The Alpha Architect mission is to empower investors through education. 1
We can’t accomplish our mission without help. Fortunately, “finance twitter” and an explosion of bloggers are helping us achieve our goal.
Of course, with so many new blogs hitting the scene, we now face an information overload problem: too many blogs and too many writers.
How do we identify who is a flash in the pan blogger versus the next Michael Kitces, Josh Brown, or Ben Carlson?
We’ve tried to do our part and help to promote and share research from up and coming “undiscovered” bloggers/writers out there. In our early days, we were helped by long-time bloggers such as Meb Faber and Tadas Viskanta, so we try and return the favor. Recent examples of up and coming guest writers we’ve highlighted include Dan Sotiroff (now heading to Morningstar!), Aaron Brask, Andrew Miller, Elisabetta Basilico, and Dan Grioli — all of whom have written interesting and insightful pieces!
One of our favorite research blogs to hit the scene is from Corey Hoffstein @ https://blog.thinknewfound.com/. Corey and his team aren’t exactly newbies in the investments business — they’ve been around since 2008 — however, Corey and his colleague Justin Sibears, have recently started pumping out some really cool content.
I reached out to Corey and Justin and asked them if we could pick their brain a bit on tactical asset allocation — one of our favorite topics on the blog.
Here is the interview. Enjoy!
Q) We recently highlighted research by Victor DeMiguel on tactical asset allocation. Prof. DeMiguel has a lot of incredibly technical papers focused on complex asset allocation schema, but perhaps his most interesting piece of research is related to his “1/N” paper. What are your thoughts on the findings?
Corey: The research echoes our own broader findings: simple tends to trump complicated.
In The Dog and the Frisbee, Andrew Haldane runs a test where he assumes that daily market return volatility follows a true GARCH(3,3) model. To this data, he fits GARCH models of increasing complexity, from GARCH(1,1) through GARCH(5,5) model.
Despite the true model being GARCH(3,3), the GARCH(1,1) model has a lower mean squared prediction error than the fitted GARCH(3,3) until 100,000 samples are used.
100,000 daily samples, by the way, is the equivalent of 400 years of data, which is realistically intractable.
Using a simpler, albeit wrong, model proved to be more robust than the correct, but more complicated one.
Justin: That’s exactly right. We find that this is a reoccurring theme for models that operate in environments characterized by a high degree of uncertainty, like financial markets. Simplicity often allows us to be vaguely right in an uncertain future instead of precisely wrong.
Simple by design is an important tenet here for us at Newfound.
Q) We’ve done our own simplistic version of DeMiguel’s paper on Meb’s “Ivy 5” asset classes. Trend-following seems to be a dominate technique. Do you find similar things?
Corey: We agree that trend-following appears to be a dominant alternative allocation strategy, particularly during periods containing significant and prolonged drawdowns.
That said, all too often we see investors drawn to the allure of trend-following fail to adhere to what we believe are best practices that apply to any tactical approach.
First, we believe that trend-following should be performed systematically. Evidence suggests that when applied consistently over time, trend-following exploits the behavioral biases of other investors. Too often we see investors override signals with discretionary views, re-introducing the very biases they seek to exploit.
Second, we believe trend-following should be applied selectively. While research suggests the long-term efficacy of the approach, it can introduce significant tracking error in the short-term that can lead investors to abandon the approach. Therefore, only using trend-following for part of an asset allocation strategy can lead to a more sustainable investment approach.
Finally, trend-following should be implemented symmetrically. By this we mean in a fashion whereby the portfolio can benefit from both the strategy’s in and out calls.
For example, if a long/flat equity trend-following strategy replaces an existing equity position, the strategy can only add value when equities crash and at best keep up the remainder of the time. This has the added effect of requiring the trend-following model to be more accurate in its out calls.
On the other hand, if the trend-following strategy were to replace a 50% equity and 50% bond position, then the strategy serves as a dial within the strategic portfolio to increase or decrease equity exposure, benefiting from both in and out calls.
Failure to adhere to these best practices often comes from trend-following’s long-term historical dominance blinding investors to its short-term risks.
Justin: One behavioral theme we run across consistently is that investors are often unaware of the substantial role timing luck can play.
For example, it is very common for simple trend-following approaches to rebalance on a monthly basis. However, applied over several decades, the investor who rebalances at the end of the month will likely have a meaningfully different experience than the investor that rebalances on the 15th of the month.
Our research does not suggest a systematic edge from investing at a given point within the month, but rather highlights the significant amount of luck that trend-following investors potentially subject themselves to with such an approach.
Using a common approach, we found that applying the 200-day moving average model from 1950 to 2015, the choice of which day of the month you rebalanced on could create up to a 5800 percentage point difference in total return.
So while we believe that trend-following is an extremely robust and valuable asset allocation methodology, we also believe that significant care must be applied in its execution and implementation within a portfolio.
Q) The key challenges of tactical asset allocation are the inputs: What are the expected return assumptions? What are the volatility/correlation assumptions? What are the constraints? Let’s tackle each of these.
First, expected returns: How do you go about estimating these?
Corey: Adhering to our simple by design principle, first and foremost we try to avoid them! So, for example, we try to retain a binary approach in our trend-following models. Where we can use an equal-weight approach, we do.
In those places we do use explicit forecasts – like our more strategic allocation models – we prefer a simple building block approach based upon long-term guideposts (e.g. breaking equities into yield, dividend growth, valuation change, and currency change).
Next, the covariance matrix: Any pointers on taming this notoriously difficult-to-estimate aspect of asset allocation?
Justin: Where possible, we try to avoid covariance matrices and prefer simple risk management tools like position limits.
When a covariance matrix is unavoidable, we first see if we can get away with simply using asset variances – which tend to be fairly rank stable – and avoid explicit correlation estimates.
When that is not possible, we will take one of two approaches:
- The first is a simulation-based approach, where we will re-run our portfolio process several times, but each time apply a factor shock to the covariance matrix. We believe this process helps us account for parameter uncertainty in our variance and correlation estimates, as well as tail risks often not captured by a standard covariance matrix.
- In the second approach, we’ll re-run our process using multiple dimensions of risk (e.g. volatility, CVaR, fundamental, etc). We apply this approach when we do not believe factor shocks can capture latent risk factors that need to be explicitly accounted for.
While these approaches may seem complicated, they boil down to a philosophy of, “assume we’re wrong on average.”
Finally, the constraints: What sort of constraints do you put on your models and systems? Why have you gone this route?
Corey: We are big proponents of simple constraints from a risk management perspective. They are not only easy to implement, but they are easy to rationalize. Generally, the types of limits we impose are: position limits, portfolio turnover limits, and portfolio aggregate risk limits.
Q) I know your team focuses a lot on trend. Can you share a bit more detail on why you’ve taken this approach?
Corey: We tend to focus on trend-following because it meets our criteria of being a sustainable approach.
First, the approach exhibits incredible robustness throughout history, across geographies, and when applied to a variety of asset classes.
Second, it is simple.
These two facts together give us confidence that the success of trend-following is likely not a data-mined artifact.
Third, it can be implemented systematically. This is important for us because we believe that quantitative approaches help us avoid the behavioral biases that often lead to poor investment decisions.
Finally – and perhaps most importantly – there is a meaningful argument as to why the approach should continue to work in the future. Specifically, under- and over-reaction biases among investors cause the persistent emergence of trends.
We consider human nature to be a fairly sustainable edge.
Q) What about valuation-based asset allocation techniques? Asness is lukewarm on the topic, as are we. Have you guys figured out how to incorporate valuation in your asset allocation mix?
Justin: Our own research shows that valuation can be used in limited cases in tactical rotation, but the jury is still out on its application in market timing.
In other words, we’ve seen it be effective when applied in cases like sector rotation or country rotation, but deciding whether to be in our out of equities based on valuations has proven to be difficult.
So we believe that value can be successfully applied to help adjust long-term return expectations in equity markets in a fairly straightforward manner.
In a multi-asset portfolio, the application of value is not as simple as momentum. Defining the richness of a bond index, a commodity index, or a currency pair is not straight forward, much less figuring out how to compare valuation across them.
That said, evidence suggests that long-term mean-reversion and value are close cousins, and there is evidence that increasing exposure to assets that have been out of favor over the last 3-5 years and decreasing exposure to those that have been in favor can be a successful strategy.
Q) How do you think about tail-risk management or “crisis alpha?” The simple man’s version is treasury bonds and the complex solution is managed futures — you have any other unique ideas?
Justin: Close to a year ago we published a presentation called The State of Risk Management. It covered a number of approaches that investors might consider for risk management, including diversification with fixed income, managed futures, equity trend-following, managed volatility, and several options strategies.
While our research suggests that the optimal approach varies based on the investor’s objective, managed futures, equity trend-following, and managed volatility approaches offered some of the best cost-benefit trade off. Of course, diversifying your approach to risk management enhances the results even further.
When it comes to crisis alpha, an important question to ask is, “what type of crisis?”
For longer, more sustained drawdowns, managed futures and trend-following approaches tend to perform quite well.
For a sharp and sudden market decline, U.S. Treasury bonds are hard to beat.
Corey: In The Search for Crisis Alpha: Weathering the Storm Using Relative Momentum (which won 2nd place in NAAIM’s 2015 Wagner Award competition), our own Nathan Faber even suggests that relative momentum can be effectively used across the Treasury term structure to enhance crisis alpha opportunities.
Another unique idea we have employed – and outlined in past research pieces – is a dynamic VIX hedging strategy using the ETF’s VXX and XIV.
In comparing approaches, we think it is key to ask: “what are we trying to protect against?” and “how much am I willing to pay for that protection?”
Q) Finally, what role does experience and human decision making matter in asset allocation? Or does it?
Corey: Our preference is for fully systematic approaches. There is simply too much evidence, in our opinion, of negative drag caused by behavioral biases.
That said, it is impossible to completely remove the human element. We do, after all, design and build the models. Which is another reason we strive for simplicity by design: to avoid embedding too many of our own biases.
We tend to think of ourselves as stewards to the model. In this capacity, experience is incredibly valuable in helping us understand if portfolio behavior is normal or abnormal for a given market environment. No model is perfect, so having reasoned expectations for what sort of market it should struggle in is important.
Furthermore, experience can help guide prudent risk limits in portfolio design, help a portfolio manager reason about the impact of potential model changes, and help guide how a strategy should be used within a portfolio context.
Q) Love your blog and your research. Do you and the team have a particular focus or interesting questions you’ll be tackling this year?
Corey: Thanks – the feeling is mutual!
Historically, we have focused on the topics of asset allocation, risk management, factor investing and behavioral finance.
This year we are trying to balance the ideas presented in our evidence-based research with the behavior-based reality of trying to implement them in a portfolio. All too often we see ideas presented that may have long-term merit, but introduce substantial short-term anxiety that few investors have the fortitude to stomach.
Our goal is not only to explore and introduce new and unique concepts, but discuss how they can be implemented in a tractable manner.
Also, if you know of any great researchers out there doing great work, but still flying under the radar — please let us know!