Post New

I am reading Stocks on the Move by Clenow and he suggests a momentum screen that uses the exponential regression slope multiplied by the coefficient of determination. GTR1 can determine the slope using rrs (x.y) Does GRT1 have a command to get the coefficient of determination?

Thanks

Craig

Well, I was able to answer my own question. GTR1 has vol(a,b). The book by Clenow uses the SP500 components. A rough approximation of the screen from 1985 using SP 500 components, BCC>0 has a CAGR of 14.09 and SAWR of 7.49.

However, why I am reading books instead of old posts I am not sure. BarryDTO proposed a method of adjusting the linear egression by subtracting the variation. linear(1,rrs(1,126),-1,vol(1,252)) top x

Using his method from 11 years ago and 4 years before the book was written has a CAGR of 14.79% and SAWR of 8.91%

Changing the stocks to nas100 has a CAGR of 22%

No more trips to the library!

Aussi


Clenow

https://gtr1.net/2013/?s19851231f0.10000::BCC:gt0:...

BarryDTO

https://gtr1.net/2013/?s19851231f0.10000::BCC:gt0:...

Does anyone know how to replicate the GTR1 vol(a,b) calculation in Excel?

Thanks in advance.

Baltassar

I think this is it for vol(a,b). I haven't checked it.

https://www.statology.org/r-squared-excel/

Aussi

The GTR1 function vol() is not R-squared. R-squared is between 0 and 1. vol() could be higher than 1.

R-squared = 1 - RSS/TSS
RSS = sum of squares of residuals
TSS = total sum of squares (proportional to the variance)

The GTR1 function vol() calculates the annualized standard deviation of the natural logarithms of total returns.

This could be checked in Excel using the functions:
n = A:A = 1,2,3...
PortValue = B:B = 1,2,1,2,1...
C:C = total return multipliers (always > 0)
D2 = ln(C2) ... fill formula down to apply to all return data
stdev(D:D)

rsq(A:A,B:B)

For example, alternating portfolio value of 1 and 2 have:
R-squared = 0
vol(ln(TRm)) = 0.71

n   PortValue  TRm  ln(TRm)
1       1
2       2       2     0.69
3       1      0.5   -0.69
4       2       2     0.69
5       1      0.5   -0.69
6       2       2     0.69
7       1      0.5   -0.69
8       2       2     0.69
9       1      0.5   -0.69
10      2       2     0.69
11      1      0.5   -0.69
12      2       2     0.69
13      1      0.5   -0.69
14      2       2     0.69
15      1      0.5   -0.69
16      2       2     0.69
17      1      0.5   -0.69
18      2       2     0.69
19      1      0.5   -0.69

vol() uses natural logarithms, and so should probably only be used with multiplication and division. Using some other base would change vol() by some multiple, but should not affect the screen picks. For example, the screen logic should be set up to pick the same stocks after vol() is multiplied by 10.

lizgdal

Thanks for the correction on vol().

Aussi

IIRC - there used to be a screen which uses the linreg ( 1 variable beta essentially) 126 day .... which was available in GTR1

One could potentially have an approximation - but I havent thought thru the math.

As lizgdal points out its (Coeff of Determination) is nothing but R-square [ I have to say - its been a long while someone referred to it as CD - rather than the common name]

If its involves pooling of stocks - you have to check the GTR1 calc - whether it would do it for the universe. If its a single stock the Linreg should work fine


Best

Post New

US Policy FAQ

---

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