Feynman Portfolio Study – Part 1

This first Part of the Study identifies the 10 Asset Groups and 18 individual EFTs that comprise the Feynman Portfolio and looks at the general market behavior, as defined by the Vanguard Total Stock Market Fund (VTSMX), over the 6 year period from 06/29/2007 to 06/30/2003.  The VTSMX will be used as the reference, or benchmark, for comparison of portfolio performance through each future phase of the Study. Ideally we would like to use a more balanced benchmark like Lowell’s customized ITA Index, but unfortunately no such standard index exists, so we will use the VTSMX.

Over the 6-year period of the Study, the value of the VTSMX increased by 10.81% or at a modest Compound Annual Growth Rate (CAGR) of 1.73%. This barely absorbs the impact of inflation. During this period the US equity markets absorbed a maximum drawdown of 56.53% (based on VTSMX) over an 18-month period before taking almost 4 years to rebound to its starting value. Remember that a 50% portfolio drawdown requires a 100% “bounce” to recover to initial portfolio values.  Volatility over the period also fluctuates significantly.

Future Parts of this Study will examine how the Feynman portfolio might have performed during this period of major market moves. Each part of the Study will focus on one aspect of portfolio construction or Risk Management and reference the performance of the portfolio to the performance of the VTSMX.

Although regular readers of this Blog will be familiar with Lowell’s use of the Hoadley Portfolio Analyzer to generate Efficient Frontier profiles, Part 1 of the Study also introduces the reader to other sections of the analyzer that Lowell does not usually show. More details will be provided in future parts of the Study and it is hoped that this will give the reader a better understanding of this analytical tool and how it is used.

Before closing this Post I would be remiss not to mention some of the benefits and pitfalls of back-testing. Back-testing inherently assumes that the analysis of past performance can help us improve future performance – this is an arguable point – especially if past history does not reflect likely future behavior. In this Study we have chosen a period of market activity with little change in total market value from the beginning of the Study to the end. Thus the Study period is not dominated by either a predominantly bull or bear market. In the interim, the market sees both major and minor moves both to the downside and to the upside. It is difficult to visualize a scenario with much more diversification, so, from this perspective it would seem fair. However, there are numerous routes the market could take to get from beginning to end with the same level of fluctuation or volatility. The use of Monte Carlo analysis is useful to examine these differences and the QPP software used by Lowell to analyze some of his portfolios is one way to examine the impact of these fluctuations. Although I have significant experience in using Monte Carlo techniques, I will not be including this type of analysis in the Study. Lowell may choose to run the Feynman Portfolio through QPP to provide additional insights.

Perhaps the biggest pitfall in back-testing is the temptation to look at the data with hindsight and develop a “system” that demonstrates good performance – this is often referred to as curve-fitting or over-optimization. Unfortunately the “system” invariably performs poorly in forward looking tests.

In this Study I go out of my way to avoid any tendency to curve-fit or over-optimize by using logical rules that can be reasonably applied under all market conditions.

Unfortunately, there is no Crystal Ball or Holy Grail and historical data is all we have to work with. I firmly believe that any price chart merely reflects investor’s collective reaction to news and to their general feelings of optimism or pessimism in the markets/economy – sometimes these reactions run contrary to what we might expect (e.g. price drops following good earnings report) but, over time the reactions are repeated. For those readers that may be familiar with Chaos Theory – there is Order in Chaos.

Please download the Word file at this link found on my PogoPlug site to view Feynman Portfolio Study – Part 1 and be sure to provide feedback as to how these documents might be improved or for clarifications if I have not explained things clearly.  If you have problems downloading the Word document, let it be known immediately.

Part 2 will examine the benefits and limitations of diversification using a “Passive” Feynman Portfolio.  Stay tuned.

Part 2 and Part 3 of this study are now available here at ITA Wealth Management.


About HedgeHunter


  1. Note that I added a new Category, Feynman Portfolio, under the ITA Portfolios.

  2. Robert Warasila says:

    I read the summary and think this will be a good portfolio to test. I have a suggestion that may already be in your plan. Several years before the inauguration of the blog Lowell and I played with testing the limits on the rebalancing trigger. We used the 6 domestic equity groups, international equities and REITs. We left out bonds and commodities. We found that a trigger between 20% to 30% maximized gains on this all equities portfolio and that is a number (25%) I still stick with. I’m sure some fraction of the subscribers share my interest in minimizing oversight chores and I think rebalancing fills that bill. I suggest one of the models you test as a benchmark in a sense, is simple rebalancing at the 25% collar.
    Bob W.

  3. Robert Warasila says:

    BTW we tested the period 1994 to 2007.

    Bob W.

  4. Richard Yalmokas says:

    The constraints should be self-explanatory but it is worth noting that equities (all groups) will be represented by a minimum 23% and a maximum 100% of available funds. Similarly, Bonds will have a 0% minimum and 55% maximum representation.

    I’m probably missing something obvious, but I didn’t think Hoadley would allow something like this (i.e if you have a 23% minimum w/equities and a 55% maximum with Bonds, that doesn’t sum to 100%).

  5. HedgeHunter says:

    Bob W,

    I need to think about what you may be suggesting, but let me make absolutely sure I am understanding – are you suggesting a 20-30% trigger on the asset gains before rebalancing?


  6. Len Suelter says:

    Per your request, I can read the comments. Took me a few minutes to figure out that I had to click on the comments header.

  7. HedgeHunter says:


    These constraints are ok – to sum to 100% we just need a) > 45% equities (easy solution with a max constraint 0f 100%) or b) a balance that includes other asset classes that fill the gap. I’m glad you raised this question because it reflects the fact that considerable thought needs to go into setting these constraints – the optimizer can then help “fine tune” the portfolio within the constraints – this may become clearer when you read Part 2 which we will post before the end of the week.


  8. HedgeHunter says:

    Bob W,

    Have you tracked the performance of this “trigger” on the same portfolio through the 2007 – 2013 period? What did you do in the 2008 bear market?


  9. HedgeHunter says:


    Sounds like a comment that is worth considering in your planned website re-design.


  10. I suspect the Genesis overlay prevents me from using “Recent Comments” in other sidebars other than the Primary Sidebar.

    The new version may be different. Otherwise one always needs to go back to the Home Page to pick the Recent Comments.


  11. Robert Warasila says:

    Not to the extent we did up to 2007.

    Bob W.

  12. Robert Warasila says:

    Yes what we did was look for an asset gain or loss of 10%, 15%, 20%, 25%, 30%, 35%. The complication is how do you redistribute the $ of a gain or draw in money to reallocate in the loss. What I did was reinvest in the assets that were trailing or gaining the most.
    Bob W.

  13. Robert Warasila says:

    I basically did buy and hold, sold call options, hung tough. Call options cost me a number of stocks that I would have been better off holding onto. I regained all my losses, but that probably says something about what I was invested in. As an aside in Fall 2009 I started a new normal portfolio based on PIMCO suggestions and Considine’s (QPP creator) suggestions. That portfolio performed OK until the current “bull” started in late 2012 and continued through now. The IRR is now trailing VSTMX by some 5% while it had led by 2% or remained even from 2010 through late 2012. The market has definitely changed. I’m maintaining that portfolio until we hit the 5 year mark, fall 2014.

    Bob W.

  14. HedgeHunter says:

    Bob W,

    Ok, thanks for the clarification. Rather than get into lengthy discussions of these issues here I’ll keep your suggestions in mind as I go through future parts of the planned Study and add something if it fits in naturally. To some extent, the parts of the Study that examine the impact of Optimization and rebalancing at least partially address some of these questions. When the planned Study is complete, if necessary, we can go back and look at some of these options.

    A word of caution regarding the method you used to establish your profit targets – there is a danger of “curve-fitting” or over-optimization in doing this – I’m not saying that your levels are right/wrong/inappropriate etc – just be careful and convince yourself that the “system” is robust – i.e. works over a wide range of market conditions over long periods of time.

    Maybe a quick comment here on pros and cons of setting targets. There are differing views on whether this is beneficial but I think it is true to say that the most successful traders let winners run and cut losers quickly i.e. take small losses (note I’m distinguishing traders from Investors here – although even Buffet lets his winners run) – Maybe I’ll post an explanation of “expectancy” at some point if readers aren’t familiar with the concept and there is an interest – but again, it’s probably more relevant to “traders”. In my trading accounts I will sometimes set profit targets based on Fibonacci extensions and take at least partial profits if these targets are met – but I’m not sure this is the right thing for an Investor, especially if managing a portfolio based on asset allocation.


  15. Bob,

    Here is my memory of the study we performed. 1) It was an 18-year study as that is as far back as I was able to come up with data. 2) We used eight asset classes as that is what the Callan Periodic Table covered back in late 1980s and early 1990s. 3) The asset classes included several of the “Big Nine” U.S. Equity classes, international, and perhaps U.S. Bonds. 4) The portfolio was $80,000 as we “invested” $10,000 in each asset class. 5) Rebalancing was done only once a year – at the end of the year and we rebalanced when an asset class moved above or below target by the various percentages going as high as 35%.

    The idea was to find the “sweet spot” for rebalancing as we know it is frequently a good thing to let an asset class run longer than one might assume is reasonable. As I recall, and it depended on whether 2007 was included in the study, the “sweet spot” was around 30%. In other words, the original $80,000 original portfolio ended up higher in value if we rebalanced the eight asset classes only when they exceed 30% of the target range.

    This is my memory of the study and it could well be flawed.


  16. Hamilton Arnold says:

    I’m quite interested in this work, especially whether the optimizer can be used to manage a more tactical portfolio (one whose weights change over time). I’m a bit puzzled, though, over what I see so far in the two Word documents. If I read them right, the “Passive Portfolio” weights are chosen on 3/31/08, presumably using performance data up to then. But then performance of the overall portfolio is shown beginning three quarters earlier, starting on 6/29/07. I read the comments on rebalancing – I assume the Passive Portfolio is rebalanced at some points to return it to the original asset allocation. I’m also puzzled over the statement that the Passive Portfolio remains on the Efficient Frontier as it moves over time. If the calculations always use the latest available performance data, I could picture a situation where an asset class that was included in the original portfolio completely blew up and would then receive a zero weight throughout the newer Efficient Frontier. Any portfolio that included it would then fall below the newer Efficient Frontier. Finally, I don’t see the Appendix to the second document that shows the Efficient Frontier evolving over time. I’d love to see that, and also how the asset class weights for the optimized portfolio change over time (the first step toward a tactical portfolio). Keep the good work coming!

    Pete Arnold

    P.S. Lowell, your last Sunday’s mention of Renaissance brass music was great. I bought the mp3 version (instant gratification) and now have it in my car. The difference between modern and period instruments is amazing t0 hear. Thanks.

  17. HedgeHunter says:


    Let me try to address your questions:

    1. Sufficient data for all 18 ETFs was only available from 3/31/08 so the portfolio is “passive” from that date i.e. no further optimization. Because it was desirable to start the analysis prior to this date (i.e. 6/29/07) I started the analysis on 6/27/2007 using the 16 ETFs for which data was available, RWX and BWX were added in the interim period, as data became available, and the portfolio was re-optimized on the date at which these assets were added.

    2. The answer to your initial question is probably implied in the above answer – i.e. yes, the analyzer can be used to track the performance of a TAA portfolio – but it is a manual process, not automated. Part 3, which should be online later this weekend, will show a demonstration of this.

    3. No – as in 1 above, there is no re-allocation/re-distribution of funds to other assets after 3/31/08 – no attempt is deliberately made to keep the portfolio on the EF – that’s just where it stays – even though the EF moves, the “passive” portfolio stays on it –return/risk ratios change, but it stays on the line – that’s why the observation is so interesting and potentially important. Take a close look at Figure 4 and notice e.g. that IJS,IVW and IJK can all have allocations between 0% and 15% and yet lie on the EF. I agree that it’s surprising that a “passive” portfolio seems to “adjust” to the new EF – but that’s the observation we should probably try to understand.

    4. To see the EF evolving over time in Appendix 1 you will have to study the EF charts carefully, one by one. However, if you wait until I post the Part 3 findings you will see this on one chart and also find (hopefully) answers/clarifications to some of your other questions.

    Thanks for your interest – I’m hoping I can post Part 3 tomorrow – but there’s a lot of work behind this and it takes time and care.


  18. HedgeHunter says:


    Check the “Shares Held” Column of the top table in Appendix 1 – this is the number of shares held in the portfolio in the period up to the date at which the portfolio value is calculated.

    The numbers in the “Optimal Shares” column are irrelevant in Part 2 of the Study – these will be introduced in Part 3 when we rebalance.


  19. All Readers: This issue comes up from time to time as to whether dividends are included in the Hoadley SS calculations. One ITA subscriber asked Peter Hoadley and here is his response.

    “If you are referring to the PriceHistory toolbar method of downloading price
    history (see sample sheet 182) then adjusted close is used (see
    documentation for the price history toolbar which explicitly mentions this).

    The price history class (see sample sheet 183) returns both close and
    adjusted close.

    Applications that use the price history class like the Historic Volatility
    Calculator and the Portfolio Optimizer use adjusted close as adjusted prices
    should always be used when calculating volatility, beta etc.”

  20. HedgeHunter says:

    I have double-checked this and can confirm that the prices downloaded into the Hoadley Optimizer, and therefore used in all calculations in the modified MOM file, are, in fact adjusted values – i.e. they are adjusted for dividends and splits. I thought I had looked at this in the past and decided that they were actual historical (unadjusted) values – but not so.


Speak Your Mind

%d bloggers like this: