Forecasting GDP growth with stock returns: Time-series or cross-sectional information?

This paper investigates whether the predictive content of stock-market information for macroeconomic activity reflects high-frequency time-series dynamics or cross-sectional aggregation. Using a factor-augmented mixed-data sampling (MIDAS) framework applied to Japan, we find that aggregate market indices and high-frequency variation provide limited forecasting gains, whereas factor-based predictors extracted from large cross-sections of individual stock returns can improve forecast accuracy relative to an autoregressive benchmark. Overall, the results suggest that the informational value of stock prices for GDP forecasting arises primarily from effective cross-sectional aggregation rather than from higher-frequency variation. • Stock returns contain useful information for forecasting Japan’s GDP growth. • Forecast gains are selective and concentrated at the one-quarter horizon. • MIDAS yields little extra gain over quarterly stock-return specifications. • Factor-based models outperform models based on aggregate stock indices. • Cross-sectional aggregation matters more than high-frequency variation.