I’m not a “stocks for the long run” guy.

I’m a “probably stocks for the long run, most of the time” guy.

See, I’m pretty confident that in order to get rich, you’ve got to own equities. You probably also have to own equities to stay rich (to support drawing cash from a portfolio while preserving purchasing power).

*BUT*

Usually when people say “stocks for the long run” what they really mean is “US stocks for the long run.” And usually what they’ve done to arrive at this conclusion is extrapolate past returns from the US stock market since about 1926 or so.

We like to pretend this is a disciplined asset allocation process when really it’s just a massive directional bet on the US equity market. A massive directional bet based on a relatively limited historical data sample. (btw , your “diversified” RIA and wirehouse models typically make this same bet but with a dash of Chili P for flavor)

When we do this with fund managers and stocks it’s performance chasing.

When we do it with asset classes and countries it’s asset allocation.

Classic.

Particularly since we *know* major economies and empires have *all* mean-reverted historically. (There are literally no exceptions I can think of)

Now, I’m certainly not going to argue a bet on US stocks is a *bad* bet over the next 20 to 30 years. Especially considering the alternatives. In the grand scheme of things, if you’re going to make a massive directional bet, this is probably one of the better ones you can make. But there sure are a lot of assumptions embedded in that kind of allocation.

The ur-assumption is, of course, that asset allocation is an exercise in decision making under risk, like placing bets in casino games where the odds and payoffs are both known and fixed.

It isn’t.

**Asset allocation is an exercise in decision making under uncertainty. **

A metaphor we often use to teach basic probability is that of picking colored balls from a bag. If you know there’s one red ball and nine green balls in the bag and the proportion remains static over time, you’ll always have a 10% chance of pulling a red ball.* This is the world as modeled by modern portfolio theory and mean-variance optimization.

Financial markets work more like this: every time you pull a ball from the bag, you have to turn your back, and the person holding the bag may or may not place another ball, either red OR green, into the bag. You can continue to assume a 10% chance of pulling a red ball, but the true distribution may turn out to be dramatically different over time.**

Most of what we think we know about asset allocation is a noble lie. We treat asset allocation as an exercise in decision making under risk because doing so makes it more amenable to neat and tidy mathematical models (not to mention neat and tidy sales pitches). In reality, we have no idea what the “true” distribution of returns looks like.

In fact, it’s extremely unlikely a “true” distribution of returns exists. Even if it did, it probably wouldn’t remain static. Why would it, given that we *know* economies and markets are complex, chaotic systems that are constantly changing? It should hardly come as a surprise that fancy statistical models based on decision making under risk repeatedly fail in the wild (see: Long-Term Capital Management; The Gaussian Copula)*. *

As I’ve grown increasingly fond of saying: there’s no *there* there.

The single biggest change in my personal investment philosophy over time has been shifting from a utility maximization mindset to a regret minimization mindset. To me there are two key components to regret minimization:

(1) Get balanced beta exposure cheaply and efficiently. A little leverage is okay to help balance it all out. Emphasize robustness over maximization.

(2) When you do take shots at alpha generation, make them count.

This is why over time I’ve become increasingly convinced strategies such as risk parity or leveraged permanent portfolio should be core building blocks for folks who want **truly** diversified portfolios. Grind out 5% real or so in the core. Make your high risk/high reward bets in a dedicated alpha sleeve.

However, I’d be remiss to conclude without noting that regret functions don’t generalize well. *Your* regret function is probably different from mine. In fact, it’s entirely possible *your* maximum regret is not maximizing utility (“leaving returns on the table”).

In that case, by all means, go ahead and maximize utility! But it’s still worthwhile to be explicit about the assumptions embedded in what you are doing.

* If we assign a value of 1 to “pick a red ball” and 0 to “pick a green ball” we can compute an “expected return” and standard deviation (“volatility”) for “pick a red ball.” Those values are 10% and 30%, respectively. Assuming T-bills yield 2%, “pick a red ball” has a Sharpe ratio of about .27. Somewhat amusingly, this is not too far off the long-run average Sharpe for the S&P 500.

** You should therefore be updating your views of the distribution over time. And it behooves you to assign low confidence levels to your views. A detailed examination of the math behind this is beyond the scope of this post but you can read an excellent discussion of the issue here.