Colorful Horse and Wagon outside Store in Three Rivers, New Mexico.
This post addresses comments made and questions asked by an ITA Platinum member related to a recent post that I wrote in review of the Rutherford Portfolio (https://itawealth.com/rutherford-portfolio-review-tranche-2-15-june-2018/).
In addition to specifically answering the questions raised, the post may also offer readers more insight into the value of information contained in the Kipling Workbook/Spreadsheet (SS). The post is likely to appeal more to ITA members interested in a deeper understanding of the SS than those just interested in using it as a simple mechanical tool.
Whilst the member’s comments are generally valid and deserving of this detailed response I want to put everything into perspective. First of all, the original post was a review of the Rutherford Portfolio – that is a relatively simple “real” live portfolio and is based on the selection of assets from a small (fixed/static) list of 10 ETFs (the investment “quiver”) that offer global diversification over all major asset classes. As such, each ETF is (generally) very broadly based in its representation of the general asset class and correlations between asset classes are minimized. This usually means that the portfolio will experience less volatility/movement than portfolios that select from a quiver that might contain different ETFs (e.g. specific market sectors) and/or might not be quite as diversified.
The first comment/observation:
“using the Kipling spreadsheet but using prices back to only February 12 (a low point of the recent correction) you would have been in securities such as IJT, IJS, IJR (all small-cap); FTEC, IGV (Tech); IHI (health – medical appliances); FDIS (consumer discretionary); and GNR (commodities). These are examples of ETFs that have been doing very well over the last month or two”
… raises two issues:
- Selection of assets to be included in the investment “quiver”. Since the suggested assets are not included in the Rutherford “quiver” this is a mute point in terms of it’s relevancy to the post – but it is a relevant question as related to portfolio construction. However, this should not be viewed as a reflection of the investment “system” – a good system should be acceptably “robust” when used with all quivers and under a wide range of market conditions. If it is not, then yes, there may be deficiencies in the quiver. This is a whole separate/different subject.
- The comment infers a question related to appropriate look-back periods – and this is a “system” issue. More on this below.
“I am fairly convinced that using a model such as the Kipling linear regression approach for ranking purposes after a major correction that does not shoot back up to the trend that had been occurring prior to the correction will be very slow in catching up to any new trends that might begin after the correction occurs. I’ve been trying to think of an example that illustrates what I am trying to say and might have finally come up with one. I think of a data set with two regression lines created from it, both with equal slopes (trends). However, one is developed using data prior to February 12, and the other is developed using data after February 12. There are two differences between the two trend lines, one attributable to modeling in general and one attributable to changes in the market leaders: the latter regression line has a lower y-axis intercept and ETFs supporting the newer trend after February 12 are different than the ETFs supporting the trend before February 12.”
This is an expansion of point 2 identified above – i.e. it is a “system” consideration.
I will try address both issues with some examples that use the “Kipling” ranking spreadsheet. However, I first need to ask how we should/can decide that a “correction” is complete (and “major”) and not the start of a major trend reversal (bull to bear market)? Specifically, we experienced a ~10% pullback in February but have not yet “bounced” enough to take out the January highs – so we could still be in a secular bull market – or we could be at the start of a major secular bear market (that would take out the February lows) – how do we define these trends in relation to our own personal investment horizons? How long does it take for us to decide that the February low was a major pivot low – and that we should be looking for a continuation of the prior bullish trend? These are not easy questions to answer without the benefits of hindsight so, ideally, our system should be oblivious to these definitions/decisions.
While there is some “cheating” in that the member’s suggested assets/ETFs benefit from “hindsight” (i.e. we now know that these assets have performed well over the past few months), let’s assume that the suggested ETFs were in our investment “quiver” in mid-February. I have added the suggested assets to the Rutherford list – so let’s take a look at how the rankings would have looked on February 16:
This spreadsheet is slightly different from the spreadsheets that members will be used to seeing but there is no fundamental difference in ranking/suggested selection methodology – just a few cosmetic changes that might add value to the evaluation toolbox.
Let’s look first at the left-hand side of the spreadsheet and focus on momentum rankings using default look-back periods (60- and 100-days):
This reflects our “original” ranking methodology in which we would only consider the inclusion of assets ranked higher than SHY – i.e. with positive (weighted) absolute momentum. In the above example we have 13 ETFs (Rank Column) that would pass this test. Perhaps not unexpectedly (since we have benefitted from hindsight in adding the more specialized ETFs outside the Rutherford quiver) most of these new ETFs pass this test although the (highly correlated) US small-cap sector ETFs are all borderline. At that time, Emerging Markets (VWO) were looking stronger.
However, FDIS (Consumer Discretionary) was top ranked (and a valid buy based on momentum rank) – so let’s take a look at the chart:
I have chosen FDIS because it is ranked #1 – but most charts look very similar following the February pullback, particularly the “new” ETFs at the bottom of the list. Would you have the confidence to buy at this point after that 10% drop?
FDIS saw a ~$4 drop (~$43 to ~$39 – ~10%) from 26 Jan to 8 Feb – 9 trading days – so no “investor” holding the ETF would be likely to avoid this draw-down without having a stop loss order in place – unless, of course, they might be using shorter look-back periods. Let’s assume an investor were using 22- and 65-days as the look-backs (some ITA members have suggested these values):
… now we see that all but GLD (Gold) are ranked lower than SHY – so FDIS would not be purchased (or would be sold if currently held) based solely on rankings. As we now know, with hindsight, selling this asset would not have been a good exit – and there is no entry signal.
Before moving on to other signals on the right-hand side of the spreadsheet I will point out some new features that I am working on that appear in the above screenshots. First of all, you will note that I have set the number of days between offsets to 5 (1 week) that leads to a picture of changes in rank over a 55 day period (a little less than 3 months). I have also added a new column (Trend) that provides a measure of trend over this period – so a positive number suggests that the asset is moving higher in the rankings with a negative trend indicating a fall in ranking. The number is just another linear regression measure of the slope/strength of the rank trend. So, GLD is the big mover in the above example with a trend slope of 0.27 and a change in rank from a low of 19 (lowest possible rank) to the current #1 position.
The color formatting of rank numbers is also changed so that dark green will always be #1 ranking and dark red will be the lowest rank (#19 in the above examples).
Let’s now move to the right-hand side of the spreadsheet:
….here’s what we see using the default (60- and 100-day) settings for ranking. The relevance to the member’s comments on slope and intercept are reflected in columns LRS1 and LRS 2. For example, the strength of the short-term trend (slope) for IHI (0.12%) is the same as the strength of the long-term trend (slope) for IGV – however, IGV is ranked higher than IHI based on it’s stronger acceleration from long-term trend to short-term trend (0.12% to 0.15% for IGV vs 0.11% to 0.12% for IHI) – this is all reflected in the ranking calculations (and influenced by the chosen weightings).
For the sake of completeness, here’s what the LRS1 and LRS2 numbers look like if we change the look-back values to 22- and 65-days:
…. Here we clearly see the negative short-term trends (negative values for LRS1) whilst some longer-term trends are still positive.
You will note that Projection (P) and Convolution (C) values are identical irrespective of the look-back periods used to calculate rankings. These are totally different (independent) calculations – although look-back periods for P-C calculations are also user selectable parameters. Note also that FDIS is now a recommended buy based on its positive P-C value.
One of the objectives of introducing the P-C calculations is to try to identify early entry and exit signals irrespective of ranking/absolute momentum. For example, using default settings, IJS would not be considered for purchase based on its low momentum ranking (negative relative to SHY) – however. It might be considered based on it’s positive P-C value.
As a back-up to factoring-in short-term performance we can also look at the H-A signals on the extreme right of the sheet. Here we see that GLD and IGV are looking strong in the short term (say 2-week period) – and this information is irrespective of any look-back periods chosen for ranking or P-C calculations.
Now, let’s follow the spreadsheet (using default values) through a monthly update (nearest Friday to 15th of month) to the present time. March ….
Assuming we were using a 4 asset maximum holding we note that VWO has now dropped out of the top 4 rankings and recommendations are focused on the “new” ETFs at the bottom of the list – with small-cap equities still out of favor despite positive H-A signals.
Let’s move on to April …..
…. Where we see a move to Commodities and Gold with IJT (small-cap growth) coming into the picture.
Moving on to May …..
… not a big change, but stronger short-term signals in the H-A candles – so, on to June ….
… where small-caps are showing strongly. VTI (that includes small-caps) is also moving up the rankings (positive trend) but not as strongly due to it’s wider exposure to (large- and mid-cap) US Equities.
A similar exercise can be run using the shorter-term look-backs (22- and 65-day) but I will not show the details here since the post will get too long. However, I can show the return performance of portfolios using the different look-backs:
The portfolio JimH_2 (blue line) shows the performance of the above portfolio (using default 60- and 100-day lookbacks) with monthly review/adjustments as described above.
The portfolio JimH_3 shows the performance of the same portfolio but using 22- and 65-day lookbacks. There is very little difference in performance (at least over this short time period).
Obviously, if the portfolios were reviewed on a different schedule (say weekly) the performance would likely be different.
So, is there an advantage to using shorter look-back periods? I think the answer to this question lies in how frequently an “investor” will check the rankings. By definition I assume that an “investor” will check his/her portfolio – and make adjustments – only on a monthly basis. Under this scenario I do not see an advantage of moving to shorter look-backs and this is based on the evidence of 10 year back-tests. Review date “luck” is also likely to play a bigger role than we already encounter and whipsaw trades are likely to be more prevalent (with added costs). If an investor prefers to check/adjust more frequently (becoming more a ”trader”) then there may be some merit to the change – but this would have to be analyzed with consideration of added trading costs and tax implications. This is why I offer the option of “tranching” as a compromise system.
Finally, the performance of the standard Rutherford portfolio (without the “new” ETFs) can be seen by looking at the green line in the above chart. The obvious differences lie in the much lower volatility and the lower returns (over this specific time period) – although there was a period (mid-March – end April) when the Rutherford out-performed the JimH portfolio.
In summary, there are 2 major considerations that an “investor” must consider:
- The list of assets from which the portfolio will be constructed and
- The system (and associated parameters) that will be used to manage the portfolio.
…. Both are important considerations – and we need to make choices based on performance over a relevant time-frame (and this should not be too long or too short) – depending on our investment objectives and tolerance for risk. The Kipling spreadsheet is designed with the flexibility to allow investors to modify the tool to match their own personal preferences with default parameters that offer a “robust” system (based on back-test performance) for an investor reviewing their portfolios on a monthly basis. The default parameters are intended to provide a starting point for more “adventurous” investors – but we obviously don’t have longer term back-tests for all combinations of different portfolios and different parameter settings.