Return Predictability Research Articles

Forecasting equity returns with oil prices: addressing model choice uncertainty

ABSTRACT We introduce a method to adapt Sala-i-Martin’s (1997) Extreme Bounds Analysis (EBA) to out-of-sample forecasting. This method allows us to robustly detect if the information in a particular variable improves forecasting performance. As an application of our modified EBA, we test whether the information in oil prices can be used to forecast stock returns. Using multiple models with different combinations of predictor variables, we find moderate evidence to suggest that oil prices provide incremental information about out-of-sample stock returns at the aggregate level. However, after combining oil prices with other oil-related variables, we find strong evidence of out-of-sample predictability for returns at the aggregate and sectoral levels. A Monte Carlo simulation exercise confirms that the methodology has the power to identify predictability.

Source of US market predictability in international equities: the investor attention perspective

US stock returns lead other countries due to delays in information transmission caused by limited investor attention. But two conditions must be held for this argument to be tenable. One is attention’s domestic return predictability, and the other is the US’s internationally central position among attention lead-lag relationships. While many recent studies suggest investors’ limited attention materially influences future asset returns, there lacks direct evidence of investors focusing on the US first and then gradually turning their attention to the rest of the world. This paper fills this gap by demonstrating the international leadership of investor attention on the US equity market.

Granular information and sectoral movements

This paper shows a strong link between the granular information contained in individual stock prices and sectoral movements. We find that a predictor aggregating the price movements of a broad cross section of individual stocks predicts intraday returns of sector ETF. When we further incorporate the information from structural models, the resulting information signal has even stronger return predictability. These results support theories of granular and network origins of aggregate shocks.

CNN-CBAM-LSTM: Enhancing Stock Return Prediction Through Long and Short Information Mining in Stock Prediction

Deep learning, a foundational technology in artificial intelligence, facilitates the identification of complex associations between stock prices and various influential factors through comprehensive data analysis. Stock price data exhibits unique time-series characteristics; models emphasizing long-term data may miss short-term fluctuations, while those focusing solely on short-term data may not capture cyclical trends. Existing models that integrate long short-term memory (LSTM) and convolutional neural networks (CNNs) face limitations in capturing both long- and short-term dependencies due to LSTM’s gated transmission mechanism and CNNs’ limited receptive field. This study introduces an innovative deep learning model, CNN-CBAM-LSTM, which integrates the convolutional block attention module (CBAM) to enhance the extraction of both long- and short-term features. The model’s performance is assessed using the Australian Standard & Poor’s 200 Index (AS51), showing improvement over traditional models across metrics such as RMSE, MAE, R2, and RETURN. To further confirm its robustness and generalizability, Diebold–Mariano (DM) tests and model confidence set experiments are conducted, with results indicating the consistently high performance of the CNN-CBAM-LSTM model. Additional tests on six globally recognized stock indices reinforce the model’s predictive strength and adaptability, establishing it as a reliable tool for forecasting in the stock market.

Deep neural network model enhanced with data preparation for the directional predictability of multi-stock returns

This study presents novel deep neural networks (DNNs) that integrate a thorough data preparation process, including dollar bar sampling, trend scanning labeling, and piecewise aggregate approximation (PAA), to extract features for predicting the directional returns of leading technology stocks on the US stock exchange, such as Apple, Google, Microsoft, and Intel. Our DNN approach seeks to enhance the quality of input data for the models, thereby improving their predictive accuracy. The empirical analysis, conducted with high-frequency data from 2012 to 2022, reveals that DNN models—particularly the one employing dollar bars and trend scanning labeling (DB-TSC)—exhibit vital accuracy and generalizability in forecasting stock return directions. Our research emphasizes the importance of data preparation and fine-tuning model parameters for reliable predictions. This study presents valuable insights into applying deep neural networks (DNNs) for financial forecasting. It is an effective tool for investors and financial analysts aiming to navigate the complexities of the stock market.

Forecasting the realized variance in the presence of intraday periodicity

This paper examines the impact of intraday periodicity on forecasting realized volatility using a heterogeneous autoregressive model (HAR) framework. We show that periodicity inflates the variance of the realized volatility and biases jump estimators. This combined effect adversely affects forecasting. To account for this, we propose a periodicity-adjusted HAR model, HARP, where predictors are constructed from the periodicity-filtered data. We demonstrate empirically (using 30 stocks from various business sectors and the SPY for the period 2000–2020) and via Monte Carlo simulations that the HARP models produce significantly better forecasts across all forecasting horizons. We also show that adjusting for periodicity when estimating the variance risk premium improves return predictability.

Optimal attention allocation: picking alpha or betting on beta?

We investigate a problem of attention allocation and portfolio selection with information capacity constraint and return predictability in a multi-asset framework. In a two-phase formulation, the optimal attention strategy maximizes the combined expected alpha payoffs and expected beta payoffs of the portfolio. Return predictors taking extreme values incentivize the investor to learn about them and this leads to competition among information sources for attention. Moreover, the investor trades with varying skills including picking alphas and betting on beta, depending on the magnitude of the related predictors. Our multi-period analysis using reinforcement learning demonstrates time-horizon effects on attention and investment strategies.

Deconstructing the Yield Curve

Abstract We introduce a novel nonparametric bootstrap for the yield curve that is agnostic to the true factor structure of interest rates. We deconstruct the yield curve into primitive objects, with weak cross-sectional and time-series dependence, that serve as building blocks for resampling the data. We analyze the properties of the bootstrap for mimicking salient features of the data and conducting valid inference. We demonstrate the benefits of our general method by revisiting the predictability of bond returns based on slow-moving fundamentals. We find that trend inflation, but not the equilibrium real rate, has predictive power for future bond returns. (JEL G10, G12, C15, C58)

The return predictability of technological links: evidence from the Chinese STAR Market

ABSTRACT This paper examines the predictive power of technology-linked returns in the Chinese Science and Technology Innovation Market (STAR Market). Using a sample from August 2019 to June 2023, we construct technology-linked returns based on patent portfolio similarities among firms. Our findings indicate that these returns significantly predict future stock returns, with a long-short strategy yielding weekly excess returns between 0.50% and 0.53%. Analysis of trade sizes reveals that institutional investors are more likely to drive this predictability, as the direction of large trades aligns with technology-linked returns. This study explores how technology-peer information affects stock returns in an emerging, technology-intensive market, highlighting the significant role of traders with large capital in leveraging technology-peer information.

Conformal prediction with local weights: randomization enables robust guarantees

Abstract In this work, we consider the problem of building distribution-free prediction intervals with finite-sample conditional coverage guarantees. Conformal prediction (CP) is an increasingly popular framework for building such intervals with distribution-free guarantees, but these guarantees only ensure marginal coverage: the probability of coverage is averaged over both the training and test data, meaning that there might be substantial undercoverage within certain subpopulations. Instead, ideally we would want to have local coverage guarantees that hold for each possible value of the test point’s features. While the impossibility of achieving pointwise local coverage is well established in the literature, many variants of conformal prediction algorithm show favourable local coverage properties empirically. Relaxing the definition of local coverage can allow for a theoretical understanding of this empirical phenomenon. We propose randomly localized conformal prediction (RLCP), a method that builds on localized CP and weighted CP techniques to return prediction intervals that are not only marginally valid but also offer relaxed local coverage guarantees and validity under covariate shift. Through a series of simulations and real data experiments, we validate these coverage guarantees of RLCP while comparing it with the other local conformal prediction methods.

Bonferroni‐Type Tests for Return Predictability With Possibly Trending Predictors

ABSTRACTThe Bonferroni test is widely used in empirical studies investigating predictability in asset returns by strongly persistent and endogenous predictors. Its formulation, however, only allows for a constant mean in the predictor, seemingly at odds with many of the predictors used in practice. We establish the asymptotic size and local power properties of the test, and the corresponding Bonferroni ‐test, under a local‐to‐zero specification for a linear trend in the predictor, revealing that size and power depend on the magnitude of the trend for both. To rectify this, we develop with‐trend variants of the operational Bonferroni and tests. However, where a trend is not present in the predictor, we show that these tests lose (both finite sample and asymptotic local) power relative to the extant constant‐only versions of the tests. In practice, uncertainty will necessarily exist over whether a linear trend is genuinely present in the predictor or not. To deal with this, we also develop hybrid tests based on union‐of‐rejections and switching mechanisms to capitalise on the relative power advantages of the constant‐only tests when a trend is absent (or very weak) and the with‐trend tests otherwise. A further extension allows the use of a conventional ‐test where the predictor appears to be weakly persistent. We show that, overall, our recommended hybrid test can offer excellent size and power properties regardless of whether or not a linear trend is present in the predictor, or the predictor's degrees of persistence and endogeneity. An empirical application illustrates the practical relevance of our new approach.JEL Classifications: C22, C12, G14

Factor momentum versus price momentum: Insights from international markets

Does factor momentum drive stock price momentum? We examine this relationship across 51 countries. Factor momentum proves strong across many markets and international portfolios, independent of typical return predictability drivers. However, its ability to capture stock momentum profits depends on methodological and dataset choices. Empirical factor momentum cannot entirely subsume stock or industry momentum globally. Conversely, price momentum often better explains its factor counterpart than vice versa. Notably, factor momentum based on principal components is more robust, capturing a major share of price momentum gains in developed and emerging markets. Our findings challenge the view that momentum merely times other factors rather than constituting a distinct anomaly.

The tail risk premium in the oil market

This paper studies tail risk and its option-implied risk compensation in the crude oil market. We identify economically large premia for upside and downside tail risks that significantly forecast crude oil futures returns. These premia are also reflected in the convenience yield for physical oil, which amplifies the predictive power for spot returns. Oil tail risk premia are not spanned by aggregate uncertainty measures, suggesting that shifts in market-specific risk attitudes contribute to commodity price volatility and return predictability.

Forecasting crude oil prices: Does global financial uncertainty matter?

In this paper, we introduce an informative uncertainty measure, global financial uncertainty (GFU), for the prediction of crude oil price returns. We find that GFU exhibits significant and remarkable forecasting power for crude oil price returns both in- and out-of-sample with monthly R2 of 13.63% and 11.32%, respectively. This predictive power outperforms and complements those of popular economic variables and uncertainty measures. Further analysis shows that a mean–variance investor can obtain considerable economic gains based on the return forecasts of GFU. By dissecting the GFU’s predictability, we observe that the strong forecasting efficacy of GFU for crude oil price returns may stem from its notable power during high-risk conditions and its significant effects on oil demand dynamics.

Financial regulatory policy uncertainty: An informative predictor for financial industry stock returns

We find that financial regulatory policy uncertainty is an informative indicator for predicting returns of financial industry stocks, outperforming popular predictive variables both in-sample and out-of-sample. Mean-variance investors can achieve substantial economic gains by allocating assets based on the information provided by the financial regulatory policy uncertainty index. Placebo tests suggest that other policy uncertainty indices do not provide predictive information for stock returns of the financial industry, and financial regulatory policy uncertainty cannot forecast stock returns of other industries. We demonstrate that the predictive power of financial regulatory policy uncertainty stems from the cash flow channel, potentially due to the inhibitory effect of uncertainty on firms’ economic activities.

Revisiting time-varying dynamics in stock market forecasting: A Multi-source sentiment analysis approach with large language model

This paper presents the Heterogeneous Dynamic Seemingly Unrelated Regression with Dynamic Linear Models (HD-SURDLM), an innovative framework for stock return prediction that combines cutting-edge sentiment analysis with dynamic financial modeling. The model integrates sentiment data from 2.5 million Twitter posts and various news sources, utilizing state-of-the-art sentiment analysis tools such as VADER, TextBlob, and RoBERTa. HD-SURDLM refines Gibbs sampling for enhanced numerical stability and efficiency while capturing cross-sectional dependencies across multiple assets such as a portfolio. The model consistently outperforms traditional methods like LSTM, Random Forest, and RNN in forecasting accuracy. Empirical results show a 1.02% improvement in 1-day horizon forecasts, a 0.42% gain for 20-day predictions, and a 0.36% increase for 50-day forecasts. By effectively merging public sentiment with dynamic asset modeling, HD-SURDLM offers substantial improvements in short- and long-term prediction accuracy. Its capacity to capture both cross-sectional insights and temporal dynamics makes it an invaluable tool for investors, traders, and financial institutions navigating sentiment-driven markets. HD-SURDLM not only enhances predictive accuracy but also provides a robust decision-support system for financial stakeholders.

Optimal asset allocation and nonlinear return predictability from the dividend-price ratio

AbstractWe study non-linear predictability of stock returns arising from the dividend-price ratio and its implications for asset allocation decisions. Using data from five countries — U.S., U.K., France, Germany and Japan — we find empirical evidence supporting non-linear and time-varying models for the equity risk premium. Building on this, we examine several model specifications that can account for non-linear return predictability, including Markov switching models, regression trees, random forests and neural networks. Although in-sample return regressions and portfolio allocation results support the use of non-linear predictability models, the out-of-sample evidence is notably weaker, highlighting the difficulty in exploiting non-linear predictability in real time.

Institutional trading and satellite data

This study investigates the impact of satellite data availability on the market timing abilities of institutional investors. By analyzing the introduction of satellite coverage for major retailers, this study finds that access to satellite data significantly enhances the return predictability of daily institutional trading. This improvement is more pronounced for stocks with severe information asymmetry and is driven by non-hedge funds. These results suggest that institutions, especially non-hedge funds, actively adjust their trading positions on a daily basis in response to satellite data.

Forecasting U.S. Stock Returns Conditional on Geopolitical Risk and Business Cycles

Using standard predictors in the forecasting literature, we forecast the U.S. stock market returns conditional on geopolitical risk and business cycles over the 1927–2021 period. We find that out-of-sample forecasting performance is significantly better in times of high geopolitical risk versus low geopolitical risk. Consistent with previous research, we find further evidence of improved return predictability in recessions. However, we find little difference in forecasting performance in recessions versus expansions once the level of geopolitical risk is controlled for. We find similar results when using stock market cycles and periods of positive/negative industrial production growth in place of recessions/expansions. Our study contributes to the forecasting literature by documenting that geopolitical risk by itself and in combination with business cycle indicators impacts the forecasting ability of standard forecasting variables in the literature. We also contribute to the literature on the adaptive markets hypothesis with evidence of time-varying return predictability. We find inconclusive evidence as to whether our results are based on time-varying predictability or time-varying risk.

Stock price delay and the cross-section of expected returns: A story of night and day

We examine how individual stock price reacts to intraday and overnight market information. By examining stock betas, we document a stark gap between day beta and night beta and nontrivial asynchronous beta, which captures the slow diffusion of information. We provide evidence that this information delay can predict future stock returns over a one-month period. Long-short portfolios sorted on the gap between day beta and night beta and asynchronous beta generate raw returns of 0.8% and 0.39% and risk-adjusted alphas of 0.77% and 0.28% per month. These results are robust to alternative asset pricing models and when controlling for firm characteristics, such as size, book-to-market ratios, liquidity, investor recognition and limits-to-arbitrage characteristics. We also explore the heterogeneity of the information delay premium and find that the predictive power of information delay for future return is stronger among small, value firms, less visible firms, illiquid firms and firms with higher limit-to-arbitrage. We further provide evidence that our measure of price delay indeed offers incremental information for cross-section return prediction after we explicitly control for the conventional measure of Hou and Moskowitz (2005)'s price delay.