Obtaining good return data for infrastructure investing is hard. We have seen some investors avoid the asset class altogether, such as Norway’s Global Government Pension Fund, partly due to this issue. Risk-adjusting return data is harder still, as we shall see. Nevertheless, modern portfolio theory – and asset allocations by chief investment officers the world over – require us to estimate infrastructure risks, on a standalone basis and for their impact on wider investment portfolios.
We find that although absolute post-fee returns appear lower for unlisted infrastructure funds, once those returns are risk-adjusted – through widely-used approaches such as Shape Ratios, CAPM or Fama-French factors – unlisted infrastructure evidences performance superior to listed infrastructure and the wider equity market. However, there are several issues that standard risk measures do not take into account, such as how to deal with appraisal valuations and illiquidity.
Portfolio theory looks at periodic asset price returns, volatilities and correlations to estimate expected return distributions of various possible portfolios. This allows a risk and return profile to be matched with investors’ requirements and risk appetite. For listed assets, with regular prices from liquid secondary markets, this approach is relatively straightforward. It becomes far harder for unlisted, irregularly traded assets, such as infrastructure fund allocations. It is important, though, both for investors and for the infrastructure sector as a whole in arguing for growing capital commitment.
PAST IS NOT PROLOGUE
We start by reviewing what we know about past returns achieved from investing in infrastructure. Of course, the past is not necessarily a good guide to the future; and for a young asset class such as infrastructure, there are few long-term return time series available. This is particularly true for unlisted fund returns, where there were relatively few funds prior to 2000. We have a sample of 66 unlisted funds, with an aggregate fund value of approximately $100 billion, for which periodic NAVs, fees and distributions are available, referred to as the Unlisted Infrastructure series. To compare these to listed infrastructure – including listed utilities, transportation infrastructure companies, listed yieldcos, etc. – we use aggregated data collated by the Global Listed Infrastructure Organisation, referred to as Listed Infrastructure.
A further issue for unlisted infrastructure is the lack of complete performance data for all funds, due to the private nature of management agreements with limited partner investors, hence we are dealing with a sample. Internal rates of return are relatively widely available for funds’ performance and are collated by a number of data providers. Issues with IRR have been widely documented. We have argued for alternative return measures, originally developed to assess private equity fund performance, including Modified Internal Rate of Return and Public Market Equivalent.
While these measures can be shown to be more representative than IRR – and are useful in an overall assessment of asset class returns – they are insufficient for portfolio risk-return assessment.
Portfolio management models require a time series of periodic returns for each asset class that can be invested in. For listed assets, periodic returns are equal to the proportionate change in market price plus any dividends received. Unlisted funds do not, by definition, have listed prices and secondary market trades are rare. Funds do have periodically reported NAVs, typically every three months. As a result, quarterly returns can be calculated as for a listed asset, but substituting “change in NAV” for “change in market price”. Market prices and NAVs are different conceptually, though, with NAVs being appraisal values based on the long-term expected value of a fund’s assets. The asset values are re-assessed regularly, independently verified and subject to being written down for value impairment. However, this is not the same as being subject to day-to-day market volatility in the manner of listed equities, as discussed further below.
Accepting for the moment the comparability of listed and unlisted returns, we can observe cumulative total returns for unlisted and listed infrastructure, and compare them with the returns from listed equities (MSCI World index) and investing in a risk-free asset. These return indices are illustrated in Figure 1.
These time series show $100 invested in the MSCI World index on June 2003 would have increased to $270 12 years later. This is very similar to the $268 resulting from investing in the Unlisted Infrastructure fund sample. Both are far higher than the risk-free total return index of $117, but significantly less than the $394 that would have resulted from investing in Listed Infrastructure. For a risk-neutral investor, this suggests Listed Infrastructure was the best asset class to have invested in.
However, Figure 1 also clearly shows far greater swings in value for the MSCI World and Listed Infrastructure indices than for Unlisted Infrastructure, supported by measuring their volatility through taking the annualised standard deviation of quarterly returns, shown in Table 1. Listed Infrastructure’s standard deviation was 14.7 percent, close to the 15.8 percent of MSCI World. Volatility of Unlisted Infrastructure was far lower at 8 percent, although this is based on NAV changes rather than market prices.
Using these standard deviations, we calculate the assets’ Sharpe Ratios (equal to excess returns above the risk-free rate, divided by standard deviation of returns) and based on this risk-adjusted performance measure, Unlisted Infrastructure shows a higher risk-adjusted return than both Listed Infrastructure and MSCI World, seen in Table 1.
There are widely used alternative measures of risk-adjusted returns to the Sharpe Ratio. These typically look at the co-movement of an asset’s returns to the overall equity market and potentially other asset pricing factors. The rationale for these approaches is that the right measure of risk for well-diversified investors is not total volatility (measured by standard deviation), but the extent to which an asset affects the investor’s overall portfolio.
One of the original measures of this so-called systemic or non-diversifiable risk is the Capital Asset Pricing Model, which regresses an asset’s return (above the risk-free benchmark) against that of the overall market. Despite widespread theoretical and empirical criticism, CAPM remains a starting point for much risk-adjusted return analysis. For any asset, CAPM analysis gives both the co-movement of the asset’s return with the market (its CAPM beta) as well as any return outperformance against the CAPM expected return based on this co-movement (its CAPM alpha). Using our 2003-15 data, we estimate CAPM alphas and betas for Unlisted and Listed Infrastructure in Table 2.
Unlisted infrastructure has a low CAPM beta of 0.17, i.e. the expected change in unlisted returns is relatively unrelated to movements in the wider equity market. Hence, under CAPM it would have a low expected return, which results in a high level of risk-adjusted outperformance, measured by its CAPM alpha of 5.8 percent. In other words, Unlisted Infrastructure appears to offer almost 6 percent better returns annually than would be expected for its low level of market risk. Again, remember that “risk” here is measured by variation in NAV.
In contrast, Listed Infrastructure has a relatively high CAPM beta of 0.76, so although less systemically risky than the overall market (which has a beta of 1 by definition), it nevertheless has a higher CAPM expected return than Unlisted Infrastructure. As a result, its risk-adjusted outperformance is lower, with a CAPM alpha of 4.9 percent. This still represents significantly higher returns than the market as a whole and those predicted by CAPM.
THE FIVE-FACTOR ALTERNATIVE
Given the criticism of CAPM, particularly empirical observation that factors other than market co-movement appear to affect returns, researchers have developed many alternative asset pricing approaches. We limit ourselves here to considering one recent model, the Fama-French Five Factor Model, published in 2014 and itself an updated version of their 1993 Three Factor model. In this model, equity market co-movement is still considered, but only as one of five factors affecting expected returns. The other four factors are size (where small companies have higher returns); value versus growth (where value stocks have a premium); profitability; and levels of investment (both of which lead to higher return expectations).
We do not argue that this is necessarily a better model compared to competing alternatives, but we believe it may be illuminating to compare listed and unlisted infrastructure returns using a more up-to-date approach than CAPM, which was, after all, developed in the 1960s. Under this model, the results of which are shown in Table 3, there is continued risk-adjusted outperformance of Unlisted Infrastructure versus Listed Infrastructure and the wider equity markets. Unlisted Infrastructure continues to show a high annualised FF5 alpha of 4.8 percent, and whilst Listed Infrastructure’s alpha remains positive (i.e. implies outperformance against the market and its asset characteristics), FF5 alpha is 1.3 percent, significantly less than its CAPM alpha.
The Five-Factor model also shows an asset class’s exposure to ‘asset factors’ that are assumed to affect returns (in addition to sensitivity to overall equity market risk). The Listed Infrastructure factor coefficients, shown in Table 3, are as expected based on infrastructure’s widely accepted asset characteristics: exhibiting loadings towards large companies, value (rather than growth) stocks and having high levels of profitability and investment intensity. The FF5 market beta is 0.88, showing a high correlation with overall equity returns, slightly higher than that seen with CAPM. The model explains 83 percent of return variation (the Adjusted R2 value), a high level of explanatory power for models of this type.
In contrast, the Five-Factor model is less effective in explaining the variation of Unlisted Infrastructure returns, with an Adjusted R2 of 11 percent, less than the explanatory power of the simpler CAPM approach. Hence, one should be more circumspect in interpreting the model’s parameters for Unlisted Infrastructure. The FF5 market beta is very low (at 0.11), suggesting an insignificant correlation between unlisted asset values and the wider equity market. The other factor loadings suggest return correlations with small, value-oriented and profitable stocks, perhaps in line with expected infrastructure fund investments, but also correlation with low investment intensity stocks, which is less expected.
To summarise our findings on risk-adjusted returns: using each of Sharpe Ratio, CAPM and Five Factors, both Unlisted and Listed Infrastructure outperform the overall equity market, with Unlisted Infrastructure showing the best risk-adjusted performance. This is despite Unlisted Infrastructure having lower absolute returns than either Listed Infrastructure or MCSI World.
However, we note two significant issues in confirming these findings: the effect of NAV appraisal value on apparent return volatility and adjusting expected returns for illiquidity.
Firstly, it is important to reflect on the use of Unlisted Infrastructure’s periodic NAVs to assess return volatility. If unlisted funds invested solely in listed assets, they could report daily NAVs which would reflect the market volatility of the underlying assets. However, unlisted infrastructure funds invest in illiquid, long-term real assets, often with uniquely specific characteristics. As a result, market prices are not directly available for funds to report periodic values and instead they provide appraisal valuations, which reflect estimates of the long-term value based on then current economic and market conditions. It is widely viewed that NAVs do not fully reflect market volatility, although this is hard to prove given the unique nature of the underlying assets.
If NAVs systematically understate the true change in market value of funds’ assets, on an equivalent basis to the value of identical listed assets, then we cannot accurately compare market-based and NAV-based return volatilities, as NAV volatility will be understated (albeit to an unknown extent). It is possible to use statistical filtering techniques to estimate this effect – which has been undertaken for real estate investment vehicles, but not for infrastructure funds to our knowledge – and this may be a fruitful area for further research.
There is some evidence of NAV appraisal value having a “smoothing” effect on return volatility, based on the auto-correlation of returns. If returns are independent between consecutive periods, there should be little or no correlation between the returns from one period to another. If there is positive autocorrelation, this suggests the previous period’s return affects the next period, which could be the case if appraisal based methods smooth NAV returns. Looking at autocorrelations in our data, we note Unlisted Infrastructure exhibits a positive autocorrelation of +0.26, although we also see Listed Infrastructure has autocorrelation of +0.18. This is not conclusive, but it does suggest further work is needed to assess this effect, as it could significantly distort the conclusion of higher risk-adjusted returns for Unlisted Infrastructure.
A similar statistical suggestion of return smoothing can be seen from the measure of skewness in the return indices. Both Listed Infrastructure and MSCI World returns exhibit negative skew, i.e. their probability distributions have a long tail of rare but large negative outcomes. In contrast, Unlisted Infrastructure exhibits positive skew, where these downwards moves are far rarer.
The second issue is illiquidity: it is self-evident that Unlisted Infrastructure is an illiquid asset class, with investors unable to exit their investment during the life of the fund and with a limited secondary market for trading fund investments. There is significant academic debate over the extent of an illiquidity discount for unlisted asset returns. However, the approach outlined by Prof. Andrew Ang (of Colombia University and BlackRock) implies potential illiquidity discounts of 0.5-6 percent per annum. Although a wide range, at the upper end this would entirely eliminate the observed risk-adjusted outperformance of Unlisted Infrastructure (as CAPM and FF5 alphas are less than 6 percent). Investors will have differing investment horizons and appetite for illiquidity risks. Although further theoretical and empirical research is required to examine these preferences for infrastructure investment, we note this will be an important consideration for each investor to address, and potentially come to differing decisions. For example, at a recent investor conference, infrastructure was discussed with two UK local authority pension funds, with one concluding listed infrastructure offered greater flexibility regarding changing asset allocations over time, whilst the other preferred unlisted for its targeted investment strategy and low-risk returns.
Like these two local authority funds, every investor needs to assess the relative merits of Unlisted and Listed Infrastructure against other asset classes. The low-risk nature of infrastructure is one of its key selling points and hence it is important investors assess risk-adjusted returns. The good news for the sector is that both listed and unlisted risk-adjusted returns appear attractive for the 2003-2015 period we have examined. And the superior performance, across a range of measures, of unlisted funds seems in line with their rapid growth in capital raised during this time. However, we would highlight the potential issues raised here and encourage investors, fund managers and infrastructure associations to look further at these areas. ?