12 Month Unemployment Forecast

[bcassill]

Just ran a quick forecast based on the latest BLS using single series ARIMA on the U-6 unemployment numbers. The 12 month forecast is not pretty (click on the attached image to see the graph). Seasonally unadjusted, we are at 16.5% for August. Look for 20%+ by next August (19% to 24% at the 90% CI level).

[plinko]

Re: 12 Month Unemployment Forecast

I'm kinda confused. Is this type of analysis valid for estimating future unemployment? Partly curious because I have no idea, partly skeptical because I would expect more political events to possibly shift the numbers.

My personal theory so far has been that we'll have inflation with stagnating wages via an increase in "Discouraged employment" numbers. Update: Of course, since you're measuring U6, you're estimations seem to make more sense perhaps.

ripping wikipedia's unemployment indexes here for reference.

• U1: Percentage of labor force unemployed 15 weeks or longer.
• U2: Percentage of labor force who lost jobs or completed temporary work.
• U3: Official unemployment rate per ILO definition.
• U4: U3 + "discouraged workers", or those who have stopped looking for work because current economic conditions make them believe that no work is available for them.
• U5: U4 + other "marginally attached workers", or "loosely attached workers", or those who "would like" and are able to work, but have not looked for work recently.
• U6: U5 + Part time workers who want to work full time, but cannot due to economic reasons.

[*T*]

Re: 12 Month Unemployment Forecast

Just ran a quick forecast based on the latest BLS using single series ARIMA on the U-6 unemployment numbers. The 12 month forecast is not pretty (click on the attached image to see the graph). Seasonally unadjusted, we are at 16.5% for August. Look for 20%+ by next August (19% to 24% at the 90% CI level).

Interesting -- how reliable a predictor has this modelling method proved in the past?

[bcassill]

Re: 12 Month Unemployment Forecast

Interesting -- how reliable a predictor has this modelling method proved in the past?

Single series ARIMA (i.e. autoregressive integrated moving average) models used auto-correlated information to create forecasting models. Although there are some checks you need to make along the way and the process is a little trial and error to get a good model (i.e. the art part of the science), these models can be accurate if there is a deterministic component to the series. To your point, shocks can throw a wrench into the forecast, but shocks will do that to any forecast, no matter what the method used. However, in this case, I'll stick with the forecast I've made. The 90% confidence intervals of 19% to 24% basically mean that we can be 90% sure that the true future value will fall somewhere between the high and the low estimate 12 months out.

[ThePythonicCow]

Re: 12 Month Unemployment Forecast

Single series ARIMA ... can be accurate if there is a deterministic component to the series. ... The 90% confidence intervals of 19% to 24% basically mean that we can be 90% sure that the true future value will fall somewhere between the high and the low estimate 12 months out.

That's 90% confident if the past fortells the future; if there is not only a deterministic component to the series, but if the sum of the deterministic components (including possible random variations constrained to some center weighted probability curve) is the series.

This is drawing a line with a French Curve instead of a straight edge ruler to extend the past trend.

It works until it doesn't work.

Estimates of confidence are deceptive in such uses, because the underlying assumption is guaranteed to fail sooner or later in large and small ways.

One could get more useful estimates of confidence by back-testing the model. For example, one could take the unemployment rates for each month going back many years and then for each period of N months, run the model to see what it says for the following M months. Then over all such runs, see how well the model did.

That works better; until an even bigger black swan flies by.