ARIMA-driven memory market insights: Forecasting DRAM spot price

Abstract

The semiconductor sector is a vital cornerstone of Taiwan's economy, pivotal in bolstering the nation's global technology prowess. Dynamic Random Access Memory (DRAM) stands out among its various outputs. However, the price of DRAM exhibits significant volatility, leading to substantial financial fluctuations for manufacturers in the semiconductor sector. This unpredictability poses a considerable challenge, placing undue strain on their financial stability.

Hence, this study aims to establish a quantitatively-based prediction model departing from conventional industry heuristics. Empirical findings reveal that DRAM spot prices exhibit non-stationary time series characteristics, prompting the development of an ARIMA model to capture their price dynamics. Furthermore, we enriched the original ARIMA model by incorporating four additional variables: Hynix DSI, Micron DSI, European PMI, and US PMI, resulting in a more robust ARIMAX model with enhanced explanatory power for predicting DRAM prices.

Our analysis demonstrates the ARIMAX model's effectiveness in explaining and predicting DRAM prices. When combined with the Rolling prediction method, the final predicted values closely align with actual outcomes. Our prediction model promises to inform future DRAM purchasing decisions within the company, potentially yielding cost savings and alleviating inventory pressures. In the subsequent scenario analysis, it was observed that implementing procurement strategies using this prediction model effectively reduced costs.