Whose GDP is it anyway?

They say what gets measured gets managed, but a measure as nebulous as GDP needs careful interpretation when used as a guide for economic management.  Putting aside the conceptual problems surrounding the use of GDP as a measure of progress, there is still the practical problem of taking estimates of production or expenditure in current prices and deflating these prices to determine a real change in the volume of good and services produced and consumed between time periods.

What good is GDP at all if we can’t separate price changes from volume changes?

We can get a feel for the challenges of separating price changes from volume changes by examining the way prices are deflated when using the expenditure method to calculate GDP.

* It is worth noting that the ABS has used an integrated and balanced accounting method for GDP calculations since 1994-95, which means that the income, expenditure and production methods for determining GDP must give the same result each quarter.

GDP nominal prices estimates are converted into chain volume measures (real terms) using implicit price deflators for each type of expenditure (consumption, investment and government). The data shows that consumption expenditure is approximately 70% of total GDP, while the household component of consumption is 53% of GDP. Having spent quite some time researching the statistical experiment known as the CPI, some odd patterns have appeared in the relationship between the CPI and the GDP deflators.

The ABS explains that its implicit price deflator for household final consumption expenditure (HFCE) should be somewhat similar to the CPI:

Movements in the chain price index for HFCE are generally very close to movements in the CPI due to the fact that most parts of HFCE are deflated by components of the CPI. However, differences do occur between the two price indexes in some quarters

The table below shows the difference in the weights in the CPI and the deflator, which can explain some variation in the short term:

But the data doesn’t seem to support the ABS assertion that the HFCE deflator and CPI will ‘generally be very close’.  Below is a chart of the cumulative underestimation of the HFCE implicit price deflator compared to the CPI since 1972. Since about 1993 the GDP deflator has been substantially lower than the CPI and the cumulative difference is now approximately 11%. It is definitely interesting that the variation between these two measures was minimal and corrected quickly prior to the mid 1990s.

The red line is the total GDP deflator (covering all items using the expenditure method), which seems to have recovered its extended discrepancy with CPI since 2003.

This observation leads to questions about the impact on GDP from miscalculation of the deflators.  An underestimate of the implicit price deflators for the GDP calculation (treated as positive in the graph) will overestimate real (chain volume) GDP.  Since 1997 the household final expenditure component of GDP (around 53%) would have been underestimated by 11%, resulting in a net overestimate of GDP by 5.8%.

The graph below makes adjustments for the cumulative overestimation of GDP since the late 1990s, and from this new measure is appears that GDP per capita is back at a level first seen in 2006.  This gives a slightly different picture to the one in yesterdays Chart of the day showing GDP per capita at 2008 levels.

This analysis does not mean that the ABS has their methods all wrong.  Indeed, there may be other biases that more than cancel out this single bias resulting in an underestimated GDP.  But it does show that the GDP figure is a product of the assumptions and methods used.  It should also raise the question of just how much of the measured economic growth over the past decade was tangible, and how much was simply the result of measurement methods. Historically the CPI and HFCE deflator diverged by small margins for a short periods, so the divergence over the past 15 years appears unique. Whether this pattern will correct and we will see a period of underestimation of GDP I am not sure.

The lesson here is that all economic measurement has bias.  Tread carefully.


  1. Who is the director of ABS accountable for the consumer goods basket and the method used in ABS measurements?

  2. I can’t see why there would be such a big variance between the deflator and CPI. It must have something to do with the make-up of the basket of goods and services in the CPI. Something seems shoddy though.

    • “Something seems shoddy though”

      Yep, that the point. Economic measurement is imperfect. While this appears shoddy, there may very well be other biases working in the opposite direction (which I have yet to find).

      For me it is also interesting because the CPI methodology changed dramatically in 1998. Prior to this time it was a cost of living index, and afterwards became a pure price index. There is pretty strong evidence that if CPI was still calculated as a cost-of-living index it would be much higher, meaning that GDP growth could have been much lower even than my recalculation.

      • Yeah I think the variance must have something to do with the CPI measurement. Have you ever looked in to the CPI? ie. the goods included in the basket, their weightings etc.?

        • That was discussed a bit recently, and one that got my attention was floor coverings having a higher weighting than boys clothing…done by a committee in Canberra with no growing male children I’d guess.

  3. Thanks Rumple, doing this comparison was on my list so its most appreciated 🙂

    Reason being, its been clear to me for a while that the GDP figures have not squared with the ‘vibe’ of the economy on the ground, and at the same time CPI figures have that same feel. It seems to me that GDP, which i think isnt useful at the best of times, has become less and less valuable over time.

    This kind of analysis offers at least some partial evidence for that ‘vibe’

  4. Great piece: it’s worth visiting macrobusiness.com.au just to read your posts!

    Isn’t the main difference between the two measures that the CPI is based on a fixed basket of goods (whose composition and weight is not often changed), while the GDP deflators measures changes in prices of goods that are actually consumed, so it’s much more dynamic?

    For instance, the price of bananas went up 6 times recently and that’s measured in the CPI, but consumption of bananas is way down, because consumers have decided to eat something else for now and supply is limited. So, in a sense, the increase in the price of bananas is not as significant, if it can be replaced with alternative goods.

    In the US the PCE is a very important measure, in Australia it does not seem to have the same importance in economic commentary.

    Q2-11 deflator in Australia was 0.4 Q/Q, if I am not mistaken, and that’s a very low number.

    • “Isn’t the main difference between the two measures that the CPI is based on a fixed basket of goods (whose composition and weight is not often changed), while the GDP deflators measures changes in prices of goods that are actually consumed, so it’s much more dynamic?”

      Yes. But the question is why these two measures travelled closely together until the mid-1990s then went their own separate ways. The CPI weights were revised in 1998, 2000 and 2005, so there has been opportunity to correct the discrepancy.

      You are right that the HCFE (the PCE equivalent) is not given any attention by the media – which is probably why no-one has yet cared to comment about the growing deviation between this measure and the official CPI.

      @sweeper, yes I have been a student of the CPI recently and i will have much more to say soon.

      • “Yes. But the question is why these two measures travelled closely together until the mid-1990s then went their own separate ways. The CPI weights were revised in 1998, 2000 and 2005, so there has been opportunity to correct the discrepancy.”

        Probably an answer could be found in the details of which of the components of the two measures went separate ways. The CPI does not reflect the fact that when the prices of certain items spike, due to supply constraints or other factors, consumers can opt for replacement items. Maybe consumer ability of opting for replacement items has increased recently and the consumer spending patterns were more “static” then. For instance, an example is traveling and accommodation: the CPI measures price of hotels and traveling in Australia, but consumers are opting to have holidays abroad (which got a lot cheaper recently) and that’s what the deflator should measure.

  5. Very interesting Cameron. Also when the IMF looks at CPI’s globally do they look at the different methods of calculation when producing their outlooks?

    • CPIs across countries are not directly comparable. The EU requires member states to produce a common CPI metric (the HICP), but each country produces other CPI measures for domestic use.


      The biggest variation between CPI methods across countries is the treatment of owner-occupied housing (since it is such a large portion of household expenditure). There are conceptual differences between the various methods used, but the international economic organisations do a reasonably good job of recognising these differences.

      eg. http://www.oecd.org/dataoecd/14/18/34987270.pdf

      There has been a global trend away from using the CPI as a cost-of-living index, and towards constructing a CPI as a pure price index to use as a guide for monetary policy.

      As I said last week, this difference in CPI construction since the 1990s is important to realise when comparing current economic conditions to past economic conditions.