No. of Recommendations: 3
The Research Affiliates article is not behind a paywall. One might have to 'register', I do so with a fake name and burner email. But these tend to be smart guys, so stuff they might send you might be interesting stuff.
Here's a quick synopsis:
A recent example illustrates the difference between CC CAPE and Shiller CAPE. In December 2020, the S&P 500 Index added Tesla and deleted Apartment Investment and Management (AIV). As we noted at the time (see here and also here), it was a classic example of a cap-weighted index buying high and selling low.
The calculation of Shiller CAPE would not have included Tesla's earnings losses in the decade before being added to the index, but it would have included AIV's modest earnings. The denominator would have been larger as a result of including AIV's positive earnings instead of Tesla's negative earnings. In contrast, the calculation of CC CAPE would have excluded AIV's earnings history as soon as the stock was dropped from the index and would have included Tesla's previous decade of negative earnings as soon as it was added. This approach would have made the denominator smaller. If Tesla's market cap at the time of inclusion was also proportionally smaller than that of AIV, the impact on CC CAPE versus Shiller CAPE would not have been so large. On the contrary, Tesla's market cap at the time of inclusion dwarfed that of AIV, exacerbating the difference and making CC CAPE higher than Shiller CAPE.
They plot the CC CAPE ('Current Constituent' CAPE) and Shiller CAPE over time, showing that they differ significantly, on same plot is the difference of CC - Shiller i.e. which they call the 'spread'.
They show that CC_CAPE predicts signifcantly better than Shiller_CAPE over 3 and 5 year future return horizons
Ultimately, the best explanatory power for future returns comes from combining Shiller CAPE and CAPE Spread because this combination contains information about both starting-point valuations and sentiment. As the table below shows, combining them (i.e., Shiller + Spread) more than doubles the correlation with future returns at the shorter horizons of 1, 3, and 5 years compared with using Shiller starting valuation alone.
The use a notation of "Shiller + Spread", which if taken literally is actually just their CC_CAPE.
Perhaps they mean they do a fit of Shiller and Spread to future returns, like this:
FutureReturns = a*ShillerCAPE + b*Spread + c
where a, b, c are adjustable constants.
If so, it would have been useful for them to fit a, c, b using data only up to a given date, i.e. walk the date of evaluation forward in time using only data available to that date to fit the coefficients.
Later they show a table that shows that CC_CAPE predicts a somewhat rosier return future than does Shiller CAPE.
