Risk drives everything about your investments, but all risk is not created equal. For most investors (and most of the financial industry) dealing with risk is a lot like making sausage. People want what it gets them, the investment returns, but they don’t want to deal with the process and the uncertainty that risk implies. They want to get their returns without paying for them. This is impossible.
Not only must you accept risk to get your returns, you also need to choose the right types of risk. We want to focus on building our portfolios around the risks that actually make sense. We want to find the risks that have positive expected returns. These risk factors are not necessarily things that everyone will need, or want, in their portfolio, but they are, by and large, the pieces on the board when you are constructing your portfolio1.
Now, we can identify all sorts of interesting relationships in the data if we want. There are massive numbers of really smart people who do nothing but look for new and interesting relationships in the data. It would be weird if they couldn’t find relationships in the data. There are relationships all around us, the question is if they are meaningful relationships, or if it’s just an artifact in the data.
As an example, let’s say that I grouped all of the stocks on the New York Stock Exchange by the first letter of their ticker symbol. If we look at the performance of each of these 26 portfolios, one of them will have the best returns, and one will have the worst returns. You might find a pattern in there somehow; maybe stocks with tickers that start with vowels have really great returns through time. Does this indicate a meaningful relationship? Do you think that the letter a company’s ticker starts with has any relationship to it’s stock performance? I think everyone’s going to realize that any relationship we find here is probably going to just be random noise.
There are all sorts of these things in real life. One of the most famous is butter production in Bangladesh predicting the returns of the S&P 500. My personal favorite is the Super Bowl indicator. This says that if a team from the NFC wins the Super Bowl, the markets will be up. If a team from the AFC wins the Super Bowl, the markets will be down. The crazy thing is, it’s been right about 80% of the time. Again, do you really think that which conference the team who wins the Super Bowl is from has any meaningful impact on the stock market? Patterns are everywhere. As a species, we are hard wired to find patterns. It’s a compulsion. Think about the constellations, or the pictures we see in clouds. Both the shape of clouds and where the stars are in the sky is essentially random, but we invariably find patterns and tell stories based on those patterns. It’s what we do.
As we review the patterns we find in the data, we know that some of them are meaningful. We know that some of them actually do describe real risk factors that you can use to better target your portfolio. The trick is being able to separate out the real stuff from the things that just appear in the data.
The best way that I know of is a 5 step test from Marlena Lee at Dimensional Fund Advisors. She says that for a risk factor to be real, and for you to be able to use it in your portfolio, it has to meet all five of these criteria:
- The relationship must be sensible.
- The relationship must be persistent across time periods.
- The relationship must be pervasive across markets.
- The relationship must be robust to alternative specifications.
- The relationship must be cost effective to capture in a well diversified portfolio.
So let’s walk through these quickly.
The first test is the one that stops most relationships in their tracks (there’s a reason she put it first). Before anything else, the relationship needs to be sensible. You need to be able to explain what’s going on in a reasonably simple manner, preferably without numbers, and you need to be able to do it with a straight face.
If the relationship passes the first test, then we can move on to the second. Here, we want to see that the relationship is persistent across time periods. This is basically a fancy way of asking if you see the relationship in different time periods. It doesn’t do you any good if you found a really powerful risk factor that had great returns in the 80s, but hasn’t really done anything since then. If something is a true risk relationship, then you would expect to see evidence of it over the long-term.
The third test is getting at the same idea as the previous test, but instead of testing the relationship temporally, it’s testing the relationship across different markets and countries. If you find a relationship in the US, does that same relationship show up in the EU, or Asian markets? If something works in one place, but not anywhere else, that’s a pretty big red flag.
The fourth test is probably the most technical test, but it’s incredibly important. Being “robust to alternative specifications” means that you can look at the relationship in a number of different ways, and use different metrics, and get basically the same numbers. So for instance, the value factor – this is normally described as the relationship between a stocks book (accounting) value, and what the market values it at. We could plug pretty much any fundamental value of the company in for it’s book value, and still come up with roughly the same results. Whether we use Net Sales, or Cost of Goods Sold, or almost anything else that you want that gives some sense of the economic activity of the business, we would be very likely to see the basic relationship pop out. Some metrics will be better than others, but we could tell that something is there.
If you need to look at a relationship in a very specific way, and squint at a very specific angle, then the relationship just isn’t all that robust. If there is something real there, then it will show up under multiple different metrics.
The fifth test pretty much comes down to whether or not you can actually do anything about the relationship. Can you capture the relationship, in a cost-effective way, in a diversified portfolio? Both parts of that question are important here. If you have to constantly turn the portfolio over to capture the risk factor, your transaction costs are going to eat into the premium you were expecting from the relationship. On the other hand, even if you don’t have to worry about the cost-effective part, if capturing the premium for the relationship requires you to build a portfolio that is not effectively diversified, then you’re digging yourself a pretty deep hole – especially during those periods when the premium doesn’t show up.
I generally regard this as the least significant of the tests. If a relationship fails one of the first four tests, then it’s pretty much over. If it fails this test, well, give it a couple of years. Just like there are a lot of really smart people finding all of the different relationships, there are a lot of really smart people figuring out how to actually trade these things effectively, though it’s a full time job. When small cap stock funds first started to come out, no one, including a lot of the academics working on the small cap premium, thought that it was possible to capture the small cap premium – it would get eaten up by transaction costs. That obviously turned out not to be the case. If a relationship gets through the other four tests, but fails this one, that just means that you need to circle back around in a few years to see if the situation has changed.
By using these five tests you can tease out whether a relationship that you’re seeing is actually meaningful or not. The vast majority of the relationships that people are talking about are nothing but noise, but occasionally there is something that’s worth a closer look.