The Portfolio Optimizer for Wealth Management

Tax-Sensitive Rebalancing

Summary

We rebalance less aggressively if it helps us avoid incurring capital gains. Our optimization-based approach allows the rebalancing aggressiveness to scale smoothly against tax cost, vs. a simple rules-based approach that uses hard cut-offs.

Background

A basic service financial advisors provide is monitoring a client's mix of assets, and rebalancing that mix back to the desired target allocation (proportions for different asset classes) if, over a period of time, some asset classes go up in value more than others. This prevents the portfolio from being riskier than a client's risk preferences warrant.

Rebalancing entails selling the asset classes that are "overweight" (i.e. their proportion of the total portfolio is higher than desired) and buying the underweight ones. However, in the case where selling incurs capital gains, there is a tradeoff between:

better portfolio tracking
realizing capital gains, which effectively is the opposite of tax loss harvesting

Some ways to consider the effect of taxes during a rebalance are:

Ignore it altogether: this is simplest, but can create a big tax bill by selling gains.
Use a rule such as "never sell short-term gains": this is simple, but very coarse. It can result in either too much upfront tax paid (e.g. by selling at a large long-term capital gain) or a portfolio that is not balanced enough (e.g. by not selling very small short-term gains).
Use a rule such as "never sell long-term gains of more than 15%": any such rule would necessarily be inaccurate around the cut-off point. Selling a 15.1% gain that improves tracking by a lot is better than selling a 14.9% gain that barely improves tracking, but the rule would prefer selling the 14.9% gain.
Consider the cost/benefit of tax and portfolio tracking jointly: this is difficult to build, but is the best alternative. As the taxes incurred by a portfolio rebalance increase, our system gradually rebalances less. We do it by quantifying the tradeoff of tax efficiency and portfolio tracking into some common "currency", and then finding a solution that optimizes that mix.

The purpose of our optimization algorithm is to trade off different objectives against each other. We want to track a target mix of asset classes closely (including assets held externally), but also want to minimize capital gains taxes and transaction costs. These goals can clearly conflict; by trading every day we could achieve almost perfect asset class tracking, but incur large short-term capital gains taxes and trading costs. Conversely, if we were to focus solely on taxes, our portfolio could drift quite far from its target mix. Our system is able to find the best portfolio subject to these trade-offs.

The goal of this backtest is to demonstrate how we evaluate this trade-off to create the best possible portfolio:

In the tracking only scenario, we track a target portfolio without considering the tradeoff between tax and tracking. However, we add a hard constraint that no short-term capital gains may be incurred while rebalancing. This makes this scenario more realistic against the intelligent tax-sensitive logic (next item), even though it makes the improvement look smaller.
In the tax-sensitive scenario, we again attempt to track the target portfolio, but introduce a tax minimization objective into the optimization. The tax objective has a small weight; its goal is to prevent the trades with the worst tax consequences while preserving excellent asset class tracking.

Scenario parameters

Held by us: $1,000,000 in cash, which we are free to invest.
External: although we can handle external assets in parallel, there are none here, to keep things simple.
Target allocation: in both cases, a balanced mix of 7 asset classes.
AUM fees: 25 bps annually, charged ~2 bps / month.

Results

We can see the trade-off between tracking and taxes clearly in the Asset tracking tab for each scenario. In the tax aware scenario, the tracking is somewhat worse than in the tracking only scenario, as we would expect. On the other hand, in the Net tax owed tabs, we see that the tax aware scenario ends with more taxes still owed (i.e. having paid less in taxes) than the tracking scenario. You can see the taxes for both scenarios in the same place here.

It may be counterintuitive, but owing more taxes in the tax-sensitive case is a good thing; it usually means that realized taxes were even lower. Avoiding rebalancing when it would cause large tax gains to be realized increases after-tax returns in this scenario slightly, by about 32 basis points (annualized).

There is no guarantee that tax-sensitive rebalancing will always do that. If, by tracking slightly less well, we end up overweight an asset class that happens to perform poorly, then after-tax returns would be worse. However, since that can go either way, tax sensitive rebalancing should increase after-tax returns, on average. It may help to think about it as a mild form of tax-loss harvesting.

Note: Both scenarios do their first large rebalance trades exactly one year after the backtest begins. Since stocks had risen steadily for the year and we had optionally specified 'no short-term capital gain sales', the system could not rebalance until the gains had turned long-term.

Conclusion

Our system suggests optimal orders by considering the trade-off between tax and portfolio tracking, without resorting to rules. We can do this because we have built a sophisticated optimization infrastructure (see our blog for more). This results in higher after-tax returns and saves money for clients.