Monday, Aug 3: 2:00 PM - 3:50 PM
6043
Contributed Papers
Thomas M. Menino Convention & Exhibition Center
Room: CC-105
Main Sponsor
Business and Economic Statistics Section
Presentations
Cross-sectional stock return prediction is a high-dimensional problem characterized by weak signals, strong collinearity, and heavy-tailed outcomes. In such settings, model performance often depends less on flexibility than on how regularization is imposed. This paper studies the role of economically motivated structure as a regularization device, treating thirteen economic themes as exogenous organizing principles used exclusively to discipline shrinkage rather than to identify new predictive mechanisms. We estimate an additive spline model with a theme-structured Sparse Group LASSO that applies shrinkage both across and within these themes. Using a strict rolling out-of-sample forecasting design on U.S. equities from 2013--2023, conventional forecast metrics such as out-of-sample R^2 and MAE ratios provide limited separation across structured and unstructured penalties. In contrast, structured regularization consistently yields stronger long--short decile portfolio performance --- measured by Sharpe ratio and cumulative return --- where portfolios are formed by sorting stocks into deciles based on predicted returns, analogous to standard anomaly portfolio construction. Theme-level exposures are also more coherent under structured penalties, with results robust to excluding the January 2021 meme-stock episode. The findings suggest that in weak-signal environments, economic structure adds value by disciplining regularization and shaping how predictive weight is allocated --- not by improving statistical forecast accuracy.
Keywords
High-dimensional data and Regularization
Model interpretability / Statistical learning
Sparse Group LASSO
Financial economics (Asset Pricing)
Portfolio construction and performance
Additive models / splines
This paper analyses daily coffee price returns over a two-decade period for 17 varieties across the United States, Germany, and France. We examine the coffee price relationships considering coffee quality, origin, and trade location, using a high-dimensional CoVaR-copula network approach. By exploring CoVaR connectedness, we assess patterns of risk co-movement and potential spillovers, particularly during periods of market stress. Our findings suggest that higher-quality coffees tend to exhibit stronger within-market connections, with distinct clusters emerging across different markets. The United States appears as a central node within the risk network, with notable spillover effects from both Germany and France – likely reflecting its position as the world's largest coffee importer. Additionally, trade location is associated with varying connectedness patterns, with marked differences observed across the US, German, and French markets.
Keywords
Coffee prices
Chemical composition
Tail risk contagion
CoVaR
Copula
Elastic-Net-VAR
The Net Promoter Score (NPS) is widely used to measure customer loyalty, yet theoretical challenges limit its practical application. Using simulations and an unbiased variance estimator, we evaluate coverage and width of Wald, bootstrap t, and adjusted Wald confidence intervals under four population shapes: extreme, triangular, uniform, and left-skewed. With increasing sample size, coverage approached the nominal 95% level for all methods except under the extreme population shape. The adjusted Wald method with triangular or uniform weights was most robust when the population shape was unknown, with comparable interval widths at moderate to large sample sizes. Interval width depended on population shape, and the Wald and bootstrap t methods performed poorly at small sample sizes. These results clarify the sampling behavior of NPS, provide a theoretical basis for unbiased variance estimation, support reliable confidence interval construction, and inform applied use and future methodology.
Keywords
Confidence interval
Sampling distribution
Net Promoter Score (NPS)
Adjusted Wald
This study systematically diagnoses the bubbles and crashes in the Bitcoin. The number of active addresses as a proxy of the number of active users is used in the generalized Metcalfe's Law to evaluate the fundamental values of cryptocurrency market capitalizations. The results show that the market capitalizations of some cryptocurrencies have a statistically significant relationship with the number of active addresses, while the market capitalizations of others are not associated with the number of active addresses, indicating that the generalized Metcalfe's Law is not a universal law for evaluating the basic values of cryptocurrency market capitalizations. Further, we develop a novel bubble diagnosis framework by combining the generalized Metcalfe's Law and the log-periodic power law singularity model to detect the bubble in the cryptocurrency market. This study creates a paradigm for future studies in bubble and crash detection in not only the cryptocurrency market, but also other financial markets.
Keywords
Cryptocurrency Market
log-periodic power law singularity
generalized Metcalfe’s Law
bubble and crash
Bitcoin
active addresses
The Gravity Equation has been the empirical workhorse of the international trade literature. Originally, this technique regressed the logarithm of trade volumes on the logarithm of GDP and other controls. More recently, use of PPML techniques which provides various advantages to the logarithmic counterpart. This paper illustrates some shortcomings of this technique, and evaluate through a simulation study whether an alternative model based on additive and multiplicative effect network models can improve on these limitations.
Keywords
Gravity
AMEN
PPML
Social Networks
International trade
The methods of econometric connectedness-which involve estimating a vector autoregression (VAR) and decomposing the forecast error variance-are popular techniques for measuring financial market integration and shock spillovers between banks, markets, or assets. Previous papers that employ these tools often motivate their results by saying that their connectedness indices can help monitor volatility contagions and serve as an early warning sign of financial crises. In this paper, we test that statement. Namely, can connectedness indices help forecast high volatility and thus serve as an early warning sign? Using a 2-step procedure, we first estimate a VAR over rolling windows using stock market volatility data of 16 countries, which yields a sequence of connectedness indices. In the 2nd step, we re-estimate the VAR but include a connectedness index sequence in the model. Granger causality tests indicate that, yes, connectedness indices can help forecast market volatility. But the improvement in forecast accuracy by including these connectedness indices is relatively minor. Thus, the predictive power of these indices is statistically significant but economically small.
Keywords
Crisis Forecastability
Volatility Prediction
Vector Autoregression
Connectedness
Market Integration
Forecast Error Variance Decomposition
Traditional pairs trading focuses on selecting a pair of individual stocks. However, in practice, it is difficult to identify pairs that may exhibit similar behavior over an extended period. In this work, we take a more drastic approach: paired-portfolio trading, where, using statistical techniques, we carefully construct paired portfolios. We do so by using multivariate techniques such as canonical correlations and cointegration. We apply and illustrate our approach to daily data on ten large-cap U.S. equities from January 2021 to December 2023. We construct a host of paired portfolios from these ten equities, which have a high possibility of cointegrating and thus offer profitable paired-portfolio trading. Our strategy is almost market-neutral, as the net investment in the market is minimal. We develop several entry and exit rules and refine our approach by using a variety of modelling techniques, such as orthogonal regression. We also evaluate the strategy for the future out-of-sample data from January 2024 through November 2025. Our analysis indicates that most of the selected paired portfolios remain profitable in future data, yielding positive returns, often substantial.
Keywords
Canonical Correlation Analysis
Cointegration
Orthogonal Regression
Paired-Portfolio Trading