Diversification? Portfolio testing almost always just gives 1 for number of positions

Hi this is just a general question for other portfolio model developers out there. I've been working on developing rotational trading systems using models made up of ETF's of major stock indexes, fixed income, commodities, and other asset classes. And while everything I read talks about the benefits of diversification, almost every single one of the models I design perform "best" focusing on a single ETF for the rotational period. For example, just as an example I'm testing a model with a basket of about 14 ETF's, that ranks all of them by, say, Rate of Change over the past 3 months, and then rotates them on a monthly basis. My code is optimizing anywhere from 1 to 10 positions. Almost every single test shows a higher net profit as well as a higher CAR/MDD and higher Ulcer Performance Index when investing in just one ETF for the month, rather than a basket of 3 or 4 or 5. It seems to defy logic and what we've all been taught from day one, but I'm seeing it in several different ranking methodologies I'm using--not just 1 ROC but using ROC for 3 different time frames all weighted differently, etc.
Are other model developers noticing this phenomenon as well? If so, should we just "go with the flow" and trade that way? Do you have any recommendations? I'm certainly not comfortable putting all a client's money into a single ETF for a month, but the numbers don't lie. I'm going back at least 18 years in my testing to make sure I get both bear markets included in my results, and still it almost always picks just one ETF.
It makes me think that momentum does work in the market, but the market favors just the winner, there are no points for second place.

Diversification is not for increasing performance. Diversification is for lowering the risk. Lower risk (usually) means lower profit.

@TonyR you appear to be discussing Tactical Asset Allocation (sometimes referred to as Adaptive Asset Allocation). I think depending on the securities you are back testing with, and the time frames, you can find a wide variety of results using those rotational approaches. Not only do the assets you choose make a big difference, but the day of the month you decide to rotate can make a huge difference in your results. I personally do not find that to be very reassuring in terms of robustness.

The academic literature is where you might consider looking into this question, and there has been much written about this in the past few years. But in general I have noticed that we find greater volatility (often referred to as a measure of "risk") if you focus your allocations on one asset.

I know you are familiar with @TrendXplorer 's blog so read through his posts and his whitepapers. But I suggest you review the writing of Mebane Faber who popularized this entire field with his original publication around 2007. He has a blog, more papers and even a free book on the subject.

Lastly some asset managers that write copiously on the subject would be worth a read, like ReSolve Asset Management. They too have an educational blog, whitepapers and a book.

Since some of those are commercial sites I haven't included web links but they are easy to find with the information I have listed.

Good luck!

1 Like

Ah yes TJ, I guess it would be better to look at diversification as more of an "insurance policy". I think the risk numbers are skewed because with 1 security the result show a huge net profit, but also a drawdown of say 30-40% (which to me is unbearable), while the results that have 2 or more securities have a much more reasonable net profit (maybe a quarter or a fifth of the single security), but it's drawdown numbers will be closer to 20% (tolerable). I just found this all very interesting and wanted to make sure I wasn't alone. I appreciate the programming, coding, technical side of this forum very much, but would also like to talk trading concepts as well. There are so many brilliant and talented people on this board.
Thank you,

@TonyR IMHO a site that offers a lot of interesting research in this field is CXO Advisory (to access some the most relevant content it requires a subscription, but it is a cheap one - and can be just a month).

In particular, I suggest you to read about their ETF based Momentum Strategy (SACEMS).

@portfoliobuilder all the self-published e-books of Meb Faber now are all free to download (here is why - a pretty interesting story!)


Thank you portfoliobuilder, I always enjoy your posts and very often learn something new. Yes I'm working on different iterations of @trendxplorer's ideas and code that he's very generously shared, as well as a few other ideas. I will research the other sources you recommended as well.
Thank you for the ideas and the reply.

Thank you @Beppe, I've had CXO Advisory in my bookmarks to dive into when I had more time, but I'll get into it right now based on your recommendation.

@TonyR A quick and limited test using the GTAA code as previously shared, but expanded with AddCustomMetrics for UPI(0%) and K(20%), and running 2D sweeps on Topsize (1-10) and the SMA's lookback (1m-13m), results in diversified portfolios (Topsize >= 5) as being optimal when targeting on MAR, UPI(0%), and K(20%). Only when using CAR as optimization target, concentrated portfolios (topsize = 2) appear to be "optimal" (small ridge instead of the preferred plateau).


GTAA N10+2 2000-2018 MAR:

GTAA N10+2 2000-2018 UPI(0%):

GTAA N10+2 2000-2018 K(20%):

GTAA N10+2 2000-2018 CAR:


These are great topics for discussion and I miss it here.
A few things:

  1. ETF's ok, for indices, maybe even some bonds.
  2. commodities less so, key word roll-over / cost of storage etc.

For a really diversified portfolio I would consider a good 40+ markets, among these all asset classes. I would strongly consider using futures.

When you optimize the run, your ETF won't go back much further than 2000.
In addition to having too few markets, your optimization may simply curve fit to the best performing ETF's and do much better so, the fewer you choose.
Play with the parameter settings, and adjust the position number +/-2 if your performance jumps down drastically, you have been sitting on a peak and over-optimized things. I would trade what appears to be as robust as possible, opposed to what makes most performance. CAR/MDD over 1 is certainly not sustainable over a 20 year run. Someone may get lucky, but I wouldn't hope for it. Settle with something that keeps you trading.,

There are many variables, i.e. also how do you adjust positionsize? a simple 10% allocation or risk parity of what u are doing there make a huge difference. How about rebalancing?


Thank you @TrendXplorer for the additional information and analysis. It's obvious that I need to dig deeper into the results of the system tests as you have shown. My post was inspired by 3 different tests using radically different ways to measure momentum, and each one had a position size as 1 as it's best performance not only in net profit, but also CAR/MDD--where the ratio remains strong because of the outperformance of the CAR even though the MDD is probably worse than most investors could tolerate. Thank you for sharing!

That's an excellent point as well, a lot of the ETF's we can use to build a diversified portfolio among different asset classes didn't exist prior to 2007, and many not till much later. Thanks for the ideas @Dionysos I'd like to see more discussion about investing and trading as well. There are so many brilliant people here and I'd really like us Amibroker users to be more of a tight-knit community.