Demand forecasting of antarctic krill meal : an automatic model for comparison of time series methods

• Main Authors: Takseth, Miriam Slagnes, Newermann, Tove Fotland
• Other Authors: Guajardo, Mario, Andersson, Jonas
• Format: Master Thesis
• Language: English
• Published: 2019
• Subjects: business analytics; Antarctic; Antarc*; Antarctic Krill
• Online Access: https://hdl.handle.net/11250/2645871

The world’s population is growing faster than ever. As a consequence, it is challenging to maintain a sustainable food production to satisfy all needs. In recent years, krill has emerged as a viable and effective supplement, especially for fish- and animal feed. In an industry characterized by increasing demand and harvesting limitations, it is particularly interesting to investigate whether time series forecasting can be a useful tool to aid effective decision making and long-term strategic planning. Demand forecasting in the krill market is an area in which little previous research is attributed. However, research within related areas such as fisheries harvesting and food production have shown positive results from applying ARIMA and exponential smoothing models. This thesis therefore considers univariate demand forecasting of krill meal for twelve months ahead, applying both of these methods, as well as a combination of decomposition and exponential smoothing. We use historical sales data over a seven-year period from Aker BioMarine as a case study to test the accuracy of the proposed methods. This is done through an automatic model built using R, which chooses the best model from each method based on a variety of criteria. The performance of the models is evaluated using the mean absolute error and the mean absolute scaled error and compared to simple benchmarks. According to our results, the benchmarks seem to perform better than the more complex methods. However, the chosen models from the automatic modeling procedure generally yield a high forecasting error. The provided forecasts should therefore be interpreted by someone with expert knowledge about the krill market and the specific customer, in order to be useful for resource allocation and strategic planning purposes. Since the chosen models do not give satisfying results in terms of forecast error, this opens an opportunity for further research within demand forecasting of krill meal. Keywords – Demand forecasting, time series, krill, krill meal, ARIMA, exponential smoothing, ETS, decomposition, STL nhhmas