I need to carry a non-contractual accounts behavoiural study for a bank. The objective is to estimate core/non core ratios and then bucket and ftp them. Any recipe where to start? I have 3yrs of historical data, daily closing balances. From what I googled I understand that I need some kind of seasonal vs growth trend segregation. But only guidelines, nothing in particular. Visually represented my data has (e.g. current accounts) very heavy seasonal bias with highs in shoulder seasons and lows in the festive seasons ;)). How to isolate it? How do I then calculate the true core/volatile ratio?

  • $\begingroup$ welcome to Quant. S.E.! If there is seasonal bias, then you may take seasonal difference. $\endgroup$ – Neeraj Feb 22 '16 at 17:11
  • $\begingroup$ Hi Neeraj, thanks for the reply. Could you please elaborate? I'm not a retail banker, donno much about it. How to isolate the seasonal bias? Do I just take the seasonal lows, draw the line and everything below the line becomes core? sounds too simple. do i still need the regression and how do i use the two together? $\endgroup$ – Peaches Feb 23 '16 at 12:10
  • $\begingroup$ I can only help you to make your data stationary. But you wanted to estimate some ratio. I do not know would it serve your purpose or not. I think you must refer some literature on your topic and check how they dealt with such problem. $\endgroup$ – Neeraj Feb 23 '16 at 18:16
  • $\begingroup$ It would be better if you provide some data or results here. It would help the other users to understand your problem more clearly. $\endgroup$ – Neeraj Feb 23 '16 at 18:19
  • $\begingroup$ You have several time periods of seasonality to consider. There will be multiple time periods - they're the obvious ones, yearly (ranting from more spending around Christmas, different bill costs in winter vs summer, etc), monthly (for some, maybe you have some accounts of people living paycheck to paycheck - balance gets very close to zero towards the end of each period, effecting vol), and even daily seasonality (mote spending at meal times, etc). These seasonalities will differ between accounts too, but you can probably bucket them into categories. $\endgroup$ – will May 21 '17 at 22:23

Here is one of the easier ways to value a non-contractual book. 1.Get your trend line. Using moving averages smooth the data to remove the seasonal bias. Select n that gives the straightest line but still shows the growth trend. 2.Find your percentile extreme value (EV) based on chosen conf. interval. 3.Calculate the ratio of EV to your trend line for each data point. This is essentially your volatile ratio. 4.Put the volatile portion into the shortest bucket, the core into the longer selected tractor. Points: The higher the conf. interval the higher the volatile ratio will be. It will also shorted the duration of the book. For rates sensitivity must use the two components separately. Hope that helps


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.