Publishing media stock prediction based on TransXFormer

Amid global economic restructuring and China's "cultural power" strategy, the publishing and media industry is transforming drastically. Traditional print media decline, fierce content platform competition, capital withdrawal and policy shifts bring systemic valuation fluctuations and capital desensitization risks. Listed firms' stock prices, key for industry monitoring, have financial time series with high non-linearity hard to capture via existing methods, causing forecast errors and risk misjudgments, a bottleneck for cultural-financial integration. To address these challenges, a TransXFormer end-to-end prediction model is proposed, which achieves collaborative modeling of long-memory structures and global attention via architectural innovation, realizes endogenous cross-scale fusion of macro fundamentals and multi-view trading data via the MPA-Macro Boost mechanism, and effectively captures long-range dependencies and dynamic uncertainty in financial time series via uncertainty-aware and temporal constraint mechanisms. In the stock price prediction task for two publishing and media ETF datasets and five publishing and media listed company datasets, end-to-end point prediction and interval prediction of listed companies' stock prices were achieved. Although the level-based R 2 does not constitute direct evidence of return predictability, it nevertheless provides a measure of the model's short-horizon price-level predictive performance. Under this interpretation, the average R 2 across the seven prediction tasks was 90%, with the highest value of 98.33% obtained on the 2007-2025 dataset for the 601999 Publishing and Media stock.