applying a risk-based smart beta approach to fixed income investing · pdf file ·...

Applying a risk-based smart beta approach to fixed income investing

A theoretical and empirical case for a smart beta approach to investing in fixed income

Claudio Ferrarese, Quantitative Analyst, Fixed Income

Peter Khan, Portfolio Manager, Fixed Income

David Buckle, Head of Quantitative Research, Fixed Income

Smart beta approaches are well documented in equity markets and in asset allocation, but rarely discussed and implemented in the context of fixed income.

We argue, however, that the intuitive and theoretical case for such indices is arguably strongest in fixed income, and set out to show this empirically in high yield.

Our analysis reveals that smart beta approaches might indeed work well in fixed income, generating superior returns compared to more conventional strategies particularly when based on equal credit risk contribution (ECRC) benchmarking.

April 2015

Executive Summary

‘Smart beta’ is typically regarded as a better form of exposure to beta than Market Capitalisation Weighted (MCW) indices. Smart beta is widely researched and applied in equity markets and asset allocation but rarely discussed and implemented in the context of fixed income.

In this paper we argue that the intuitive and theoretical case for such indices is possibly strongest in fixed income. An MCW approach is simply not the most efficient way to construct a fixed income portfolio, as there are few instances where it is sound to link the amount invested in a company to how much debt it has issued.

We set out to show this empirically in high yield, using a range of smart beta alternatives. In line with previous studies on the other asset classes, we find indeed that smart beta outperforms MCW-based high yield portfolios in our study.

Of all smart beta alternatives considered in this paper, an equal credit risk contribution approach based on regional weights appears the most promising. We apply this form of smart beta to the Bank of America global high yield corporate debt universe. This approach not only generates better optimal risk-adjusted returns, but also allows for any breakdown in future regional correlation patterns.

Having established the theoretical and empirical arguments in favour of smart beta in fixed income, we then assess some of the practical implications of managing a portfolio around such a benchmark. These include liquidity issues, transparency risk and additional scrutiny necessary in the security selection process. We show how these practical obstacles can be addressed through careful portfolio construction.

Importantly, due to the practical considerations around the implementation of smart beta approaches in fixed income, the role of the portfolio manager is central for funds based on managed beta exposure in fixed income.

Given the nature of the high yield debt market, this is an asset class that benefits from active management. Such active management can enhance portfolio returns in capturing alpha alongside the more efficient allocation of portfolio risk.

3

Contents

Introduction ...................................................................................................................................... 4

Part one: The rationale for smart beta ........................................................................................... 5

What is smart beta? ...................................................................................................................... 5

The case for smart beta ................................................................................................................ 6

Why does smart beta work? .......................................................................................................... 6

Smart beta is not a panacea ......................................................................................................... 7

Part two: Smart beta in fixed income ............................................................................................. 8

Is smart beta justifiable in fixed income? ....................................................................................... 8

Typical smart beta obstacles in high yield ..................................................................................... 8

Part three: Composing ECRC for global high yield .................................................................... 10

Which data sets to select? .......................................................................................................... 10

Equalising the risk contribution .................................................................................................... 11

Superior returns validate smart beta approach ........................................................................... 11

How to implement ECRC in practice? ......................................................................................... 15

Controlling systematic risk exposure in high yield smart beta ..................................................... 16

Practical implications of managing an ECRC based portfolio ...................................................... 17

Conclusion ..................................................................................................................................... 19

Key references ............................................................................................................................... 20

Authors:

Claudio Ferrarese Quantitative Analyst, Fixed Income [email protected]

Peter Khan Portfolio Manager Fixed Income Peter [email protected]

David Buckle Head of Quantitative Research, Fixed Income [email protected]

4

Introduction

Smart beta approaches are well documented in equity markets and in asset allocation, but rarely discussed and implemented in the context of fixed income. We argue, however, that the intuitive and theoretical case for such indices is arguably strongest in fixed income, and set out to show this empirically in high yield.

Part 1 of this paper briefly explains the smart beta premise, what its benefits are and why it is not a panacea for all situations. Part 2 sets out the case for smart beta in fixed income, and how the theoretical obstacles in constructing smart beta benchmarks in fixed income can be overcome.

In Part 3 we compose a smart beta fixed income approach for a global high yield portfolio, based on equal risk contribution, and compare the performance of such a portfolio against other smart beta alternatives as well as a conventional MCW weighted portfolio. We also look at the practical implementation of such a strategy including liquidity issues, transparency risk and the additional scrutiny necessary in the security selection process.

We conclude that smart beta approaches do indeed work well in fixed income, generating superior returns compared to conventional theories particularly when based on equal risk contribution (ERC) benchmarking.

It almost goes without saying that the role of the portfolio manager is central for funds based on an ERC approach, but we also highlight how combining an ERC-based allocation with active management means alpha can be captured alongside the more efficient allocation of portfolio risk.

5

Part one: The rationale for smart beta

Explaining the theoretical background to smart beta investing, this section illustrates different smart beta approaches and sets out the shortcomings of conventional methods that cause the empirical outperformance of alternative indexation. It also flags the limitations of smart beta strategies.

What is smart beta?

‘Smart beta’ has become a buzzword in the investment community, but its roots go far back. The terms beta and alpha come from index models1 with beta representing the non-diversifiable component of a portfolio measured against the market portfolio and alpha being the residual return. Seeking beta became synonymous with harvesting risk premium via market exposure, predominantly using market capitalisation weighted (MCW) indices.

The Capital Asset Pricing Model (CAPM) as developed by Sharpe and others referenced one single market factor while Fama and French showed that multifactor models explain about 90% of the market volatility. Therefore the definition of beta has been extended to indicate a generic exposure to market risk factors.

‘Smart beta’ is typically regarded as the better exposure to beta than MCW. Its application is well documented in equity markets2 and in asset allocation3. A commonly accepted definition of smart beta is simply portfolio construction not weighted by market capitalisation. Smart beta indexation can be either fundamental/factor based, or risk-based.

Fundamental/factor based indexation

Arnott et al. (2005) introduced fundamental weighting schemes, weighting stocks according to their fundamental scoring using metrics like earnings, dividends, equity book value or sales, and originated the FTSE RAFI fundamental-weighted index family.

Risk-based indexation

Risk-based indices differ in that they use no information other than the covariance matrix. Typical of risk-based weighting schemes are minimum variance indices, which aim to minimize ex-post return volatility of an equity portfolio.

The Risk Parity (RP) approach is another example of risk-based indexation. This strategy aims at equalising the stand alone risk of the various portfolio components. The term ‘risk parity’, now standard industry jargon, was first coined by Qian in 2005, although the approach is older. An early application is Bridgewater Associates’ ‘All Weather’ asset allocation fund launched in 1996.

Equal Risk Contribution (ERC), Maillard et al. (2010), is a generalisation of risk parity. Rather than equalising risk, it equalises the marginal risk contributions of the various portfolio components, thereby accounting for correlations between components. It can be shown that risk parity is a special case of ERC when the security return correlations are equal.

DeMiguel et al. (2009) found that the simple “1/n” approach outperforms many sophisticated techniques, including MCW indices, arguably due to the lack of parameter estimation error. This too is considered a smart beta approach.

1SeeforexampleSharpe19612Demeyetal.20103Maillardetal.2008

6

The case for smart beta

There are a number of benefits to MCW indices, both practical and theoretical, but empirically they tend to underperform due to a tendency towards portfolio concentration.

The two-fund separation theorem4 states that investors should hold only two assets: a single risky portfolio and the risk free rate. The CAPM5 proved that this risky portfolio was the MCW index. The CAPM assumed that investors use the Markowitz framework, which means that the MCW index is efficient in a risk-adjusted sense. This theoretical support for the MCW approach is compelling, and not matched for smart beta approaches.

Furthermore, MCW is simple. No complicated rules are needed to construct the index. It is relatively straightforward to anticipate future index composition. Moreover, theoretically MCW indices have no turnover because the index weights move with market prices. MCW indices also tend to favour liquid securities where there is more capacity to invest. Therefore MCW indices have seen widespread adoption.

However, empirically, smart beta approaches often outperform them. For example, Grinold (1992) tested five major market MCW indices for market statistical efficiency, revealing that four out of five indices were inefficient and that value-based investing has been successful in all five markets. Recently Clare et al. (2013) have shown that equity indices constructed randomly (by ‘monkeys’) would have produced higher risk-adjusted returns than MCW indices over the last 40 years.

ERC and RP became a popular alternatives to market cap weighting, showing better risk-adjusted returns compared to traditional weighting schemes in a range of empirical studies6.

A theoretical shortcoming of MCW is that it does not account for estimation risk in the implicit expected return estimates embedded in it. As a consequence the classic approach usually brings a lack of diversification. MCW indices can be quite concentrated in few names, sectors or geographical regions, explaining the inferior ex-post risk adjusted returns. In contrast, smart beta approaches tend to be more diversified, thereby reducing the impact of estimation error.

Why does smart beta work?

As noted above, in empirical studies smart beta approaches have generated superior performance. They appear to do so by exploiting well known factor exposures such as size, value, volatility and momentum that have historically been remunerated by the market. Seemingly smart beta and factor/style exposure tilting behave similarly; the latter having been used extensively by quantitative equity managers for many years. Similarly risk based strategies such as minimum volatility allocate to stocks with a low volatility factor bias. Low volatility is also well known to provide superior risk adjusted performance owing to the yield chasing bias of investors, lottery preference, and leverage constraints. Yield chasing behaviour has probably been exacerbated by years of ever decreasing interest rates forcing investors who target a fixed expected return along the efficient frontier to a higher level of risk.

Smart beta and factor tilting are not however identical. It has been shown that even after controlling for these factor biases, smart beta still generates excess returns. Recently Amenc et al. (2013) show that existing smart beta methodologies have inherent systematic and specific risks that are neither documented nor explicitly controlled by their promoters. They show that those factor exposures can be largely controlled and after doing so continue to outperform. Importantly from an implementation perspective, they show that these smart beta strategies tend to survive the incorporation of restrictions on liquidity.

4Tobin19585Sharpe19646SeeforexampleMaillardetal.(2008and2018),andtheabovementionedBridgewaterAssociates’AllWeather”portfolio

7

Smart beta is not a panacea

In addition to the lack of theoretical support, there are several other significant problems when using a smart beta approach. Unlike MCW, smart beta indices can have higher turnover. For this reason, transaction costs are an important consideration in smart beta, while inconsequential in MCW equity indices. Methods of reducing turnover can be adopted, e.g. varying the rebalancing frequency, but they introduce new parameters into the process.

When using company fundamentals to construct portfolios we are implicitly estimating expected returns. Estimation risk of this is known to be high. For example, fundamental weighting is affected by data window choice. Accounting data are typically backward looking and estimates are rarely available for a full universe for a lengthy history. Past accounting ratios might introduce additional errors and respond badly to changes in economic regimes. Companies that were successful or scored high on, say, cash flow metrics might fail to do so in the future due to changing regulation or industry cycles. Unbiased forecasts of future accounting ratios would be an ideal substitute for fundamental weighting but unfortunately they are rarely available for the whole universe. Therefore, in the same way that an average historic return is a poor estimate of future return, so an out of sample experience of a fundamental index can differ different from in-sample expectation.

Of course the MCW also has embedded expected returns estimates, but it uses the market portfolio to imply them. This lack of parameter estimation is a significant benefit of the MCW. Even so, the concentration shortcoming of MCW is due to this implicit estimation.

Risk based methodologies have a lower parameter estimation risk (as long as the covariance matrix is robust) than those incorporating expected return estimates. Nevertheless they still use statistical estimation of covariance and thus have some estimation risk. In particular the risk based approaches can be very sensitive when correlations are high, exhibiting weight instability especially when the volatilities are similar. Some of these issues can be mitigated with techniques like shrinkage of the covariance matrix, or good choice of prior in a Bayesian setting. Another issue with ERC is deciding how granularly to allocate the risk. At a single security level it is not always clear how to treat company splits or mergers, but should we then equalise risk over sectors, or geographies, or some other partition?

In general, any methodology which tries to maximise diversification, like equally –weighted, minimum variance, RP or ERC, will increase the weight to less liquid stocks/issuers, so loading up more on a liquidity premium – a systemic risk in its own right. Conversely MCW indices benefit from a natural bias toward more liquid securities.

8

Part two: Smart beta in fixed income

In theory, an MCW approach is simply not the most efficient way to construct a fixed income portfolio, as apart from the issues relating to turnover and the assumption of normality, there are few instances where it is sound to link the amount invested in a company to how much debt it has issued. This section deals with typical obstacles to high yield smart beta indices, introducing an alternative high yield smart beta approach based on equal credit risk contribution at a wider geographical level.

Is smart beta justifiable in fixed income?

Smart beta is frequently applied to equities, or across asset allocation, yet the theoretical case to move away from MCW indices is arguably strongest in fixed income:

For one thing bonds mature and are called, and newly issued debt surfaces virtually continuously, violating the zero turnover features of equity MCW indices.

Furthermore, in bond indices the weight of each issuer in the traditional MCW index is

proportional to how much debt has been issued. But Modigliani–Miller’s seminal capital structure irrelevance principle suggests the value of a firm is not dependent on how it is financed. It is therefore not sound to link the amount invested in a company to how much debt it has issued. (Even intuitively it seems unsafe to invest most capital in the most indebted companies or nations.) This rather aggressive investment philosophy would represent a good strategy only if the market were efficient in the way it prices the risk of each additional unit of leverage demanded by borrowers. Yet investors do indeed tend to favour deeply indebted companies. This so called lottery-seeker preference bias was shown empirically by Ilmanen (2011), with CCC rated corporate bonds having generally delivered poorer excess returns in the last 20 years than BB and B peers. The issue is recognised by index providers who often constrain indices to limit the maximum exposure to a single issuer, sector and sometimes geographical area. This might itself be considered a version of smart beta.

Additionally, the CAPM assumes that security returns are normally distributed. To a degree

even equities violate this assumption, though perhaps tolerably so. But high yield bonds exhibit large enough levels of both skew7 and kurtosis8 that the normality assumption is poor. This implies MCW will underestimate tail risk and therefore create a too concentrated portfolio.

Typical smart beta obstacles in high yield

There are a number of obstacles to creating smart beta indices in high yield, which need to be addressed to create a viable smart beta alternative.

First, in global high yield indices only a fraction of the issuers are public companies and so accounting data or estimates are rarely available. Therefore fundamental based smart beta is problematic for high yield.

We can see this in the two fundamental-weighted fixed income indices created by Research Affiliates (RAFI): RAFI® Bonds US High Yield Master and RAFI® Bonds US Investment Grade Master, Shepherd, S. (2011).

Second, unlike equities which trade on exchanges, price discovery of high yield debt securities is more challenging. Often only indicative prices are available, and many bonds can go days without

7Ameasureoftheasymmetryofaprobabilitydistribution

8Ameasureofthe‘peakedness’ofadistribution,interpretedtheaswidthofthepeak,thetailweight,orthelackofshouldersinbetweenthepeakandtails

9

trading. Therefore model prices are commonly used to create historic time series, and these may well differ from actual tradable prices. This illiquidity of small issuers can make strategies that look good on paper un-investable in practice.

A third problem is issuance size. Pure equally-weighted strategies will give the same weight to a company that issues only $200 million as to a large issuer with several billion dollars of debt outstanding. Clearly such strategies can have scalability problems.

Fourth, smart beta for high yield cannot be applied at a single issue level, as we would need an estimate of issue risk, and correlation. For equities we can use price history to do that, but for newly issued bonds, there is no such history.

To circumvent this we could use a market implied risk measure like Duration Times Spread (DTS). We term the smart beta approach based on this as ‘DTS parity’. Unfortunately, unlike equities, one issuer could have issued many different bonds. DTS parity applied on this basis would imply big allocations to issuers of many bonds.

Instead it might be tempting to deploy an ERC approach at an issuer level. We’d simply need the notion of the risk of an issuer, possibly again leaning on DTS and we could make some assumption on correlation, such as average same pairwise correlation. However such an approach would place a substantial weight on low risk and small issuers. In reality that would not be plausible for even moderate-size investors.

Instead, we prefer to apply ERC at a coarser geographical level. This combines the advantages of MCW indices in a geography, but reduces their greatest disadvantage by achieving better diversification. Furthermore, aggregating high yield returns at the geographic level somewhat diminishes the skew and kurtosis prevalent in single securities.

It is common practice to decompose the debt risk premium into credit and interest rate components (for our purpose any liquidity premium can be thought of as part of the credit premium). The following analysis aims to identify the optimal mix to extract credit premia, thus the name Equal Credit Risk Contribution (ECRC). Isolating exposure to the interest rate factor, we look at excess returns over government bonds to measure the performance of our indices, leaving the returns due to pure interest rate duration exposure out of our study.

10

Part three: Composing ECRC for global high yield

Our empirical study on the global high yield universe shows that smart beta outperforms MCW for global high yield. Some modified MCW indices can be thought of as a form of smart beta but we find better-performing alternatives, including the ECRC, Risk Parity, DTS parity and the simple 1/N approach. The similarities in performance between these alternative indexation techniques can be explained fully by a combination of similar volatilities and high correlation between regions during the sample period. As these volatilities and correlations may shift, we prefer ERC because it gives a better guarantee for capturing such possible future shifts. Practical implications of managing a portfolio around such a benchmark, including liquidity issues, transparency risk and security selection, are addressed and highlight the key role of the portfolio manager for this kind of product.

Which data sets to select?

ECRC provides an alternative way to construct broad high yield market exposure, improving on traditional benchmarks. A popular MCW benchmark in this space is BofA Merrill Lynch Global High Yield Constrained Index. We use monthly excess returns data from 1998; analytics like spread and duration are calculated by the index provider. Sector distribution and returns are also taken directly from the index provider.

Six regions that make up the universe have been considered: US, Europe, UK, CEEMEA (Central and Eastern Europe, the Middle East and Africa), Latin America and Asia. We have used historical data from 1998 to September 2014, but we believe that all geographical regions would have been fully investable only from 2004. This limitation is due to the small number of issues from the Asian, CEEMEA and Latin American segments prior to 2004. From 2007 the number of issues in those buckets has increased substantially and the amount of bonds outstanding was sufficient to launch funds investing in these countries. Chart 1 depicts the market size by region for the last 15 years. If we set the threshold for market size at $10 billion, it is only from 2004 that this is achieved by almost all sectors. Today all six geographical regions are issuing in excess of $100 billion annually, making this approach not only academically interesting but also fully investable.

In this analysis other alternative choices of subdivisions like single countries, industrial sectors and rating distribution have also been considered. The preference for geographical regions is due to the dynamics of the specific global high yield universe, which clearly shows an imbalanced distribution under this metric. This was therefore a critical dimension of the problem and the most promising source to search for better diversification. For other datasets it could be helpful to consider industrial sector distribution. A combination of different subdivisions - e.g. both sector and geographical distributions - could theoretically also be attractive but would increase the dimension of the problem and potentially create sample size and error estimation issues in the variance-covariance (VCV) estimation, which would then not be robust.

Chart 1: Market cap by geographical split in log scale

Source: Fidelity Worldwide Investment, BofA Merrill Lynch. Data as of 30th September 2014.

100

1,000

10,000

100,000

1,000,000

'99 '00 '01 '02 '03 '04 '05 '06 '07 '08 '09 '10 '11 '12 '13 '14

EU UK

Asia CEEMEA

LATAM US

US$, log scale

11

Equalising the risk contribution

Maillard et al. (2008) introduced ERC from the idea that each asset should contribute in the same way to the volatility of the overall portfolio to create a more balanced mix and maximise diversification. We present a quick look at the mathematics of their approach here, but recommend the original paper for a deeper explanation of the technique. Using the Euler’s decomposition we can rewrite the total risk of a portfolio as the sum of risk contributions:

√ Ω ∑ ∑ (1)

where σ(w) is the risk of the portfolio w, Ω is the VCV matrix, σi (w) is the marginal risk contributions

to the overall portfolio risk of the i(th ) asset and T is the number of periods. In order to find a solution

for the ERC portfolio we need to make sure that the risk budget σi (w) is identical for every asset i

and j:

for any i and j (2)

A special case of ERC is when correlations across assets are the same, so the weight associated with each asset is given by the ratio of the inverse of its volatility with the harmonic average of the volatilities (also called simple or naive risk parity). The solution of risk parity is closed form and the weights are:

1/ ∑ (3)

where sj is the estimate for the stand alone volatility for each asset.

In general correlations are not pairwise identical, so finding the weights to satisfy equation (1) requires solving a quadratic minimisation program with non-linear constraints. A solution to the ERC problem can be obtained using numerical techniques applied to equation (1).

Recently Spinu (2013) proved existence and uniqueness of the ERC solution when the matrix covariance is non-degenerate, and provided efficient numerical techniques to solve the ERC problem.

We investigated the stability of the solution for our empirical study by randomly selecting one million different starting weights and observing differences in the final weights obtained using our numerical technique. There was no instability problem in the solution in our empirical study, and in all cases we found only one solution when we used the following constraints: 0 1, and ∑ 1.

Our methodology extends the conventional risk strategy by not only accounting for the risk of the six regions that make up the universe, but also taking into account the change in correlation between these regions. Equalising the risk contribution means that regions with lower estimated volatility and correlation with the other geographical regions receive larger index weightings.

The end product is a risk-balanced benchmark which increases diversification benefits and delivers higher risk-adjusted and absolute returns, while maintaining full exposure to the investable underlying universe.

Superior returns validate smart beta approach

The BofA Merrill Lynch Global High Yield Constrained Index is traditionally skewed to a few geographical regions. The US and Europe together currently constitute approximately 80% of the index weight, and until 5 years ago the US alone represented 80% of the index weight, as illustrated in Charts 2 and 4 below.

12

Chart 2: ECRC weighting addresses imbalance that characterises MCW index

US EU UK Asia CEEMEA Latin America

Chart 3: Historical geographical ECRC weights Chart 4: Historical geographical split of the BofA Merrill Lynch GHY Constrained Index

GHY”: Global High Yield. Source: Fidelity Worldwide Investment, BofA Merrill Lynch. Data as of 30th September 2014.

Past performance demonstrates that the ECRC strategy delivered superior results compared to the MCW index.

Chart 5 overleaf compares ECRC with the MCW index as well as other alternative indexation techniques: 1/N, Risk Parity and DTS Parity. The latter is an extension of the Risk Parity where we measure the risk by relative DTS of the sector and allocate inversely to the risk. The main difference is that traditional risk parity uses historical data for risk estimation while DTS is usually considered a forward looking measure as it is largely based on current market pricing.

21.0%

5.3%

4.6%

4.9%

6.4%

57.8%

Market cap weighted

16.5%

16.6%

13.1%15.4%

18.1%

20.3%

ECRC weighted

0%

20%

40%

60%

80%

100%

'99 '02 '05 '08 '11 '14

US

LATAM

CEEMEA

Asia

UK

EU

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

'99 '02 '05 '08 '11 '14

EU

UK

Asia

CEEMEA

LATAM

US

13

Chart 5: ECRC delivered superior results

Source: “IE” = information ratio. Fidelity Worldwide Investment, BofA Merrill Lynch. Data as of 30th September 2014.

As chart 5 above reveals, all alternative indexation techniques beat market classical indices on a risk adjusted basis, with similar performances for ECRC, Risk Parity and DTS parity. Interestingly, the naive diversification rule 1/N is superior in the back test, although only marginally so. We attribute the similarity in performance of the four methodologies to the similar correlation and volatilities over this period (more about this later on).

From 2004, ECRC risk-adjusted returns are higher, outperforming the traditional benchmark by an average of 1.7% annually, generating annual returns of 6.4% (Table 1). Overall risk is slightly higher, due to the presence of emerging markets (EM), but the approach delivers a higher information ratio. Maximum drawdown is approximately the same for both strategies.

Table 1: Risk-based approaches consistently outperform global MCW high yield

MCW Indices Alternative smart beta strategies

Global HY US HY ECRC 1/N Risk Parity DTS Parity

Since 1999

Hist. return 3.12% 2.95% 4.16% 4.40% 4.13% 3.83%

Volatility 11.00% 10.76% 12.01% 12.07% 11.97% 11.84%

IR 0.284 0.274 0.346 0.364 0.345 0.324

Max DD 41.6% 43.7% 41.8% 41.4% 41.8% 41.8%

From 2004

Hist. return 4.68% 4.31% 6.37% 6.57% 6.35% 6.11%

Volatility 11.31% 11.04% 12.73% 12.77% 12.66% 12.75%

IR 0.414 0.390 0.501 0.515 0.501 0.480

From 2010

Hist. return 6.54% 6.21% 7.60% 7.71% 7.59% 7.48%

Volatility 7.50% 6.81% 8.51% 8.59% 8.55% 8.47%

IR 0.872 0.913 0.892 0.897 0.887 0.883

Source: IE: information ratio. Max DD: Maximum drawdown. Fidelity Worldwide Investment, BofA Merrill Lynch. Data as of 30th September 2014.

There is a large outperformance of the strategies during the 2008/2009 crisis. If we exclude this period, from 2010 ECRC risk-adjusted returns are still higher and outperforming the traditional benchmark by an average of 1.06% per year, generating returns of 7.6% (Table 1 above). This performance comes with higher volatility but it does not seem to be exclusively due to the highest beta, because the risk adjusted performance is modestly superior. In general for both 10 and 15

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

'99 '00 '01 '02 '03 '04 '05 '06 '07 '08 '09 '10 '11 '12 '13 '14

ERC HY Global HY

US HY 1/N

Risk Parity DTS Parity

14

years, the ECRC outperformed the Global MCW High Yield, on average producing the best ranking together with the 1/n methodology (Tables 2 and 3 below).

Table 2: Outperformance of risk-based approaches: annual performance



1999 6.7% 5.8% 11.0% 13.3% 10.8% 10.7%

2000 -13.8% -13.4% -13.5% -13.5% -13.5% -13.3%

2001 -3.3% -2.5% -14.5% -16.7% -14.0% -16.2%

2002 -10.5% -9.6% -6.0% -3.2% -6.5% -3.5%

2003 25.1% 25.0% 25.2% 24.3% 25.2% 21.7%

2004 7.9% 7.6% 9.6% 10.0% 9.6% 9.8%

2005 1.0% 0.3% 3.3% 3.9% 3.3% 4.0%

2006 7.2% 7.0% 7.0% 6.9% 7.0% 6.7%

2007 -6.1% -6.4% -5.1% -5.0% -5.3% -5.1%

2008 -32.7% -32.4% -34.3% -33.9% -34.3% -34.7%

2009 64.5% 62.1% 83.0% 82.9% 83.1% 79.1%

2010 9.6% 9.2% 16.0% 16.3% 16.1% 14.7%

2011 -3.5% -1.7% -6.5% -6.6% -6.5% -6.9%

2012 16.4% 13.1% 20.4% 20.9% 20.3% 20.9%

2013 9.1% 9.4% 7.3% 7.3% 7.3% 8.2%

2014 2.3% 2.2% 2.0% 1.9% 2.0% 2.1%


Table 3: Outperformance of risk-based approaches: performance ranking



1999 5 6 2 1 3 4

2000 6 2 4 5 3 1

2001 2 1 4 6 3 5

2002 6 5 3 1 4 2

2003 3 4 1 5 2 6

2004 5 6 3 1 4 2

2005 5 6 3 2 4 1

2006 1 3 2 5 4 6

2007 5 6 3 1 4 2

2008 2 1 4 3 5 6

2009 5 6 2 3 1 4

2010 5 6 3 1 2 4

2011 2 1 3 5 4 6

2012 5 6 3 2 4 1

2013 2 1 5 4 6 3

Average rank

3.93 4.00 3.00 3.00 3.53 3.53


15

How to implement ECRC in practice?

A numerical solution can be obtained following Maillard et al. (2008). A full treatment of the numerical techniques available to solve this optimisation problem is out the scope of this paper. However, as can easily be verified from the weights in table 7, which is derived from the data in

tables 4, 5 and 6, the contributions to the portfolio’s risk - w Σw √ Σ⁄ - satisfy the ERC condition (2) for all geographical regions, namely that the risk contribution for each region is the same, resulting in different weights for each region. In the example below we use a 10 year equally-weighted VCV matrix for simplicity. Sample size considerations are always important when a VCV matrix is estimated, but in this case it is not a particularly severe limitation because the number of periods T is usually an order of magnitude larger than the number of assets N, i.e. 120 months versus 6 assets.

Table 4: Correlation matrix, last 10 years, monthly

EU UK Asia CEEMEA LATAM US

EU 100% 93% 70% 70% 44% 88%

UK 93% 100% 71% 69% 49% 84%

Asia 70% 71% 100% 70% 67% 61%

CEEMEA 70% 69% 70% 100% 58% 68%

LATAM 44% 49% 67% 58% 100% 56%

US 88% 84% 61% 68% 56% 100%

Source: Fidelity Worldwide Investment, Bloomberg, BofA Merrill Lynch. Data as of 30th September 2014.

Table 5: VCV matrix


EU 0.0015 0.0014 0.0017 0.0015 0.0013 0.0012

UK 0.0014 0.0018 0.0017 0.0013 0.0012 0.0011

Asia 0.0017 0.0017 0.0025 0.0020 0.0017 0.0014

CEEMEA 0.0015 0.0013 0.0020 0.0019 0.0014 0.0012

LATAM 0.0013 0.0012 0.0017 0.0014 0.0014 0.0010

US 0.0012 0.0011 0.0014 0.0012 0.0010 0.0011


Table 6: Volatility, last 10 years, monthly annualised


Volatility 13% 14% 19% 16% 19% 11%


Table 7: Total risk

Contributions to risk Weights

EU 0.625% 16.52%

UK 0.625% 16.64%

Asia 0.625% 13.08%

CEEMEA 0.625% 15.42%

LATAM 0.625% 18.07%

US 0.625% 20.27%

Total 3.75% 100%

Monthly √ Σ ∑ 3.75%, or approximately 13% annualised. Source: Fidelity Worldwide Investment, Bloomberg, BofA Merrill Lynch. Data as of 30th September 2014.

16

Correlations of excess returns across global high yield geographical regions are high and have increased since the 2008 financial crisis, as can be observed from Charts 6 and 7 below. This increase in co-movements of credit spreads is not only affecting the global high yield market but it has also been a common theme in other asset classes, particularly amongst fixed income assets. It is due to the so called ‘risk on/risk off’ dynamic that has been in place after Lehman Brothers’ default and the following years of exceptionally benign monetary conditions.

With this in mind, we can easily explain the similar performance of ERC, Risk Parity and DTS parity because the implicit assumption of similar correlations held up well during the 2008-2014 period. There is no guarantee that this will continue to be the case going forward. Evidence from a recent reading of cross-sectional correlation highlights a reduction in correlation and an increase in its dispersion. Therefore we anticipate that, should this divergence continue, the ERC and RP based methodologies could deliver substantially different weight allocations.

Chart 6: Cross-sectional correlation


Controlling systematic risk exposure in high yield smart beta

To round off our analysis, we assess the inherent systematic risk loading of this ECRC strategy and compare it to the MCW index. The illiquidity premium can be thought of as part of the credit risk premium and it is well known that liquidity in EM high yield is more challenging given the relative immaturity of these markets compared to the US. A liquidity premium is embedded in valuations to help compensate for this downside. A strategy that allocates more aggressively into EM is likely going to load relatively more into the liquidity premium.

To measure the overall liquidity/systematic risk we can use DTS rather than historical excess return volatility to assess the credit beta. This is a common measure of spread exposure because future excess return volatility is proportional to the overall current DTS 9.The main advantage of DTS is that it does not suffer from a historical lag effect, as it is the product of two analytics calculated daily, namely: spread duration and option adjusted spread.

Our normalised measure of credit beta is a simple ratio of the DTS of the ECRC index vs the DTS of the classic MCW index. Overall exposure to credit premium can then be measured.

9BenDoretal.(2007)

-20%

0%

20%

40%

60%

80%

100%

'04 '05 '06 '07 '08 '09 '10 '11 '12 '13 '14

Average

Avg + 1std

Avg - 1 std

17

Table 8: Credit beta distribution of ECRC global high yield vs MCW index

Percentile Credit beta

5% 90.7%

20% 94.5%

50% 99.8%

80% 106.5%

95% 115.6%


From table 8 we can see that the average credit beta is 101.2% and a median credit beta is 99.8%, so although not always perfectly identical we observe no persistent active beta positioning, supporting the superior risk adjusted performance of the ERC benchmark.

The ECRC portfolio was running a higher DTS/credit beta from Q3 2008 until Q4 2009, contributing to the good performance of the strategy during the full back test period. Including only the performance from 2010, we still see a better total return and also a modest risk-adjusted outperformance of the ECRC vs the global high yield MCW portfolio.

ECRC shows higher ex-post volatility but this does not suggest we are simply moving along the efficient frontier, because the information ratio marginally improves. The efficient frontier is typically concave, i.e. less return is typically offered for an additional unit of risk, so if we were simply increasing the beta we would expect a decline in the information ratio rather than an increase. In other words, and other things being equal, increasing risk would usually come with an overall lower information ratio, so despite the reduced sample size we think this is a good result and we attribute it to the better diversification captured by the risk-adjusted alternative indexation techniques.

Practical implications of managing an ECRC based portfolio

The ECRC approach developed here provides us with an excellent framework to deliver superior risk-adjusted returns. There are, however, a number of significant factors that influence portfolio construction and risk management in an actively managed ECRC portfolio. Translating the theoretical advantages into actual returns requires an approach that addresses these sources of friction.

While not a comprehensive list, the most significant risk factors are:

liquidity, transparency, pace of evolution of capital markets for emerging economy corporate debt and, correlation between assets across regions.

Naturally, these are generally familiar to participants in the high yield market, but their impact is magnified in an ECRC framework. Below we will briefly address each of these topics and their disproportionate influence.

Liquidity management is first and foremost among the key variables to consider. Allocating assets in an ECRC framework naturally compels the investor to take more exposure to traditionally less liquid emerging market issuers. This has direct consequences for both managing liquidity at the aggregate portfolio level and dimensioning exposure once the bottom up security selection process is complete. Individual managers will naturally differ in their preferences for diversification, but there is little doubt that EM high yield’s relatively smaller secondary market capacity and a more concentrated investor base warrant careful consideration.

Transparency risk - both with respect to corporate governance and creditor rights - also increases in an ECRC universe. Market standards for disclosure and corporate governance vary widely. Of course, investors believe that best practice will ultimately converge upon developed market levels over time as issuers become more accustomed to capital markets. Nonetheless, progression takes time and ebbs and flows with the credit cycle. Insolvency regimes, on the other hand, will remain

18

complex and idiosyncratic. Appetite to hold stressed and distressed names in such jurisdictions will clearly vary as a function of experience and access to supplementary resources.

ECRC security selection therefore demands additional scrutiny. Expanding the investment horizon for portfolio constituents assists in minimizing risk management problems associated with lower liquidity and reduced transparency. Correspondingly, embracing an ERC approach raises the threshold for the fundamental conviction required to allocate risk in an actively managed portfolio.

From the top-down risk management perspective, correlations between risky asset classes are always a moveable feast. Expectations for increasing breadth and depth of emerging high yield markets imply the long-term underlying trend is for correlations between developed high yield markets (US, EU, UK) and emerging high yield markets to increase and therefore for volatilities to converge. These relationships are inherently unpredictable; they can and do fluctuate with shifts in market risk appetite and liquidity profile, creating practical implications for tactical asset allocation and liquidity management at the portfolio level. Rebalancing the ECRC reference benchmark’s regional allocations less frequently than traditional indices should help reduce the noise factor. Perhaps more importantly, adhering to an ECRC rather than 1/N approach embeds an important risk calibration into the process, given the inherent uncertainties in measuring ex-ante correlations and volatilities.

Taking all these significant factors into consideration is paramount in constructing an ECRC based portfolio.

19

Conclusion

We show in our empirical study on the global high yield universe that, similar to previous studies on other asset classes, smart beta outperforms MCW for the global high yield asset class. We argue that the case to move away from MCW indices, from both an intuitive and theoretical point of view, is even stronger in fixed income than in equity markets.

MCW indices have been modified in many ways to reduce their inherent inefficiencies. Most importantly, constrained indices can be thought of as a form of smart beta. In our analysis, however, we find better alternatives to such smart beta benchmarks, including the ECRC, Risk Parity, DTS parity and the simple 1/N approach.

Furthermore we witness similarities in performance between these alternative indexation techniques, which can be explained fully by a combination of similar volatilities and high correlation during the sample period. Despite the similar performance we choose ERC as our preferred approach because it gives a better guarantee for capturing possible future shifts in correlation and volatility between assets from different parts of the world.

We therefore think that these results are very promising and prove that alternative indexation can be an interesting proposition in the fixed income spectrum, generating higher absolute and risk-adjusted returns.

Having looked at practical implications of managing a portfolio around such a benchmark, including liquidity issues, transparency risk and the additional scrutiny necessary in the security selection process, we consider the role of the portfolio manager to be central for this kind of product. Combining a smart beta approach with active management means alpha can be captured alongside the more efficient allocation of portfolio risk.

20

Key references

Amenc, N., and L. Martellini. 2002. Portfolio optimization and hedge fund style allocation decisions. Journal of Alternative Investments 5 (2): 7-20.

Amenc, N., Goltz F, and L. Martellini. 2013. Smart beta 2.0. EDHEC-Risk Institute

Arnott, R., J. Hsu, and P. Moore. 2005. Fundamental indexation. Financial Analysts Journal 60 (2), 83 9

Amenc, N.; Goltz, F.; Le Sourd, V. (2006). Assessing the Quality of Stock Market Indices. EDHEC Publication.

Ben Dor, A., Dynkin, L., Houweling, P., Hyman, J., Leeuwen, E., and Penninga, O., “DTS (Duration Times Spread) A New Measure of Spread Exposure in Credit Portfolios”, Journal of Portfolio Management, Winter 2007

Clare, Motson, and Thomas (2013) “An Evaluation Of Alternative Equity Indices. Part 1: Heuristic And Optimised Weighting Schemes” Cass Working Paper

Clare, Motson, and Thomas (2013) “An Evaluation Of Alternative Equity Indices. Part 2: Fundamental Weighting Schemes” Cass Working Paper

Demey, P., S. Maillard and T. Roncalli. 2010. Risk-Based Indexation. Lyxor. White Paper:http://www.lyxor.com/fileadmin/_fileup/lyxor_wp/document/16/docs/all.pdf.

DeMiguel, V., Garlappi, L., Uppal, R., 2009. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? Review of Financial Studies 22.5, 1915-1953.

Fama, E., and K. French. 1992. The cross-section of expected stock returns. Journal of Finance 47 (2): 427–65.

Falkenstein, E. 1994. Mutual Funds, Idiosyncratic Variance, and Asset Returns. NortWestern University, Doctro of Phillosxopy dissertation.

Haugen, R. A.; Baker, N. L. (1991). "The Efficient Market Inefficiency of Capitalization-Weighted Stock Portfolios". Journal of Portfolio Management 17 (3): 35–40. doi:10.3905/jpm.1991.409335.

Hsu, Jason (2006). "Cap-Weighted Portfolios are Sub-optimal Portfolios". Journal of Investment Management 4 (3): 1–10.

Grinold, R. 1992. Are benchmark portfolios efficient? Journal of Portfolio Management (fall).

Ilmanen, A, 2011. Expected Returns: An investor’s Guide to Harvesting Market Rewards; Chichester, UK: John Wiley & Son

Ledoit, O., Wolf, M., 2003. Honey, I Shrunk the Sample Covariance Matrix. https://repositori.upf.edu/bitstream/handle/10230/560/691.pdf

Modigliani, F.; Miller, M. (1958). "The Cost of Capital, Corporation Finance and the Theory of Investment". American Economic Review 48 (3): 261–297. JSTOR 1809766.

21

Martellini, L. 2008. Towards the design of better equity benchmarks: Rehabilitating the tangency portfolio from modern portfolio theory. Journal of Portfolio Management 34 (4): 34-41.

Maillard S., Roncalli T. and Teiletche J. (2008), On the Property of Equally-weighted Risk Contributions Portfolios, Working Paper, Available at http://ssrn.com/abstract=1271972

Michaud, R O. "The Markowitz Optimization Enigma: Is 'Optimized' Optimal?," Financial Analyst Journal, 1989, v45(1), 31-42.

Nielsen, F., and R. Aylursubramanian. 2008. Far from the madding crowd: Volatility efficient indices. Working paper.

Sharpe, W. (1961). “Portfolio Analysis Based on a Simplified Model of the Relationships Among Securities,” PhD Dissertation, University of California at Los Angeles, June 1961.

Sharpe, W. (1963). “A Simplified Model for Portfolio Analysis,” Management Science, 9:277-293.

Sharpe, W. 1964. “Capital asset prices: A theory of market equilibrium under conditions of risk”. Journal of Finance 19 (3): 425–42.

Shepherd, S. (2011). “A Fundamentally Weighted Broad-Based Fixed-Income Index”. Journal of Indexes.

Spinu, F., (2013). An Algorithm for Computing Risk Parity Weights (2013), www.ssrn.com/abstract=2297383.

Qian, E. (2005). "Risk parity portfolios: Efficient portfolios through true diversification". Panagora Asset Management

Tobin, J [1958b], “Liquidity Preference as Behaviour Towards Risk”, Review of Economic Studies, 25, No. 67, pp. 65-86.

22

Important Information

This document is for Investment Professionals only and should not be relied on by private investors.

This document is provided for information purposes only and is intended only for the person or entity to which it is sent. It must not be reproduced or circulated to any other party without prior permission of Fidelity.

This document does not constitute a distribution, an offer or solicitation to engage the investment management services of Fidelity, or an offer to buy or sell or the solicitation of any offer to buy or sell any securities in any jurisdiction or country where such distribution or offer is not authorised or would be contrary to local laws or regulations. Fidelity makes no representations that the contents are appropriate for use in all locations or that the transactions or services discussed are available or appropriate for sale or use in all jurisdictions or countries or by all investors or counterparties.

This communication is not directed at, and must not be acted on by persons inside the United States and is otherwise only directed at persons residing in jurisdictions where the relevant funds are authorised for distribution or where no such authorisation is required. Fidelity is not authorised to manage or distribute investment funds or products in, or to provide investment management or advisory services to persons resident in, mainland China. All persons and entities accessing the information do so on their own initiative and are responsible for compliance with applicable local laws and regulations and should consult their professional advisers.

Reference in this document to specific securities should not be interpreted as a recommendation to buy or sell these securities, but is included for the purposes of illustration only. Investors should also note that the views expressed may no longer be current and may have already been acted upon by Fidelity. The research and analysis used in this documentation is gathered by Fidelity for its use as an investment manager and may have already been acted upon for its own purposes. This material was created by Fidelity Worldwide Investment.

Past performance is not a reliable indicator of future results.

This document may contain materials from third-parties which are supplied by companies that are not affiliated with any Fidelity entity (Third-Party Content). Fidelity has not been involved in the preparation, adoption or editing of such third-party materials and does not explicitly or implicitly endorse or approve such content.

Fidelity Worldwide investment refers to the group of companies which form the global investment management organization that provides products and services in designated jurisdictions outside of North America Fidelity, Fidelity Worldwide Investment, the Fidelity Worldwide Investment logo and F symbol are trademarks of FIL Limited. Fidelity only offers information on products and services and does not provide investment advice based on an individual circumstances.

Issued in Europe: Issued by FIL Investments International (FCA registered number 122170) a firm authorised and regulated by the Financial Conduct Authority, FIL (Luxembourg) S.A., authorised and supervised by the CSSF (Commission de Surveillance du Secteur Financier) and FIL Investment Switzerland AG, authorised and supervised by the Swiss Financial Market Supervisory Authority FINMA. For German wholesale clients issued by FIL Investment Services GmbH, Kastanienhöhe 1, 61476 Kronberg im Taunus. For German institutional clients issued by FIL Investments International – Niederlassung Frankfurt on behalf of FIL Pension Management, Oakhill House, 130 Tonbridge Road, Hildenborough, Tonbridge, Kent TN11 9DZ.

In Hong Kong, this document is issued by FIL Investment Management (Hong Kong) Limited and it has not been reviewed by the Securities and Future Commission. FIL Investment Management (Singapore) Limited (Co. Reg. No: 199006300E) is the legal representative of Fidelity Worldwide Investment in Singapore. FIL Asset Management (Korea) Limited is the legal representative of Fidelity Worldwide Investment in Korea. In Taiwan, Independently operated by FIL Securities (Taiwan ) Limited 15F, 207 Tun Hwa South Road, Section 2, Taipei 106, Taiwan, R.O.C. Customer Service Number: 0800-00-9911#2

ITL15-15

applying a risk-based smart beta approach to fixed income investing · pdf file ·...

Documents