Yesterday I successfully defended my dissertation proposal (hurrah!). The topic is holdings and transaction-based attribution, and, as often happens for the student, I found myself providing insights into a topic my committee members generally know little about. In theory, the student is to be the "expert" on the subject, given the amount of research that's expected. Fortunately, this is a topic I've spent a lot of time on over many years.
We briefly touched on the issue of data. I mentioned that a colleague told me that transaction-based attribution requires a lot more data, which opens it up to the risk of errors being introduced. Two quick responses arise:
1) Is it true that transaction-based attribution requires more data than holdings?
2) How dirty is the data?
The first point I'll discuss today; the second, tomorrow.
I'm of the belief (though I haven't confirmed this yet) that monthly transaction-based attribution is as accurate as daily. If this holds, then the only additional data that is needed are transaction details.
Firms that use the holdings-based method are increasingly moving to daily, in an attempt to reduce the residual that is common with this approach. This means that every day their entire portfolio must be revalued and repopulated. Talk about a lot of data!
My bet is that we'll find that transaction-based models, in actuality, require less data. We'll see! If you have empirical evidence you wish to share, please let me know!