Cliff Asness, Antti Ilmanen, and Thomas Maloney of AQR are out with a new paper about market timing with value, titled: Market Timing: Sin a Little.

Specifically, the paper explores whether the Shiller PE (also known as the cyclically adjusted P/E, or CAPE) can be effectively used to directionally time equity market exposure.

Hot off the presses, we wanted to provide our take on the results.

The Value Timing Opportunity

At first blush, the opportunity for timing seems massive.  A simple sort shows that rich markets perform poorly while cheap markets perform well.

Returns by CAPE Quintile

The authors are quick to point out, however, a key problem with this graph: it has embedded hindsight bias.  The graph uses the full period of valuation data define the quintiles, which is data an investor would not have had in real time.

Nevertheless, using rolling 60-year windows seems to still suggest some power in using valuations to predict forward returns.

Returns by CAPE Quintile Rolling

We would like to humbly submit our own flaw in this data.

While this approach of evaluating the opportunity is common for factor investing, it may not apply to market timing.

The subtle difference is that in factor investing, securities are ranked simultaneously and then held for a given period.

The performance of a factor quintile, then, represents an implementable strategy.

Here, the performance of a quintile is the arithmetic mean rate of return of all the associated time periods.  Time periods which may be overlapping.

Consider the early 1980s, a period of incredibly low Shiller PE value.  We would expect both 1980-1990 and 1980-1991 to be in the cheapest quintile.  But how would you actually implement that trade?

We’ll return to this point later.

AQR’s Approach to Value Timing

To test whether value can be utilized as a timing mechanism, AQR takes the following approach to calculate the portfolio allocation to equities.

On a monthly basis, using the prior 60-years of Shiller PE data,

  1. Trim the current Shiller PE (so it stays within the 5th and 95th percentile bands)
  2. Calculate (trimmed Shiller PE – median Shiller PE) / (95th – 5th percentile range)
  3. Add the value to 100%
  4. Floor the value at 50% and cap it at 150%

This approach creates a tactical equity strategy that pivots around 100% exposure based upon how cheap or rich equities currently are.

As a side note, we think it is important to point out that this test was designed symmetrically.  All too often we see tactical approaches benchmarked in such a way that they only benefit from calls in a single direction.

Consider if AQR had designed this test with a floor of 0% and a cap at 100% and compared it to a 100% equity benchmark.  The only way the portfolio could add value above the benchmark would be during periods of significant pullback from high equity valuations.  Otherwise, the best the portfolio can do is keep up.

Disappointing Initial Results

Initial results presented in the paper, by the authors’ own admission, are not spectacular.  Particularly in the latter half of the 1900s, value timing appears to add little value.

Initial Paper Results

These results appear to be because the strategy was largely underinvested in the latter half of the century, with only an average position of 89% exposure since 1958.

This is akin to only collecting 89% of the equity risk premium for over five decades.  Even if your timing is impeccable, that’s still a lot of ground you need to make up.

The authors adjust for this by shifting the signal over each period to give an average position of 100%, ex post.

While unrealistic (as you cannot go back in time to adjust your average position size), it does help explore whether value timing is at fault or whether the realized draw of history was biased against a value timing approach.

The authors find that, indeed, the underexposure may be at significant fault for the failures of value timing over the last 50+ years.

Ex Post Average Adjustment Results


Momentum + Value: A Winning Combination

A potential headwind faced by contrarian timing strategies is the short-term momentum effects that are commonly exhibited by markets.

Running a simple ordinary least squares regression, the authors find that indeed there is a significantly negative beta component to the momentum factor.

OLS Results

The implication of these results is that, like using value and momentum in cross-sectional security selection, combining value and momentum in market timing may be an effective approach.

Indeed, using momentum to help curb the enthusiasm – in effect creating a more patient value investor – appears to help make the approach value-add over the full period.

VM Timing

The Patient Value Timer

AQR also explores a second way of being a patient value timer: an approach we’ve been advocates of for a very long time.

To help reduce exposure to short-term momentum, AQR uses a moving average to “slow down” the value signal, which is the equivalent of making overlapping value bets.

This is an approach we’ve been advocates of for quite some time.

The authors test the approach using a variety of lags and find that holding 2-5 years maximizes the value timing Sharpe ratio while simultaneously neutralizing the negative correlation to the value factor.

Overlapping Value Timing Bets

We believe the benefit of implementing via this approach is that it explicitly addresses the construction flaws that make Exhibit 2 irreproducible in a traditional manner.  Given that the results are based upon overlapping time periods, we believe that the strategy must too use overlapping bets.

A Variety of Value Timing Variants

In the appendix, the authors test a number of approaches to value timing, with varying results.

Appendix Results

We would draw your attention, specifically, to two of the tests.

First is “EP – Real Bond Yield.”  In this approach, instead of using Shiller PE, the Shiller EP (earnings yield) minus the 10-year real yield (nominal yield minus forecasted inflation) is utilized.

We think this is an important control step because it acknowledges that PE is comprised of two components: Price and Earnings.  Traditional discount cash-flow models or dividend growth models hold that all else held equal, when discount rates decline, prices go up (this was a topic we explored in Are Stocks Actually Undervalued?).

Second is “EP 10-Year Lock-in” which uses the smoothing method mentioned above to create overlapping value bets, and should create results more inline with those found in the paper’s Exhibit 2.

We believe this overlapping bets method is one of the fews ways to actually create an implementable strategy that should match the results seen in that figure.  Without overlapping bets, there is no way to address the overlapping data.


Timing with Value: Worth It?

The results of the paper appear to point towards a hesitant yes, with the caveat being that some sort of approach likely needs to be embedded to help investors stay patient in their contrarian enthusiasm.

While incorporating momentum seems to be the most effective means, even simply using a moving average on the value signal (creating overlapping value bets) appears to work.

Nevertheless, the value add in both cases is fairly marginal and so the advice of the authors returns to the title of the paper: if you’re going to sin at all, sin only a little.


Our Take

First, we would like to thank Cliff, Antti, and Thomas for their valuable contribution to the literature.

We would argue that the initial strategy implementation in the paper is not designed to capture the opportunity set laid out in their introduction.  While our preferred method of addressing this problem would be through overlapping bets (to keep the approach as a pure value study), the authors leverage momentum as a unique means to create a patient contrarian investor.

The use of value and momentum together as timing mechanisms introduces an interesting question.  In Capital Efficiency in Multi-Factor Portfolios, we demonstrated that cross-sectional security selection using multiple factors simultaneously was beneficial because it created implicit leverage.  Arguably, combining value and momentum signals together as timing tools does the same: you are using a single dollar to implement two timing systems.

However, the math is complicated because with timing you do not retain full investment as you do with a multi-factor approach.

So the question we have is: is it that momentum makes you a more patient value investor, or does the implicit leverage help the portfolio overcome the under-allocation drag from the 1960s onward?

Perhaps both.

We also feel that the appendix points to some interesting areas of future research.  Specifically, we would like to see 2-5 year lock-in periods explored, as well as a combination of approaches used together.

For example, we expect that a 2-5 year lock in period using Shiller EP – Real Bond Yield as the signal could not only better align the strategy with the opportunity, but also help prevent false positives where price is artificially inflated due to low real bond yields.

All in all, a stellar piece worth the read.

Corey is co-founder and Chief Investment Officer of Newfound Research, a quantitative asset manager offering a suite of separately managed accounts and mutual funds. At Newfound, Corey is responsible for portfolio management, investment research, strategy development, and communication of the firm's views to clients. Prior to offering asset management services, Newfound licensed research from the quantitative investment models developed by Corey. At peak, this research helped steer the tactical allocation decisions for upwards of $10bn. Corey holds a Master of Science in Computational Finance from Carnegie Mellon University and a Bachelor of Science in Computer Science, cum laude, from Cornell University. You can connect with Corey on LinkedIn or Twitter.