Subject: Re: OT Grantham Sounding AI Alarm
It looks like you have about 1,100 weeks of data (Jan 1995 until Dec, 2015). If not too big of an ask, could you show the other deciles? The first half of the sample (1995 plus 4 to 10 years until 2005 plus 4 to 10 years) includes the Dotcom bust and the GFC, so a pretty tough period.

Actually I have about 6600 weeks, but I figured it's of limited relevance to modern markets. Everything averaged much cheaper back in the day.

For this table I didn't use deciles, since all the interesting stuff happens near the extremes.
Possible summary: really high valuations are a poor bet, but only REALLY high ones. Moderately expensive valuations lead to a wide variety of outcomes. Cheap valuations do very well quite reliably, even "pretty cheap" ones (and also, not visible in this table, cheap valuations tend to pay off pretty quickly).

All figures include reinvested dividends and are adjusted for inflation.

                                 Lowest    Pctl 10    Average    Pctl 90    Highest
Most expensive     1% of time     -4.1%      -3.9%      -3.5%      -3.0%      -2.9%
next               3% of time     -3.9%      -3.7%      -3.1%      -2.4%      -2.2%
next               6% of time     -3.4%      -3.1%      -2.1%      -1.2%      -0.3%
next               8% of time     -2.4%      -1.8%       0.1%       2.1%       2.3%
next              16% of time     -0.5%       0.3%       3.0%       7.4%       8.7%
next              16% of time      1.4%       2.0%       5.5%       9.3%       9.8%
next              16% of time      2.0%       2.7%       7.4%      10.5%      11.2%
next              16% of time      3.5%       4.2%       9.4%      12.5%      12.9%
next               8% of time      5.3%       8.6%      11.9%      13.4%      14.0%
next               6% of time     10.6%      11.8%      12.7%      13.8%      14.6%
next               3% of time     12.7%      13.3%      14.0%      14.5%      14.7%
Cheapest           1% of time     14.9%      14.9%      15.7%      16.6%      16.9%

And I should add that the prior comment about poor past returns starting at valuation levels similar to the current levels since 1995 is more than a bit misleading, since there were few such times. Most of those starting weeks were during or near the dotcom bubble, or were so recent that there aren't 4-10 year forward results in yet, so it's really a sample of one bubble if you squint.

The valuation method used to divide up the weeks is very similar to CAPE, just that the smoothing is a bit smoother than the 10 year average traditionally used.

Jim