Defining the precise scale advantages for DB pension funds

Most people working in the institutional asset management space have an intuitive understanding that the size of the institution, measured by total assets under management (AUM), has an impact on performance – that bigger funds tend to perform slightly better. On the other hand, there are plenty of stories of successful hedge funds that got too large and lost their way, unable to continue delivering on past success due to their size. So which is it? Do larger institutional investors outperform their smaller kin, or is AUM the proverbial millstone in terms of performance?

A quick and easy answer to the question of how size impacts on performance can be found by looking at net value added, defined as the difference between a fund’s net return and its policy return.

Size matters

We run a simple exercise using CEM Benchmarking’s database of over 1,100 large institutional investors, which dates from 1992 and includes, for the most part, defined benefit (DB) pension funds and DB pension asset managers. We segment the sample into size cohorts and look for a performance differential. The analysis clearly suggests that size does matter. Net of investment expenses, investors with less than $1bn (€900m) in AUM produced a net value added of -3bps, whereas investors with more than $10bn AUM produced a net value added of 29bps.

However, diseconomies of scale could hide in the result. For example, coverage of European funds in the CEM database skews towards the very large (above $50bn, on average), whereas in the US and Canada it skews towards smaller funds. Both markets could exhibit diseconomies of scale, but if European investors had an environment more conducive to generating value added, then simple statistics would show economies of scale that are not actually there.

To mitigate against these issues, we use a simple but robust regression analysis that takes into account the elements we know matter in driving value added. First, size as measured by the log of AUM. Second, how actively managed a fund is. Third, the extent to which a fund is internally managed. Finally, geography, which is incorporated into the analysis as a region-specific constant that determines the baseline expectation of value added, all things being equal.

Regression techniques allow turning on and off fund characteristics such as cost, returns, benchmarks and allocations all by asset class. This is important because, for example, size might not be important gross of costs, but net of costs it might be critical. Since cost is the only thing that changes in this example, we could conclude that cost savings drives the observed economies of scale.

When a regression is performed using value added gross of costs, we observe that the main drivers of gross value added are how actively managed a fund is and its size. The result suggests that investors that are more active obtain more gross value added, which makes sense, but it gives little indication of what the role of size might be.

By removing from gross value added all the contributions from private equity and unlisted real estate, the impact of size on gross value added disappears. This suggests that the size dependence in gross value added is caused by private equity and unlisted real estate. Why?

The answer is that larger funds allocate more of their AUM to private equity and unlisted real estate, and private equity in particular has historically been a great source of gross value added. Indeed, private equity produces zero or slightly negative net value added, but that’s after netting an average of around 350bps of cost on a net asset value basis. Larger funds have more unlisted assets, and unlisted assets tend to have high gross value added.

When private equity and unlisted real estate are excluded, we observed no size dependence in gross value added. A further regression, based on net value added and including the impact of costs (while still excluding private equity and unlisted real estate) suggests that size does matter. For each 10-fold increase in size, net value added improves by 11bps. If size was immaterial before costs, and material net of costs, then cost savings must be the source of the economies of scale.

Economies of scale in public markets are something that can be observed in other ways as well. For example, when looking at staffing, which is great proxy for internal costs, we have continuously found strong economies of scale over the years. As AUM increases, the number of full-time equivalent (FTE) staff required for internally managed active public equities programmes increases, but slower than the AUM itself. Indeed, for each doubling of AUM, staffing for internal active public equities increases by only 65%. The same behaviour is found for other public-market asset classes like fixed income, with stronger economies of scale for passively managed programmes than active ones.

Internal costs are only a part of the picture. We also see economies of scale in costs paid to third-party money managers. For example, our data indicates that an actively managed $200m US small-cap stock portfolio costs about 55bps, but the cost is only 40bps for a $2bn portfolio.

In a fourth regression, we look at net value added but now with the addition of private equity and unlisted real estate. This generates two interesting observations. First, the impact of size on net value added grows from 11bps to 20bps per every 10-fold increase in AUM. Second, internalisation becomes important. Since internalisation was statistically unimportant excluding private equity and unlisted real estate, but statistically important with their addition, it means that the benefit of internalisation to net value added is due to the allocation to those alternative asset classes.

We can therefore say that lower-cost illiquid asset portfolios consistently outperform. Running an internally managed private equity or unlisted real estate portfolio can cost 50bps, whereas fund investments cost two to six times as much. Fund of funds layer on another set of fees to the tune of 100 to 200bps. While the impact of internalisation in public markets is much smaller, in private markets it is very prominent.

Slight advantage

However, internalising private equity and to a lesser degree unlisted real estate requires scale. The smallest fund in the CEM database with any internally managed private equity has nearly $20bn in AUM, while the average investor with internal private equity has $150bn of AUM.

The data speaks with a clear voice, saying that larger funds have advantages that smaller funds do not. The advantage is largely driven by costs and the efficiencies that scale can provide. We stress that the advantages are slight and will usually be dwarfed by the year-to-year gyrations of markets. Large funds are not always going to outperform year after year, but over time size matters, and it matters because costs matter.

Alexander Beath is a senior researcher at CEM Benchmarking 

*Regression model details: 

1. Methods: multivariate regression models of GVA and NVA, both with and without contributions from PE and RE, are carried out for each of the 29 years of data spanning 1992-2020. For an investor x in year y, we have GVA x,y / NVA x,y :

GVA / NVAx, y = A y x (% Active) x,y + B y x (% Internal) x,y + C y x (log 10 AUM) x,y + D y x (Country logit variable) + ϵx,y

with A y the co-efficient for ‘Percent Active’, B y the co-efficient for ‘Percent Internal’, C y the size co-efficient measuring economies of scale, D y a region dependant variable (regions used are U.S., Canada, Euro region, and Other), and ϵx,y residuals the sum of squares of which are minimized to estimate the set Ay,By,Cy,Dy for year y with standard errors δAy,δBy,δCy,δDy. For each model, the 29 year specific variables Ay,By,Cy,Dy are meta-analyzed via maximum likelihood estimation to yield A,B,C above (region specific constants D are available upon

2. Filtering: Not all GVA/NVA data provided to CEM Benchmarking are of equal quality, either because of poor benchmark construction at an asset class level or because the policy portfolio is not a good representation of the actual portfolio. To ensure that only the best GVA/NVA data is used, we run a regression of net return vs. policy return, selecting from the set: (i) investors with 5+ years of data to ensure we can do a regression at all, and (ii) investors for which we achieve a beta between 0.75 and 1.25, and (iii) investors for which we achieve a correlation in excess of 0.92 (the 10th percentile of correlations). Mutliple filters in terms of years, beta’s, and correlations were considered in order to ensure that the results presented above have both maximum accuracy and precision. Of the total 8,692 investor/year observations, 7,175 are used. The average regression alpha, beta and correlation of the data set used is 0.20%, 99.3%, and 98.1% respectively.