We’ll be talking about Paul Novell’s flagship SPY-COMP strategy on the blog tomorrow. The strategy uses monthly economic data, like the kind available from the FRED database.
We’ve covered a handful of strategies like this in the past (think Philosophical Economics’ Growth Trend-Timing). Whenever we do, we invariably get a ton of questions, because it can be confusing to determine how to properly lag economic data to prevent lookahead bias. We wanted to take a moment to write this short post on how we backtest these types of strategies. We’ll refer back here in the future whenever this subject comes up. Fair warning: non-geeks should look away now.
* * *
Let’s say the strategy executes trades at the close on the last trading day of March. For most strategies, which are based on asset prices alone, we can use the data that is available up to that moment in time.
For monthly economic data however, the data available at that moment in time isn’t March data, because March data hasn’t been released yet; it’s February data. February data doesn’t get released until some point intramonth in March. The exact date depends on the data series.
Hence, all of our backtests based on monthly economic data include a 1-month lag. In our example, the trade at end-of-month March would use February’s data point.
FRED’s funny date convention:
What trips a lot of folks up is the funny date convention FRED uses for monthly data series. FRED’s March data point would read 03/01/2020, rather than 03/31/2020 or 03/2020 or something more intuitive.
Users often misinterpret that as the data being released on 03/01/2020. That’s not the case. FRED doesn’t maintain a day and simply uses a 1 in its place. The 03/01/2020 data represents March, but is released at some point in April.
Finessing by guessing:
In many cases you could try to estimate what that March data point would look like on 03/31 based on consensus estimates (meaning you wouldn’t need a 1-month lag). Depending on the data series, you might even be able to do it well. But that’s beyond the scope of our tests. We don’t account for a user finessing the strategy by guessing at future data points.
The good news is that, for all of these types of strategies that we’ve tested so far, the 1-month lag hasn’t had a significant impact on long-term performance. By their nature, these strategies tend to be slower moving, and include some other price-based input for confirmation.
How we handle alternate trading days:
These types of strategies tend to trade once per month at month-end, but like all monthly strategies that we track, we also show the results of trading on other days of the month as well. How do we handle the 1-month lag on these alternate trading days? With a butcher’s knife, not a scalpel.
We assume that all alt trading days prior to month-end are lagged by more than 1-month. So, in our previous example, trading on 03/31 would rely on February data, but trading on 03/30 would rely on January data.
Why? Some of this monthly data isn’t released until very deep into the following month. Trying to determine the exact date it was released decades ago, and then continue to do so on an ongoing basis, would be a nightmare. So we make a simplifying assumption that data isn’t available until the following month-end.
Lagging by more than 1-month:
Some data series are released even further out than the following month. We haven’t encountered one yet with any of the strategies we’ve added, but if/when we do, we’ll add the necessary additional lag to account for it.
Point-in-time versus revised economic data:
Economic data is often released at some initial value and then later revised, often multiple times. All of the historical results shown on this site are based on the historical economic data as it looks today. That raises the possibility that the historical signal shown today doesn’t match what would have been signaled in real-time. That’s a risk inherent in trading based on economic data.
Philosophical Economics did an excellent analysis of this issue for his Growth-Trend Timing strategies. In short, while it could lead to a different signal in any given month (sometimes for the worse, sometimes for the better), those differences tend to wash over time and have had little impact on long-term performance. Our previous analyses of some of the relevant strategies that we track concurs. All of these strategies include some element of price trend-following (which is unaffected by this data revision issue) that likely serves to keep signals somewhat in-line through any possible data revisions.
A list of strategies that use economic data:
Members can find a list of strategies that trade based on economic data by visiting the All Strategies list, clicking “Filter by Category” and then “Economic Data”.
New here?
We invite you to become a member for about a $1 a day, or take our platform for a test drive with a free limited membership. Put the industry’s best tactical asset allocation strategies to the test, combine them into your own custom portfolio, and follow them in near real-time. Not a DIY investor? There’s also a managed solution. Learn more about what we do.
