US Stock Returns Research Articles

Choices Matter When Training Machine Learning Models for Return Prediction

Applying machine learning to cross-sectional stock return prediction requires careful consideration of modeling choices. Common approaches that fail to account for heterogeneity or imbalanced stock representation in training data can lead to suboptimal performance. I study two strategies to address these issues: training group-specific models and predicting relative returns. Both approaches yield similar economic improvements over models trained on the full cross-section of US stock returns, with value-weighted trading strategies benefiting significantly. The findings underscore the importance of aligning machine learning modeling decisions with desired economic outcomes and provide guidance for researchers and practitioners seeking robust machine learning models.

Dynamic partial correlation models

We introduce a new scalable model for dynamic conditional correlation matrices based on a recursion of dynamic bivariate partial correlation models. By exploiting the model’s recursive structure and the theory of perturbed stochastic recurrence equations, we establish stationarity, ergodicity, and filter invertibility in the multivariate setting using conditions for bivariate slices of the data only. From this, we establish consistency and asymptotic normality of the maximum likelihood estimator for the model’s static parameters. The new model outperforms benchmarks like the t-cDCC and the multivariate t-GAS, both in simulations and in an in-sample and out-of-sample asset pricing application to US stock returns.

US efficient factors in a Bayesian model scan framework

PurposeThe author examines the impact these efficient factors have on factor model comparison tests in US returns using the Bayesian model scan approach of Chib et al. (2020), and Chib et al.(2022).Design/methodology/approachEhsani and Linnainmaa (2022) show that time-series efficient investment factors in US stock returns span and earn 40% higher Sharpe ratios than the original factors.FindingsThe author shows that the optimal asset pricing model is an eight-factor model which contains efficient versions of the market factor, value factor (HML) and long-horizon behavioral factor (FIN). The findings show that efficient factors enhance the performance of US factor model performance. The top performing asset pricing model does not change in recent data.Originality/valueThe author is the only one to examine if the efficient factors developed by Ehsani and Linnainmaa (2022) have an impact on model comparison tests in US stock returns.

Intangible Assets and US Stock Returns: An analysis using the Index Method, Panel Regression, and Machine Learning

This study examines the impact of intangible assets on stock returns in the U.S. using the Drucker Institute indices, which assess companies based on customer satisfaction, employee engagement and development, innovation, social responsibility, and financial stability. The relevance of this study lies in the growing importance of considering non-financial indicators in investment decision-making. The objective is to determine how these indices affect stock returns across different sectors. The hypotheses posit that each index has a positive impact. The study employs both panel regression with fixed effects and machine learning methods using XGBoost with Shapley values to analyze data from U.S. companies for the period from June 30, 2016, to June 30, 2023. The results indicate that social responsibility has a broadly positive impact on stock returns across various sectors. Innovation significantly affects returns only in the technology sector. Customer satisfaction and financial stability exhibit varying effects depending on the sector, while employee engagement and development show only negative impacts in the energy sector. The significance of this research lies in its contribution to understanding the role of intangible assets in shaping stock performance. We show that investors can achieve both ethical satisfaction and higher financial returns by prioritizing investments in companies with strong social responsibility records. Additionally, we draw the attention of investors and researchers to the importance of considering sectoral affiliation when analyzing companies. The use of advanced analytical tools, such as XGBoost with Shapley values, underscores the potential of machine learning in uncovering complex relationships in financial data. This approach proves to be highly promising for future research.

The historical lepto-variance of the US stock returns

<abstract> <p>Regression trees (RT) involve sorting samples based on a particular feature and identifying the splitting point that yields the highest drop in variance from a parent node to its children. The optimal factor for reducing mean squared error (MSE) is the target variable itself. Consequently, employing the target variable as the basis for splitting sets an upper limit on the reduction of MSE and, equivalently, a lower limit on the residual MSE. Building upon this observation, we define lepto-regression as the process of constructing an RT of a target feature on itself. Lepto-variance pertains to the portion of variance that cannot be mitigated by any regression tree, providing a measure of inherent variance at a specific tree depth. This concept is valuable as it offers insights into the intrinsic structure of the dataset by establishing an upper boundary on the "resolving power" of RTs for a sample. The maximal variance that can be accounted for by RTs with depths up to <italic>k</italic> is termed the sample k-bit macro-variance. At each depth, the overall variance within a dataset is thus broken into lepto- and macro-variance. We perform 1- and 2-bit lepto-variance analysis for the entire US stock universe for a large historical period since 1926. We find that the optimal 1-bit split is a 30–70 balance. The two children subsets are centered roughly at −1% and 0.5%. The 1-bit macro-variance is almost 42% of the total US stock variability. The other 58% is structure beyond the resolving power of a 1-bit RT. The 2-bit lepto-variance equals 26.3% of the total, with 42% and 47% of the 1-bit lepto-variance of the left and right subtree, respectively.</p> </abstract>

ESG news and long-run stock returns

This paper investigates the persistence of ESG news' effects on US stock returns. Utilizing a rolling-portfolio approach to calculate overlapping returns of news-based portfolios, extending up to twelve months post-news release, we find that market reactions to ESG news are fleeting and reversing within a month in the most recent sample period. This suggests ESG news has no enduring influence on stock returns. Notably, the reactions are prominent for social and governance themes but not for environmental-related news. This study is one of the first attempts to understand how ESG news can affect a company's value over the long term.

The Impact of the Russo-Ukraine War on US Stock Return

This paper makes an event study on the Russo-Ukrainian War effect of the stock return, including that of the S&P500, renewable energy industry, military industry, and catering industry, with a time window of 29 days (39 days for Teslas event). This article will gain the normal and abnormal returns by using linear regression and the least square method. Then, the hypothesis test will be conducted using the Summation of the abnormal rate of return (CAR). The result will be shown by whether the cumulative abnormal return breaks the confidence interval according to the graph. Research results show that even though the war harms the economy, it does not entirely reduce the stock return. This empirical study would be helpful for the investors to make their investment strategy during the War period.

Precious Metals Comovements in Turbulent Times: COVID-19 and the Ukrainian Conflict

We examined the evolution of cross-market linkages between four major precious metals and US stock returns, before (Phase I) and after (Phase II) the COVID-19 outbreak. Phase II was also extended to encompass the Ukrainian conflict, which prolonged the period of uncertainty in financial markets. Due to the increase in volatility observed in Phase II, we used a heteroskedasticity-adjusted correlation coefficient to examine the evolution of correlation changes since the COVID-19 outbreak. We also propose a relevant dissimilarity measure in multidimensional scaling analysis that can be used for depicting associations between financial returns in turbulent times. Our results suggest that (i) the correlation levels of gold, silver, platinum, and palladium returns with US stock returns have not changed substantially since the COVID-19 outbreak, and (ii) all precious metal returns exhibit movements that are less synchronized with US stock returns, with palladium and gold being the least synchronized.

On IVX-based structural break tests in univariate predictive regressions

ABSTRACT Testing for structural break in predictive regressions is important for empirical finance. In this article, we contribute to the literature by exploring asymptotic and finite-sample properties of three tests built upon IVX estimator proposed by Kostakis, Magdalinos and Stamatogiannis. Although they are shown to be non-pivotal, simulation studies suggest that limit distributions are highly insensitive to the variation of nuisance parameters. Furthermore, satisfactory small-sample size control and power performance can be achieved by applying critical values already tabulated in the literature. These tests are therefore easy to implement for practitioners and can serve as handy complements to other more sophisticated approaches. An empirical application concerning the break in the predictability of US stock returns is provided as an illustration.

Time to Retire: The 4% Withdrawal Rule

The so-called “<i>4% rule</i>” is perhaps the single most commonly referenced retirement withdrawal approach. Most discussion of this rule centers on the appropriate percentage to use, 4%, 3%, or 5%. My objective is to suggest a superior retirement rule. Superior in the sense that it will deliver a higher standard of living during one’s retirement years. I start with the observation that past retirement withdrawal analysis suffers from two key deficiencies, that is, a reliance on past US stock and bond market returns (ex-post cherry picking the best performing market) and an assumption that asset class returns can be modeled with independent and identically distributed random variables, serving to mask their important time series properties. This article repairs these two deficiencies. A new retirement distribution rule is proposed. One that distributes/liquidates an ever-increasing proportion of the retiree’s then-current portfolio as they age, but subject to a minimum monthly distribution expressed in dollar terms. This approach provides simplicity, ease of execution, and transparency. It also results in a standard of living for the retiree as expressed by the median and average monthly distributions that are 101.3% and 174.2%, respectively, above those delivered by the best possible constant-dollar withdrawal rule. This pleasing result occurs for two primary reasons. First, this new rule realizes significant benefit from selling more/less shares when markets are high/low (which is the exact opposite of what transpires with all constant-dollar rules). Or to use other words, it delivers a measurably higher IRR despite utilizing an identical asset mix. Second, the new rule effectively consumes the retiree’s entire portfolio over the retirement life, leaving no unused residual balance (again, the exact opposite of all constant-dollar rules).

The predictive power of Bitcoin prices for the realized volatility of US stock sector returns

This paper is motivated by Bitcoin’s rapid ascension into mainstream finance and recent evidence of a strong relationship between Bitcoin and US stock markets. It is also motivated by a lack of empirical studies on whether Bitcoin prices contain useful information for the volatility of US stock returns, particularly at the sectoral level of data. We specifically assess Bitcoin prices’ ability to predict the volatility of US composite and sectoral stock indices using both in-sample and out-of-sample analyses over multiple forecast horizons, based on daily data from November 22, 2017, to December, 30, 2021. The findings show that Bitcoin prices have significant predictive power for US stock volatility, with an inverse relationship between Bitcoin prices and stock sector volatility. Regardless of the stock sectors or number of forecast horizons, the model that includes Bitcoin prices consistently outperforms the benchmark historical average model. These findings are independent of the volatility measure used. Using Bitcoin prices as a predictor yields higher economic gains. These findings emphasize the importance and utility of tracking Bitcoin prices when forecasting the volatility of US stock sectors, which is important for practitioners and policymakers.

A flexible predictive density combination for large financial data sets in regular and crisis periods

A flexible predictive density combination is introduced for large financial data sets which allows for model set incompleteness. Dimension reduction procedures that include learning allocate the large sets of predictive densities and combination weights to relatively small subsets. Given the representation of the probability model in extended nonlinear state-space form, efficient simulation-based Bayesian inference is proposed using parallel dynamic clustering as well as nonlinear filtering, implemented on graphics processing units. The approach is applied to combine predictive densities based on a large number of individual US stock returns of daily observations over a period that includes the Covid-19 crisis period. Evidence on dynamic cluster composition, weight patterns and model set incompleteness gives valuable signals for improved modelling. This enables higher predictive accuracy and better assessment of uncertainty and risk for investment fund management.

The Historical Lepto-Variance of the US Stock Returns

The Historical Lepto-Variance of the US Stock Returns

Strategic approach to analyze the effect of Covid-19 on the stock market volatility and uncertainty: a first and second wave perspective

PurposeThis paper aims to analyze the impact of Covid-19 on the stock market volatility and uncertainty during the first and second waves.Design/methodology/approachThis study has applied event study and autoregressive integrated moving average models using daily data of confirmed and death cases of Covid-19, US S&amp;P 500, volatility index, economic policy uncertainty and S&amp;P 500 of Bombay Stock Exchange to attain the purpose.FindingsIt is observed that, during the first wave, the confirmed cases and the fiscal measure have a significant impact, while the vaccination initiative and the abnormal hike of confirmed cases have a significant impact on the US stock returns during the second wave. It is further observed that the volatility of Indian and US stock markets spillovers during the sample period. Moreover, a perpetual correlation between the Covid-19 and the stock market variables has been noticed.Research limitations/implicationsAt present, the world is experiencing the third wave of Covid-19. This paper has considered the first and second waves.Practical implicationsIt is expected that business leaders, stock market regulators and the policymakers will be highly benefitted from the research outcomes of this study.Originality/valueThis paper briefly highlights the drawbacks of existing policies and suggests appropriate guidelines to successfully implement the forthcoming initiatives to reduce the catastrophic impact of Covid-19 on the stock market volatility and uncertainty.

How Have ETFs Changed Market Macro Efficiency and Risk Structure?

In the late 1990s, Samuelson offered a dictum suggesting that the equity market, although micro efficient, is also macro inefficient. That is, at the aggregate level, the market does not accurately reflect the available information. We further explore this dictum in the context of a market that has evolved to provide larger options for trading across the whole spectrum of investors. We argue that the macro efficiency of the stock market has been increasing over the past 20 years. We show that the development and adoption of exchange-traded funds (ETFs) have played a major role in the transmission of information to the market, resulting in an increase in market macro efficiency. Our approach is based on the R-squared of the market model explaining US stock and sector returns. First, we propose these R-squared values as a proxy for market macro efficiency. Then, we use regression analyses at the individual stock and sector levels to observe the relationship between the rise in ETFs and market macro efficiency. Finally, we use extrafinancial data related to the awareness and adoption of ETFs to confirm their contribution.

Predicting the equity market risk premium: A model selection approach

We present a novel approach to investigate US stock return predictability. Our analysis utilizes the best subset method to construct a single predictive regression from a set of fundamental factors and hence, it is robust to data snooping. We consider models with non-Gaussian distributions as a first in the literature. We find that our selected predictive regression is Student’s t-distributed and has both in sample and out of sample forecasting power, with a high degree of economic significance.

Robust Estimation of Conditional Factor Models

This paper develops estimation and inference methods for conditional quantile factor models. We first introduce a simple sieve estimation, and establish asymptotic properties of the estimators under large $N$. We then provide a bootstrap procedure for estimating the distributions of the estimators. We also provide two consistent estimators for the number of factors. The methods allow us not only to estimate conditional factor structures of distributions of asset returns utilizing characteristics, but also to conduct robust inference in conditional factor models, which enables us to analyze the cross section of asset returns with heavy tails. We apply the methods to analyze the cross section of individual US stock returns.

Time‐varying neural network for stock return prediction

AbstractWe consider the problem of neural network training in a time‐varying context. Machine learning algorithms have excelled in problems that do not change over time. However, problems encountered in financial markets are often time varying. We propose the online early stopping algorithm and show that a neural network trained using this algorithm can track a function changing with unknown dynamics. We compare the proposed algorithm to current approaches on predicting monthly US stock returns and show its superiority. We also show that prominent factors (such as the size and momentum effects) and industry indicators exhibit time‐varying predictive power on stock returns. We find that during market distress, industry indicators experience an increase in importance at the expense of firm level features. This indicates that industries play a role in explaining stock returns during periods of heightened risk.

Oil price shocks and the US stock market: A nonlinear approach

We study the response of US stock market returns to oil price shocks and to what extent it behaves asymmetrically over the different phases of the business cycle. For this purpose, we decompose the oil price changes into supply and demand shocks in the oil market and assess the state-dependent dynamics of structural shocks on US stock returns using a smooth transition vector autoregression model. When nonlinearity is considered, quantitatively very different asymmetric dynamics are observed. Our findings show that the responses of US stock returns to disaggregated shocks are asymmetric over the business cycle and that the impact of demand-driven shocks on US stock returns is stronger and more persistent, especially when economic activity is depressed. Furthermore, the contribution of shocks to expectation-driven precautionary demand in recessions accounts for a larger share of the variability of US stock market returns than that predicted by standard linear vector autoregressions. • State-dependent effect of oil price shocks on US stock returns is investigated. • We decompose oil price changes into supply and demand shocks. • A smooth transition VAR is used for asymmetric dynamics over the business cycle. • Demand-driven shocks are larger and more persistent during recessions. • Oil price shocks explain around 40% for US stock market variations in recessions.

US Stock return predictability with high dimensional models

We examine the role of large information sets in the predictability of US stock using a large data set of over 400 predictors covering macro-, financial-, trade- and commodity-related variables over the period of 1960:Q1 to 2018:Q4. We consider 13 alternative models ranging from autoregressive models with no predictors to 5-factor, 60-factor and high dimensional models with over 400 predictors including assumptions of constant and time varying coefficients. We find that models that incorporate large predictors improve US stock return predictability. The outcome particularly favours models involving Dynamic Variable Selection prior with Variational Bayes (VBDV) for density forecast.