"Larger bid-ask spreads ... are also drawbacks to consider, he added."
In theory this sounds right, but I don't think this matters in practice. Let me substantiate this, for those patient enough to read on:
When stock-level tax-loss harvesting was initially being discussed at Wealthfront (before they renamed it to 'Direct Indexing' - a term they invented, to my knowledge), the higher spreads were my first concern. So I went to a Bloomberg terminal and ran some macros to suck out historical spreads for the 500-ish constituents of a "major index". I calculated an average spread that was weighted by the index constituents' weights, so that bigger index constituents would matter more, since you'd be trading more of them (at least in an initial portfolio buy-in). I recall the average spread being ~1 basis points for the ETF (you don't need Bloomberg for this) and ~3 for the DI portfolio. This was ~2013; I imagine the numbers are even smaller now, especially for the DI portfolio, since the ETFs can't get much tighter, since many already have a publicly quoted spread of 1 penny, which is the minimum allowed. Note also that the spread is the round-trip cost, so you are really looking at half that cost for each leg of a trade.
Now, whether you are using ETF pairs/tuples or the - considerably more sophisticated - DI flavor of tax-loss harvesting, you typically only harvest losses that are large enough. Let's assume you use a threshold of a 2% capital loss. [Incidentally, our company's (Rowboat Advisors) software uses losses in 'tax space', which is more accurate, because it accounts for the differential in short- vs. long-term capital gains]. Assuming a combined federal + state marginal capital gains rate of 40%, this would result in postponing tax that's equal to at least 0.8% of the notional being harvested.
Net-present-valuing the postponed tax doesn't have a single right answer. It depends on several things:
1. future tax rate expectations; obviously if they go down for a client, e.g. in retirement when other income is reduced, harvesting has a bigger benefit.
2. similarly, how long it will be until the client withdraws money; clearly, if there's a full withdrawal in a month, then today's TLH is irrelevant.
3. the possibility that the low-cost-basis / high-embedded-gains holdings have zero tax applied to them eventually, e.g. due to a stock charitable contribution, or the client passing assets to their heirs after death
4. the extent to which there are external short- or long-term capital gains to offset harvested losses against.
5. the absolute (not relative) tax rates applicable to a client. E.g. if a client is at a 0% marginal capital gains tax rate today, harvesting is pointless.6. the liquidity (a.k.a. 'bird in hand') of having the extra cash available until it is to be paid eventually.
Given that the set of clients using TLH is somewhat self-selecting - which reduces uncertainty around most of the above - a reasonable assumption I've seen investment researchers use is that the NPV is about 10% to 25% of the projected tax savings of a tax loss harvest. That is, if I postpone $100 of taxes today, it's the same as adding roughly $10 to $25 to the account.
Going back to my previous (realistic) example, the NPV of the 0.8% postponed tax is 10% to 25% of those 80 basis points, so 8 to 20 bps. This is considerably bigger than either the ~3 bps.
The numbers are slightly worse for a stock index with a higher average spread (e.g. the top ~1500 US stocks, which I also recall researching similarly), but the benefit still exceeds the cost.
In summary, unless the tax-loss harvesting thresholds are unreasonably low, the benefit of DI would trump the additional spread cost. Of course, there are other indirect costs to holding many small positions, and my comment is already too long to go over these. However, spreads are not a reason not to do DI.
