Rebalancing Can Be Hazardous to Your Portfolio

An excerpt from the new book Yes, You Can Supercharge Your Portfolio - reprinted with permission of authors Ben Stein & Philip DeMuth and publisher:

• • •

Yes, You Can Supercharge Your Portfolio

Portfolio Rebalancing

The financial services industry loves to tout portfolio rebalancing as a value-adding strategy. The idea is that every year we should sell a little of that 12 months’ winners and use them to buy the current losers so that we bring our portfolios back into alignment with their original specs. Thus, if we had a portfolio of 60 percent stocks and 40 percent bonds initially, and at the end of the year find ourselves with 62 percent stocks and 38 percent bonds, we’d sell the extra 2 percent of from the stock side and add it to the bond side. Although it can incur tax and transaction costs, rebalancing is promoted on the idea that it gets us into a righteous “sell high, buy low” discipline.

Most of the studies we’ve read on the topic are either theory-based or derived from simple 60/40 stock/bond portfolios drawn from the historical record. This greatly oversimplifies an important question. We looked at 10,000 Monte Carlo simulations involving complex, seven-asset-class portfolios. We experimented with several calendar-based strategies, rebalancing the portfolios monthly, annually, or never. We also looked at tolerance-based rebalancing approaches: rebalancing when the portfolio wandered 10, 15, and 20 percent from its original allocations. (Your authors are grateful to Chartered Financial Analyst Bill Swerbenski for modifying his Portfolio Survival Simulator to enable us to make these calculations.) The results of these different approaches are all shown in the graph below.

Each triangle in the graph represents a specific rebalancing strategy and its effects on the 10,000 portfolios’ returns and risks. Note especially the “line of best fit” that indicates the average trade-off between risk and return of all these approaches.

The first thing that stands out is that the more frequently we rebalance, the worse our returns. The next obvious point is that, as is commonly observed, rebalancing cuts our risks. What we’d hasten to add, however, is that it doesn’t cut our risks efficiently. Specifically, the most frequently recommended ritual—annual rebalancing—puts our portfolio’s returns-to-risk profile below the efficient frontier.

Tolerance-based approaches, on the other hand, seem to have more merit: Those portfolios that were allowed some leeway to roam before being rebalanced had better risk-adjusted returns (above the line) than those adjusted according to the calendar (below the line).

This shouldn’t be surprising. When we go through this procedure frequently, it just shuffles short-term market noise into our portfolios. But rebalancing only when market action has had a chance to push our portfolios significantly out of alignment allows us to take advantage of market momentum misvaluation effects that aren’t otherwise captured by asset-allocation strategies.

When the obvious costs of rebalancing are pointed out, advocates quickly shift their ground to argue that it’s really being done to control risk. This may be true, if the only risk we’re talking about is standard deviation. But what about the risk of running out of money? The figure below shows how, starting with $1,000 invested across seven asset classes, the worst 5th percentile portfolio out of 10,000 grew after 25 years in the markets when following each of these rebalancing strategies.

Even in this bad-case scenario, overactive realignment costs us money if we’re long-term investors. In sum, one way we can supercharge our long-term returns is by not rebalancing our holdings too often.

Comments

[richjoy]

THANK YOU...I always suspected the 'annual rebalance' mantra was a drumbeat to lemming-like results; furthermore, I try to make it a habit to be suspect when faced with one-size-fits-all rules. This Monte Carlo study is convincing enough for me.

[Phil DeMuth]

RIchjoy: By the way, there is an recent article in the Journal of Financial Planning that makes the same point.

www.fpanet.org/journal...

[DavidJackson]

Phil, this is fascinating -- thank you. My question is this:

If we assume that
(1) asset classes trend together in the long term, and
(2) in the short term they deviate from their long term trajectories,

then rebalancing should increase the return of your portfolio, not just reduce the risk/volatility.

This is the argument I outlined here:
seekingalpha.com/artic...

My question is: why do you think you are not seeing that in the monte carlo studies for less-frequent rebalancing strategies?

[Phil DeMuth]

Hello David --

Thanks for your comment.

I think the reason that rebalancing is not increasing returns here -- even long term -- is that some of the asset classes were fixed income, which are inherently lower-return than equities, so that even as the equity asset classes forged ahead, their wings were clipped by being periodically re-weighted down with bonds. It would be interesting to follow up and look at rebalancing with an all-equity portfolio and see whether long-term rebalancing improves returns while cutting risks.

[DavidJackson]

Phil, that makes total sense.

So the implication is that a long term portfolio shouldn't include any bonds; and if you rebalance between asset classes that do trend up in the long term, that should raise returns?

David

[Phil DeMuth]

Hi David,

The purpose of bonds in a portfolio is either for income or to control volatility, so an investor should have whatever allocation to bonds is appropriate to his or her goals. My guess is that rebalancing just among equity classes would slightly lower risks and possibly raise returns as well but I would rather see some numbers put to this before assuming that this would be so. Rebalancing is basically a grudge match between two effects: momentum and valuation. In theory, rebalancing too often cuts off momentum at the knees, but rebalancing too infrequently fails to take advantage of misvaluations. Or something like that.

[snowmman]

"We looked at 10,000 Monte Carlo simulations involving complex, seven-asset-class portfolios."

well, isn't that the problem? Assuming you did normal distribution
with correlations of 0 between asset classes? So you're modelling something that doesn't match asset class behavior historically?

Why didn't you use an analytic or historical model?

see
www.fpanet.org/journal...
for detailed critique of monte carlo simulations.

I haven't given it much thought, but assuming returns are
symmetrically distributed, normal, around a mean, with
no correlation, then yeah, it doesn't make sense to do anything,
because just waiting will get you your normal distribution
for each asset. But that's not what our asset classes do.

Isn't this just a bad experiment?

[Phil DeMuth]

Hi Snowman --

Thanks for your comment. Fear not: the historical correlations among the seven asset classes were built into the experiment. The classes were : Large Cap US Stocks, Microcap US Stocks, MSCI EAFE Stocks, Emerging Market Stocks, REITs, Treasury Bills, Long Bonds. The point of Monte Carlo simulation is to try to look at a broader range of outcomes than just the one entombed in the historical record -- exactly as the article you cited recommends.

[snowmman]

Thanks for the replay Phil.
I suppose it's a little muddy since you're mixing historic info (correlations) with non-historic (monte-carlo).

In any case, here's an odd thought.

Assume I have two portfolios. One I decided on an
asset allocation 5 years ago and let it drift without
rebalancing till today. It has some defacto allocation now.

Another portfolio I want to start and allocate today.

Assume that the first portfolio is an ideal state in terms of
risk/return (it's drift has led to more risk, but more return,
but I want that).

And then I want the second portfolio to have the same
predicted risk/return. So I allocate exactly like the state
of the first portfolio.

Take that to the limit, and it says that the optimal risk/return
decision, is made by using a portfolio decision that was
made in the past, and allowed to drift to "some state" today.

That's basically saying the optimal risk/return decision is
made with historic data.

Right?

Isn't this just portfolio optimization using historic data
(and the appropriate software)

It basically is saying you can drift to a different risk/return
point? If so, shouldn't we just design the portfolio with
the higher predicted risk in the first place?

Or is the short term historic data more important. We can
still design using short term data, though.

see what I'm thinking?

-s

[VennData]

By letting your winners ride, you're increasing the riskiness of the portfolio, so of course if you don't rebalance your returns go up on average. The problem is when you tend of the years to go from a 60/40 allocation to an 80/20, your risk is greatly increased, hence the increased returns.

[JOINVESTOR]

I sort of agree. What about people in their 60's and 70's who cannot afford to wait out a market decline and are taking income from their accounts?

[Phil DeMuth]

Snowman: I wouldn't call the data "muddy" -- the correlations, means and standard deviations were all derived from historical data. Since we don't have 250,000 years of historical data, we use Monte Carlo to fill in the gaps as best we can.
The point is not that we could just choose a higher risk/return profile initially -- it's that rebalancing at a 20% tolerance produces a more efficient risk/return trade off than rebalancing frequently or rebalancing not at all. We are looking at the rebalancing effect here separate from the % equities effect. The returns and risks are determined by the % equities --but the 20% tolerance rebalancing strategy seems to tweak the risk-adjusted returns for a small free lunch (above the line of best fit in the first figure).

VennData: I think the above addresses your point as well.

Joinvestor: The biggest risk is really to new retirees. They need enough in the way of fixed-income investments so that they don't have to sell stocks when they are down just to raise money for living expenses, so their stocks have time to recover. In our book on retirement (Yes, You Can Still Retire Comfortably), we recommended that new retirees rebalance their portfolios annually for the first five years for this very reason.
Against this, the second figure in the article shows that even in very bad cases (the fifth percentile) infrequent rebalancers had more money 25 years later. But (and this is the critical fact) they were not making withdrawals along the way. To the extent that lowering risk supersedes all other goals, more frequent rebalancing is useful. So would more studies be along these lines.

[Roger Beep]

Phil;

Thank you for your posts.

A parallel question for you; as a retiree if one does not currently need the dividends generted by a porfolio as income is it
(1)better to have the dividends reinvested in the company of orgin or
to (2)wait and invest in low performing stocks within the porfolio or
(3)buy new stock holdings to round out a the portfolio or (4) invest them in the bond portion of one's porfolio?

Ken

[Roger Beep]

Phil;

Thanks for your posts.

A parallel question, as a retiree if one does not currently need the dividend income generted by a porfolio is it
(1)better to have the dividends reinvested in the company of orgin or
to (2)wait and invest in low performing stocks within the porfolio (akin to rebalancing) or
(3)buy new stock holdings rounding out the portfolio?

[snowmman]

If there's a free lunch, I think this author's data on opportunistic rebalancing is more plausible (taking advantage of volatility)

https://fpanet.org/journal/art...

and any argument against rebalancing because of costs in a taxable account also makes a lot of sense

www.fpanet.org/journal...


Also, it's an interesting question whether the state of a portfolio with a "let it ride strategy" would outperform one that used the same historic behaviors, but tuned with portfolio optimization software for some new risk/return profile. If it did, it kind of says the portfolio optimization software is broken.

[Phil DeMuth]

Hi Roger/Ken --

I don't think your question can be answered apart from a consideration of the specific portfolio and stocks in question Even if I had the portfolio in front of me, I suspect the answer would be above my pay grade. Sorry!

[JOINVESTOR]

Another thought? Pull out a Callan Chart and notice no one market segment dominates the marketplace for more than two consecutive years. What happens when you don't take some profits off the table. I would agree rebalancing is overblown as insurance companies and big brokerage firms moved towards a risk managed approach to protect themselves from lawsuits in declining markets. It's rare an investor wins a settlement for underperforming the market with positive real rates of return. America as a whole lives in a world were the reward for risk no longer justifies the adverse impact of failure.

[Smart ETF]

You are so right – when using an academic model from 1960. However, we are not in Kansas anymore. You fool yourself by believing in an out-dated model known as Modern Portfolio Theory (MPT) and by using Mean Variance Optimization (MVO) to provide your asset allocation and rebalance solution; circa 1952 & 1959.

For starts, show me one security or portfolio that has a normal distribution; yet this is what you are using in your asset allocation model; therefore your risk measurement is fatally wrong from the start.

Next, you consistently refer to the 60/40mix; I’ll bet you can’t tell me where the 60/40 mix originates! I’ll bet none of your MVO brethren can tell me where the 60/40 mix comes from……just what I thought. It came from 10 years of historical data during the 1950’s; the time period known as the Nifty Fifties (the biggest Bull market prior to the recent decade). Try running your 60/40 mixes for the 20 years after the 50’s (60’s and 70’s) and tell me if any of your clients survived. The answer is no because no combination of stocks or bonds made positive returns for that 20 year period. MVO used past performance to predict future returns, and lost.

In your fantasy world of long-term ‘mean-variance’ you are correct to assume rebalancing is a waste of transaction costs and most likely inefficient. But what happens when you add 3 months of new data to your model; a model that is built on decades of data (let’s guess 40 years))? Absolutely nothing! The 3 months of new data gets averaged out over the 40 years so it has no effect on the portfolio. In other words, your mean-variance model ignores the last three months of this market decline because it really doesn’t matter in your world (but it does in mine!).

Next, what good is Monte-Carlo modeling on a mean variance model? It provides little if any assistance because the recommendation always tracks to a mean. Now try it using a M-C simulation model on a dynamic asset allocation solution, like Mandelbrot’s Extreme Value Theory, and you will get vastly superior outcomes. BTW, you will want to run 100,000 simulations once you upgrade to a more sophisticated model.

Lastly, you refer to a seven asset class portfolio as complex; a seven class portfolio is ANYTHING but complex; maybe for my Blackberry, but not any meaningful asset allocation solution.

I have patently read your articles but it is time you move on in your learning curve. I recommend you read Benoit Mandelbrot’s ‘The (Mis) Behavior of Markets. Welcome to the brave new world of portfolio optimization where rebalancing does matter!

[Phil DeMuth]

Smart ETF: The excerpt says nothing about 60/40 portfolios except that rebalancing studies based on simple 60/40 stock/bond portfolios are inadequate. It makes no mention whatever of Mean-Variance-Optimiza... I have no idea what you are referencing here.

You asked me to run a 60/40 stock/bond mix after the 1950s so I obliged: a portfolio of 60% S&P 500/40% Treasury Bonds had an average annualized return of 7.1% for 1960-1980. This hardly jives with your statement that "no combination of stocks or bonds made positive returns for that 20 year period." It made 321%.

You reference Mandelbrot, but the very book you cite urges portfolio managers to build portfolios using Monte Carlo simulations ("The (Mis)Behavior of Markets" p. 267). Mandelbrot says of the people trying to apply his methods in finance, "...in truth, our knowledge is still so limited that no one has yet to report great success." If you are having great success, why not publish and collect the Nobel Prize that is your due?

I like your point about the number of asset classes. I used seven because it was more than 2 (stocks/bonds) I had often seen in rebalancing discussions. I don't know how many asset classes there really are: my guess is that a factor analysis would reveal not many more than 7, and very possibly fewer (Fama/French find 3 factors account for almost all the variance, for example). Asset classes are determined by behavior, not by nomenclature, as I am sure you would agree.

[Roger Beep]

Hi Phil;

Regarding investing dividends in the context of rebalancing and borrowing form a previous post on "opportunistic rebalancing" it seems to me that QPP might be used to identify individual holdings on a forward looking basis that may be down in price in the future.

A possible strategy for identifying potential holding candidates: Each share's closing price could be entered as the "Current Portfolio Value" on page 1 of QPP and a future price forecast using the 5th percentile might be stipulated. For example, a forecasted price of 5% below the current closing price at the 5th percentile might flag a holding for possible future additional purchase or initial acquisition. One is assuming a previously well thought out portfolio using QPP.

The problem of the Opportunistic Rebalancing software for the average investor is that it costs $10,000. Unaforadable.

[Phil DeMuth]

Hi Roger --

You might want to put your question to Geoff Considine, who is the resident expert on QPP.

[Smart ETF]

Thank you for your response. I meant to ask you to run an MVO model during the 60’s and 70’s, not since the 50’s. Running MVO since the 50’s gives you the positive returns of the 80’s & 90’s.

The point I was stressing on Monte-Carlo was its ineffectiveness in combination with MVO models. I am a big believer in Monte-Carlo in Extreme Value Theory models; thus recommending 100,000 simulations over 10,000.

You are correct there is a big difference between Fama’s 3 & 4 factor models. My main point is there are many asset allocation solutions, such as MPT (MVO, CAPM, I-CAPM, C-CAPM), Arbitrage Pricing Theory (MacroAPT, Multiple APT, Black-Litterman), and Extreme Value Theory (Dynamic Portfolio Optimization), and the type solution determines the effectiveness of rebalancing. MPT models are completely static, APT models are semi-static (tilting MVO models to favor market conditions), whereas EVT models are completely dynamic (letting market conditions predict results). I agree that rebalancing an MVO model is wrong (but I argue MVO is wrong; even its founders are admitting its flawed). I would say that rebalancing an APT model is a good thing and rebalancing an EVT model is a mandatory event.

The second point I’m stressing is that we need to scrub all discussion on MVO models and cease the perpetuation of this myth. Sharpe & Mandelbrot state that normal distributions miscalculate risk, whereas, stable distributions allow for fat-tails that scale. Sharpe quotes CAPM is ready for a make-over; that MVO assumes all investors have the same beliefs about the market and the relationship among different assets. Mandelbrot pretty much trashes all of it. BTW, Mandelbrot was Fama’s PhD advisor!

All I ask is that we up our discussions to new methodologies and quit making headlines that represent all asset allocation models.

[GeoffConsidine]

There are many ways to solve the dilemme of rebalancing and I wrote about one last year:

seekingalpha.com/artic...

IMHO, it is far more productive to think about rebalancing only if your portfolio's risk/return characteristics meaningfully shift from what you have designed in--not just because one asset class goes beyond is 'policy' allocation. It is also going to be far more cost effective to do this as you rebalance less often.

To Smart ETF: It sounds like you have not read the responses to these same issues that you raised with Phil's last article. You seem convinced that this is historical MVO (its not). Further, Phil noted that Mandelbrot himself has said that his tools / models are not appropriate for application in actual asset allocation--they are descriptive rather than prognostic. Fama, who was Mandelbrot's grad student as you note, is on record making this point too.

[MarketRiders]

Phil -- all of this sounds interesting from an academic standpoint. But frankly when I read it, my eyes glaze over a little as I do when I read other studies like these. Please don't take offense -- here's why. Most investors add and subtract capital from their investment accounts during the year. This presents a very different context for the rebalancing argument. A fixed portfolio that has no new invested capital or withdrawals is different than a dynamic one where the investor is adding and subtracting capital.

Let's assume a dynamic situation where someone is adding or subtracting capital each year. This investor can choose to do so in a way that rebalances back to the target allocations -- or not.

So If I told you I was adding 5% each year to my portfolio for 20 years, would you recommend that I rebalance when adding or subtracting funds, or not? I think this is a more relevant question.

[Phil DeMuth]

Hi Micht --

Yes, the simplest approach would be to rebalance back to specs whenever you add or subtract cash.

[lazytrader]

rebalance is important, but to what specs? I didn't find the simple 60/40 mix etc. to be sufficiently risk mitigating and have developed my own balanced portfolio method which doesn't only rebalances, but also evaluates reallocation and holding elimination. I'm publishing my ongoing experimental results here: notiming.com , you're welcome to visit and provide me your feedback.

[Augustus]

lazytrader, that is a very interesting concept as I note that it incorporates the short side of the market as well. Nothing showing of backtest results or evan a slight description of the methodology. Certainly a successful few weeks shown. Hope it continues for you.

[lazytrader]

Hi Augustus,
There is a high level description of the methodology, essentially based on the Moderm Portfolio Theory along with an asset allocation optimization based on meeting portfolio targets, as listed in NoTiming.com/portfolio... and optimizing for best Sharpe Ratio. The methodology is using historical correlations adjusted with time-based trending for non-uniform weighting of market risk.
The triggers to re-balance/re-optimize are still experimental.

[Vig Oren]

Hi Phil, a bit late to the party. Glancing at the two studies from the JFP mentioned above, I see these results. Are there any other important issues involved?

Questions:

1) Why multi year frequency was not checked? Jim Otar, and Bill Bernstein are suggesting such strategies.

2) While rebalancing in taxable accounts are not the tested 35% and 15% taxrates sort of naiveté? See evolution of past taxrates since the 1960s. Would not such rates dictate the results?


Studies:

Rebalancing for Tax-Deferred Accounts: Just Do It—Don't Worry How
APR 2006

by Mark W. Riepe, CFA, and Bill Swerbenski, CFA


Two top results out of 7:

1. There's no free lunch. In all five models, the 21 techniques lined up close to linearly when plotted in risk-return space. At least with this set of techniques and the assumptions that form the basis for the study, no one technique resulted in a superior trade-off between risk and return.

2. Absolute differences between strategies are small. The values for the axes in the five figures were chosen to spread out letter codes as much as possible so that they could be visually distinguished. Inspection of the magnitude of the absolute differences in average return and risk reveals that they are small. Our conclusion is that the decision to rebalance is far more important than the decision of how exactly to do it.


And

Rebalancing for Taxable Accounts

APR 2007

Tips When Rebalancing in a Taxable Account:

1. Exert more care when rebalancing in taxable accounts.
2. Avoid generating rebalancing trades by directing new money into underweighted asset classes.
3. When sensible, execute trades to generate tax losses that can then be used to offset any capital gains generated by
rebalancing trades.
4. Be patient and wait until eligible for long-term capital gains treatment.
5. If taxable and tax-deferred accounts are both allocated toward the same goal, have the tax-deferred account bear as much of the
rebalancing load as possible.

Vig Oren

[Mark L]

the only purpose of asset allocation for the long term - 25 yrs would qualify as in the article - is to control risk so that one doesn't get scared out due to a huge drop. If one doesn't rebalance to control risk, then it seems it would make much more sense to simply put 100% of one's portfolio in the asset class with the highest historical return and don't look at the statement for 25 years.

The article seems to end up making 2 points.

a) that tolerance based rebalancing is superior to time period based rebalancing - I agree with this - it decreases transactional/tax costs yet maintains the risk profile.
b) it might be better not to rebalance at all - this i disagree with as noted in my comments above

[Vig Oren]

Mark, notice the contrast in above report with the discussions with the following. Isn't it just mindbogglingly?:

Source Link:

www.financial-planning...=

Excerpt:

Rebalance, Rebalance, Rebalance

During this period of market turmoil, we have been following the advice of Gobind Daryanani, creator of the rules-based rebalancing software iRebal, who has conducted detailed research on the importance and logistics of portfolio rebalancing. He believes it's prudent to "look frequently" for rebalancing opportunities—instead of rebalancing over time, do it by setting a narrow tolerance band for allocation changes. Daryanani estimates that across a wide range of market conditions, the benefits of rebalancing based on returns can be more than double those garnered from traditional annual rebalancing. His data suggests a rebalancing benefit of 55bps, with 22bps of the total attributable to capturing short-term momentum and mean reversion effects of asset class performance. In a period of potentially lower and more volatile returns, this incremental return adds significant value.

According to Daryanani, it is particularly crucial to look for rebalancing opportunities during volatile market environments. Otherwise, "you are likely to miss the rebalancing benefits during periods of short-term fluctuations." His most recent research proposes an approach he calls "opportunistic rebalancing," which is designed not only to control portfolio drift, but also to capture buy-low/sell-high opportunities as asset class performances drift apart in the short term. These transient rebalancing opportunities occur sporadically and can only be captured if the advisor is monitoring client portfolios frequently—on a daily, weekly or biweekly schedule. Operational challenges would indicate that advisors would do well to look at client portfolios on a biweekly basis to ensure they identify and capture these opportunities.
----------------------...
p.s. I would go with Gobind Daryanani at any time, b/c he is thee expert! Even if it's just a bonus of 35 bps annually. Heck, isn't "correct asset location" and "tax- loss harvesting" at same scale? At least enough to cover advisors' fees.

[User 220633]

It sounds as though your model includes the tax and transaction costs associated with each rebalancing transaction. Wouldn't the comparison be more fair if you convert each portfolio to all cash at the end of 25 years? That way taxes would also be assessed on the portfolio that has never been rebalanced.