What Is Factor-Based Investing?

Factor-based investing is one attempt to answer that question. By focusing on the underlying factors that define risk, return, and correlation this approach seeks to explain why some asset classes move together and to offer more efficient portfolio construction. Asset managers are starting to incorporate the idea into their portfolios, and a number of firms are offering “factor-based” mutual funds and ETFs.

In an interview with Yale Insights, Podkaminer lays out the theoretical groundwork for factor-based investing. He explains how constructing a portfolio around risk factors could lead to lower risk at equivalent rates of return. But he also explains why it’s been impossible thus far to bring theory fully into reality.

Q: What is factor-based investing?

The way that we have traditionally invested has focused on asset classes. We invest in equities or fixed income or real estate or commodities or private equity. But just under the surface each asset class is made up of multiple factors driving risk and return. A metaphor I use to explain factors is, if asset classes are like complex molecules, factors are the atoms that serve as building blocks.

During a crisis, molecules that were supposed to have different characteristics turn out to be composed of atoms that move together; in other words, asset classes that you typically wouldn’t think of as being correlated actually are.

From 2008 onwards, looking at the cross-sectional data, you definitely see correlation has increased significantly and noticeably. Even for people who don’t know which Greek letter corresponds to correlation, they felt it in their 401Ks and college savings plans.

It’s important to keep in mind as we go forward and build correlation assumptions that there is more there than meets the eye.

If we combine risk factors together into a portfolio instead of asset classes, potentially, we can get to a more efficient portfolio—returns that are comparable to the traditional model with a much better risk-adjusted tradeoff.

Q: Was the increased correlation across asset classes a temporary shock caused by the financial crisis?

Have we gone through an inflection point where disequilibrium jarred correlations? I think the answer to that is yes. If you look back through history, there have been many violent periods where correlations did change because what was happening from a geopolitical perspective, from an inflationary perspective, from a commodities-influenced perspective, resulted in seemingly unrelated asset classes moving together.

That said, there’s the concept of DINO—diversification in name only. Over time, we tend to put more and more “asset classes” into our models. But they’re not truly distinct asset classes; they’re just ever-finer slices of existing asset classes. Why separate, say, large cap and small cap? Domestic versus non-domestic equity? It doesn’t matter whether the domestic reference point is the U.S. or Belgium or South Korea. If you’re a South Korean investor, why make a big distinction between South Korean equities and the rest of the world?

The resulting pretty colored pie slices make an investment committee member feel good about the diversified portfolio. But if you just look at the names, without doing any sort of empirical analysis, intuitively we know that a lot of them are highly correlated.

Q: Where should correlation fit in portfolio design?

Correlation is the unloved redheaded stepchild of mean variance optimization. Traditional models for large, sophisticated, institutional asset owners have evolved to focus on expected return, expected volatility, and correlation.

What do we think the equity markets will return over the next ten year period? Or the bond markets? Or inflation? Or cash? Everybody has an opinion on expected return. Some people will have an opinion on expected risk as well. I find that folks typically don’t have as tight of a grasp on correlation. But correlation is a key determinant of not just the portfolios that you structure but the way that you define the assumptions in your model and the internal consistency therein.

Q: How do things look different using risk factors?

At least in my lexicon, to call something a risk factor, it shouldn’t be divisible into any smaller part. For example, inflation is very hard to decompose further, versus a bond, which is sensitive to numerous risk and return factors that are macroeconomic in nature—what happens with GDP, real interest rates, and inflation along with asset class-specific things like duration, convexity, and spread.

There’s also an important terminology issue that I want to bring up, which is risk factor versus risk premia. A risk factor doesn’t necessarily need to be compensated by the market, where risk premia do. Investors looking to maximize wealth are interested in holding compensated premia. That’s one way to distill the universe down instead of looking at 10,000 factors. It needs to make sense from a first principles economic standpoint. Is this a compensated premium?

For a moment, let’s ignore messy reality and talk about theoretical finance. We can group factors into different buckets. The macroeconomic bucket has things like exposure to GDP growth and productivity, real rates, inflation, and volatility. A regional bucket includes things like currency, emerging or developed market, and sovereign exposure. The equity-specific bucket includes size, value, and momentum. For fixed income there is risk of default, where this bond is in the capital structure of an organization, and, as I mentioned, duration, convexity, and credit spreads. Other factors don’t neatly fit into any of those buckets.

In the traditional model, when I hold a U.S. bond and a U.S. stock, they’re viewed as separate. But there’s very important overlap and crosstalk. In a risk factor context, I can put factors together in such a way that I explicitly capture that crosstalk and understand where the overlaps are and where the gaps are.

In theory, I could do this at a very granular level with every possible risk factor. In practice, it’s a little bit harder, because ultimately most factors aren’t naturally investable. I can’t go out there and buy GDP growth or size exposure or momentum or credit spread in a single, easily traded asset that’s quoted in the back of the Wall Street Journal. So I have to create factor-mimicking portfolios typically using long-short spread exposures.

If I want to replicate something like size, I would be long a global small-cap index and short a global large-cap index to get exposure to that particular factor premium.

Q: Your paper for the CFA Institute went into detail on all these issues. Could you explain here how factor-based investing moves from theory to practice?

The first real step is to define the parsimonious set of factors. How many do you need to cover the investable universe? Do you have any gaps? Did you have any overlaps? Do you have things that are truly uncorrelated? Once we decide which factors or premia to look at, the next challenge is trying to come up with ex-ante predictions for risk, return, and correlation for each of those factors.

That’s complicated because instead of looking at a handful of asset classes, all of a sudden, I could be looking at something like 10 to 20 factors where I need to be making an ex-ante prediction on what that factor’s characteristics are going to be, going forward. On the one hand, that’s daunting, because there’s more of them. On the other hand, it’s a much more focused look. For instance, what do I think is going to happen with real interest rates? That may take some of the measurement error out of the estimation. But it’s still a challenge.

Q: These ex-ante assumptions aren’t unique to factors, right?

Any portfolio, whether it’s constructed with asset classes or risk factors or some other system, needs to be based on ex-ante, forward-looking assumptions. Period. Go back to Markowitz, Sharpe, or any of the pioneers in this space, and it’s clear that when putting together portfolios, you really should be looking through the windshield versus the rearview mirror. So in a mean variance optimization model, I need to make a judgment on what the expected returns, and the risk, and the correlations of the different pieces of my portfolio are going to be before I put them together.

I think that there is a power in focusing on the smaller, granular units used for risk factors. Now, we also have less experience doing that. We’ve gotten comfortable with a variety of different models for looking at asset class returns. It’ll take some stretching to get a little bit more comfortable with factors as well.

Q: How often do you expect to be questioning ex-ante assumptions?

This is a great question, because it also applies not just to factors but to asset classes as well. If I’m a strategic, long-term investor, with hopefully a near-perpetual horizon, I want to be looking at equilibrium relationships amongst asset classes or amongst factors. So I’m not necessarily concerned with performance for next year or three years or even five years out. I’m thinking of a 10- or 15-year horizon. So for something to change, it needs to be not just a marginal change. It needs to be something real like a change in market structure or a long-term secular trend that is going to impact a particular factor or asset class.

I don’t have to be reactive to every tick of the market in changing these assumptions. They ought to be durable and robust. I think that the right frequency to revisit them is probably annually, because we’re long-term investors. Even then, they may not change much, at least that’s been my experience with asset class returns.

Q: What does a factor-based portfolio look like?

The factor-replicating portfolio is not buy and hold. That long-short component is going to have to be rebalanced continuously. It is a very intensive management process in terms of making sure that you’re still getting the factor exposure given that your longs and your shorts are moving around quite a bit on a daily basis. If we’re talking about a portfolio made up of 10 factors, you may have 20 positions with the long and a short for each of them. That’s a technical challenge.

There are other issues, especially around plan governance. If I’m a chief investment officer, I need to define my portfolio through factors or premia. That means that I really can’t use any peer groups for comparisons. There’s a whole host of softer issues, including headline risk, that would be material for organizations. Andrew Ang, a professor at Columbia, has written extensively on governance issues with risk factor portfolios.

Q: Given the practical challenges, how are risk factors being used?

Factors have been used for a long time in implementing a policy portfolio. When I put together portfolios of asset managers or asset management products, I look at volatility, liquidity, size, and style, among others. In some sense, we are just letting these factors and premia bubble up from the implementation arena to the principles that guide the portfolios of institutional investors.

Beyond that, there’s a middle ground between where we are today and really building portfolios of risk factors, which is where many practitioners would eventually like to be. The middle ground is to take a deeper and more inquisitive look at the asset class building blocks of your current portfolio, and to explicitly make those connections between asset classes which may not come through in the standard risk, return, and correlation modeling.

If I have 10 asset classes, each of which has equity somewhere in the name, whether it’s private equity or U.S. equity or high-yield, which has a huge equity component to it, I should be looking at those together. They’re going to move similarly. Let me not fool myself into thinking I’m diversified, because I have 10 different kinds of equity in my portfolio. That’s really not going to help when equity, as a whole, does poorly. And it’s not going to help me attribute things when it does really well.

Understanding the basic risk-return drivers of the asset classes and running a fuzzy optimizer in your head can go quite a ways.

Extending that, there’s a really simple matrix framework that we can use, and it’s just the two-by-two matrix of looking to see what happens in rising and falling inflation and economic growth environments. It’s no secret that many portfolios, because they’re so equity heavy, are really counting on low, stable inflation and relatively robust GDP growth. That’s the sweet spot for those portfolios. But there are three other quadrants and different asset classes perform best in those different quadrants because of the underlying risk factors.

Even if you don’t go all the way to risk factor portfolios, understanding from a regime perspective which asset classes do well in different scenarios can avoid the Goldilocks portfolio that performs remarkably well in one type of economic scenario, in favor of something that performs reasonably well in multiple scenarios.

Q: The risk factor portfolio you describe is difficult to implement, but there are white paper descriptions that make it sound easy to slot it into a portfolio.

There’s been a lot of marketing by asset managers about risk factor-type strategies which don’t meet some of the criteria that I’ve set out, which is a multi-factor long-short approach that’s trying to capture premia, or spreads. Many of the existing offerings are single asset class, long-only.

I’d describe those as falling into the alternative indexing or smart beta universe. There’s been a lot of academic literature showing that alternative indexes often have factor tilts, say, towards size, or value. But it’s accomplished without the same far-reaching framework I’ve described.

The tagline for alternative indexes is, potentially I can give you more return for less risk, or a better risk-return tradeoff, or more diversification with your portfolio. That is done by introducing tilts based on factors but without really talking about the tilts or factors directly.

Ultimately they’re just a slightly different type of return-stream to what we already have in a portfolio. While it’s an open question, I’d make the very strong case that these are active strategies.

From a governance perspective, given that I’ve defined them as active, they can be used to complement a large passive core. You could take part of your investment that is held purely passively and carve out a piece and say, I would like to put this in alternative index because I want to have an explicit factor tilt, and this is an efficient way to do that. I can also look at my stable of active managers and try to achieve some of their systematic exposure through alternative indexes with greater transparency and at lower cost. It becomes another way to incorporate factors without going all the way.

Interview conducted and edited by Ted O'Callahan.