The defense in depth we use to prevent tax management from eroding investment strategies.
Smartleaf is a platform for automating portfolio customization and tax management. A question we sometimes hear is: how do you make sure that tax management doesn’t systematically degrade pre-tax returns? Or, more colloquially: how do you keep the tax tail from wagging the investment dog?”
It’s a good question. We’ll answer it here in the specific context of our own system, but we suspect that the principles are broadly applicable.
A closer look at the problem
Suppose you have a beat-the-benchmark strategy – a favored list of stocks, a factor-tilt strategy, a sector rotation strategy, or what have you. Or maybe your goal isn’t alpha at all. Maybe you promised a retiree a high-dividend strategy, or you promised an energy-company executive low exposure to the energy sector.
Either way, the expectation is clear: portfolios should reflect strategy.
Active tax management changes the holdings in a portfolio. (If it doesn't, it’s not active tax management.) How do you make sure that those changes won’t dilute—or even undo—what you’re trying to achieve?
To be clear, some deviation is inevitable.If those deviations are random—just as likely to help as hurt—you’re basically fine, as long as the level of noise is tolerable. (More on that later.)
What’s not fine is systematic drift away from your strategy. For alpha strategies, that can mean lost returns. But even for strategies that are, over time, neutral—neither good nor bad, working ½ the time and failing ½ the time—inconsistency creates real-world risk. If your CIO publicly backs a bet and portfolios don’t reflect it, that disconnect can cause problems—especially if the bet ends up paying off.
The solution: defense in depth
We don’t solve the problem using just one method. We deploy four. Call it, if you will, defense in depth. Our methods are: trade-off analytics, differential treatment of legacy holdings, buy lists, and range constraints.
Tradeoff analytics
Our tax management isn’t a simple set of tax management rules that get implemented in isolation. Our analytics are optimization based, except we use the term “trade-off engine” instead of “optimizer.” Conventionally, optimizers are used to create a strategy. That’s not how we use it. We call it a trade-off engine because we’re trying to balance a bunch of possibly conflicting goals: follow your strategy AND obey constraints AND reduce taxes and transaction costs.
That’s where the optimizer—err, trade-off engine—comes in. Tax management is never implemented in isolation. It’s balanced against faithfulness to your strategy, as measured by tracking error. This keeps tax management on a, well, leash.
That is, the system does not blindly follow tax management rules. For example, in a down market, the system will not loss-harvest everything. It will stop when further loss harvesting sales would cause excess tracking error. And while the system tries to avoid realizing capital gains, especially short-term gains, it’s not an absolute. It would, for instance, sell a large concentrated position with small gains, even if it were short term.
In practice, this approach alone—tradeoff analytics—has proven sufficient. To date, it doesn’t appear that any tax-tail-wagging-the-investment-dog problem has ever pierced this first line of defense.
Nevertheless, we have defense in depth. The reason is that, in principle, at least, the tracking-error aware approach could fail. We use a multi-factor risk (covariance) model. It works well, but there is always the possibility that what drives your strategy is invisible to our model.
To give a mostly — but not entirely — silly example, suppose your strategy is to underweight securities of firms whose CEOs win lots of accolades. For better or worse, “CEOs winning awards” is not one of our risk model’s factors, so, theoretically, we could be blind to your goals and accidentally work against them.
We’re not aware of something like this ever happening, but better safe than sorry. So we support additional defenses. Let us continue.
Differential treatment of legacy holdings
A (really important) component of tax management is gains deferral – deciding to hold overweighted positions with unrealized gains. That includes securities that are no longer in the target model. A special case is legacy holdings with low tax basis.,. The optimization-based, tracking-aware analytics described above can figure out what to underweight in order to minimize the drift (tracking error) that holding the legacy position would cause, but it can’t capture your return beliefs regarding those holdings. We address this by allowing users to assign negative rankings to legacy holdings, individually or in groups. This allows the user to tilt the system towards selling some or all legacy holdings.
Buy lists
When you loss harvest, you need to reinvest the proceeds elsewhere. While we support designated substitution lists for mutual funds and ETFs, our default is to just set the buy list to “any other security in the target portfolio”. That ensures that all purchases are from a list you’ve expressly approved—models you’ve chosen to incorporate into your target. The vast majority of our clients use this approach, and it works quite well.
However…this, too, can, in principle, fail. The problem is that not all securities in a model are equally loved. It’s fairly common for a model to consist of something like 10% - 20% weights that reflect strategy, with the 80% - 90% of the remainder serving as “beta filler” that keeps the model fairly close to its benchmark. Yes, the beta-filler securities are “approved”, but not in the sense that they embody your strategy. In theory, tax-driven trades could overweight the filler and underweight the core.
If security movements were uncorrelated, the probability of this happening would likely be vanishingly small. But the returns securities embedding a strategy typically are correlated, so you could imagine a scenario where they all dip in value in unison. You then loss harvest and buy out-of-strategy substitutes, which then fail to participate in a rebound.
Again, we’ve never seen this happen, but, again, never say never. So we continue. (As we said, it’s defense in depth.)
Asset-class-, model-, and security-level min/max range constraints
We support minimum and maximum constraints on drift at the asset class, model, and security level. So, if you have specific asset classes, models, or securities that do most of your strategy’s heavy lifting, you can mandate that tax management never causes the portfolio to stray too far from home.
We support these controls but discourage their use because, as far as we can tell, they are unnecessary, a solution in search of a problem. The standard solutions described above (tracking-error-aware tax management, setting the buy list to the securities in the model, and favorably ranking those securities) seem to work. And if you don’t need range constraints, you shouldn’t use them. They’re a thumb-on-the-scale override of the trade-off logic we use to balance low taxes and faithfulness to your strategy, which degrades tax management and/or tracking error.
Nevertheless, they’re available as yet another defense against tax management overwriting a strategy.
What about “noise” from tax management?
We mentioned that active tax management will inevitably introduce some tracking error, . If it’s unbiased and modest, it’s typically not a problem. But what counts as “modest”?
Excessive noise can cause a “heads I don’t win, tails I lose” conundrum. If the random tracking error causes better-than-expected performance, you may not really get much credit (‘heads I don't win”), but if you dramatically underperform, you get fired (“tails I lose”). How do you fix this?
We address this by setting the balance between tax savings and tracking error such that the tax benefits reliably outweigh performance noise. This turns “heads I don’t win, tails I lose” into “heads I win, tails I don’t lose”. If the tracking error is randomly in your favor, everybody is happy. If it isn’t, the client is still likely better off given that the decline pre-tax performance is more than compensated by taxes saved or deferred.
So, is the problem solved?
Does this put all concerns about the tax tail wagging the investment dog to bed? Not entirely. Suppose your CIO talks publicly about specific securities. We support security level min/max range constraints, so we can guarantee that every portfolio (except those with a specific never-hold constraint) will hold some. But we might end up holding more or less than the CIO recommends. You can always put tighter min/max range constraints on the holdings but, at some point, the constraints become so binding that there isn’t much room left for tax management.
That’s a choice you can make–surrendering the substantial benefits of tax management to make sure that there is no perceptible gap between your strategies and your portfolios, independent of whether that gap actually erodes value. But it’s not where the industry is heading. Wealth management is increasingly financial-planning centric. It’s about acting as the client’s lifetime financial coach, not their stock-selection guru. If that’s you, you’re good. Tax management won’t destroy value. The tax tail won’t wag the investment dog. You’ll just have a happy dog.
1 For us, this is definitional. We have a report—the “Taxes Saved or Deferred Report”—that we generate on demand for each account and household, but only decisions that result in an under or overweight are considered. So, for example, neither selling at legacy holding at a loss, nor optimal tax-lot selection are included.
2 https://econpapers.repec.org/article/oupqjecon/v_3a124_3ay_3a2009_3ai_3a4_3ap_3a1593-1638..htm
3 These are real numbers given to us by a well-known asset manager whose name I will not reveal.
4 How could having range constraints degrade tracking error? Range constraints set limits on what we can do to compensate underweights and overweights. Suppose, in some asset class, you have a bunch of low-basis, overweighted positions—still within the permitted range, but nevertheless overweighted. The right response might be to sell off holdings of other securities in the same asset class, thereby achieving perfect asset class balance at low tax cost. But this would be disallowed if you had security level min/max range constraints.