Estimate of Unemployment

With the unemployment figures steady for the last three months and the fiscal policy adopted by the Australian government given almost universal plaudit in saving jobs I thought I would try my hand at estimating whether unemployment will rise or fall in the coming months.

I should preface my comments with I have no knowledge in statistics or trend line regression that I attempt in this scenario. *It is all just a best guess I could be using a completely inappropriate process and be totally wrong.*

Above we have unemployment statistics as recorded by the RBA to a line graph and a third order polynomial regression trend line attempting to forecast six months into the future.

In the graph above we see the trend line just above 5.8% in September at 5.9% as estimated by a product called Engauge^{1}. It stays stable at that throughout October and November before falling in December back to a trend of 5.8%, 5.7% in January, 5.5% in February and 5.2% in March.

Using this analysis, suggests that Unemployment could begin to fall rapidly as early as December.

**UPDATE:**

It looks like I am going to have two bites of the cheesecake on this one. I mistakenly placed 5.8% in the September 2009 unemployment figures when that figure is true only for August 2009 at this stage. We will find out at 11.30am on the 8th of October whether this figure also holds true for September. If it does the above estimate analysis should hold true.

However, since I erroneously placed some non-existent data into the original analysis, let us reassess with the correct data.

Using the same techniques as described above, the trend line sneaks up to 6% in October and stays steady until about February where it begins to fall, first to 5.9% and then to 5.7% in March.

Assuming these estimates are even remotely done correctly, unemployment should begin to fall in either December 2009 or February 2010.

Either way unemployment should be falling by the end of the third quarter in the 2009/2010 financial year.

^{1 Another product brought to you by the Stubborn Mule.}

### Like this:

Like Loading...

*Related*

Senex(x):you did provide the caveat that you could be using a“completely inappropriate process“, and I’m afraid you are.Time series analysis can be a tricky business, as I’ve described elsewhere. Polynomials are particularly bad for extrapolation because they are quickly dominated by the largest power term. First, it’s important to note that the fitting process does not attempt to put any constraints on the trend outside the data being fitted. In your case, the coefficient of the third order term in the polynomial trendline is small, but negative. This forces it to go down as you move further out in time. If you tried a degree six polynomial you would find what looks like a good fit to the data, but since it has a positive coefficient for the sixth power term, it starts to rise rapidly as you move forward in time.

The other thing to add is that, although I am a fan of Engauge, you are probably making life harder for yourself than necessary. As I understand it, you used Excel to add a trendline to a chart, saved it as a picture and used Engauge to pull off the data points. While Engauge excels (no pun intended) at extracting data from images, since you have everything you need in Excel, you don’t need to use this technique here. You can use the LINEST function in Excel to calculate the same trendline for yourself and read off the predicted values in your spreadsheet. I’ve uploaded an example.

And there you have it folks, a

completely inappropriate processbut valuable lessons learnt along the way.I did try the whole LINEST function now I have seen it at work but could not get my head completely around it.