I often read that valuations don't matter over the short-term (a case often cited against market timing). Over very short periods (hours, days, etc...) this certainly may be true, but while there can be a lot of variability around month-to-month or year-to-year performance, I completely disagree with the sentiment that it doesn't matter. That said, there are better ways than just using current valuation levels for an investor to time markets which I will outline below.
Debunking the Case that Valuations Don't Matter
Before getting into my analysis supporting the theory that valuations do in fact matter a LOT, especially when you use the information embedded in momentum as support, I did want to highlight a recent research paper / article outlining valuations don't matter that simply appears to be wrong.
Larry Swedroe from ETF.com shared research in an article Valuation Metrics In Perspective outlining the research done by Javier Estrada which concluded that one year forward performance of the S&P composite from 1899-2014 was independent of preceding valuations:
when the current P/E was between 10.4 and 13.3, the one-year forward return was 7.3 percent. When it was higher, between 16.4 and 18.9, the one-year forward return averaged 11.7 percent. And when the current P/E was above 19, the one-year forward return averaged 10.0 percent.
While I completely agree that one year returns are noisier than 10 year returns, the above results truly surprised me. While any one year period could have much higher / lower returns (i.e. the range of outcomes should be wider), over long periods of time the average one year forward performance when valuations are cheap shouldn't be that different than the average 10 year forward performance (it should average out to roughly the same figure).
Given my surprise, I decided to recreate the results in the paper using Shiller CAPE for my P/E over that same 1899-2014 time frame. The results in the table below make it likely that the analysis in the paper is either flawed (they do use a weird 3-month lag for earnings) or is simply wrong (my personal view). As the table outlines, the results not only show that high valuations = lower returns over one year, but the average one year return is even more dependent on valuations (on average) than ten year returns. You can also see that the range of outcomes is much more dispersed over one year than ten years.
Back to the Original Point of the Post... Momentum to Time Value
I had run the analysis below prior to my attempt at recreating the above data, so my methodology is a bit different. In the example below, I use data going back to 1881 (which is aligned to the inception of the Shiller CAPE time series) and I separate one month forward returns by whether or not the S&P composite had a one year backward looking return that was positive or negative. Below are the same data points in a table form and in a scatter plot.
What is easily discerned is that regardless of momentum, cheap valuations are generally good for returns (though extremely cheap and negative momentum has had some severe volatility - see standard deviation for periods when CAPE < 10), while high valuations have generally resulted in pretty strong returns when momentum is strong. Valuations seemingly only matter when valuations are high and momentum has turned negative (as seen by the negative absolute performance for the three buckets when CAPE > 15 and momentum is negative).