Forecasting – the unemployment numbers
Over the weekend the place where I worked was shut down due to partner dispute so I did not do my usual posting on Sunday. Now I’m going to make up today and maybe post more frequently until something else comes up, jobwise.
Continuing on the topic of Kaiser Fung’s Numbersense book, I’m going to talk about a small section of chapter 6 which is mainly about unemployment numbers. He has a great section on using raw numbers versus adjusted numbers. He also has fabulous graphs on the various unemployment numbers U1 to U6, which I won’t talk about here.
If you don’t know it, the unemployment numbers are seasonally adjusted to take into account the normal rise and fall of seasonal employment. Page 138 shows the remarkable consistency of the rise and fall from year to year. There appears to be no variations whatsoever. The BLS folks extract out these seasonal rise and fall to give us the big employment trends.
Apparently some folks say we should use the raw data because they are the “real” data. The graph on page 136 gives the raw data as the grayed dotted line. The black line gives the unemployment trends with the seasonal variations (page 138) subtracted out. The black line gives you a trend whereas the gray dotted just gives you noise. In this case, “real” data is not really telling you anything meaningful.
Kaiser Fung says: When looking at other’s prognostications, you need to look deep at the source of the data. You need to dig out how it is collected and how it is used/manipulated before you really understand the interpretation.
When doing your own forecasting, you will need to make your own adjustments. Kaiser Fung on page 140 lists some factors you need to be aware and adjust when doing your own forecasting:
- Number of days per month
- Number of weekdays per month
- Number of workdays in a paycheck
A lot of financial analysts or project controllers will do a straight line forecast because it is the easiest (and lazy) way of doing forecast. But if your expenses are predominantly labor, then holidays or workdays per month can swing your numbers. Make your adjustments to show the variability of the business. Unlike the unemployment numbers, quarter results (or month end results) with all of its variability will be compared to your forecast.
But if you want to discern trends, then you should try to figure out how to extract the “seasonality” to get a sense of the underlying trends. This is the method the folks at BLS are using: they are trying to get at the sense of the underlying trends, not perfectly forecast the numbers.