We don’t forecast anymore at Practical Economics. We learned the hard way that we were not very good at it.
Nevertheless, State and Territory Treasuries must undertake the unenviable task of reading the tea leaves.
Growth in real (constant price) Gross State Product (GSP) is a key measure of the health of a State or Territory’s economy. Yes we know it’s not the be all and end all of well being, but it’s up there.
A problem with time series forecasting of Australian State of Territory economies is that the Australian Bureau of Statistics (ABS) does not publish quarterly GSP.
The ABS publishes quarterly national Gross Domestic Product (GDP) from the September quarter 1959 onward.
The ABS publishes annual GSP (Catalogue Number 5220.0) around November each year, as well as quarterly State Final Demand (SFD, Cat. No. 5206.0) four times a year. However, while SFD makes up a major portion of GSP from the demand side, the two measures diverge.
Annual GSP from the demand side also includes the (in)famous balancing item. According to the ABS:
The difference between the sum of these components (SFD plus international trade) and GSP(E) is known as the balancing item. The balancing item reflects: changes in inventories; interstate trade in goods and services; and the balancing item discrepancy.
https://www.abs.gov.au/methodologies/australian-national-accounts-state-accounts-methodology/2019-20 . The ABS calculates GDP from the income, expenditure and production approaches.
Good luck estimating the balancing item quarterly!
There are 30 years of annual GSP published in Cat. No. 5220.0, but this isn’t enough to do any serious econometrics. Hence the need for a quarterly GSP series for each jurisdiction.
We haven’t looked for a while, but back in the day the ABS used to argue that it couldn’t produce quarterly GSP because it could not produce estimates that met its statistical standards.
Thankfully we here at Practical Economics have no such standards!
We make no claim that improved GSP data will improve forecasting, but it’s a start.
This article is an attempt to generate a quarterly GSP-like data series States and Territories using publicly available data.
To be clear, we do not claim that the data presented here are quarterly State and Territory GSP. They are a quarterly allocation of annual GSP data using publicly available quarterly data and methods commonly used by statistical agencies to generate quarterly GDP.
This is our hobby so our quality control is not where we’d like it to be yet. We rushed this out to line up with the release of the December quarter 2020 National Accounts and can’t be sure at this stage that there are no errors in our code.
Now that is out of the way, here are the data series we’re attempting to impersonate, spelled out in a way that might get this post noticed, if you know what we mean:
- New South Wales (NSW) Gross State Product (GSP);
- Victoria (Vic) Gross State Product (GSP);
- Queensland (Qld) Gross State Product (GSP);
- South Australia (SA) Gross Sate Product (GSP);
- Western Australia (WA) Gross State Product (GSP);
- Tasmania (Tas) Gross State Product (GSP);
- Northern Territory (NT) Gross State Product (GSP); and
- Australian Capital Territory (ACT) Gross State Product (GSP).
This analysis, done for both current prices and chain volume (constant prices). For a discussion of price adjustment see pp73-76 here) and for each State and Territory:
- Takes the quarterly national Gross Value Added (GVA) by industry found here and here as the quarterly industry indicators for each state (Indicator 1). That is, the industry indicator for each State and Territory is initially the same.
- Bench-marks the State and Territory industry indicators against annual Total Factor Income (TFI, in the ABS State Accounts, for example for NSW, Benchmark 1) for each State and Territory. The difference between GVA and TFI is ‘Other taxes and subsidies on production’.
- Sums the State quarterly series for each industry into a ‘synthetic’ national quarterly TFI series for each industry.
- Adjusts the synthetic series to sum to the national quarterly GVA.
- This now becomes the new quarterly indicator for each industry in each State and Territory (Indicator 1a). These indicators are now reinserted to Step 1, and the process is looped until the adjustment in Step 4 is unnecessary.
- Uses this bench-marked TFI/GVA as an indicator of quarterly GSP for each State and Territory (Indicator 2).
- Benchmarks this new indicator against annual GSP for each State and Territory (Benchmark 2) to gain quarterly GSP for each State and Territory.
- Sums the initial State and Territory quarterly GSP to a gain an initial national GSP series. This national GSP series is then adjusted so that it sums to the national quarterly GDP series. This adjustment is applied to the State and Territory GSP indicators uniformly (Indicator 2a).
- Loops the adjusted series to Step 7 with the aim of convergence where quarterly GSP is completely consistent with quarterly GDP and annual GSP.
- This leads to series for quarterly series for GSP for every State and Territory.
Steps 5 and 9 are not fully implemented at this stage because the COVID-19 shutdowns have caused some inconsistencies in the 2019-20 data.
To illustrate this, consider the difference between the sum (over States and Territories) of our State-quarterly bench-marked TFI (Step 2) and the national GVA figure for each industry (each line in the next two charts represents an industry).
Up until 2015-16 the difference isn’t that great (note this is bench-marking over the whole sample, with the results truncated), mostly less than $1 million per quarter on most occasions. Convergence might be possible.
And then 2019-20. What a year to attempt to publish the first installment of quarterly State and Territory GSP!
Can’t wait to see what 2020-21 has in store for the statisticians!
The main reason for this divergence seems to be a major switch in indirect taxes and subsidies (which is the major difference between GVA and TFI) from small and positive up prior to the December quarter 2020, but then large and negative afterwards as major government assistance packages were implemented.
Therefore in Steps 5 and 9 we restrict the re-balancing to 1 round and send the data forward consistent with the quarterly national GDP series.
The national quarterly industry GVA estimates are available in chain volume form from the mid-1970s onward, but current price estimates are only available from 2002-03. Consequently, the estimates presented here are from 2002-03.
Bench-marking here is done with the Cholette Method, whereas the ABS uses the Denton method. The Cholette method has an advantage because, unlike the Denton method, the first final quarterly estimate need not equal the first quarterly indicator observation (Cholette, p37).
The bench-marking procedure we use requires that the length of the quarterly indicator series exactly map to the length of the annual benchmark series. For the quarterly data here, there must be n*4 quarterly indicators, where n is the number of years of annual data.
We have 72 quarterly indicators between September 2002 and June 2020 and 18 annual benchmarks between 2002-03 and 2019-20.
At the time of writing we also have indicators for the September and December quarters 2020.
There is no annual benchmark for quarterly indicators in incomplete years. Consequently, we use the ratio of ratio of (Indicator 1)/(Benchmark 1) then (Indicator 2)/(Benchmark 2) for the last equivalent bench-marked quarter.
For example, for September 2020, we use the ratio of (Indicator 1a)/(Benchmark 1) and then (Indicator 2a)/(Benchmark 2) for the September quarter 2019.
We produce seasonally-adjusted and trend estimates of the Chain Volume State and Territory GSP series produced in Steps 1-10 above. We use the STL seasonal adjustment procedure, available in Python, whereas the ABS uses the X11 procedure (p102, to our knowledge not available in Python).
We haven’t given seasonal adjustment methods much thought but we aim to revisit this issue later.
Additionally, owing to the severity of the COVID-19 economic contraction the ABS has suspended trend estimates and implemented new measures for seasonal adjustment estimates. We’ve presented seasonally-adjusted and trend estimates, but please use at your own risk as these smoothing techniques are not designed to sensibly deal with such large changes.
Let’s look at the original levels series for the current price data. Current price data are most useful for considering the relative size of each State and Territory economy.
As you would expect, GSP increases over time in most cases, which is a combination of volumes and prices. Seasonal patterns are apparent.
More commonly, we often compare quarterly growth in chain volume seasonally-adjusted data as a measure of economic heath. Quarterly growth in State and Territory GSP from September 2002 to December 2020 is shown in the following panel of charts.
Obviously the larger States tend to mirror most closely the national quarterly pattern .
We’re not 100% sure what’s going on in Tasmania in the December quarter 2020, but seems to be because the national Agriculture Industry grew by 33% in seasonally-adjusted terms.
Were any of the States in ‘technical recession’, i.e. two quarters of GSP contraction, over our sample? We’re not sure this definition means anything as it was originally conceived for the large and relatively stable US economy, not small and volatile Australian State and Territories economies’.
Almost every State and Territory contracted due to COVID-19 restrictions in the March and June quarters of 2020 (the Northern Territory contracted in the March and September quarters). Queensland’s economy began contracting in the last half of 2019.
A full list of consecutive quarters of GSP contraction is shown in the Table below.
| State/Territory | ‘Recessions’ |
| NSW | Dec-08 to Jun-09, Dec-17 to Mar-18, Mar-20 to Jun-20 |
| VIC | Mar-09 to Jun-09, Mar-20 to Jun-20 |
| QLD | Mar-09 to Jun-09, Sep-18 to Dec-18, Sep-19 to Jun-20 |
| SA | Sep-05 to Dec-05, Mar-09 to Jun-09, Sep-12 to Dec-12, Sep-15 to Mar-16, Mar-20 to Jun-20 |
| WA | Dec-15 to Mar-16, Sep-18 to Dec-18, Mar-20 to Jun-20 |
| TAS | Mar-09 to Jun-09, Sep-12 to Dec-12, Mar-20 to Jun-20 |
| NT | Jun-07 to Dec-07, Mar-09 to Sep-09, Sep-18 to Dec-18, Jun-20 to Dec-20 |
| ACT | Sep-13 to Dec-13, Mar-20 to Jun-20 |
| AUS | Mar-09 to Jun-09, Mar-20 to Jun-20 |
The ‘GFC’ in 2008-09 is apparent (yes we can do a pre-COVID-19 trendy initialisation, remember how commonly ‘GFC’ was used by someone try to prove they were ‘in the know’?).
The September and December quarters of 2018 were bad for the mining-reliant States and Territories.
OK, now for the problems.
As the charts above indicate through the close movement of the State and national quarterly GSP series, the original quarterly GVA indicator data series is the same for every State and Territory (Step 1 above).
Obviously this isn’t an ideal situation, especially for the quarters for which we do not yet have annual State and Territory GSP benchmarks.
The best way would be to construct separate industry indicators for each industry in each State and Territory. We could also take advantage of the fact that Queensland publishes its own quarterly GSP.
It may also allow us to construct a longer times series than the one here from 2002-03 onward. For example, we might also be able to join it up to the ABS quarterly GSP series (Cat. No. 5242.0) that covered 1984-85 to 1996-97.
We are getting around to this. In fact we have notes on this from 2016, but our progress has been slow, actually non-existent. Potential data sources are scattered all over the place and are difficult to aggregate. We eventually decided to get the code working with less than ideal data.
We’ll continue to chip away at collating sources. Maybe next year?
Practically speaking, the national quarterly industry GVA indicator drags the differing growth rates amongst annual State and Territory GSP towards each other in the bench-marked quarterly GSP series.
This is shown in the table below, where there is perfect correlation between growth in the quarterly indicator data (GVA). It’s the same data after all. Annual GSP growth correlation varies between the States and Territories.
The growth in the Practical Economics quarterly GSP series is someone between the two, depending on the industry structure in each State and Territory.
Correlation Between Economic Growth (Chain Volume) in Australian States and Territories
| Annual GSP (Original, ABS) | Quarterly GSP (Seasonally-Adjusted, Practical Economics) | Quarterly Indicator (Sum of Industry GVA, Original, ABS) | |
| NSW | 0.56 | 0.95 | 1.00 |
| VIC | 0.67 | 0.95 | 1.00 |
| QLD | 0.89 | 0.96 | 1.00 |
| SA | 0.63 | 0.91 | 1.00 |
| WA | 0.58 | 0.80 | 1.00 |
| TAS | 0.58 | 0.84 | 1.00 |
| NT | 0.05 | 0.62 | 1.00 |
| ACT | 0.38 | 0.69 | 1.00 |
| AUS | 1.00 | 1.00 | 1.00 |
We’ve played a bit fast and loose with chain volume data. The best (pp20-21) way to construct a chain volume series is to apply price information to available current price data. However, we don’t have said price information for States and Territories, so we’ve bench-marked chain volume data on chain volume data.
Additionally, on a couple of occasions we’ve aggregated chain volume State and Territory data to form national aggregates, but there’s no reason why aggregated chain volume data will be the sum of component chain volume data, except in the base year for prices (p23).
Regardless, our aim was to make a start on constructing quarterly Australian State and Territory GSP. We’ll improve it as we have time.
Use of the code requires that you construct data files from the ABS source data.
We have a few things we want to further examine in the code, especially to wee whether we can get the quarterly national and State/Territory data to converge better than currently (Steps 5 and 9 in the list above).
If you have any comments, suggestions or want to purchase full results then please contact us. This isn’t our full-time job and we generate Practical Economics content when we can, so we might not reply immediately, but we will get back to you.
Good luck with your forecasting this year.