Awkward: After Bashing Cold Weather Excuses, Bank Of America Jumps On The "2nd Seasonal Adjustment" Bandwagon

A little over one year ago, Europe did not like the fact that it was stuck in a perpetual recession so it did something about it: it arbitrarily raised the GDP number by hundreds of billions in estimated "growth" when it added the "contribution" from prostitutes and drug dealers, and hey presto: GDP jumped in every European country (alas, Greece has since descended once more into recession).

In the US, where the populist outcry to such an arbitrary "sinful" strategy to boost GDP would not work, especially not so recently after the US itself revised its own historical GDP higher by about $500 billion when it retroactively added the benefits of intangibles, trademarks, and changed the way pensions were capitalized, economists have been stumped how to rejigger numbers which refuse to comply with central-planning's "best "intentions of boosting not only the S&P to record highs, but also the economy which somehow crashes every time there is snow in the winter.

Which brings us to the most recent idiotic proposal, one which had been hinted at several months ago by the Chicago Fed, and which has gotten significant prominence in recent days after the San Fran Fed came out of the closet, and said it's not its fault it has been perpetually wrong with its permabullish forecasts (unlike the Atlanta Fed of course, whose impartial, unbiased, numbers-driven model has been spot on): it is the seasonal adjustments. Or rather, lack of a second seasonal adjustment. Because, you see, the "big thing" in economics right now is that seasonally-adjusted economic data is simply not seasonally-adjusted enough!

One wonders if the Fed looked at the Q3 GDP print of +5% with the same alarm, and said the number was too high so clearly it is time to apply a "summer seasonal adjustment" reduction to outlier numbers... to the upside. Turns out the answer is no: the only numbers the Fed cares to keep massaging, are those which are below where they should be.

And now, with the Fed giving its blessing to economists around the globe to be not only wrong, but to blame their "inaccuracy", here comes Wall Street: the one place whose economists have been even more dead wrong than those of the Fed.

Enter Bank of America, with its overnight report "Smarter seasonal dummies" in which it does precisely what has just made every economist an even greater joke, and has also jumped the shark with not only but two seasonal adjustments to GDP.

From BofA:

There has been a flood of papers on “residual seasonality” in recent weeks; here are some thoughts on the main papers. The bottom line: adjustment problems are quite real and probably bias down 1Q GDP growth by at least 1.5%, and bias up other quarters by about 0.5%.

Quite real, for sure. Unambiguously so. And this is how the data would look like in a world in which data is irrelevant, but 2x seasonally adjusted, aka goalseeked data, mattered:

BofA is quite correct when it says that "Currently, 1Q GDP is likely to come in around -1.2%, so fixing the data gets us back above zero." Fixing indeed.

Some other observations from BofA on this quite humorous topic for anyone still paying attention:

Glenn Rudebusch and colleagues at the San Francisco Fed argue that residual seasonality could exist because not all the components of GDP are adjusted and the adding-up process could create further seasonality issues. They note that the risk of problems is high in 1Q, because the overall economy has a major “seasonal recession” in 1Q and, hence, big adjustments are required to normalize the data.


Rudebusch, et al, re-adjust the data using the usual X-12-ARIMA statistical filter and find significant effects in 1Q. Notably, the new factors boost 1Q GDP growth from 0.2 to 1.8%. This seems sensible to us. Unfortunately, as we note above, GDP continues to track lower. One simple approach to update their analysis is to apply their corrected seasonal correction factor to our -1.2% tracking estimate. This suggests 1Q real GDP growth of 0.3%—still quite weak, but no longer negative.


Alternatively, one could replicate their approach by re-applying the X-12-ARIMA statistical filter to the originally adjusted GDP data, substituting our tracking estimate for the initial 1Q release of 0.2%. This allows the seasonal adjustment process to apply to the full range of data. In that case, we estimate 1Q growth would be not quite as soft, at 0.9% (Chart 1). The gap between these two revised estimates highlights the unavoidable uncertainty in this exercise: at the end of a data  sample, there isn’t as much information to clearly signal whether a surprise is just a seasonal distortion or something more fundamental. What is clear is that the current 1Q GDP data still have seasonal components that we should look past to infer the state of the economy. While tracking has drifted lower as new data have come in, 1Q may not have outright contracted once the seasonals are corrected.


Tom Stark at the Philadelphia Fed takes a different track, using “dummy variables” to pin down which parts of the data have the biggest problem. As with many other papers, he finds that the adjustment problem seems to have started in the mid- 1980s and has gotten successively worse. He finds the biggest problem for government consumption and gross investment, with weaker effects for residential and trade. From 1985 onward, GDP growth has averaged 1.9% in 1Q and 3.3%, 2.9% and 2.7%, respectively, for the subsequent quarters. Note that the implied seasonal distortion in Stark’s work is about 1%, which is smaller than Rudebusch, et al. That is because Stark’s approach gives equal weight to every year, while Rudebusch, et al’s, approach puts greater weight on recent data.


One of the compelling things about the paper is the robustness of the results. Readers should be skeptical about statistical results that don’t hold up when there are minor changes in the exact test. Stark shows that the bad 1Q is not the result of a few really bad numbers: it holds up when you exclude the extreme results of the last four years, and it is relatively consistent in terms of which sectors create the problem.


Stark also looks at alternative measures of overall economic activity and finds much smaller seasonal adjustment problems. GDP is based on an adding-up of the expenditure side of the economy. An alternative approach is to add up incomes: Gross Domestic Income (GDI). The two can diverge, particularly on a quarterly basis. Unfortunately, GDI is unavailable in the advance release, so it remains to be seen whether there are any significant 1Q distortions. But, the Philly Fed’s GDPplus—a composite of GDP and GDI; the latter is presumably estimated—has insignificant seasonal issues. It grew 1.65% in 1Q.

Things get awkward when as BofA admits, in a separate paper, none other than the Fed (?!) did not find a reason to doubt the original numbers.

Some recent pieces have been more skeptical of the significance of these results, including a paper from the Fed Board by Gilbert, et al. For example, looking at data for 2010 to 2014, they confirm that GDP growth was 1.7% lower in 1Q.  However, excluding 2011 and 2014, the drop off is just 0.2%. Moreover, they argue that the proper test for seasonal distortions is not to zero in on 1Q, but to test whether any quarter is significantly distorted (this makes it harder to prove statistical significance). The probability that any quarter averages 1.7% less than other quarters is 7%, flunking the usual 5% threshold.




A recent paper from the Chicago Fed—“The effect of winter weather on U.S. economic activity”—is the best attempt we have seen in recent years. They look at detailed data on snowfall and temperature by state and for the nation as a whole. The results show that weather effects can have a significant impact on local employment and housing activity, however, when you add it up for the nation it becomes very hard to quantify. Looking back at the very severe winter of 2013-14 they find that bad weather can only explain part of the weakness at the start of the year.



How does this finding impact the debate? It suggests some skepticism is warranted. However, after slicing and dicing the data in many ways, the result still seems economically important. Moreover, in the real world of forecasting, ultimately, a judgment has to be made: do we wait for more data to nail down the case statistically, or go with a reasonable story and reasonable evidence? We choose the latter.

BofA' conclusion: it is not our models that are wrong, it is reality!

Macroeconomic Advisers has a model that captures both the seasonal distortion and the impact of worse-than-normal winter weather. They argue that bad seasonals subtracted 1.6% from the quarter and unusually bad snow subtracted another 1.6%. We think the seasonal estimate makes sense, but the weather variable seems a bit high. The only truly bad month this winter was February. Moreover, estimates of bad winter effects are sensitive to the exact measure used. Hence, we assume a bad weather effect less than half as big.

But where things for BofA get really awkward is when one recall that just one month ago it was Bank of America itself which said that it was "in an awkward spot" because the Q1 weakness can not be explained by weather, saying that "While we would love to blame the weather for all of our bad forecasts, in reality it is hard to pin down  weather effects" To wit:

Is weather the main reason for recent weak economic data? While we would love to blame the weather for all of our bad forecasts, in reality it is hard to pin down  weather effects.


This puts us in an awkward spot today. There is a bit of an urban legend that weather can explain all of the weakness in the first quarter of last year and hence could explain all the weakness today. However, hindsight is always 20-20. In real time, the slowdown last year was a major surprise to economists even though we get data on the weather before we get data on the economy. Moreover, this winter is not nearly as bad as last winter—last year we had three bad months, this year only February was unusually bad (Chart 1). Economic fundamentals point to stronger growth ahead, and that remains our forecast. However, we can’t completely explain the recent weakness and hence there is a risk that growth does not pick up.

And here is the actual evidence. Unadjusted.

But where things get most awkward, is that none other than a Federal Reserve Bank, that of Atlanta, continues to insist that the US is effectively in a technical recession with a H1 GDP print that will be negative, thanks to a Q2 GDP of just 0.7% following as -1.0% or worse Q1 GDP.

And when even the central planners in control of the economy can't keep their lies straight, then it may be time to panic.

As for the the double, triple and so on seasonal adjustments, we eagerly await the Fed to downward revise Q2 and Q3 data for the same but inverse reason: the weather was too nice. We may have a while to wait.