Estimating UK regional growth in Q2 2020


Estimating UK regional growth in Q2 2020

By Stuart McIntyre

UK GDP declined by a record 20.4% in the second quarter of 2020, following a fall of 2.2% in Q1. On a rolling four quarter basis (comparing the latest four quarters to the previous four quarters) the drop was 5.3%. These truly are unprecedented times.

There is a lot of interest in how the pandemic restrictions have affected different parts of the UK economically and socially.

For the past two years, we have been using an econometric model to allocate headline UK growth estimates across the regions and nations of the UK.

Given that these are unprecedented times, there are no historical data that our model can use to understand how an episode like this will play out at a regional level. This is the clear downside to using econometric models that rely on historical patterns and known uncertainties in past economic data at times of great uncertainty and structural change.

We have spent some time since the UK data were released exploring what our model can tell us about sub-UK growth in 2020 Q2, and we report some findings in this update. However, we emphasise the huge uncertainties that exist at this time.

Other sub-national estimates

So far this year, we have seen estimates of quarterly growth for 2020 published by the Northern Ireland Statistics and Research Agency (NISRA) as well as the Scottish Government.

NISRA estimate that output in Northern Ireland fell by 2.8% in Q1 2020 relative to Q4 2019 and fell by 3.2% relative to Q1 2019. On a rolling four quarter basis, comparable to our own estimates below, the economy declined by 0.44%.

The Scottish Government have in the pandemic period begun to produce monthly GDP estimates on an experimental basis as well as their usual ‘National Statistics’ publication of quarterly output.

The Scottish Government’s ‘National Statistics’ estimates run to Q1 2020 at the time of writing and suggest that output fell by 2.5% in Q1 relative to Q4 2019, fell by 2.6% relative to the same quarter in 2019, and on a rolling four quarter basis by 0.3%.

Meanwhile the experimental monthly estimates provide some insight on possible “growth” in Q2 2020, placing it at -19.7%. On a monthly basis, the Scottish Government estimates that GDP increased by 5.7% during June, following an increase of 2.3% in May and a fall of 19.2% in April.

Our latest estimates

Our headline estimates of growth for the regions and nations of the UK estimating our econometric model in the usual (or “original”) way are reported in column (2) of Table 1. As usual in this blog, we report the growth estimates as ‘growth in the year to’ a particular quarter, in other words on a rolling four quarter basis. In addition, in this blog, we present both the central (median) estimates and the 16th and 84th quantiles of the underlying density estimates to indicate some of the uncertainties present.

Table 1: Original estimates of growth in the year to 2020 Q2

Table 1 indicates considerable uncertainty about the regional growth estimates; and arguably, in some cases, the results are implausible. For example, regrettably, it seems unlikely given the uniformity of the shutdown across the UK that the economy in the North East has barely contracted in the aftermath of the pandemic.

So we caution the reader from putting too much stock in the estimates reported in Table 1. We emphasise that these results are based on estimating our econometric model in the “usual” way i.e. not making any additional adjustments for the disruptions and extreme observations caused by the COVID-19 induced economic shutdown of the UK economy.

As noted earlier, the fundamental problem in using econometric models during the current pandemic is the absence of any prior period of this kind in the data to inform the model. As set out in our last blog update, the key features of our model that drive its regional estimates are:

  1. estimated historical relationships between regional growth and UK growth (in essence: this reflects how sensitive regional growth is to UK growth);
  2. estimated historical relationships between the growth of particular regions (in essence: this captures how growth in region x has translated into growth in region y);
  3. estimated historical relationships within the regions (in essence: this captures the persistence of regional growth from one quarter to the next), and;
  4. estimated historical relationships between other macroeconomic variables (like the oil price) and regional growth.

In re-estimating our model in the “usual” way (effectively ignoring the outlying observations due to the pandemic), the extreme 2020Q2 UK data mean that the estimated relationships set out in 1-4 above have changed, in some cases quite substantially.

To explore this directly, we re-estimated our model including data only through to the end of 2019; in other words the relationships in points 1-4 above are estimated based only on experience up to the end of 2019 and hence are unaffected by the pandemic “shock”. We then use these estimated parameters but condition on UK data during the pandemic of 2020 to estimate regional output during 2020. We undertake this conditioning by imposing the important restriction that ensures our regional estimates aggregate to the ONS’s latest GDP growth estimates for the UK as a whole.

We have used this approach to conditional forecasting in some of our earlier regional nowcasting work. It is called ‘entropic tilting’. It delivers estimates of regional growth in the year to Q2 2020 conditional on UK Q1 and Q2 growth equalling the estimates produced by the ONS in the first half of 2020. In this way, we ensure our regional estimates during the pandemic period aggregate to the published UK data but the estimated relationships used to inform this regional allocation are not distorted by the extreme observations observed especially in 2020Q2.  

This provides our preferred nowcasts of regional growth in the year to 2020 Q2.

Table 2: Preferred estimates of growth in the year to 2020 Q2

Table 2 column (1) reports the results from entropically tilting our regional nowcasts to condition on the outturn of UK growth. Column (2) repeats the results from Column (2) of Table 1.

These preferred estimates of regional growth from our entropic tilting exercise (column 1) suggest that growth in the year to Q2 2020 was considerably more similar across regions than suggested by our headline model (which, remember, incorporates the latest UK data into estimation of the model parameters).

One conclusion from this exercise is that it’s the estimated model parameters that are driving the somewhat surprisingly varied regional nowcasts reported in Table 1. That is, the pandemic has led to considerable revisions in the estimated parameters of our original econometric model. Which of the two sets of estimates in Table 2 turns out to be better is an empirical matter that we will investigate in due course. But, as indicated, our money is on those in Column (1) of Table (2), given our belief that the extreme data from the pandemic period should not be treated similarly to pre-pandemic data.


Forecasting and nowcasting at the present time is fraught with many challenges. Econometric models struggle to accommodate the exceptionally severe downturn that the economy experienced in Q2 2020, given the complete absence of any comparable episode in the underlying data.

That said, as we have shown, economically sensible nowcasts (statistical accuracy is something to be judged after the event) can be produced. This can be achieved by combining our usual model, estimated on data not contaminated by the current pandemic, with a method of conditional forecasting known as entropic tilting that lets us exploit the recent but extreme GDP data for the UK as a whole.

Like many researchers, we will reflect on and grapple with the emerging economic data in the period ahead and seek to develop methods of extending our toolkit to capture this unprecedented economic episode and the extreme observations it is causing.

You can access all of our historic quarterly estimates here.

ESCoE blogs are published to further debate.  Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.

About the authors

Research Projects

Related publications