In the context of the expansion of the new energy vehicle industry and the growth in copper demand, fluctuations in copper prices directly affect the procurement costs and budget stability of enterprises. This paper uses the COMEX copper futures prices from July 20, 2016 to July 10, 2026, the US dollar index, and WTI crude oil prices to construct a rolling forecast for the next 20 trading days, comparing Naive, rolling mean, ARIMA, random forest, HistGradientBoosting, and combined models. Further, using the ARIMA prediction interval and 5,000 Monte Carlo paths, the price uncertainty is converted into the distribution of procurement costs, and the Cost at Risk concept is borrowed to construct the Purchase Cost Risk Value (PCaR). The results show that the complex machine learning models do not outperform the simple benchmark; the MAE of the ARIMA development period is $259.0 per ton, slightly better than the 260.3 per ton of Naive, but the error in the independent test period significantly increases, with the 80% and 95% coverage rates being only 69.5% and 87.6%. Under the assumption of uniform procurement of 1,000 metric tons in the next 20 trading days, the expected cost is approximately $1,392,000, and the 95% PCaR is approximately $93,000, accounting for 6.7% of the expected cost. Based on this, this paper proposes three levels of risk signals (routine, attention, and review), as well as batch procurement, budget buffer, and dynamic manual review mechanisms. The research value lies in connecting short-term price prediction, uncertainty, and procurement budget risk into an interpretable decision support process, rather than pursuing the lowest deterministic price prediction.