The purpose of the Feynman Study was to take a list of 18 diversified assets and to demonstrate how that list could be used to construct portfolios based on a number of different strategies:
- Passive (Buy and Hold) Strategic Asset Allocation (SAA) Strategy;
- Active (Buy…..Don’t Hold) Tactical Asset Allocation (Re-Balance) Strategy;
- “Classic” US Equity/US Bond Portfolios;
- “Dynamic” Strategies based on Modern Portfolio Theory (MPT) and Minimum Variance Optimization (MVO) – Efficient Frontier Portfolios;
- “Momentum” Strategies
Overlaid on these strategies were the concepts of Risk Management and the use of “filtering” techniques such as ITARR and SHY Momentum to reduce risk and large drawdowns in bear markets.
Thus we have covered the essence of this web site – Portfolio Construction and Management.
I hope that I have succeeded in providing a smorgasbord of options from which Investors of all temperaments, and with different financial objectives/needs, can choose to apply to their own portfolios.
Although there is an obvious comparison of portfolio performance between the different strategies I have deliberately avoided the biggest weakness of back-testing historical data – the temptation to “optimize” performance over the period of analysis. Parameters used in the study are “discretionary”, based on my personal experience and familiarity with the techniques, but are not “optimized” to demonstrate optimal performance – I consider them “robust” in that they will probably provide acceptable performance over a wide range of market conditions but will not be the “best” parameters to use in any specific time period.
Part 9 of the Study described a “state-of-the-art” method to quantify the level of “diversification” in a portfolio and introduced the concept of asset “clusters”. A number of Platinum members requested a modified Excel Ranking spreadsheet that would help in the construction of portfolios incorporating these concepts and that SS is available as a downloadable Excel File here. This is the corrected spreadsheet.
This SS is not simple and needs a reasonably good familiarity with Excel and an even better understanding of Correlation Analysis as described in Part 9 – it requires the user to have purchased a license to use the Hoadley Finance Add-in for Excel and that it be installed. It also requires the use of the Hoadley Correlation Analyzer App that comes with the license.
Certain calculations can be automated – others cannot (without a level of effort that I don’t consider necessary and a technical ability that I don’t have). To use this spreadsheet follow the process provided in the “Instructions” sheet of the Excel Ranking File.
The SS uses my “proprietary” algorithm for calculating asset allocations (weightings) – although many sheets in the File are protected, the calculated weightings can be overridden if desired (be sure to save a copy of the original file if you intend to go back to the calculated values, since you will be overriding cell formulae).
One final point that I would like to make here regards the “Cluster Analysis” as described in Part 9 and implemented in the “Sample” File. If you look at the Cluster Diagram in the “Cluster” sheet you will notice that VTV and SDS are identified in the same “Cluster” with a 0.98 “Strength” linkage between the two. At first glance this doesn’t seem to make sense since VTV is a large-cap US Equity ETF and SDS is a (double) inverse US Equity ETF. However, it is important to fully understand the Cluster Hierarchy Diagram – the “Strength” indicates the level of confidence that can be given to the correlation – but it does not imply anything about the sign of that correlation – thus the fact that VTV and SDS are classified in the same cluster with a high “Strength” (r-squared) value simply confirms that they are highly correlated – in this case, negatively.
In an application sense, this will not be a problem since, for example, the “Ranking” of the 2 assets will be at opposite ends of the spectrum and (at least) one will be discarded by the SHY Filter.
The above is a little complex and needs some serious thought to be understood. This will likely not be important unless you include inverse (or very highly negatively correlated assets) in your portfolio.
David
Discover more from ITA Wealth Management
Subscribe to get the latest posts sent to your email.
David,
The spreadsheet is really well done and I can confirm works well. I will attempt to use it this month for analysis. One helpful hint I would make is that if one goes to the correlation analyzer and selects the step-by-step tab, runs the analysis, and then works their way from bottom to top, it makes the manual process much faster.
I am curious, where one changes how many etfs are used. The cluster for me showed 6 etfs and if I were to do the cluster4 method, how could I get the spreadsheet to allocate percentages properly?
Once again great job on the spreadsheet!
Steve
Steve,
You may want to go back to Part 9-1 where I go through the process of “building” clusters from 2 (weakest linkage) to 3, 4, 5 etc by “splitting” clusters.
David
Steve,
Thanks for your feedback and pleased to hear that the SS is working ok for you. Let me address your comments/questions:
1. Yes, running the step-by-step may help – I had problems knowing what to do with this until I fully understood the Diagram – now it makes sense to me – but now I also don’t really need to use it – except maybe for confirmation that I’ve done the breakdown correctly. This may be a personal preference.
2. In the SS I have (arbitrarily) chosen to use a 0.80 strength cut-off – this has given me 10 clusters as identified in the “Clusters” sheet. To reduce the number of clusters you would simply lower the cut-off level e.g. if I chose 0.65 this would combine clusters 1 and 2 to give me a single cluster – It would also combine clusters 7 and 8 (TLT) to form a single cluster, 9 and 10 (BWX and PCY) to form a single cluster and 3 and 4 (VNQ and JNK) to form a single cluster – a total of 6 clusters.
3. Having identified the clusters (in the salmon colored cells) in the “Cluster” sheet the SS will (should) calculate the weightings correctly and automatically – i.e. if you reduce the clusters to 4 you will have a max of 4 ETFs that will be weighted (provided they pass the SHY Filter).
Let me know if this does not make sense – it’s a little difficult to explain in writing.
David
David,
It makes sense. I did not communicate effectively what I was attempting to accomplish. I was attempting to keep 10 clusters and only take the top 4. I finally figured out that I could change the weighting factor to get the max 4 clusters. I somehow missed in your instructions the following quote: “The Weighting Filter can be used to limit the number of assets to be included in the portfolio if required, but the portfolio is also SHY Filtered automatically.”
I assume if I only wanted the four top clusters I should change the weighting portfolio to get those numbers. In the case of your sample portfolio I had to change the weighting filter to 13 to get 4 clusters (for 10/25/13).
Attached is a screenshot:
https://dl.dropboxusercontent.com/u/2586806/samplecluster.jpg
Steve,
You got it!
David
Test of latest SS: Add one more security (ETF) to the list on the Portfolio worksheet and let me know what happens when you try to rank the holdings.
Lowell
Lowell,
When I ranked I had it all go down to the bottom with an apostrophe in one of the fields. After I copied and pasted the correct formula it appeared to sort fine. Is this what you saw?
Steve
Steve,
Yes, there was 1 corrupted cell formula – Lowell will be re-linking a new file – but no other changes so no need to reload data if you have fixed your file.
Please let me know if you find any other “bugs”.
David
David,
Everything is working spectacularly even with mutual funds. The only issue I had and its a miniscule one is that on the BHS sheet, a couple of the rows heights doubled, presumably due to a 5 letter ticker, and the sheet is protected so I could not fix that. But to be honest I don’t think its worth bothering with.
Any thoughts on using the cluster strategy on mutual funds? My work is switching companies for 401K and I believe I will be losing my self directed account in April.
Thanks to you and Lowell,
Steve
Steve,
No reason to think the Cluster Analysis should be any less effective/valid using Mutual Funds.
David.
Lowell:
Please let us know when the new file David mentions above is re-linked.
Thank you,
Dick
The corrected spreadsheet (4.2.1CW) is now correctly linked in the blog post.
Lowell
David, Lowell,
Thank’s for the new SS. Very useful with the Hoadley add on I think. Clustering does show the utility of not over doing following too many funds for diversification. Now all we need is a tool to find the non-correlated ETFs a year from now :>). I ran a test of funds I had a “hunch” would have low correlations. Results were intestine but not promising. I’ll post for fun and to show the challenge
Robert
Robert,
Looking forward to seeing your low correlated securities.
David deserves all the credit for these momentum and cluster weighting spreadsheets. I’m still on the learning curve with the “Cluster Weighting” spreadsheet.
Lowell
I’d be interested in seeing the list too Robert.
Check in a year from now and I’ll give you next year’s correlations 🙂
David
Testing my understanding:
During the 2008 “financial crisis” all the correlations went to 1, and there was no place to hide. So, all the attempts to structure a portfolio of low-correlated assets went for naught in terms of buy-and-hold.
That’s why, for me, I’ve set up the best low-correlated port (based on cluster analysis) I can come up with; but I monitor it once a month. I don’t try to forecast correlations a year out, because they also change over time.
What are other folks here doing vis a vis their portfolios?
Dick
PS,
I mean no criticism of Robert or anyone else. I’m just trying to keep my own house in order.
PS,
I should have added that I use a momentum-based approach; and take things a month at a time, rather than trying to forecast correlations or anything else a year out. YMMV.
Dick
I’m pretty sure Robert was just joking Dick – but your approach is correct – we can adjust our portfolios month-to-month based on current correlations and allow the SHY filter to keep us out of the markets when everything correlates and the market (momentum) is bearish.
Go Sox!
David
Has all of your backtesting in support of the Feynman series done manually using the guass spreadsheet?
Robert,
I’m not sure I understand the question. The back-testing does not involved the Gauss Portfolio. All the back testing is posted under the category, The Feynman Study. The Gauss Portfolio is one of the thirteen portfolios I track on this blog.
Lowell
“All of the back testing is posted under the category…” <– I see all the summaries of your back-testing in the Feynman category, yes. I'm asking how you perform your back-testing. Do you have an automated tool/spreadsheet that applies your decision inputs and runs a test over the backtest time period? Or is it a manual process where you run your spreadsheet repeatedly with different end dates and compile a running performance data?
In short, my question is simply how you perform back-testing – do you manually use your spreadsheets or do you have an automated tool?
Why I'm asking: One challenge I have with the Feynman series is that it is all entirely based on a single back-test period. I have come across a simpler momentum strategy but it only uses lookback performance as a criteria – I really like the idea of including volatility in the decision. But for the simpler model, I have a spreadsheet with vba that I can set my parameters and press one key that will run 20 or more backtests over various periods going back as far as I have data for. I am using ETFs that go back 17 years (they map reasonably well to the same indexes you're using – smaller pool of course). The combined CAGR and MaxDD depending on whether the backtest period is 1997-2005 vs 2001-2012 vs 1998-2010 vs 2004-2013 etc. etc. is dramatic and important. Averaging all those periods for a particular approach gives me a much better view and expectations. All of your Feynman work is looking at a single window. It's all fine and good and useful but I am wondering if you have an automated tool you use for backtesting and if so, and it's based on one of your spreadsheets, is that something we could get our hands on so I can write a broader tool that will run it multiple times over multiple windows for a single set of input parameters.
Robert,
Unfortunately my VBA competence level does not go beyond the ability to generate simple macros – therefore the backtests were all performed manually and were not automated.
The six year period that was chosen for the Feynman study was the maximum period for which I could find data for all asset classes/groups included in the study. Even 2 of those had to be “eased in” during the first 9 months – but I didn’t want to miss the 2008 downturn in the market.
David
Would it be possible to refresh the link? I received this message upon clicking on the link from the article: Sorry! This share is no longer available.
Thanks,
Ken
I’ll see if I can find it or upload a substitute.
Lowell
Ken,
The file should be in your mailbox. My PogoPlug would not accept this new file.
Lowell