This article is in response to Michael Edesess’ article, The Tax Harvesting Mirage, which appeared Aug. 12. This article is also part of an ongoing conversation about Edesess’ article on APViewpoint, which you can view here. If you are not a member of APViewpoint, you can join here.
Michael Edesess’ response to this article appears here as well as in the APViewpoint conversation.
Advisor Perspectives welcomes guest contributions. The views presented here do not necessarily represent those of Advisor Perspectives.
What is the value of tax-loss harvesting (TLH) to individual investors? Automated investment services have democratized investing strategies that were once the exclusive domain of highly sophisticated investors and planners. TLH is a prime example of such a strategy, and with greater exposure comes greater awareness among the investing public.
The automated strategy itself can be complex enough – far more so is any attempt to produce a reasonable estimate of its long-term value. Any such analysis is hugely dependent on a sizable list of assumptions with respect to general market conditions over the time period, investor-specific cash flows and individual tax circumstances extrinsic to the portfolio. Varying just one of these assumptions could dramatically alter the estimate.
It is therefore not surprising that attempts to pin down a set of reasonable assumptions have been met with alternate analyses that reach different results. In a recent article, poetically titled “The Tax Harvesting Mirage,” Michael Edesess referenced our firm’s Betterment TLH+ service and the performance estimate we have published (77 bps annually) and attempted to estimate the value of TLH on his own. He determines the benefit (14-17 bps annually) to be notably lower.
For reference, peer-reviewed studies found higher figures than the ones Edesess reported. Arnott, Berkin and Ye found a 50-basis-point post-liquidation ‘tax alpha’ from tax-loss harvesting over a 25-year Monte Carlo simulated period.1 Smith and Smith found a 37-basis-point premium from loss harvesting combined with rebalancing over a 40-year period.2
Edesess’ effort is admirable, and he clearly understands many of the nuances of long-term TLH strategies. However, as a starting point, he assumed an algorithm far less sophisticated than the one powering our TLH+ service (which our white paper describes in detail). Then, in running his own analysis, he diverged significantly from the assumptions made in ours, with no discussion as to why his are preferable.
Edesess suggested that purveyors of automated TLH services gravitate towards assumptions that overestimate the benefit of TLH. We would argue that he did the opposite – he failed to factor in a number of circumstances that we believe apply to a majority of investors. In doing so, he understated the potential benefit of TLH+.
His piece, and others like it, yielded lively discussion amongst advisors and investors on the value of automated TLH, in light of its higher profile. We took great care to publicly disclose how we arrived at our estimates. We would like to take the opportunity to highlight where our assumptions differ from Edesess’ and why we believe ours are appropriate.
Material assumptions
As Edesess noted, we properly employed an after-tax IRR in our calculations, thereby eliminating a key source of distortion as to the benefits of TLH. However, unlike asset selection, TLH performance will be materially affected by a number of taxpayer-specific facts and circumstances.
The following assumptions specific to an individual taxpayer will inflate the estimated value of TLH:
Assuming only the highest possible tax rates.
-
Assuming the taxpayer has infinite like-for-like short-term or long-term capital gains to offset every year, no matter how large the loss harvested. Assuming bottomless long-term capital gains (LTCGs) one can trigger at the end of the year is reasonable. Assuming large amounts of short-term capital gains (STCGs) in the exact same year that you harvested STCLis very aggressive.
Our white paper makes none of these assumptions in the headline estimate we report. We use a moderate tax rate, and $3,000 ordinary income and LTCG offsets only.
Notably, Betterment has gone further than any other automated investment service by illustrating the full lifecycle of TLH – i.e., eventual liquidation. We considered 0%, 50% and 100% liquidation scenarios. Experts took note.
We gave a lot of thought to the assumptions we did make in our communications, using the profile of a typical Betterment customer as reference, where appropriate. Though he set out to analyze our claims, Edesess departed from our assumptions early and often, always in a direction that downplays the benefit of TLH. He then asked: “Why are these results so much different from the claims of tax-loss harvesters?”
We can answer his question, at least with respect to Betterment’s findings.
Deposit frequency
Edesess assumed a single, annual contribution over 35 years. We assumed that the annual amount is broken up into 24 contributions, to mimic auto-deposit schedules we observe from our customers (set up to coincide with paychecks). Frequent deposits create multiple price points in each security, capturing local maxima which provide additional harvesting opportunities. Our assumption is favorable to TLH, and we plainly say so. It also happens to be a far more accurate representation of actual client cash flows. In our Best Practices discussion, we encourage customers who make use of TLH+ to deposit frequently.
Deposit size relative to portfolio
With no additional flows into a portfolio, TLH eventually ceases to deliver additional value (on top of whatever tax deferral was already captured in earlier years). Why? Markets climb in the long run, leaving the original cost basis far behind. An equity index fund purchased in the mid-1990s would still have shown a gain during the crash of 2008. While Edesess assumed that deposits will continue, he kept them constant at $10,000/year. The effect is that as a portfolio grows over the decades, his ongoing contributions approached zero as a percentage of the portfolio. The benefit of TLH won’t disappear completely, but Edesess effectively put it on life support, ensuring that its annual benefit (measured in reference to the portfolio) continually shrinks.
Of course, a more realistic assumption is that as an investor gets older, he or she earns more (due to both inflation and career growth) and will contribute more each year. We assumed twice-monthly deposits of $750, but increased them by 5% annually, to account for both inflation and real salary growth. When measuring the value of TLH over many decades, ensuring that your assumptions still make sense in the future is crucial.
Number of assets in portfolio
Edesess chose four assets for his model, though he doesn’t disclose which. The Betterment portfolio consists of 12 assets, 10 of which (excluding SHV and VTIP) are substantively volatile and have been chosen in part because their varying correlations decrease volatility in the overall portfolio. Why this provides additional fodder for a TLH strategy does not require an explanation. Our study was first and foremost an attempt to communicate the value of our own TLH+ strategy on our own portfolio. This was a rare instance where there was only one “correct” assumption we could have made.
Gains outside portfolio
Edesess assumed that an investor could only take advantage of the $3,000 ordinary income offset after offsetting any gains generated only within the portfolio running the TLH model. This is a reasonable assumption, and surely describes many investors. However, it greatly undersells the benefit of TLH. To his credit, Edesess noted that capital gains outside the portfolio “would provide additional reason to accumulate carryover tax losses” in his concluding substantive passage. However, this is not an afterthought. It is a top-line assumption.
In our experience, many customers seeking to make a transition from a legacy portfolio to a lower-cost automated offering such as Betterment’s are quite aware of the tax impact of such a transition and are eager to mitigate it. These investors are able to realize as much in LTCG as necessary to use however much in losses that TLH+ can generate. Note, however, that we do not assume that every STCL harvested will be matched with STCG. The availability of endless STCG in any year does notaccurately represent a typical Betterment customer, so our top-line assumption is that STCL will offset LTCG, which is more conservative than assuming that each loss will match against like.
30-day switchback to primary security
The assumptions above will all have material impact on how much benefit a TLH algorithm will deliver. However, Edesess’ assumption about the algorithm itself makes any comparison to Betterment’s TLH+ meaningless.
Early in his piece, Edesess noted that the substitute security will be sold 31 days after purchase “at least according to the strategy used by some advisors.” Betterment is not one of those advisors, and we devote substantial space in our white paper to why this approach hurts tax alpha and how we built our service to avoid the negative tax arbitrage that results from this approach.
When discussing his results, Edesess confirms that he employed the blind switchback: “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.” His insight is only relevant in the case where the substitute security is always sold, even if it realizes STCG. Betterment took great care in engineering a “Parallel Position Management” system, in no small part because we found this approach to be twice as effective as this switchback strategy.
This is a critical distinction between ours and other TLH algorithms whose descriptions are publicly available. By painting over all automated, daily TLH with the same brush, Edesess arrived at an assessment that inevitably failed to do justice to the benefit TLH+ is expected to deliver.
Baseline transparency
These departures would clearly alter any semblance to our findings beyond recognition. Rather than digging into these divergent assumptions, Edesess simply states that we are unclear when we say that excess return due to TLH+ is calculated “relative to the baseline Betterment portfolio” and leaves it at that.
To clarify – the baseline Betterment portfolio is our core offering. The exchange-traded funds (ETFs) are the primary ETFs listed here. The core offering would never include any positions in the alternate ETFs. TaxMin lot selling, which selects lots in order to minimize tax burden, is part of our core offering and is the default selling method for every Betterment customer for every transaction, so the baseline includes it. The baseline also includes smart dividend reinvestment, which shores up underweight positions rather than buying pro rata, as well as a number of other tax optimizations. The purpose of our white paper was to isolate only the effect of the TLH+ service against the backdrop of all other features provided by Betterment.
Monte Carlo and historical backtests
Edesess correctly noted that actual historical data is “anecdotal evidence only.” Therefore, he concluded, Monte Carlo is “the only way to do the evaluation.” We disagree. Monte Carlo is the only way to do a calibration. Once the model is calibrated, backtesting it against actual historical data is standard, and if done right, uncontroversial, provided that the models are not actually designed with such data in mind.
Monte Carlo has two strengths which make it the ideal calibration tool:
It considers all possible distributions: A mechanical Monte Carlo can deal with projecting anychosen set of returns: normal, non-normal, auto-correlated or reverse-chronological.
It supports path-dependent processes: Monte Carlo is necessary when modeling a path-dependent process. Investment strategies that make decisions dynamically (such as TLH) cannot be accurately modeled except through a Monte Carlo process.
Accordingly, the variables that steer Betterment’s TLH+ were calibrated exclusively through our proprietary Monte Carlo engine, built in-house. It can take in any given set of:
Daily price returns
Dividends
User cashflows
Fees paid to us (the advisor)
Lot sale selection methods (TaxMin, FIFO, HIFO)
Wash sale implications, including the complications around qualified dividend income (QDI) holding periods and municipal bond wash sales.
Rebalancing triggers and methods
TLH triggers, at a lot and ETF level
End-of-year taxes, including:
Carry-forwards from previous years
QDI versus non-QDI taxation of dividends, including holding-period restrictions
Federal and state income taxes, including income exemption for municipal bonds.
Up to $3,000 income offsets at ordinary income tax rates.
Proper netting of ST versus LT capital gains at year-end, prior to income tax offsetting.
This granularity gives us a high degree of confidence in the accuracy of any given run. It also allows us to test all the incremental improvements and get a correct picture of the impact to a change in strategy.
While indispensable to model designers, Monte Carlo has a number of weaknesses when it comes to communication with the general public. Customers virtually never ask to see how our strategies would have performed across 10,000 possible futures, but rather how they would have performed historically.
Whereas the concept of forward-looking returns carries with it a thorny implication for future performance, customers generally understand that past performance is only illustrative and not predictive of the future. Yes, the events of 2001 and 2008 will never repeat in exactly the same pattern. However, backtests allow customers to visualize a concept as complex as TLH in the context of those familiar market events and observe what TLH could and could not do in those circumstances.
As long as the model itself was not calibrated based on historical data, the use of backtests when communicating the model allows us to educate retail investors in a framework most intuitive to them.
Tax-planning as rent-seeking
Edesess wrapped up by addressing tax-planning generally, and we will do the same. In his words, efforts to reduce tax liability by legal means constitute “rent-seeking” because they do not create new wealth – they merely redistribute it. He finds such activity especially troubling when it seems to run counter to legislative intent, as he sees it.
In designing TLH+, we made a number of conservative assumptions with an eye towards firm accordance with existing law, as it is written. We are skeptical of Edesess’ implication that each of hundreds of millions of taxpayers should interpret legislative intent, rather than follow written law. Rather than get into a longer discussion on the appropriate process by which laws should be drafted and if necessary amended to match lawmaker intent, we’ll defer to Learned Hand, one of the most storied judges in U.S. history:
“Over and over again, courts have said that there is nothing sinister in arranging one’s affairs as to keep taxes as low as possible. Everybody does so, rich and poor; and all do right, for nobody owes any public duty to pay more than the law demands: taxes are enforced exactions, not voluntary contributions.”
Conclusion
We are in the process of launching Betterment Institutional, our platform for advisor-managed accounts. We are eager to work with advisors and believe that our offering is strongly complementary to advisors’ services. The Betterment platform excels at optimal investment management, and advisors excel at delivering a multitude of services, well-tailored to their clients’ needs. Our automated daily TLH+ service is just one component of an offering that is executed with the highest level of efficiency, and at far lower cost than individual advisors would bear if they attempted to do it themselves.
Daniel Egan is the director of behavioral finance and investing at Betterment, an online investment manager. He is responsible for maximizing Betterments customers’ take-home returns - returns after tax, costs, risk and misbehavior.
Boris Khentov is operations manager and legal counsel at Betterment, an online investment manager. A software engineer and tax lawyer in previous lives, he now designs and scales automated investment products, with a focus on tax-efficiency and compliance.
Read more articles by Daniel Egan and Boris Khentov
