Wednesday, December 19, 2018

Cash or Bonds at Low Yields and a Flat Yield Curve?

The End of an Era?

While there have been a few cyclical periods of rising rates over the past 40 years, we've largely been in one large downtrend... meaning that it has consistently paid to own bonds vs cash** or take duration risk for nearly my / many investment lives. 

Now that we've moved away from a zero interest rate policy on cash in the U.S. and the yield curve is essentially flat, this post is an attempt to pose the question of whether it still makes sense own bonds at the same scale.



The Historical Benefit of Extending Duration

Mapping the forward 10-year return of t-bills, a constant maturity 5-year Treasury bond, and a 10-year constant maturity Treasury bond against the starting yield of the 10-year Treasury, it should come as no surprise that 1) higher nominal starting yields have led to higher forward returns and given the yield curve is usually upwards sloping, that 2) longer duration bonds have generally outperformed shorter-duration bonds given their higher starting yield. 


The data in the chart above may be more easily digested when the average forward returns are "bucketed" by the starting yields of less than 4%, 4-8%, and more than 8%. Here we can more clearly see that the benefit of bonds / duration historically occurred when rates were quite elevated. 



Adjusting for a Flat Yield Curve

What the above chart does not account for is the relative starting yield of t-bills, the five year Treasury bond, and the 10 year Treasury bond. The below charts "adjust" the returns of t-bills and 5 year Treasuries to a yield equal to the starting yield of the 10 year Treasury. For example... if at T=0 t-bills yielded 1% and 10 year Treasuries yielded 3%, I added 2% / year to the t-bill ten year return. This is obviously inexact given it assumes the path of yield movements are identical in each situation despite the different yield levels.


And again... adjusted returns bucketed by starting yield of the 10 year Treasury. Now we can see the improved opportunity for cash / reducing duration; an investor can potentially (if history is a guide and this framework makes sense) reduce risk, while capturing similar (or potentially increased) return.


For those focused on tactical asset allocation, bonds are likely to outperform cash if we enter a deflationary / disinflationary environment, while cash will likely outperform if markets continue to normalize or if there are inflationary pressures. Rather than pretend to guess which situation is more likely, I would frame it as follows... are investors being fairly compensated for the increased volatility to own bonds vs cash?

With cash, you know the value will increase by the short-term rate, you just don't know what that short-term rate will be in the future. Importantly, the daily volatility of cash can be assumed to be pretty much 0% irrespective of what happens in the market. With bonds, you know roughly what that the nominal return will be, but you don't know if that return will compensate you over cash. As important, the value of bonds will fluctuate daily (historical volatility has been ~6% for 10 year Treasuries).

So, while the risk of owning bonds has likely been exaggerated for longer-term holders, there is a real increased risk of ownership relative to cash. Whether it’s the risk of less proceeds available when it comes to reallocating to other opportunities / taking withdrawals or the behavioral impact of a fluctuating account value, it’s a risk. So, if you believe the return of cash / lower duration bonds will be the same (or more) than bonds, why take that risk?


** in the above analysis I considered t-bills as a cash equivalent because they are liquid and not subject to material fluctuations in value

Wednesday, September 19, 2018

Market Timing The Credit Cycle

Over the last few years, you’ve likely heard the following competing narratives: ­

  • “Credit spreads are tight, a sign of exuberance among investors that are willing to overlook risk. This will end in tears.” 
  • “Credit spreads are tight, reflecting an environment of high economic growth and low default rates. This supports risk assets.” 
This post will outline why both of the above comments may be correct (or incorrect) by looking at asset class performance over various time frames / over recent credit cycles.


THE CYCLICALITY OF CREDIT

Corporate bond spreads can be thought of as an indicator of the overall creditworthiness of the private sector, with widening spreads either reflecting a difficult environment for companies to service their debt or the perception by investors that it may be difficult for these companies to service their debt. In this post, credit risk is defined as the difference in the option adjusted spread “OAS” (the difference in yield between a corporate bond and similar duration Treasury bond) between junk bonds and investment-grade corporate bonds, which I’ll refer to as “quality spread”.

The chart below outlines the month-end "quality spread" in percent terms going back to 1994, a time frame that goes as far back as I can get the high yield OAS data, as well as two bands reflecting one standard deviation above and below the three year month-end average spread.


Quality Spread Since 1994 (as far back as Barclays reports High Yield OAS) - %


As the chart above highlights, credit spreads can be highly cyclical, which I’ll bucket as: ­
  • Tight: More than one standard deviation below average ­
  • Normal: Within a one standard deviation band ­
  • Wide: More than one standard deviation above average 
Are tight or wide spreads a better indicator for forward risk taking? Let’s take a look.


Credit Spreads vs Longer-term Returns 

Bucketing each starting month-end period into tight, normal, or wide buckets, the forward five-year average performance of investment grade corporate bonds, high yield corporate bonds, and the S&P 500 for the 1997-2018 time frame is shown below (1997 is the first data point for the bands three years forward from the 1994 starting date).


The result is that while riskier asset classes returned more on average over the whole period: ­
  • Forward five year returns of all three asset classes were noticeably lower when the starting yield of the “quality spread” was low ­
  • In fact, both high yield bond and the S&P 500 average returns were less over the subsequent five years than the returns on the Treasury index when the starting spread was more than one standard deviation below average
  • When the “quality spread” was elevated, average excess performance was exceptionally high in both absolute and relative terms five years forward

In summary… over these longer-term windows, a low spread = a lower return (which would seem to indicate longer-term investors may currently be taking excess / uncompensated risk).


Credit Spreads vs Shorter-term Returns

Using the same 1997-2018 time frame and spread buckets, we can see that over the short-term (one-month forward time frame) the opposite narrative appears to be true: ­
  • Risk assets performed better when spreads were tight than when spreads were wide ­
  • In fact, the best short-term period for stocks were when spreads were more than one standard deviation below their average ­
  • Both high yield and stocks performed worse than Treasuries when spreads were wide 

In summary… over the shorter-term, low spread seems to = a higher return (which would seem to indicate investors may be more than fairly compensated to take risk given current fundamentals).



RISK / RETURN BY THE LEVEL AND DIRECTION OF SPREAD

The below charts break out investment grade bonds, high yield bonds, and S&P 500 further, charting the return (geometric one month forward annualized) ­and risk (standard deviation) by:
  • Spread levels (narrow, normal, or wide) ­
  • Spread direction (i.e. whether the "quality spread" has narrowed or tightened) 

Investment Grade Corporates

Geometric Annualized Forward One-Month Return vs Month-End Starting Yield / Direction of Spreads 


Investment grade corporates provided positive performance in all of the various environments in this time frame, but volatility did pick up when spreads were both elevated and widening (as a frame of reference, IG Corporate bonds did outperform Treasuries in all of these environments except when spreads were elevated and widening - a period where they underperformed by 6%).


High Yield Corporates

Geometric Annualized Forward One-Month Return vs Month-End Starting Yield / Direction of Spreads


The results for high yield seem more interesting. High yield returns were relatively steady in both low spread and “normal” spread environments, but the volatility of high yield was materially lower when spreads were narrow. Things were especially interesting at higher spread environments as there was: ­
  • Underperformance: when spreads were wide and widening (“catching a falling knife”) ­ 
  • Outperformance: when spreads were wide and narrowing (an investor was able to successfully capture these higher than normal yields in this window) 

US Stocks

Geometric Annualized Forward One-Month Return vs Month-End Starting Yield / Direction of Spreads 


S&P 500 returns were especially strong when spreads were narrow, as well as when spreads were “normal” and moving wider (perhaps noise over the previous month presented a buying opportunity). What I found interesting was the linear relationship between the level of spread and volatility (narrow spreads = much lower volatility). Finally, I found it interesting that the performance of US stocks was poor when spreads were elevated (irrespective of whether spreads were narrowing or widening, unlike the divergence in performance within high yield).


SUMMARY

The good news is the above analysis may provide some interesting signals for those with the flexibility to allocate tactically.

The bad news is that while I think the relationship between spread and the shorter-term performance outlined above makes logical sense, it may not work going forward.

So the next time someone asks for a quick answer as to whether tight spreads have been a sign of exuberance among investors who were willing to overlook risk (which will end in tears) or an environment of high economic growth and low default rates, supporting risk assets... you can now respond.

It depends.

Wednesday, June 20, 2018

CAPE of Good Hope? P/E Divergence as a Performance Signal

Lawrence Hamtil recently shared a Vanguard paper with me that was surprising given it indicated the trailing twelve month price-to-earnings ratio "TTM P/E" was nearly as strong a predictor of forward 10-year equity returns as the cyclically adjusted price-to-earnings "CAPE" ratio going back to 1926. My assumption had been that the CAPE ratio (which uses smoothed 10-year real earnings) would be the much better of the two ratios given it reflects the longer-term earnings power of companies within the index, rather than the (potentially at times) cyclical peak. 

This post will dig into:
  1. the historical relationship between the TTM P/E and CAPE ratios and forward returns
  2. the historical relationship between the TTM P/E and CAPE ratios, and how that relationship has changed in recent years
  3. how these ratios may potentially be used together to help predict shorter term market performance

Backdrop: The Surprising Predictive Power of TTM P/E

While perma-bears seem to enjoy highlighting metrics (debt, rates, growth rates, etc...) that have no predictive power for either short or long-term forward equity returns, valuations themselves have mattered. As I’ve highlighted in previous posts, higher valuations (as defined by an elevated CAPE ratio) have historically resulted in lower long-term forward returns. Vanguard replicated this result for trailing P/E, which surprised me given the backward looking / shorter-term / cyclical nature of the TTM earnings component in the denominator of the P/E ratio.

Per Vanguard:
We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio).
  

Given my need to replicate anything I see to personally believe, the below charts replicate this analysis with a scatter plot for each updated through May 2018 (the Vanguard piece is through 2011) using data from Shiller (the dotted line shows the ratio as of May 2018). We see both relationships remain strong, though the CAPE’s predictive power has improved quite a bit (more on that below) since 2011.


The Changing Relationship Between the CAPE and TTM P/E

The reason the CAPE shows a higher predictive power in updated results is due to the divergence of the two ratios leading up to and through the global financial crisis “GFC” when earnings collapsed, causing the TTM P/E to spike, which in turn made the US equity market seemingly more expensive as it sold off. 

Meanwhile the US equity market appeared quite cheap on a CAPE basis (it hit a ~30 year low), which turned out to be the correct signal. In the following chart we can see the close relationship between the two ratios following the Great Depression through late 1990’s, then the divergence seen first during and after the technology bubble (note the chart stops at 50 to show the data more clearly, but the TTM P/E spiked to 86 in October 2008).



The impact of this divergence is especially clear in the rolling ten year correlation of the TTM P/E and CAPE ratios.


As a result, in the more recent periods that capture the Internet Bubble and/or GFC at the back, middle, or the front of a 10 year rolling period, the CAPE has been extremely predictive (89%), while the TTM P/E has been less so


The Potential Use of the CAPE and TTM P/E to Make Allocation Decisions

The following chart shows the difference between the two ratios over time. We can see that for a ~60 year window following the Great Depression to the beginning stages of the Internet Bubble they moved together closely. We can also see the more recent divergence.


And this is where I think things get interesting and potentially less intuitive.

Historically, when the CAPE was elevated (meaning markets were potentially at risk from a valuation standpoint) and the CAPE > TTM P/E (meaning recent earnings in the TTM denominator are higher than the smoothed 10-year real earnings), forward short-term performance has been just fine. It's when the CAPE was elevated (again… meaning markets were potentially at risk from a valuation standpoint) and CAPE < TTM P/E (meaning recent earnings have lagged the smoothed 10-year real earnings) that short-term performance hasn't just been poor, but outright negative.

In fact, looking back at the chart above we can see the CAPE ratio exceeded the TTM P/E by a substantial margin before the major market corrections of the Great Depression, Internet Bubble, and GFC, but when the CAPE flipped below the TTM P/E is when each sell-off really took hold. Note that in the Great Depression the US equity market continued to sell-off even after the CAPE got to seemingly attractive levels.


My takeaway from all of this remains that forward long-term returns are likely to be low relative to history (both CAPE and TTM P/E point to that likelihood), while the shorter-term outlook looks better. Investors tactically holding US stocks may be well served by what has historically been strong equity performance in elevated valuation environments when current earnings remain strong and/or the upward trend of the market stays intact. But buyer beware... should either earnings or the positive trend of the market shift, current valuations increase the risk that this may end up viewed as a period of calm before the storm.

Wednesday, January 31, 2018

The Behavioral and Performance Benefits of Trend Following

When we tell our investors to invest for the long run, we have to make sure the short run doesn’t kill them first… Investing for the long run isn’t bad advice, it’s just unrealistic. It doesn’t take into account human behavior.

-Andrew Lo (HT: Andrew Thrasher)


Trend following has historically provided strong long-term returns with materially reduced drawdowns relative to a traditional buy and hold investment, but none of this matters if an investor cannot stick with the strategy through periods of relative underperformance. This “opportunity cost” is often felt the most during periods when more traditional allocations outperform.

The consistency of a trend following strategy’s relative performance vs a 60/40 portfolio (impacting the ability for investors to stick with trend following) is the basis of an argument that’s taken place offline (yes, I also argue offline) with a FinTwit friend who is a huge proponent of buy and hold. It’s progressed to the point that we’ve discussed making a mini (very mini) Buffett style bet related to whether trend following or a 60% US Stock / 40% Bond allocation will outperform over the next five years (with money going to the winner's charity of choice).

Given we've hit a dead-end due to his view that the result will be a coin flip (i.e. random whether trend or buy and hold outperforms, thus even if he were to lose it would be random as well), I thought I would put my case forward in this post outlining why I think trend following has a much higher probability of outperforming a 60/40 portfolio in most environments and especially in the current environment.


Backdrop: What is Trend?

As outlined in a previous post:

In a nutshell, trend following is simply a means of determining if you will own an asset based on its recent price history. 
One simple set of trend following rules are: ­
  • If the S&P 500 Total Return Index > 12-Month Moving Average, Own Stocks ­ 
  • Otherwise Own Bonds
The diagrams below depict how those without an understanding of trend following often believe it works vs how it really works. The original image on the left is used with permission from Carl Richards and outlines how poorly behaved investors often act, while my revised version on the right outlines what trend followers attempt to do (follow Carl on Twitter @behaviorgap).


Buy and hold investors seem to perceive trend followers in a similar light as these poorly behaved investors, chasing strong returns higher and selling out once markets have completely rolled over. In practice, trend following buys / sells after major turning points, thus gets back into markets once a new trend forms and holds until there is another turning point... often much later than when the investor would have otherwise preferred to sell. The opportunity cost of trend following is the willingness to miss initial turns and to be wrong over many intermediate periods until a new long-term trend emerges.


How Often Does Trend Following Outperform?

As the chart below shows, trend-following outperformance has occurred at a much higher rate than a coin flip and that beat rate has increased over longer periods. This specific 12-month trend model outperformed a 60/40 portfolio over ~80% of rolling 60-month time frames since 1926 and 90%+ of the time over 10 and 15 year periods. 


To get a better sense of when these periods of outperformance and underperformance occurred, the following chart breaks out when trend following outperformed a 60/40 portfolio (blue) and when it did not (white) over rolling 60-month time frames. 


A few things to note:
  1. the trend following model outperformed a 60/40 portfolio consistently
  2. the trend following model outperformed over extended periods of time
  3. periods of trend following underperformance were short lived and clustered
  4. periods of relative underperformance were more likely to occur when the opportunity cost associated with moving away from a 60/40 portfolio were high
Specific to point #4, the chart below shows the historical yield of a 60/40 portfolio (i.e. what can be viewed as the likely opportunity cost of trend following). We can see that a high starting yield that compressed quickly in the mid 1930's coincided with the underperformance that took place in the late 1930's / early 1940's, while the huge yield compression of a 60/40 portfolio in the mid 1980's coincided with the challenging relative performance in the late 1980's / early 1990's.



Current Expectation

Simply put, the starting yield / opportunity cost of a 60/40 portfolio is extremely low at ~3%. Historically, when we split the universe into buckets when the starting 60/40 yield was above or below the long-term 6% average (3% higher than the current level), we can see the increased likelihood a trend following strategy outperforms at lower starting yields. In fact, the historical beat rate over rolling 5 year time frames moves up to 85%.


Conclusion

I really like the way my friend Wes from Alpha Architect framed this decision:
Flip it and make trend following the benchmark and consider buy and hold. Works sometimes, doesn't work other times, but you eat massive tail risk with buy and hold, therefore isn't worth the risk/effort.
In other words, if trend following is your base allocation, would you as an investor allocate to a different strategy that underperformed at a 70%+ rate over 3-5 year rolling time frames and 90-100% of rolling 10-15 year time frames, and had materially more downside risk? Not only would I not make that allocation, I'd love to bet someone with proceeds going to charity that trend following would outperform.



Appendix: Tax Efficiency of Trend Following

A common question that comes up related to trend following is its tax efficiency. The reason I ended up ignoring the tax issue as a detriment to trend following in the analysis is that trend following may actually be more tax efficient than a buy and hold allocation that includes taxable bonds.

A few reasons:
  1. A trend following solution can be structured utilizing futures, which are taxed at 60% the long-term rate and 40% the short-term rate, making it pretty similar to a 60/40 portfolio
  2. Even with cash holdings, trend following can be tax efficient as the large gains in stocks are often held much longer than 12 months (i.e. are taxed at the long-term rate). The chart below shows the length of each past isolated trend, along with the gains earned in these trends showing that significant gains were often taken years after the initial trend signal.