This paper has a simple and valid point to make: Attribution which relies on the Modified Dietz method is prone to distortions and those distortions become greater as we measure smaller pieces of the portfolio.  This causes transaction-based attribution to be potentially less accurate than holdings-based attribution.