How often should you rebalance?

In this post:

Rebalancing frequency and scheduling seems simple at first, but is actually a tricky topic

• Costs matter obviously. How sure are you that your estimate of rebalancing cost is realistic?

• Information decay is faster in some factors and slower in others. What does the literature say about this?

• A checklist to decide on “a rebalancing frequency that is right for you”.

Rebalancing is not straight forward

When one is starting out in the factor world, frequent rebalancing seems like a really good idea. At least it did to me. Whether you are running tests to check for correlations between signal and return or move on to backtests to confirm and model out the effects on portfolios over time, the results will on the surface tell you that you should rebalance as frequently as possible to extract the maximum possible premium (or alpha if you are doing non-factor(beta) stuff, this is still relevant outside the factor realm).

There are other positives to rebalancing more frequently as well, you are all else equal less prone to rebalancing timing luck effects (see the paper “Rebalance Timing Luck - The Dumb Luck of Smart Beta” by Corey Hoffstein: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3673910 or this post by Alphalayer: https://www.alphalayer.ai/news/timing-luck-in-factor-rebalancing which discusses its contents. How sensitive your model is to these “RTL” effects depends on both how often you rebalance and how quickly the signal itself turns over.

In general, I think the following is true: The fewer instruments you trade, the higher the risk of luck playing a big part. The more frequently the factor signal changes, the more prone you are to luck effects in your performance measurements. That means high decay strategies such as momentum are more luck prone than low vol for example. Also generally: the more seldom you rebalance, the more luck/randomness plays a part in your performance figures.

The most luck-prone strategy, or the one likely to show the most variability in results, would therefore be one that:

1. runs a concentrated portfolio,

2. uses high turnover / signal decay factor signals to generate trades and

3. also rebalances quite infrequently.

The least variable strategy would be one that diversifies over a larger number of instruments, uses slower-decay signals to generate trades and also rebalances often enough to, or on a schedule that, minimizes variability (for example a staggered date part-rebalancing scheme or running several instances of the same strategy on different rebalancing schedules, or in Hoffstein’s words: “equal weighting across N sub-indexes reduces rebalance timing luck by 1/N.”

There is a trade-off between optimal rebalancing strategy and practicality of execution, however. Put simply, if you are runnning 20 instances of the same strategy in parallell it will certainly remove a lot of luck effects and probably decrease portfolio volatility as well, but it will be very cumbersome and challenging to execute.

People have different preferences when it comes to this, as with everything else. What RTL risk are you willing to take? What volatility are you willing to stomach?

How sure are you that your estimate of rebalancing cost is realistic?

Speaking generally, a lot of people seem unrealistic in their assumptions on rebalancing costs. Sometimes wildly off! Since commissions are now so low that you can trade for just a few bips even as a retail client, trading frequently seems realistic at first. It’s (again, generally) not. If you are not very, very small player, you will very quickly realize that you are not only having to pay the bid-ask spread, but also having an additional impact on the price in whichever direction you are trading. Sure, some super liquid stocks and ETF’s can be traded without much of an impact even in size. But you have to remember that as a rule, factor premia are lower in the most liquid, largest stocks. So to maximize premia, it pays to trade smaller stocks as well (to a point). The real issue then, is slippage (the difference between the price at which you would like to transact and where you end up transacting), or market impact.

Jean Phillippe Bocheaud wrote a paper called “Price Impact” in 2010 that is considered foundational when it comes to estimating market impact of trading. According to Boucheaud, market impact can be estimated using the formula I(V)= σ⋅ SQRT(V/V daily), where

• I(V) is the estimated market impact.

• σ is the daily volatility of the asset.

• V is the volume of the trade you're executing.

  Vdaily​ is the average daily trading volume of the asset.

  This paper is often quoted in the literature on market microstructure, and Bouchaud and others have empirically validated this relationship, especially for large institutional trades. So it seems reasonably accurate. I use it to estimate how many days I need to spread my trades over to avoid impacting the market too much, as I trade both liquid and a little less liquid stocks (I try to avoid the really small and illiquid stuff).

  To sum up, the most important cost of trading - the market impact - is unknown a priori. You can get a good estimate by thinking ahead using empirical data from your own trading or calculations like the Boucheaud formula for example, but there is no clear answer. Additionally, there may be other costs to consider. If you are trading foreign stocks, you are by definition incurring other costs as well due to currency conversions. And brokers screw you on those, big time. Not on commissions, but on spreads. For some reason, they seem to get away with spreads on the order of 0.2-0.4% where I live. That has a huge impact on a high turnover strategy! So think carefully about the real costs of rebalancing before committing to a high turnover strategy. You can easily rack up 1% total costs from trading illiquid stocks aggressively and paying a high conversion fee on currency transactions. Good luck finding an edge that overcomes 1% monthly friction costs.

  Information decay is faster in some factors and slower in others. What factors you use matter.

  Consistent with other findings in the literature, a recent paper by Emlyn Flint and Rademeyer Vermaak (“Factor Information Decay - A Global Study”) finds that information decay is fastest in pure momentum strategies and slowest in value, with quality and low vol falling somewhere in between. All else equal then, momentum strategies should be rebalanced more frequently than other factor portfolios. Note that the authors use pure factor strategies in this research. The paper is not freely available online, but Alpha Architect has written a summary of it here.

  Since trading frictions can kill an otherwise attractive strategy, and decreasing the rebalance frequency can lead to rapid performance decline, a good solution might be adding a complementary factor to your model - if the authors of the above study are correct. which they probably are. Combining factors looks like a good idea more broadly (see for example this paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4080705,, so why not exploit it to reduce rebalancing costs as well?

  You might also want to see for yourself if this holds true for the factor you use in your models. Running tests on portfolios of different sizes in different periods with different rebalancing schedules is tedious and time consuming, but informative - it may at least give you some idea of what exposure your strategy has to rebalancing luck effects. Of course, the future will not be exactly like the past, so take the results with a grain of salt, but I think this has been a useful excercise when deciding on how to structure my portfolio. Overall, what I’ve found in my own research is consistent with that of the authors above as well, so I try to reduce rebalancing frequency by combining factors.

  Note that when you are researching this, you need to remember that in order to overcome the confounding effects of periodic underperformance of one or more factors, testing over very long periods of time as well as different geographies is required. Otherwise you don’t get a sense of whether or not it is effective in general. E.g adding value to a momentum strategy over the last few years would not have improved your strategy, quite the opposite. Over shorter time frames, it’s very hard to see these effects clearly. Over long periods and different markets, it becomes clear(er) though.

  Summing up: A brief list of questions that you should answer before deciding on a rebalancing frequency for a given strategy:

• How many instruments are you trading? Does trading more instruments really hurt the expected returns or might it be a good idea overall?

• How liquid are the markets you are trading / what is your estimated REALISTIC cost of rebalancing, including commissions, slippage, currency spreads?

• How quickly do your factor signals decay? Can I reduce the decay rate by adding a complementary factor (and not hurt my returns)?

• How much time and effort are you willing to spend on trading and rebalancing a strategy?

• And finally, and beyond the scope of this post: have you considered adding more markets and more strategies instead of tweaking this one? It’s often much more important for your results as a whole than optimizing a single strategy to death.