Dr. Alan O'Sullivan

And that logic has empirically also tended to work.

But what we find is that in 2000, for example, those survey-based expectations were very bullish.

Investors, retail investors, also CFOs had high forecasts.

My favorite is to look at analysts' earnings growth forecasts, and they were very high at that time.

So this...

academic rational idea that we had low expected returns because there were good times and people accept a lower equity premium at that time didn't jibe with what we saw in surveys.

Surveys were rather telling that investors had very aggressive growth expectations and analysts had high growth expectations

And that was pushing valuations too high, and that was sort of causing the low prospective returns.

And indeed, so the decade of 2000s turned out to be a very bad decade for equities.

And so now I would say that we are in a sort of little sister situation of that.

I'm not sure that it could work out another way, but that classic case of very high valuations predicting low future returns is sort of a scary precedent for the current situation.

It is with the extra twist.

So again, conditional is very much this objective that you are trying to use something, some useful current information to predict future returns.

By the way, it's not easy and you are not going to be right like always, but you tend to be.

So the correlation between these predictors I use and next decade returns might be 0.5.

And so that leaves lots of scope for doing badly, but at least it's the right direction.

Conditional is just like you said, but with unconditional, so I think Cam Harby would say that past should be a very long-term past.

It should be last 50 years, 100, so really long-run data.

Whereas the rearview mirror that I worry about is typically like last five to 10 years history.

That's sort of the natural memory length that we tend to have.