abc132 wrote: ↑Wed Sep 21, 2022 12:26 pm
JackoC wrote: ↑Wed Sep 21, 2022 12:04 pm
abc132 wrote: ↑Wed Sep 21, 2022 11:34 am
JackoC wrote: ↑Wed Sep 21, 2022 10:48 am
.. when it comes to 'factor tilt' particularly it tends to rely more heavily on estimating the difference in expected return (for risk) between/among factors and that's harder than a general estimate of expected return for the market. Non-US IMO can easily be justified with a diversification argument. Some also believe the expected return of ex-US is higher but that belief isn't as key.
Looking at how the expected stock CAPE returns compared to actual returns, they have shown fairly poor future predictability. Taking even less certain estimations for factors, we should expect future predictability from the models to be less than what CAPE has done. I suppose we can always say that analyzing data after the fact is predictive (it's not), but that can't ever earn us any actual returns or premiums. Given the uncertainty in realized returns from market expectations, we should expect these factor models to be poor at predicting future performance and premiums if they have significantly less certainty than market/beta predictions.
The models rely on factors being independent sources of risk, which they are not. A large solar flare, a war, a pandemic - any black swan event can do completely opposite things to any given factor depending on the specifics of the black swan. If/when the market doesn't hold up in a manner big enough that we need diversification beyond fixed income, we can't say what factors will do.
This is basically going around in the usual circle on this question. We are trying to estimate the *expected return*, not predict the realized return. If anyone could come up with a measure which closely predicted the realized return, it would prove that the asset in question had no risk, which we know it does. So we can't ever expect an estimate of the expected return of a risk asset to correlate very highly with the subsequent realized return. The issue is whether the estimate has a fundamental basis and whether it can be shown biased high or low, or else it should be accepted. 'CAPE' is so confused now with regressions attempting to show that higher valuation now correlates with lower valuation later, different issue, it should probably be left out for clarity. Although the derivation which says E[r, real]=1/PE is fundamental, it's not an empirical result. But again let's put CAPE aside and just take another fundamental formula for stock expected return E[r]=div yield+real div growth. Realized return can be higher/lower than that estimate because either a) 'div yield' isn't what it seems* b) we misestimate *expected* real div growth (Ilmanen estimated it as the past 100 yr average of real EPS growth, 1.5%, what's your different assumption?) or c) there is a speculative return over the given period...which there always will be, positive or negative. But to invalidate that equation you can't just say 'there could be solar flares' which we all know. You have to show why it's a *biased estimate of the midpoint* and in which direction, by coming up with different estimates for a) and b) or arguing why c) the *expected* speculative return, has a non-zero value (back to CAPE regressions, some would argue the expected value of the speculative return now is negative because CAPE is so high, but that's not a necessary assumption, E[r]=div yield+real div growth with no 3rd term implicitly assumes expected speculative return=0).
This sort of analysis leads to a relatively contained range of plausible expected return estimates (with which 1/CAPE pretty closely agrees right now). It makes clear the assumption 'expected return is past long run avg return' is biased to the optimistic side with valuations as high as now. Again not because valuation has to go down, but by basic fundamentals or less earnings/div you get for each $ invested. However again, to make such estimates for factors and say firmly which factor has higher expected return than another is more difficult. Too difficult IMO to really believe in factor investing, you should really believe in it to do it (so you don't just quit when 'it doesn't work for you') and that's a big reason I don't.
*as in 'how about buybacks?' but see Ilmanen on this, little convincing evidence net anti-dilutive buybacks are significant, US companies buy back more stock now than they used to but mainly when things are going well, they dilute more all the time than they used to. Especially if we define the asset class as global stocks which we probably should on a theoretical basis, and if people want to tilt toward/away from US within that, fine.
I can't tell you in which direction factors are biased because they are not a fundamental property - that is any factor will have the opposite of it's expected direction/performance at some point in time. Nor do I need to do so to say the estimates being given are not very predictive or reliable. There is a phenomenon known as the butterfly effect that clearly demonstrates that some things about the future are not known. This is well-understood in the scientific community where we can back-test simulations to see what might have happened in the past, how galaxy's might form, etc, but this community recognizes the limitations of future predictions for complex systems. Things that are not currently even measurable are likely to change the end/future result.
In no way does anyone need to make more accurate future predictions to reject models that have been demonstrated to have low future predictability. You can't take "average annual stock performance" as the measure of regression for a system that compounds error over 20-50 years of investing. You are not even predicting a final result (portfolio size) within a factor of 2 over a 20 year period and are not looking at the actual error of the prediction.
I can say that human population is likely to increase and technology is likely to increase production per energy usage. All of these things should benefit the companies of the future and contribute to more earnings per dollar invested. That is the reason the market as a whole is very likely to hold up.
Your misconception is more basic than factors as in 'factor investing' so I'll reiterate based on the more basic idea of estimating expected return. The estimate of expected return cannot be validly discarded because the past estimate of ex ante expected return did not period after period fall right on top of the subsequent ex post realized return. To ask the estimate to do that, for a risk asset, must be to basically misunderstand what expected return means.
And in actual practice as was covered ad nauseum in the recent thread on Ilmanen's recent book, as well as various other threads, it's easy to see that in general the range of both realized returns and ex ante expected return of 60/40 by simple fundamental measures, div yield+avg past EPS growth and nominal bond yield minus survey based inflation expectation respectively, were in the same ballpark for nearly 1900-1990, then expected returns steadily dropped to levels never seen in the previous period but realized returns did not as valuations marched basically monotonically higher since around 1990. There is IOW no obvious evidence from the past that either of those fundamental measures, E[r, real, stock]=div yield+long term past avg real EPS growth or E[r, real, nominal bond]=yield-expected inflation is a biased estimate. And you do in fact have to show that to reject them. You can't just say 'I' don't know': you can't make any decisions about investing without any assumptions about future returns except in the most extreme boundary case. 'There will be innovation and the market will hold up' is some kind of assumption, and not at odds with E[r]=2.4% div+1.5% EPS growth, (VT's div yield, assume world growth trend around 3% real, US EPS growth in 20 th century was 1.5% less than GDP growth), as wishy washy qualitative statements are generally unlikely to be at odds with specific numerical estimates. But one must make some assumption to choose how much to invest, the exceptions to which would tend to just prove the rule. And 3.9% is very significantly lower than the assumption 'expected return=past avg return', so it's important.
OTOH this risk asset/factor has higher expected return than that one, more difficult, especially for things which give no direct read on expected return. The whole stock market does partly (in the div yield though requires us to assess if expected EPS growth is higher/lower than past long run, here I'm assuming same), bonds do pretty completely (no better estimate than the yield to the investing horizon), the 'SCV factor' doesn't at all, it's basically only derived from past return history. That's the difference I see, if ever we could get past the basic confusion about estimating expected return vs predicting realized return.