An allocation based on correlations

[eukrate]

imported post

From the TSP site, I took daily prices of the 5 funds, found Friday-Friday weekly
returns and analyzed the correlations. The results can be useful for those looking
to diversify into uncorrelated funds to minmize volatility.


Weekly Correlations and Means june 6 03- july 15 05

correl G F +7.8%

correl G C -0.3% correl F C -16.8%

correl G S -9.4% correl F S -18.1% correl C S +90.0%

correl G I -2.3% correl F I -7.4% correl C I +73.9% correl S I +71.1%


mean G 0.08% mean F 0.06% mean C 0.24% mean S 0.40% mean I 0.38%

What's it mean?
G & F are mildly correlated, G does better (mean) so forget F.
C S and I are strongly positively correlated to each other. So pick one of them.
Looking at the GC,GS and GI corrs, GC is least correlated and would be the best
from a diversification point of view. But S & I do 50% better than G (mean)
and G is less correlated to I than to S. So pick I. If this data is representative
and you have a 2 yr horizon, put your money into G and I and skip the others.

 

[rokid]

imported post

Eukrate,

You might want to consider the fund correlations, rates of return, and volatility for a longer period of time, e.g. 10-15 years. In addition, Markowitz Mean-Variance Optimization uses those three statistics to plot the efficient frontier. As you move along the efficient frontier you can discoverthe optimal portfolio for adesired level of risk. Markowitz won the Nobel Prize in 1990 for this work.

You can also download a 30 day, free, trial version, single period, PC-based, MVO program from Efficient Solutions http://www.effisols.com/basics/. The trial version handlesfive asset classes - just perfect for TSP. Enter the fund returns, the fund correlations, and the fund standard deviations and the MVO will generate the efficient frontier for the TSP funds. You can then push the button up and down the curve to see what percentage of each fund should be invested to meet your risk and return needs, e.g. I want the C Fund return with less risk than 100% C Fund.

Finally, if you're interested, I've already calculated, and can provide you with, thelonger term statistics, i.e. return, standard deviation for each fund and the correlations between the funds.

If nothing else, it's instructive and interesting to see how the funds interact to produce a given level of risk/return.

 

[rokid]

imported post

I just got home and checked my spreadsheet. I've got 17 years of data, i.e. 1988 to 2004.

 

[eukrate]

imported post

Thanks for the offer. I'm aware of the optimization and efficient frontier studies.
I use the Matlab finance toolbox for these. I posted the short-term weekly correlations
because they were suprisingly strong. The auto-correlation results are actually more
interesting. I don't put too much faith in efficient frontier results sinces they're built on a faulty assumption - that the market instantaneously accomodates new information ( in fact it over- and underestimates usually, and requires days to correct).

 

[rokid]

imported post

Eukrate,

Well that issue has certainly received a lot of study. Intuitively, I like the concept of an efficient market, i.e. the current price is the best estimate of the correct price.