There was some confusion about core CPI yesterday, thanks to the divergence between trimmed mean CPI & weighted median CPI . This is not surprising as they are fairly esoteric concepts. The bottom line is that the trimmed mean is the best measure of inflation — as it preserves a lot more information about the distribution of inflation, but still cuts out a lot of the noise.
This is most easily described with the help of a chart. The below chart shows the distribution of QoQ CPI changes. The blue line is the weighted median, and the green line is the trimmed mean.
These are not weighted by how important each of the 87 expenditure categories are — but most folks agree that is an important next step. Say we want to cut out some noise — how much do we cut out? Do we cut it all out and just take the thing in the middle (the weighted median) or do we just cut out the outliers (the trimmed mean)?
The Blue line is what we have left if we look at the weights and cut out 86 of the 87 spending categories, and are just left with the expenditure category that’s in the middle (could be Bread one qtr, fuel the next, and vets the quarter after that). The green line is what we have if we just cut out the 15% of weighted spending categories at the top and bottom (the bits outside the red lines), and take the weighted average.
So which is best?
The RBA has looked at this question, and it’s found that the trim is the best quide if you want to forecast CPI. That means that if you want to forecast future inflation, the best guide is current trimmed mean CPI (in these charts you want the line to be as low as possible).
The RBA went with the 15% trim (slicing 15% from either side) as it performed best — the 50% trim, of course, is the same thing as the weighted median.
Another way of looking at this is the above chart — in this chart, you want the line to be as high as possible. The performance of the core measure peaks ~15% for Australian CPI, and starts falling steadily ~30% for the trim. There’s a notable loss of performance by the time we reach the weighted median (of ~50%).
This is why I focus on the trimmed mean, and not the weighted median.
I don’t know why the ABS even publishes the weighted median — it just confuses the issue. The weighted median is a case of cutting too much of the signal in an attempt to avoid the noise…
On this basis, the situation in Australian inflation is clear — the best measure shows that inflation has been running below 2% AR for the last two quarters. The RBA’s job is to hit 2.5%, so they are failing. They must cut rates.
With the unemployment rate trending up, it’s unlikely we’ve seen the low print for CPI — which means it’s unlikely that this is going to be the last rate cut. It might be if the AUD drops another 10c, but I do not think that’s likely (Fed-Taper is mostly already in the price of FX).