Do You QoQ?

"Prediction is very difficult, especially if it's about the future."

-Niels Bohr

Most (if not all) of you are at least somewhat familiar with Nobel Laureate Niels Bohr and his contributions to the field of physics – namely developing a working model of the atom and laying the groundwork for modern-day quantum theory.

The study of various quantitative disciplines was never really my strong suit as an undergraduate, but I did manage to sneak in a physics course during my time at Yale. Thankfully as a result, I am able to at least hazily recall learning about Bohr’s Principle of Complementarity, which states that the more accurately one property is measured, the less accurately the complimentary property is measured. Some eight or nine years later, I now wish I took better notes during that lecture(s)…

Back to the Global Macro Grind

One debate Keith and I often find ourselves engaged in with clients is the merit of reacting to headline (i.e. QoQ SAAR) GDP prints versus the merit of focusing on the underlying growth rate of the economy (i.e. YoY % change). For the purposes of retroactively explaining financial market returns and, ultimately, factor exposure selection, both growth rates are important to contextualize.

Specifically, when you backtest our GIP Model quadrants using historical return data across key asset classes and factor exposures, the key takeaways are overwhelmingly similar regardless if the second derivative of real GDP (i.e. the rate of change of the growth rate in this instance) is a function of the aforementioned tangent or secant.

For better or for worse, however, we are firmly entrenched in our preference for the latter (i.e. YoY % change) and have built a proprietary asset allocation process that responds appropriately to meaningful deviations in this key economic variable, among others. Like most frameworks, our asset allocation process remains ever-expanding alongside the cumulative intelligence, experience and bandwidth of our now six-person macro team, but one thing is for sure according to Bohr’s Principle of Complementarity: the more you are able to learn about an object’s momentum, the less you are able to discern about its position – and vice versa.

In the context of modeling the economy, the more we learn about sequential momentum, the less we are able to know about the underlying growth rate of the economy. Recall that headline GDP growth accelerated +660bps to +7.8% in the 2nd quarter of 2000 and that it accelerated +470bps to +2% in the 2nd quarter of 2008. If you were prescient in forecasting these second-derivative deltas, you could’ve bought all the stocks you wanted en route to peak-to-trough declines on the order of -49.1% and -56.8%, respectively (S&P 500).

Oh, and by the way for all the QE4 bulls out there: the Fed cut rates by -525bps during the former downturn and by -500bps (in addition to introducing QE1) during the latter downturn. If our #LateCycle Slowdown view proves prescient, investors would do well to keep that in mind as the Consensus Macro bull case for U.S. stocks shifts from “we really like them despite $100-120 crude oil” to “we love them because of falling gas prices” to “growth missed our expectations, but that’s OK because the Fed is likely to ease monetary policy again”…

Going back to the aforementioned head-fakes, it’s clear that those read-throughs on sequential momentum failed to signal pending material changes to the underlying growth rate of the economy. To cite a lesson from macroeconomics 101: the annual growth rate of real GDP is calculated by averaging the YoY growth rate recorded in each quarter. In light of this, what does 2Q15’s revised growth rate of +3.7% QoQ SAAR signal to you about the forward outlook for the underlying growth rate of the U.S. economy?

Another reason we like to focus on the secant rather than the tangent is because the former is simply easier to predict on an out-quarter basis. Intra-quarter forecasting is fairly straightforward if, like us and the Atlanta Fed’s GDPNow Tracker, you apply a predictive tracking algorithm to record and coagulate trends across key high-frequency economic data. Out-quarter forecasting is a far more difficult task.

But don’t take my whining for it; just look at our competitors’ track records:

  • Over the past five years, Bloomberg consensus forecasts for headline GDP just one quarter out have demonstrated a quarterly average tracking error of 145bps. This means that at some point within 3-6 months of any given quarter-end, Wall St. economists’ estimates for QoQ SAAR real GDP growth were off by an average of 145bps. That’s flat-out terrible in the context of actual reported QoQ SAAR growth rates averaging just 2.1% over this period.
  • Over the past five years, the FOMC’s intra-year U.S. GDP forecasts have demonstrated an annual average tracking error of 100bps. Worse, the maximum deviation of their intra-year forecasts from the actual reported annual real GDP growth rate was an upside deviation in every single year, meaning that the Fed’s growth forecasts are consistently far too optimistic.

Moving along, as the guy on our team responsible for generating our GDP estimates, how am I going have a reasonable basis for predicting the sequential growth rate in 4Q15 if I do not yet know what GDP is in 3Q15? What if my 3Q15 estimate is wrong? In the context of tens of basis points making all the difference between noteworthy accelerations or decelerations, multiplying a mistake by four for the purposes of annualizing the growth rate can lead to costly errors.

At least in attempting to calculate out-quarter YoY estimates, we are equipped with a base rate that is far more useful than the prior QoQ SAAR reading and a Bayes factor that is substantially more robust:

  • Base rate (i.e. the prior reported growth rate): Over the trailing 10Y, the standard deviation of the YoY growth rate of real GDP is 30% less than that of the headline growth rate (186bps vs. 267bps). Translation: YoY readings are considerably less volatile, which implies the most recently reported growth rate is a far better starting point for the purposes of forecasting YoY % changes than it is for forecasting QoQ SAAR % changes.
  • Bayes factor (i.e. the base effect): Roughly two-thirds of the time, the second derivative of GDP in the forecast period carries the opposite sign of the second derivative of GDP in the comparative base period. Moreover, as the Chart of the Day below shows, there is exists a considerable degree of negative cointegration between the comparative base effect and the subsequent YoY growth rate. Translation: we have a reasonable basis for knowing which direction (up or down) to adjust the base rate.

Going back to Bohr’s Principle of Complementarity one last time, what buy-side analyst in their right mind would analyze a cyclical on a sequential growth rate basis? If that made even a lick of sense, we’d all be buying retailer stocks hand-over-fist into every fourth quarter of every year, in which this thing called “the holiday season” occurs. Sure, you could probably apply a seasonal adjustment overlay to smooth obvious calendar-related deviations, but who’s to know what duration is appropriate for the historical observation period? Do you minimize it to keep current with changing dynamics within the industry or do you elongate it to account for key mean reversion thresholds across cycles? I, for one, have no idea what the right answer is and this dilemma is at least part of the reason why the BEA – i.e. the government organization responsible for reporting GDP – struggles mightily with its own seasonal adjustment process (CLICK HERE to learn more).

Do You QoQ? - Residual Seasonality

Again, I don’t purport to know a ton about analyzing individual companies – certainly not on a relative basis to anyone reading this note – and I’m sure there are industries where the aforementioned practices might make sense, but, on the whole, analyzing corporate operating metrics on a QoQ SAAR basis doesn’t really tell you a whole heck of a lot about the underlying health of the business on either a retroactive or prospective basis. But don’t take my word for it; try it at home for yourself.

So why the long rant about modeling principles? Because the U.S. economy is one giant cyclical – especially its most meaningful component, personal consumption (~70% of GDP), which we highlight in the chart below. As said chart emphasizes, when cyclicals bump up against a series of outright tough or incrementally tougher compares, recorded growth rates tend to slow on a trending basis; the opposite is true of recorded growth rates when encountering outright easy or incrementally easier compares.

Given that the strength of the U.S. consumer is one of the remaining catalysts (if not the only remaining catalyst) underpinning largely sanguine expectations for domestic economic growth among sell-side and Federal Reserve economists, market participants must hope that the underlying growth rate of the U.S. economy is not poised to inflect and decelerate on a trending basis as our forecasts imply it likely to do for the foreseeable future. Hope, however, is not an investment process.

But then again, the science of prediction is very difficult – especially when said predictions are about the future…

Our immediate-term Global Macro Risk Ranges are now:

UST 10yr Yield 1.99-2.21% (bearish)

SPX 1 (bearish)
VIX 21.69-42.27 (bullish)
EUR/USD 1.09-1.15 (neutral)
YEN 117.71-124.89 (neutral)
Oil (WTI) 35.84-48.42 (bearish)

Gold 1115-1165 (bullish)

Keep your head on a swivel,


Darius Dale


Do You QoQ? - Chart of the Day