Lowell Herr
Follow this author
Daily digest on all my authors
Send Message

x

Author Following Options:

See their articles on your Seeking Alpha homepage and in your feed.

Real-time alerts on this author

Get their new articles emailed to you as they are published.

Get all your favorite authors’ new articles emailed to you in one daily digest.

Lowell Herr

Stop FollowingLowell Herr

View as an RSS Feed

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

We have tried using more funds and the performance actually drops off. The best guess is that as one adds more funds we don't add all that much in the way of low correlations. Other than GLD all the funds we use are commission free with TDAmeritrade. One can easily find equivalent commission free ETFs from other discount brokers.

For larger portfolios I tend to invest in more than the top two, but again, performance falls off.

With the Kipling Tranche spreadsheet, manual tranching is very easy. There are a few assumptions, but not many. We find that tranche investing increases the number of trades per year. That is not a big issue for commission free ETFs. Taxes are always another factor.

We find the 12-month look-back period is too long and does not give best results. However, it may be the better choice for taxable accounts due to our current tax laws.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

Monte Carlo (MC) analysis does not predict the future. No model does. MC's usefulness is in providing probabilities of what is likely to happen, not what will happen. Those of us involved in back-testing, and I happen to be quite skeptical of most tests, need to be aware of the "Schwert rule." The "Schwert rule" goes like this. "After studying several claimed anomalies that had been observed in historical data, Schwert concluded - After they are documented and analyzed in the academic literature, anomalies often seem to disappear, reverse or attenuate."

Momentum seems to be bucking the "disappear, reverse or attenuate" trend, but it too could pass.

Instead of recommending, I'm explaining a momentum model I am testing with several portfolio. I'm about to prepare my weekly blog post as to how well the portfolios are performing. To be honest, so far I would have been better off to put 100% in VTSMX and go on a long cruise. Another fact is that we have experienced a great bull market in U.S. Equities since March of 2009.

I have not been using the above momentum model sufficiently long to come to a conclusion. Right now, most, but not all portfolios using the momentum model are gaining ground on the VTSMX, a hopeful sign.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

For our "cutoff" or "circuit breaker" security for the out-of-sample back-test we used VFISX as it is a low volatility treasury. Going forward I use SHY as the cutoff ETF.

It sounds like we are testing similar variables.

Going forward I am measuring the performance trend of the portfolios against VTSMX and a benchmark that is customized for each portfolio. This is known as the ITA Index. Most benchmarks are not appropriate for the portfolios under examination. The S&P 500 is most frequently misused as a benchmark. I realize that VTSMX falls into the inappropriate class of benchmarks.

Lowell

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

There is no Holy Grail to investing. If you look at the heat maps in this reference you will see where we look at probabilities for "best" look-back periods and weighting for volatility. One always needs to remember that this is what worked well in the past for this group of ETFs.

http://bit.ly/1Nz3KT6

This model was tested running a Monte Carlo analysis with out-of-sample data and I am testing the model going forward with several portfolios. I'm skeptical of single back-tested runs. Too much luck is involved in the metrics selected for the various decisions.

There is also something else to consider and that is the ratio of winning/losing trades. It was reported to me yesterday, that this model has about a 60% chance of generating a winning trade. One would like that to be higher. Lots of testing remains. Nevertheless, using SHY as the "circuit breaker" ETF keeps one out of major bear markets, and that is definitely a positive feature of this model.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

TMD,

The above brief description sounds very similar to one of the Momentum Models. If TLT is the treasury, I am skeptical as TLT has had an unusually good run. We used TLT in some of our back-tests and it skewed results to the upside. Bonds have had an excellent run since 1980 - something I don't expect to see over the next ten year due to the expectation of rising interest rates.

You can see the low correlated ETFs I am using from the correlation diagram in the above article.

Lowell

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

This blog post provides a look into how Antonacci's Dual Momentum model works.

http://bit.ly/1SwFyRo

Lowell

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

Were these single runs or multiple runs with "noise" introduced to simulate real world investing?

I'm interested in "robust" models vs. optimized models. In other words, will the model provide an edge during most market conditions.

"I know Antonacci's work to some degree at least. If I recall correctly, he advocates a 10-month SMA with SPX. That's not bad, but there are other moving averages that work better."

Antonacci's "Dual Momentum" model is a little more complex than using a SMA with SPX. I think you might find it of interest.

Lowell

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

"With a CAGR of 11.1%, the MAR works out to 0.58. That's not very good at all, in my book. MAR values for a momentum strategy ought to be well over 1.0. Generally we shoot for somewhere between 1.5 and 2.0, and we consider anything over 2.0 to be outstanding (especially if CAGR is in the vicinity of 10% or greater)."

If 1.5 or 2.0 is the MAR standard, then this model fails. I looked up data Gary Antonacci presents in his Dual Momentum book and I did not spot any MAR values as high as 1.5. I did find long running data where MAR = 0.77. This includes the great bull market from 1982 - 2000, something our data does not do.

Are you able to generate MAR values >1.5? If the historical market growth is around 9% and it is not unusual to see a 15% market variation in any given year, it would take an astute investor to turn in a return consistently 1.5 times better than a high probability draw-down. Maybe I am missing something. My train needs to move to a different track. ;)

Lowell

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

Still not quite right. When you are looking at the graphs that show a red line or graph, that is the performance of the benchmark. Instead of running just one run for a portfolio, multiple runs are generate to better simulate the actual behavior of an investor as it would be rare for one to be so disciplined to invest exactly on a specific date and right down to the precise number of odd shares. This is where back-tests fall short or at least convey information that is misleading.

To get around some of these problems, "noise" is introduced into the back-test. Back-tests are run on either side of the trade date to introduce variations of prices an actual investor might obtain. This still does not get at the price variations that occur on the day of a trade.

The 19.1% is the average draw-down from all the runs and as an investor, one would have the highest probability of seeing this draw-down, although it could be smaller or larger.

While I am pontificating on the inherent errors tied to back-tests, here is another one. When back-tests are run, end-of-day prices are used. However, if we deal with ETFs or stocks instead of mutual funds, errors are introduced into the back-tests as investors do not buy and sell using end-of-day prices. We buy and sell during the day and those prices will vary compared to the end-of-day prices. By introducing the "noise" mentioned earlier, some of the price variation is better understood and measured by all the gray graphs shown on my blog.

In so many investments books I see back-test projections showing only one set of data. This should raise red flags to the reader as "real life" investing will vary widely from these printed results. Once more, those gray line graphs you see on my blog site point out the wide variations one might expect from following the model and the longer the run takes place the wide the fan of the end results. Hope I am making myself somewhat clear.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

You and Cliff focused on the Maximum Draw-Down. In other words, you concentrate on the worst case DD from 50 runs. Were you aware of this?

The average DD was 19.1%, still high in my book. My preference is to limit the DD to something in the order of 12% to 15% max. The DD for VTSMX was 49.2%, so the portfolio compared favorably with respect to the benchmark.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

Those figures are close enough. That maximum DD is of concern and something we are working on to reduce. In the spreadsheet I am using, there is an additional worksheet that provides some guidance as to how to know what risk is involved in the portfolio as constructed. While this will not prevent draw-downs, at least it makes the investor aware of what risks are involved.

Note that the results in the first "Robust" link I provide are for index funds, not the ETFs I used in the above example. Just want to make that clear.

The index fund portfolio generated a total return of 860% (rounded) to 574% (rounded) for VTSMX over the 20-year period.

As a general rule, tranching reduces both return and risk. If you look at the graphs in the links I provided, you see the shaded or light gray graphs vary widely due to introducing "noise" from slightly different starting dates. Just yesterday I finished reading an investment book that is getting some attention. The authors presented only one set of data touted excellent returns. They may be accurate or the good returns are only due to the luck of the starting and ending dates. The published results were "accurate" to the hundredths place. Now that is a joke.

Lowell

http://itawealth.com

## Robust Portfolio Construction Using Momentum And Correlation Analysis [View article]

Did you click on the link at the very end of the article. You will find out-of-sample data going back to 1995. Here it is again.

http://bit.ly/1KyVFY1

Also there is a lot more information on my blog about results. Here is another link.

http://bit.ly/1Pdqsll

Lowell

http://itawealth.com

## Protect Profits By Implementing A Risk Reduction Strategy [View article]

http://tinyurl.com/p7p...

Lowell

http://itawealth.com

## Tranche Model Applied To The 'Swensen Six' Portfolio [View article]

Correct - the Tranching Model I am using is not set up as multiple portfolios. The Tranche Model I am following is really a combination of absolute momentum, relative momentum, and portfolio risk all wrapped into one package where we review a portfolio, but do not rebalance strictly based on the recommendations that show up on the review day. We look back to see what the recommendations were several days back - 2, 4, 4, 8, 10, 12, 14, and 16 in my example. Those recommendations are averaged and show up in the Required column.

You could easily follow the "traditional tranche" model if the family had multiple portfolios as we do. For example, my wife and I each have a taxable and tax deferred accounts giving us four different portfolios to review at different times.

Lowell

http://itawealth.com

Lowell

## Tranche Model Applied To The 'Swensen Six' Portfolio [View article]

This article on my blog may provide some answers.

http://bit.ly/1KyVFY1

There are several other articles showing tests on different models. A search for "robust" is a place to start. I am also testing momentum-tranche models going forward. It is still too early to come to any conclusions.

Lowell

http://itawealth.com