Identifying differential transaction patterns in coffee commodity tradingn : a fuzzy clustering analysis of differential risk and financial outcomes
Reuter, Miriam (2025)
Pro gradu -tutkielma
2025
School of Business and Management, Kauppatieteet
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2025061668836
https://urn.fi/URN:NBN:fi-fe2025061668836
Tiivistelmä
This thesis investigated the differential risk and profitability patterns for a coffee trading company using fuzzy k-means clustering. The research was based on a dataset of approximately 17,500 matched sale and purchase transactions. The analysis focuses on the Purchase-, Market- and Sale Differential Results as well as operational variables such as transaction volume, holding period, and net result. These variables reflect the financial performance and risk exposure associated with market dynamics and customer behaviour.
Fuzzy clustering was used to group transactions based on similarities in their risk and profitability characteristics, allowing each transaction to partially belong to multiple clusters. The optimal number of clusters was determined using Modified Partition Coefficient, Fuzzy Silhouette Score and Xie-Beni-Index. The results reveal four distinct transaction clusters, each characterized by different risk-return profiles and trading strategies, ranging from high-margin positioning to low-risk fast execution models. These clusters were further analysed by their association with the different customers and coffee origins. Based on these findings strategy recommendations were formulated which aim to help the company to reduce its risk exposure and increase profitability.
The thesis demonstrates how unsupervised machine learning techniques, specifically fuzzy k-means clustering, can be leveraged for customer analytics and financial risk management in volatile commodity markets.
Fuzzy clustering was used to group transactions based on similarities in their risk and profitability characteristics, allowing each transaction to partially belong to multiple clusters. The optimal number of clusters was determined using Modified Partition Coefficient, Fuzzy Silhouette Score and Xie-Beni-Index. The results reveal four distinct transaction clusters, each characterized by different risk-return profiles and trading strategies, ranging from high-margin positioning to low-risk fast execution models. These clusters were further analysed by their association with the different customers and coffee origins. Based on these findings strategy recommendations were formulated which aim to help the company to reduce its risk exposure and increase profitability.
The thesis demonstrates how unsupervised machine learning techniques, specifically fuzzy k-means clustering, can be leveraged for customer analytics and financial risk management in volatile commodity markets.