The subject of rebalancing investment portfolios has generated much discussion as well as confusion among advisors and researchers. A recent example is the October 2020 Advisor Perspectives article by Michael Edesess, which has drawn more than 100 comments on APViewpoint. There are upsides and downsides associated with various rebalancing strategies (including the strategy of not rebalancing), and it’s important to look at the full picture. For this analysis, I’ll mainly focus on rebalancing versus buy-and-hold (not rebalancing) with a focus on retirement withdrawals. I’ll utilize Monte Carlo simulations and provide measures of benefits and risks.


My interest in the subject of rebalancing stemmed from my earlier analysis of asset allocation where I made the case (in this article) that modest differences in stock/bond mixes were not an important determinant of retirement results – such allocation differences tended to get swamped by variability of year-by-year performance of the underlying asset classes, principally stocks. There is considerable overlap between rebalancing and asset allocation, so I wondered if those arguing for and against rebalancing were debating something that does not matter much.


For this analysis I’ll use an example of a 35-year retirement that starts with $1 million of savings. I’ll assume variable retirement withdrawals based on RMD factors, and a target 60/40 stock/bond allocation.

I will use a Monte Carlo generation based on historical stock and bond returns, but with a downward adjustment to reflect my personal view about future returns. More specifically, I’ll use bootstrapped annual real returns (after inflation) by randomly selecting 20-year blocks from Ibbotson’s 1926 – 2019 data and subtract 4.85% from each of the stock returns generated and 1.95% from bond returns. The result is that arithmetic real returns will average 0% for bonds, reflecting current yields, and 4% for stocks, assuming a lower-than-historical equity premium. The purpose of using bootstrapped historical returns is to capture any correlation between stock and bond returns as well as any year-by-year correlation.

For each analysis, I generate 5,000 Monte Carlo simulations, but I caution that the end results are limited by the scarcity of historical data. Although there are 94 individual years of data between 1926 and 2019, that is less than three independent 35-year retirement periods. This points to the more general limitations in the use of historical data – there’s both the statistical significance of the data, and the question of whether historical financial relationships will persist into the future.

But I do the best I can with what I have.