Conceptually, one may define seasonal adjustment as the purging of any variations in economic data that are predictable using the calendar alone. This includes not only effects associated with the time of year but factors such as the timing of Easter or the number of business days in a month. It does not include variations in economic data owing to deviations in weather from the norms for a given time of year.
What makes estimation of seasonal effects dif cult is that they can change over time. For example, the rise of air conditioning changed the peak of electricity demand from the winter to the summer (this is, for example, documented in Energy Efficient Strategies 2005). Demographic trends affect the number of school-and college-age people seeking employment primarily during the summer. Climate change may also affect seasonal patterns. If seasonal effects were constant over time, econometricians could eventually learn the “true” seasonal patterns. But given that seasonal effects do vary over time, the seasonal factor is an unobserved component that can be estimated but never perfectly identi ed.
Unfortunately, in academic economic and econometric research, issues of seasonal adjustment are typically given short shrift. A great deal of work has been done on the question of how to do seasonal adjustment, but these papers get limited outside attention and are seldom published in leading journals. Most academics treat seasonal adjustment as a very mundane job, rumored to be undertaken by hobbits living in holes in the ground. I believe that this is a terrible mistake, though it is one in which the statistical agencies share at least a little of the blame. Statistical agencies emphasize SA data (and in some cases do not even publish NSA data), and while they generally document their seasonal adjustment process thoroughly, they do not always do so in a way that facilitates replication or encourages entry into this research area. Yet seasonality is both substantively important and difficult. It essentially involves issues such as bandwidth choice, or choosing between parametric and nonparametric approaches, that are all quite standard in modern econometrics. In short, seasonal adjustment could and should be better integrated into mainstream econometrics.
This paper therefore revisits the question of seasonal adjustment, including the dif culty of disentangling seasonality from cyclical factors. It focuses on seasonal adjustment of the BLS current employment statistics (CES) survey (the “establishment” survey), which includes total nonfarm payrolls, since this is the most widely followed monthly economic indicator.
In my paper “Unseasonal Seasonals?” I argue that a longer window should be used to estimate seasonal effects. I find that using a different seasonal filter, known as the 3×9 filter, produces better results and more accurate forecasts by emphasizing more years of data. The 3×9 filter spreads weight over the most recent six years in estimating seasonal patterns, which makes them more stable over time than in the current BLS seasonal adjustment method.I calculate the month-over-month change in total nonfarm payrolls, seasonally adjusted by the 3×9 filter, for the most recent month (column Wright SA). The corresponding data as published by the BLS are shown for comparison purposes (column BLS Official). According to the alternative seasonal adjustment, the economy gained 190,000 new jobs last month, compared to the official BLS total of 215,000. Data updates released today for prior months also reveal some differences between my figure and the official jobs gains from prior months. The official BLS numbers for February were revised up from 242,000 to 245,000 new jobs; my alternative adjustment shows the total actually falling slightly, from 261,000 to 258,000 new jobs. For January, the official numbers were revised down slightly from 172,000 to 168,000, leading my alternative seasonal adjustment to be revised down from 109,000 to 100,000. On net for January and February, official jobs gains were revised down by 1,000, while my alternative seasonal adjustment revised down by 12,000 jobs. The discrepancies between the two series are explained in my paper.In addition to seasonal effects, abnormal weather can also affect month-to-month fluctuations in job growth. In my paper “Weather Adjusting Economic Data” I and my coauthor Michael Boldin implement a statistical methodology for adjusting employment data for the effects of deviations in weather from seasonal norms. This is distinct from seasonal adjustment, which only controls for the normal variation in weather across the year. We use several indicators of weather, including temperature and snowfall.We calculate that warm temperatures and the lack of a major snowstorm in March added roughly 14,000 jobs (column Weather Effect), on top of a February jobs level that was itself inflated by unseasonably mild weather. Controlling for the effects of unusual weather produces a weather-adjusted figure of 201,000 new payrolls in March (column Boldin-Wright SWA), down from the official number of 215,000, but still yielding significant positive jobs growth.At some point over the next few months, the weather effects are likely to turn negative as temperatures and snowfall return to their seasonal norms, because the model ties the level of employment in a given month to unusual weather in that month. Our weather adjustments are designed to be able to see through transient weather-induced effects, both positive and negative, and to be used to gain a better handle on underlying job trends in the economy.
Chosen excerpts by Job Market Monitor. Read the whole story at However you look at it, March 2016 was another strong month for job growth | Brookings Institution