April 2012

Smarter Rebalancing: Using Single Period

Optimization In a Multi-period World
Northfield Research

Dan diBartolomeo

Abstract

Modern Portfolio Theory as introduced by Markowitz (1952) frames the time dimension of investing as a single
period over which the parameters of the probability distribution of asset returns are both known with certainty and
are unchanging. We know that neither assumption is true in the real world. The second assumption has received
attention in the theoretical literature but there has been little progress in terms of practical implementations
available to financial practitioners. In this paper, we will discuss simple approximations that can improve the ability
of investment professionals to rebalance portfolios in ways that efficiently control portfolio turnover and the related
costs of rebalancing.

Introduction

Modern Portfolio Theory as introduced by Markowitz (1952) frames the time dimension of investing as a single
period over which the parameters of the probability distribution of asset returns are both known with certainty and
are unchanging. In essence, we treat the future as a single period which begins at the present but ends only at
some unknown moment in the future. This second assumption has received attention in the theoretical literature
but there has been little progress in terms of practical advances available to financial practitioners.

The underlying assumption that legitimizes the “single-period” framework for portfolio optimization is that all
frictions that impact portfolio formation and rebalancing are inconsequentially small. If the cost of rebalancing is
zero, then the single period assumption does no harm. If conditions change such that our portfolio should be
changed, we can simply declare that the current single period has ended and a new one has begun. We can then
rebalance our portfolio accordingly at no cost. By doing so, we wish away the relevance of routine trading costs,
market impact, and taxes on capital gains. While these costs may be very small for some investment assets held
by some investors (e.g. liquid futures contracts held by tax-exempt entities) the necessary conditions are not true in
most cases.

1 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

For almost all real world investors, portfolio rebalancing is not free, and often is very costly. For taxable investors
holding illiquid assets such as real estate, private equity or investment collectibles (e.g. art), the cost of transactions
often dominant risk and return considerations unless holding periods exceed multiple decades.

Almost all widely available portfolio optimization systems are based on mean-variance optimization methods which
maximize the expectation of the risk adjusted return on a portfolio over a single period, as described in Markowitz
and Levy (1979).

Multi-Period Optimization and Approximations Thereof

Much research has been done on the formulation of full multi-period optimization. Mossin (1968) suggested an
explicit multi-period formulation for portfolio optimization. A paper by Cargill and Meyer (1987) focused on the risk
side of the multi-period problem. This was followed by Merton (1990) who introduced a continuous time analog to
mean-variance optimization, and by Pliska (1997) who provided a discrete time analog to the single period approach.

Other papers such as Li and Ng (2000) provide a framework for multi-period mean variance optimization (MVO)
using dynamic stochastic programming. Such stochastic programming methods investigate a range of potential
paths of future events and outcomes, and select the set of asset weights that best fulfills a stated objective and set
of constraints. Such techniques are often used for asset allocation for complex taxable investors such as insurance
companies and high net worth individuals. The computational intensity of the method often limits the number of
assets that can be handled. Even with today’s improvements in computational efficiency, the methodology is still
only viable for portfolios with a small number of assets. Using a set of simplifying assumptions, Sneddon (2005)
provides a closed form solution to the multi-period optimization including optimal turnover that could be applied to
problems with large numbers of assets.

Explicit multi-period optimization and related processes have had very limited adoption among practitioners because
of these constraints. The most significant impediment is parameter estimation error. Investment practitioners
often have great difficulty accurately estimating the future distribution and correlations among the assets in their
universe of choices. To fully exploit the benefits of multi-period optimization we must be able to not only forecast
these distributions, but also forecast how these future distributions will change from one future moment of time to
the next and so on. Obviously, being able to accurately forecast such changes appears to be beyond the predictive
power of real world investors.

Another line of research in this area focuses on the idea of creating rules that inform investors when it is really
necessary to rebalance their portfolio weights. In the absence of statistically significant and economically material
advantage such rules simply tell the investor to do nothing, hence avoiding rebalancing costs altogether. Early
research in this area includes Rubenstein (1991) who examines the efficiency of continuous rebalancing and
proposes a rule for avoiding unnecessary turnover. Kroner and Sultan (1993) propose a “hurdle” rule for rebalancing
currency hedges when return distributions are time varying, while Engle, Mezrich and Yu (1998) propose a similar
hurdle on alpha improvement as the trigger for rebalancing actively managed asset allocations.

2 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

Bey, Burgess, Cook (1990) use bootstrap resampling to identify “indifference” regions along the efficient frontier.
In Gold (1995), a similar technique is to define indifference to rebalancing portfolios for illiquid asset classes such as
real estate. Michaud (1998) uses a parametric form of resampling to measure the confidence interval on portfolio
return and risk to form a “when to trade rule.” The rule was elaborated upon in Michaud and Michaud (2002) and
patented.

Markowitz and Van Dijk (2003) propose a rebalancing rule (“MvD”) based on game theory to approximate multi-
period optimization, but argue it is mathematically intractable (at least in closed form) for large problems. In
Kritzman, Mygren and Paige (2007) the authors are able to test the efficiency of MvD against full dynamic
programming for cases up to a maximum of five assets, as dynamic programming becomes computationally too
burdensome for larger numbers of assets. The authors are able to extend the MvD method to one hundred assets.

Practitioner adoption of some form of “all or nothing” rebalancing rule is fairly common, but at least two negative
arguments have come to light. The first is that when active managers are “inactive” because the potential benefits
of rebalancing are too small, this lack of trading is perceived by clients as the manager being neglectful rather than
as an analytically-driven decision to reduce trading costs. Presumably this objection could be overcome by
appropriate communication between the asset manager and their investors.

The second argument is that after a period of inactivity, the eventual rebalancing concentrates the required trading
into a particular moment in time. For large investors the market impact arising from doing trades that are a larger
fraction of available trading volume per unit time will create higher transaction costs than if the trading had been
done gradually between the previous portfolio rebalancing and the current one.

For portfolios that are composed of assets with homogeneous transaction costs it is common to simply place
heuristic limits on the amount of turnover allowable in a given rebalancing. Grinold and Stuckelman (1993) consider
a value added/turnover efficient frontier. They derive that under certain common assumptions, value added
(improvement in utility) is approximately a square root function of turnover (“half the turnover gets you three
quarters of the available improvement”). As such, investors can optimize their portfolios without consideration of
transaction costs, and then simply choose the intermediate point (e.g. the amount of turnover) between the initial
portfolio and the optimal portfolio that gives the most favorable balance of utility improvement relative to the
expense of trading costs.

A Single Period Model with Trading Costs

A typical way of accounting for transaction costs in an optimization is to deduct these costs from the mean-variance
objective function, amortizing the expected costs over the expected holding period of the investment. Put
differently, the expected holding period is just 100% / M%.

3 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

U1 = α – λσ2 – (Σj = 1 to N(abs(Wij – Wfj) * Kj ) * M

Where
N = the number of assets in the portfolio
Wij = the initial weight of asset J
Wfj = the optimal weight of asset J
Kj = the unit cost of trading asset J
M = the expected % one way turnover per annum
α = the expected annual % return on the portfolio
σ = % expected volatility of annual portfolio returns
λ = investor risk aversion scalar

One obvious improvement to the simple linear amortization of costs is to realize that expending transaction costs
today reduces the value of the portfolio on which returns can be earned. For portfolios with small costs and low
returns, the linear approximation is sufficient, but if costs or returns are large we need to consider the effects of
return compounding. For example, assume a trade with 20% trading costs and an expected holding period of one
year that provides a return improvement of 20% per year. Despite the apparent equivalence of costs and benefits,
if we give up 20% of our portfolio value to costs now we would end the year with only 96% of our initial wealth.
The solution is to adjust the amortization rate to reflect the opportunity costs of the lost returns.

Mα = M * (1 + α/100)
It should be noted that multiple distinct amortization rates may be required in dealing with taxable portfolios at the
individual tax lot level. For example, we might have a portfolio that normally has 20% per annum turnover.
However, within the portfolio there is a single security that has a negative expected return, but a substantial
embedded short-term capital gain. We would like to sell the security because of the negative return expectation,
but we would also not like to not sell the security because of the substantial tax consequences. If the point in time
at which our short term gain would revert to a long term capital gain that is less taxed is near, we would hold. If
the point of conversion to long term capital gain status is far away, we would sell. To optimize this portfolio
correctly we must use our regular amortization rate for the long term portion of cost of the capital gain tax, while
the incremental portion of the tax cost of selling a short term capital gain should be amortized over the time interval
between today and the conversion to long term status.

The Probability of Realization

Imagine we have a portfolio, P1 with return α (net of fees and expenses) and standard deviation σ. Our usual utility
function would say:

4 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

U1 = α – λσ2

Now let’s imagine that another portfolio, P2 has a higher utility, because either the return is higher or the standard
deviation is lower. This portfolio has completely different positions than the initial portfolio. Let’s assume that this
portfolio has a higher return by positive increment ∆, so:

U2 = (α+∆) – λσ2
Since U2 is greater than U1, we should be willing to pay some transaction costs to switch from P1 to P2. Now let’s
consider a different way to improve our returns. We go back to the manager of Portfolio 1 and ask them to reduce
their fees by ∆, so now our revised utility on P1 is U1L for “lowered fees.” Notice that U1L and U2 are equal. So if
we invest our money in either P2 or P1L (after lowering the fees), the expected value of wealth at the end of time is
the same. This suggests that we should be willing to pay the manager an upfront fee to lower ongoing
management costs equal to the trading costs we would be willing to pay to switch from the initial portfolio P1 to
P2. As long as conditions never change, this is valid.

However, since P2 and P1L have different securities, the performance will be different from day to day and month to
month, even if the long term average return and volatility are identical. So over any finite time horizon, we cannot
be sure which of the portfolios will perform better. On the other hand, P1L will always perform better than P1 over all
time horizons, because it is just the same portfolio with lower fees. For P1L the probability of outperforming P1 is
always one. P2 is guaranteed to be better than P1 in the long run if conditions don’t change, but the probability that
P2 will actually outperform P1 over any finite horizon is between one half and one. We call this probability value the
“probability of realization” of the utility increase.

Amortizing transaction costs over a single period is equivalent to assuming that the probability of realization is
always one. Let’s assume a single period optimization for a strategy with expected turnover of M% per annum. In
our multi-period world, we want to amortize by M divided by a scalar that reflects the probability that the revised
portfolio will actually realize a better risk adjusted return over the finite holding period. That is:

M* = Mα / K
Where
M* is the adjusted amortization rate
K is a chosen scalar in the range [Z:1] where Z is the probability of realization

To calculate Z we can use the tracking error (expected volatility of the return differences) between any two
portfolios as a standard error on the expected differences in utility. We obtain the tracking error value from
whatever model we are using to estimate σ in our utility function. We can then calculate a T-statistic on the

5 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

expected difference in utility, calculating Z under a one-tailed test and our choice of probability distribution. In our
example above, the value of Z is one for the incremental utility of P1L relative to P1. This is because the tracking
error between P1 and P1L is zero, as they are the same portfolio with different fee structures. Highly risk-averse
investors will chose values for K close to Z (the expected benefit in utility over a finite horizon is uncertain while the
costs of trading are certain so we want to reduce turnover). More risk-tolerant investors will chose K close to one.

The Next Rebalancing

The process above to amortize rebalancing costs reflects only the trading costs that are going to be experienced as
a result of the current portfolio revisions. Another way to get closer to multi-period optimization is to give some
forethought to the expectation of ongoing costs that are likely to be experienced in future rebalancing events. In
high turnover strategies, the changes in the asset weights will be dominated by the volatility of the asset values
and the lack of correlation across the asset returns. The expected amount of trading required at the next
rebalancing will be greater for asset universes with higher variety (i.e. cross-sectional dispersion of returns). For
low turnover strategies, the differences in asset weights at the next rebalancing will be dominated by differences in
the mean returns achieved by the assets. The larger the difference in expected returns across the asset universe,
the larger the required turnover is likely to be at the next rebalancing. If some assets have materially higher trading
costs than others, we can decrement the expected returns of the individual assets to reflect the expectation of
higher ongoing trading costs that will be experienced in future rebalancing events.

Many portfolio strategies also have implicit beliefs about auto-correlation in asset returns. For example in a stock
portfolio with a momentum strategy, we are implicitly expecting positive serial correlation in returns. Stocks that
realize a high return in one period will have an increased expected return for the next period. In that an above
average return relative to other assets implies an increase in portfolio weight, the portfolio is already positioned to
reflect the higher expected return for the next period that would justify an increase in the asset’s weight in the new
optimal portfolio. This congruence reduces turnover.

Conversely, a value strategy implicitly expects negative serial correlation in returns. Stocks that realize negative
returns in one period will have a higher expected return for the next period. The negative return will reduce the
weight in these assets, while the higher expected return is apt to call for a higher optimal weight for the next
period. This discrepancy of direction implies greater portfolio turnover and costs over future rebalancing cycles. To
the extent that a portfolio strategy leads to a larger or smaller expectation of turnover we may choose to change
the magnitude of the suggested decrement to expected returns.

Conclusion

While full multi-period optimization seems obviously preferable to the single period representation there are many
impediments which prevent practical implementation. “When to rebalance” rules are often helpful in reducing the

6 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

economic cost of portfolio turnover, but may be detrimental to large investors where the market impact of trading
is a material issue.

The MvD solution has been demonstrated to be useful for small cases, but implementation of the technique for
portfolios of large numbers of assets positions or large numbers of active positions (i.e. a small portfolio measured
against a broad benchmark) is problematic.

We suggest that much of the potential benefit of multi-period optimization can be obtained by incorporating trading
costs of the current rebalancing into the optimization process in a thoughtful way. Additional benefits can be
obtained by considering the long-run rebalancing costs of different investment assets, and adjusting return
expectation to be net of these long-term costs.

References

Bey, Roger, Richard Burgess and Peyton Cook. “Measurement of Estimation Risk in Markowitz Portfolios,”
University of Tulsa Working Paper, 1990.
Cargill, Thomas F. and Robert A. Meyer. "Multiperiod Portfolio Optimization And The Value Of Risk Information,"
Advances in Financial Planning and Forecasting, 1987, v2(1), 245-268.
Engle, Robert, Joseph Mezrich and Linfeng You. “Optimal Asset Allocation,” Smith-Barney Market Commentary,
January 28, 1998.
Gold, Richard B. “Why the Efficient Frontier for Real Estate is Fuzzy,” Journal of Real Estate Portfolio
Management, 1995, v1(1), 59-66.
Grinold, Richard C. and Mark Stuckelman. "The Value-Added/Turnover Frontier," Journal of Portfolio Management,
1993, v19(4), 8-17.
Kritzman, Mark, Simon Mygren and Sebastien Page. “Optimal Rebalancing: A Scalable Solution,” Revere Street
Working Papers, 2007.
Kroner, Kenneth F. and Jahangir Sultan. "Time-Varying Distributions And Dynamic Hedging With Foreign Currency
Futures," Journal of Financial and Quantitative Analysis, 1993, v28(4), 535-551.
Levy, H. and H. M. Markowitz. "Approximating Expected Utility By A Function Of Mean And Variance," American
Economic Review, 1979, v69(3), 308-317.
Li, Duan and Wan-Lung Ng. "Optimal Dynamic Portfolio Selection: Multiperiod Mean-Variance Formulation,"
Mathematical Finance, 2000, v10(3,Jul), 387-406.
Markowitz, Harry. "Portfolio Selection," Journal of Finance, 1952, v7(1), 77-91.
Merton, Robert, Continuous-Time Finance, Oxford, Blackwell, 1990

7 www.northinfo.com
Smarter Rebalancing: Using Single Period Optimization In a Multi-period World

Markowitz, Harry M and Erik L. van Dijk. “Single-Period Mean–Variance Analysis in a Changing World,” Financial
Analysts Journal, Vol. 59 (March/April 2003.), No. 2, pp. 30-44
Michaud, Richard, Efficient Asset Management, Boston, Harvard Business School Press, 1998
Michaud, Robert and Richard Michaud. “Resampled Portfolio Rebalancing and Monitoring,” New Frontier Advisors
Research Release, 2002.
Mossin, Jan. "Optimal Multiperiod Portfolio Policies," Journal of Business, 1968, v41(2), 215-229.
Pliska, Stanley, Introduction to Mathematical Finance: Discrete Time Models, Oxford, Blackwell, 1997
Rubinstein, Mark. "Continuously Rebalanced Investment Strategies," Journal of Portfolio Management, 1991,
v18(1), 78-81.
Sneddon, Leigh. “The Dynamics of Active Portfolios,” Proceedings of the Northfield Research Conference 2005,
http://www.northinfo.com/documents/180.pdf

8 www.northinfo.com