Designing a Difference-in-Differences Study of Fund Fee Disclosure
Summary
The document considers whether mandatory disclosure of inducements affects mutual fund fees, using mutual funds as the treatment group and ETFs as the comparison group in a difference-in-differences design. It highlights two identification problems: ETFs may differ from mutual funds in ways that make them a poor control, and the mutual fund sample includes an unknown share of funds that did not pay inducements and therefore were not exposed to the regulation.
The answer argues that the ideal comparison would distinguish affected from unaffected mutual funds, with unaffected funds serving as controls. If the available data cannot identify those groups, the treatment effect and its statistical uncertainty are difficult to estimate reliably. The discussion is conceptual and provides no empirical results or formal diagnostic tests; its conclusion depends on the stated data limitation and does not assess other possible identification strategies.
Key ideas
- ETFs may not provide a comparable control group for mutual funds in a fee study.
- An unknown share of unaffected mutual funds creates uncertainty about treatment assignment.
- Unaffected mutual funds would provide a more direct comparison for affected funds.
- Data limitations can undermine the interpretability of a difference-in-differences estimate.
Tags
Full text
# How to measure the impact of regulation on fund fees? # How to measure the impact of regulation on fund fees? I want to measure the impact of mandatory disclosure on inducements on fees. I thought about doing a difference-in-differences analysis around the date of the new regulation with mutual funds as the treatment group and ETFs as control group (since they do not pay kickback to advisors). My problem of course is that that there are a lot of mutual funds that do also not pay inducements but using Thomson Reuters it is not possible to filter them out. Do you think the research design would still be valid? ## Answer by Attack68 (score 2, accepted) https://quant.stackexchange.com/a/41503 I think you will struggle for the following reason; you are essentially trying to create a statistical test along the lines of: > $H_0$: impact (mean performance) on impacted mutual funds equals the mean performance on control group, i.e. ETFs, vs, $H_1$: impact (mean performance) on impacted mutual funds is less than (or different to) the mean performance on control group, i.e. ETFs. The problem is the parameter assumptions and your sample data; - The assumption of equivalence between ETFs as control and mutual funds in the first place is open to criticism, and potentially of a larger variance than your underlying parameter of interest. - You have a latent variable which is the proportion of mutual funds in your sample data that do not conform to the impacted group, meaning deriving the critical values (p-values) is unknown, even if you can isolate a test. I believe you know the answer - the unimpacted mutual funds are the best control group and the impacted mutual funds provide the comparative data, which directly solves the above two problems. Your question boils down to making estimates due to Thomson Reuters limitations whether or not you can still derive statistically significant results. In my opinion, I'm afraid not, especially since the very nature of the research suggests there may or may not be an effect so it is hard to observe. Sorry...
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.