# Does More Diversification Result in Better Performance?

Do highly concentrated managers outperform their more diversified peers? We find that concentration can be a double-edged sword. Learn more.

Hit Enter to Search or X to close

Novus Editorial# Does More Diversification Result in Better Performance?

Do highly concentrated managers outperform their more diversified peers? We find that concentration can be a double-edged sword. Learn more.

Do highly concentrated managers outperform their more diversified peers? We find that concentration can be a double-edged sword. On the one hand, concentration pays off handsomely when the top trades succeed. Some of our published work supports this – we have shown that the highest conviction stocks within the hedge fund industry consistently outperform the broader market.

On the flip side, concentration increases the risk of top trades damaging performance when they begin to underperform. In this article we explore data of over 1,000 public hedge fund portfolios to determine what concentration means for performance and volatility. We also highlight some trends we found when analyzing historical data on the portfolios through the lens of public ownership.

Key Points

- Larger managers are more diversified than smaller ones
- Concentration has a very weak relationship to performance
- A higher level of concentration corresponds to a higher level of volatility
- Managers have been getting more concentrated over time

It is important to underscore that the data in this study is sourced exclusively from public sources. The data used here omits the short side, non-equity securities, many non-US securities and all non-public information such as actual fund performance. Manager performance is simulated based on the returns of disclosed holdings.

We use the Herfindahl Index to measure portfolio concentration. The measure was originally designed to gauge concentration in industries and identify monopolies or oligopolies (by measuring industry market share). We find it is a reasonable metric to use on securities within a portfolio as well. It is defined as the sum of squares of position sizes. For example, the Herfindahl index for an equal-weighted portfolio of 50 positions is 2%; the Herfindahl for an equal-weighted portfolio of 100 positions is 1%. The appendix at the end contains more details and examples on the methodology and intuition.

We studied the distribution of Herfindahl across managers in our Hedge Fund Universe as of May 2015. Later in the article we will put this more recent period in historical context.

We see that most managers have a Herfindahl index in the 1-9% range. This can be thought of as the weighted average position size (for levered portfolios this can be likened to average percent of exposure). With a mean of 10.7% vs. a median of 4.7%, we have a distribution that’s skewed to the right (there can be a number reasons for this, one of which is a reflection of the structure of public data). Now that we have a base line of concentration for managers, we can also look for relationships it has with AUM, performance and volatility.

To see if managers with a larger asset base tend to be more diversified, we regress total reported assets against Herfindahl.

We find that higher AUMs mean more diversified the portfolios. A low R-squared (of 2%) tells us that variation in AUM explains a very small portion of variation in concentration, and a standard error of 11% on the regression means that if we solely relied on AUM to predict how concentrated a manager is, it would be an imprecise model. But a statistically significant coefficient on total filing value means that on average, the higher the AUM the more diversified the manager.

To test for this we plot Herfindahl against managers’ trailing 24-month performance as of May 2015 (log scale).

Here we actually see a negative coefficient – higher concentration generally corresponds with worse performance but the *relationship is very weak*. But could it be that a few outliers are skewing this (a couple managers with very high concentration that happened to do very badly)? Interestingly, when we exclude all the managers in the last quartile (with the highest concentration levels), there still doesn’t seem to be much of a relationship between concentration and performance.

In the above we see a marginally positive coefficient and a very low fit. Taking a step back, intuitively concentration just means moving exposure into fewer positions – there is no fundamental reason to believe that higher concentration leads to better performance either. But because the stakes are higher, it’s reasonable to suspect that it leads to higher volatility. Next we plot concentration against standard deviation for this universe of managers.

Now the scatter and regression show (also with axis on log scale) that the coefficient on concentration is positive. The R-squared is still small but has improved – to about 8%; a regression also showed that the coefficient is statistically significant at 99%. What this means is that while concentration does not fully explain volatility across managers (it could be driven by a host of other factors, such as the volatility of their underlying holdings), higher concentration does on average lead to higher volatility (it explains about 8% of the variation in manager volatility).

Instead of looking at a snapshot in time, we can look at how Herfindahl has evolved historically to better contextualize the current distribution.

Remembering that this tends to be a right skewed distribution, we include the 25th and the 75th percentile as well as the median. We see in the chart above that while the 25th percentile has remained steady and the median has shifted up only slightly, the 75th percentile has risen from the 6% range pre-crisis to around 10% recently. That is, even though the average manager (median) is not becoming much more concentrated, those who were already relatively concentrated are getting more concentrated. Another way to view the spread between concentrated and more diversified managers is to look at the standard deviation of Herfindahls over time.

Standard Deviation in concentration across managers has been increasing since 2005 and is at an all-time high. As shown earlier, this has been largely driven by concentrated managers becoming more concentrated (rather than diversified managers becoming more diversified). For managers that are increasing concentration, while it doesn’t necessarily mean better performance, it is a contributing factor to potential future volatility.

To summarize the high-level takeaways, we found that:

This study is based on quarterly data and from public disclosures, but for private portfolios, concentration affects managers in at least 2 more ways:

In light of the caveats, it’s useful to reiterate the interpretations so we don’t make too much or too little of these numbers. Through the lens of public data, AUM is an imprecise predictor of concentration, and concentration is an imprecise predictor of portfolio volatility. But on average, higher AUM corresponds to more diversification, and more concentration corresponds to higher volatility. Finally, concentrated managers have been getting more concentrated – over the last 15 years the average Herfindahl for the most concentrated set of managers has increased from below 6% to above 10%.

Liquidity has been an ongoing concern for market participants since the financial crisis. This concern is partly due to structural changes, as well as the rise in High Frequency Trading. It is also due in part to the lessons investors learned about the dangers of investment illiquidity during market stress. Those with legacy side pockets from 2008 alternative investments know this all too well; some are still in the process of liquidating even today.

In our latest paper, the question we are essentially trying to answer is how funds with impaired liquidity profiles fare in periods of market stress, and discover whether hedge fund liquidity is an alpha generator or risk factor.

Download our latest report to learn more.

The Herfindahl Index (originally constructed to measure concentration of industry market share) can be used to measure portfolio concentration. It is defined as the sum of squares of position sizes:

The intuition is similar to why differences are squared to calculate standard deviation. If we simply added each position’s percent of exposure it would always add up to 100%, so we square the position sizes and sum them to capture what is essentially a “weighted” average of each position’s contribution to total exposure.

As with any measure of concentration, Herfindahl can be calculated on long and short sides separately. But for the purpose of intuition let’s interpret it in the context of long-only portfolios. For example, a manager has an equal-weighted portfolio of 50 positions, each taking up 2% of portfolio AUM. The concentration Herfindahl is then (0.02^2)*50 = 2%. From this it’s easy to see that for equal-weighted portfolios, the concentration Herfindahl equals the average position size. Since this post is all based on public data, percent of total filing value is same as percent of exposure. When it comes to levered portfolios, their percent of portfolio sizes would be different, but from a concentration standpoint the average percent of exposure size is the same, i.e. they could have the same Herfindahl.

For portfolios that are not equally weighted, the concentration Herfindahl essentially captures a *weighted average* position size rather than the *average* position size. Suppose manager A has 5 positions with 10% position size and 10 positions with 5%. The concentration Herfindahl for manager A is 7.5% but the average position size is 6.67%.

Published on

October 14, 2015

related Posts

> All Posts