Fundamentals Update: US Cold Weather

US Cold Weather

Fundamentals Update as at 11 February 2014 by Lorenzo Beriozza

The challenge is to quantify the drag to economic growth and determine how it manifests itself in the high frequency data. Focusing on the winter months of December and January, weather can explain about a third of the deviation from the underlying NFP trend (non-farm payrolls). The correlation is a bit lower for housing starts, with weather explaining about a quarter of the deviations from normal in these months. There is no significant weather impact for retail sales, even for specific components. If the weather is indeed depressing the data, as it is to some degree, a healthy bounce back is likely in the spring.

Concern over the weather started after the unexpectedly weak December jobs report. Non-farm payrolls were reported to have increased only 74,000 (subsequently revised to 75,000), a sharp slowdown from the three-month prior trend of 185,000. Weather played also a role in the January report, with job growth of only 113,000. There was a bounce back in the goods-producing sectors of construction and manufacturing, but weakness in the services side of the economy. This outcome was surprising to many since weather was worse in January than December.

But it is not that simple. The following factors must be controlled:

  1. How does the temperature compare to normal during this time of year?
  2. Was there a seasonal distortion the month prior?
  3. What were weather conditions during the survey week?

For January, the answers are:

  1. Colder than normal;
  2. Yes, there was a seasonal distortion in December; and
  3. The survey week was warmer than normal.

As a result, the goods-producing sectors that weakened in December seemingly reversed in January from one survey week to the next. But poor weather conditions in the month likely weighed on overall economic activity.

By the time economic data are released to the public, they have undergone a vigorous adjustment process. Part of the smoothing is to remove seasonality, which includes controlling for the impact of normal weather patterns, holidays and other special events such as the start of the school season. The typical seasonal adjustment process is the X-12, a statistical method that estimates the underlying trend, seasonal effects, and irregular effects based on historical and future data if available. Of course, this is imperfect and, when there is abnormal weather, the seasonal adjustment will not capture it in real time.

An attempt to quantify the bias to the jobs numbers created by poor seasonal adjustments can be tried. Focusing on the winter months of December and January, deviation in temperature from the prior 5-year average is calculated. The 5-year average is used rather than the average of the full history since the seasonal adjustment process puts a greater emphasis on recent trends. A comparison to the deviation from trend in payrolls is performed, which is measured as the change in payrolls in the current month compared to the average change over the prior three months. Over the full sample from 1980 to 2013, the deviation in temperature can explain 22% of the miss in payrolls in December and 23% in January.

If the sample is restricted to just significant misses, which are defined as anything greater than two degrees in either direction, the R2 increases to 34% on average after combining the sample to December and January. Using this equation, payrolls should have been biased lower by 41,000 in December and 56,000 in January. Instead, 111,000 fewer jobs were registered relative to the three month trend in December and 47,000 in January (note that the three-month moving average for January is lower since it captures weakness in December).

A test can also be run to see if comparing the deviation in temperature in the survey week, which is the pay period containing the 12th of the month, improves the significance. Using this restricted sample, the survey week has no more significance than average temperature for the month. This could be because what is being measured is the deviation in temperature rather than capturing extreme weather events. Looking at the latter, the survey week would likely be more relevant since snowstorms will prevent people from getting to work and close businesses.

In addition to the labor market, abnormal winter weather impacts housing construction, but there is no significant relationship with retail sales. Running the same exercise from 1963 to 2013, weather can explain 14% of the deviation in monthly housing starts over December and January. The relationship strengthens to reveal a R2 of 23% with a restricted sample as above.

There is no significant relationship with retail sales. Perhaps this is because the weather causes a reallocation of spending – toward necessary goods to combat the poor weather or shopping online versus big-ticket items or restaurants. But digging into the details does not improve the significance. In fact, there are counterintuitive results. For example, it was abnormally warm in January 2012, but there was only a small upward surprise relative to trend and motor vehicle sales were surprisingly weak. In January 2008, the weather was colder than normal, but online shopping was weaker than the recent trend while motor vehicles were stronger. The market will likely be looking for a weather distorted weak retail sales report in January, but it is arguable that the risk is to the upside given the historical relationship.

The main takeaway from this exercise is that the cold winter weather is likely responsible for a modest slowdown in the data, but there isn’t a consistent bias.

 

Source: Bank of America Merrill Lynch
2017-05-07T22:18:41+00:00