation model or the DECO model.

The Performance of Characteristic-Sorted Portfolios: Evaluating the Past and Predicting the Future

The expected returns of most portfolios are likely to fluctuate over time. We present a statistical model that allows for such fluctuations and apply the model to analyze the returns of characteristic-sorted portfolios, such as value minus growth. We find that accounting for plausible magnitudes of persistent variation in returns doubles the standard errors of these portfolios' expected return estimates. We also analyze characteristic-sorted portfolios from the perspective of Bayesian investors and show that investors' posterior beliefs about expected returns are highly dependent on their priors about persistence.

Housing Risk and the Cross-Section of Returns Across Many Asset Classes

When households consume both nondurable goods and housing services, external habit preference over nondurable consumption generates procyclical demand for housing. Marginal utility falls when housing demand rises and innovations to housing demand arise as a risk factor. Motivated by theory, we use shocks to the ratio of residential-to-aggregate investment to capture the housing demand risk. The single-factor model exhibits strong explanatory power for expected returns across various equity characteristic-sorted portfolios and non-equity asset classes with positive risk price estimates that are similar in magnitude. The model is robust to controlling for other factor models based on durable consumption, financial intermediaries, household heterogeneity, and return-based multifactor models designed to price these assets.

Optimal Characteristic Portfolios

Characteristic-sorted portfolios are the workhorses of modern empirical finance, deployed widely to evaluate anomalies and construct asset pricing models. We propose a new method for their estimation that is simple to compute; makes no ex-ante assumption on the nature of the relationship between the characteristic and returns; and does not require ad hoc selections of percentile breakpoints or portfolio weighting schemes. Characteristic portfolio weights are implied directly from data, through maximizing the expected value of a mean-variance utility function estimated non-parametrically over the cross section of the assets. To illustrate the method we use it to evaluate the size, value and momentum anomalies.

Do Growth Expectations Help Explain Characteristic-Sorted Portfolio Returns?

We find that accounting ratios (asset and sales growth, profitability, and equity dilution) that predict stock returns are associated with errors in analyst long-term growth forecasts. Specifically, accounting information that is associated with favorable long-term growth forecasts tends to predict negative analyst forecast errors and negative future excess returns. This and other evidence we present is consistent with the idea that biased long-term growth forecasts generate the observed return premia of popular characteristic-sorted portfolios.

What explains the investment anomaly in the Chinese stock market?

PurposeThis paper aims to examine the Chinese investment anomaly and dissect it from a perspective of rational expectation framework.Design/methodology/approachCharacteristic-based sorting and Fama–MacBeth two-stage cross-sectional regression are adopted to test the relationship between corporate investment and expected returns in both portfolio and individual stock levels. Under the framework of pricing kernels, an investment-based common risk factor is constructed to test the role of risk played in the negative investment-return relationship. Moreover, a Markov regime switching model is adopted to investigate the time-varying risk premium across market regimes.FindingsEmpirical results provide ample evidence showing that there is a negative relationship between investment and expected returns in the Chinese stock market. The new investment-based risk factor is found to capture the return differences across characteristic-based portfolios. In addition, risk premium of the new risk factor is not only statistically positive throughout the sample period, but also has an asymmetry that is higher during market downturn but lower under bull market.Research limitations/implicationsThis paper merely tests the hypotheses derived from rational school.Practical implicationsInvestment strategies based on characteristic-sorted portfolios should be adjusted to different market regimes.Originality/valueFirst, this paper provides comprehensive empirical results by adopting different methodologies for investigating the investment anomaly in China. Second, an investment-based factor is constructed specifically for the Chinese stock market for the first time. Finally, this is the first paper to investigate the asymmetric risk premium across the Chinese bear and bull regimes by using a multivariate Markov regime switching model.

Metals: resources or financial assets? A multivariate cross-sectional analysis

Metals are very important resources for industrial production, but recently they have attracted more and more attention from investors. While certainly industrial producers, consumers, and financial investors do have some influence on metal price development, the role of relevant price factors is not yet quite clear. Therefore, in this paper, we examine the explanatory power of various fundamental factors and characteristics known from financial markets, specifically on the expected returns in a unique data sample of 30 metals. We apply—to our knowledge for the first time in this context—the widely accepted method of characteristic-sorted portfolios, extended by the very recent method of two-way portfolio sorts as an alternative to classical multivariate regressions. This mostly nonparametric approach, combined with portfolio aggregation, provides very robust results. Our major finding is that the financial characteristics value and momentum have a very high predictive power for monthly returns of metal portfolios. Metal-specific fundamental factors like stocks, secondary production, apparent consumption, country concentration, mine production, or reserves perform depending on the interpretation moderately well or rather poorly, regarding some economically interpretable transformations and when using multivariate two-way sorts. Hence, from the perspective of expected returns, metals are predominantly assets, while fundamental metal-specific factors still play a non-negligible role. Thus, to a much lesser extent, metals can still be regarded as resources. Overall, the combination of financial characteristics and metal-specific fundamental factors yields the best results. With these robust results, we hope to contribute to a better understanding of metal prices and their underlying factors.

Conditional Benchmarks and the Identification of Skill in Active Management

Recent studies link mutual fund performance to measures of active management, and this evidence often takes the form of large spreads in unconditional alphas for characteristic-sorted portfolios. Unconditional benchmarks can, however, produce misleading inferences on managerial skill for strategies that exhibit substantial turnover and unstable factor exposures. We propose an approach to performance attribution that accounts for predictable changes in portfolio style. Compared to existing methods, our benchmarks yield superior tracking performance and a more powerful statistical assessment of abnormal returns. We reevaluate six proxies for active management using our method and conclude that these measures are largely unrelated to managerial ability.

Anomalies, risk adjustment and seasonality: Australian evidence

On the basis of raw return analysis, economically significant anomalies appear to exist in relation to the size, momentum, book-to-market and profitability of Australian firms. However, characteristic-sorted portfolios are shown to load in very particular ways on multiple risk factors. After adjusting for exposure to risk, convincing evidence only remains for the size premium. An analysis of seasonality shows that, rather than being consistent throughout the year, anomaly returns are concentrated in a handful of months. We provide and test preliminary explanations of the observed seasonality in these well-known anomalies.

Modelling Dependence Structure and Forecasting Portfolio Value-at-Risk with Dynamic Copulas

We investigate the dynamic and asymmetric dependence structure between equity portfolios from the US and UK. We demonstrate the statistical significance of dynamic asymmetric copula models in modelling and forecasting market risk. First, we construct "high-minus-low" equity portfolios sorted on beta, coskewness, and cokurtosis. We find substantial evidence of dynamic and asymmetric dependence between characteristic-sorted portfolios. Second, we consider a dynamic asymmetric copula model by combining the generalized hyperbolic skewed t copula with the generalized autoregressive score (GAS) model to capture both the multivariate non-normality and the dynamic and asymmetric dependence between equity portfolios. We demonstrate its usefulness by evaluating the forecasting performance of Value-at-Risk and Expected Shortfall for the high-minus-low portfolios. From backtesting, we find consistent and robust evidence that our dynamic asymmetric copula model provides the most accurate forecasts, indicating the importance of incorporating the dynamic and asymmetric dependence structure in risk management.

An alternative measure of the “world market portfolio”: Determinants, efficiency, and information content

The world market portfolio plays an important role in international asset pricing, but is unobservable in practice. We first propose a framework for constructing a market proxy that corresponds to the “market portfolio” of financial theory. We then construct this proxy, analyze its determinants and test its efficiency and explanatory power over the period 1975–2007 with respect to the return generating processes of a broad asset universe. We show that its major determinants are traded assets and that it is not efficient. However, it is significant for explaining individual asset returns over an asset universe that includes stocks, bonds, money markets and commodities. The explanatory information is incremental to what is available in traded asset prices and the significance of this information is robust with respect to diversified portfolios generated by factor analysis and to characteristic-sorted portfolios as well as to various model specifications, including the single-index model, the Fama–French (1992) three factor model for stocks, and various specifications of multi-index models hedged and unhedged for foreign currency risk.

Characteristic-Based Covariances and Cross-Sectional Expected Returns

I suggest a characteristic-based covariance model that directly links the predetermined fi rm characteristics to time-varying covariance risk. Using a large cross section of individual stock-level data, I parsimoniously estimate both conditional expected returns and conditional covariances as functions of fi rm characteristics. I fi nd a strong and positive intertemporal risk-return relation on individual stocks. In comparison to the Fama-French three-factor model, the characteristic-based covariance model substantially reduces the pricing errors of characteristic-sorted portfolios on size, book-to-market, accruals, asset growth, investment-to-assets, return-on-assets, net stock issues, financial distress, and momentum. Portfolio tests similar to Daniel and Titman (1997) suggest that firm characteristics, mainly through the characteristic-based covariance structure of returns, appear to explain the cross section of average stock returns.

Rational Sources of Momentum Profits: Evidence from the Indian Equity Market

In this paper, we evaluate if there are any momentum patterns in Indian stock returns. We report strong momentum profits for individual stocks as well as a wide range of characteristic-sorted portfolios. The momentum returns are not captured by standard CAPM. However, the Fama-French three-factor model explains momentum as the characteristic-sorted momentum portfolios tend to load on size and book equity to market equity factors. Our results suggest that there are rational sources of momentum profits in the Indian environment. The findings are in contrast with those of US where the momentum explanation is tilted towards delayed price reaction to firm-specific information. Perhaps, a more complex market like US requires an asset pricing model that has a factor structure which is more elaborate than the Fama-French model. Our work is a step towards resolving the momentum controversy.