In briefDiversifying Across Predictors: Mixed-Model Ensembles for Stock Return Forecasting
SummaryThis paper studies cross-sectional stock return prediction and proposes a heterogeneous ensemble that bags linear and nonlinear predictors so their errors partly cancel. The authors use a bias-variance-covariance decomposition to show how predictor aggregation parallels portfolio diversification. On CSI300 daily Alpha158 factor data, the ensemble raises Rank IC by over 13% versus standalone neural networks and outperforms tree-based methods such as XGBoost and Double Ensemble.
In briefPortfolio risk measures that combine several risk models to reduce reliance on one.
SummaryThis paper proposes a Diversified-Risk-Perception (DRP) framework that diversifies both assets and risk models to reduce dependence on a single probability model. It introduces DRP expected shortfall (DRP-ES), establishes its theoretical properties, and derives bounds on its sensitivity to model parameters. Empirical tests in U.S., Chinese A-share, and South Korean markets report that DRP-ES-based strategies achieve higher Sharpe ratios and lower maximum drawdowns than market benchmarks.
In briefAdjoint kernel sensitivity of the rough Heston characteristic exponent
SummaryThis paper derives an adjoint representation for the first-order sensitivity of the rough Heston characteristic exponent with respect to its memory kernel, holding model parameters and the input variance curve fixed. A continuous sensitivity kernel, obtained from a backward linear Volterra equation, gives the Fréchet derivative with respect to the kernel. On the structural kernel class, which includes multi-factor approximations, the paper establishes Hadamard directional differentiability and a first-order expansion with a quadratic remainder. Numerical experiments at multi-factor reference kernels are included.
Carlos Escudero, Francisco J. Luguera-Par\'uas, Miguel Sama
