The Test of ARIMA's Stock Price Prediction Capability in Stable and High-Variability Scenarios - A Comparative Study Based on Guizhou Moutai and NVIDIA

Abstract

The stock market is the "barometer" of the economy. Stock price forecasting is core focus of investors. Time series models like ARIMA are widely used for this purpose. However, the stock market’s high volatility challenges the model’s universality. This study selects two stocks with distinct price characteristics: Kweichow Moutai (relatively stable stock prices) and NVIDIA (significant price fluctuations driven by major positive news) as research objects. ARIMA is used to predict their price. RMSE is used to assess how accurate the predictions are. Major events that happened to the two companies are examined to explain prediction deviations. Results show that when predicting Kweichow Moutai’s stock price, the ARIMA model achieves high accuracy. However, when it comes to the task of predicting NVIDIA’s stock price, a striking and persistent gap emerges between the forecasted figures and the actual market outcomes. This pronounced discrepancy underscores a critical weakness in the models employed: they appear to falter precisely when confronted with sudden, high-impact shocks—such as unexpected regulatory announcements, geopolitical flashpoints, or abrupt shifts in consumer demand—that lie outside the historical patterns on which their parameters were trained.