Post New

I have been looking at finding a screen that is a fairly steady screen but switches to something like QLD when a major bottom detector fires or platykurtic's 90/86.

Anyway using GTR1 with NAS100 as a universe and screening on low vol(1,253) I get CAGR 14%, SAWR of 10% and MaxDD -33% since 1985. All without timing. The CAGR is slightly less than ^N1T, but a much smoother ride. About 1% less CAGR than ^N1T with timing, but again with a much smoother ride,

https://gtr1.net/2013/?s19850201h1::nas100.a:et1:r...

Thoughts on the concept and screen would be appreciated.

Also, does anyone know how vol(1,253) is calculated in GTR1?

Aussi

vol:[<lag_days>|<param_ref_1>],[<num_of_periods>|<param_ref_2>],[<period_length>|<param_ref_3>]

As shown, the function vol takes three arguments, each of which can be either a number or a parameter reference. The value of lag_days must be a non-negative integer, the values of num_of_periods and period_length must be positive integers, and

lag_days + num_of_periods * period_length

must not exceed the retrospective limit. For each investment, the function vol calculates the annualized standard deviation of the natural logarithms of total returns (obtained by dividing the investment's appropriate closing g-prices) over the num_of_periods number of disjoint consecutive periods consisting of period_length number of market days, with the most recent period ending lag_days number of market days prior to the current trading date; the result is assigned to the calling field.

For example, vol:3,52,5 calculates each investment's volatility using its last 52 disjointly measured 5-day total returns, with the most recent 5-day total return measured through the close of the market date 3 days prior to the close of the current trading date.

Note that if lag_days is zero, then volatility is measured through the close of the current trading date, which corresponds to the same g-prices at which trades take place.

By default, a meaningful volatility measurement is always calculated for every investment on every market date, regardless of how recently the stock associated with an investment may have begun trading. This is possible because GTR1 Linearization defines each investment's daily g-prices indefinitely into the past using the history of parent companies and, when the latter does not exist, interpolation. However, to filter out investments with inadequate actual pricing history for a conventional volatility computation, use the field function dsio or dspo.

RAMC

I only used 2 parameters, vol(1,253), thinking that was volatility over the last year. You stated the requirement for 3 parameters. Do you know how to interpret what I used?

Aussi

RamMc

I tried with vol(1,253,1) instead of vol(1,253) and got the same result. I guess it defaults to "1" if not specified.

Aussi

I should have noted the above vol definition wasn't my words it was from Robbie Geary's GTR1 Glossary documentation from back in 2008.
It's still available at: https://mechinvesting.wdfiles.com/local--files/bac...

RAMc

Using the 8 best performing Vanguard country funds rebalanced every 6 months by 6 month momentum outperforms a world index over 10, 20, and 30 years; $100 invested in the LMMS compounded at its
annualized return would turn into $342 (T10), $1,171 (T20), and $4,007 (T30), which amount to
nearly 28%, 64%, and 110% more than the same $100 invested in the World benchmark. With a higher risk-adjusted return also.

https://papers.ssrn.com/sol3/papers.cfm?abstract_i...

Surely it must be Blackrock country funds (ETFs)?

Baltassar

I simulated LMMS on P123 using the following code:

TICKER("EWA,EWO,EWK,EWC,EDEN,EFNL,EWQ,EWG,EWH,EIRL,EIS,EWI,EWJ,EWN,ENZL,ENOR,EWS,EWP,EWD,
EWL,EWU,IVV")
FOrderOLD("100*((Close(0)-Close(130)) / Close(130))",#all,#desc,#previous,true)<=8

Backtested 6-month rebalance since 1999:
6.64% annualized return, S&P 8.47%
-62% Max DD, S&P -55%
0.46 Sortino, S&P 0.63

Taz

I reran your screen on Portfolio 123. With the same start date I got the same poor results you did.
But then I did a rolling weekly start date with 27 sample periods.

                Ret     Bench  Excess
Average          17.75%  10.73% 7.02%
Up Markets   22  18.95%  13.34% 5.60%
Down Markets  5  12.49%  -0.77% 13.26%

So it appears the paper's conclusion was valid for most start dates.

Disregard the results of my previous post. I made an error in the input parameters.
Rerunning the test with different start dates does make a difference but the overall results are still poor.

Yes I misspoke saying the article used Vanguard funds; Blackrock ETFs are now available for all but Portugal (he made a reference to Vanguard and I misread that as use of their funds). But the study
used
data "of the entire MSCI database of developed markets, from each country’s
inception in the database and through the end of 2024. The number of countries varies over time,
with a minimum of 17 and a maximum of 23. The benchmark against which the momentum
strategy considered here is evaluated is the MSCI World index, consisting of all the developed
countries in the sample, weighted by market cap. All returns are monthly, nominal, in dollars, and
include capital gains/losses and dividends. Exhibit A1 in the appendix reports some summary
statistics for all the countries in the sample and the World benchmark.
Of the 23 countries in the sample, 17 have their inception date in Dec/1969. With these
countries, the first portfolio is built at the end of Jun/1970, after observing their return
performance over the first six months of 1970. The eight countries with the highest return are
placed in the ‘Winners’ portfolio, and the remaining nine countries in the ‘Losers’ portfolio, with
equal weights in both cases. At the end of Dec/1970 the same process is repeated, based on the
return performance of each country in the second half of the year. The same process is then
repeated year after year through the end of 2024, with countries being added to the analysis as
they are incorporated into the database."

So he used data not ETFs since the latter did not exist, and he only used developed markets.

Nuzzo fast!

I read this paper with interest and thought to try a backtest. By
happenstance I was recently working on another project which would
make a backtest easily possible.
After a few trips down box canyons and blind alleys I was able to
backtest this Lazy Man’s Momentum Strategy.
https://papers.ssrn.com/sol3/papers.cfm?abstract_i...

From the Exhibit A2 of the paper, there are 16 ETFs which have been
available since July 1996, a period of 29 years.



Tested with
Number of dates: 58
Starting 1996/06/01 ending 2024/12/01

29 years. All the ETFs were available on June 1996.
Equal weight recast in July and December, for a 6 month hold. Buy the
top N stocks ranked by 6 month Relative Strength (momentum).
Returns include reinvested dividends. RS is price-only.


6 month relative strength using 16 ETFs.
Take the top N in equal weight, reconstituted in July and December

Performance calculated for each portfolio run.
Long term results are the average of all 58 runs.

CAGR and standard deviation:
Top 2: 11.6%    Stdev: 33%
Top 5: 11.4%    Stdev: 28%
Top 10: 9.5%    Stdev: 27%
Top 12: 9.7%    Stdev: 26%
All 16: 9.8%    Stdev: 26%

SPY buy & hold: 12.1%    Stdev: 21%

Another great idea destroyed by reality.

nice job on the testing!

I also found nothing worthwhile. The drawdowns were too high compared to the slight gains in CAGRs compared to the SP. The SDs of 26-33% above confirm the 50-65% DDs I saw. Having said that, if you test the final picks with 200 day MA and stay with less than 5-6 picks of the 16, it ain't bad. I am guessing if you used some BCC level and used Qs and SPY you would blow it away. just my thoughts.

Post New

Berkshire Hathaway FAQ

---

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