Last time in part one, I discussed the theory of factor investing used by an increasingly large portion of the active management investment complex and the practical impediments to using it well.
Today, I will more specifically address the foundational flaws in the practice and how they can lead to substantial investment mistakes. Furthermore, the current market environment has heightened the risk of a mishap as the factors are more likely to be misused than normal.
Garbage In, Garbage Out
The core flaw in using factor models is…the inputs to the factor model. The models are built based on historic market behavior and typically emphasizes the last six months trading history – or less.
This means the weightings on the factor are constantly changing, often materially. For example, a stock’s beta might be 1.2 in the current model but six months from now it might be 0.9 or 1.5.
When the model weightings change, your attributed risk changes. That means the same portfolio can be considered well hedged today and poorly hedged a month from now just because a new month of data came into the model. This lack of stationarity makes using the model wisely terribly difficult.
Now, you might say, well things don’t really change that much that quickly, do they? Most of the time perhaps they don’t. However, most of the time managing your factor risk isn’t a significant driver of investment performance.
The times that excessive factor risk hurt (or inadvertently help) you is when markets get volatile. Guess what? That’s exactly when the weightings of the factors are the most likely to be out of date!
Help When You Don’t Need It
So just when you need the help, you are making decisions based on meaningless data. Worse, so is everyone else! Remember from last time the prevalence of factor models means investors tend to have common holdings.
Thus, everyone is going to sell out of the same stocks to try to manage their risk which means those stocks have more risk than the model says they did.
But that’s not all. Because those stocks go down, when the model next updates, it will say those tickers have more risk (volatility up, momentum down, beta up) which means even if you never owned them, you now want to short them to help you buy “safer” names.
But guess what happens when the selloff ends? The model says the stocks down the most are now super, duper risky! Don’t own them under any circumstances!
This is because factor models largely ignore valuation (“value” is a factor but not the change in valuation), so even though a stock may be down 50% and have lots of valuation support, the model says it’s beta and volatility have doubled and it is punitive to your overall risk to buy it at the lows. Thus, you miss the recovery rally.
To recap, the model pushes you to own the same stuff as everyone else, sell them at the same time everyone else wants to sell, and then makes it onerous to buy them back after they’ve been washed out. But at least you were true to your model!!!
What The Model Doesn’t Know Can Hurt You
In addition to using backward looking information, factor models treat all information that impacts stock prices, but isn’t a factor, as something an investor can control for and manage.
Remember, in theory, by controlling your factor risk, you are only betting on company fundamentals. But that’s theory. In practice, what happens is you are highly susceptible to macro risks that aren’t captured by factor models.
What often happens is hedging your factor risk leaves you exposed to big market trends like the Euro crisis, Brexit, or, now, Covid.
Because these big events change trading relationships, the model weightings end up lagging reality as I described earlier. Your risk model tells you that you’re hedged, but your performance swings violently.
You can try to adjust your portfolio to where you think the models are headed, but, if you do that, you are offsides on the current model, so the risk police give you a warning. You’re trapped! Waiting hopelessly for an out of date model to catch up to reality.
Cheating the Model
But there’s a bigger problem than that. There are the cheaters. These are investors who know that the model is backward looking and hunt for outdated factors that let them buy the stuff they want to – typically, stocks that are higher risk but performing well.
Anytime I wanted to I could have built a portfolio that screened as having low risk but that I knew probably had high risk and thus more upside potential.
I didn’t play that game because I was focused on my best estimate of the real risk and, if it didn’t screen well, I would try to make small adjustments to get inside the box without creating undue forward risk. (Over the course of my career, my worst performance stretches usually came when the models said my risk was lowest.)
There are a lot more cheaters than you might guess. Just like any constraint in any business, people find ways to manage to the metric rather than manage to the purpose of the metric. To be honest, few investors actually understand the models and thus look at it as something to game rather than as a tool to keep them out of trouble.
The Vaccine Factor…
This brings us to the long awaited payoff of this missive…the impact of vaccine news on factor models. As mentioned above, hedging your factor risk can expose you to large macro events. There is no bigger macro event than covid, specifically how long will it last and when can we return to “normal”.
The obvious problem is covid changes all trading relationships which makes the factor weightings useless. Additionally, it increases correlations. On days with positive vaccine trial news, most stocks go up and, often, in not so obvious relationships.
The increased correlation makes factor trading very difficult. If everything goes up together and to the same degree, you can’t make any profit pair trading regardless of your factor bets.
Worse, any bets you think you’re making are likely to actually be bets on covid. Let’s take technology. It has obviously been leading the market for some time here. However, on days where there is positive vaccine news, tech actually lags, if not outright declines.
Why? Because so much of the tech gains are tied to the “stay at home” trade, e.g. Amazon, Netflix, Zoom. So any bet on momentum, which is really a bet on continued tech leadership, is really a bet on there being no vaccine.
Conversely, we typically think of the performance of financials as being driven by the direction of interest rates. However, the direction of interest rates is now a function of vaccine progress.
How’s that? As long as there’s no vaccine, the Fed will remain in full accommodation mode and the government will continue to pursue more stimulus. These actions keep rates low and thus financial stocks underperforming. If we get a vaccine and the Fed starts to ease up off the gas, rates will drift higher and financials will do better.
Energy is in the same boat as financials. Energy prices will remain low as long as the economy is depressed. Most energy investors understand economic activity is a key factor in oil prices. However, the return of economic activity is a function of vaccine progress, so energy too is nothing more than a vaccine proxy.
…Will Wreak Havoc on Performance
So if you remember in part one where I discussed all the reasons tech has outperformed while energy and financials have underperformed (ESG inflows, increased retail trading, changes in index membership), the continuation of that trend is a function of one thing…when do we get a vaccine?
Thus, when we do get a vaccine, all the market dynamics will change. Some time ago, I predicted a vaccine would lead the market lower. I also suggested it would be accompanied by a rotation out of leaders into laggards.
The combination of these two means the factor weightings will become irrelevant. Betas will compress, momentum performance will flip, value won’t be correlated as much to interest rates, etc.
Those who lazily bet on continuation of the trends of the last few years will be lost. Those who diligently followed the model and remained hedged will realize they were hedging the past, not the future.
But those who will be in the worst way will be the ones who “cheated” and gamed the system. They will find that trying to game things pushes them to buy all the stocks they don’t want to own.
Without being able to cheat, their next best option is to take the model seriously and hedge more, but, as noted, they will be hedging the wrong thing because of the outdated weightings and find themselves losing money.
The Way Forward
So what happens next? Hopefully, a realization that factor models have outlived their usefulness. They have a place, but they have been overused and now abused. It would be more fruitful to identify environments where they work better (low correlation, low volatility) and those where they don’t.
In the environments where the models are more likely to be stable, they are a good risk control tool. However, there needs to be more effort to identify periods where the models are likely to fail and identify substitutes that work better in those situations.