It is now the end of June, so I figure it’s time to update the Google “unemployment benefits” data – to see what the full month tells us.
The story is basically unchanged, in seasonally adjusted terms searches for unemployment benefits declined a little in June. Historically, this has been a reasonable guide to the official ABS unemployment data (released 7 July).
As it stands, the data suggests that the unemployment rate will remain ~4.9% in June. This would not normally be news, except that the Aussie short end has been flirting with rate cuts (and the equity guys are talking “recession outside mining”).
The BoE used an AR framework for their out of sample test, but I dislike those for my purposes as they tend to lag at turning points: if you want to make money, you need to have an idea when the data is turning.
I prefer a simple OLS — I start in Oct’01, when I have 20 observations, and forecast the current period unemployment rate using the google search data (which we have at month end, giving us a lead of a week or so). I then increase the history by one observation (so next using 21 pairs) , re-estimate and forecast the next period (step two is Nov’01), and so on …
This method predicts the increase in the unemployment rate in Q2’08, and the predicts the peak (a little early) in Q1’09. In both cases, this was early enough to make large in AUD fixed income markets.
Aside from predicting the unemployment rate, the search data also gives us interesting colour on who’s searching. The pain in Qld is clear – as the very large gas investment projects come online in H2’11, we should see that ease back a little.
Finally, google correlate allows see what unemployed Aussies do with their spare time — by uploading the data and letting google find the best match —
— apparently we are into “piercing pictures”… or something boring about spurious correlation and the importance of theory (note the recession in piercing picture searches suggest the unemployment rate has further to fall!).