Determine smoothing method for performance attribution

AI & Machine learning
1

Hello,



I am a Data Scientist . Can you please answer to the following question?



The smoothing algorithm (Carino) will smoothen each attribute to make the sum of all the attributes daily to sum up to the active return.



One goal of smoothing is to make specific sub samples additive for some purposes (across time, or across investment groups or asset classes), however the Carino method does not allow for this. There is a variation of Carino which allow smoothing across time in a way that the sum of attributes can be additive across say month, or quarters, however this is still not enough for getting additivity across asset classes.



Using some acronyms here…

  • AR is active return, or benchmark return minus portfolio return
  • AA asset allocation attribution component
  • SS security selection attribution component
  • IF interaction factor

The goal is to have 
  • additivity at the total portfolio (AR = AA + SS)
  • additivity at each asset class (AR for asset class = AA for asset class + SS for asset class)
  • addivity 'down', so the ARs for each asset class add up to the portfolio AR and so on for AA and SS.
Calculating AA and SS and IR is simple. How can we make them additive in this fashion?
2

Hi Tigran, 



To understand the performance attribution of the portfolio by dividing them into different components, you can use Brinson attribution. 



You can read more about it here. 



Thanks,

Bhavika

3

Hi Bhavika,



Thank you for your reply. I have read your blog called Portfolio Analysis - Performance Measurement And Evaluation. It contains very useful information.



I have sample data  where I have calculated allocation, selection and interaction effects. I have  received additivity for totals but not for individual asset classes. Can you please help to fix this issue by using the Geometric linking method or any other smoothing algorithm.

4

Hi Tigran, 



To understand the issue better, can you please share a sample of the data and the procedure that is followed? 



Thanks,

Bhavika