vineviz wrote: ↑Tue Mar 31, 2020 8:08 am
afan wrote: ↑Tue Mar 31, 2020 7:11 am
I tilt towards all factors, proven and unproven, known and unknown. New factors pop up all the time. Some that once appeared to be valid turned out to be proxies for something else. That new something else is probably a proxy for yet another unknown or unproven factor. One can constantly chase individual factors, revising the portfolio every time the supposedly optimal ones change. Or one can buy them all and consider the factor literature interesting without letting it affect one's asset allocation.

I find the cheapest way to do this is TSM.

I see this misunderstanding perpetuated nowhere but on this forum.

**A total stock market is a completely tilt-less investment.** TSM is a perfectly reasonable approach to investing, but it is most definitely NOT a "tilt towards all factors".

It has exposure only to one factor, namely market beta, and by definition eschews all other equity tilts completely.

Remember that factor analysis of stock returns is never final.

Factors are derived from the behavior of markets. All factors are potential correlates of market behavior. Some work better than others.

Some appear to work well until better analysis or better data show that their explanatory values are low. Some appear to work poorly until better analysis or better data show their explanatory value to be high. Some go between high and low power as studies and data evolve.

Some factors are found to be proxies of others. Or at least so it may appear at any point in time.

The number of candidate factors can be larger than the number of portfolios, which is larger than the number of stocks.

When one defines a factor, one has also defined the inverse as everything that is excluded.

"Small" is a factor, as is "not small". "Quality" is a factor as is "not quality".

The only bounds on the number of possible factors is the ability to define and measure them. With current computer technology, the number of candiate factors should be infinite. At any given time, few of these will appear to have explanatory value. The elements of the set of candiate factors that appear to have value will vary over time.

At any time, the number of factors that appear to explain returns will be relatively small. But that need not mean that any of them are correct, that they are not proxies for unknown factors, or that that are not artifacts of problems with data or analysis.

Human cognition and current methods for analyzing factor analysis favor small numbers of factors. A model that "explains" stock returns with 1,000 factors would be viewed with skepticism. One would intuit that nearly all of the factors had such minor roles as to be lost in the noise. The model itself would appear to be an error of data mining. No one knows how many "true" factors there should be. One would also conclude that thousand-factor models are useless, no matter how well one might fit the data used. If there were consistent, reproducible, robust thousand-factor models that consistently outperformed far smaller models, one would conclude that factor analysis itself was useless. It could be both useless and return valid models that are too large and unwieldy to be of any value.

Consequently, popular models, distinguished from the unknowable "correct" factors, must be parsimonious, relatively easy to define, easy to test and survive application to subsets of the data. Ideally, they would retain their explanatory value when applied to new data sets.

Unfortunately, the world is running out of new datasets and applying models prospectively means waiting decades for sufficient new data to accumulate.

Excluding some factors or investing in their inverses will leave the investor without exposure to those omitted. Depending on how the omitted factors perform, this may be desirable, or not.

To be exposed to "small" and "not small" factors, one must invest in both.

This results in having no net exposure to size.

The same applies to any other factor.

One can get exposure to a subset of factors, which is what most tilters seem to want, without going to TSM. That is possible exactly because one decides to exclude a universe of factors that are not in the chosen set.

The alternative is to have exposure to all candidate factors that can be accessed with stock investing, whether or not previously defined or analyzed.

Exposure to all possible stock factors=exposure to none, other than beta.

We don't know how to beat the market on a risk-adjusted basis, and we don't know anyone that does know either |
--Swedroe |
We assume that markets are efficient, that prices are right |
--Fama