Portformer Methodology - Network

Creating Better ETF Portfolios: Finding the Competitive Edge

We wanted to thank those that attended the webinar on November 19, 2019. We had a great conversation and it was exciting to reveal our “Portformer ETF Influencer Portfolio“. This portfolio is designed to look at our highest scoring ETFs across the entire ETF landscape and to organize them into clusters. We picked the ETFs with the highest ‘Influencer Score’ in each cluster and backtested a simple inverse-vol weighted strategy with compelling results. For us, the most interesting thing was that the portfolio was organized differently than most standard ‘model portfolios’. Although some of the clusters corresponded to the typical style boxes, real-estate and commodities had very unusual (and yet intuitive) allocations.

Let us know what you think in the comments below or on social media!

Key Takeaways:

  • Building a portfolio is a like a chef preparing a meal with a recipe and a selection of ingredients
  • Portformer™ creates head to head comparisons of funds generating a replacement score considering the 4 C’s: Consistency, Completion, Concision and Confidence
  • Looking at a simple 60/40 portfolio strategy, it was presented how straightforward it is to look beyond traditional style boxes and create a strategy of similar return characteristics with lower expenses, reduced risk and increased availability of capital; improvements in drawdown profiles were noted.
  • The presented research highlighted the Portformer ETF Influencer portfolio using mutual fund/ETF data, network analysis and advanced statistical techniques. The portfolio was the result of reviewing the entire ETF universe and identifying the best opportunities for increased return and improved diversification.
  • The resulting selection of Portformer ETFs presented is a better approach and what we consider as the core building blocks for portfolio construction.

Regardless of how you build your portfolio, make sure your buckets are diversified and your investments are the best ones for each bucket

We get it. The webinar its 20 minutes, and you’d rather skim this article rather than view the webinar. No worries: here are the key slides:

Building a portfolio is like creating a meal

However most advisors focus on the Recipe rather than the Ingredients

Part 1: Choosing better Ingredients: How to use Portformer to select better securities

We talk about this in more detail this during our WealthStack 2019 Presentation and in our two-minute video tutorial for new clients. The main takeaway here is that having the right asset allocation, but the wrong securities in a portfolio can have massive implications on each of your client’s wealth and the value of your practice.

This plot represents the distribution of 10 year returns from choosing 70 different Large Cap funds and the AGG ETF in a standard 60% stocks / 40% bonds strategy.

Part 2: Improving the Recipe: How to use network analysis to make better portfolios

In the future we’ll write a more comprehensive ‘Intro to networks for advisors’ piece. Until then please watch the video from 7 min 0 sec to 9 min 40 sec. The crux of the argument is that we can treat each of the Portformer Relative Replacement Scores like edges in a directed graph. We can then use the same techniques used to analyze social networks to create diversified and efficient portfolios.

Confused? Excited? Curious? This is a relatively new technique to financial markets who have previously been obsessed with Factor Models and Principal Component Analysis (PCA). If you’d love to geek out on this with us, schedule a quick call with our founder, Sean Kruzel.

How the top 1,461 ETFs are organized into networks

Traditionally there are three steps to create a portfolio:

  1. Filter and divide the investable universe into ‘buckets’
  2. Decide the best investment for each bucket
  3. Determine how much to invest in each bucket at each point in time for each type of client

During step 1, typically buckets are defined using Geography (US, Asia ex-Japan, etc..), Asset Class (Stocks, Bonds, Commodities), Styles (Small Cap, Large Cap, Momentum, etc..), Factors or a combination of them all. In our case we are going to use Network Components that were automatically detected based on how our Portformer Scores connected ETFs in our universe.

An ETF can be used grouped based on how ‘popular’ it is

Rather than looking at popularity metrics to group ETFs, we used a clustering technique called Louvian Groups to automatically detect 15 clusters. This analysis is used to segment regions of the brain and portions of the internet; it’s a powerful approach to apply as we try to discover patterns in the ETF universe.

The graphic above shows each ETF as a circle (also called a ‘node’) and each grey arrow (the ‘directed edge’) is a Portformer Score. We show the 1,461 ETFs along with over 40,000 top Portformer scores in a two-dimensional plot that allows for highly connected nodes to be grouped together and ETFs with fewer connections to be placed closer to the edge. The nodes are then colored based on the 15 clustered that were identified using the Louvian Group analysis. You can observe that some of the colors are often grouped throughout the plot. We have manually circled some clusters of ETFs that correspond to certain niches. The large cluster in the middle likely corresponds to the ETFs with a high ‘beta’.

We then aggregated all of the nodes with the same color to show the connections between the clusters. In the network above we can see 15 nodes created from each of the clusters and with edges corresponding to the existence of Portformer Replacement scores between clusters. Note that some of these clusters only contain a couple ETFs (like the clusters for Sugar, Coffee and Natural Gas), while others like Global Growth contain hundreds.

After careful review of the clusters of ETFs identified by each of the colors above, we manually labeled each of them to loosely describe the collections of ETFs contained within each cluster. This is an approximate description and does not contain or necessarily correlate with the funds’ self-reported categories. For our purposes, we are going to treat each of these as the ‘buckets’ in our portfolio analysis.

How to pick the ingredients for each cluster in the network

After running the clustering step in the previous section we now have 15 clusters with between 2 and several hundred ETFs located within them. There are many ways to select which ETF to use in a portfolio. This is not an endorsement or recommendation for or against a particular ETF. Please consult with a professional before making any investment decision.

Some of the ways we could have selected the ETF from each cluster is:

  • One with the most assets under management (AUM)
  • Lowest Fee
  • Highest Aggregate Portformer Score
  • Highest Influencer Score
  • Randomly or some other method

In this case, we only evaluated the portfolio with the ‘Highest Influencer Score’ This score is determined by selecting the ETF within our network that had the highest ‘Page Rank‘ score. The details of Page Rank are beyond the scope of this article, but this can be thought of as simple version of the ‘Google 1.0 search algorithm’. In our case we are estimating the most important ETFs in the network by assuming that “important ETFs are connected to other ETFs”.

We now have the ingredients and the buckets to which they correspond.

Portformer ETF Influencer Portfolio v1.0 Constituents

These ETFs were selected using Portformer’s 2019 vintage scores, grouped by Louvian Clustering into buckets and then Page Rank was used to selected the ETFs with the highest influencer scores in each bucket.

CategoryBucketETF Ticker
US EquitiesUS EquitiesIWY
Utilities / Real Estate / Low VolVPU
Global EquitiesGlobal / EM EquitiesVT
Global GrowthVONG
BondsFixed IncomeVCLT
Misc Fixed IncomeBSCP
CommoditiesPrecious MetalsOUNZ
Natural GasUNL
SHORTSShort ‘Risk Assets’EFZ
Short Precious MetalsDGZ
Short TreasuriesDTYS
Short Real EstateSRS

The Recipe: Portformer Trading Strategy

These are preliminary results and are strictly theoretical. However we only ran this backtest only once using standard portfolio construction techniques to minimize the over-fitting of this investment strategy. The strategy is reballanced quarterly with a historical turnover rate of approximately 14%.

The portfolios are weighted using Inverse Volatility Sizing based on trailing realized volatility of each of the 15 ETFs. Since the Portformer Replacement Scores are calculated with data between 1 Jan 2014 and 31 Dec 2018, we are considering all days since 1 Jan 2019 to be ‘out-of-sample’ / ‘Live’.

This portfolio is clearly performing well year-to-date but still remains within the expected two-standard deviation uncertainty cone. Below we show that volatility has been stable between 5-7% and that the rolling Sharpe ratios have mostly positive throughout its history.

Here are some more detailed performance statistics for the 48 months performance data.

Thank you for reading the whole piece! I hope this inspired you to think about how you could improve your investment process by breaking down the portfolio into its ingredients and recipe.

I would love to hear what you think of this work and if you love it, I would sincerely appreciate your sharing this on social media (use the buttons on the left).

Sean Kruzel

Based in Boston, Sean is a portfolio manager, MIT grad, and founder of Portformer. When he’s not investing or programming, he’s probably sharpening his skis for the next icy, New England snowstorm. Check out his full bio here: https://www.portformer.com/our-team


Recent posts

  • Workflows & implementable insights — to elevate each step of the advisor/client experience

    Use our head-to-head Replacement Scores™ to quickly create client proposals, discover funds, improve model portfolios, and monitor investment plans