Algorithm and Design Updates (8/17/2021)

The weekly Focused 15 Investing publications now show two perspectives on the historical performance of the model portfolios. Part I shows the performance figures that would have been produced by the "current" model portfolio design and processes. These figures are based on the designs and processes as they exist today and represent the best picture of the relative performance characteristics going forward for all the portfolios in the publication. We do this so you can make fair and reasonable comparisons across the model portfolios and different Focused 15 Investment publications to determine which model portfolio is right for you. 

Parts II and III show (except as noted) performance figures as published over time period for the model portfolio. In the investment industry this is called "real-time" performance. These real-time/as-published figures reflect the performance one would have seen each week in the publications in the past. Since some portfolios have not experienced updates, Parts II and III may show the same figures as Part I.  

We do not track updates for the add-in sleeves and theme portfolio because these may be temporary portfolios.  The note at the center left of each detail page (green box) indicates whether the performance figures are either real-time or current design.  

These two perspectives (current design and real-time) differ when there have been updates to the design and processes used to create the model portfolios. This post describes the main updates that have taken place since the Focused 15 Investing approach was developed in 2007 that have an important impact on model portfolio performance.

As of September 2021, the main updates that cause the current performance to differ from the real-time performance are those for the:
  • D5 Signal Set for Avoiding Losses in the Dow Jones Industrial Average (DJIA) done in 2017 and in September 2021
  • Maximum allocation to the US 10-year Treasury Bond Index done in 2015
The table below shows the rates of return (annualized) figures for the Diamond model portfolios in a) real-time (as published) and b) based on current design of the model portfolio. The period shown is from July 18, 2014 to September 2021, which is shown on the second page of the weekly publication.


We periodically discover data errors from Bloomberg that affect both perspectives.  We correct both series when these are discovered.  

If one had invested using Diamond (sg131) since 2014, the return for your account would have been close to 15.4% (not adjusted for structural costs), as shown in column "a" for real-time performance.   

Looking forward, if we were to encounter the exact same market environment in the future, your account would return 17.3%, as shown in column "b" above for Diamond (sg131), because we would be applying the current portfolio to that future environment.  

As a side note, I did not include "Diamond Plus 80|20 (sg231)" in the table above. This portfolio was introduced as a temporarily more aggressive version of Diamond (sg131), and I did not report its higher performance figures at that time. Subscribers with long time horizons are now using it and I have decided to keep it in the publication and will show its higher historical performance.

Update to the D5 Signal Set for Avoiding Losses in the Dow Jones Industrial Average (DJIA)

The D5 signal set is used in most Focused 15 Investing model portfolios and consists of five different versions of the algorithms for avoiding losses in the DJIA.  All five are calculated each week and help determine the target weights, with each having a 20% weight in determining the final target weight for the DJIA-linked ETFs.  The five members of the D5 signal set are shown below with their objectives:

  1. The original version optimized to avoid losses in the DJIA from to 2009, plus an added stop-loss technique (discussed later)
  2. Similar to version #1 but also sought to avoid losses that were experienced in the stock market crash from 1929 to 1931
  3. Use the smallest number of parameters while delivering returns similar to the prior algorithms
  4. Use the least amount of “data smoothing” while still delivering competitive returns. Excessive data smoothing produces good simulations but poor real-time performance.
  5. Use explicit seasonal variables to capture returns throughout the calendar year

In 2017, I revised the stop-loss technique for version #1. The original stop-loss technique was optimized for use in portfolios that short the DJIA, which was well-suited to #1's original purpose. However, we don’t short the DJIA in the Focused 15 Investing model portfolios. The 2017 revision made the technique fit our long-only portfolios.

I am now (as of September 2021) revising member #1 again - we will not use any stop-loss technique at all for D5 signal-set member #1. The recently established Focused 15 Investing practice is to increase cash levels (Box #2 Cash) during periods when the algorithms are not well synchronized with current market dynamics. This makes the stop-loss technique in #1 redundant. In addition, if we need to reduce risk in the portfolios, applying a technique to just 20% of the loss-avoiding algorithm is not sufficient.

Using the current design of the model portfolio over the 20-year period, our returns would have been about 3% better each year on average for Diamond (sg131). Thus, the real-time performance figures, which capture the performance before the updates were made, are lower than what we would have obtained using the current model portfolio over that time period.  As indicated above, the performance figures labelled "current design" (beginning Friday, September 17, 2021) do not include this stop-loss technique.

Update to the Maximum Allocation to the US 10-year Treasury Bond Index

In 2009, I developed algorithms for rotating between ETFs linked to the US 10-year bond and the US 2-year bond indexes. They create a sleeve that is used in many model portfolios. While the algorithms themselves have been stable since their development, I have modified the design of the sleeve by reducing the maximum allowed in the riskiest of the two assets, the US 10-year bond.

Interest rates have been falling since the early 1980s. When interest rates fall, investments in the 10-year bond ETFs rise in value, all else equal. When rates rise, investments in the 10-year bond will decline in value. Over the 20+ year period shown in the publications, a high allocation to the 10-year bond ETFs has been beneficial. This was especially true in the early part of the period, when interest rates declined more dramatically.

In 2015, I concluded that interest rates would not continue to decline as they had for the prior 15 years and I decided to reduce to 50% from 100% the maximum allowed target weight to the US 10-year bond ETFs in the sleeve. Thus, the real-time historical figures are higher than what we could expect if we encountered the same environment going forward. The impact of this overstatement is about 1%, annualized, over the 20-year period for Diamond (sg131).

I review the maximum allocations to the ETFs linked to the US 10-year bond index on a regular basis and may make future changes to the design of the model portfolios based on how much interest rate risk we should take in the model portfolios. Thus, the maximum allocation could be changed again in the future.

The Effects of These Updates

If we consider these two updates, real-time performance understates the returns we could expect from the current design of Diamond (sg131) going forward by about 2%, annualized. Assuming we encounter a period like the last 20 years in the future, we will get about 3% more per year from the D5 signal set and about 1% less from our investments in the US 10-year bond ETFs. These figures will vary by the aggressiveness of the model portfolio, and, of course, the next 20 years may not be similar to the last 20+ years. On this last point, recall that the D5 signal set has provided strong returns over the last 100+ years and I believe our processes will be relevant well into the future.

Relative to the total investment process, the changes described above are small and don't change the nature of the investment approach.   

Returning to the Original Objective of Part I

In the early years of Focused 15 Investing, I manually adjusted the returns in Part I to promote more realistic comparability so users could develop realistic expectations for the different portfolios.  However, the manual adjustments became unworkable.  The recent modification brings the first two pages back to their original purpose, and does so in a more robust way.    


I evaluate many aspects of the investment process on a regular basis and will change the algorithms and elements of the process as needed to keep the entire approach well-suited to the markets and how people use Focused 15 Investing.  The key elements I review are checking for data errors from Bloomberg, the maximum allocation we should have for US 10-years bonds and the key drivers of Onyx as interest rates change, the success of each of the five members of the D5 signal set and determining when they should be retired and replaced by their backups. These are likely to be the areas where updates will occur. Potential changes are evaluated over many months or years before being made.  I have a great deal of confidence that the approach will be strong for many years.

In addition, I am always interested in how subscribers are using and would like to use the model portfolios.  I will adapt the process new needed and circumstances - like the use of the add-in sleeves in 2020. I continue to have a great deal of confidence in the approach.