Post New

A Tactical Asset Allocation ETF screen Generalized Protective Asset Allocation Fuzzy a modified version of Keuning and Keller's Generalized Protective Momentum. An analysis of the GPM results for each period shows that a nonlinear what I call fuzzy decision threshold makes a better separator between safe and risk periods. This was developed with GTR1 help from robbie in 2020.

zi>0  K&K'sSafe  FuzSafe
  12      0.000    0.000
  11      0.167    0.000
  10      0.333    0.000
   9      0.500    0.000
   8      0.667    0.167
   7      0.833    0.333
   6      1.000    0.500
   5      1.000    0.667
   4      1.000    0.833
   3      1.000    1.000
   2      1.000    1.000
   1      1.000    1.000
   0      1.000    1.000

Compared to Vanguard’s VWELX balanced fund from 10/2007

	GPMF	VWELX
CAGR:	10.1	7.5
TR:	393.9	231.0
Log2TR:	2.3	1.7
SAWR(	8.2	6.0
GSD(42)	9.1	12.2
DIGSD	8.7	15.0
LDD	3.9	8.3
LDDD3	4.4	8.7
MDD	-15.1	-36.0
UI(42)	2.5	7.2
Sharpe	1.0	0.6
Beta	0.1	0.7
TI	121.4	10.4
AT:	3.0	0.0

Year       GPMF    VWELX
20071231	8.1	-1.6
20081231	24.4	-22.3
20091231	9.4	22.2
20101231	10.2	10.9
20111230	9.5	3.9
20121231	0.4	12.6
20131231	17.3	19.7
20141231	5.6	9.8
20151231	-6.2	0.0
20161230	2.7	11.0
20171229	14.2	14.7
20181231	1.9	-3.4
20191231	14.5	22.5
20201231	21.7	10.6
20211231	22.5	19.0
20221230	2.1	-14.4
20231229	3.9	14.3
20240520	10.3	7.0

http://gtr1.net/2013/?!!QlpoMTFBWSZTWXkX7oAAACQfgA...

I’m currently using a blend of 3 non correlated TAA strategies based on momentum and risk premium.

RAMc

Corrected Link to GTR1 GPM fuzzy

http://gtr1.net/2013/?!!QlpoMTFBWSZTWYDDoxcAA8efgA...

Is there an explanation of how this works anywhere?

Is there an explanation of how this works anywhere?

TrendXplorer originally posted on seekingalph.com in 2016

Generalized Protective Momentum | Seeking Alpha

I ran my own simulations in Python and analyzing his the decision to switch between risk and safe equities would be much more effective using a nonlinear curve similar to fuzzy logic’s method.
I attempted to run it in GTR1 but needed help from Robbie to accomplish the current version.
You might like Keller an JW Keuning’s newest TAA strategy Hybrid Asset Allocation a little better.
Dual and Canary Momentum with Rising Yields/Inflation: Hybrid Asset Allocation (HAA) by Wouter J. Keller, Jan Willem Keuning :: SSRN

Independently verified here

Hybrid Asset Allocation - Allocate Smartly

And just for good measure a TAA strategy based on Risk Premium Value
Testing a Risk Premium Value Strategy - Allocate Smartly

Hybrid and Risk Premium are individually nice historical performers but they are based on two completely different principals with less than 40% correlation in their returns. A 50/50 portfolio of each of these gives a nice mix. Allocate smartly has a feature where you can test a blend of strategy’s.

I like the concept of not putting all my eggs in one basket.

I'm having trouble posting links! Here is another try.
https://seekingalpha.com/article/3985525-generaliz...
https://papers.ssrn.com/sol3/papers.cfm?abstract_i...
https://allocatesmartly.com/hybrid-asset-allocatio...
https://allocatesmartly.com/testing-a-risk-premium...

Can you summarize what
* zi is
* what each tier means (is it individual stocks passing/having a zi score, or an asset class', or what?

12 0.000 0.000
11 0.167 0.000
10 0.333 0.000
9 0.500 0.000
etc

From K & K’s paper
“we use a simplified formula for GPM: only return and correlation information are combined for measuring absolute and relative momentum. By mitigating the return information with a correlation hedge, risk-adjusted performance can be improved in two separate ways: a boost of portfolio return is possible without compromising left tail risk (drawdowns) or, alternatively, risk can be reduced while preserving portfolio return. For these two variations we combine return and correlation into a generalized momentum measure as follows: . . .
GPMxM: the correlation multiplied return metric ri * ( 1 - ci )”

“Each month 3 out of 12 assets with the highest correlation hedged return readings are eligible for capital allocation next to the safety asset's allocation. For crash protection we will use the same "high protection" level as introduced in our PAA-contribution. With the protection level set to "high (2)" the capital fraction of the safety assets equals twice the fraction of assets with non-positive momentum in the risky universe. The backtests in the original paper cover the 45+ year period December 1970 until May 2016.”

The GTR1 GPM fuzzy doesn’t use the either the same momentum or correlation equations but the way they are put together follows the same concept.

Correlation: tscorr(TR21,RiskyEWTR21,0,231,1,step0)
Momentum: linear(0.5,rrs(1,84),1,rrs(1,210))
ZiScore: product(Momentum,minus(1,Correlation))

Allocate Smartly’s write up on GPM states it a little more clearly.
https://allocatesmartly.com/keuning-kellers-genera...

In trying to understand and validate the underlying data and calculations of this PAA version, I see something that confuses me. One possible conclusion is that the "RiskyEWTR21" measure is off by one day, which causes the 231-day correlation measure to be incorrect. A workaround I used to get matching results changes the current order of the 3rd and 4th top stocks now, which would change what one would invest in now. Not sure if the workaround is the best way to do this though.

The PAA link given in this thread by RAMc recently: https://gtr1.net/2013/?!!QlpoMTFBWSZTWYDDoxcAA8efg...

The "workaround" I'm trying: https://gtr1.net/2013/?s20071015i42f0.15000::dspo%...

The workaround is to change the "trp(1,21)" in the imported screen to "trp(0,21)". This doesn't necessarily make sense to me, but it matches what I can calculate from the underlying data.

If you "Run Screener" for the original link, with data ending on 20240522, you can see that for all stocks, the "RiskyEWTR21" measure is 4.684806. For the top stock, GLD, the correlation is 0.257699.

If you do the same for workaround link, you see "RiskyEWTR21" = 3.286588 and the GLD correlation is 0.2737858.

I downloaded daily data for the set of stocks used and calculated the TR1 (and other) measures. I took the average of TR1 each day (for the 12 "risky" stocks). Taking the product of the most recent 21 average TR1 returns and subtracting 1 gives the "RiskyEWTR21" for 240522.

Using this direct calculation approach shows that using the 240522 end date gives a "RiskyEWTR21" measure of 3.286588, which matches the workaround link. My calculated correlation for GLD is of 0.2737858, which matches the correlation from the workaround link (and these correlations matched for all the other stocks). All the other measures for all stocks are the same for the original link and the workaround link, and match what I calculate separately from the underlying daily data.

Can anyone validate what I'm seeing? If so, is there a more appropriate workaround? I'm always confused with the lags to use, and the field and price adjustments.

Post New

Mechanical Investing FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.