Thu Sep 26 14:00:00 UTC 2024: ## New Hybrid Model Accurately Predicts Crop Prices, Providing Insights for Supply and Demand Control

**Naju, South Korea** – A team of researchers from the Korea Rural Economic Institute, Korea University, and Jeonnam Agricultural Research & Extention Services has developed a novel hybrid model, Parametric Seasonal-Trend Autoregressive Neural Network (PaSTANet), for long-term crop price forecasting. This groundbreaking model combines the strengths of both statistical and deep learning approaches, outperforming existing methods in predicting onion, radish, Chinese cabbage, and green onion prices.

The research, published in PLOS ONE, addresses the challenges of forecasting crop prices, which are influenced by factors such as weather, growing conditions, yields, and demand fluctuations. These uncertainties pose economic risks to both producers and consumers.

“Long-term forecasting is crucial for regulating supply and demand, stabilizing the agricultural economy, and enhancing the efficiency of agricultural markets,” explains lead author Dr. Won Hong.

PaSTANet combines two modules:

* **Multi-Kernel Residual Convolution Neural Network (MRCNN):** Utilizes previous price information and different kernels to capture weekly and monthly price variations.
* **Gaussian Seasonality-Trend (GaST):** Learns trend and seasonality models through piecewise regression and Fourier series analysis, reflecting monthly, weekly, and daily volatility.

The model then estimates the parameters of a Gaussian distribution for each module and combines them to predict crop prices with a confidence interval (CI). This CI enables classification into three zones:

* **Predictability:** The price is expected to remain within a stable range.
* **Caution:** There is a potential for moderate price volatility.
* **Warning:** A significant price fluctuation is anticipated.

The researchers tested PaSTANet using daily data from the Garak market in Seoul from 2010 to 2023. For onion prices, the model achieved the best performance across various metrics, surpassing the second-best model (Prophet) by 29.6% in mean absolute error (MAE). PaSTANet also showed significant improvements for the other three crops.

Furthermore, the researchers developed an Abnormal Price Detection System (APDS) based on the CI, which helps to predict and anticipate price swings. This tool provides a valuable basis for agricultural supply and demand control strategies.

“PaSTANet’s ability to forecast price distributions with confidence intervals and identify potential price fluctuations offers valuable insights for agricultural policymakers and market participants,” states Dr. Hong.

While PaSTANet demonstrates impressive accuracy, the researchers emphasize the need for further research to incorporate exogenous variables such as environmental conditions and policy changes. Future work will also aim to expand the model to include other crops, considering complementary and substitutive relationships between prices.

The development of PaSTANet represents a significant step toward more accurate and reliable crop price forecasting, empowering stakeholders to make informed decisions and manage risks effectively.

Read More