The Do-It-Yourself Market-Neutral Portfolio

The hedge fund business has been growing rapidly in recent years—with no end in sight. Wealthy investors are seeking alternative investments to meet their needs and hedge fund managers have an array of tools and strategies at their disposal. One of the core strategies employed by hedge funds is what is called market-neutral investing. A market-neutral portfolio is designed to deliver returns that are not impacted by the movements of the broader market. This does not mean that a market-neutral approach is low risk, however. Goldman Sachs' (GS) largest hedge fund employs market-neutral strategies and has attributed their very poor recent performance partly to this approach.

The defining objective of a market-neutral portfolio is that its returns have low correlation to the broader market. Investors use such funds as a diversifier to their other investments in equities. Low-correlation between assets improves their diversification value. In practical terms, increased diversification means that you can achieve a higher return for the total amount of portfolio risk.

Market-neutral portfolios are achieved through a range of strategies. The simplest and best-known approach to market-neutral investing is to take long and short positions in stocks so that market exposure from the long positions is offset by the short positions. There are a number of mutual funds that exploit long-short strategies and this article is a good overview for the layperson.

A list of top-performing long-short mutual funds can be found here.

Long-short portfolios tend to ride out market downturns quite well, but they tend to miss a fraction of the earnings growth in the companies that they invest in—simply because some of their bets are short. The fees for gaining access to long-short portfolios tend to be high. Hedge funds typically charge substantial fees and the mutual funds that that exploit this strategy are typically quite expensive. The Diamond Hill Long-Short Fund [DIAMX], for example, has an expense ration of 1.5% and a front-end load of 5%.

There is another way to build a portfolio that is (almost) market-neutral that is much cheaper and quite simple. It is possible to build a portfolio that exhibits very low correlation to the broader market (say the S&P500) in which there are no short positions—you simply use a carefully selected group of stocks that exhibit very low correlation to the broader market and, ideally, to each other. The effectiveness of such an approach will surprise many people, but I have been studying this type of portfolio for some time and this is not a fluke.

I have built a sample (almost) market-neutral portfolio by screening for stocks and closed-end funds (CEFs) that have moderate or low price-to-earnings ratios (P/E) as well as low Betas. I wanted to look at some stress tests during the last bear market, so I also wanted companies with at least eight years of data. In the next step, I rejected a set of these stocks and CEFs that were in the higher levels of R-squared (also written as R^2) for the group and some that exhibited correlations to other portfolio components that were a bit high. I ended up with a portfolio of twenty components and I am decided to test the performance of a simple-minded portfolio with equal allocation to each—this is our model market-neutral portfolio:

click to enlarge

If you have a look at these companies, several things become apparent. First, all of these selections have P/E ratios below the S&P500—which is above fifteen at the time of this writing. There is also quite a bit of concentration in terms of sectors. There is a major concentration in oil, refining, and related services such as drilling and shipping. There are a couple of property and casualty insurance firms. There are also four closed-end debt funds. The final major sector is banks. There is also a position in a CEF that invests in gold and silver. This specific list of companies is quite arbitrary, and is a result of the screening procedure—but it is a worthy case study.

First, let’s look at the correlations between all of these stocks and funds over the last eight years (below). The first thing to understand is that these are correlations between monthly total returns---not prices. Correlations in returns between portfolio components determine how well your portfolio exploits diversification effects. The correlation matrix shows the correlation between the ticker in a column with the ticker in the intersecting row. The correlation of BPT (first column) with OMM (fourth row) is 37%, for example. The correlation of anything with itself is 100%, which is why the diagonal of a correlation matrix is always 100%.

There are some very low correlations between stocks and CEFs that are ostensibly in the same industry. BPT and SJT are both oil royalty trusts, but they have a correlation in returns over the last eight years of only 43%. Many of these stocks have maximum correlations to other portfolio components of less than 30%, with median correlations far below this level. Companies that are in the same sector may exhibit low correlation to each other—depending on industry. The typical correlations between broad market sectors are far higher than most of the correlations shown here. The correlation between the S&P500 and EAFE index, for example, is around 75%-80% and the correlation between the S&P500 and the NASDAQ index is greater than 85%.

The chart above shows that the correlations between these stocks and funds have been very low over the past eight years. An issue that has gotten some attention is whether correlations between certain sectors are increasing in time. If true, this would make portfolio diversification based on history appear more effective than it will be in the future. Fortunately, this is easily analyzed.

In the tables below, I show the correlation matrix for the same tickers, but having generated one matrix for the first four years of the period (May 1999 through April 2003) and another matrix for the later four years (May 2003 through April 2007). If you examine the numbers, you will see that plenty of the correlations have changed between the first and second four-year periods. This is not crucially important for the portfolio, however. The average change in correlation between portfolio components is zero. If we saw an average change in correlation that was greater than zero, we might conclude that there was a trend towards more correlation between these stocks (and vice versa). There is no trend towards increasing correlation in this portfolio. The median change in correlation between any two of these stocks and funds is a 2% decrease.

Looking at changes in correlation in different time periods can be important. If correlation effects are increasing in time, your portfolio is getting less diversified in time—something you do not want to see. Ideally, on a portfolio basis, no such trend will show up. The stability of the overall correlations is a good sign for this portfolio. The fact that the correlations are so low for a number of companies in the same industry does bear some careful thought. Over the eight years of data in this analysis, the average return for companies in the oil industry has been higher than in many longer periods, and the average return is not a component of correlation.

To make sure that this effect has not biased our correlation, I also had a look at fifteen years of history—going back to May 1992. The correlation between BPT and SJT, both oil royalty trusts, is lower for the fifteen year period than it has been in recent years (48% in the longer period and 64% in the most recent four years), but higher than the last eight years (43%).

Having established that there are no major trends in the correlations in this portfolio as a whole, we can look at the performance on an historical and projected basis—this is where things get interesting. I have analyzed this portfolio using Quantext Portfolio Planner [QPP], a portfolio planning tool that generates outlooks for portfolio volatility and return for individual assets and for total portfolios. QPP calculates both historical performance and an outlook using a specific historical period to generate the predicted performance. The table below shows the results using QPP to generate an outlook for our portfolio using eight years of data as input (and all baseline settings for QPP). The Portfolio Stats table shows the predicted future average return and standard deviation in annual return. The Historical Data shows the historical performance.

Over the last eight years, the market as a whole has generated low returns (average annual return of 2.6% for the S&P500 before dividends). Even including dividends, the S&P500 has generated an average return of less than 4.5% per year. Our market-neutral portfolio has performed very well over this period, with average returns of 20.5% per year and with a lower volatility than the S&P500 (11% vs. 14.1% for the S&P500). The real proof that this portfolio is, indeed, market-neutral comes from looking at Beta and R-squared (R^2). A perfectly market-neutral portfolio would have Beta of zero and R^2 of zero. Our portfolio has Beta of 32% and R^2 of 17%.

For purposes of reference, an S&P500 index fund would have Beta and R^2 of 100% and a government bond index fund would have Beta and R^2 of 0%. The values observed for our (mostly) market-neutral portfolio are very low. We could play with the allocations to try to get Beta and R^2 even lower, but this is not terribly practical. For investment purposes, we simply want a portfolio that weathers market downturns well---we are not worried about making a portfolio absolutely uncorrelated to the broader market.

Consider the historical and projected performance from QPP when we split our eight-year period into two four-year periods (below). When I ran QPP using market data from 5/1/1999 through 4/30/2003 as the only inputs, QPP generated the results shown on the left side of the figure below. This period encompassed our most recent bear market—note that the S&P500 returned -7.3% per year (before accounting for dividends) in this four-year period. This was also a high volatility period for the S&P500, with standard deviation in annual return of 18.2% (also shown below). Our market-neutral portfolio generated 16% per year, with a standard deviation of 12.6%---much higher return with lower volatility. Over this bear market, R^2 was about 19% and Beta was 30%. QPP’s projection for the future performance of this portfolio was for an average return 14%, with a standard deviation of 11% (Portfolio Stats), using only the four years of data through 4/30/2003 as input.

When I ran the portfolio through QPP using historical data from 5/1/2003 through 4/30/2007 and all baseline settings, QPP generated the table on the right side of the figure below. Over this period, the S&P500 generated an average annual return of 11.3% per year (before dividends) --- about 13% per year when we include dividends—with a standard deviation of 7.3%. The market-neutral portfolio generated an average annual return of 23.8%, with a standard deviation of 8.95%. This four year period has been characterized by fairly high average returns but, more significantly, very low market volatility—as evidenced by the low standard deviation. Recent years have seen market volatility that is historically low. This portfolio has generated annual returns that are much higher than we can expect for this volatility level—and something will have to give.

QPP, using data from this most recent four years, predicts that this portfolio will generate returns of 21% per year, but with volatility that is slightly more than twice what we have seen over the past four years. When we used the entire eight-year period, QPP projected an average annual return of 15% per year, with a standard deviation of 12.5% (see previous chart). These results provide a reasonable range for what might be expected in the future from this portfolio.

Over the most recent four years, the portfolio has exhibited even lower R^2 (8.9%), but slightly higher Beta, than we saw over the previous four years. R^2 (R-squared) measures that percentage of the variability in a portfolio that can be explained by the movement of the broader market. These results mean that less than 9% of the performance of this portfolio is due to movements in the S&P500 in the last four years. R^2 is higher for the bear market period, but still very low at 19% (see table above). Even during this bear market, less than 20% of the returns of this portfolio were driven by the broader market. The R^2 value for this fund is less than the R^2 for DIAMX, for example, despite the fact that DIAMX takes short positions and also has a concentration in energy.

While our simple portfolio is obviously not perfectly market-neutral, it has generated very solid returns in recent bear and bull markets and the R-squared of this portfolio has been very low in both situations. It is certainly possible to ‘tune’ this portfolio or one like it to be perfectly neutral on an historical basis but this may not be productive because we would end up over-tuning to history.

Using three sets of historical data (the eight years and the two four-year sub-periods), QPP projected that this portfolio would generate average annual returns that are slightly greater than the projected standard deviations in annual return. As I have discussed previously, this is about as good a return-to-risk ratio as you can realistically plan for, even if trailing performance is much higher relative to volatility.

If you want the general features of a market-neutral strategy, you could do a lot worse that following the general strategy shown here. Please understand that I am not advocating this specific portfolio---each investor must consider the specific themes that he or she wishes to pursue. The strategy outlined here is not limited to this specific set of stocks—this is just an example. My principal point here is that you do not need to invest in a hedge fund or a long-short fund to garner the benefits of de-coupling your portfolio from the gyrations of the major market indices. This type of almost-market-neutral approach can be used to manage the total market sensitivity of a portfolio at very little cost.

Note: all calculations performed here and all tables were generated using Quantext Portfolio Planner with standard settings.

Disclosure: author is long BAC.

Comments

[Ralph F:]

This is a really thought provoking (and well written) article -- thank you.

Question: is your analysis based just on back-testing? Could we find that stocks or funds with energy exposure do indeed exhibit correlation with the broad market going forward if there's an economic downturn?

2007 Jun 04 12:19 PM

[Geoff Consi...:]

Ralph:

Thanks for your comments. QPP's projections do depend on historical correlations, volatility, etc. I have gone to some lengths to avoid 'over fitting' but every analysis using historical data is subject to the specifics of these data. QPP's projections do not, however, simply rehash historical returns, volatilities, and correlations. I have research showing how QPP works for a real portfolio of stocks over 30+ years. Also, this idea of a very low Beta / low R^2 / low correlation portfolio using only long positions has been demonstrated in a number of my other papers. Could correlations shift so much that this approach fails? Certainly. Any analysis is subject to the potential for the world to shift.

The fact that this portfolio has done well in recent bull and bear markets AND that the monthly returns exhibit such low R^2 and Beta makes me feel that these results are fairly robust. Obviously this portfolio has too much energy exposure to be a real choice for a total equity portfolio. I would be worried about how this portfolio would do in the event that oil drops a lot...that can can tested using the correlation of the portfolio to an oil index--I didn't do that in the article, but it would be easy enough using QPP.

BTW, DIAMX has a lot of energy exposure, too. If you look on Morningstar here:

quicktake.morningstar....;Symbol=DIAMX&...

you will see that DIAMX has an R^2 with respect to the Goldman Sachs Natural Resources Index of 71%---that is quite high. This means that movements in this index explain 71% of the variability in the returns on DIAMX.

2007 Jun 04 05:39 PM

[Everyday Fi...:]

Nice article; great variety of options (no pun intended) out there for the build-it-yourself portfolio. A few additional considerations and options out there:

Some of the instruments listed, especially in the energy sector are additionally favorable in that they provide significant dividend yields; there are several CANROYs, commodity ETF and international REITs that yield between say, 7 and 10% annually. I've performed some analysis on these and own a couple myself. Not only does the high yield help differentiate the correlation of the total return, but it moves the return in a favorable direction, regardless of the major market indices. I bought into a few that started off yielding around 10% and have since increased in share price signficantly as well (of course, now the yield has dropped into the high single digits, but I'm still earning 10%+ on my initial investment).

Additionally, a great way to get some additional non-correlated performance in the 9-12%/yr range is through lending on Prosper.com. When diversifying your loans, you can asymptotically approach the returns of the overall performance measures as lists on the site for "all loans" in particular credit categories. This can be achieved simply through "average" performance. If you're good at researching and choosing your loans, you can exceed these returns, inclusive of defaults. To date, I have over 20 loans out with an average of 13.4% returns. I've also included top lender groups, strategies and learnings on my everydayfinance blog. Feel free to visit, review and leave comments by clicking on my name for this post.

Dan

2007 Jun 04 04:26 PM

[dsassyt@aol...:]

I have not closely studied your suggested portfolio, but have you taken into account international allocation? Is this primarily a US portfolio?

2007 Jun 04 05:09 PM

[Geoff Consi...:]

Well, you will note that several of the holdings are Canadian. Second, we have Mitsui ADR's. Oil companies tend to have performance that is correlated to emerging markets--just because a number of these economies are driven by oil. This study was performed in the U.S. currency and there are more domestic companies than non-U.S. companies, but the strategy will hold up with quite a few variations. Again, I stress that this article is a demonstration of a concept rather than being an ideal portfolio.

2007 Jun 04 05:45 PM

[Dan Sweet:]

Geoff, interesting article. I am a trial user of QPP and finally got it working properly last night. I have a question regarding the different sample portfolios you have profiled in you articles over the last year or so. You've done simple equally divided asset class diversification, more complex portfolios with differing percentages held in each class, sample portfolios incorporating SPY and QQQQ, mixes of ETFs and individual stocks and now another portfolio made up of primarily individual equities. Your performance / risk numbers and ratio have continued to improve as your articles have evolved. Is a portfolio of the sort profiled in this article is a fairly late stage evolution of the work you have been doing with QPP or are you just illustrating different possibilities? Or will you have a new sample portfolio next month with even better numbers an different allocations?

When you screened for the sample stocks in this portfolio were you looking for individual companies to provide exposure to utilities, materials, emerging markets and other sectors that have been performign well or is it really a straight screen for low P/E and low betas with 8 years of data? I'd like to reproduce these results myself but am not sure where to start with screening for companies.

2007 Jun 04 07:54 PM

[Dan Sweet:]

I'm screening for stocks using the MSN Money Deluxe Stock Screener, what tool do you use to screen closed-end funds? Is there a better tool that will scan for both concurrently?

And to restate my question from above more explicitly...did you just screen and end up with concentrations in these asset classes? Or did you say "I want some debt funds, some utilities, some energy" and run screens for picks within those specific sectors?

2007 Jun 04 11:06 PM

[Geoff Consi...:]

Hi Dan:

Yahoo's stock screener also gives CEF's--it is free and easy. In this case, with the P/E criteria that I used, these industries ended up with high representation--this is a consequence of looking for low Beta and low P/E.

2007 Jun 05 12:20 AM

[manhapf:]

Hello Geoff,

Thanks for your software and articles; they've been insightful and helpful. I've been interested in portfolio asset allocation for many years and am a user of some complex software tools; love QPP's format, nicely done! I've been trial using your software and have some questions/observations...

Rebalancing/Optimizati... of Allocations:
One of the implementation details in using QPP is when to future adjust and at what frequency the 'final' asset/security allocations? At present, you either define an allocation weighting of equal or some proportion that 'suits' the investor's preferences for risk/return ideally based on empirical knowledge; eg, 70% equities/30% bonds.

I'm not a fan of software optimization in general as applied to portfolio allocations, I'm quite familiar with the pitfalls and the bad results that typically are produced from even excellent out-of-sample and supposedly robust models. However, since with QPP we're using monte carlo simulations of forward looking return possibilities and 'experimenting' with various assets/security combinations/allocatio... weightings anyways, there is a temptation to use Excel's built-in Solver tool to optimizatize. For example, we can maximize the forward returns by modifying the asset/security allocation weights constrained by the forward looking standard deviation for example.

Ideally, the optimized forward returns/SD yield the best weights and are robust in the sense that they deliver similar optimized/simulated results compared to the historical results over 1, 3 & 5 year time horizons; would you consider longer time horizons? I've attached a before/after example using your market-neutral article to demonstrate the effect of optimizing the allocation weights which appears quite compelling since it also improves on the portfolio beta and diversification values.

So, what frequency of re-running the simulations to adjust the allocation weightings do you normally recommend once they're established to ensure the expected return/SD results are still meaningful; annually? What's your opinion of optimizing the weightings in this manner?

Disclaimers:
1) To what index is Beta used to capture correlation; am assuming S&P500, but should be stated.
3) Based on research I've done, asset/security returns are not normally distributed. I realize that this makes the programming/math easier, but this is an area that could be improved and make QPP a more valuable tool and one that I'd be willing to pay more for!
6) Based on your articles, I assume that this refers to the 1:1 relationship normally displayed between returns and standard deviation; is this correct?

Thanks much for your time and consideration of my questions/observations...

2007 Dec 28 08:20 AM

[Steveal50:]

Geoff,

Any comment on this article in the light of a subsequent years performance?

The banks were hammered, of course, but the overall performance seems right in line with the indices.

Steve

2008 Jul 08 06:07 AM

[Geoff Consi...:]

Steve:

Over the last 12 months, I get about the same result--depending on whether you drop OMM when it was delisted due to acquisition or substitute its acquirer. Either way, the portfolio is down by an amount close to the S&P500, as you suggest. That said, note that the R^2 of 20% or less over long periods of time means that it is not really a good idea to benchmark against something like the S&P500--most of the volatility in the portfolio is not due to moves in the S&P500. A "market neutral" apporoach that has fairly high volatility (like this one) may not drop when the market drops--or it may--the market direction is not the driver. In this case, the credit crisis has been a big driver because of the exposure to banks here.

This kind of portfolio is a good one to monitor over the long haul--thanks for the reminder :)

Geoff

2008 Jul 08 10:37 AM