Coupling Macro-Sector-Micro Financial Indicators for Learning Stock Representations with Less Uncertainty

While the stock movement prediction has been intensively studied, existing work suffers from weak generalization because of uncertainty in both data and modeling. On one hand, training a representation on stochastic an end-to-end manner may lead to excessive modeling, which involves model uncertainty. other, analysis correlating with its relevant factors To simultaneously address such modeling perspectives, fundamental yet challenging task is learn better less by considering hierarchical couplings macro-level sector-and micro-level. Accordingly, we propose copula-based contrastive predictive coding (Co-CPC) method. Co-CPC first models dependence between certain sector macroeconomic variables that are sequential heterogeneous, e.g., macro-variables associated different time intervals, scales, distributions. Then, involving macro-sector context, representations learned self-supervised way can further be used for downstream tasks like prediction. Extensive experiments two typical datasets verify effectiveness our