| Summary: | Accurate credit scoring prediction is vital for financial institutions and loan takers. Researchers have long been interested in small and medium sized enterprises (SMEs) due to their distinct characteristics compared to large enterprises and the notable influence of board members and directors. In fact, 99.8% of all enterprises in Europe are defined as SMEs, making a substantial annual contribution to the economy. Since the financial crash of 2008, interest in credit scoring has surged, with unusual data being used to predict defaults more accurately. In this thesis, traditional financial data is enriched with board members’ and directors’ credit scores, using a causal approach called matching to reduce bias and noise. We applied three commonly used machine learning models: logistic regression, decision trees, and random forests. Additionally, we explored the impact of enterprise size in combination with different data sets, enriched by board members’ and directors’ personal credit scores. To evaluate the performance, the metrics AUC, AUPR, F1 and Brier scores were applied. Our results, indicate that SME credit scores can achieve improved accuracy when considering the board and the individuals in control. This project is conducted in collaboration with Credit Info, Iceland. Nákvæmir útreikningar á lánstrausti eru mikilvægir fyrir bæði fjármálafyrirtæki og lántakendur. Rannsakendur hafa lengi haft áhuga á smáum og millistórum fyrirtækjum (SME) vegna einstakra eiginleika í samanburði við stór fyrirtæki sem og áhrifa stjórnenda á rekstur. 99.8% allra fyrirtækja í Evrópu flokkast sem SME og hafa þau þar af leiðandi þónokkur áhrif á hagkerfið. Frá fjármálahruninu 2008 hefur áhugi á lánshæfni aukist verulega, og óhefðbundin gögn hafa verið rannsökuð til að auka nákvæmni matsins. Í þessari ritgerð eru hefðbundin fjármálagögn auðguð með persónulegu lánshæfnismati stjórnarmeðlima sem og framkvæmdastjórnar. Notast er við orsakaaðferð sem kallast “matching” til að minnka hlutdrægni gagnanna.Við notuðum þrjú þekkt líkön ...
|
|---|