Unintended Factor Bias
Why your factor beta comes with a lot of noise.
BY Scot Blythe | August 16, 2018
Not all beta is smart, nor are all smart-beta products. Some are very well constructed, says Bill DeRoche, chief investment officer at AGFiQ. Others, not so much. Nor does every investment theme offer a persistent return premia. Value, size, and momentum are factors that do.
Key is what the product is actually tapping. DeRoche uses the example of a portfolio of 200 momentum stocks sorted from a pool of 1,000. There’s a lot more than momentum driving performance of those stocks.
“With active risk, there are two sources,” he explains. “We want exposure to momentum, so we want to maximize the amount of risk, the amount of tracking error that’s coming from momentum. We’re going to try to minimize all the others. We can’t eliminate them entirely because there will always be some impurities. At the end of the day, we are going to have some unintended sources of risk.”
And it can be substantial, he points out. “I only have 20% of the risk coming from the momentum factor, so I have 80% of my risk coming from what I’ll call noise.”
FACTORS THAT ENDURE
One problem is factor decay: stocks that rank highly based on momentum characteristics can see their ranking change after a period — often less than six months. Value and small-cap characteristics tend to endure longer. But part of it has to do with the mechanics of tracking error.
“Typically, you think that, if the level of my tracking error goes up, my expected return should increase,” he explains. “We all know intuitively that they’re not related in a linear way. At some point, as I increase tracking error, I’m not going to be able to increase my expected return.”
A transfer co-efficient is at work, he says, which can hound conventional portfolios. “If I could follow my ranking signal perfectly, I would end up with a tracking error transfer co-efficient of one,” DeRoche adds. “Unfortunately, there are constraints when you invest, long-only being one of them. As a result, you’re not going to have a transfer co-efficient of one. It’s going to be lower.”
Interestingly, he finds that while the tracking error from a factor might be expected to be 2% to 4%, with some products it’s actually higher. “From our perspective, you’re getting a lot of noise; you’re not getting a lot of the exposure you were hoping to get with the particular factor.