Existing Factor Models Research Articles

Estimating Asset Pricing Models in the Presence of Cross-Sectionally Correlated Pricing Errors

In this study, we propose an adversarial learning approach to the asset pricing model estimation problem which aims to find estimates of factors and loadings that capture time-series covariations while minimizing the worst-case cross-sectional pricing errors. The proposed estimator is defined by a novel min-max optimization problem in which finding a solution is known to be difficult. This contrasts with other related estimators that admit a well-defined analytic solution but do not effectively account for correlations among the pricing errors. To this end, we propose an approximate algorithm based on the alternating optimization procedure and empirically demonstrate that our proposed adversarial estimation framework outperforms other existing factor models, especially when the explanatory power of the pricing model is limited.

Production similarity and the cross‐section of stock returns: A machine learning approach

AbstractThis paper employs a machine learning approach to capture firm‐pair production similarity, which depicts how firms' production processes resemble each other using textual data in corporate MD&As. We show that production‐linked firms' average return has strong predictive power on focal firm's future stock return. A hedging portfolio yields an annualised return of 11.69%, which cannot be subsumed by existing factor models. For mechanism tests, we find that the main findings are stronger in firms with higher information asymmetry and higher costs of arbitrage. The production‐linkage measure also predicts future unexpected earnings, suggesting it possibly includes valuable information on firm fundamentals.

Using, taming or avoiding the factor zoo? A double-shrinkage estimator for covariance matrices

Existing factor models struggle to model the covariance matrix for a large number of stocks and factors. Therefore, we introduce a new covariance matrix estimator that first shrinks the factor model coefficients and then applies nonlinear shrinkage to the residuals and factors. The estimator blends a regularized factor structure with conditional heteroskedasticity of residuals and factors and displays superior all-around performance against various competitors. We show that for the proposed double-shrinkage estimator, it is enough to use only the market factor or the most important latent factor(s). Thus there is no need for laboriously taking into account the factor zoo.

How is the change in left-tail risk priced in China?

In this paper, we find a negative cross-sectional relationship between the change in left-tail risk and expected returns in the Chinese stock market. This effect cannot be explained by the common control variables and existing factor models in China, including the four-factor model (CH4) proposed by Liu, Stambaugh, and Yuan (2019). The predictive power of the change in left tail risk on expected returns can persist several months into the future, which may due to the strong prior beliefs of investors and the slow diffusion of fundamental information. Additionally, we observe that investors exhibit a stronger preference toward change in left-tail risk for stocks with lower capital gains overhang (CGO).

A trend factor in commodity futures markets: Any economic gains from using information over investment horizons?

AbstractThis paper identifies a trend factor that exploits the short‐, intermediate‐, and long‐run moving averages of settlement prices in commodity futures markets. The trend factor generates statistically and economically large returns during the post‐financialization period 2004–2020. It outperforms the well‐known momentum factor by more than nine times the Sharpe ratio and has less downside risk. The trend factor cannot be explained by the existing factor models and is priced cross‐sectionally. Finally, we find that the trend factor is correlated with funding liquidity measured by the TED spread. Overall, the results indicate that there are significant economic gains from using the information on historical prices in commodity futures markets.

How Does the Change in Left-Tail Risk Priced in China?

In this paper, we find a negative cross-sectional relationship between the change in left-tail risk and expected returns in the Chinese stock market. This effect cannot be explained by the common control variables and existing factor models in China, including the four-factor model (CH4) proposed by Liu, Stambaugh, and Yuan (2019). Additionally, we observe that investors exhibit a stronger preference towards change in left-tail risk for stocks with lower capital gains overhang (CGO).

A Trend Factor in Commodity Futures Markets: Any Economic Gains From Using Information Over Investment Horizons?

This paper identifies a trend factor that exploits the short-, intermediate-, and long-run moving averages of settlement prices in commodity futures markets. The trend factor generates statistically and economically large returns during the post-financialization period 2004-2020. It outperforms the well-known momentum factor by more than nine times the Sharpe ratio and has less downside risk. The trend factor cannot be explained by the existing factor models and is priced cross-sectionally. Finally, we find that the trend factor is correlated with funding liquidity measured by the TED spread. Overall, the results indicate that there are significant economic gains from using the information on historical prices in commodity futures markets.

Benchmark Currency Stochastic Discount Factors

We examine the pricing performance of out-of-sample pricing factors in the broad cross-section of currency returns. To this end, we develop a methodology for estimating empirical minimum-entropy stochastic discount factors (SDFs) under economically-motivated constraints on position leverage. Our empirical SDFs deliver superior out-of-sample fit and smaller pricing errors than existing factor models in the cross-section of currency portfolio returns, and are priced in individual currency and hedge fund cross-sections. After transaction costs, an investable SDF portfolio delivers a Sharpe ratio of around 0.8 and positively skewed returns. These empirical SDFs offer tractable benchmarks for candidate currency pricing models.

Using, Taming or Avoiding the Factor Zoo? A Double-Shrinkage Estimator for Covariance Matrices

Existing factor models struggle to model the covariance matrix for a large number of stocks and factors. Therefore, we introduce a new covariance matrix estimator that first shrinks the factor model coefficients and then applies nonlinear shrinkage to the residuals and factors. The estimator blends a regularized factor structure with conditional heteroskedasticity of residuals and factors and displays superior all-around performance against various competitors. We show that for the proposed double- shrinkage estimator, it is enough to use only the market factor or the most important latent factor(s). Thus there is no need for laboriously taking into account the factor zoo. Supplementary material for this article is available online.

Manager Uncertainty and Cross-Sectional Stock Returns

This paper evidences the explanatory power of managers’ uncertainty for cross-sectional stock returns. I introduce a novel measure of the degree of managers’ uncertain beliefs about future states: manager uncertainty (MU), defined as the count of the word “uncertainty” over the sum of the count of the word “uncertainty” and the count of the word “risk” in filings and conference calls. I find that manager’s level of uncertainty reveals valuation information about real options and thereby has significantly negative explanatory power for cross-sectional stock returns. Beyond existing market-based uncertainty measures, the manager uncertainty measure has incremental pricing power by capturing information frictions between managers’ reported uncertainty and investors’ perception of uncertainty. Moreover, a short-long portfolio sorted by manager uncertainty has a significantly positive premium and cannot be spanned by existing factor models. An application on COVID-19 uncertainty shows consistent results.

Factorization Meets Neural Networks: A Scalable and Efficient Recommender for Solving the New User Problem

Predicting new user behavior has always been a challenging issue in intelligent recommender systems. This challenge is mainly due to the extreme asymmetry of information between new users and old users. Existing factorization models can efficiently process and map asymmetric information, but they are not good at mining deep relationships between contexts when compressing high-dimensional data. In contrast, neural network methods can deeply exploit the relationship between contexts; however, their training cost is much larger than factorization approaches. Therefore, this paper proposes a scalable and efficient recommender to solve the new user problem by filling the gap between factorization models and neural networks. The scalable part is a neural network that can jointly encode, compress, and fuse various types of contexts. The efficient part is a factorization model based on a correlation constraint mechanism and a projection strategy, which enables an asymmetric mapping of information between old and new users. The entire recommender fuses the two parts so that the factorization model and the neural network can complement each other. The experimental results show that our approach can achieve a good balance between performance and training efficiency compared to state-of-the-art methods.

Characteristics are covariances: A unified model of risk and return

We propose a new modeling approach for the cross section of returns. Our method, Instrumented Principal Component Analysis (IPCA), allows for latent factors and time-varying loadings by introducing observable characteristics that instrument for the unobservable dynamic loadings. If the characteristics/expected return relationship is driven by compensation for exposure to latent risk factors, IPCA will identify the corresponding latent factors. If no such factors exist, IPCA infers that the characteristic effect is compensation without risk and allocates it to an “anomaly” intercept. Studying returns and characteristics at the stock-level, we find that five IPCA factors explain the cross section of average returns significantly more accurately than existing factor models and produce characteristic-associated anomaly intercepts that are small and statistically insignificant. Furthermore, among a large collection of characteristics explored in the literature, only ten are statistically significant at the 1% level in the IPCA specification and are responsible for nearly 100% of the model’s accuracy.

Factor Models for Portfolio Selection in Large Dimensions: The Good, the Better and the Ugly

Abstract This paper injects factor structure into the estimation of time-varying, large-dimensional covariance matrices of stock returns. Existing factor models struggle to model the covariance matrix of residuals in the presence of time-varying conditional heteroskedasticity in large universes. Conversely, rotation-equivariant estimators of large-dimensional time-varying covariance matrices forsake directional information embedded in market-wide risk factors. We introduce a new covariance matrix estimator that blends factor structure with time-varying conditional heteroskedasticity of residuals in large dimensions up to 1000 stocks. It displays superior all-around performance on historical data against a variety of state-of-the-art competitors, including static factor models, exogenous factor models, sparsity-based models, and structure-free dynamic models. This new estimator can be used to deliver more efficient portfolio selection and detection of anomalies in the cross-section of stock returns.

The factor structure of the Quality of Life Inventory (QOLI) following traumatic brain injury

ABSTRACTQuality of life is a key indicator of outcome following traumatic brain injury (TBI). Research has reported several different factor structures for the Quality of Life Inventory (QOLI, Frisch, 1994). We compared the fit of existing factor models and examined the clinical utility of the QOLI’s factors in a sample of Australian adults with TBI. Archival data from 901 participants were provided by the Neurotrauma Register of Tasmania. Participants were aged 16–80 years and 63% were male. Approximately 69% had mild TBI (PTA < 24 h), approximately 24% had moderate TBI (PTA >1 day, <7 days) and 7% had severe TBI. Both cross sectional and longitudinal analyses were utilized, as participants provided data at one or more of seven time-points, up to 3 years following injury. The results showed the data best fitted a three-factor model, comprising Self-functioning and activity, Self-actualization and Family and environment factors, and a second order Overall QOL factor. Differences in the trajectory of recovery were noted between the QOLI factor scores over time and in relation to demographic and injury variables. In conclusion, the three-factor structure of the QOLI provided useful clinical information about the recovery of patients’ subjective quality of life following TBI.

Factor Models for Portfolio Selection in Large Dimensions: The Good, the Better and the Ugly

This paper injects factor structure into the estimation of time-varying, large-dimensional covariance matrices of stock returns. Existing factor models struggle to model the covariance matrix of residuals in the presence of time-varying conditional heteroskedasticity in large universes. Conversely, rotation-equivariant estimators of large-dimensional time-varying covariance matrices forsake directional information embedded in market-wide risk factors. We introduce a new covariance matrix estimator that blends factor structure with time-varying conditional heteroskedasticity of residuals in large dimensions up to 1000 stocks. It displays superior all-around performance on historical data against a variety of state-of-the-art competitors, including static factor models, exogenous factor models, sparsity-based models, and structure-free dynamic models. This new estimator can be used to deliver more efficient portfolio selection and detection of anomalies in the cross-section of stock returns.

Some Characteristics Are Risk Exposures, and the Rest Are Irrelevant

We propose a new modeling approach for the cross section of returns. Our method, Instrumented Principal Components Analysis (IPCA), allows for latent factors and time-varying loadings by introducing observable characteristics that instrument for the unobservable dynamic loadings. If the characteristics/expected return relationship is driven by compensation for exposure to latent risk factors, IPCA will identify the corresponding latent factors. If no such factors exist, IPCA infers that the characteristic effect is compensation without risk and allocates it to an intercept. Studying returns and characteristics at the stock-level, we find that four IPCA factors explain the cross section of average returns significantly more accurately than existing factor models and produce characteristic-associated anomaly intercepts that are small and statistically insignificant. Furthermore, among a large collection of characteristics explored in the literature, only eight are statistically significant in the IPCA specification and are responsible for nearly 100% of the model's accuracy. Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.

How does corporate governace pay off? Evidence from Korean stock listings

Corporate governance is an envelope for the mechanisms, processes and relations through which corporations are controlled and guided. Consequently, corporate governance affects operational performance and, in turn, stock returns, as Gompers et al. (2003) find. In this research, we use the Korea Corporate Governance Stock Price Index (KOGI) to test a possible linkage between corporate governance and shareholder wealth in Korea.Factor mimicking portfolios sorted per KOGI are constructed to estimate a corporate governance risk factor (“good minus bad”). By augmenting this new factor to the existing factor models (Fama and French, 1993; Carhart, 1997) to fit multiply imputed data, we find evidence that corporate governanceinfluences stock pricing in Korea. Keywords: CG; Risk factor; Factor-mimicking portfolio; Long-short portfolio; Multiple imputation. JEL Classification: G11, G12, G34, C11

Activation as an overlooked factor in the BDI-II: a factor model based on core symptoms and qualitative aspects of depression.

An adequate assessment of depression has been of concern to many researchers over the last half-century. These efforts have brought forth a manifold of depression rating scales, of which the Beck Depression Inventory (BDI) is 1 of the most commonly used self-assessment scales. Since its revision, the item structure of the BDI-II has been examined in many factor analytic studies, yet it has not been possible to achieve a consensus about the underlying factor structure. Recent findings from a nonmetric multidimensional scaling (NMDS) analysis (Bühler, Keller, & Läge, 2012) of the German norming sample of the BDI-II emphasized a structure with different qualitative aspects of depression, which suggested that the existing factor models do not adequately represent the data. The NMDS results were reviewed, and on the basis of these findings, a different factor model is proposed. In contrast to the common factor models in the literature, the presented model includes an additional factor, which is associated with the activation level of the BDI-II symptoms. The model was evaluated with a 2nd sample of patients diagnosed with a primary affective disorder (N = 569) and obtained good fit indices that even exceeded the fit of the most reliable factor model (Ward, 2006) described in the literature so far. Furthermore, emphasis is placed on the methodological question of how factor models may be derived from the results of NMDS analyses.

Nowcasting and Forecasting Economic Growth in the Euro Area Using Principal Components

Many empirical studies have shown that factor models produce relatively accurate forecasts compared to alternative short-term forecasting models. These empirical ndings have been established for dierent macroeconomic data sets and dierent forecast horizons. However, various specications of the factor model exist and it is a topic of debate which specication is most eective in its forecasting performance. Furthermore, the forecast performances of the dierent specications during the recent nancial crisis are also not well documented. In this study we investigate these two issues in depth. We empirically verify the forecast performance of three factor model approaches and report our ndings in an extended empirical out-ofsample forecasting competition for quarterly growth of gross domestic product in the euro area and its ve largest countries over the period 1992-2012. We also introduce two extensions of existing factor models to make them more suitable for real-time forecasting. We show that the factor models have been able to systematically beat the benchmark autoregressive model, both before as well as during the nancial crisis. The recently proposed collapsed dynamic factor model shows the highest forecast accuracy for the euro area and the majority of countries that we have analyzed. The forecast precision improvements against the benchmark model can range up to 77% in mean square error reduction, depending on the country and forecast horizon.

Nowcasting and Forecasting Economic Growth in the Euro Area Using Principal Components

Many empirical studies have shown that factor models produce relatively accurate forecasts compared to alternative short-term forecasting models. These empirical findings have been established for different macroeconomic data sets and different forecast horizons. However, various specifications of the factor model exist and it is a topic of debate which specification is most effective in its forecasting performance. Furthermore, the forecast performances of the different specifications during the recent financial crisis are also not well documented. In this study we investigate these two issues in depth. We empirically verify the forecast performance of three factor model approaches and report our findings in an extended empirical out-of-sample forecasting competition for quarterly growth of gross domestic product in the euro area and its five largest countries over the period 1992-2012. We also introduce two extensions of existing factor models to make them more suitable for real-time forecasting. We show that the factor models have been able to systematically beat the benchmark autoregressive model, both before as well as during the financial crisis. The recently proposed collapsed dynamic factor model shows the highest forecast accuracy for the euro area and the majority of countries that we have analyzed. The forecast precision improvements against the benchmark model can range up to 77% in mean square error reduction, depending on the country and forecast horizon.