This study presents the effects of adversarial training on irregular time-series. While it has often been leveraged in forecasting, interpretation has been neglected in research, leading to a gap in our understanding of their performance under adversarial conditions. To address this, we conduct an empirical study under time-series forecasting framework. Our findings suggest that 1) existing outperforming methods often produce results that are overly unrealistic, 2) adversarial learning can effectively address this issue, particularly in the underexplored area of irregular time-series. Therefore, we consider forecasting quality in terms of the global distribution - by the mean absolute error (MAE) of the standard deviation between forecasted values and true values - to address the issue of unrealistic and misaligned predictions with human intuition. Along with the discriminator, we also enriched latent information by additional prediction of overall scale and also explored the benefits of incorporating statistical models as auxiliary information in neural models.