Incremental Predictive Value of Management Forward - Looking Disclosures with Respect to Subsequent Fiscal - Year Profitability: Evidence from Chinese A - Share Listed Companies
DOI:
https://doi.org/10.70693/xn2x4s83Keywords:
Forward-looking disclosure(FLD), MD&A, Profitability forecasting, Machine learning, A-share listed firmsAbstract
This study checks if management forward - looking disclosure (FLD) brings any additional predictive power with regard to subsequent fiscal-year profitability—beyond that of ordinary accounting data— for Chinese A-share non-financial firms. FLD is recognized at the sentence level via a TF - IDF logistic classifier; it is measured as the fraction of forward-looking sentences in MD&A. A fixed stratified sample is taken in chronological order and split into training, validation, and test periods. Ridge and XGBoost models compare a purely financial specification against one of identical form but with added FLDIntensity. With regard to the central ROA test and to robustness analyses involving ROE, rolling test years, or non - ST/*ST firms, FLDIntensity fails to yield a consistent improvement in forecast accuracy out-of-sample. That result does not prove that managerial narratives is of no value; it allows for the possibility that a crude prevalence count omits tone or detail present in more complex text forms.