John Dishman sent the following to me, having tried to post the table in the Forum section. I was not able to do that easily so I am posting it here as a blog entry. If you have questions for John, state your questions in the comment section associated with this blog entry.
“I found this to be an interesting result using the Tranche 1.6 spreadsheet. I have a portfolio of about 10 ETFs I’m tracking. Using a date of last Friday I decided to look at how the offset interval (period between offsets) impacted the results. I used 8 offset portfolios and selected 4 ETFs to be included at any one offset. I then looked at 3 different offset intervals: 5, 6 & 7 days. The results are shown in the table below. What surprised me was how sensitive the results were to the offset interval. For example, ETF1 that had the greatest percentage for 6 days, dropped to 3rd place for 5 days & 7 days. This seems to show that tranching–which is used to reduce timing luck–has an element of luck itself, namely, the offset interval used to do the tranching. This may be reflective of the choppy market we’re in where no momentum trend dominates.”
PS: the columns below don’t add to 100% since I don’t show the full list that the spreadsheet includes. I just shown a subset to illustrate how these 6 vary with offset interval.
John/Lowell,
What you have illustrated in fine form is exactly the nature of timing luck itself. It is, in fact, a better illustration than the post I made a couple of months ago showing how timing luck arises. The tranche method is not immune to timing luck and what it attempts to do is to smooth out the effects. The current sideways market is no doubt susceptible to timing luck because the indices are probably toggling back and forth as the look back values settle into a steady state.
I suspect the momentum engine we are using goes a bit unstable in sideways markets, but I have no data to back up that hunch.
HedgeHunter and I have had some pretty large back tests running on tranche for the last week or so but the results are not ready for publication yet. I would just say, beware of drawing any firm conclusions about the likely long term results using the tranche method based on short term tests. One thing that is already apparent is the success of using the tranche method to modulate the effects of timing luck is very much dependent on the nature of the market. As we have pointed out before, the current market is not typical.
As soon as we have something to report, we’ll do so.
Ernie
Ernie, John, et al.,
We know there is timing luck based on when the review date arrives and the above table shows this. Given the daily timing luck, we are also faced with “timing luck” due to volatility within the daily price gyrations. Buy and hold investors are not bothered by price fluctuation as they hang in regardless of market changes. However, this can be difficult to stomach when major bear markets hit.
My recommendation is to be patient and to stick with the portfolio management model that fits your comfort level. There are a variety of models available on this blog. Read the reviews as they come up and find a model that meets your retirement needs.
Tomorrow the Kepler is up for review and I am using the Kipling 8.0 with this portfolio. With the Curie, to be reviewed on Wednesday, I am using the Tranche 1.6 spreadsheet. Friday I will review the Pasteur and it is passively “managed.” So you will see three examples this week.
Lowell
Lowell,
I frequently check the Kipling 8.0 each evening after the closing prices are posted, just for curiosity, not for making adjustments. There were some days when the ranking shifted significantly, then in the next day or two went back to the prior trend. When the option to add smoothing to the closing prices was offered, I started checking 1 day vs 4 days. The 4 day smoothing avoided the whipsaw in ranking that occurred with 1-day prices. I would like to see the closing price smoothing offered in the Tranche 1.6 spreadsheet. I have a hunch that it would smooth out much of what John Dishman found in his small study.
Another factor in play with having a fixed number of ETFs but different number of days in the offset interval is that the time period being assessed was different for each of the offsets. 8 ETF x 5 days = 40 day performance period. 8 x 6 = 48 days. 8 x 7 = 56 days. In the fluctuating market we’ve seen, that could generate different allocations because the periods are different.
Just a few thoughts.
…Lee
Correction for above comment. I referred to portfolio offsets as ETFs. My last paragraph should have read:
Another factor in play with having a fixed number of portfolio offsets but different number of days in the offset interval is that the time period being assessed was different for each of the offsets. 8 x 5 days = 40 day performance period. 8 x 6 = 48 days. 8 x 7 = 56 days. In the fluctuating market we’ve seen, that could generate different allocations because the periods are different.
Lee,
I’ll check with HedgeHunter as he programmed the Tranche 1.6 spreadsheet.
Depending on the back-testing results, it may not be worth the programming time. The Tranche model is still in the testing mode. As I recall, we came to the conclusion that EMA smoothing does not make a significant difference in returns over the long run. I don’t recall if it reduces trading, but if it does, then it has merit.
Lowell
In response to the above comments my response would be that readers should wait for any definitive response and guidance on how the Tranche SS might be used for maximum effect. As Ernie notes above we are presently analyzing numerous tests using the Tranche methodology – but this takes many hours of testing and analysis and we don’t want to fall into the trap of (over) optimization without robustness testing over different portfolios and/or time periods of different market behavior.
However, here are some quick responses to some of the comments/suggestions:
1. Although tranching does not completely remove the element of luck in selecting the check date/review period, it does reduce the deviation in total returns/performance;
2. This reduction in deviation is achieved at the expense of total returns (surprise, surprise!) however, for investors not comfortable with the “luck” factor it might be a more acceptable approach to asset allocation;
3. Lee makes an important observation i.e. that the different combinations of number of offset portfolios/tranches and separation between these offsets results in different “lookback” periods – Ernie & I will be addressing this subject when we are comfortable publishing our results;
4. EMA smoothing and using the tranche methodology are, in some ways, similar in that they both attempt to smooth volatility fluctuations. They both also introduce an element of “lag” into the equation. The impact of both approaches is, in itself, quite small and adding EMA smoothing to tranching is probably getting too subtle for the influence this might have on the end results. It is also a significant amount of work to add into the tranching sheet and I do not plan to add this feature unless something comes up to suggest that it is really worthwhile.
I hope this answers most of the immediate questions. We would ask readers to be patient in waiting for more analysis using the tranching system – we are running computers overnight and spending many hours analyzing outputs. This is not a simple exercise that can be accomplished quickly. We don’t want to publish results that may be confusing and/or misleading – at least without highlighting strong caveats that results may not be typical of anticipated trends.
David