On the use of case estimate and transactional payment data in neural networks for individual loss reserving
The use of neural networks trained on individual claims data has become increasingly popular in the actuarial reserving literature. We consider how to best input historical payment data in neural network models. Additionally, case estimates are also available in the format of a time series, and we extend our analysis to assessing their predictive power.
In this paper, we compare a feed-forward neural network trained on summarised transactions to a recurrent neural network equipped to analyse a claim’s entire payment history and/or case estimate development history. We draw conclusions from training and comparing the performance the models on multiple, comparable highly complex datasets simulated from SPLICE (Avanzi, Taylor & Wang, 2023). We find evidence that case estimates will improve predictions significantly, but that equipping the neural network with memory only leads to meagre improvements. Although the case estimation process and quality will vary significantly between insurers, we provide a standardised methodology for assessing their value.
See main for the files submitted as part of the paper. The masters-thesis branch refers to previous work from the Master's thesis titled 'Individual Loss Reserving with Recurrent Neural Networks' (Lambrianidis, 2024).
The anaconda environment used for all Python code, as well as a list of R packages and versions used, can be found here. These should be installed prior to running any of the files.
The files should be examined and run in the following order:
1. Generate Dataset.R
As the name suggests, this file is responsible for simulating the datasets from SPLICE (Avanzi, Taylor & Wang, 2023) and SynthETIC (Avanzi, Taylor, Wang & Wong, 2020).
2. Data Manipulation.R
Contains the main data manipulation, as well as train-test splitting. Prepares the raw data for input into the LSTM(+) and FNN(+) models.
3. Model Training.ipynb files
These jupyter notebooks rely on 'Functions.py'. This script contains all the functions and classes to be called from each of the model training notebooks.
For any questions or further information, please contact Matthew Lambrianidis (matthew.lambrianidis@gmail.com)