Valuation-Informed Indexing #358 on the S&P 500 and how it impacts safe withdrawal rates for retirees
By Rob Bennett
Michael Kitces was recently interviewed on his views re safe withdrawal rates at the Madfientist blog. He offered an interesting observation about the correlation between valuations (the P/E10 level) and long-term returns that I had never considered before.
The correlation between valuations and long-term returns is the backbone of the Valuation-Informed Indexing strategy. Prior to 1981, the thinking among investors who believed in research-based strategies was that the market is efficient, that is, that stock prices fall in the pattern of a random walk (hence the title of the famous book promoting the Buy-and-Hold concept, A Random Walk Down Wall Street). What made Robert Shiller’s “revolutionary” (his word) research findings of 1981 so revolutionary is that he showed that that’s not so. Yes, stock prices really do fall in the pattern of a random walk over the course of one or two or three years, Shiller showed. But there is nothing even a tiny bit random about the walk of stock prices when you look 10 years out or 15 years out or 20 years out.
Over those longer long-periods, stock prices are strongly correlated to the P/E10 level that applied on the day the time-period being considered began. In other words, the market is not efficient and prices are not determined by unforeseen economic developments. In other words, investors are not rational actors pursuing their self-interest but highly emotional humans, driven by a Get Rich Quick impulse to let prices rise to unsustainable levels and then by a common-sense fear over what they have done to send them crashing back to fair-value levels or lower.
There is a good bit of controversy over the strength of the correlation between today’s P/E10 level and the going forward ten-year return. There’s just about universal acceptance that the correlation is statistically significant. But no one claims that it is anything close to perfect. The correlation is strong, far too strong to ignore in my assessment. But most investors are highly reluctant to use it to engage in long-term market timing, fearing that it could cause them to miss out on gains enjoyed by investors who stick with the same high stock allocation at all times.
Kitces pointed out that the correlation between valuations and safe withdrawal rates is a good bit stronger than the correlation between valuations and 10-year returns. He said that he has done research in the psychology field and that correlations of even 0.1 percent get people’s attention in that field while correlations of 0.2 percent are strong enough to support the publication of research in respected journals. The correlation between P/E10 levels and 10-year returns for stocks is about .55 and the correlation between P/E10 levels and 18-year returns is about .68. The correlation between P/E10 levels and safe withdrawal rates is .79.
Why so strong a correlation?
I believe that the biggest reason why the correlation is stronger is that safe withdrawal rates is a concept that focuses on long-term results. The correlation for returns generally gets stronger as the time horizon lengthens. But there is a complication. At some point the time horizon lengthens enough for the bear market predicted by the high valuation level to end and for a bull market to replace it. Thus, at some point the correlation for returns actually gets weaker rather than stronger. The year in which this happens differs in each historical return sequence examined. So there is no one year in which we can say that the correlation will always be stronger. In the data available to us today, the correlation is strongest at 18 years. But it could be that in a future time that will change to 16 years or 20 years.
The particular year in which the change from a secular bear market to a secular bull market takes place does not matter as much when it is the safe withdrawal rates that are being examined. When one is calculating safe withdrawal rates , one is not seeking to identify the return in any one particular year but how loss years and gain years interact to determine whether a given portfolio amount will produce enough income to survive 30 years of a specified withdrawal amount or not.
The correlations for returns are weakened by the fact that it is always one year for which the correlation is being tested. In some historical return patterns, that will be a year in which the effect of the starting-point valuation level was strong and, in other historical return patterns, that will be a year in which the effect of the starting-point valuation level was weak (depending on whether the secular bear market had been transformed into a secular bull market yet or not). So the overall historical record shows a mixed result -- a correlation that is very strong in some cases but not quite as strong in others (even though the correlation was equally strong in those cases, only for different years).
However, every year of a 30-year time-period influences the safe withdrawal rate for that time-period. Weak correlations for short time-periods don’t affect the overall correlation as much. So we see better numbers when we generate correlation figures for safe withdrawal rates.
I believe that the correlation between valuations and long-term returns is stronger than most of us realize because the statistical tools that we use to measure the strength of correlations are not designed to filter out the noise that is generated when a bear market is transformed into a bull market . Kitces’ observation that the correlation between valuations and safe withdrawal rates is amazingly strong supports this possibility.
Rob’s bio is here.