UK GDP growth is projected to have remained sluggish in Q2 2017, growing by just 0.3%, according to the latest PwC UK Economic Outlook report. This follows the pattern of growth easing from 0.7% quarter-on-quarter in Q4 2016 to 0.2% in the first quarter of 2017.
PwC’s projections for Q2 growth are based on their new ‘nowcasting’ model, developed to combine expert human judgement and artificial intelligence (AI) machine learning techniques to produce more timely GDP estimates, ahead of those published by the ONS.
John Hawksworth, chief economist at PwC, said that, while there is no way to predict GDP growth with perfect accuracy, PwC’s new nowcasting model uses the latest AI machine learning techniques to augment human judgement and has performed well in tests using data for the past four years.
PwC’s nowcasting report comes just days after the firm’s latest Northern Ireland Economic Outlook (NIEO) said that Northern Ireland can expect economic growth of 1% in 2017, falling to around 0.9% in 2018.
Historical performance of PwC nowcasting model vs actual GDP growth
PwC economists have tested the performance of their nowcasting model over the past four years using the data that would have been available at the end of each quarter (almost a month in advance of the ONS preliminary GDP estimate). The results of this exercise are summarised in the chart below.
PwC’s nowcasting model would have been able to pick up changes in the direction of GDP growth correctly 94% of the time over the four year testing period with an average error of less than 0.2 percentage points. This improves on the commonly referenced Reuters Poll of forecasters and comes close to matching the accuracy of the ONS’s own preliminary estimate of real GDP growth (taken as a forecast of final real GDP growth), while being available several weeks earlier.
Notes to editors:
1. Nowcasting models are becoming an increasingly popular tool for policymakers globally and are predominantly used by Central Banks, such as the Bank of England, and sub-national monetary authorities such as The Federal Reserve Bank of Atlanta. They can also be applied more generally to leverage the benefits of ‘Big Data’ to help businesses understand the present state of their markets better.
2. PwC’s nowcasting model takes a broad set of frequently released indicators and uses a machine learning technique known as Elastic Net Regularisation and Variable Selection (‘Elastic Net’) to estimate the relationship between these different indicators and past GDP growth. Using this method (i) accounts for the co-movement between different variables over time; (ii) automatically selects variables in the model by learning which ones are most useful at predicting GDP; and (iii) predicts GDP more accurately by using a type of cross-validation (fitting the model over different sub-sets of the data). But expert human input is still required to frame the problem, evaluate alternative modelling approaches and interpret results.
3. As the services sector now accounts for almost 80% of UK GDP, a number of services-focused variables have been included in the consolidated set of variables that the model considers to nowcast GDP growth. These have also been complemented with other indicators to account for the industrial production and construction sectors, as well as “softer” indicators to capture expectations.
4. Full details of this research will be included in the next edition of PwC’s regular UK Economic Outlook report, which is due to be published on 18th July.