The Tax Harvesting Mirage

A critique of this article by Daniel Egan and Boris Khentov of Betterment appears here. Michael Edesess' response to that critique appears here.

As passive investments have gained at a rapid pace in recent years over active investment management, many investment advisors are turning toward other sources of alpha (market-beating performance) to justify their fees. One of these sources is the “rebalancing bonus”, which I have previously shown to be a phantom. More recently, some advisors have been competing to show that their tax loss harvesting strategies produce a substantial “tax alpha.”

While this source of alpha is not wholly mythical, its benefits are vastly overstated.

Indeed, they may be negligible.

I will first discuss the appropriate framework to measure the potential benefit of tax-loss harvesting. I will then explain the simulations I did to measure those benefits, and then compare my results to those of Wealthfront and Betterment, two so-called robo-advisors.

The basics of tax-loss harvesting

A specific strategy being advertised is that an advisor will monitor the portfolio daily for losses in the values of securities and determine whether it may be advantageous to sell them in order to take a tax write-off. If the value of a security such as a passively managed exchange-traded fund (ETF) falls well below its original purchase price – its cost – then a tax loss can be realized by selling it.

One challenge with this strategy is that it needs to navigate the “wash rule.” The rule states that losses from a sale of a security cannot be deducted if a substantially identical security is purchased with the proceeds within 30 days. The Internal Revenue Service (IRS) established this rule in an attempt to prevent investors from taking artificial losses solely for the purpose of tax avoidance. I will come back to the intent of the rule later.

To avoid the wash-sale penalty while still adhering closely to an investment strategy, the investor needs to use the proceeds from the loss to purchase a security that the IRS would not construe to be “substantially identical,” but which is nonetheless close enough to identical for the investor’s purposes. This “not quite substantially identical” security can then be sold 31 days later and replaced with the security that was originally held.

Advisor-practitioners of the daily tax-loss harvesting approach profess to have determined which passively-managed ETFs are “not quite substantially identical” in the eyes of the IRS and can therefore be legitimately substituted temporarily for one another in the process of harvesting a tax loss and avoiding the wash-rule penalty.

The specifics of the strategy

When a tax loss is harvested and a substitute security is purchased, the new security will have a lower cost basis than the security sold. Hence, in the future, the eventual capital gains will be increased by the exact same amount as the amount harvested as a loss. Therefore, the tax write-off on the loss that was realized is only temporary (except in the event that the security is held until the investor’s death, when the security’s cost basis will receive a “step up” to its current value).

Because the substitute security will — at least according to the strategy used by some advisors — be sold 31 days after its purchase, in order to restore the “integrity” of the intended mix of assets, that sale could produce a short-term gain. The tax cost of that short-term gain needs to be weighed against the benefit of the tax loss that was harvested. Hence, some practitioners of the daily tax-loss harvesting strategy establish a loss threshold below which losses will not be harvested. Only when the loss passes that threshold will the security be sold and a substitute security purchased in its place, to be sold in 31 days and the original security repurchased and placed back in the portfolio.

It is a complex but accomplishable task to evaluate how much the savings from this form of tax-loss harvesting can add to after-tax investment returns. Some practitioners have referred to this positive increment to an investor’s after-tax return as “tax alpha.” Unfortunately, as we shall see, their definitions of tax alpha leave much to be desired. They can overestimate the tax alpha by a factor of 5 to 10 times.

How to compare, and to what
The value of a tax-loss harvesting strategy can only be evaluated over an investor’s lifetime, in order to capture the value of the tax deferral and the step-up in basis at death. Given that the investor will add and withdraw funds each year, the appropriate measure of after-tax return is the internal rate of return (IRR) over the investor’s lifetime, from the time the investor begins an investment portfolio until that portfolio is depleted or the investor’s death.

The next question is, to what should it be compared? That is, if tax harvesting is an improvement, it is an improvement over what?

Let us assume that if daily tax-loss monitoring is not practiced, then the portfolio will be adjusted only once yearly. That adjustment will be made as money is contributed or withdrawn, the portfolio is rebalanced or asset percentages are readjusted. The practitioners of daily tax-loss monitoring that I have reviewed apply a strategy of rebalancing the portfolio’s asset mix to the original mix, so let us assume that is the strategy adhered to once each year.

When the once-yearly rebalancing and readjustment is accomplished, it will entail tax implications. It is standard practice to minimize the negative tax implications when selling an asset by selling the highest-cost tax lots first.

Hence, the appropriate comparison for determining any “tax alpha” is between a portfolio that rebalances annually and practices daily tax-loss monitoring and occasional loss harvesting, and one that does not practice daily tax-loss monitoring and harvesting. In both cases it will be assumed that the annual rebalancing and readjustment process, when it entails the sale of assets, sells the highest-cost tax lots first.

Dividends will be assumed reinvested and the taxes on them will be ignored on the assumption that they will be the same for the portfolio that practices daily tax harvesting and the portfolio that does not. In accordance with the assumptions made by one of the tax-harvesting advisors, Wealthfront, the assumption will be made that the tax rates are those of top-bracket California investors: 56.7% for income and short-term capital gains and 37.1% for long-term capital gains. Hence, the “tax alpha”, if any, will be as high as any tax alpha realized by an actual investor. That is, it uses the assumption that is most beneficial to those claiming a tax alpha.

Assumptions of the model

If we were to use actual daily prices to make these comparisons, we would discover very quickly that we do not have enough data. We need to make the comparisons over investors’ lifetimes, from their first contribution to their portfolio until their death. That means that if an investor begins investing at age 25 or 30, a time span of 60 years or more needs to be used to evaluate the comparative advantage of daily tax-loss harvesting. But even if we had daily returns data for the desired assets for 60 years, that would comprise only one time period of record – in other words, it would provide anecdotal evidence only. One time period would not be enough to evaluate the average or expected advantage of daily tax-loss harvesting, much less the probability that that advantage would be in some particular range.

Therefore the only way to do the evaluation is to simulate daily price changes using a standard Monte Carlo methodology. Simulated daily price changes are nearly identical in their patterns over time to actual daily price changes.

I assumed that the investor began contributing at age 30 and contributed $10,000 a year until age 65, then withdrew $50,000 a year until death. The age at death for each simulation was determined randomly from a table of actuarial survival likelihoods. Different sets of expected returns, standard deviations and correlations among the assets were assumed in order to survey a range of results. These results are reported below.

When a tax is levied on a realized capital gain, the tax either reduces contributions to the portfolio in that year or increases withdrawals from it. When a tax loss is taken and thus reduces taxes that year, that reduction is added to the contributions in that year or deducted from the withdrawals.

No more than $3,000 of net tax loss can be credited against income in a given year. The remainder of the tax loss is carried over to subsequent years, keeping account separately of net long-term losses and net short-term losses in accordance with IRS rules.

Results of the simulations

I ran simulations for two portfolios of four assets each, one in which the expected returns and standard deviations were all the same, and one in which they were different for different assets. For each portfolio, 1,000 simulations were run of each strategy:

Adding net contributions to the portfolio or extracting net withdrawals and rebalancing only once yearly.
In addition to the actions in #1, harvesting a loss whenever daily monitoring determines that the loss is above a threshold.

Through running a few simulations, I determined that a threshold for harvesting a loss of approximately the expected gain on the substitute security, plus one standard deviation above that expected gain, maximized the tax alpha over time. Therefore, I used that threshold in the simulations to determine when to harvest a loss in Strategy 2.

In the case of the portfolio in which it was assumed that the expected returns and standard deviations were equal, a buy-and-hold strategy was also run. New net moneys contributed were invested in accordance with the original allocation, while net withdrawals were distributed in accordance with the currently existing allocation. In the buy-and-hold strategy, no daily loss harvesting was performed.

The result in the case of the portfolio in which it was assumed that the expected returns and standard deviations for the assets were unequal was that the average increment in after-tax IRR for Strategy 2, the daily loss-harvesting strategy with year-end rebalancing, over Strategy 1 with only year-end rebalancing, was 14 basis points a year. However, the average annual portfolio turnover for Strategy 2 was 28% while for Strategy 1 it was only 9%, a difference of 19%.

The result in the case of the portfolio in which it was assumed that the expected returns and standard deviations were all equal was that the average increment in after-tax IRR for Strategy 2 over Strategy 1 was 17 basis points a year. The average annual portfolio turnover for Strategy 2 was 28%, while for Strategy 1 it was only 8%, a difference of 20%. In this case, a strategy of buy-and-hold was also run, since it would not be presumed to have an innate non-tax-related return advantage over rebalancing, due to the expected returns and standard deviations all being the same. The buy-and-hold strategy without daily loss monitoring and harvesting had a tax alpha of 16 basis points, compared to the rebalancing strategy without daily loss monitoring. Its tax alpha was thus nearly the same as that of Strategy 2.

To summarize, in the case of unequal portfolio expectations, the average annual tax alpha for daily loss monitoring was 14 basis points, but it created incremental portfolio turnover of 19%. In the case of equal portfolio expectations, the tax alpha for daily loss monitoring was 17 basis points and created incremental portfolio turnover of 20%.

Why are these results so much different from the claims of tax-loss harvesters?

Two so-called robo-advisors have come out with white papers on their tax-loss harvesting methodologies: Wealthfront and Betterment. In both cases, they claim tax alphas much greater than the ones I found in this study. Wealthfront claims that its studies show that historically a tax alpha of 1.13% could have been realized, while Betterment claims 0.77%.

Why are these claims so much greater than the results I have shown here? In the case of Wealthfront, its measure of tax alpha is incomplete, because it takes credit for the losses harvested but conveniently does not debit its tax alpha for the compensating gains that must inevitably be realized eventually (prior to death), as financial advisor Michael Kitces has pointed out.

Betterment uses a better measure, employing IRR, though it is not applied over a time period extending over an investor’s lifetime. However, Betterment’s white paper on tax loss harvesting, which it calls TLH+, is not transparent as to what baseline strategy its tax-loss harvesting strategy is compared with. It merely says “we used IRR to calculate the excess return that a TLH+ portfolio would have generated relative to the baseline Betterment portfolio over the last 13 years.” It does not specify what that baseline portfolio is. Perhaps the baseline portfolio is one that practices a strategy that takes no advantage whatsoever of any tax-avoidance measures, such as selling highest-cost tax lots first. It is impossible to tell.

My results suggest that the advantages with daily tax-loss harvesting may be minimal or negligible. The increased turnover with Strategy 2 will increase transaction costs, though to what extent is not known. The greater complexity of the strategy increases the likelihood of something going wrong, such as programming bugs or other pitfalls. Further, the IRS may deem that the securities substituted for others “not quite substantially identical” are, in fact, substantially identical.

Furthermore, there is no guarantee whatsoever that tax rates will not rise in the future, in which case tax deferral could have detrimental effect. My study, and presumably those of Wealthfront and Betterment, assumes that tax rates will remain constant. But if tax rates go up in the future, the tax-deferral benefits of loss harvesting could evaporate, while if tax rates decline the benefits could be enhanced.

Possible additional advantages to tax harvesting

My study assumed that the investor could only take advantage of the portfolio’s net losses to the extent of $3,000 in tax losses per year, with any excess carried over to future years. It is possible, however, that the investor might have incurred other capital gains outside his or her portfolio of investment securities that these losses could offset. For example, the capital gain on the sale of a house or other property could be offset by portfolio losses. If such a capital gain is anticipated, in the near future or even in the distant future, that would provide additional reason to accumulate carryover tax losses.

Rent-seeking and regulatory arbitrage

I mentioned that the intent of the wash rule was to prevent investors from taking artificial losses solely for the purpose of tax avoidance. The tax-loss harvesting strategies are clearly efforts to take artificial losses solely for the purpose of tax avoidance. In Kurt Eichenwald's excellent book on the rise and fall of Enron, Conspiracy of Fools, he writes, "Ultimately, it was Enron's tragedy to be filled with people smart enough to know how to maneuver around the rules, but not wise enough to understand why the rules had been written in the first place."

Something should be said about the ethics of going to such lengths to get around the IRS rules that were written for a reason. This is a clear case of regulatory arbitrage for the sole purpose of rent-seeking.

Rent-seeking is the practice of effecting wealth or income redistribution, often by manipulation of the rules or regulatory arbitrage, in order to benefit one set of individuals or entities at the expense of another, but without increasing the total amount of wealth. It is difficult to see how artificial tax-loss harvesting, in contempt of the reasons why the wash rule was written, will increase the total amount of wealth for everyone. It will only redistribute the tax burden from those in a position to practice daily tax-loss harvesting due to having a substantial investment portfolio to others who are not so positioned.

While this may not be the measure by which the behavior should be judged, it is particularly petty to engage in such dedicated rent-seeking behavior merely for so trivial a benefit.

Michael Edesess, a mathematician and economist, is a visiting fellow with the Centre for Systems Informatics Engineering at City University of Hong Kong, a principal and chief strategist of Compendium Finance and a research associate at EDHEC-Risk Institute. In 2007, he authored a book about the investment services industry titled The Big Investment Lie, published by Berrett-Koehler. His new book, The Three Simple Rules of Investing, co-authored with Kwok L. Tsui, Carol Fabbri and George Peacock, has just been published by Berrett-Koehler.