Is there a better way of calculating a country's GDP?

03 June 2019

By Ana Galvão

"Shock 0.7% fall in UK GDP deepens double-dip recession,” screamed The Guardian font page in July 2012.

After the Great Recession of 2007 to 2009 the start of 2012 signalled another downturn in the UK economy and prompted economists and politicians to attack Chancellor George Osborne’s austerity programme, which had seen public spending slashed in an attempt to balance the country’s books, as the cause, with some even calling for him to be sacked.

Rachel Reeves, Labour’s shadow chief secretary to the Treasury, said the contraction was a "disastrous verdict” on austerity, while the Bank of England tore up its growth and inflation forecasts, with the Financial Times reporting it “planned to restart quantitative easing… to try to give the economy a lift”.

And yet a year later revised figures from the Office for National Statistic (ONS) found there was no recession at all. The panicked revising of policy by the Bank of England and the demands for an increase in Government spending were, in the end, not necessary.

It seems economists are not just uncertain about the future prospects of an economy, but also what is going on now with current growth as well.

It is because what we call ‘nowcasting’ is incredibly complicated and difficult. Calculating a nation’s Gross Domestic Product sounds simple enough - adding up the value of what is produced, with inputs taken out and weighting each according to its economic importance, while adjusting for inflation.

But when you delve into the actual mechanics of recording every company’s output every month - it was every three months until recently - it becomes fiendishly tortuous and can lead to severe misjudgements as happened in 2012. Outside statistical offices, economists observe the flow of data from surveys and PMIs (Purchasing Managers’ Indexes), but these may be affected by business sentiment, exacerbating the problem of finding out what is going on ‘now’.

The ONS makes it clear that the first GDP announcement is an estimate and will be revised, but it is the one that makes all the headlines and, as we saw in 2012, can cause huge ripples in the economy itself. The ONS, though, makes three revisions to the GDP figure, after eight and 13 weeks and then a year later, but it continues to be updated for another two years with the initial output method of calculating GDP then compared with the other two methods of totalling income and expenditure.

This is hard enough to do for a manufacturing-based economy producing endless, and easy to count, widgets. But in a service-based economy like the UK’s, which is essentially producing knowledge and information with houses becoming Airbnb hotels and any car turning into an Uber taxi, while millions download free apps like Facebook and watch hours of YouTube videos, it becomes doubly hard.

Thus, history is literally re-written as new data comes in and is compared to new calculations. And as we saw in 2012, revisions to the initial estimate can be substantial, sometimes leading to a complete reversal of assessments.

GDP is calculated by the ONS by surveying the output of every company in the UK, with the first estimate, which was published 25 days after the end of the referenced quarter until July 2018, using a representative sample amounting to 44 per cent of the required survey data.

How can nowcasting an economy be improved?

Hence, accounting for data uncertainty when nowcasting the UK economy is important and why my research partners Nikoleta Anesti and Silvia Miranda-Agrippino, of the Bank of England, and I have devised a potentially more accurate way of calculating an early GDP estimate.

We show in our research paper that it is best to enlarge the information set available to nowcast GDP growth and combine that with the pattern of revisions that have emerged.

We have built a model using 33 different economic indicators, covering the indexes of production and services, labour market indicators, macroeconomic aggregates such as consumption, investment and international trade, as well as surveys, and credit measures and financial variables. It features all the ‘market movers’ distributed by the likes of Bloomberg and Thomson Reuters.

Many nowcasting models use this data, but we have augmented it with the pattern of revisions that have built up since the UK started calculating GDP in 1955.

Subsequent releases relative to the same reference period can in fact be thought of as increasingly more accurate estimates of the same GDP figure. Initial rounds of revisions are typically due to the fact that as time goes by and more information is accumulated, the ONS can review its assessment of past events, so this pattern of revisions can be used to inform an estimate of the likely revisions needed for the first GDP estimate.

We call this model the Release-Augmented Dynamic Factor Model (RA-DFM) and it is as close to real-time GDP as you can get, with it being updated each time a new data set is published. We tested the model, which runs over many hundreds of lines of code, against historic GDP first estimates and subsequent revisions from 2006 to 2016 and found it to give a more accurate reflection of the ‘true’ GDP, being in line with future revisions.

Indeed, when looking back to 2012 our model does not show a double-dip recession. Instead, just as the latest revisions show, our RA-DFM model reveals positive growth rates for 2011 Q4 and 2012 Q1, with it only turning negative after the publication of the ONS’ preliminary estimate.

Related course: MSc Global Central Banking and Financial Regulation

The period illustrates how the RA-DFM can extract a reliable signal to measure GDP growth by filtering out the noise that may contaminate the early statistical releases, while providing a good assessment of the uncertainty around the first GDP estimate.

For the testing period, we compute 90 per cent confidence predictive intervals, which means we expect GDP values to be outside of the intervals 10 per cent of the time. This equates to about four times for our 10-year sample and yet this happens only three times in our study and two of them are during the 2009 recession when the subsequent downward revisions were quite unexpected. 

Further improvements and refinements will increase the reliability of our predictions, but we have shown that using the RA-DFM would help the UK avoid unwelcome and subsequently inaccurate headlines from first GDP estimates that cause an economic impact in themselves. And it might just help whoever is Chancellor to avoid unnecessary criticism and stay in their job a little longer.

Further reading:

Anesti, N, Galvão, A. B, & Miranda-Agrippino, S. (2018). Staff Working Paper No. 764 Uncertain Kingdom: nowcasting GDP and its revisions. Bank of England.

Carriero, A., Galvão, A. B. and Kapetanios, G. (2019). A comprehensive evaluation of macroeconomic forecasting methodsInternational Journal of Forecasting.

 

Ana Galvão is Professor of Economic Modelling and Forecasting and part of the Warwick Business School Forecasting System. She teaches Advanced Monetary Policy on the MSc Global Central Banking and Financial Regulation and Forecasting for Decision Makers on the MSc Management.

For more articles like this download Core magazine here.

Learn more about central banking and download a brochure for the MSc Global Central Banking & Financial Regulation.

Join the conversation

WBS on social media