Car sales have fallen off a cliff. The latest figures from the motor trade body smmt show that in November sales were more than a third down on the same month a year ago.

This news reminded me of a spookily strong relationship between new car sales and the official data for private housing repair, maintenance and improvement.

The figures, roughly speaking, suggest that the level of private housing rmi work follows a similar path to car sales, but about six months to a year later.

If this relationship holds Britain's local builders can expect a very nasty beating in the very near future.

Graph 1 below shows new car sales and private housing rmi up to the third quarter of this year.

car sales v priv housing rmi 1.GIF

The fall in car sales since then has been even more pronounced. Broadly car sales started to sag badly in the summer and the decline has gathered pace.

I looked at the relationship a few years ago, interested to see if there was a link given that they are both big ticket buys for households.

I was surprised at how closely the paths tracked. I have rerun the numbers thanks to smmt providing me with new data. They are on a 12 month rolling total basis, so smooth out the lumps and bumps.

I have also put in gross domestic product and real household disposable income, which you would have thought would be major factors influencing both. What is clear is that there are other and apparently similar factors influencing both housing rmi and car sales - levels of debt perhaps?

Mucking around with the numbers the best fit I could get was with a nine month lag. You can see that more clearly in graph 2.

car sales v priv housing rmi 2.GIF

Introducing that lag provided a correlation of 0.93. If you ignore the last three quarters figures for private housing rmi, which I have long thought are suspect, you get a correlation of 0.95.

Also the data supplied included fleet sales, so one might assume the relationship would be closer if it were simply for private sale.

I can't claim to understand the causal link, if indeed there is one. This is only some rough and ready calcs I felt like doing.

But if some clever statistician out there with a few hours spare on their hands cares to have a look at the data and find out what if anything links the two I would be most interested to hear what they have to say.