Triangle Export

[KiY0On]

Hi,

I am working on using the Triangle object from the chain-ladder package and seeking some help in using it in my project.

At the current stages:

1. I have the flat data, which I have it as a pandas dataframe as I needed to format before feeding it as the triangle data.
2. I have 2 index - one is the type of amounts such as incurred loss, case outstanding, paid loss, written prem and earned prem; the other is the reserve study group. I am working on the syntax to create these triangles.

I am concerned that once I create the triangles, would I be able to export all of the triangles as a single flat csv file to be used for analysis. I would like to use the CAS repository for the analytical part, but I wanted to try turning the flat data into the triangle and exporting out first.

I would appreciate any idea or advice on this subject. Thank you!

[jbogaardt]

The Triangle object has a to_frame method. This should allow you to export a multidimensional triangle to a pandas dataframe with all relevant info intact.

>>> import chainladder as cl
>>> print(cl.Triangle.to_frame.__doc__)
 Converts a triangle to a pandas.DataFrame.

        Parameters
        ----------
        origin_as_datetime : bool
            Whether the origin vector should be converted from PeriodIndex
            into a datetime dtype. Default is False.
        keepdims : bool
            If True, the triangle will be converted to a DataFrame with all
            dimensions intact.  The argument will force a consistent DataFrame
            format regardless of whether any dimensions are of length 1.
        implicit_axis : bool
            When keepdims is True, this denotes whether to include the implicit
            valuation axis in addition to the origin and development.
        Returns
        -------
            pandas.DataFrame representation of the Triangle.

[kennethshsu]

Hi @KiY0On, it looks like you are new. Welcome! There's a starter tutorial which introduces you to most of the basic functionalities of the package here. Did you get a chance to take a look?

[kennethshsu]

Also, I think the calendar year data needs to be January and July, not January and June. Can you try to make sure your data is at the right granularity level?

[KiY0On]

Yes, I will post the sample data.

In the meantime, just one more quick question. Once I feed the data in to create a multi-indexed data, can I output the multi indexed triangles as a csv file? But output them in the triangle formats.

Edit: I was able to output one triangle in the triangle format. However, with the multi-indexed triangles, I was trying to figure out how to iterate each triangle to output as csv. What should be a syntax to access one triangle out of multi-indexed triangle?

[kennethshsu]

For that, you'll need to loop through your indexes and do .to_frame() at each index. I don't think there's another way. @jbogaardt?

[KiY0On]

I know how to access a column of the dataframe like df['paid']. But to access by using index, what would the syntax look like? df(columns = 'paid', index = '192') ?

[kennethshsu]

You can do something like this:

raa = cl.load_sample("raa")
raa[raa.origin == "1983"][raa.development == 36]

[KiY0On]

Sample Data.csv
Here is the sample data of one study group.

NDT EU TFtest.csv
Here is the output I got from the initial run.

There must be something wrong with when I create the development column...

Thank you so much for the help and apologies for asking too many questions.

[kennethshsu]

Can you also post your python code as well?

Am I understanding this correctly? I think you used under_yr as your origin for underwriting_year, and Development Period as your ages? Development Period will need to be a date-like value, so instead of age 12, put the same year as origin. Is your Commission Amount your values?

See if you can post your python code I can help take a look.

[KiY0On]

Yes, I am trying to create Age factor by using under_yr + Development Period, which is in a 6month step in order to create a semi annual development triangle. Sorry, I am not familiar with the data-like value.

df_input["und_yr"] = pd.to_datetime(df_input["und_yr"],format = '%Y')
df_input['Cal Year'] = (df_input["und_yr"].dt.year + df_input["im"] // 12).astype(str)
df_input['Cal Season'] = (df_input["im"].mod(12)).astype(str)
df_input['development'] = (df_input["Cal Year"] +'-'+ df_input["Cal Season"])
df_input['development'] = pd.to_datetime(df_input['development'],format = '%Y-%M')

[kennethshsu]

Date-like values are what they sound like, the value should look like a date, and not an integer. For example, for und_yr = 1996 at age 12: instead of 12, we want 1996.

Consider this code that converts the ages to valuation.

df_input = pd.read_csv("/Users/kennethhsu/Downloads/Sample.Data.csv")

df_input["und_yr"] = pd.to_datetime(df_input["und_yr"], format="%Y")
df_input["Development Period"] = df_input["Development Period"].astype(int)

f = lambda x: x["und_yr"] + pd.DateOffset(months=x["Development Period"])
df_input["valuation"] = df_input.apply(f, axis=1) - pd.DateOffset(days=1)

df_input["und_yr"] = df_input["und_yr"].dt.strftime("%Y")
df_input["valuation"] = df_input["valuation"].dt.strftime("%Y-%m")
df_input

Once we do this, we will then be able to port the data frame into the triangle format by doing cl.Triangle(...).

Also, I noticed that the sample data that you posted contains lots of blank rows, you probably removed them on purpose. But on the off chance that it's a mistake, be extra careful when you are working with a data frame that contains blank rows.

[KiY0On]

Thank you so much.