The average cost accounting method was first created to allow a taxpayer to simply report the gain on partial sales based on the average cost of all shares purchased (instead of the default FIFO treatment, or by using specific share identification), but was reserved exclusively for mutual funds and not for individual equity securities. However, it appears now that the rules may be a little broader than anyone realized - because technically, an exchange-traded fund (ETF) may also be eligible, notwithstanding the fact that it trades more like a stock than a mutual fund.
The reasoning is that technically, (most) exchange-traded funds are registered as a regulated investment company in a manner consistent with the Treasury Regulation (1.1012-1(e)(5)) that specifies the types of investments eligible for average cost treatment. However, as it turns out, not all ETFs are quite structured this way.
In reality, ETFs are typically structured in one of three ways: an open-end index fund, a unit investment trust (UIT), or a grantor trust. ETFs structured as an open-end index fund are registered under the Investment Company Act of 1940, and consequently appear to be eligible for average cost treatment. However, grantor trust-based ETFs (most commonly, used to hold investments like commodities, because of certain SEC regulatory complications that arise when otherwise trying to hold investments like commodities in '40 Act ETFs) would not be eligible for the average cost treatment, as their structure does not fit within the Treasury Regulations. On the other hand, the treatment of UIT funds is less clear; although they appear to be a subset of the open-end fund structure, the biggest provider of UIT-based ETFs (State Street Global Advisors) refused to provide me with any guidance whatsoever about their position on this, after an excruciating 7-week delay to my request for information, so the issue remains open at this point. Fortunately, though, UIT-based ETFs are currently less common, as most ETFs are structured as '40 Act funds, or grantor trusts where necessary, since the SEC amended the rules in 1996 to allow ETFs to be structured as open-end funds. The uncertainty of UIT ETFs can be a bit of a complication, though, because many of the oldest and most popular ETFs - including the DIAMONDs, SPDRs, and Qubes - are UITs.
So why would anyone want to use average cost for ETFs? There are a few reasons. First of all, the tax rules for securities stipulate that when an investment is sold, if you can't use the specific share identification method, you are required to use FIFO (first-in first-out) treatment, which generally will result in selling the oldest shares with the lowest cost basis that produce the largest gain (assuming your investments have on average grown in value over time). So if you haven't tracked the individual lots, and/or your custodian or broker can't accommodate your request to sell specific lots, you'll be stuck with what's usually the least favorable cost basis accounting treatment. Using average cost provides an alternative to this. Beyond that, some people simply consider it more convenient and easier to use average cost accounting, where it's available.
In addition, the average cost method may actually be necessary in some situations. The original context of the question - can an ETF use the average cost method of accounting - was from an advisor working with Vanguard, which allows their investors to convert certain Vanguard mutual funds into Vanguard ETFs shares (assuming you want to do such a conversion). So if you were using the average cost method for a mutual fund, and you convert it into an ETFs, you would theoretically be required to use the average cost method for the ETF (since you were already using it for the mutual fund). Thanks to how ETFs are registered, that appears to be not only required, but in fact the proper way to account for such a transaction.
The biggest caveat? Under the tax law, once you use average cost accounting, you have to stick with it for that security position in the future. So if you think you might want to use the specific share identification method at some point down the road to maximize your tax planning, you'll probably want to take the time to use that method all along. In addition, it's worth noting that because this is a new interpretation of ETF cost basis accounting, the computer systems for most brokers and custodians won't know how to handle it, which means you'll have some manual work to do just to track your own average cost, because your custodian probably won't be able to help. But hopefully, that systems issue will be resolved with time.
In the end, we're talking about cost accounting for ETFs. This isn't exactly a big tax savings opportunity in most situations - in point of fact, for most investors the best tax planning for investments involves using specific share identification to optimize which shares to sell and which gains to recognize in the tax year based on the current year's overall tax situation. In addition, it's only relevant if you're selling a portion of your position (if you sell all the shares, it doesn't matter which cost basis method you use!), and furthermore it only matters if you were holding an open-end ETF (or maybe a UIT-based ETF). But this is still a nice planning opportunity for some, and for conversions from mutual funds to ETFs it may be a requirement for investors already using the average cost method.
Note: Special thanks to Vanguard and Barclays for their assistance in providing interpretations for some of the research in this post. Unfortunately, the fund administration and legal groups at State Street Global Advisors weren't nearly as helpful.