Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/kasaai/rsvr

"reserver"
https://github.com/kasaai/rsvr

Last synced: 9 days ago
JSON representation

"reserver"

Awesome Lists containing this project

README 

        
---

output: github_document

---



```{r, include = FALSE}

knitr::opts_chunk$set(

  collapse = TRUE,

  comment = "#>",

  fig.path = "man/figures/README-",

  out.width = "100%"

)

```

# rsvr



[**Live design doc**](https://docs.google.com/document/d/1x2Pi7tujWLAQlqd0chBXb0Ml0Ga2HK797OVJSaQQjv0/edit)



Initial prototyping for a tidy approach for loss reserving



```{r, eval=FALSE}

library(tidyverse)

library(rsvr)



data <- ChainLadder::ABC %>%

  as.data.frame() %>%

  as_tibble() %>%

  mutate(

    origin = as.integer(origin),

    dev = as.integer(dev),

    type = "paid_loss",

    segment = "west-auto",

    currency = "USD",

    origin_interval = 1,

    dev_interval = 1

  )



mack_spec <- mack_chain_ladder()

mack_spec



result <- mack_spec %>% 

  fit(data)

result

```



# Reserving Process Flow



Here is a rough outline of what an actuary would go through in a typical reserving analysis on a periodical base.



1. Starting from the Claims Data Warehouse, an actuary collects the important data fields that are required for analysis. 

    - Usually needs: `loss payments`, `lae payments`, `case reserves`, `transaction date`, `accident date` at a minimum

2. After verifying the data, we transform the claims level data into loss development triangles.

    - Here we can produce various triangles such as `paid triangle`, `incurred triangle`, `reported counts triangle`, `closed counts triangle`, etc

    - We can further segment the data into various groupings (state, injury type, large vs small), e.g., `California triangle` vs `New York triangle`

3. At the diagnostics step, we examine the triangle data to see if there's any outliers or anomaly we need to adjust before producing the models

    - For example, we can look at how the reported frequency and severity change over time

    - We can also examine if there's any pattern in the triangles due to changes in claims practices or regulatory environment

4. After going through diagnostics, the reserving methods should be automatically populated

    - Paid, Incurred CL, BF, GB, etc

    - Bayesian Methods (Compartmental Reserving, Clark, etc)

    - Individual claims using ML such as [Tychobra](https://www.tychobra.com/posts/claims-ml/) or [Loyalty Program](https://www.casact.org/pubs/forum/17sforum/01-Llaguno_Bardis_Chin_Gwilliam_Hagerstrand_Petzoldt.pdf)

5. Step 4 and 5 are intertwined to the extent actuaries need to make adjustments to the models

    - Adjusting for outliers in link ratios

    - Coming up with a priori LR for BF method

    - Selecting prior distributions for various Bayesian models

    - Tuning hyperparameters for ML models

    - Here we should also keep track of changes we make to the models for documentation

6. Finally we will select the Ultimate losses for the book of business

7. We communicate the results within the actuarial team, and we may circle back to step 4/5 for additional adjustments

8. After finalizing reserves selections, we communicate the results to audience outside of reserving

    - We present results to relevant stakeholders (Finance, UW, etc)

    - We also supply any actuarial data required by our colleagues

9. In the last step, we save our final reserves numbers and any relevant data, documentation into a historical database for future reference





![](flow.png)