CEU Electronic Theses and Dissertations, 2020
| Author | Dóró, Róbert Ferenc |
|---|---|
| Title | Modeling Agency CMBS prepayments using Independently Recurrent Neural Networks |
| Summary | Recurrent Neural Network (RNN) architectures have been applied to study and predict financial time series due to their ability to process sequential data. It is well documented that networks that have memory cells can easily outperform basic RNNs. The Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) cells have gained popularity among financial professionals. This paper examines whether the most recent developments in the field of recurrent neural networks allow us to take a step further to leverage such networks to model mortgage prepayment/default events. Namely, Independently Recurrent Neural Networks (IndRNN) will be tested against LSTM structures via publicly available US agency multifamily loan data. The goal of this paper is to assess whether IndRNNs can serve as an alternative to LSTM architectures to model longer term prepayment/default events using monthly data. |
| Supervisor | Szilágyi, Peter G. |
| Department | Economics MSc |
| Full text | https://www.etd.ceu.edu/2020/doro_robert.pdf |
Visit the CEU Library.
© 2007-2021, Central European University