Goldman's Meteoronomists Have A Dire Forecast: "Winter Is Coming"

What little credibility the shamanistic voodoo religion that is economics had, it lost over the past 2 years when even the most modest downtick in economic activity was blamed on the "weather." It appears that as part of their conversion from "economist" to pure-play weathermen, nobody advised Wall Street's if not best and brightest, then certainly dumbest Keynesians, that adjusting for the seasons, is precisely what seasonal adjustments are for, and why they spend hundreds of hours goalseeking every data point with Arima-X-13 models until they get the result they want.

It was not enough, and in the winter of 2013 and 2014, the farce was indeed complete, when none other than the Bureau of Weather Economic "Analysis" incorporated double seasonal adjustments, to smoothe away what to most was an "inexplicable" slowdown in the US economy, and which was simply a function of two consecutive credit crises hitting China in the latter part of 2013 and 2014.

However, instead of modeling how two consecutive years of China's slowing credit impulse slammed US growth, the economisseds instead decided to blame it all on the unprecedented events of cold and snow in the winter as they relied on their favorite forecasting tool...


So with the winter of 2015 so far shaping up to be what some have dubbed "abnormally hot", we thought that at least this year the weatherconomists would keep their mouth shut: after all, if you blame cold weather for an underperforming economy, you better say nothing at all if the weather is warmer than usual as it has been in October and November.

Alas, it was not meant to be, and so, without further ado, here are everyone favorite economweathermen from Goldman Sachs, warning everyone that, drumroll, yes, Winter Is Coming.

No really, that's the title.

Here is the full 2000-word "explanation" from Goldman's team of merry weathermen:

Winter is Coming

  • Growth decelerated sharply in Q1 in 2014 and 2015, and we suspect that unusually harsh winter weather contributed. With the winter season now upon us, we revisit old lessons learned and develop new rules of thumb for estimating the economic impact of weather fluctuations.
  • We focus on two weather indicators that measure temperature and snowfall. The first is the deviation of “heating degree days” (HDD), a measure of cold temperatures, from seasonal norms. The second is the Regional Snowfall Index, a measure of the societal impact of snowfall that includes scores for hundreds of major snowstorms.
  • We draw three sets of conclusions about the impact of weather on the economy. First, we find that both temperatures and snowfall matter for growth. In particular, we estimate that a 1 standard deviation (SD) increase in HDD is associated with a 0.4 percentage point (pp) reduction in GDP growth and a 0.1pp reduction in our current activity indicator (CAI), while a 1SD snowstorm is associated with about a 0.3pp reduction in both. Admittedly, there is considerable uncertainty around our estimates due to both collinearity between temperatures and snowstorms and the limited sample size at the aggregate level.
  • Second, we find that weather effects have a “tell” in the form of an uneven pattern of impact across the economy. Comparing the effect of weather variables across top-tier indicators, across sectors in the Gross State Product data, and across industries in the payrolls report, we find that weather typically has the largest impact on construction, retail trade, leisure and hospitality, foreign trade, and manufacturing.
  • Third, we find that weather variables have an important but somewhat more nuanced effect on the payrolls report during the winter months. Using state-level data, we find that the intra-month pattern of weather conditions is important, with conditions during the reference week carrying the greatest weight. We estimate that a 1SD colder month and 1SD of snowfall during the reference week are each associated with a roughly 35k reduction in payroll growth.

The sharp deceleration of Q1 GDP growth in both 2014 and 2015 has provoked some anxiety about what to expect this winter. At the time, both we and Fed officials pointed to unusually severe winter weather as one contributor to the first-quarter slowdowns. To some skeptical investors, economists who attributed weak Q1 growth to weather effects sounded a bit like Peter Sellers’ Chance the Gardener promising that “there will be growth in the spring.” But in both years, growth did rebound strongly in Q2, suggesting that weather effects had in fact contributed to the weak Q1 performance. While we certainly do not claim to be able to predict this winter’s weather, we can estimate the impact of weather deviations from seasonal norms once they occur. In this week’s Analyst, we revisit some old lessons learned and develop some new rules of thumb for assessing the impact of weather conditions on growth and employment.

Why Weather Matters

Most economic data are seasonally adjusted to account for weather patterns as well as other calendar effects such as holidays. But weather can still affect economic data when it departs significantly from seasonal norms. For indicators such as housing starts normal seasonal fluctuations are very large and mostly weather-driven, meaning that even moderate deviations from normal seasonal weather patterns can have large effects not captured by seasonal adjustment.

We focus on two weather variables that capture temperature and snowfall. We measure temperature effects using the deviation of the number of heating degree days from a trailing 10-year average. Heating degree days (HDD) are a measure of cold temperatures that we have found in past research help to predict a range of economic data. We use both state-level and national population-weighted series constructed by the National Oceanic and Atmosphere Administration (NOAA). We measure snowstorms using the Regional Snowfall Index (RSI), a measure designed to capture the societal impact of major snowstorms. The RSI provides dates and scores for over 600 storms since 1900 across six regions of the US, and we construct a monthly national index—shown in Exhibit 2—by aggregating the regional indices using relative population weights. We also convert the regional series into state series by assuming that the impact of a given storm is equal across all states in a region.

Weather and Growth

We start by estimating the growth effects of weather deviations. Exhibit 3 shows suggestive evidence that at least in the most severe deviations from normal weather patterns—in this case, the 25 months with the greatest snowfall since 1972—our current activity indicator (CAI) has dipped by a bit more than 0.5pp during the month of the storm before rebounding the next month. The average dip in the 25 coldest months as measured by HDD (nine of which also had top-25 snowfall) is more modest at about 0.25pp.

We next use simple models to estimate the impact of the weather variables on both quarterly GDP growth and the CAI. Using data since 1985, we regress each growth variable on its own lag as well as both contemporaneous and lagged weather variables. We impose on the models the constraint that the coefficients on the several HDD and snowfall variables, respectively, must sum to zero, so that there is no permanent effect of weather fluctuations on the level of output. Exhibit 4 shows the resulting estimates.

The models imply that a 1 standard deviation increase in HDD relative to the trailing 10-year average (calculated as the standard deviation among only cold-weather months, equal to roughly 150 in a quarter in the case of GDP or 70 in a month in the case of the CAI) is associated with a 0.4pp reduction in GDP growth and a 0.1pp reduction in the CAI. The models also imply that 1 standard deviation of additional snowfall subtracts about 0.3pp from both the CAI and GDP growth. Our top-down finding of a smaller impact on the CAI is in line with our previous bottom-up analysis of weather effects on the CAI and is also consistent with the more modest deceleration of the CAI seen over the last two winters.

We caution that that there is considerable uncertainty around our estimates. In particular, collinearity between temperatures and snowstorms and the limited sample size at the aggregate level, especially for GDP, mean that the results are sensitive to model specification. That said, the models imply that snow and temperature deviations combined subtracted about 0.8pp from GDP growth in both 2014Q1 and 2015Q1, and we think they provide reasonable rules of thumb for the growth impact of weather fluctuations.

The Weather “Tell”

Investors are sometimes skeptical of alleged weather effects on the economy, viewing them as simply excuses to explain away weak data. How can we be confident that we are seeing the effect of weather conditions as opposed to weak growth caused by other factors?

While it is impossible to be certain, weather effects have a “tell” in the form of an uneven pattern of impact across sectors of the economy. Exhibit 5 shows our estimates of the impact of temperatures and snowfall on a number of top-tier indicators. We find that weather tends to have the largest impact on economic data related to housing and construction, retail spending, and trade. We also assess the relative impact on different sectors of the economy using state-level panel data on Gross State Product by sector, available quarterly since 2005. We find that the most weather-sensitive sectors include construction, mining, manufacturing, retail trade, and accommodation.

Finally, we can also look at the cross-sectional or category-level data within a particular report for typical weather patterns. For example, we have shown in past research that the impact of harsh winter weather differs across categories of retail sales, with the largest effects on building materials, vehicle sales, and furniture, and a positive impact on non-store sales, which include online purchases.

Weather and the Payrolls Report

We conclude by assessing the impact of weather conditions on payrolls. The impact of weather on the employment report is more nuanced because the precise timing of the payrolls reference week is important. We construct a state panel that includes payrolls and weekly HDD, recording for each month the degree days deviation during the reference week and the three previous weeks. We include all weekly variables for both the current and prior month in an initial regression in order to estimate the optimal relative weighting of the weekly weather observations. We find that the reference week is about twice as important as any other week, and we use the regression coefficients—shown in Exhibit 6—to calculate optimally-weighted HDD summary variables for the current and prior months.

We next add other weather data to our panel to see if those series are important too. We find that while the snowfall variable again has an economically and statistically significant impact, a parallel temperature measure called cooling degree days does not. We also find that the effect of precipitation is statistically significant, but quite small.

The richness of the state-level payrolls data enables us to address several more subtle questions that are difficult to answer convincingly with aggregate national time series:

  1. Do weather effects matter year-round? We find that the effect of HDD deviations is not statistically significant in single-month samples from June to October.
  2. Is the impact asymmetric between warmer-than-usual and colder-than-usual months? By splitting the sample, we find that the per-degree day impact of colder deviations is about double that of warmer deviations.
  3. Is the effect linear? We find that a quadratic degree days term is not statistically significant. While there are an endless number of ways one could specify thresholds, we think that assuming linear effects is reasonable.
  4. Do weather effects reverse? In unconstrained regressions, the coefficients on the contemporaneous and two lagged weather terms usually sum to roughly zero, suggesting that a nearly full rebound usually occurs within a couple of months.

While these conclusions do not necessarily apply to all economic indicators, we think they offer valuable broader lessons. Based on these findings, we construct both aggregate and industry-level models using the optimally-weighted HDD deviation and a version of the snowstorm index adjusted for the timing of the payrolls reference week. Exhibit 7 summarizes the model estimates.

In aggregating industry effects to produce a bottom-up estimate of the total impact, we only include industries whose temperature or snowstorm effects are statistically significant. We find that a 1 standard deviation colder month (about 70 HDD) is associated with a roughly 35k reduction in payroll growth, while 1 standard deviation of snowfall during the reference week is associated with a reduction of 25-45k. Once again, the weather impact leaves a familiar pattern, weighing primarily on employment in construction, leisure & hospitality, and retail & wholesale trade.

Slightly more favorable temperatures in the weeks leading into the November reference period should make a small positive contribution to payrolls this month. The employment components of business surveys have been mixed so far, and the labor differential included in the consumer confidence report declined. We therefore expect a gain of 200k in November, a bit softer than the 215k average gain over the last six months.

* * *

Yes, Goldman really spent a few days writing this.

And after that nearly 2000 words of worthless drivel, here is the punchline: if poor Q1 in 2014 and 2015 was blamed on the cold weather, then how many points of GDP in Q1 2016 (and Q4 2015) will be the result of abnormally warm weather (just don't ask the retailers who blame both hot and cold weather when their sales keep on declining) and will this be the first case in monetary policy history when a Fed hiked rates because it thought the economy was improving only to realize after the fact that it was merely ignoring the weather effect it had dissected so extensively in the prior two years, simply because this time it was in reverse and had been "boosting" the economy?