Dual Momentum on equities could whipsaw soon, opinions sought

[MachineGhost]

4. You can have a time sequenced tranched portfolio...in other words, break your portfolio up and trade signals exactly as they come for that portion of the portfolio based on your trade dates. Thus, you might break your port into 3/4 tranches and look at the signals once a week, month or quarterly in rotational sequence.

If you want to get the closest possible to whatever Mr. Market may give us in the future and minimize random variation results, go with option 4. The downside is it drives up portfolio costs due to more frequent trading.

Personal recommendation...go with #4 unless it is a taxable account. If psychologically you like doing something then divide your port into quarters and review 1/4 of your port each week. If you want to reduce costs then go with thirds and once a month for 1/3 of your port...the latter would totally take you away from your current worries.

Are you talking about dollar cost averging in on the same signal but over a period of time? That's pretty much what I'm doing with the PP to avoid sequence of returns risk.

[Kbg]

MG,

Not sure what exactly what you mean, but no it isn't averaging in on a signal as I would use the term. It is essentially setting up x number of pseudo portfolios with time separated signals and then trading them individually. By time separate an example would be dividing your portfolio into 4 pseudo ports and checking for signals each week. If your signal check day was Friday and it indicated a change the change would only apply to 1/4 of your port/that week's pseudo port. Let's say the next week there was a whipsaw back then for the second pseudo port you would do nothing. This assumes port 1 and port 2 were the same 3 weeks ago of course. This technique doesn't improve performance it simply (greatly) reduces random performance outcomes.

For a visual...let's say you were running a Monte Carlo analysis that produces a "broomstick" chart. The broom threads would be much tighter and closer to the median result.

[MachineGhost]

MG,

Not sure what exactly what you mean, but no it isn't averaging in on a signal as I would use the term. It is essentially setting up x number of pseudo portfolios with time separated signals and then trading them individually. By time separate an example would be dividing your portfolio into 4 pseudo ports and checking for signals each week. If your signal check day was Friday and it indicated a change the change would only apply to 1/4 of your port/that week's pseudo port. Let's say the next week there was a whipsaw back then for the second pseudo port you would do nothing. This assumes port 1 and port 2 were the same 3 weeks ago of course. This technique doesn't improve performance it simply (greatly) reduces random performance outcomes.

For a visual...let's say you were running a Monte Carlo analysis that produces a "broomstick" chart. The broom threads would be much tighter and closer to the median result.

But when do you get fully invested if theres no flip flopping? You'll miss out on the gains!

[ochotona]

We've been through period of really historic volatility since late August 2015. The US equity market has been in a trading range since October 2014. TMOM examines the differences between prices. Would it be reasonable to expect that a differencing method would work during a period of severe price volatility? Answer - it's probably not the best environment for TMOM.

That's not a prediction. That's just reality. TMOM is the wrong algorithm to use at this time.

Also, Antonacci himself and Gray, Vogel, and Foulke all demonstrate that 1-year TMOM and 10-mo MA are really about the same, in the end. Except TMOM trades less on average. We're not in "average" times.

So I'm going to fail-over from primary TMOM to backup MA, until we go a year past the two historic Valleys of Worry... either to the upside or the downside. 10-mo MA instead of 210 day MA is also a good idea, thanks.

Now if we keep doing these severe S&P 1800-2100 Valleys many more times... that will be bad, yes. That would be teh suck.

I only want to trade once a month. I think about it as the end of month approaches, then I just want to pull the trigger (or not) and forget about it until the next month.

[MachineGhost]

I disagree that the period since last year is anything in terms of "really historic volatility". Look back a lot further in history and you'll see much, much worse times that makes today's volatility look like a walk in the park.

I admit I can't really visualize what you're so concerned about here unless it is the dropoff and wiggle effect from using a ROC. If you're not willing to stick to a simple method through thick and thin, then it's just not going to work. You can't avoid whipsaws unless you introduce more lag and give up more gains.

Another thing you can try is using a separate momentum measurement for exiting than for entry. They are rarely, if ever, optimal to be exactly the same because of the nature of greed and panic.

[Kbg]

MG,

Not sure what exactly what you mean, but no it isn't averaging in on a signal as I would use the term. It is essentially setting up x number of pseudo portfolios with time separated signals and then trading them individually. By time separate an example would be dividing your portfolio into 4 pseudo ports and checking for signals each week. If your signal check day was Friday and it indicated a change the change would only apply to 1/4 of your port/that week's pseudo port. Let's say the next week there was a whipsaw back then for the second pseudo port you would do nothing. This assumes port 1 and port 2 were the same 3 weeks ago of course. This technique doesn't improve performance it simply (greatly) reduces random performance outcomes.

For a visual...let's say you were running a Monte Carlo analysis that produces a "broomstick" chart. The broom threads would be much tighter and closer to the median result.

But when do you get fully invested if theres no flip flopping? You'll miss out on the gains!

If your signal is long for four weeks you are fully invested until it turns to sell/short. Again, you are going to basically get the returns of whatever trend system you are using. In fact, you should be closer to the mean returns over many start and trading dates. This method simply reduces the impact of signal timing variability.

[MachineGhost]

If your signal is long for four weeks you are fully invested until it turns to sell/short. Again, you are going to basically get the returns of whatever trend system you are using. In fact, you should be closer to the mean returns over many start and trading dates. This method simply reduces the impact of signal timing variability.

That sounds like what I originally asked, dollar cost avergining on the same signal?

[Kbg]

If your signal is long for four weeks you are fully invested until it turns to sell/short. Again, you are going to basically get the returns of whatever trend system you are using. In fact, you should be closer to the mean returns over many start and trading dates. This method simply reduces the impact of signal timing variability.

That sounds like what I originally asked, dollar cost avergining on the same signal?

Same signal, yes. However, you trade the sub portfolio according to the signal on that day. So let's say you were long in all 4 sub ports and on week one the signal said to close. Then you would close 1/4 of the port. If on week two the signal was long you would do nothing and 3/4 of the port would stay long. If on week three the signal flipped again then another 1/4 would be closed but not 3/4 of it(i.e. You wouldn't go back and close week 2)

[ochotona]

After #brexit, trying to figure out a false "risk-on" signal could be completely a moot point.

However,

I have chosen target price levels for my ETFs for the troublesome months ahead. Sept, Oct, Feb, Mar. I am just linearly interpolating Aug-Nov, and Jan-Apr. No emotion, no fear, no panic - I'm going to execute. It just seems pointless ("stoooopid") to execute on a buy signal caused by short-term -14% price fluctuation year-prior which then quickly reversed - twice! I'm just not even going to look at those scenarios. Why?

Worst case scenario if I blow the trade? I stay in bonds a few months longer than otherwise. Not a big deal. Really not a big deal.

[ochotona]

The AlphaArchitect guys have licensed Enhanced Global Equities Momentum (E-GEM) from Gary Antonacci. That's the proprietary version of the Dual Momentum allocation described in the book. I am moving my IRA over from another Gary-affiliated advisor, which is a little sad and disruptive. It's hard for me to sever human ties based purely on cold business considerations. AlphaArchitect offering it for 25 basis points less, which amounts to a big difference over years and years.

[MachineGhost]

The AlphaArchitect guys have licensed Enhanced Global Equities Momentum (E-GEM) from Gary Antonacci. That's the proprietary version of the Dual Momentum allocation described in the book. I am moving my IRA over from another Gary-affiliated advisor, which is a little sad and disruptive. It's hard for me to sever human ties based purely on cold business considerations. AlphaArchitect offering it for 25 basis points less, which amounts to a big difference over years and years.

Human ties is why you pay for the advisor's Porsche and McMansion!

What's going to be in the "enhanced"?

I'm not sure why you're so gung ho about GEM and AA though. They're not the best implementations of momentum. Nor should you put all your eggs into one basket. There's about a handful or two of ETF's now with downside risk management and some use momentum. You would be better off with strategy diversification because of the severe tracking error with momentum.

The state of the art is now Faber's Triple whateveritscalled. AA clearly doesn't understand the concept of proper portfolio diversification and GEM is curve fitted junk to uninvestable indexes IMO.

[Mr Vacuum]

How do you figure "uninvestable indexes"? It's the S&P 500 and AWCI ex US.

[MachineGhost]

How do you figure "uninvestable indexes"? It's the S&P 500 and AWCI ex US.

No one can invest in indexes, hence the're uninvestable. Which is why performance differs. No fund literally buys all of the stocks in an index; they use shortcuts. Small tracking errors compound enormously over time which biases positive results.

My point is you have to take it with several grains of salt any system that uses uninvestable indexes compared to actual investable funds. The former always looks better (especially in the far flung past when such indexes did not even have any matching investable funds) and to what extent it is misleading depends on how skilled you are about being cognizant of all this stuff.

[Mr Vacuum]

How do you figure "uninvestable indexes"? It's the S&P 500 and AWCI ex US.

No one can invest in indexes, hence the're uninvestable. Which is why performance differs. No fund literally buys all of the stocks in an index; they use shortcuts. Small tracking errors compound enormously over time which biases positive results.

My point is you have to take it with several grains of salt any system that uses uninvestable indexes compared to actual investable funds. The former always looks better (especially in the far flung past when such indexes did not even have any matching investable funds) and to what extent it is misleading depends on how skilled you are about being cognizant of all this stuff.

Ah, right, understood. Uninvestable indexes... like you said.

I suppose you have investable funds data for all the relevant asset classes while the rest of us stumble along blindly with index data. God help the fool running a PP backtest with whatever they use for gold prices and not accounting for 5% spreads on physical gold or 1.12 tracking error on ETFs.

[MachineGhost]

I suppose you have investable funds data for all the relevant asset classes while the rest of us stumble along blindly with index data. God help the fool running a PP backtest with whatever they use for gold prices and not accounting for 5% spreads on physical gold or 1.12 tracking error on ETFs.

Exactly! Real world performance will not match uninvestable indexes (or illegal products) from the good ol' days. And the way compounding works, small gains or losses occuring at the beginning of a backtest have a ridiculously outsized influence over the ending terminal wealth.