Forecasting the Behavior of Peruvian Coffee Export Prices in International Markets Using Econometric Models, 2010–2025

High coffee price volatility increases uncertainty regarding harvest profitability and marketing price setting, affecting both coffee growers and producing and exporting firms. This study aims to analyze and forecast the unit export price of Peruvian coffee using monthly data for the 2010–2024 period, with projections for 2025, in order to provide an applied decision-support tool. To this end, six econometric and time-series specifications were compared: a linear model, a quadratic model, additive Holt–Winters exponential smoothing, a benchmark causal model, a lagged model, and an ARIMA model. Model performance was assessed using goodness-of-fit and diagnostic criteria, including the coefficient of determination (R²), the Akaike Information Criterion, and the Durbin–Watson statistic, complemented by forecast error metrics for the exponential smoothing model. The results show that the linear and quadratic models exhibit non-significant trend terms and residual autocorrelation. The Holt–Winters model reported a root mean squared error of 5.45, although it did not show a robust contribution of trend and seasonal components. Likewise, the models based on FOB value and net weight displayed low explanatory power and persistent autocorrelation. In contrast, the ARIMA(1,0,0) model showed the most consistent statistical performance, with the lowest AIC (6.220984) and a Durbin–Watson statistic close to 2.00, suggesting the absence of residual autocorrelation. Its forecasts indicate stabilization from March 2025 onward, with values converging around 8.15. Overall, the evidence supports ARIMA as an operational tool for price monitoring and export planning. As a future research agenda, this approach could be complemented with GARCH models to capture volatility and contrasted with nonlinear methods under out-of-sample validation schemes.咖啡价格的高波动性会增加收获期盈利能力与定价决策的不确定性，对咖啡种植户、生产企业及出口企业均造成影响。本研究旨在借助2010-2024年的月度数据，分析并预测秘鲁咖啡的单位出口价格，并对2025年进行展望，以期提供一款实用的决策支持工具。为此，本研究对比了六种计量与时序模型设定：线性模型、二次型模型、加法型Holt-Winters指数平滑法（Holt–Winters Exponential Smoothing）、基准因果模型、滞后模型以及ARIMA模型。模型性能通过拟合优度与诊断准则进行评估，涵盖决定系数（R²）、赤池信息准则（Akaike Information Criterion）以及德宾-沃森统计量，并辅以指数平滑模型的预测误差指标。结果显示，线性模型与二次型模型的趋势项不具备统计显著性，且存在残差自相关问题。Holt-Winters模型的均方根误差为5.45，但其趋势与季节成分并未表现出显著的正向贡献。同样，基于离岸价（FOB）与净重构建的模型解释力较低，且存在持续的自相关问题。与之形成对比的是，ARIMA(1,0,0)模型展现出最稳定的统计性能：其赤池信息准则值最低（6.220984），德宾-沃森统计量接近2.00，表明不存在残差自相关问题。该模型的预测结果显示，2025年3月起价格将趋于平稳，最终收敛至约8.15的水平。总体而言，研究结果表明ARIMA模型可作为价格监测与出口规划的实用工具。作为未来研究方向，可通过引入广义自回归条件异方差模型（GARCH）以捕捉价格波动性，并结合非线性方法在样本外验证框架下开展对比研究。