Domestic Fixed Income Factor Implementations

bySteven Braun

March 2, 2020

12 minute read

by Steven Braun, Newfound Research

This post is available as a PDF download here

Summary

Prior academic and practitioner research suggests that factor-based fixed income investing can create attractive return profiles and be useful when building fixed income portfolios.
Using an investment universe of eight domestic fixed income asset classes, we build dollar-neutral long-short portfolios targeting Value, Momentum, and Carry factors using both single-metric definitions and ensembled-metric definitions.
Moving from Long-Short to Long-Only implementations, we find that the tactical tilts created by the factor portfolios result in attractive risk-adjusted returns. Overweights to high-yield assets, however, can expose the portfolio to crash risk when credit-spreads spike.
To combat the risk of being overweight assets at inopportune times, we implement a Trend Filtering approach to the strategy, delaying purchases or exiting positions when trends are negative.

In previous commentaries, we have written about fixed income factors in the municipal credit market as well as multi-sector bonds and found that there is evidence supporting a factor-based approach to fixed income when implemented at the sector-level, as opposed to the individual security-level.

In this commentary, we explore Value, Carry, and Momentum signals in the fixed income universe with the goal of implementing the factors in a robust, long-only context. Specifically, we will apply these signals to eight domestic fixed income sectors, spanning duration and credit quality.

U.S. Treasuries: Short (1-3 Year, SHY), Mid (3-7 Year, IEI), Intermediate (7-10 Year, IEF), and Long (20+ Year, TLT).
Investment Grade Corporates: Short-term (IGSB) and long-term (LQD).
High Yield: Short- (SHYG) and intermediate-term (HYG).

As in previous commentaries, we implement these exposures using a corresponding ETF and, prior to ETF launch, the data is extended by employing data from the underlying index that each ETF seeks to track

The factors explored in this commentary are as follows:

Value: Buy cheap and sell expensive. Measured by the Z-score of deviations from the five-year average real yield[1].
Momentum: Buy recent winners and sell recent losers. Initially measured by trailing 63-, 126-, and 252-day total return.
Carry: Buy securities with high expected return and sell securities with low expected return.[2]

Long-Short Implementation

To start off, for each of the factors, we construct dollar-neutral long-short portfolios where the weights in each are calculated as:

Where c is a scaling coefficient to ensure that the portfolio is dollar-neutral, S is the score of each holding in factor f at time t, and N is the number of available holdings in the universe at time t.

Source: Bloomberg; Tiingo. Calculations by Newfound Research. Returns are hypothetical and backtested. Returns are gross of all management fees, transaction fees, and taxes, but net of underlying fund fees. Total return series assumes the reinvestment of all distributions.

From the chart above, we can see that the long-short factors provide attractive return profiles, except for the 252-day specification of Momentum. This result aligns with prior research conducted where short-term momentum signals are potentially more effective for fixed income than in other asset classes.[3]

Below we show the return statistics of each portfolio as well as their pair-wise correlations:[4]

Return Statistics:

Correlations:

Since the vast majority of total return in fixed income asset classes comes from income (as opposed to price appreciation), there is a large opportunity cost for switching out of higher-yielding instruments. Assets exhibiting higher carry, generally come with the expectation of higher returns. Certain signals, then, can lead to portfolios exhibiting negative carry when those signals call for shorting or underweighting the higher carry assets.[5]

This can be illustrated quite well in the Value factor, which has been short short- and intermediate-term high yield bonds since 2016.

The plot below shows the real yield and average real yield for the iShares iBoxx High Yield Corporate Bond ETF (ticker: HYG) and iShares 7-10 Year Treasury Bond ETF (ticker: IEF). In both cases, we can see that with the declining rate environment over recent history, the real yields have been below their respective averages. The measurement of value, though, comes from the deviations from this average real yield over time.[6]

Source: Bloomberg; St. Louis Fed. Calculations by Newfound Research.

The plot below shows the Z-Scores of deviations from average real yield for each of the above tickers.

Source: Bloomberg; St. Louis Fed. Calculations by Newfound Research.

We can see that while both tickers have Z-Scores below 0, higher yielding assets have decreased more rapidly than 7-10 year treasuries. Since we are ranking on this Z-Score, the securities with the higher Z-Score are treated more favorably. This coincides with the allocations we see below, as credit spreads have been in a down-trend since 2008 leading to a less-favorable relative valuation.

Ensembled Long-Shorts

In an attempt to refine our signals, avoid timing luck, and diversify our signal risk, an ensemble approach can be employed. Our ensembled factor definitions are as follows:

Value: Measure the deviations from average real yield, where the average is calculated for each year between one and ten years.

Momentum: Measure momentum using total return, price minus moving average, and moving average crossover for periods between 60-day and 380-day, in increments of 10 days.[7]

We will hold our carry definition constant for the remainder of this commentary, as utilizing another yield measure would obtain directionally similar results without being necessarily additive to the improvement of the portfolio.

In addition to tweaking our factor measurements, we can include an additional weighting methodology to obtain a purer tilt to the factors. Specifically, we include a weighting methodology that equal weights securities in the top and bottom thirtieth quantiles, removing the intermediate-ranked securities.

It is important to remind ourselves here why we would favor an ensemble-based approach. Using the momentum factor as an example, we can see from the initial tests that a shorter lookback outperformed longer lookbacks quite handily. However, this would not always be expected to hold using a different universe or time-period. The goal of this analysis is not to simply produce the most up-and-to-the-right-est curve; rather, it is to ensure that we are capturing the essence of the factor we are attempting to target.

By creating a robust investment process that encompasses a range of sensible specifications, we are reducing the likelihood that our portfolio will fail out-of-sample.

From the ensembles, we see that the directionality and courses for the factors broadly rhyme with our initial specification.

Below, we plot the allocations for our ensembled long-short portfolios (excluding Carry). Since the portfolios are constructed in a long-short fashion, the sum of each leg of the portfolio will not necessarily equal +/- 100%, though, they will remain dollar-neutral. This reflects conflicting signals in the underlying portfolios, causing the ensembled strategy to reduce its gross exposure.

The Value allocations remain similar with only twenty weeks out of the entire analysis period where the notional exposures of each leg are not 100%, with the lowest notional exposure reaching +/- 97%. This high level of notional exposure indicates that the specifications are largely in agreement of the relative value of asset classes, only deviating when there are dramatic shifts in the yield environment over short time-periods.

Momentum allocations see the largest shift as the average notional exposure of each leg is around 40% indicating a fairly meaningful discrepancy in the underlying momentum signals.

Implementing Long-Only Strategies

While it is comforting to know that the long-short implementations are efficacious, it poses a hurdle to any investor that is hoping to gain exposure to these factors and is either disallowed from shorting due to mandate or is simply wary of the operational burdens of the shorting process.

Additionally, a long-only portfolio will alleviate some of the pain of shorting high yielding assets. We need to keep in mind, however, that any active bets being made in a portfolio can be encapsulated as a long-short overlay on our benchmark. In this light, being underweight assets that are performing well in the benchmark will not necessarily cause absolute losses, but will result in relative underperformance (remember, we are implicitly short those assets). Furthermore, a long-only context can prevent us from expressing our factor views in their purest form, as our implicit short is limited by how much we can actually underweight a sector within the benchmark.

To convert these factors into long-only strategies, we will menially alter our weighting methodology to encompass this change. Further, to avoid timing luck, we will include weekly sub-indexes: Momentum uses four weekly sub-indexes, while Value and Carry use 52 weekly sub-indexes.

For our rank-weighting measure, our new weighting function results in:

While our quantile-cutoff weighting methodology simply removes the bottom 70% of the ranks and equally weights the top 30%.

We maintain the same ensembling parameters as before, but now aggregate the factors into a naïve equal-weighted multi-factor portfolio.[8]

Return Statistics[9]:

Correlations:

The long-only implementations create attractive strategy profiles on a risk-adjusted basis, but the glaring bruise on the factor peach is the maximum drawdown created in the Carry and Value factors. In a strong majority of cases, fixed income is relied upon for diversification purposes, and a heavy tilt toward credit during an equity drawdown can be especially damaging to investors when diversification is needed most.

Another point of contention is the high correlation that arises between Carry and Value. Since we are utilizing a naïve equal weight between the factors, the high correlation could be skewing the multi-factor portfolio too heavily into portfolios that score well on both carry and value, which coincidentally concentrates the bets being taken in a long-only portfolio.

There are a few ways to combat this. One method that we will explore in this commentary is Trend filtering.

Trend Filtering

We define trend filtering as the removal (or deferral of purchase) of assets exhibiting negative trend.

Implementing the trend filter can still allow our portfolio to tilt towards the target factors, but only if the price trend is signaling an attractive time to buy. In periods such as 2008 when yields on credit-based assets were gapping upwards, our Value and Carry signals would be signaling that the assets were becoming more attractive relative to the other assets in the universe. The trend filter would then delay the addition of these assets while yields continue to rise until yields stabilize or decline to more normalized levels.

We will apply this filter on both our Value and Carry signals using the same parameterizations of the momentum signal (i.e. 60- to 380- day total return, price minus moving average, and dual moving average crossovers). When a trend signal indicates an inopportune time to buy, the asset will be removed from the scoring mechanism and the remaining assets will be scored on the same method as before. Since the Momentum factor would already be targeting assets with the most attractive price trends, this factor will not be filtered.[10]

Return Statistics:

Correlations:

From filtering out negatively trending securities, we see that the drawdowns for both the Value and Carry styles were improved, albeit at the cost of return. Included in the above figures is HYG and the Vanguard Total Bond Market Index Fund (Ticker: VBTLX) as a point of reference.

One concern by filtering out negative trending assets is that the portfolio could be reducing the yield generated by the portfolio. For comparison purposes, below we plot the carry estimate for both the original Carry portfolio and the Filtered Carry portfolio.

Source: Bloomberg; Tiingo. Calculations by Newfound Research.

By overlaying the momentum filter, we see that the most glaring divergence is during market turmoil, but the carry of the filtered portfolio does not drop to levels that would cause a large concern for investors seeking return from investing for income.[11]

Looking into the allocations of the multi-factor strategy, we see that the allocations are changed in the directions we would expect. Tilting away from credit in favor of higher credit quality in times of market stress, then upon “market normalization”, leaning back into higher-yielding assets.

To see if our implementation is adding value through the tactical signals we employ, below we plot the ratio of the multi-factor strategy against its average allocations. When the ratio increases, the tactical decisions employed are adding value, while a decreasing ratio implies the decisions are detracting from our average allocations.

The average allocation portfolio and the multi-factor strategy stay remarkably close over the time period. The benefit of course, can be seen during the large spikes in 2008 and 2014-2015 when the tactical decisions to underweight credit-heavy assets improved the drawdown profile of the portfolio.

One implicit assumption we are making in this analysis is that U.S. Treasuries and high yield fixed income factors will continue to remain low-to-negatively correlated. In all cases, the factor portfolios had exposures to both high yield and treasuries which have been negatively correlated during the market corrections over the time period analyzed. Moving forward, this relationship will not necessarily hold if market dynamics change. Whether that caveat ever comes to fruition or not, it is important to understand the building blocks of a portfolio and recognize the assumptions governing the current strategy.

Conclusion

In this commentary we sought to look into Value, Carry, and Momentum factors using fixed income asset classes. We find that there is efficacy in these factors when viewed through a long-short lens. Importantly, over the past twenty years, it seems that longer-term momentum measures seem to add little to no value, while short- to intermediate-measures seem to capture interest rate trends and allow a portfolio to successfully rotate into (out of) outperforming (underperforming) asset classes.

We then sought to create robust methods of these factor implementations and implement these robust signals. The signals used for Value seem to remain fairly stable regardless of the formation period selected. This solidifies the notion that credit and quality interest rates tend to move in vaguely the same direction except in times of market stress, when credit-spreads can move in the opposite direction of treasury rates.

Implementing these factors in a long-only portfolio, then, created attractive portfolios from both a nominal and risk-adjusted perspective. In this implementation, Value and Carry factors had a high degree of correlation while momentum had a somewhat lower correlation with the other two factors. Combining these factors into a naïve multi-factor portfolio resulted in a reasonably attractive take on fixed income investing; however, combining these styles resulted in drawdowns that could cause investors seeking the safety of fixed income to take a second glance.

To attempt to combat the potential drawdown mostly resulting from being over-exposed to high yield bonds, we then introduced a momentum filter into the Carry and Value factors which sought to remove assets from the lineup which were exhibiting negative price trends. This attempt at mitigating drawdown realized a degree of success as the max drawdowns for the Value, Carry, and Multi-Factor portfolios were drastically reduced.

Appendix

[1] For a further explanation of this measure, see “Navigating Municipal Bond Factors”

[2] In the past, we have used yield-to-worst minus the risk-free rate as a proxy for carry. In this commentary, we utilize a more granular measure for each sector. For Treasuries, Carry is measured as (yield + roll – rf). For Investment Grade Corporates, Carry is measured as (OAS + duration-matched treasury yield + roll – rf).

[3] Our return data for short- and intermediate-term high yield begins in December 2001, so 1/1/2003 was selected to allow for a momentum formation period of one year to include sufficient data.

[4] The average annual turnover of the Carry, Value, 63-, 126-, and 252-day momentum strategies are 252%, 686%, 1784%, 1321%, and 885%, respectively.

[5] For a further discussion on fixed income total return versus price return, we covered this more thoroughly in https://blog.thinknewfound.com/2019/06/tactical-credit/

[6] For our average yield, we use an exponentially weighted average.

[7] The appendix shows further detail into the results of individual momentum versions.

[8] Average annual turnover for Value, Carry, Momentum, and Multi-Factor are 41.9%, 24.9%, 155.7%, and 62.1%, respectively.

[9] Sharpe Ratios are calculated in excess of the iShares 1-3 Yr Treasury Bond ETF (SHY).

[10] Average annual turnover for Value, Carry, Momentum, and Multi-Factor are 41.8%, 37.9%, 155.7%, and 60.0%, respectively.

[11] This statement is, of course, relative to the original carry portfolio. Yield dynamics dropping over time is a concern for income-seeking investors, but this problem remains regardless of the strategy used to seek this income.

Steven Braun

Steven is a Quantitative Analyst at Newfound Research. Steven joined Newfound in June 2019 and is responsible for investment research, strategy development, and supporting the portfolio management team. Prior to joining Newfound, Steven was an investment analyst at Frontier Asset Management where he conducted quantitative research into the ongoing maintenance and improvement of proprietary expected return and risk models. Steven holds a Master of Science in Applied Quantitative Finance from the University of Denver and a BBA with concentrations in Corporate Finance and Investment Analysis from Colorado State University.

This post was origninally published at the Newfound Research's Flirting With Models Blog