Overview
Read the full white paper here
Factor investing has its roots in the Fama-French academic papers from the early 1990s. Initially a study into the efficient market hypothesis, academics soon discovered that gaining exposure to certain factors could provide persistent sources of improved risk-adjusted return compared to the market portfolio. One of the better-known factors is value: assets that are cheap perform better over the long term than assets that are expensive. Academics showed that it was relatively simple to capture some of the value factor’s alpha by using readily-available data in a systematic way.
But there are many ways to judge value; in equites this can range from a simple metric such as a price-to-earnings multiple to more complex metrics such as a full cash flow valuation model. And decades after equity factors became widely known there are still disagreements about how they are best defined. The result is many products claiming to target the same factor but with quite different approaches. This, along with the proliferation of potential new factors, has led to what many call the ‘factor zoo’, thousands of different potential factors proposed by academics.
Factor investing in fixed income
Factor investing in fixed income has a number of potential applications, just as it does with equities. Firstly, it can be used by active managers for idea generation and attribution analysis, assessing the source and nature of excess returns. Factor investing can also be used in ‘low touch’ systematic strategies, where automation can be used to reduce costs and mitigate the biases inherent in human decision making. The drawback is that systematic strategies cannot yet provide the level of depth, context, and therefore potential alpha a human analyst can.
Instead, systematic factor strategies offer a middle path, allowing investors a more nuanced balance to meet their specific goals between the lower cost of passive and the excess return potential of active management.
Why corporate bonds present a challenge
Equities benefit from high liquidity, electronic trading, low execution costs and excellent availability of data, all of which help to make systematic strategies easy to test and implement. The same cannot be said of bond markets, where investors face the added complication of multiple securities from each issuer. Despite this, factor-based solutions can and are being implemented with bonds and the body of academic and practitioner research on the subject is growing.
Some of the same issues that arise in equity factor investing are also present in fixed income. As noted above, factor definitions are often a matter of personal choice. Where you put your faith in price-to-book, we might favour enterprise-value-to-EBITDA. Furthermore, different factor definitions perform differently as market dynamics change. The worry is that some researchers repeatedly test the same historical data and simply pick the factor definition that presents the best backtest, a practice known as data mining. We are not alone in thinking this is not a sound way to achieve good future performance.
Fixed income factor investors also face the larger question of whether to use the factors from equity research or instead something more specific to bonds. Recent academic studies have tested the main equity factors on fixed income markets but offer little clarity - they use different definitions to those used in equities and even different definitions to each other.
Our bottom-up approach to factor investing
To help overcome these difficulties, we have designed a novel bottom-up method of multifactor portfolio construction. Due to the differing drivers and limited upside of fixed income returns, we do not believe that restricting bond factors to their equity equivalents is optimal. Instead we approached the issue with an open mind.
Rather than translate the equity factor methodology onto individual bonds, the complicated nature of fixed income markets (where bonds from the same issuer have different characteristics such as maturity and seniority) means that a more sophisticated approach is warranted. Unlike most other available fixed income strategies, our multifactor model identifies issuers with the most alpha potential, rather than individual bonds. We then select the bonds that offer the best value exposure to these issuers.
At Fidelity we have built a substantial database of corporate issuer information across asset classes and data types, of around 80 variables for each issuer. The variables are assigned to one of three factor groups:
- fundamental company information
- valuation, which includes cross sectional and historical deviations from the mean for bond and equity characteristics
- sentiment, which includes momentum and trend signals for both equities and bonds.
We then use a proprietary algorithm to weight the variables within each factor group. The weights are chosen to maximise the information ratio (a measure of risk-adjusted return) of a combined factor portfolio after transactions costs (costs are an important consideration in fixed income markets and are often overlooked in research). Finally, using the same algorithm, the three factor groups are combined to leave a single credit multifactor time series per issuer.
This approach has several advantages over existing fixed income and equity factor construction. Firstly, factors are not narrowly defined by a few variables. Instead we analyse all potentially relevant data and let our algorithm decide the factor definitions that offer the most alpha potential. Secondly, the factor weights are dynamic, meaning the strategy adapts to changes in market conditions over time.
From issuers to individual bonds
Our process identifies the companies with the highest potential for the positive returns provided by our multifactor. Backtests using generic five-year issuer returns calculated from our issuer curves show this credit multifactor delivers significant alpha. Knowing which issuers to pick would be sufficient to create either long-only or long-short portfolios using equities. But fixed income strategies have an extra complication - identifying which bonds to own in order to create a long-only portfolio (long-short portfolios are not feasible in corporate bonds due to the difficulty and expense of shorting).
We have developed two further factors to solve this problem: bond roll down and bond residual. The roll down factor is the natural return from holding a bond for one year assuming no change to its issuer curve. The residual factor calculates the premium of a bond over the issuer’s credit curve, providing a measure of how cheap each bond is.
Backtest results are encouraging
Once the individual bonds have been identified, we use an optimiser to construct a portfolio that maximises exposure to our issuer, roll down and residual factors while minimising transactions costs. The optimisation controls the exposure of the portfolio relative to its respective index so that there is no residual exposure to unwanted risk factors. These include things like duration, sectors, credit beta, and rating categories. Without controlling for turnover, backtests using the US investment grade universe show that our portfolio outperformed the benchmark by 25 basis points per month, indicating our multifactor construction method has significant alpha potential.
However, portfolio turnover in the backtest described above was around 50 per cent per month, much too high to make the strategy practical. However, introducing transactions costs to the portfolio construction process constrained turnover to a more reasonable 10 per cent per month. Doing so reduced the alpha but still generated a significant excess return of 72 basis points per year after transactions costs.
US investment grade universe. Benchmark index - ICE Bank of America Merrill Lynch C0A0. Source: Fidelity International, March 2019.
The portfolio features around 150 issuers compared to more than 1000 in the comparative index, with a tracking error of 1.5 per cent. To validate the results further, we limited bond selection to only the most liquid half of the universe. Encouragingly this reduced the monthly alpha only by 0.1 basis points. We also achieved an excess return of around 75 basis points per year in US high yield and 55 basis points per year in European investment grade markets.
A differentiated factor strategy
Unlike equity factor returns, our strategy is not cyclical. Our portfolio construction approach removes unwanted macro biases by controlling for cyclical factors such as market and sector exposure. However, we have observed that alpha generation is higher in periods of higher volatility, meaning the strategy performs particularly well in times of market stress.
Our factor strategy should appeal to those investors with active mandates who are starting to question the potential for fundamental research to outperform in credit markets and who want to dip their toes in an investment style with more passive elements. For investors who currently have passive exposures, our strategy still offers an index-relative exposure with no portfolio manager but with the potential to outperform.
Factor investing is a new frontier for fixed income investors, made possible by the changing structure of the market. More reliable historical data and the introduction electronic trading in recent years has made systematic trading in the more liquid credit markets possible. Our approach using dynamic, optimised factor definitions can take advantage of this, potentially providing significant alpha at a lower cost to fully active management.
