• KEY TAKEAWAYS
• Asset location—an investment strategy that manages multiple accounts with different tax profiles as a single portfolio—can deliver additional after-tax returns, while maintaining the same level of risk.

• Intelligently applying asset location to a globally diversified portfolio is a complex, mathematically rigorous, and continuous undertaking.

• Common approaches cannot maximize the strategy’s potential because they (1) consider only tax efficiency, ignoring expected return, and (2) focus only on reducing annual tax drag, ignoring liquidation tax.

• A mathematical optimization is more effective than rules of thumb when balancing numerous trade-offs and constraints.

• Tax-Coordinated Portfolio (TCP), Betterment’s automated implementation of the strategy, relies on linear optimization to locate assets.

• Based on our research, we expect TCP to deliver an annualized tax alpha in the range 0.10% to 0.82%, depending on assumptions.

“To allocate assets is human; to auto-locate them, divine”

—Ancient Tax Proverb

## Summary

Asset location is widely regarded as the closest thing there is to a “free lunch” in the wealth management industry.1 When investments are held in at least two types of accounts (out of three possible types: taxable, tax-deferred and tax-exempt), asset location has been demonstrated to deliver additional after-tax returns, while maintaining the same level of risk.

Generally speaking, this benefit is achieved by placing the least tax-efficient assets in the accounts taxed most favorably, and the most tax-efficient assets in the accounts taxed least favorably, all while maintaining the desired asset allocation in the aggregate.

However, intelligently applying asset location to a globally diversified portfolio is a complex, mathematically rigorous, and continuous undertaking. Betterment’s Tax-Coordinated Portfolio (TCP) is a dynamic asset location service that automates the process, tailoring its approach to each investor’s personal circumstances.

TCP is the first implementation of holistic asset location by an automated investment service, and is a milestone in the rapid advancement of investing technology. It makes this sophisticated strategy accessible to investors who either do not have the desire, expertise, or the time to effectively customize it, and manage it on an ongoing basis.

Due to its undeniable value for long-term investors, automated asset location will eventually become table stakes for digital advisors. We believe that its proliferation in the investment management industry will set the stage for the gradual decline of target-date funds, for reasons we discuss below.

TCP is seamlessly compatible with Tax Loss Harvesting+ (TLH+), Betterment’s automated tax loss harvesting service. While the latter derives value from the taxable account only, both work in tandem to improve after-tax return without disturbing the desired asset allocation.

In this paper, we discuss the various factors that must be considered by an optimal asset location strategy. We then present the methodology behind TCP, as well as results from performance simulations.

Our estimates of the annualized benefit range from 0.10% to 0.82%, depending on assumptions described below. For a typical customer saving in taxable, tax-deferred, and tax-exempt accounts over 30 years, we estimate the annualized benefit across the entire portfolio to be 0.48%. This equates to 15% more after-tax return available in retirement.

## Part I: Asset Location For All: “We Have The Technology!”

### Introduction

With each year, it is increasingly accepted that most investors are better off following a passive investment strategy, which is best implemented through a diversified portfolio of index funds. As passive investing has grown more popular, focus has shifted away from attempting to beat the market, and toward maximizing the net “take-home” value of the portfolio. This is done by reducing costs, which include fees and taxes, among other things.

Maximizing after-tax return on investments can be complex. Still, most investors know that contributing to tax-advantaged (or “qualified”) accounts is a relatively straightforward way to pay less tax on their retirement savings. Millions of Americans wind up with some combination of IRAs and 401(k) accounts, both available in two types: traditional or Roth.

Investors can see the impact of a deductible contribution when they do their taxes. They may be aware that a Roth contribution makes sense in a year when one’s tax bracket is unusually low. Many will only save in a taxable account once they have maxed out their contribution limits for the qualified accounts. But while tax considerations are paramount when choosing which account to fund, less thought is given to the tax impact of which investments to then purchase across all accounts.

Meanwhile, the tax profiles of the three account types (taxable, traditional, and Roth) have implications for what to invest in, once the account has been funded. Choosing wisely can significantly improve the after-tax value of one’s savings, when more than one account is in the mix.

Almost universally, such investors can benefit from a properly executed asset location strategy. The idea behind asset location is fairly straightforward. Certain investments generate their returns in a more tax-efficient manner than others. Certain accounts shelter investment returns from tax better than others. Placing, or “locating” less tax-efficient investments in tax-sheltered accounts should increase the after-tax value of the overall portfolio.

### Allocate First, Locate Second

Let’s start with what asset location isn’t. All investors must select a mix of stocks and bonds, finding an appropriate balance of risk and expected return, in line with their goals. One common goal is retirement, in which case, the mix of assets should be tailored to match the investor’s time horizon. This initial determination is known as “asset allocation,” and it comes first.

When investing in multiple accounts, it is common for investors to simply recreate their desired asset allocation in each account. On a superficial level, this makes sense. If each account, no matter the size, holds the same assets in the same proportions, adding up all the holdings will also match the desired asset allocation. If all these funds, however scattered, are invested towards the same goal, this is the right result. The aggregate portfolio is the one that matters, and it should track the asset allocation selected for the common goal.

#### Portfolio Managed Separately in Each Account

However, as long as the desired asset allocation is maintained in the aggregate, each individual account does not necessarily need to reproduce the target portfolio. Enter asset location, which can only be applied once a desired asset allocation is selected. Each asset’s after-tax return is considered in the context of every available account. The assets are then arranged (unequally) across all coordinated accounts to maximize the after-tax performance of the overall portfolio.

#### Same Portfolio Overall—With Asset Location

To help conceptualize asset location, consider a team of runners. Some runners compete better on a track than a cross-country dirt path, as compared to their more versatile teammates. Similarly, certain asset classes benefit more than others from the tax-efficient “terrain” of a qualified account.

Asset allocation determines the composition of the team, and the overall portfolio’s after-tax return is a team effort. Asset location then seeks to match up asset and environment in a way that maximizes the overall result over time, while keeping the composition of the team intact.

### Automatic for the People

Asset location has theoretically made sense since qualified accounts were introduced in the 1970s. More recently, however, the strategy has gained increasing importance, as financial advice has expanded to cater to a wider demographic.

The ultra-wealthy certainly take advantage of qualified accounts. However, due to contribution limits, such accounts typically represent a small percentage of a massive net worth (with exceptions, of course). Planning for such estates is more concerned with generational succession to wealth, rather than liquidation of all or most of a portfolio during one’s life.

The rest of us hold a substantial portion of our savings in qualified accounts, and intend for these assets to be our source of income in retirement. These circumstances make asset location particularly valuable. Therefore, a scalable implementation of the strategy is highly desirable for advisors looking to improve outcomes for the average investor.

This is easier said than done. Michael Kitces, a leading financial planner who has written extensively about asset location, summarizes both its potential and challenge:

Asset location represents one of those unique “free lunch” opportunities for wealth creation—a mechanism by which investment strategies that are already being implemented can simply be done in a more tax-efficient manner that maximizes long-term wealth creation. Yet in practice, the idea of a “free” lunch for asset location may be slightly overstated, in that the complexity of implementing it effectively requires more work, with a far more intensive and proactive process to evaluate investments for their prospective return and tax-efficiency characteristics, establish an asset location priority list to be utilized, and then actually implement it—ostensibly with the assistance of rebalancing/trading software—on an ongoing basis.2

In other words, the theoretical value of asset location is limited by the difficulty of proper execution. Indeed, the difficulty is twofold. First, the initial calculation of a client-specific location requires a mathematically rigorous approach to what is essentially a complex optimization problem.

Second, to maximize the benefit, this calculation must be performed continuously, in response to not only market movements and revised return expectations, but also the individual’s cash flows throughout the entire investment period. Deposits, withdrawals, rollovers, distributions, conversions and dividends will all shift the relative balances of the coordinated accounts. These shifts can necessitate location adjustments, big and small.

While these challenges can be daunting and cost-prohibitive when performed manually, they are easily handled by software, which is strongest when it comes to mathematically complex, repeatable tasks. Automating these tasks takes far more work than manually solving for a single client. However, once the formulas have been derived, coded into algorithms, and deployed to a production environment, the incremental work required to apply the strategy to additional accounts is minimal.

A team of Betterment quantitative analysts, tax experts, software engineers, designers, and product managers have been working for over a year, solving and automating all of these complexities. The result is software that can efficiently apply these calculations to millions of accounts. Betterment’s tightly integrated trading and position tracking systems allow for TCP’s decisions to be seamlessly executed for any individual client’s group of accounts, no matter the mix.

### RIP, TDF?

Automated asset location makes sense for so many investors, that we believe most advice services will eventually have to provide some version of it to protect their value proposition. Seen through that lens, we believe that the release of TCP is the first step toward the eventual decline of the target-date fund (TDF).

The primary appeal of a TDF is the “set it and forget it” simplicity with which it allows investors to select and maintain a diversified asset allocation, by purchasing only one fund. That simplicity comes at a price—because each TDF is a single, indivisible security, it cannot unevenly distribute its underlying assets across multiple accounts, and thus cannot deliver the additional after-tax returns of asset location. As soon as managing a single portfolio across multiple accounts becomes as easy as buying a TDF, its popularity should wane.

Of course, TDFs are not disappearing any time soon. Disruption takes time—after all, Research In Motion recorded its highest ever profits on the Blackberry in the two years after the iPhone was released, before they started to slide. The structural friction in the investment management industry is an order of magnitude higher, and the transition will be slow.

In particular, participants who are locked into 401(k) plans without automated management may find that a cheap TDF is still their best “hands off” option (plus, a TDF’s ability to satisfy the Qualified Default Investment Alternative (QDIA) requirement under ERISA ensures its baseline survival under current law).

But change is coming—today, participants in a Betterment for Business plan can already enable TCP to manage a single portfolio across their 401(k), and any IRAs and taxable accounts they individually have with Betterment, squeezing additional after-tax returns from their aggregate long-term savings.

Automated asset location (when integrated with automated asset allocation) replicates what makes a TDF so appealing, but effectively amounts to a “TDF 2.0″—an effortless, continuously managed portfolio, but one that can straddle multiple accounts for tax benefits.

Next, we dive into the complex dynamics that need to be considered when seeking to optimize after-tax return of a diversified portfolio.

## Part II: After-Tax Return—Deep Dive

The rules governing the taxation of investment income are dense and heavy on detail, but so was Ron Chernow’s “Alexander Hamilton,” and look how that turned out. We first define a few key terms, and then dive into specifics.

### What a Drag It Is Getting Taxed

A good starting point for a discussion of investment taxation is the concept of “tax drag.” As the name implies, a growing portfolio encounters friction through the years, in the form of annual taxation. Tax drag is the portion of the return that is lost to tax on an annual basis. In particular, funds pay dividends, which are taxed in the year they are received.

##### Gimme Shelter

However, there is no annual tax in qualified accounts, also sometimes known as “tax-sheltered accounts.” Therefore, placing assets that pay a substantial amount of dividends into a qualified account, rather than a taxable account, “shelters” those dividends, and reduces tax drag. Reducing the tax drag of the overall portfolio is one way that asset location improves the portfolio’s after-tax return.

Importantly, investments are also subject to tax at liquidation, both in the taxable account, and in a traditional IRA (where tax is deferred). However, “tax drag”, as that term it is commonly used, does not include liquidation tax. So while the concept of “tax drag” is intuitive, and thus a good place to start, it cannot be the sole focus when looking to minimize taxes.

A closely related term is “tax efficiency” and this is one that most discussions of asset location will inevitably focus on. A tax-efficient asset is one that has minimal “tax drag.” The less return that is lost to annual taxation, the more tax-efficient the asset generating that return is considered to be. By this definition, tax-efficient assets are typically those that generate most of their return through capital gains (taxed at liquidation), rather than through dividends.

It is helpful to think of tax efficiency as a relative concept. An asset is only tax-efficient in the sense that it has less tax drag than another asset in the portfolio. Prioritizing assets on the basis of tax efficiency allows for asset location decisions to be made following a simple, rule-based approach.

One goal of this paper is to demonstrate why this approach is too simple to be optimal. Since tax efficiency (as that term is commonly used) is only concerned with reducing annual tax drag, it has nothing to say about eventual liquidation tax. As we shall see, this narrow focus on a subset of the overall tax that must be paid makes tax efficiency an imperfect metric for trying to maximize after-tax return. However, the concept helps facilitate a basic understanding of how asset location adds a substantial amount of its value.

Because it is a useful device, tax efficiency is referenced frequently in the discussion that follows. However, it is not worth dwelling too much on whether some asset is truly more or less tax-efficient than another. The ultimate objective is to maximize after-tax return of the portfolio, and tax efficiency is merely one piece of that puzzle.

### Tax Rates, Timing, and Account Types, Oh My!

Both “tax drag” and “tax efficiency” are concepts pertaining to taxation of returns in a taxable account. Therefore, we first consider that account, where the rules are most elaborate. With an understanding of these rules, we can layer on the impact of the two types of qualified accounts.

### Returns in a Taxable Account

There are two types of investment income, and two types of applicable tax rates, but they do not correlate neatly. Their interaction is easier to describe if we define the rates first.

Two types of investment tax rates. All investment income in a taxable brokerage account is subject to one of two rate categories (with material exceptions noted). For simplicity, and to keep the analysis universal, this section only addresses federal tax (state tax is considered when testing for performance).

• Ordinary rate: For most, this rate mirrors the marginal tax bracket applicable to earned income (primarily wages reported on a W-2). For all but the lowest earners, that bracket will range from 25% to 39.6%.
• Preferential rate: This more favorable rate ranges from 15% to 20% for most investors.

For especially high earners, both rates are subject to an additional tax of 3.8%, making the highest possible ordinary and preferential rates 43.4% and 23.8%, respectively.

Two types of investment returns. Investments generate returns in two ways: by appreciating in value, and by making cash distributions.

• Capital gains: When an investment is sold, the difference between the proceeds and the tax basis (generally, the purchase price) is taxed as capital gains. If held for longer than a year, this gain is treated as long-term capital gains (LTCG) and taxed at the preferential rate. If held for a year or less, the gain is treated as short-term capital gains (STCG), and taxed at the ordinary rate. Barring unforeseen circumstances, passive investors should be able to avoid STCG entirely. Betterment’s automated account management avoids STCG,3 and the rest of this paper assumes only LTCG on liquidation of assets.
• Dividends: Bonds pay interest, which is taxed at the ordinary rate, whereas stocks pay dividends, which are taxed at the preferential rate (both subject to the exceptions below). An exchange-traded fund (ETF) pools the cash generated by its underlying investments, and makes payments that are called dividends, even if some or all of the source was interest. These dividends inherit the tax treatment of the source payments. This means that, generally, a dividend paid by a bond ETF is taxed at the ordinary rate, and a dividend paid by a stock ETF is taxed at the preferential rate.
• Qualified Dividend Income (QDI): There is an exception to the general rule for stock dividends. Stock dividends enjoy preferential rates only if they meet the requirements of qualified dividend income (QDI). Key among those requirements is that the company issuing the dividend must be a U.S. corporation (or a qualified foreign corporation). A fund pools dividends from many companies, only some of which may qualify for QDI. To account for this, the fund assigns itself a QDI percentage each year, which the custodian uses to determine the portion of the fund’s dividends that are eligible for the preferential rate.For stock funds tracking a U.S. index, the QDI percentage is typically 100%. However, funds tracking a foreign stock index will have a lower QDI percentage, sometimes substantially. For example, VWO, Vanguard’s Emerging Markets Stock ETF, had a QDI percentage of 38% in 2015, which means that 38% of its dividends for the year were taxed at the preferential rate, and 62% were taxed at the ordinary rate.
• Tax-exempt interest: There is also an exception to the general rule for bonds. Certain bonds pay interest that is exempt from federal tax. Primarily, these are municipal bonds, issued by state and local governments. This means that an ETF which holds municipal bonds will pay a dividend that is subject to 0% federal tax—even better than the preferential rate.

The chart below summarizes these interactions. Note that this section does not consider tax treatment for those in a marginal tax bracket of 15% and below. These taxpayers are addressed in “Special Considerations.”

Dividends (taxed annually) Capital Gains (taxed when sold)
Ordinary Rate Most bonds

Non-QDI stocks (foreign)

Any security held for a year or less (STCG)
Preferential Rate QDI stocks (domestic and some foreign) Any security held for more than a year (LTCG)
No Tax Municipal bonds Any security transferred upon death or donated to charity

The impact of rates is obvious: The higher the rate, the higher the tax drag. Equally important is timing. The key difference between dividends and capital gains is that the former are taxed annually, contributing to tax drag, whereas tax on the latter is deferred.

Tax deferral is a powerful driver of after-tax return, for the simple reason that the savings, though temporary, can be reinvested in the meantime, and compounded. The longer the deferral, the more valuable it is.

Putting this all together, we arrive at the foundational piece of conventional wisdom, where the most basic approach to asset location begins and ends:

• Bond funds are expected to generate their return entirely through dividends, taxed at the ordinary rate. This return benefits neither from the preferential rate, nor from tax deferral, making bonds the classic tax-inefficient asset class. These go in your qualified account.
• Stock funds are expected to generate their return primarily through capital gains. This return benefits both from the preferential rate, and from tax deferral. Stocks are therefore the most tax-efficient asset class. These go in your taxable account.

### Tax-Efficient Status: It’s Complicated

Reality gets messy rather quickly, however. Over the long term, stocks are expected to grow faster than bonds, causing the portfolio to drift from the desired asset allocation. Rebalancing may periodically realize some capital gains, so we cannot expect full tax deferral on these returns (although if cash flows exist, investing them intelligently can reduce the need to rebalance via selling).

Furthermore, stocks do generate some return via dividends. The expected dividend yield varies with more granularity. Small cap stocks pay relatively little (these are growth companies that tend to reinvest any profits back into the business) whereas large cap stocks pay more (as these are mature companies that tend to distribute profits). Depending on the interest rate environment, stock dividends can exceed those paid by bonds.

International stocks pay dividends too, and complicating things further, some of those dividends will not qualify as QDI, and will be taxed at the ordinary rate, like bond dividends (especially emerging markets stock dividends).

Already, we can see the simple conventional wisdom straining to account for this real-world complexity. How large of a dividend must a stock fund pay before we take the resulting tax drag seriously? Is an emerging markets stock fund really more tax-efficient than a bond fund? If a bond fund is tax-exempt, doesn’t that asset class make the most sense in the taxable account?

To have the full framework for addressing these questions, we next consider taxation of investment income in the qualified accounts, to the extent there is any.

### Returns in a Tax-Deferred Account (TDA)

Compared to a taxable account, a TDA is governed by deceptively simple rules. However, earning the same return in a TDA involves trade-offs which are not intuitive. Applying a different time horizon to the same asset can swing our preference between a taxable account and a TDA.

Understanding these dynamics is crucial to appreciating why an optimal asset location methodology cannot ignore liquidation tax, time horizon, and the actual composition of each asset’s expected return. In this section, we examine these trade-offs, which are some of the toughest to grasp.

Although growth in a traditional IRA or traditional 401(k) is not taxed annually, it is subject to a liquidation tax. All the complexity of a taxable account described above is reduced to two rules. First, all tax is deferred until distributions are made from the account, which should begin only in retirement. Second, all distributions are taxed at the same rate, no matter the source of the return.

The rate applied to all distributions is the higher ordinary rate, except that the additional 3.8% tax will not apply to those whose tax bracket in retirement would otherwise be high enough.4

First, we consider income that would be taxed annually at the ordinary rate (i.e. bond dividends and non-QDI stock dividends). The benefit of shifting these returns to a TDA is clear. In a TDA, these returns will eventually be taxed at the same rate, assuming the same tax bracket in retirement. But that tax will not be applied until the end, and compounding due to deferral can only have a positive impact on the after-tax return, as compared to the same income paid in a taxable account.5

The effect on returns that would otherwise be taxed at the preferential rate is more complex. For these returns, tax deferral comes at a price: when it does get taxed, it will be at the higher ordinary rate. There is potential for “negative tax arbitrage”—a conversion of lower-taxed income into higher-taxed income, which can lower the after-tax return. This runs counter to the goal of asset location, and needs to be considered very seriously.

In particular, the risk is that LTCG (which we expect plenty of from stock funds) will be taxed like ordinary income. Under the basic assumption that in a taxable account, capital gains tax is already deferred until liquidation, favoring a TDA for an asset whose only source of return is LTCG is plainly harmful. There is no benefit from deferral, which you would have gotten anyway, and only harm from a higher tax rate. This logic supports the conventional wisdom that stocks belong in the taxable account. Not only do bonds benefit more from a TDA, the reasoning goes, but stocks are actually hurt by it!

Again, reality is more complex. First, as already discussed, stocks do generate some return via dividends, and that portion of the return will benefit from tax deferral. This is obviously true for non-QDI dividends, already taxed as ordinary income, but QDI can benefit too. If the deferral period is long enough, the value of compounding will offset the hit from the higher rate at liquidation.

Second, it is not accurate to assume that all capital gains tax will be deferred until liquidation in a taxable account. Rebalancing may realize some capital gains “prematurely” and this portion of the return could also benefit from tax deferral.

Placing stocks in a TDA is a trade-off—one that must weigh the potential harm from negative rate arbitrage against the benefit of tax deferral. Valuing the latter means making assumptions about dividend yield and turnover. On top of that, the longer the investment period, the more tax deferral is worth. Kitces demonstrates that a dividend yield representing 25% of total return (at 100% QDI), and an annual turnover of 10%, could swing the calculus in favor of holding the stocks in a TDA, assuming a 30-year horizon.6 For foreign stocks with less than perfect QDI, we would expect the tipping point to come sooner.

All of the above suggests that we should be wary of overloading a TDA with assets expected to generate substantial capital gains, if the time horizon is relatively short. However, the goal of this section is not to precisely define the circumstances under which stocks perform better in a TDA. Rather, it is to demonstrate that the problem defies a “rule of thumb” approach. Not all stocks are the same, and time horizon matters. The solution must optimize around these trade-offs.

### Returns in a Tax-Exempt Account (TEA)

Investments in a Roth IRA or Roth 401(k) grow tax free, and are also not taxed upon liquidation. Since it eliminates all possible tax, a TEA presents a particularly valuable opportunity for maximizing after-tax return. The trade-off here is managing opportunity cost—every asset does better in a TEA, so how best to use its precious capacity?

Clearly, a TEA is the most favorably taxed account. Conventional wisdom thus suggests that if a TEA is available, we use it to first place the least tax-efficient assets. But that approach is wrong.

### Everything Counts in Large Amounts—Why Expected Return Matters

The powerful yet simple advantage of a TEA helps illustrate the limitation of focusing exclusively on tax efficiency when making location choices. Returns in a TEA escape all tax, whatever the rate or timing would have been, which means that an asset’s expected after-tax return equals its expected total return.

A tax-inefficient asset is one that loses a relatively large portion of its annual return to taxes, whereas a tax-efficient one loses a relatively small portion. However, ranking assets in this way does not incorporate the absolute size of their returns. A small portion of a large return could benefit from sheltering more than a large portion of a small return.

Even the least tax-efficient asset may not be the best candidate for permanent tax-avoidance, if its total return is expected to be relatively low. Savings from avoiding any annual tax do compound, but a small enough number will not amount to much. On the other hand, avoiding all tax on a large expected return may present a larger savings opportunity, even if that return is highly tax-efficient (taxed at liquidation, at the preferential rate).

Putting this in concrete terms, when both a taxable account and a TEA are available, it may be worth putting a high-growth, low-dividend stock fund into the TEA, instead of a bond fund, even though the stock fund is vastly more tax-efficient.

Not only that, but similar reasoning can apply to placement in a TDA as well, as long as the tax-efficient asset has a large enough expected return, and presents some opportunity for tax deferral (i.e., some portion of the return comes from dividends).

In other words, an optimal asset location must consider both tax efficiency and expected return. Tax efficiency has something to say about the degree to which an asset will benefit from a qualified account, but the absolute amount of expected savings is what actually matters.

## Part III: Asset Location in Popular Culture

To be sure, much like tax efficiency, “popular” is a relative concept. Still, much has been publicly written about asset location. The strategy is also (we assume) a topic of frequent rumor, conjecture, and innuendo. In this section, we look at existing approaches to asset location, and consider their advantages and limitations. But before diving into methodology, it is worth clearing up a few popular misconceptions.

### “The Tax Is Coming From Inside the Account!”: Debunking Some Urban Legends

##### Urban Legend 1: Asset location is a one-time process. Just set it and forget it.

While an initial location may add some value, doing it properly is a continuous process, and will require adjustments in response to changing conditions. Note that overlaying asset location is not a deviation from a passive investing philosophy, because optimizing for location does not mean changing the overall asset allocation (the same goes for tax loss harvesting).

Rick Ferri, a financial analyst and respected investing expert, has this to say:

The problem with asset location is that nothing is static. Tax rates change, tax brackets change, tax preferences change, and on and on. What was a logical tax location one year may turn out to be a lousy one a few years later, but you’re stuck in the one you have.7

Other things that will change, all of which should factor into an optimal methodology: expected returns (both the risk-free rate, and the excess return), dividend yields, QDI percentages, and most importantly, relative account balances. Contributions, rollovers, and conversions can increase qualified assets relative to taxable assets, continuously providing more room for additional optimization.

Ferri is also making the point that future location adjustments may be limited if the assets we want to relocate out of the taxable account have appreciated substantially (and adjusting our asset location would not be worth the tax cost of realizing those gains). This is a legitimate consideration, but should not prevent us from optimizing around what we know today. A rigorous strategy should weigh the costs and benefits of a potential future adjustment. In particular, during one’s working years, a steady stream of deposits (as well as dividends) will provide constant opportunities to move closer to whatever the optimal location is at that time.

##### Urban Legend 2: Taking advantage of asset location means you should contribute more to a particular qualified account than you otherwise would.

Definitely not! Asset location should play no role in deciding which accounts to fund. It optimizes around account balances as it finds them, and is not concerned with which accounts should be funded in the first place. Just because the presence of a TEA makes asset location more valuable, does not mean you should contribute to a TEA, as opposed to a TDA. That decision is primarily a bet on how your tax rate today will compare to your tax rate in retirement. To hedge, some may find it optimal to make contributions to both a TDA and TEA (this is called “tax diversification”). While these decisions are out of scope for this paper, Betterment’s RetireGuide service can help clients with these choices.

##### Urban Legend 3: Asset location has very little value if one of your accounts is relatively small.

It depends. Asset location will not do much for investors with a very small taxable balance and a relatively large balance in only one type of qualified account, because most of the overall assets are already sheltered. However, a large taxable balance and a small qualified account balance (especially a TEA balance) presents a better opportunity. Under these circumstances, there may be room for only the least tax-efficient, highest-return assets in the qualified account. Sheltering a small portion of the overall portfolio can deliver a disproportionate amount of value.

##### Urban Legend 4: Asset location has no value if you are investing in both types of qualified accounts, but not in a taxable account.

Not so. A TEA offers significant advantages over a TDA. Zero tax is better than a tax deferred until liquidation. While tax efficiency (i.e. annual tax drag) plays no role in these location decisions, expected returns and liquidation tax do. The assets we expect to grow the most should be placed in a TEA, and doing so will plainly increase the overall after-tax return. There is an additional benefit as well. Required minimum distributions (RMDs) apply to TDAs but not TEAs. Shifting expected growth into the TEA, at the expense of the TDA, will mean lower RMDs, giving the investor more flexibility to control taxable income down the road. In other words, a lower balance in the TDA can mean lower tax rates in retirement, if higher RMDs would have pushed the retiree into a higher bracket. This potential benefit is not captured in our results.

##### Urban Legend 5: Enough with the fancy talk. Bonds always go in the IRA, and that’s all there is to it.

Possibly, but not necessarily. This commonly asserted rule is a simplification, and will not be optimal under all circumstances. It is discussed at more length below.

### Existing Approaches to Asset Location: Advantages and Limitations

##### Optimizing for After-Tax Return While Maintaining Separate Portfolios

One approach to increasing after-tax return on retirement savings is to maintain a separate, standalone portfolio in each account with roughly the same level of risk-adjusted return, but tailoring each portfolio somewhat to take advantage of the tax profile of the account. Effectively, this means that each account separately maintains the desired exposure to stocks, while substituting certain asset classes for others.

Generally speaking, managing a fully diversified portfolio in each account means that there is no way to avoid placing some assets with the highest expected return in the taxable account.

Based on the discussion above, it should be obvious that a strategy which completely rules out the possibility of sheltering such assets cannot be optimal. The presence of a TEA, which is unambiguously best for the highest growth assets, only underscores this limitation.

This approach does include a valuable tactic, which is to differentiate the high-quality bonds component of the allocation, depending on the account they are held in. The allocation to the component is the same in each account, but in a taxable account, it is represented by municipal bonds which are exempt from federal tax (MUB), and in a qualified account, by taxable investment grade bonds (AGG).

This variation is effective because it takes advantage of the fact that these two asset classes have very similar characteristics (expected returns, covariance and risk exposures) allowing them to play roughly the same role from an asset allocation perspective. Municipal bonds, however, are highly tax-efficient, and are very compelling in a taxable account. Taxable investment grade bonds have significant tax drag, and work best in a qualified account. Betterment has applied this substitution since 2014. As we shall see, this MUB/AGG substitution is incorporated into TCP’s methodology.

##### The Basic Priority List

For those looking to unevenly distribute a single portfolio across multiple account types with no help from software, a rule-based approach has obvious appeal. The presence of multiple assets, and a highly personal ratio of account balances make for a complex problem. Most DIY investors and professional advisors are not mathematicians, so tackling the optimization with maximum precision is not viable. In other words, when implemented manually, this is a strategy crying for a heuristic—a simple set of rules that will produce an asset location which is hopefully “good enough.”

The easiest way to determine “what goes where” is to have confidence that at least with respect to certain assets, “this goes there first.” Armed with such conviction, the casual locator can start “filling up” the qualified account, prioritizing those assets which are perceived to benefit most from sheltering. Moving down the list requires only basic arithmetic—subtract the dollar amount allocated to an asset from the available balance, look to the next asset, and repeat until the qualified account is full (and what’s left goes in the taxable).

That list, more often than not, is a ranking of assets based solely on tax efficiency. At one extreme you may have say, high-yield bonds, which generate all of their returns as dividends, taxed annually at the ordinary rate (high tax drag). At the other extreme would be say, U.S. small cap stocks—mostly capital gains, and the few dividends they do generate are QDI, taxed at the preferential rate (very little tax drag). Everything else goes in between, with bonds and stocks clustering together on opposite ends.

This basic prioritization is the source for the conventional wisdom that “bonds go in the IRA.” As one would expect, the desire to reduce complexity, if taken too far, eventually leads to suboptimal results.

One immediately obvious exception is that municipal bonds are maximally tax-efficient, and belong in a taxable account. Still, other bonds are, in fact, highly tax-inefficient. Yet, as many have pointed out, in the current low-yield environment, bonds just do not return that much, so the potential savings are low. Is there a hidden opportunity cost to blindly prioritizing bonds in the qualified accounts, no matter what?

##### “Smile” Like You Mean It

Numerous experts have published research on how to best navigate the complex trade-offs discussed so far. The conclusion is more often than not: “It just depends.” However, manual asset locators need an approach that is more sophisticated than prioritization by tax efficiency, while still being rule-based and thus easy to implement.

Gobind Daryanani and Chris Cordaro sought to balance considerations around tax efficiency and expected return, and illustrated that when both are very low, location decisions with respect to those assets have very limited impact.8 That study inspired Michael Kitces, who leverages its insights into a more sophisticated approach to building a priority list.9 To visually capture the relationship between the two considerations, Kitces bends the one-dimensional list into a “smile.”

#### Asset Location Priority List

The rendering is both effective, and delightfully simple. Assets with a high expected return that are also very tax-efficient go in the taxable account. Assets with a high expected return that are also very tax-inefficient go in the qualified accounts, starting with the TEA. The “smile” guides us in filling the accounts from both ends simultaneously, and by the time we get to the middle, whatever decisions we make with respect to those assets just “don’t matter” much.

It’s clear why this heuristic is better than a straight ranking—plotting the assets in two dimensions allows the location decisions to factor in the interaction between two independent variables, both of which are important. Prioritizing with a “smile” means that the valuable real estate in the qualified accounts will not be occupied by assets that may have high tax drag, but very little to actually drag.

However, Kitces augments the graph in short order, recognizing that the basic “smile” does not capture a third key consideration—the impact of liquidation tax. Because capital gains will eventually be realized in a taxable account, but not in a TEA, even a highly tax-efficient asset might be better off in a TEA, if its expected return is high enough. The next iteration of the “smile” illustrates this preference.

#### With Limited High Return Inefficient Assets

A two-dimensional representation of a relationship between three variables immediately raises questions. Just how high must its return be, for a tax-efficient asset to bump a tax-inefficient asset out of the TEA? How high must its tax drag be, for that tax-inefficient asset to retain its tax-exempt perch? Unlike at the bottom of the smile, these are decisions that “matter,” but the rules are no longer clear. Two dimensions are not enough!

To push the optimization further, we need to revisit the limitation of “tax efficiency” as a metric. It is a poor measure for what we are actually seeking to maximize, which is after-tax return. Tax efficiency, as it is used in asset location discussions, is only concerned with annual tax drag in the taxable account. However, “liquidation tax drag” also exists, and varies across all accounts (moderate in the taxable, highest in a TDA, and absent in a TEA).

In other words, we must also attempt to quantify “liquidation tax efficiency,” but doing so is complex, because it involves all three accounts, whereas the accepted definition of tax efficiency described the taxable account only. We would need to derive an annualized after-tax return for each asset, in each account. This metric is driven by the expected return, but also incorporates both annual and liquidation tax, and is sensitive to time horizon. Suddenly, we are quite far from rules of thumb, lists, and graphs.

This is not to say that investors taking advantage of a sophisticated heuristic, whether through their own efforts, or with the help of a skilled advisor, will not select a beneficial asset location. Far from it. As Kitces summarizes, nailing the corners of his “smile” and not worrying about the middle will undoubtedly add value: “Just get those two right and you’ve done yourself a ton of good.”10 We wholeheartedly agree.

With automation, we should be able to do even better. Too many important factors cannot be accounted for with a prioritization approach. However many rules we devise, we will always land on “good enough” when the answer is not apparent. Only a mathematical optimization can squeeze maximum benefit from a given set of inputs.

## Part IV: TCP Methodology

By now, it should be clear that there is no one-size-fits-all asset location for every set of inputs. Some circumstances apply to all investors, but shift through time—the expected return of each asset class (which combines separate assumptions for the risk-free rate and the excess return), as well as dividend yields, QDI percentages, and tax laws. Other circumstances are personal—which accounts the client has, the relative balance of each account, and the client’s time horizon.

Solving for multiple variables while respecting defined constraints is a problem that can be effectively solved by linear optimization. This method is used to maximize some value, which is represented by a formula called an “objective function.” What we seek to maximize is the after-tax value of the overall portfolio at the end of the time horizon.

We get this number by adding together the expected after-tax value of every asset in the portfolio, but because each asset can be held in more than one account, each portion must be considered separately, by applying the tax rules of that account. We must therefore derive an account-specific expected after-tax return for each asset.

### Deriving Account-Specific After-Tax Return

To define the expected after-tax return of an asset, we first need its total return (i.e., before any tax is applied). The total return is the sum of the risk-free rate (same for every asset) and the excess return (unique to every asset). Betterment derives excess returns using the Black-Litterman model as a starting point. This common industry method involves analyzing the global portfolio of investable assets and their proportions, and using them to generate forward-looking expected returns for each asset class.

Next, we must reduce each total return into an after-tax return.11 The immediate problem is that for each asset class, the after-tax return can be different, depending on the account, and for how long it is held.

1. In a TEA, the answer is simple—the after-tax return equals the total return—no calculation necessary.
2. In a TDA, we project growth of the asset by compounding the total return annually. At liquidation, we apply the ordinary rate to all of the growth.12 We use what is left of the growth after taxes to derive an annualized return, which is our after-tax return.
3. In a taxable account, we need to consider the dividend and capital gain component of the total return separately, with respect to both rate and timing. We project growth of the asset by taxing the dividend component annually at the ordinary rate (or the preferential rate, to the extent that it qualifies as QDI) and adding back the after-tax dividend (i.e., we reinvest it). Capital gains are deferred, and the LTCG is fully taxed at the preferential rate at the end of the period. We then derive the annualized return based on the after-tax value of the asset.13

Note that for both the TDA and taxable calculations, time horizon matters. More time means more value from deferral, so the same total return can result in a higher annualized after-tax return. Additionally, the risk-free rate component of the total return will also depend on the time horizon, which affects all three accounts.

William Reichenstein and William Meyer express a key insight into what this really means:

“The same asset, whether stock or bond, is effectively a different asset when held in a TDA or taxable account.”14

Because we are accounting for the possibility of a TEA, as well, we actually have three distinct after-tax returns, and thus each asset effectively becomes three assets, for any given time horizon (which is specific to each Betterment customer).

### The Objective Function

To see how this comes together, we first consider an extremely simplified example. Let’s assume we have a taxable, traditional and Roth account, $50,000 in each one, and a 30-year horizon. Our allocation calls for only two assets: 70% equities (stocks) and 30% fixed income (bonds). With a total portfolio value of$150,000, we need $105,000 of stocks and$45,000 of bonds.

1. These are constants whose value we already know (as derived above):

$$\hat{r}_{eq,tax}$$ is the after-tax return of stocks in the taxable account, over 30 years
$$\hat{r}_{eq,trad}$$ is the after-tax return of stocks in the traditional account, over 30 years
$$\hat{r}_{eq,roth}$$ is the after-tax return of stocks in the Roth account, over 30 years
$$\hat{r}_{fi,tax}$$ is the after-tax return of bonds in the taxable account, over 30 years
$$\hat{r}_{fi,trad}$$ is the after-tax return of bonds in the traditional account, over 30 years
$$\hat{r}_{fi,roth}$$ is the after-tax return of bonds in the Roth account, over 30 years

2. These are the values we are trying to solve for (called “decision variables”):

$$x_{eq,tax}$$ is the amount of stocks we will place in the taxable account
$$x_{eq,trad}$$ is the amount of stocks we will place in the traditional account
$$x_{eq,roth}$$ is the amount of stocks we will place in the Roth account
$$x_{fi,tax}$$ is the amount of bonds we will place in the taxable account
$$x_{fi,trad}$$ is the amount of bonds we will place in the traditional account
$$x_{fi,roth}$$ is the amount of bonds we will place in the Roth account

3. These are the constraints which must be respected. All positions for each asset must add up to what we have allocated to the asset overall. All positions in each account must add up to the available balance in each account.

\begin{align*} x_{eq,tax}+x_{eq,trad}+x_{eq,roth}&=105,000\\ x_{fi,tax}+x_{fi,trad}+x_{fi,roth}&=45,000\\ x_{eq,tax}+x_{fi,tax}&=50,000\\ x_{eq,trad}+x_{fi,trad}&=50,000\\ x_{eq,roth}+x_{fi,roth}&=50,000\\\end{align*}

4. This is the objective function, which uses the constants and decision variables to express the after-tax value of the entire portfolio, represented by the sum of six terms (the after-tax value of each asset in each of the three accounts).

$$\begin{multline*} \underset{x}{max} \quad \hat{r}_{eq,tax}x_{eq,tax} + \hat{r}_{eq,trad}x_{eq,trad} + \hat{r}_{eq,roth}x_{eq,roth}\\ + \hat{r}_{fi,tax}x_{fi,tax} + \hat{r}_{fi,trad}x_{fi,trad} + \hat{r}_{fi,roth}x_{fi,roth} \end{multline*}$$

Linear optimization turns all of the above into a complex geometric representation, and mathematically closes in on the optimal solution. It assigns values for all decision variables in a way that maximizes the value of the objective function, while respecting the constraints. Accordingly, each decision variable is a precise instruction for how much of which asset to put in each account. If a variable comes out as zero, then that particular account will contain none of that particular asset.

An actual Betterment portfolio can have up to twelve asset classes,15 depending on the allocation. That means TCP must effectively handle up to 36 “assets”, each with its own after-tax return. However, the full complexity behind TCP goes well beyond increasing assets from two to twelve.

Updated constants and constraints will trigger another part of the optimization, which determines what TCP is allowed to sell, in order to move an already coordinated portfolio toward the newly optimal asset location, while minimizing taxes. Reshuffling assets in a TDA or TEA is “free” in the sense that no capital gains will be realized.16 In the taxable account, however, TCP will attempt to move as close as possible towards the optimal asset location without realizing capital gains.

The full optimization on an actual set of accounts must solve for hundreds of variables subject to hundreds of constraints, and reproducing the full objective function would take many pages. Good news! The computer will do it. Even better news: Because the calculation is automated, it can run every time any of our constants or constraints change, using every opportunity to either maintain, or move closer to the optimal asset location.

Expected returns will periodically be updated, either because the risk-free rate has been adjusted, or because new excess returns have been derived via Black-Litterman.

Future cash flows may be even more material. Additional funds in one or more of the accounts could significantly alter the constraints which define the size of each account, and the target dollar allocation to each asset class. Such events (including dividend payments, subject to a de minimis threshold) will trigger a recalculation, and potentially a reshuffling of the assets.

Cash flows, in particular, can be a challenge for those managing their asset location manually. Inflows to just one account (or to multiple accounts in unequal proportions) create a tension between optimizing asset location and maintaining asset allocation, which is hard to resolve without mathematical precision.

To maintain the overall asset allocation, each position in the portfolio must be increased pro-rata. However, some of the additional assets we need to buy “belong” in other accounts from an asset location perspective, even though new cash is not available in those accounts. If the taxable account can only be partially reshuffled due to built-in gains, we must choose either to move farther away from the target allocation, or the target location.

With linear optimization, our preferences can be expressed through additional constraints, weaving these considerations into the overall problem. When solving for new cash flows, TCP penalizes allocation drift higher than it does location drift—another complexity not represented by the simplified objective function above.

### Interactive Illustration of TCP Methodology

The interactive tool below approximates the location decisions made by the optimizer behind TCP, reflecting assumptions we would make today. The ON/OFF switch toggles asset location for the set of accounts. The overall allocation, relative account balances, and time horizon are adjustable, and hovering over each circle shows the asset class it represents.

The tool helps illustrate the target asset location that TCP will manage toward, but not necessarily achieve in an actual Betterment account, especially an account that predates TCP’s release, due to its aversion to realizing capital gains (see more below under “Enabling TCP On Existing Taxable Accounts”).17 However, it is a useful approximation of the decisions being made by the linear optimizer, based on our views today.

For example, it is no surprise that high-quality bonds find their way into the taxable account, in the form of municipal bonds. Emerging markets stocks, and to a lesser extent, developed markets stocks, show a strong preference for qualified accounts, especially a TEA, and take preference there over bonds. Taxable bonds tend to gravitate towards the TDA, and U.S. stocks tend to gravitate towards the taxable, but not in all cases. In particular, time horizon matters—toggling between ten and thirty years reverses certain preferences.

### All the Math in the World Cannot Predict the Future

We can reason our way into these choices, though some may be less intuitive than others. We can debate the many assumptions discussed above—changing any of them may impact these decisions. Linear optimization is only as good as its inputs, but extremely effective within that framework.

Still, even linear optimization may not land on what will turn out to have been the optimal asset location, in retrospect. Events nobody can predict make this improbable, especially over a long horizon. But that is no reason not to select an optimized asset location based on what is known today, adjusting when new information becomes available. When asset location is automated, these adjustments are made centrally, and seamlessly propagated across all accounts. This makes the ongoing component of location management extremely efficient and consistent.

However, it cannot be emphasized enough that expected returns (the key input into TCP, and portfolio management generally) are educated guesses at best. No matter how airtight the math, reasonable people will disagree on the “correct” way to derive them. The future may not cooperate, making a mockery of our best-laid plans, especially in the short-term. There is no guarantee that any particular asset location will add the most value, or even any value at all. But given decades, our certainty in the outcome grows.

## Part V: Monte Carlo on the Amazon—Betterment’s Testing Framework

To test the output of the linear optimization method, we turned to a rigorous Monte Carlo testing framework,18 built entirely in-house by Betterment’s experts. The forward-looking simulations model the behavior of the TCP strategy down to the individual lot level. We simulate the paths of these lots, accounting for dividend reinvestment, rebalancing, and taxation. As far as we are aware, this level of analysis for an asset location strategy has never been made public in the investment management industry.

The forward market scenarios were derived from parametric sampling of a multivariate Gaussian distribution, where the mean was estimated using an equilibrium forecast method. Each set of assumptions was subject to at least 1,000 market scenarios. The covariance matrix was generated from historical returns. Forward dividends were simulated by sampling from a univariate Gaussian distribution, with the first and second moments derived from historical data, and with the same current dividend frequency. For QDI, we assumed the percentages reported by funds for 2015, over the whole period.

The simulations applied Betterment’s rebalancing methodology, which corrects drift from the target asset allocation in excess of 3%, but stops short of ever realizing STCG.
Betterment’s quarterly management fees were assessed in all accounts, and ongoing taxes were paid annually from the taxable account. All taxable sales first realized available losses before touching LTCG.

The simulations assumed no additional cash flows other than dividends. This is not because we do not expect them to happen. Rather, it is because making assumptions around these very personal circumstances does nothing to isolate the benefit of TCP specifically. Asset location is driven by the relative sizes of the accounts, and cash flows will change these ratios, but the timing and amount is highly specific to the individual.19 Avoiding the need to make specific assumptions here helps keep the analysis more universal. We used equal starting balances for the same reason.20

### R&D in the Cloud: Harder, Better, Faster, Stronger

Betterment is hosted by Amazon Web Services (AWS), which allows us to dynamically increase or decrease the computing power available to our software in response to variable demand. The purpose of this infrastructure is to provide a consistently robust experience to customers in a scalable and efficient manner. However, an added benefit is that our R&D team has access to a platform that can temporarily support an extremely demanding computing load.

Tracking positions at the tax lot level means accounting for basis on the fly, even for small trades, which Betterment executes using fractional shares. Each taxable sale is processed through our TaxMin lot selection algorithm, and all transactions (including purchases in qualified accounts) are run through our wash sale avoidance algorithms, which are described in our TLH+ white paper. All of this means that a 30-year simulation runs for multiple hours.

AWS allows us to spin up thousands of servers simultaneously, each running a different market scenario in parallel. Each set of assumptions (combination of accounts, stock allocation) was tested under at least 1,000 market scenarios. The research presented here is the product of over 150,000 computer-hours, at times running on up to 3,000 AWS servers simultaneously.

For every set of assumptions, we ran each market scenario while managing each account as a standalone (uncoordinated) Betterment portfolio as the benchmark.21 We then ran the same market scenarios with TCP enabled. In both cases, we calculated the after-tax value of the aggregate portfolio after full liquidation at the end of the period.22 Then, for each market scenario, we calculated the after-tax annualized internal rates of return (IRR) and subtracted the benchmark IRR from the TCP IRR. That delta represents the incremental tax alpha of TCP for that scenario. The median of those deltas across all market scenarios is the estimated tax alpha we present below for each set of assumptions.

It is important to note that these results do not express the value of Betterment’s management vis-à-vis some particular non-Betterment investing strategy. Rather, they measure the incremental tax alpha of TCP, as compared to Betterment’s management of its uncoordinated portfolios in the same accounts. Still, these results are a reasonable measure of the effectiveness of Betterment’s implementation of asset location as applied to a passive investing strategy.

## Part VI: Who’s Up With TCP? Our Results

If you have read this far, you are truly a tax management paladin, and our hats are off to you. We should hang out sometime. If you skipped ahead to get here, that’s okay too, as long as you promise to read the “Special Considerations” section, and the disclosures.

### 4. TAX/TRAD/ROTH (High tax, but Moderate tax for liquidation)

Asset Allocation Additional Tax Alpha with TCP (Annualized)
50% Stocks 0.77%
70% Stocks 0.49%
90% Stocks 0.34%

### 5. Two Accounts: TAX/TRAD (High tax, but Moderate tax for liquidation)

Asset Allocation Additional Tax Alpha with TCP (Annualized)
50% Stocks 0.54%
70% Stocks 0.27%
90% Stocks 0.20%

### 6. Two Accounts: TAX/ROTH (High tax, but Moderate tax for liquidation)

Asset Allocation Additional Tax Alpha with TCP (Annualized)
50% Stocks 0.71%
70% Stocks 0.46%
90% Stocks 0.38%
##### More Bonds, More Alpha

The clearest pattern is that a higher allocation to bonds leads to a dramatically higher benefit across the board. This makes sense—the heavier your allocation to tax-inefficient assets, the more asset location can do for you. To be extremely clear: this is not a reason to select a lower allocation to stocks! Over the long-term, we expect a higher stock allocation to return more (because it’s riskier), both before, and after tax. These are measurements of the additional return due to TCP, which say nothing about the absolute return of the asset allocation itself.

Conversely, a very high allocation to stocks shows a smaller (though still real) benefit. However, younger customers invested this aggressively should gradually reduce risk as they get closer to retirement (to something more like 50% stocks). Looking to the 70% stock allocation is therefore an imperfect but reasonable way to generalize the value of the strategy over a 30-year period.

##### More Roth, More Alpha

Another pattern is that the presence of a Roth makes the strategy more valuable. This also makes sense—a taxable account and a TEA are on opposite ends of the “favorably taxed” spectrum, and having both presents the biggest opportunity for TCP’s “account arbitrage.” But again, this benefit should not be interpreted as a reason to contribute to a TEA over a TDA, or to shift the balance between the two via a Roth conversion. These decisions are driven by other considerations. TCP’s job is to optimize the relative balances as it finds them.

##### Enabling TCP On Existing Taxable Accounts

All of these results assume that TCP is enabled before the taxable account is funded, meaning that the initial location can be optimized without the need to sell potentially appreciated assets. A Betterment customer with an existing taxable account who enables TCP should not expect the full incremental benefit we project here, to the extent that assets with built-in capital gains need to be sold to achieve the optimal location.

This is because TCP conservatively prioritizes avoiding a certain tax today, over potentially reducing tax in the future. However, the optimization is performed every time there is a deposit (or dividend) to any account. With future cash flows, the portfolio will move closer to whatever the optimal location is determined to be at the time of the deposit.

## Part VII: Special Considerations

### Low Bracket Taxpayers: Beware

Taxation of investment income is substantially different for those who qualify for a marginal tax bracket of 15% or below. To illustrate, we have modified the chart from Part II to apply to such low bracket taxpayers.

Dividends Capital Gains
Ordinary Rate N/A Any security held for a year or less (STCG)
Preferential Rate N/A N/A
No Tax Qualified dividends from any security are not taxed Any security held for a year or more is not taxed (LTCG)

TCP is not designed for these investors. Optimizing around this tax profile would reverse many assumptions behind TCP’s methodology. Municipal bonds no longer have an advantage over other bond funds. The arbitrage opportunity between the ordinary and preferential rate is gone. In fact, there’s barely tax of any kind. It is quite likely that such investors would not benefit much from TCP, and may even reduce their overall after-tax return.

If the low tax bracket is temporary, TCP over the long-term may still make sense.  Also note that some combinations of account balances can, in certain circumstances, still add tax alpha for investors in low tax brackets. One example is when an investor only has traditional and Roth IRA accounts, and no taxable accounts being tax coordinated. Low bracket investors should very carefully consider whether TCP is suitable for them. As a general rule, we do not recommend it.

### Potential Problems with Coordinating Accounts Meant for Different Time Horizons

We began with the premise that asset location is sensible only with respect to accounts that are generally intended for the same purpose. This is crucial, because unevenly distributing assets will result in asset allocations in each account that are not tailored towards the overall goal (or any goal at all). This is fine, as long as we expect that all coordinated accounts will be available for withdrawals at roughly the same time (e.g. at retirement). Only the aggregate portfolio matters in getting there.

However, uneven distributions are less diversified. Temporary drawdowns (e.g., the 2008 financial crisis) can mean that a single account may drop substantially more than the overall coordinated portfolio. If that account is intended for a short-term goal, it may not have a chance to recover by the time you need the money. Likewise, if you do not plan on depleting an account during your retirement, and instead plan on leaving it to be inherited for future generations, arguably this account has a longer time horizon than the others and should thus be invested more aggressively. In either case, We do not recommend managing accounts with materially different time horizons as a single portfolio.

For a similar reason, you should avoid applying asset location to an account that you expect will be long-term, but one that you may look to for emergency withdrawals. For example, a Safety Net Goal should never be managed by TCP.

### Large Upcoming Transfers/Withdrawals

If you know you will be making large transfers in or out of your tax-coordinated accounts, you may want to delay enabling our tax coordination tool until after those transfers have occurred.

This is because large changes in the balances of the underlying accounts can necessitate rebalancing, and thus may cause taxes. With incoming deposits, we can intelligently rebalance your accounts by purchasing asset classes that are underweight. But when large withdrawals or transfers out are made, despite Betterment’s intelligent management of executing trades, some taxes can be unavoidable when rebalancing to your overall target allocation.

The only exception to this rule is if the large deposit will be in your taxable account instead of your IRAs. In that case, you should enable tax-coordination before depositing money into the taxable account. This is so our system knows to tax-coordinate you immediately.

The goal of tax coordination is to reduce the drag taxes have on your investments, not cause additional taxes. So if you know an upcoming withdrawal or outbound transfer could cause rebalancing, and thus taxes, it would be prudent to delay enabling tax coordination until you have completed those transfers.

### Mitigating Behavioral Challenges Through Design

There is a broader issue that stems from locating assets with different volatility profiles at the account level, but it is behavioral. Uncoordinated portfolios with the same allocation move together. Asset location, on the other hand, will cause one account to dip more than another, testing an investor’s stomach for volatility. Those who enable TCP across their accounts should be prepared for such differentiated movement. Rationally, we should ignore this—after all, the overall allocation is the same—but that is easier said than done.

Rick Ferri has an interesting example of how this played out during the last big downturn:

“[T]here is a hidden risk with having different allocations in taxable versus non-taxable, and we saw this risk turn into reality during 2008 and early 2009. A few clients terminated their higher risk taxable portfolio because that specific portfolio was losing more money than the more conservative non-taxable portfolio. In other words, they separated their portfolios in their mind and compared returns rather than looking at the big picture.”23

This is an important consideration, and should not be ignored. A good advisor should seek to maximize a client’s returns net of tax, fees, and behavior. So while this effect is real, we believe that intelligent product design can mitigate it somewhat for the modern investor.

Clients using a digital investment service can access their balance and returns in real-time, and in our experience, rely far less on traditional statements, which are account-specific. Generally, this is behaviorally undesirable: an unavoidable cost of online investing.

In this case, however, it presents an opportunity. App design has more latitude in presenting information than a statement does. A well-designed interface can surface aggregate performance only, nudging the client towards the “big picture,” and deprioritizing counterproductive information.24 Here is an example of how TCP shows summary information for multiple coordinated accounts:

So while a client can still access each account’s balance and infer account-level performance, behaviorally sensitive UX can make an impact on what clients focus on.

### How TCP Interacts with Tax Loss Harvesting+

Asset location is highly compatible with tax loss harvesting. While the latter derives value from the taxable account only, both work in tandem to improve after-tax return. The reason they work well together is that both derive their benefit without disturbing the desired asset allocation.

##### Operational Compatibility

Betterment’s implementations of both strategies are especially compatible, because of certain design decisions made when building TLH+, Betterment’s automated tax loss harvesting service. For that strategy, effective wash sale management was paramount.

Special attention was paid to avoiding a type of wash sale that permanently disallows a realized loss. This can happen when a security is sold at a loss in a taxable account, and a substantially identical security is purchased in a qualified account within the wash sale window. TLH+ was designed around a “tertiary ticker” system, which ensures that no purchase in an IRA or 401(k) managed by Betterment will interfere with a harvested loss in a Betterment taxable account.

A sale in a taxable account, and a subsequent repurchase of the same asset class in a qualified account would be incidental for accounts managed as separate portfolios. Under TCP, however, we expect this to occasionally happen by design. When “relocating” assets, either during initial setup, or as part of ongoing optimization, TCP will sell an asset class in one account, and immediately repurchase it in another. The tertiary ticker system allows this reshuffling to happen seamlessly, while protecting any tax losses that are realized in the process.

##### Conceptualizing Blended Performance

Estimating the benefit of TCP and TLH on a standalone basis already requires making many generalized assumptions. However, modeling their combined benefit is even more difficult to generalize. Simulations of both strategies must be at the tax lot level, and combined performance is highly path dependent. They are expected to add more value together than either one would on its own, but simply adding up the two standalone estimates is clearly inaccurate, and a better answer is elusive.

TCP will affect the composition of the taxable account in ways that are hard to predict, because its decisions will be driven by changes in relative balances among the accounts. Meanwhile, the weight of specific asset classes in the taxable account is a material predictor of the potential value of TLH (more volatile assets should offer more harvesting opportunities). The precise interaction between the two strategies is far more dependent on personal circumstances, such as today’s account balance ratios and future cash flow patterns, than on generally applicable inputs like asset class return profiles and tax rules.

These dynamics are best understood as a hierarchy. Asset allocation comes first, and determines what mix of asset classes we should stick to overall. Asset location comes second, and continuously generates tax alpha across all coordinated accounts, within the constraints of the overall portfolio. Tax loss harvesting comes third, and looks for opportunities to generate tax alpha from the taxable account only, within the constraints of the asset mix dictated by asset location for that account.

Decision logic at each tier is encapsulated, following the architectural principle of “information hiding.” To illustrate in concrete terms: TCP knows only that the overall allocation must be 70% stocks, but should not be concerned with what drove that determination. Similarly, TLH does not need to know why the taxable account must hold (for example) municipal bonds and domestic equities only. Both seek to optimize after-tax return within their domain, playing the hand they are dealt.

To sum up, it is worth highlighting why asset location sets the parameters for tax loss harvesting, and not the other way around. TLH is most effective in the first several years after an initial deposit to a taxable account. Over decades, however, we expect it to generate value only from subsequent deposits and dividend reinvestments. Eventually, even a substantial dip is unlikely to bring the market price below the purchase price of the older tax lots. Meanwhile, TCP continues to deliver tax alpha over the entire balance of all three accounts for the entire holding period.

## Conclusion

An investment strategy intended to maximize the after-tax value of a portfolio should focus on increasing annualized after-tax return. Traditional approaches to asset location are concerned with prioritizing assets in certain accounts based on their relative “tax efficiency”, with the objective of sheltering income from annual tax.

This narrow focus ignores numerous factors which affect annualized after-tax return, but are hard to account for without a mathematically rigorous approach. To reach its potential, an optimal asset location methodology must incorporate these realities, which include liquidation tax, time horizon, expected total return, and the actual composition of each expected return.

Automation provides us with an opportunity to tackle these complexities. Betterment’s TCP considers both generally applicable inputs and customer-specific circumstances when optimizing an asset location for each set of accounts.

TCP’s reliance on linear optimization makes it a “living implementation” of asset location. As material assumptions change, TCP can be easily adjusted across the entire customer base, without the need to rewrite its fundamental rules. Accordingly, the service can optimize after-tax returns for Betterment’s customers based on today’s expectations, without sacrificing operational flexibility, which will be utilized in the years to come.

1Boost Your After-Tax Investment Returns.” Susan B. Garland. Kiplinger.com, April 2014.

2Pg. 13, The Kitces Report. March/April 2014.

3This does not include withdrawals or customer-directed allocation changes. If these are substantial enough, realizing STCG may be unavoidable. In such cases, Betterment customers are notified before they confirm their instruction, via Tax Impact Preview.

4But see “How IRA Withdrawals In The Crossover Zone Can Trigger The 3.8% Medicare Surtax,” Michael Kitces, July 23, 2014.

5It is worth emphasizing that asset location optimizes around account balances as it finds them, and has nothing to say about which account to fund in the first place. Asset location considers which account is best for holding a specified dollar amount of a particular asset. However, contributions to a TDA are tax-deductible, whereas getting a dollar into a taxable account requires more than a dollar of income.

6Pg. 5, The Kitces Report. January/February 2014.

7 Problems with Reichenstein’s “Asset Location Decision Revisted.” Bogleheads.org. Nov. 10, 2013.

8Daryanani, Gobind, and Chris Cordaro. 2005. “Asset Location: A Generic Framework for Maximizing After-Tax Wealth.” Journal of Financial Planning (18) 1: 44–54.

9The Kitces Report, March/April 2014.

10 Minimizing the Tax Drag on Your Investments,” By Carla Fried. The New York Times, Feb. 7, 2014.

11While the significance of ordinary versus preferential tax treatment of income has been made clear, the impact of an individual’s specific tax bracket has not yet been addressed. Does it matter which ordinary rate, and which preferential rate is applicable, when locating assets? After all, calculating the after-tax return of each asset means applying a specific rate. It is certainly true that different rates should result in different after-tax returns. However, we found that while the specific rate used to derive the after-tax return can and does affect the level of resulting returns for different asset classes, it makes a negligible difference on resulting location decisions. The one exception is when considering using very low rates as inputs (the implication of which is discussed under “Special Considerations”). This should feel intuitive: Because the optimization is driven primarily by the relative size of the after-tax returns of different asset classes, moving between brackets moves all rates in the same direction, generally maintaining these relationships monotonically. The specific rates do matter a lot when it comes to estimating the benefit of the asset location chosen, so rate assumptions are laid out in the “Results” section. In other words, if one taxpayer is in a moderate tax bracket, and another in a high bracket, their optimal asset location will be very similar and often identical, but the high bracket investor may benefit more from the same location.

12In reality, the ordinary rate is applied to the entire value of the TDA, both the principal (i.e., the deductible contributions) and the growth. However, this will happen to the principal whether we use asset location or not. Therefore, we are measuring here only that which we can optimize.

13TCP today does not account for the potential benefit of a foreign tax credit (FTC). The FTC is intended to mitigate the potential for double taxation with respect to income that has already been taxed in a foreign country. The scope of the benefit is hard to quantify and its applicability depends on personal circumstances. All else being equal, we would expect that incorporating the FTC may somewhat increase the after-tax return of certain asset classes in a taxable account—in particular developed and emerging markets stocks. If maximizing your available FTC is important to your tax planning, you should carefully consider whether TCP is the optimal strategy for you.

14Note that Reichenstein and Meyer develop this concept to optimize asset allocation—an approach known as “after-tax asset allocation.” They stress that pre-tax contributions to TDAs are “partnerships”, and the government is effectively a minority partner, with an interest in part of the return and principal. This view has implications for what the asset allocation should be, when factoring in liquidation tax. After-tax asset location prioritizes allocation over after-tax return, and is not without its detractors. TCP today does not optimize for after-tax allocation.

15One component of the portfolio gets special treatment, as previewed in an earlier section. The overall asset allocation treats high-quality bonds as one asset class. However, when solving for this asset class, the optimizer uses MUB’s return for the after-tax return in the taxable account, and AGG’s return for the after-tax return in the qualified accounts. The resulting decision variables tell us the dollar value of high-quality bonds that goes in each account, but the actual purchases follow the same split—MUB in the taxable, AGG in the qualified. The sum of the MUB position and AGG positions will equal the overall allocation to high-quality bonds.

16Standard market bid-ask spread costs will still apply. These are relatively low, as Betterment considers liquidity as a factor in its investment selection process. Betterment customers do not pay for trades.

17Additionally, in the interest of making interaction with the tool maximally responsive, certain computationally demanding aspects of the methodology were simplified for purposes of the tool only. This could result in a deviation from the target asset location imposed by the TCP service in an actual Betterment account.

18Another way to test performance is with a backtest on actual market data. One advantage of this approach is that it tests the strategy on what actually happened. Conversely, a forward projection allows us to test thousands of scenarios instead of one, and the future is unlikely to look like the past. Another limitation of a backtest in this context—sufficiently granular data for the entire Betterment portfolio is only available for the last 15 years. Because asset location is fundamentally a long-term strategy, we felt it was important to test it over 30 years, which was only possible with Monte Carlo. Additionally, Monte Carlo actually allows us to test tweaks to the algorithm with some confidence, whereas adjusting the algorithm based on how it would have performed in the past is effectively a type of “data snooping”.

19In simulations, as in the actual product, Betterment always uses cash flows to purchase underweight assets first, helping reduce portfolio drift and reducing the likelihood that a taxable sale will be required to rebalance. This will increase the after-tax return both for TCP, and for uncoordinated accounts baseline. However, note that when accounts are coordinated, there is a higher likelihood that taxable sales can be avoided. Because we are now managing a single portfolio, TCP will make whatever use it can of the qualified accounts to reduce overall drift, and to the extent that the overweight asset is available in a qualified account, it will sell that position. For all transactions, the algorithms have a preference for correcting allocation drift first, but will always attempt to optimize for location as part of the same transaction, if possible.

20That said, the strategy is expected to change the relative balances dramatically over the course of the period, due to unequal allocations. We expect a Roth balance in particular to eventually outpace the others, since the optimization will favor assets with the highest expected return for the TEA. This is exactly what we want to happen.

21For the uncoordinated taxable portfolio, we assume an allocation to municipal bonds (MUB) for the high-quality bonds component, but use investment grade taxable bonds (AGG) in the uncoordinated portfolio for the qualified accounts. While TCP makes use of this substitution, Betterment has offered it since 2014, and we want to isolate the additional tax alpha of TCP specifically, without conflating the benefits.

22Full liquidation of a taxable or TDA portfolio that has been growing for 30 years will realize income that is guaranteed to push the taxpayer into a higher tax bracket. We assume this does not happen, because in reality, a taxpayer in retirement will make withdrawals gradually. The strategies around timing and sequencing decumulation from multiple account types in a tax-efficient manner are out of scope for this paper.

24Since inception, Betterment has never displayed daily performance of individual ETFs in a portfolio, and has generally de-emphasized daily performance of a portfolio, since focusing on short-term volatility is not productive, and instead increases the probability of suboptimal investor behavior. Instead, the interface was designed to emphasize information that an investor can and should act on—such as an indication of whether the portfolio is on track to reaching its specified goal, and if not, what can be done to put it back on track.

### Authors

Boris Khentov, J.D., is VP of Operations and Legal Counsel at Betterment. He has a B.A. in Computer Science from Harvard University and a J.D. from Northwestern University School of Law. Prior to Betterment, Boris was a software engineer at Antenna Software, and practiced tax and capital markets law at Cleary Gottlieb Steen & Hamilton LLP. Boris helped oversee the development of Tax-Coordinated Portfolio, and wrote so very many words. No capital gains were realized in the writing of this paper.

Rukun Vaidya is a Product Manager at Betterment. He has a degree in Operations Research from Cornell University. Prior to Betterment, Ruk spent five years performing risk research at Highbridge Capital, where he focused on statistical arbitrage, quantitative macro, and quantitative commodities. As the lead Product Manager for Tax-Coordinated Portfolio, Ruk conceived its methodology and led the implementation team. Ruk had a joke in here, but it did not make it through compliance.

Lisa Huang, Ph.D., is Head of Quantitative Analysis & Research at Betterment. She holds a degree in mathematics and biochemistry from UCLA, and a Ph.D. in theoretical physics from Harvard University. Prior to Betterment, Lisa was a quantitative strategist at Goldman Sachs, leading research collaborations and building models for fixed income strategies. She is the brains behind the “Cloud Ferrari” that is Betterment’s Monte Carlo performance testing framework. Lisa is always right, especially when the problem is so hard that everyone else barely has a clue.

Berkin. A. “A Scenario Based Approach to After-Tax Asset Allocation.” 2013. Journal of Financial Planning.

Jaconetti, Colleen M., CPA, CFP®. Asset Location for Taxable Investors, 2007. https://personal.vanguard.com/pdf/s556.pdf.

Poterba, James, John Shoven, and Clemens Sialm. “Asset Location for Retirement Savers.” November 2000. http://faculty.mccombs.utexas.edu/Clemens.Sialm/PSSChap10.pdf.

Reed, Chris. “Rethinking Asset Location – Between Tax-Deferred, Tax-Exempt and Taxable Accounts.” Accessed 2015. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2317970.

Reichenstein, William, and William Meyer. “The Asset Location Decision Revisited.” 2013. Journal of Financial Planning 26 (11): 48–55.

Reichenstein, William. 2007. “Calculating After-Tax Asset Allocation is Key to Determining Risk, Returns, and Asset Location.” Journal of Financial Planning (20) 7: 44–53.

## Disclosures

The value provided by Tax-Coordinated Portfolio (TCP) will vary depending on each investor’s personal circumstances. Investors who have a short time horizon may get little to no value from the asset location strategy employed by TCP. Under certain circumstances, investors could conceivably even decrease their after-tax returns by enabling TCP.

As a general matter, asset location strategies like the one implemented by TCP distribute assets unevenly across multiple accounts based on the varying return profiles of each asset (separately considering the potential for capital appreciation and dividend yield). These figures are “expected returns”—projections for what future returns might be. While these figures are empirically derived (using the Black-Litterman method as a starting point) they are still speculative, and actual returns will differ year-to-year, often substantially.

Therefore, while the reasonable expectation may be that asset A will pay more dividends than asset B (and/or appreciate more than asset B), the opposite could happen in any given year, or across a number of years, such that asset location decisions based on differing expectations do not maximize after-tax returns. The longer the investing period, the more likely it is that the relative performance of the various assets in the portfolio will converge on what is expected. Shorter periods, however, are more likely to produce unexpected results.

Asset location seeks to place assets with higher expected returns into tax-advantaged accounts. Therefore, toward the end of the accumulation phase, an investor should expect his or her tax-advantaged accounts to have a higher balance than they otherwise would have had (and the taxable retirement account to have a lower balance than it would have had, had an asset location strategy not been deployed).

While under most circumstances, more growth in a tax-advantaged account (in exchange for less growth in a taxable account) is desirable, there are potential trade-offs which should be considered. For instance, tax-advantaged accounts incur penalties for early withdrawal, so to the extent that access to funds prior to retirement becomes necessary, liquidity may come at a higher cost for a portion of funds that might otherwise be accessible penalty-free (had that appreciation taken place in the taxable account instead).

If the tax-advantaged assets are primarily or exclusively tax-deferred, rather than tax-exempt (e.g., a traditional IRA vs. a Roth IRA), then additional considerations should be weighed. Because all distributions from a tax-deferred account are taxed at ordinary rates, including amounts that would be taxed as long-term capital gains when realized in a taxable account, shifting such appreciation could amount to a conversion of lower taxed income into higher taxed income.

Over a long enough period, the tax deferral (i.e., the ability to continually reinvest the tax savings and compound that growth, before eventually paying the tax) is expected to be valuable enough to justify such a conversion. This is especially likely when the taxpayer expects to be in a lower income tax bracket in retirement than during the accumulation phase (often, though not always the case). However, when asset location is practiced over a short period (years, not decades) and the taxpayer maintains a high income tax bracket at the time of distribution, conversion of some capital gains into ordinary income may dominate the after-tax return, thereby rendering the asset location strategy counterproductive.

As a separate matter, a higher tax-deferred balance could mean higher required minimum distributions (RMDs) in retirement, which could be an important consideration for those seeking to minimize their RMDs.

No performance estimates are based on actual client trading history, and actual Betterment clients may experience different results. Factors which will determine the actual benefit of TCP include, but are not limited to, market performance, the relative size of each account included in TCP, the equity exposure of the portfolio, the frequency and size of deposits into the various accounts, the tax rates applicable to the investor in a given tax year and in future years, and the time elapsed before liquidation of any of the accounts becomes necessary.

Nothing herein should be interpreted as tax advice, and Betterment does not represent in any manner that the tax consequences described herein will be obtained, or that any Betterment product will result in any particular tax consequence. Please consult your personal tax advisor as to whether TCP is a suitable strategy for you, given your particular circumstances. The tax consequences of asset location are complex and uncertain. You and your tax advisor are responsible for how transactions conducted in your account are reported to the IRS on your personal tax return. Betterment assumes no responsibility for the tax consequences to any client of any transaction.