Recently I’ve noticed a few posts here discussing various interesting beta-correcting techniques. Usually, the code provided is something you can tack onto the end of any existing strategy, and it works basically like this:
# If beta is positive, sell some SPY. # If beta is negative, buy some SPY.
Rinse and repeat.
First, I want to point out the merits of these strategies, because they do have many great uses. For personal trading algorithms where you are essentially your own fund, they are a great way for you to control your own market exposure! You can even adjust the target beta to anything that suits your needs. These can be a very powerful tool for anyone who has a great personal trading strategy, but wants to maintain a solid 50% or so correlation to the market as a sort of insurance policy against the unknown quirks of their code.
But, more often than not, this code is promoted as a way to boost yourself to the top of the contest leaderboards.
I see two primary issues with using the code for this purpose:
It actually works to defeat the purpose of Q seeking a beta-neutral strategy.
The reason Q and other funds like their underlying strategies to be “pure alpha” is so that they can control their own market exposure, as outlined above. Sometimes, this means targeting zero, but that’s not always the case. Justin Lent does a great job of explaining this here.
What they want is a beta-neutral strategy, not a beta-neutral algorithm. Why? Because if all they cared about was zero-beta, period, they could just correct every allocation algorithm with the above code snippets; just short SPY whenever things are looking a little too much like the market. The thing is, a simple pairs strategy (like ones Q itself has repeatedly touted) in theory has a zero beta and positive returns. But why bother with pairs if you can just buy the “good” half and short SPY instead? Boom, zero beta.
In the end, this isn’t pitching Q a strategy, its pitching Q a carefully beta-controlled fund. As a standalone, it’s fine, but this is unhelpful when taken in the context of dozens or hundreds of other algos in aggregate, as they plan to do.
These strategies are by definition reactive, not proactive.
You need to wait to observe the beta of your strategy to decide how much to correct for with SPY longs / shorts. This is manipulating the metrics after the fact at the expense of the purity and intent of your underlying strategy, plain and simple. If your true strategy is actually good, it won’t need a hedging component that is entirely SPY, VOO, or AAPL & GOOG. In my mind, this is no different than when contest winners were shorting one share of one stock to get the hedging badge.
Often, people realize after adding these “fixes” that the returns, Sharpe, and other metrics of their algorithm has been compromised. It’s easy to hope that fifteen lines of copied-and-pasted code will optimize one metric without affecting anything else, and sure, it can be fun to game the contest, but please, please, please don’t deceive yourself or others about what you have accomplished by doing so.
I'd love for Q and others to weigh in on what I've said here; I'm new to the field relative to many of you, and love being told that I'm wrong for the opportunity to learn about and discuss topics such as these.
Edited for readability and tone.