Research on World Agricultural Economy

Improving Accuracy in Agricultural Forecasts: Evaluation of SARIMA and Forecast Models for Artichoke Prices

Axel Zevallos

Cesar Vallejo University, Trujillo, Peru

Preferred Public Name

Ricardo Palma University, Lima, Peru

Karen Palomino-Salcedo

Universidad Internacional de La Rioja (UNIR), Logroño, Spain;

Rosa Alejandra Salas-Bolaños

Cesar Vallejo University, Trujillo, Peru

DOI: https://doi.org/10.36956/rwae.v6i3.1880

Received: 21 March 2025; Published: 15 July 2025

Abstract

The volatility of agricultural prices in Peru poses a critical challenge to food security and economic stability, particularly for key crops such as artichoke, whose price fluctuations directly affect producers’ incomes and the accessibility of food for the population. This study evaluates the accuracy of several time series forecasting models—namely, SARIMA, ARIMA, Holt-Winters, Holt, and Naive—in predicting the average monthly price of artichoke in Lima, using historical data collected between January 2021 and December 2024. A comprehensive methodological approach was implemented that combines automated parameter optimization (using the Akaike Information Criterion, AIC), a 12-month retrospective validation, and the assessment of percentage errors against actual price values observed in January and February 2025. The results indicate that the SARIMA model ((0,0,1)(0,1,0),12) achieved the lowest average error of 12.16%, and it demonstrated exceptional accuracy in February, with only a 5.62% deviation. This superior performance is attributed to its ability to capture complex seasonal patterns inherent in the data. In contrast, the Holt-Winters model exhibited the poorest performance, recording an average error of 17.41% and a particularly high error of 32.99% in February, which underscores its limitations in managing nontraditional seasonal fluctuations. Additionally, while the Naive model proved highly accurate for very short-term forecasts in January (0.50% error), it was found to be unsuitable for extended forecasting horizons, as evidenced by a 28.94% error in February. Residual analysis further confirmed that SARIMA generates more robust predictions, with residual correlations that closely approximate white noise.

Keywords: Agricultural Forecasting, Artichoke Prices, Time Series Analysis, SARIMA, Python

References