By Jack Forehand, CFA (@practicalquant)
Those of us who run quant portfolios sometimes like to prove how smart we are. We like to throw out advanced statistical concepts that show how well we can measure every conceivable attribute of our portfolios and their risk and return profiles. But one thing often gets lost in this contest to see who can do the most advanced calculations to highlight the robustness of their process – most of these concepts don’t mean very much to your average investor and don’t tie all that well into the way they actually invest.
When we started running factor models, I was as guilty of this as anyone. I wanted to show how much I knew so I would try to build the most complicated processes possible and would explain them in an equally complicated way. But over time I have realized that most of the time common sense and simple approaches trump more advanced and complex ones.
One of the best ways I see this evolution in myself is in my approach to blending individual factor models into a combined portfolio. The high-level principles of blending factors are pretty straight forward. For example, if value produces an excess return over time and momentum does too, but they produce those excess returns at different times, then it probably makes sense to have exposure to both of them. But despite that, many of us have a tendency to overcomplicate this process, when in reality some very simple rules, and some easy-to-understand metrics used to measure success, can often achieve better results than much more complex modeling.
Here are some basic principles that I think can achieve success when building multi-factor models.
[1] Recognize What is Not Possible
Typically, when an investor uses factors to build an active portfolio, one of their primary goals is to outperform the market. So a good first step in building multi-factor portfolios is understanding that falling short of that goal over shorter-term periods is going to be necessary in order to achieve it over the long-term. We all know that factors like value and momentum will go through periods where they don’t work. Combining factors can mitigate that problem, but it can’t completely fix it.
Our goal on Validea is to try to search through books and academic papers to find models with long-term track records of outperformance. We currently have 45 factor-based models we track in our system. They cover all the major investing factors, and their approaches range from deep value to focused momentum. But none of them beat the market every year, and neither do any combination of them.
To illustrate this, I did a simple experiment. Let’s say I want to create the model with the most consistent outperformance over the market possible using any three of these 45 models. So I will build a thirty stock model containing ten stocks from each of the three strategies I select. Let’s also say that I was blessed with the ability to see the future and was able to identify the best possible mix of those models in advance out of the 14,190 possible combinations. If I look at the performance of that perfect foresight portfolio in every rolling one-year period (253 or so per year) in the past decade, was it able to beat the market in every one of them? It was not. It outperformed 77% of the time.
So even with perfect foresight, 45 factor-based strategies, and over 14,000 combinations of them to choose from, there was still a 23% chance that if I looked at my portfolio on any given day in the past decade that my one-year return would be worse than the S&P 500. The lesson here is that periods of underperformance are an unavoidable price that must be paid to invest using factors. No matter which factors you use and how you combine them, you are going to trail the market at times. Accepting that fact in advance is a prerequisite for even attempting the process of using a multi-factor approach to build portfolios.
[2] Measure What Matters
I think a great way to look at combining factors is to start by thinking about what derails factor-based portfolios in the real world. We have accepted the fact that no matter how we combine factors together, we will have periods where we underperform. The next important step in the process is to ask ourselves what makes that underperformance matter. If we all were Warren Buffett and just stayed the course no matter how much we underperformed, then it wouldn’t matter at all. But in the real-world, investor behavior is a major consideration in building investment strategies.
I have found that optimizing portfolios to limit the things that lead investors to abandon strategies at the wrong time can lead to better outcomes. So instead of worrying about things like standard deviation, I instead like to focus on the percentage of time that any combination of portfolios has underperformed the market over 1-, 3-, and 5-year periods. I also like to look at the magnitude of that underperformance and see how often a combination of portfolios has underperformed by over 5% and over 10% over one-year periods. By focusing on the length and magnitude of underperforming periods, which in my experience are the major factors that lead an investor to abandon a strategy, these metrics allow us to blend them in a way that tries to limit the impact of behavior as much as possible.
This same process can apply to anyone who uses active strategies, regardless of whether they involve factors or not. For example, if I was looking to invest in a portfolio of discretionary active mutual funds, I could download their price history from a place like Yahoo Finance and look at how often each of them underperforms on their own and the magnitude of that underperformance. And then I could look at the same statistics with a combination of them. If those results improve, then I am likely improving my odds of success by combining them.
[3] Make Sure the End Result Makes Sense
One of the main criteria for a quality investing factor is that it is intuitive. Or in other words, the idea that the factor would explain stock returns needs to make sense.
I think the same is true for blends of strategies as well. It is important to take a look at the end result of any process that combines factors and make sure it has exposures to a diverse group of factors that have the potential to perform well in a diverse group of economic and market environments. For example, if I used my process outlined above to build a portfolio using only the past ten years of data, it would likely result in a portfolio that had nothing but exposure to growth and momentum (maybe with a little low volatility mixed in). Value and quality would likely not make the cut. But it seems likely that we will see an environment again where those factors will outperform, and a multi-factor system that seeks consistency over time should likely include them. This is one of the places where having a person behind a quant process can add a lot of value.
The Important Link Between Factors and Behavior
There are many ways that investors can build quant portfolios. I don’t pretend that the process I outlined in this article is the best option, or one that will work for everyone. I think the most important point here isn’t necessarily what the process is, but rather that any process that builds portfolios needs to consider how investors will actually use them. This requires that behavior play a major role at every step in the process. If you are an individual investor, this can mean taking an honest look at any strategy before you follow it, focusing on what it will look like when it goes against you, and making sure you can live with that. For someone who builds portfolios like me, it means that I can’t just create things in my quant lab without thinking deeply about how they will be used in the real world. But either way, the success of a portfolio is inextricably linked to the behavior of the person following it and it is essential to build portfolios with an understanding of that reality. Sometimes simple metrics can help with that more than complex ones.
