This is in response to Dan Egan and Boris Khentov’s article, The Tax Harvesting Oasis, which was in response to my Advisor Perspectives article, The Tax Harvesting Mirage, which appeared Aug. 12. This article is also part of an ongoing conversation about my article on APViewpoint, which you can view here. If you are not a member of APViewpoint, you can join here.
Dan Egan and Boris Khentov of Betterment, a so-called robo-advisory company, have written a response to my Advisor Perspectives article, "The Tax Harvesting Mirage.” Their response is a mixture of disparate parts.
In part, Egan and Khentov are responding not to me, but to their competition. In part, they are explaining the reasons for using their methodologies. In part, they are responding to me.
And they are, of course, also marketing their product.
Their response is complicated by the fact that they sometimes get these diverse purposes mixed up. They draw paragraphs from other material they have written, then add further material in response to my article. But they don’t integrate it all very well.
For example, they say “Edesess suggests that purveyors of automated TLH [tax loss harvesting] services gravitate towards assumptions that overestimate the benefit of TLH. We would argue that he does the opposite.” Later they add, “Edesess departs from our assumptions early and often, always in a direction that downplays the benefit of TLH.”
They don’t seem to have noticed that in between these statements, they list the assumptions that “will inflate the estimated value of TLH.” These include, for example, “Assuming the highest possible tax rates.”
They overlook that in my article I assumed the highest possible tax rates – as does Betterment’s competitor Wealthfront. This assumption, of course, has the effect of biasing the results in favor of TLH. And yet, even with this TLH-friendly assumption, I found only a small TLH benefit. None of my other assumptions were chosen for the purpose of downplaying the benefit of TLH.
Betterment versus Wealthfront
Their paragraph about the assumptions that will inflate the estimated value of TLH was written, I suspect, not for the purpose of responding to me but as a response to Wealthfront.
I understand Betterment’s predicament. It has launched a service for a laudable purpose, to make it easier for investors to invest in a reasonably sound manner without paying a hefty fee to an advisor. It also faces competitors, Wealthfront chief among them. Wealthfront advertises that it offers a TLH service that can add more than a 1% “tax alpha” annually to an investor’s bottom line. Wealthfront uses an invalid methodology to calculate that number, but few people will understand that.
Betterment must preempt its competition or lose customers. To its credit, Betterment uses a valid measure to calculate tax alpha. It explains why its tax alpha of 0.77% is right. But still, its alpha must not be too much lower than Wealthfront’s tax alpha or the comparison will ruin Betterment.
The complexity of calculating tax alpha
Egan and Khentov are absolutely right in saying that the tax alpha one derives from either a simulation or a backtest is highly dependent on numerous assumptions. It is quite complicated to calculate. I made that plain in my article.
One must also specify that it is a tax alpha compared to a baseline. I believe I made it clear in my article with what baseline case I was comparing.
Both Betterment and Wealthfront use a methodology that monitors daily prices and harvests losses on any day on which the value of one of their assets falls sufficiently below its cost basis. I compared that methodology to one that does not harvest daily but only at year-end, as part of the process of rebalancing to a constant mix. (I also compared with a process that does not rebalance to a constant mix – in other words a buy-and-hold strategy. Its tax alpha is similar to that of daily TLH with much lower portfolio turnover.) In both the daily TLH case and my base case, year-end rebalancing is performed in a tax-efficient manner.
Egan and Khentov dismiss too lightly my complaint that, unlike me, they do not, in their white paper, specify the base case upon which they base their comparisons. They explain what that base case is slightly better in their response, but I still find their explanation lacking in specifics. They also cite in their response two articles that computed a tax alpha lower than the one Betterment claims but higher than mine. Those articles, however, calculate the benefit not of daily TLH but of a combination of ordinary tax-harvesting strategies. Thus, the articles do not provide validation for daily TLH.
Departure from Betterment assumptions
My article was not about Betterment alone — it was about daily TLH as practiced by tax-harvesting advisors. I looked at the white papers posted online by Wealthfront and Betterment. I used Wealthfront’s high TLH-friendly tax rate assumptions because I could not thereby be accused of choosing assumptions that are biased against TLH (and yet Egan and Khentov do so accuse me). I used Wealthfront’s strategy of replacing its substitute asset after the wash-sale period is over, because I could easily understand that – and it is what Wealthfront says it does. I could not understand Betterment’s strategy, which it calls Parallel Position Management, well enough from its white paper to embed it in a simulation, so I did not use it. I still don’t understand it well enough after Egan and Khentov’s response.
Thus, my article was aimed at both Wealthfront and Betterment. My assumptions include a hybrid of what those two advisory firms use.
Otherwise, I believe my assumptions were reasonable. They represent a typical investor case. Other investors might be different, and hence the results might be different for different investor scenarios. However, my results are likely to be representative.
Betterment’s specific criticisms of the assumptions
Besides my failure to test Parallel Position Management and my use of Wealthfront’s methodology instead, Egan and Khentov list specific assumptions that I made that they believe bias the results against daily TLH. I will address each of these.
They say that instead of assuming a single annual contribution, they break theirs up into 24 contributions annually. They say, “Frequent deposits create multiple price points in each security, capturing local maxima which provide additional harvesting opportunities. Our assumption is favorable to TLH.”
I find it implausible that this would make much of a difference, if any. Yes, volatility means that some of those 24 contributions will be invested at cost bases higher than if they were invested in one annual lump-sum but a roughly equal number will have lower cost bases. The net result is likely to be a negligible difference one way or the other.
Their second criticism is more plausible. They argue that it was unrealistic of me to assume a $10,000 annual contribution that does not increase over the 35-year period. Here, they have a point. Their argument that having this contribution increase by 5% annually would increase the tax alpha seemed reasonable. Furthermore, it was an assumption that could easily be incorporated into my simulation program. So I tried it. The result was that it increased the tax alpha by a single basis point, from 14 to 15 basis points.
Egan and Khentov argue that their use of 12 assets, “10 of which … provide substantive volatility,” will increase the tax alpha as compared to my four assets.
Here they again have a point. The point is that the volatility assumption for the assets used in the simulation has a significant effect on the resulting tax alpha.
This is true. In one of my two sets of expected returns and volatilities for the four assets, I used the same expected returns and volatilities for all assets – 8% expected return and 16% volatility (annual standard deviation).
After reading Egan and Khentov’s response, I tried running the simulation again using 20% volatility. It increased the tax alpha for daily tax harvesting to 25 basis points from the original 17, although it also increased portfolio turnover by 7%.
Indeed, the volatility assumptions for the various assets can play a major role in producing the results. Possibly, the volatility assumptions that Betterment used in its simulations were much higher than mine.
Egan and Khentov rightly point out – as I did, and as they give me credit for – that if there are substantial capital gains to be offset that lie outside the investor’s securities portfolio, such as capital gains on the sale of a property, then the benefit of daily TLH could be higher. This may be true for some investors who, at the same time that they invest in a portfolio of liquid securities, are also regularly buying and selling properties and realizing large capital gains.
Even so, the benefits of tax harvesting may largely be captured by the process of tax-efficient year-end rebalancing. They may not be very much enhanced beyond that by daily TLH. In fact, when I reran the simulation using a $50,000 annual allowable tax write-off instead of the standard $3,000, it did not increase tax alpha by much – only 3 basis points. There are several possible reasons for this. As write-off losses in excess of the $3,000 maximum can be carried over to following years indefinitely, and are written off at the short-term capital gains tax rate, most TLH write-offs can be realized eventually. There is only a limited quantity of daily TLH opportunities every year (admittedly, the greater the volatility, the more such opportunities there will be), so $50,000 may be too much to make use of all of them.
Monte Carlo and historical backtests
Egan and Khentov give an excellent account, with which I mostly agree, of the use of Monte Carlo simulation vis-à-vis the use of actual historical data. However, it contains, in my view, two oddities. First, they say, “Edesess correctly notes that actual historical data is ‘anecdotal evidence only.’ Therefore, he concludes, Monte Carlo is ‘the only way to do the evaluation.’ We cannot agree. Monte Carlo is the only way to do a calibration.”
It seems to me that they are agreeing with me, but then they claim to disagree by making a distinction without a difference. The problem with using only historical data is that there simply isn’t enough of it. We have, for example, about 90 years of monthly stock market returns for some asset classes from the Center for Research in Securities Prices database. That comprises, at most, two independent investing lifetimes of 45 years each – not enough to draw statistical conclusions. And these are only monthly data in any case, not daily. In order to correctly evaluate a TLH strategy that only defers taxes but does not eliminate them, except for assets held at investor’s death, the backtest needs to be run daily to the end of the investor’s life and for a multitude of scenarios. This can only be done with a simulation.
Second, Egan and Khentov say, “Monte Carlo is necessary when doing modeling a path dependent process. Investment strategies which make decisions dynamically (such as TLH) cannot be accurately modeled except through a Monte Carlo process.” This is not so. Some path-dependent processes can be modeled accurately using mathematics without simulation, although it is usually very difficult — and true, it would be too difficult to do it for TLH.
Everything else Egan and Khentov say about Monte Carlo simulation vis-à-vis the use of actual historical data is correct, and in fact very well stated.
Tax planning as rent-seeking
I have said that tax avoidance that barely skirts the wash rule’s boundaries of legality is rent-seeking — that is, it raises some boats only by lowering others. It is not an activity that has the potential to raise all boats.
We have had news stories recently about U.S. corporations buying much smaller foreign companies for the sole purpose of avoiding U.S. taxes. Many people frown on this behavior. I think it would be salutary for taxpayers – and their financial advisors – to be conscious that they are doing the same thing when they exploit a loophole to evade the intent of a law.
Not everything that is legal is equally admirable. Not everything that is done because everyone else is doing it is equally right. I’m not saying that people shouldn’t engage in this activity or that they don’t have a perfect right to do so. But advisors and their clients must, at a minimum, consider whether such activity meets their ethical standards, even if it is perfectly legal.
If these kinds of economically wasteful activities are looked upon as being just as admirable as pursuits that help build wealth for everybody, there is a danger that it will become – as it already has become – more difficult to curtail them. The next thing you know, Betterment and Wealthfront and others like them may pay lobbyists to ensure that the intent of the law is not reinforced by some improved version, because they and their clients have become accustomed to thriving off the benefits. That’s why there should be, at least, a stigma attached to tax harvesting.
The cost of complexity itself
There may be some small benefit to daily TLH, though it is meager. It can be enhanced by choosing high-volatility investments.
However, both implementing the strategy and evaluating whether it is successful are complex tasks. Complexity itself has unintended consequences. By touting the benefit of their tax-loss harvesting as an integral part of their low-cost service, Betterment and Wealthfront could find themselves burdened with increasingly costly operations for which they had not bargained.
Complexity also means it is difficult to tell whether the benefit exceeds the cost. As anyone who has been involved in complicated software development knows, there is always the possibility – indeed the likelihood – that there will be bugs lurking somewhere in the program. The more complicated the program, the more likely the bugs. For this reason, when it is unclear which strategy is better between a simple one or a complex one, it is better to choose the simple one.
Read more articles by Michael Edesess