This post is available as a PDF download here.

 

Please see our updated post Disproving a Signal.

Summary­

  • Value investing has not only underperformed with regard to security selection, but also country selection over the last decade.
  • In an effort to avoid country value traps, we set out to design two signals that might better confirm when a country is likely to exhibit positive re-valuation.
  • We find that one of the signals exhibits curious results, leading us to develop an entirely new metric for country rotation.
  • Initial tests indicate that the signal appears distinct from both traditional value and momentum models.
  • From 1976 – 2019, a dollar-neutral long/short portfolio that implements this signal exhibits a gross-of-cost 3.7% annualized return with a 9.6% annualized volatility, implying a Sharpe ratio of 0.39.

As has been well publicized, it has been a tough decade for value.  Not just in the realm of stock picking, either: pretty much value anywherehas been a tough go.

For example,despite the ample evidence suggestingthat valuation-driven country rotation works, it certainly has not worked as of late.  To demonstrate, below we plot dollar-neutral long/short portfolios that capture “cheap minus expensive” for developed global markets.12

Source: MSCI and Bloomberg.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

(It is worth noting that none of these long/shorts are actually statistically significant at the 5% level, even if we rewind the clock back to 12/31/2006. Nevertheless, we will charge blindly forward.)

Lawrence Hamtil has written quite a bit about how relative valuations may be distorted by sector-based differences in country index composition.  One idea for correcting for this is by first normalizing each country’s valuation to its own historical valuation distribution.  This might help adjust for structural level differences in country valuations that may emerge for sector-driven, sentiment, demographic, or other reasons.

Therefore, before using our value signal to rank across countries, we first transform it by calculating a z-score using the country’s prior valuation data.  In theory, this should help better normalize the metric and identify meaningful deviations from that country’s own “normal” valuation levels.

Source: MSCI and Bloomberg.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

We can see that self-normalization does little to help the problem (and, arguably, made valuation-driven country rotation worse).

The problem here might simply be more trivial: expensive just has out-performed cheap.

Fighting the Zeitgeist

Share prices fall when market participants adopt a negative view on a security, industry, or country’s fundamental outlook.

The relationship is clear if we adopt a simple dividend growth model for share prices; i.e. P = D1/ (R – G), where P is price, D1is the annual dividend level, R is the required return, and G is the growth rate.  As G declines, the denominator increases, and price decreases.

Note that if Dividend1is held constant and the growth rate declines, we will expect price to also decline, which will lead to an increase in yield.  And if we replace D1with Payout Ratio x Earnings1, we can also see a clear picture of how earnings yield and growth rates are similarly related.

Re-arranging the formula to solve for return, we find that R = (D1/ P) + G. This formula tells us that return should be equal to dividend yield plus our long-term dividend growth rate. Thus, if our requiredrate of return does not change and the outlook for a company’s growth declines, we need to see a decrease in a price to create a commensurate increase in yield.

Taken together, we can see that as the market outlook for a given security deteriorates, valuation multiples (e.g. P/E) should decline.  Should earnings growth realities prove to be less gloomy than forecasts imply, however, price will appreciate and valuation multiples will expand.

We should also note that a change in aggregate risk preferences among market participants should have a similar effect.  A decrease in risk appetite should manifest in an increase in the required rate of return. In our first formula, this informs us that price will decrease.  In our re-written formula, if our outlook on growth does not change but our appetite for risk does, then price must change to affect yield.

However, given that fundamental measures tend to be less volatile than price, an investor who purchases a security at a low valuation multiple and sells at a high valuation multiple will earn a return not necessarily due to higher realized yield, but rather a re-rating of future fundamentals that materializes in a price increase.

For an investor purchasing a stock today, we might say that their future return is: R = (D1/ P) + G + V, where V capturers re-valuation over the holding period.

What does this all have to do with country rotation?  Re-valuation will be driven not only by changes in growth outlook, but also changes in risk sentiment.  For example, an aggregate decrease in risk appetite among market participants will appear as a contraction in valuation multiples, while an aggregate increase in risk appetite will appear as an expansion.

Thus, our question: can we avoid country-level value traps by understanding country-level sentiment?

Measuring Sentiment with Value Dispersion

To measure evolving sentiment / risk appetite for a given country, we will examine two different metrics:

  1. Value/Growth Valuation Spreads: High valuation dispersion between growth and value indices may reflect a better entry point for buying cheap countries.Specifically, for each country we calculate the trailing 12-month yield of that country’s value index minus the trailing 12-month yield of the corresponding growth index.  We then subtract the long-term mean (using an expanding window) to better identify cyclical changes.
  2. Value/Growth Momentum: When value is out-performing growth within a country, it may reflect a better opportunity for capturing re-valuation changes in cheap countries. Specifically, for each country we will calculate the trailing 12-month return of that country’s value index minus the trailing 12-month return of the corresponding growth index.

To construct our portfolios, we will first rank each country on the value metric.  We will then rank each country on the sentiment metric.  We will then add these ranks to create a composite score, and re-rank on this composite score and construct our dollar-neutral long/short as before.

Source: MSCI.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

Well, that’s tragic.

Growing Sentiment

Something is a bit odd though.  The addition of the value/growth momentum metric turned a positive (albeit statistically insignificant) return stream negative.  This is a bit of a head-scratcher and perhaps worth exploring more deeply.  What does this signal look like on its own?

Below we plot a dollar-neutral long/short constructed on this signal alone (i.e. rank countries based upon prior return spread of respective growth and value indices).

Source: MSCI.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

That return stream is so bad, it might just be good.

After all, if we can short this return stream, we could have a pretty good result on our hands.  And in this case, shorting the return stream is effectively the same as just inverting our signals.  In other words, instead of ranking on the prior return of value minus growth, we should rank on the prior return of growth minus value.

Source: MSCI.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

But is this signal just another way to extract momentum?  After all, we are using a measure of prior return.  To answer this question, we computed a prior 12-month return signal on country indices explicitly and constructed the corresponding long/short index.  Below we plot the 36-month realized correlation between the indices constructed on TTM yield signals (“Value”), prior 12-month return signals (“Momentum”), and the 12-month return spread between growth and value indices (“G/V Momentum”).

Source: MSCI.  Calculations by Newfound Research.  Returns are hypothetical and backtested.  Returns are gross of all fees, transaction costs, and taxes.  Returns assume the reinvestment of all distributions.  You cannot invest in an index.

While both Value and Momentum have sporadic periods of strongly positive or negative correlation to G/V Momentum (with the mirror-like relationship reflecting the well-documented negative relationship between Value and Momentum), we can see that, for the most part, absolute correlations are benign.  Furthermore, the correlations are meaningfully time-varying, suggesting that there might be unique information found in the G/V Momentum signal.

What is most curious about this signal is that the strategy does not invest in the spread itself.  We’re not saying, for example, “when growth is outperforming value, buy the country’s growth index.”  Rather, a higher (lower) cross-sectional return spread between growth and value implies a higher (lower) cross-section return for a given country index.

Why that relationship exists is certainly worth pondering.

One possible explanation is that the prior return of growth minus value captures some measure of market cycle sentiment.  An increasing spread may reflect an increasing intra-market risk appetite for growth stocks over value stocks, potentially signaling a positive sentiment shift.  This relative appetite change may, therefore, foreshadow a broader, market-wide increase in risk-appetite, leading to a positive re-valuation of economic expansion.

Conclusion

In this commentary, we set out to fix value-driven country rotation, an approach which has largely exhibited negative returns for the last decade.  To do so, we designed two signals that aimed to capture internal market sentiment for a given country.  The first signal – measured as the spread in yield between value and growth – aimed to capture intra-country value spreads.  The second signal – measured as the prior total return of value minus growth – aimed to capture the intra-market risk appetite for value stocks.

In combining these signals with country value signals, our hope was to buy cheap countries with market internals that implied a greater likelihood of re-valuation.

Unfortunately, we failed miserably.

In our failure, however, we identified that as a stand-alone signal, the prior return of value minus growth lead to a long/short return profile that lost money consistently.  In flipping the signal to the prior return of growth minus value, we are able to generate a long/short strategy that not only exhibited an attractive return profile but had also exhibited meaningfully positive returns over the last decade.

Interestingly, it appears that the signal is not merely a proxy for momentum and exhibits low realized correlations to both momentum and value strategies, indicating that it might contribute unique information on its own.  What, precisely, this signal is capturing remains a mystery, but we believe it warrants further research.

Identifying the whybehind a signal can be particularly risky, as we can fall prey to constructing a narrative to fit a signal, rather than identifying a signal from a starting hypothesis.  This risk might be heightened in this case where the signal we identified was specifically the antithesis of the hypothesis we set out to test.

Even if we were convinced of the why, it is worth noting that this brief note only scratches the surface of research that would need to be performed before even considering implementing a signal like this.  For example:

  • We should determine how sensitive the strategy is to changes in the investment universe. Not only should we see if we can replicate this signal in an entirely different universe (e.g. timing sectors with intra-sector growth vs value), but also determine how sensitive the results are to changes in the universe examined (e.g. robustness tests via subset resampling).
  • We would want to determine strategy sensitivity to model specification. This would require changing not only the return model employed (e.g. risk-adjusted returns, regression slopes, or short-minus-long moving averages) as well as the formation period.
  • We would want to determine strategy sensitivity to rebalance timing luck, as the model examined in this brief note only evaluated end-of-month rebalancing.
  • We would want to quantify the decay speed of the signal to better determine the appropriate holding period.
  • We would need to identify how much return is available after cost and tax adjustments, as well as appropriate discounting for data-mining risk.
  • We would want to explore the shape of quintile returns and volatility profiles to determine what type of signal we’re working with. For example, is the excess return potential generated from a monotonic increase in returns across quintiles, or is it simply from avoiding the worst quintile?  This will have important implications for portfolio construction as well as whether this approach can be ported into a long-only model.

It is only after we are comfortable that this signal is reasonably robust across specifications and survives after conservative cost assumptions are applied should we consider employing it.

 


 

  1. Countries include Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland, Israel, Italy, Japan, Netherlands, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, United Kingdom, and USA; inclusion is based upon available data.
  2. The portfolio employs twelve overlapping sub-indexes, each which rebalances monthly on a different month of the year.  Weight for a given country, wi, with rank riis calculated as: wi= c(ri– average(r)), where cis a scaling coefficient set so that the portfolio is dollar neutral.

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. Or schedule a time to connect.