The "weather impact" debate has been ongoing for much of this year and looks set to continue for a while longer. Initially there seemed to be a clear narrative that colder-than-normal temperatures were slowing economic activity and markets seemed to gladly discount weak figures. But several recent data releases have brought confusion to the tidy story.
On Wednesday, US housing starts for January were weaker than expected, with weather the obvious explanation. But a closer look at the report showed that while housing starts were exceptionally weak in the Midwest, as expected, they rose in the Northeast, which was also hit by severe weather. In addition, starts declined in the West, where temperatures have been warmer than average.
US PMI manufacturing data for February was released Thursday and while some economists were warning that weather would be a factor, the measurement rose more than forecasts. The latest reading was the highest since May 2010. Of course, like any data point it could be an outlier and is subject to revision, but it certainly disrupted the "cold weather hurting data" narrative.
Statisticians attempt to take account of factors such as changes in weather by using seasonal adjustments. Retail sales is a data point particularly vulnerable to changes in the calendar. Without seasonal adjustment, it would be impractical to compare changes in retail activity between December and January. There would always be an expected decline, given the surge in spending around holiday season, but this should not be indicative of weakness in consumer activity.
The US Bureau of the Census, which compiles the report, assumes that retail sales drop by 21% in January as consumers wind back spending and weather typically creates harsher conditions. So January's data is increased by 21% enabling the figure to be compared with that of December. But a problem arises when temperatures are more extreme than statisticians have accounted for. There is no definitive way to quantify what impact weather has on a data report.
BofA Merrill Lynch published a study on February 4th that examined changes in average weather temperatures versus changes in US payrolls and retail sales figures. The correlation between deviations from average weather and changes in payrolls from their three-month average is 0.72 for December, 0.42 for January and 0.29 in February. This indicates that weather changes have the biggest impact on payrolls in December, with the relationship weakening in subsequent months.
Conversely, for retail sales the correlation with temperatures is just 0.16 in December, then strengthens to 0.27 in January and 0.41 in February. The authors note that this could be explained by shoppers facing a deadline for holiday gifts, but no such pressure exists in subsequent months.
While the study is interesting, it really only tells us about the relationship, not the actual impact of weather on the data. Economists seem to agree that any delay in consumer demand brought about by adverse temperatures is usually "pent-up" and released when weather returns to normal. Markets will be hoping for such a scenario and that the perplexing data isn't a sign of broader weakness.