Intrinsic capacity as an indicator of the risk of depression and anxiety: a prospective cohort study based on the UK Biobank.

Forecasting corporate earnings plays a vital role in financial decision-making, benefiting a wide array of stakeholders. Financial analysts depend on earnings forecasts to develop reliable investment recommendations, while investors use them to identify stable, high-performing companies (). Lenders evaluate earnings potential to assess credit risk and repayment capacity (). Within organizations, earnings forecasts are critical for strategic planning, enabling management to allocate resources efficiently and pursue growth objectives. Consequently, the demand for accurate and transparent forecasting tools is substantial. These models often rely on indicators such as profitability ratios, cash flow performance, leverage, liquidity, and other accounting metrics. Although recent studies highlight the predictive value of alternative data sources such as textual sentiment, ESG scores, and macroeconomic indicators, this study deliberately focuses on audited financial-statement variables to establish a controlled and regulator-aligned baseline. The proposed TRuE-XAI framework is explicitly designed to be extensible to such alternative data modalities in future work. The current U.S. SEC(United States Securities and Exchange Commission) implementation demonstrates that trustworthy, causally grounded earnings-growth forecasting can be achieved using standardized regulatory data alone, thereby strengthening its applicability in compliance-sensitive financial environments. A natural question is why investors and institutions require earnings forecasts when firms already publish quarterly or annual reports. In practice, capital markets are forward-looking: prices, credit spreads and regulatory assessments depend on expectations of future profitability rather than past realizations. Forward-looking earnings estimates enable investors and lenders to anticipate turning points before they appear in official disclosures, rebalance portfolios, hedge exposures and adjust credit terms proactively. Directional forecasts (increase vs. decrease) are particularly useful for trading and risk management decisions, while forecasts of magnitude support valuation models, scenario analysis and capital planning. Moreover, earnings outcomes are influenced by non-financial drivers such as governance quality, managerial tone, macroeconomic and geopolitical sentiment, supply-chain conditions and ESG indicators. These factors are conceptually important but are not modeled explicitly in this study due to data availability and scope constraints; we instead focus on financial-statement variables as a first, controlled setting, with non-financial extensions identified as future work. Despite substantial progress in earnings forecasting, current approaches remain limited in several ways. Classical statistical and econometric models typically rely on linearity and stationarity assumptions, which restrict their ability to capture nonlinear interactions and regime changes in financial data. Modern machine learning (ML) models–such as Random Forest, gradient boosting and deep networks–address these limitations and often achieve superior predictive performance, but they frequently operate as opaque “black boxes”, offering limited insight into why a particular forecast is issued. Existing explainable AI (XAI) tools, including SHAP and LIME, provide post-hoc feature attributions but do not, by themselves, distinguish between causal drivers and spurious correlations, which is problematic in regulated, high-stakes environments where accountability is essential. Furthermore, most prior work treats prediction, explainability and causal reasoning as separate tasks: causal inference is rarely integrated into the same pipeline as predictive optimization and visual analytics, and comparisons to standard XAI tools are often qualitative rather than systematic.