A1241
Title: Conformal inference for multi-horizon forecasting in financial panel data: A block-wise segmentation approach
Authors: Sehee Kim - Inha University (Korea, South) [presenting]
Seonghun Cho - Inha University (Korea, South)
Abstract: The problem of multi-horizon forecasting in financial panel data is investigated, a domain where uncertainty quantification is indispensable due to the high-stakes nature of financial decision-making. While traditional approaches often depend on the validity of underlying model assumptions, modern high-performance models frequently lack formal uncertainty guarantees. To bridge this gap, the conformal prediction framework is employed, a leading approach for distribution-free inference. In particular, a block-wise segmentation strategy that partitions long time series into shorter segments is introduced, integrated with block-wise Z-score standardization to mitigate scale heterogeneity across blocks. Through extensive simulation studies and empirical analysis using U.S. stock market data, it is demonstrated that the proposed methodology, when integrated with diverse predictive models, consistently attains stable joint coverage in multi-horizon forecasting settings.
