Browsing by Author "Kausiala, Roope"
Now showing 1 - 2 of 2
- Results Per Page
- Sort Options
- Forecasting Retail Investor Fund Flow Using Machine Learning
School of Science | Master's thesis(2024-11-16) Kausiala, RoopeRetail investor fund flow is an important concept for financial services companies when determining their liquidity needs. In this thesis, different machine learning methods and types of data are evaluated for the task of forecasting retail investor fund flow. A Linear Regression model from prior research is used as a benchmark model for the rest of the methods. The rest of the methods include classical forecasting methods ARIMA and ETS, as well as Linear Regression, together with a more flexible and state of the art approach Gradient Boosted Trees, and a more recent Deep Learning architecture developed for forecasting, Temporal Fusion Transformer. The research builds on existing literature by considering a linear regression model from prior research. A range of different datasets are comprehensively evaluated for predictive signal and the models are further explained using explainable machine learning technique to determine the data sources useful for prediction. The best models did not beat the baseline predictions on test evaluation with considerable margin. However, the results showed promise on validation set and identified areas which could provide further performance gain on the models. Based on validation results, Temporal Fusion Transformer was determined the best performing model. Fund flow data was determined the most important dataset for prediction, and returns of the funds together with Citigroup index provided further performance gain on validation set, depending on the model used. - SMC ja hylkäysotanta ABC-menetelminä
Perustieteiden korkeakoulu | Bachelor's thesis(2020-04-26) Kausiala, Roope