Abstract：As global warming increases significantly, the forecast of carbon dioxide emissions became the key point for worldwide carbon reductions. The traditional carbon emissions prediction model is based on co-frequency data. Transferring high-frequency data to low-frequency data not only ignores the effective information with it but also influences the timeliness and reduces the accuracy of prediction. This paper uses mixed-data sampling (MIDAS) to predict carbon emissions and analyzes the effect of high-frequency quarterly GDP hysteresis and low-frequency carbon emissions. According to the research, quarterly GDP has both positive and negative impacts on carbon emissions and positive effects dominate over six quarters. Besides that, there are mutual influences between carbon emissions and these influences will last about four years, which coincide with Chinese economy and illustrates the rationality of the mixed-data sampling (MIDAS) for carbon dioxide emissions. In addition, mixed-data sampling (MIDAS) has accurate short-term prediction in carbon dioxide emissions and has significant feasibility and timeliness in real-time forecasting.