Volatility Dynamics of NVIDIA Stock During the AI Boom: An ARIMA-Based Anomaly Detection Approach

This paper uses a model of price fluctuations and anomalies detection based on a time-series fashion to examine the price behavior of the NVIDIA (NVDA) stock within the context of the recent AI-based market boom. The 20192025 functions are the weekly closing prices, which are converted to weekly log returns, and an ARIMA model is determined using an estimated training sample, aiming to overfit linear dependence in the returns. Residual values are normalized to standardized values, whereby the abnormal weeks are determined using a z-score. In order to measure regime level differences, we divide the sample by January 1, 2023, which is generally agreed upon as a point of inflection in the trend of price variability, related market sentiment, and compare pre- and post-cutoff statistics on price variability-related statistics. The findings reveal that ARIMA may be used to give a base characterization of characteristics of returns, but the performance of the forecast reduces as market moves come to a sudden change, which is in line with the evidence of the existence of non-linear shocks and structural changes. Anomaly detection using residuals identifies a few outlier weeks that are consistent with increased uncertainty and blistering repricing. The sensitivity analysis of the alternative thresholds argues that the frequencies of shock change in their direction but remain constant in magnitude, which validates the strength of the anomaly identification process. Generally speaking, this work shows that using ARIMA modelling and simple statistical detection rules is an interpretable tool that can be used to detect instances of abnormal price changes in technology shares in times of fast narrative-driven repricing, hence it can be used in real-time risk detection in the technology markets.