University of Twente Student Theses
The application of hidden Markov models for economic state prediction and the calculation of economic state transition probabilities
Rikken, T.W. (2022) The application of hidden Markov models for economic state prediction and the calculation of economic state transition probabilities.
|
PDF
4MB |
| Abstract: | The International Financial Reporting Standards (IFRS) require financial institutions to estimate potential credit losses with a forward-looking view. Most financial institutions (77\%) are taking a scenario-based approach to include forward-looking macro-economic impact in their estimations of potential credit risk losses. The weights of the three scenarios used by most financial institutions are often quite basic where the most likely scenario (baseline) accounts for 50\% and the remaining two (upside and downside) share the other 50\% equally. These weights are currently not determined by any quantitative method, hence, this research aims to determine these weights based on a quantitative method. To meet this goal we used several hidden Markov models. We experimented with the number of hidden states, the amount of historical input-data used, the initialization of the hidden Markov model, cutoff points for extreme outliers, and the features used to predict the economic states, to evaluate the best performance of the hidden Markov models. In short, we can conclude that the two-state, and especially the three-state higher-order hidden Markov models can predict economic states. The three-state model outperforms all other hidden Markov models and viable benchmarks with an average accuracy of 0.85 and an average F-score of 0.68. Lastly, the economic state transition probabilities (weights) calculated by the HMM are not in line with the weight distribution currently used by most financial institutions (0.5, 0.25, 0.25). This deviation indicates that the current method is not optimal and that it is worth to research the impact of using the weights calculated by the HMM for the expected credit loss calculations. |
| Item Type: | Essay (Master) |
| Clients: | EY |
| Faculty: | BMS: Behavioural, Management and Social Sciences |
| Subject: | 31 mathematics, 54 computer science, 83 economics |
| Programme: | Industrial Engineering and Management MSc (60029) |
| Link to this item: | https://purl.utwente.nl/essays/92468 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page