Correlation and volatility reduction

[IndependentlyPoor]

Warning: Long and Geeky

I have been interested in "modern portfolio theory" for some time. Basically, it is the something-for-nothing aspect of reducing a portfolio's volatility be combining funds with negative correlations that intrigues me.

While there is nothing wrong with combining a money market fund or a short term bond fund with a high volatility stock fund, you miss out on the something-for-nothing aspect. When you combine two such funds, the resulting portfolio will always have greater volatility than the money market fund, along with the greater return (usually). You are just making a risk/return trade off, depending on the relative allocation of the "risky" and "safe" funds. This is a result on the two funds having essentially zero correlation.

However, if you combine two funds that are negatively correlated, you can (theoretically) produce a portfolio that has less volatility than either fund.

To test this out, I wrote a script that found the optimum allocation of all possible two-fund portfolios of Vanguard stock and bond funds (not including money market funds, because I couldn't get data for them), 700 some-odd portfolios. I defined optimum as the allocation that produced the minimum volatility, as measured by the standard deviation of the percentage change in NAV.

Here are the results from the portfolio that enjoyed the greatest reduction in volatility, the FTSE All-World ex-US Small-Cap Index (VFSVX), and the Long-Term Treasury Fund (VUSTX).
negative correlation.gif
The graph shows the the Equivalent APR vs. the standard deviation. The minimum volatility of about 0.6% daily change in NAV occurs at about a 30/70 allocation. Now VFSVX has been around for less than a year, and has a spectacular rise, so a portfolio of these two funds is not a candidate for a core holding, but it does show how the risk return of portfolios comprised of funds with high negative correlations look.

Notice that the minimum volatility point is less than that for either fund by itself. A portfolio of 100% VFSVX had a 60%+ return last year, but also more than 1% standard deviation in the daily NAV, Ouch. VUSTX also had a good year with about 6% return and 0.8% standard deviation, but the 30/70 mix more than doubled the yield of a 100% VUSTX portfolio, while reducing its volatility. That is the something-for-nothing aspect that interested me.

On the other hand, portfolios constructed of positively correlated tended to look like this graph of portfolios of the Emerging Markets Index Fund and the High-Yield Bond Fund:
positive correlation.gif
The minimum volatility occurred at 0% VEIEX and increased almost linearly with increasing VEIEX allocation.

Unfortunately, more popular combinations, such as Total Stock (VTSMX) and Total Bond (VBMFX) don't show much benefit in volatility reduction, and the minimum occurs at about 10% VTSMX.
vtsmx vbmfx.gif

All of this is sort of interesting, but the effects are small. The best standard deviation reduction in percent daily NAV change was about 0.2%... not earth shattering. It seems like that while the claims made for "modern portfolio theory" are true, the effect is just too small to be very important. No free lunch after all.

[ERD50]

Yes, and the problem is that while you can mathematically take advantage of the non-correlation, you need to find funds that are both non-correlated and provide an upward bias. I haven't looked through your examples in detail, but I suspect that your conclusion is correct - 'the effect is just too small to be very important.'.

But it is interesting and it is real - thanks for posting. Maybe we will stumble upon the magic combination here on the forum! Well, at least the historically magic combination...

-ERD50

[IndependentlyPoor]

Quote:

Originally Posted by ERD50 Maybe we will stumble upon the magic combination here on the forum! Well, at least the historically magic combination... -ERD50

Yep. Hours of grinding numbers with sophisticated software on a powerful computer and it tells you that you should have bought AAPL in 1985...

[Gone4Good]

Quote:

Originally Posted by IndependentlyPoor It seems like that while the claims made for "modern portfolio theory" are true, the effect is just too small to be very important. No free lunch after all.

Correlations change over time, so you're not getting the same result that Markowitz did in the 1950's. We've seen correlations increase dramatically across asset classes recently. Even commodities are now reasonably positively correlated with equities. It's likely due to the "mutual-fundization" of investments. The great virtue of mutual funds is that it opens asset classes up to the masses. But when the masses herd in and herd out of everything at the same time you kind of loose the benefit you were striving for.

The one big stand-out in the last correction was treasury bonds. I suspect the flight to safety trade will almost always be in style when other assets are cracking. So instead of using the bond market index as a diversifier, or worse yet, a corporate bond fund, you probably want to use intermediate treasuries.

Oh, and don't forget the massive risk reduction you get because of the diversified equity market portfolio. Run those same calculations with an individual stock, try Citigroup for example, and see what you get.

[kyounge1956]

IP, out of curiosity, of all the fund pairs you looked at, which one had the lowest standard deviation, and what was its return? Which pair had the highest return per unit of standard deviation? And how long a period did your data cover? i.e. is this the last year, last decade, longest available period for any given fund, or what? When I was following the Asset Allocation Tutorial thread, I looked into lots of different asset classes, but I was working from information published in books, and for many of the asset classes I wanted to look at there were only a few years of data, and sometimes I could not get data over the same period for all the asset classes I wanted to run through my spreadsheet. I take it this fund data is available online? I will have to check it out and see if I can improve on my current target allocation.

[IndependentlyPoor]

Quote:

Originally Posted by kyounge1956 for many of the asset classes I wanted to look at there were only a few years of data, and sometimes I could not get data over the same period for all the asset classes I wanted to run through my spreadsheet. I take it this fund data is available online? I will have to check it out and see if I can improve on my current target allocation.

You have run into the same problem as I have. Comparisons between portfolios seems to be only valid if the time periods are the same. Unfortunately, many of the most interesting funds (promising for diversification) haven't been around for long. I wanted to run this scan over the broadest possible range of portfolios, so calculations for each pair are done over the maximum time frame for which data is available for both funds. The range of time frames is extreme. I should include the starting date in the results. Oops.

The data comes from Mathematica's server.

I should point out that I am doing this only as a substitute for crossword puzzles or knitting. I do not expect to discover anything significant about investing.

I'll try to post some more data that address your specific questions later. Right now my poor laptop is cranking away at another 770 portfolios.

[kyounge1956]

Quote:

Originally Posted by Gone4Good (snip)Oh, and don't forget the massive risk reduction you get because of the diversified equity market portfolio. Run those same calculations with an individual stock, try Citigroup for example, and see what you get.

Gone4good, is there a rule of thumb about how many individual stocks are required to reduce the volatility of the group to about that of the market as a whole? I vacillate between wanting to be an index investor (for all the usual reasons) and wanting to invest in individual stocks (to be more socially responsible).

Of course, there are such things as SRI funds, but I don't know if any of them filter for exactly the same set of things I'd want to, nor if I could do the filtering myself on a few stocks for less than the expenses of a fund. I've looked at a few of the Roman Catholic oriented funds, which are probably the closest thing available to what I would filter for, but as I recall the expense ratios were quite high, up to around 2.3% (this was some years ago and expenses may have changed since then). I also don't remember how actively the funds were managed. There might have been efforts toward market timing as well as the issues-based filtering, and I wouldn't want that in any case. Could an individual investor buy and hold a basket of individual, filtered stocks at lower cost than that?

[Gone4Good]

Quote:

Originally Posted by kyounge1956 Gone4good, is there a rule of thumb about how many individual stocks are required to reduce the volatility of the group to about that of the market as a whole? I vacillate between wanting to be an index investor (for all the usual reasons) and wanting to invest in individual stocks (to be more socially responsible).

I seem to recall the magic number to be around 20. That assumes, of course, that those 20 stocks are pretty broadly diversified. It wouldn't do much good, for example, to own 20 makers of wind turbines.

[IndependentlyPoor]

Quote:

Originally Posted by Gone4Good Correlations change over time, so you're not getting the same result that Markowitz did in the 1950's. We've seen correlations increase dramatically across asset classes recently. Even commodities are now reasonably positively correlated with equities. It's likely due to the "mutual-fundization" of investments. The great virtue of mutual funds is that it opens asset classes up to the masses. But when the masses herd in and herd out of everything at the same time you kind of loose the benefit you were striving for. The one big stand-out in the last correction was treasury bonds. I suspect the flight to safety trade will almost always be in style when other assets are cracking. So instead of using the bond market index as a diversifier, or worse yet, a corporate bond fund, you probably want to use intermediate treasuries. Oh, and don't forget the massive risk reduction you get because of the diversified equity market portfolio. Run those same calculations with an individual stock, try Citigroup for example, and see what you get.

Yep, yep, and yep
Correlations do seem to have changed. In fact, they don't seem stable to me at all. Possibly so unstable as to call into question using them like this at all. Somewhere I have a bunch of plots of rolling 60 or 90 day correlations and they are all over the place.

The recent unpleasantness has tossed all this sort of analysis for a loop. Risk/return curves are upside down, with treasuries out performing stocks. (Well, at least that was true the last time I looked - a couple of thousand points ago.)

My recent runs tell me that if I had just put everything into health care and long term treasuries I would be rich now.

Running portfolios of individual stocks would be interesting, except that I am drawing a blank on which to run as examples. Suggest some ticker symbols, and I'll run'em. Give me two lists, 5 or 10 symbols each, and my script will run find the optimum allocation for each portfolio composed on one stock from one list and the other from the second list.

[Gone4Good]

Quote:

Originally Posted by IndependentlyPoor Yep, yep, and yep Correlations do seem to have changed. In fact, they don't seem stable to me at all. Possibly so unstable as to call into question using them like this at all.

I personally think they're unstable enough so that using them to "optimize" a portfolio is pointless. Some folks tie themselves in knots trying to construct the perfect portfolio and completely overlook the fact that the benchmarks they're using to build that portfolio are variable.

[kyounge1956]

Quote:

Originally Posted by IndependentlyPoor You have run into the same problem as I have. Comparisons between portfolios seems to be only valid if the time periods are the same. (snip)

I was only guessing that the data should all come from the same time period. It seemed like mixing apples and oranges otherwise. Seems my intuition was correct, since as G4G has pointed out, the correlations can change over time. The references I was using said the same thing.

Quote:

I should point out that I am doing this only as a substitute for crossword puzzles or knitting. I do not expect to discover anything significant about investing. I'll try to post some more data that address your specific questions later. Right now my poor laptop is cranking away at another 770 portfolios.

You can do the crossword while your laptop is slaving away. And if you ever have any questions about knitting, don't hesitate to ask. I'm an old hand.

P.S. Can your laptop compare groups of three funds, or only pairs?

[IndependentlyPoor]

Quote:

Originally Posted by kyounge1956 P.S. Can your laptop compare groups of three funds, or only pairs?

Yep, it can do threebies. Actually, any number is theoretically possible, it just becomes a hassle to program. I'll try to post some results with consistent time periods.

Right now my overworked MacBook is optimizing portfolio 588 out of 770.

There is a big difference between optimizing for minimum risk (lowest standard deviation) and return/risk. Right now I am just looking at minimum risk. These days (after the crash) looking at risk/return tends to just tell you to buy treasuries. Maybe that is changing some now that the market is recovering.

[kyounge1956]

Quote:

Originally Posted by IndependentlyPoor Yep, it can do threebies. Actually, any number is theoretically possible, it just becomes a hassle to program. I'll try to post some results with consistent time periods.

I found the formula in a book and made spreadsheets to do the calculations. The most asset classes I can handle at one time is 6, which is the most I have simultaneous data for anyhow.

Quote:

Right now my overworked MacBook is optimizing portfolio 588 out of 770.

I can almost here the numbers crunching!

Quote:

There is a big difference between optimizing for minimum risk (lowest standard deviation) and return/risk. Right now I am just looking at minimum risk. These days (after the crash) looking at risk/return tends to just tell you to buy treasuries. Maybe that is changing some now that the market is recovering.

Are treasury bonds also the lowest standard deviation overall? Maybe the highest return/risk isn't actually what I meant. When I was looking for a target allocation I picked a maximum standard deviation and then looked for the combination of assets that would give me the most return without exceeding the volatility ceiling. Then I plugged that return into the online Monte Carlo modeler I use to make sure it was adequate for portfolio survival. It's probably time to check again and see if I'm on course or not.

[kyounge1956]

Quote:

Originally Posted by Gone4Good I personally think they're unstable enough so that using them to "optimize" a portfolio is pointless. Some folks tie themselves in knots trying to construct the perfect portfolio and completely overlook the fact that the benchmarks they're using to build that portfolio are variable.

I have to plead a little guilty there, but I hope they're good enough to get me in the ballpark, or for broad generalizations like "any significant allocation to such and such an asset class makes the portfolio more volatile than I care for", or "....makes it return less than I need it to", or "...doesn't make much of a difference either way".

[IndependentlyPoor]

Okey dokey, I generated 770 portfolios, each composed of one Vanguard stock fund and one other Vanguard fund (no money market funds though). For each portfolio, I found the asset allocation that produces the minimum volatility.

Here is the amount of volatility reduction achieved in minimum volatility portfolios plotted against correlation. It is clear that something is affecting volatility other than correlation.
portfolios1.gif

The best way to view the data is to download the Mathematica viewer. It is free, works on both Mac and PC, and is easy to install. Here is the link.

I need a drink.
Wolfram Mathematica Player: Download
Each point represents a portfolio with the asset allocation chosen to produce the minimum volatility. The charts are interactive. When you hover the cursor over a point, a popup displays the asset allocation and return for that portfolio.
Here is the link for the interactive chart:
http://www.wolfram.com/solutions/int...nd+publish.nbp

If you want to see the shape of the risk/return curves underlying each minimum volatility portfolio, download and view this interactive chart.
http://www.wolfram.com/solutions/int...+2+publish.nbp
The popups look like this.
portfolios2.gif
They show the asset allocation and the return of the minimum volatility portfolio, in addition to the risk/return curve of several asset allocations. The minimum volatility point is the leftmost point on the popup graphs.

[Midpack]

During the meltdown of the 2008-09 I would agree that correlations broke down and everything declined. However, I still think it's wise to be aware of the correlation of all assets and to try to utilize some negative correlation. While it may not pan out all the time, what is the downside of attempting this strategy to reduce volatility? It has worked (to some extent) for me most of the time.

The Benefits of Low Correlation - JOI Articles - IndexUniverse.com - ETFs, Index Funds, Indexes

[Texas Proud]

Quote:

Originally Posted by Gone4Good I seem to recall the magic number to be around 20. That assumes, of course, that those 20 stocks are pretty broadly diversified. It wouldn't do much good, for example, to own 20 makers of wind turbines.

When I took portfolio analysis in college.... it was 30... who knows now...


I always find it interesting that some people say the S&P 500 is not diversified... it is... it might not be the portfolio you want, but it IS diversified...

[IndependentlyPoor]

Here are the top few portfolios sorted by return/volatility.
Again, please note that these portfolios were calculated with different time periods. I don't know if there is any useful information here at all other than that bonds have been good recently.
correlation and risk reduction.gif
Here is a pdf with the 100 best return/volatility ratios.
correlation and risk reduction.pdf

Finally, I should apologize for the misleading titles on the popup graphs in the previous post. The graphs all all labeled at the top with something like 63/37 VMNFX/VIPSX. The 63/37 is the asset allocation of the minimum volatility VMNFX/VIPSX portfolio, but the graph is of the risk/return of various allocations of VMNFX/VIPSX from 100/0 to 0/100.

[audreyh1]

Quote:

Originally Posted by Texas Proud When I took portfolio analysis in college.... it was 30... who knows now... I always find it interesting that some people say the S&P 500 is not diversified... it is... it might not be the portfolio you want, but it IS diversified...

I think when they say that, they mean S&P500 is all US-based large-cap stocks. If you want small-cap, mid-cap or foreign stocks, you're not going to find them in the S&P500.

Audrey

[ziggy29]

Quote:

Originally Posted by IndependentlyPoor Yep. Hours of grinding numbers with sophisticated software on a powerful computer and it tells you that you should have bought AAPL in 1985...

Even if you missed that go-round, you still had another shot around 1997 or so, when the company was in a Gil Amelio-led death spiral and the stock was basically wallpaper. Of course, you'd need to know Steve Jobs would come back and resurrect the product line, or else it would have felt like investing in Enron.