Keywords

4.1 Background and Method

Our approach in this volume will be to take the univariate approach and use GDP as a summary measure of activity for dating the business cycle and deriving key metrics under the different approaches. Other data will be brought into the picture in Chapter 6 when we undertake the narrative history of cycles over the past three and a half centuries.

As is well known, GDP can be measured using expenditure, income and output-based approaches. Estimates using each of the three approaches can be averaged to create “compromise” measures that implicitly place equal weight on all three estimates. Alternatively the measures can be balanced where judgement is placed on the reliability of each series, and the relative degree of reliability is then used to weight the series together. Balanced estimates can also be produced for particular benchmark years using input-output analysis where the researcher attempts to make the three estimates consistent using knowledge about the industrial structure and spending patterns in the economy for that year. Typically for the C19th such analysis is only feasible for years when a population census was carried out.

Various historical estimates for the UK exist using each of the three approaches. The main work on the UK in the C19th and first half of the C20th was carried out by Charles Feinstein in his 1972 volume National Income, Expenditure and Output of the United Kingdom 1855–1965. This monumental work, which is still the bedrock of UK estimates over this period, built on the earlier contributions of many scholars such as Bowley, Stamp, Clark, Deane, Stone, Jeffreys, Walters, Prest, Chapman, Hoffman, and Lewis among others. Work improving the estimates of industrial production and output during the industrial revolution was subsequently carried out by Crafts and Harley in the 1980s and restated in Crafts and Harley (1992). Martin Weale in a series of collaborations (Horrell et al. 1994; Sefton and Weale 1995; Solomou and Weale 1991) pioneered the work on balanced estimates in the UK and constructed an input-output table for the UK in the census year of 1841. Ongoing work by Mark Thomas, building on earlier collaboration with Charles Feinstein, is aiming for a similar input-output table for 1851. More recently the path-breaking work of Broadberry et al. (2015), who took on the task of reconstructing the growth of output and population in England and Great Britain between 1270 and 1870, now allows comparisons of GDP per capita to be made over seven centuries.

For many years little was known about the exact pattern of growth in Ireland especially during the C19th when Ireland experienced the Great Famine and the large overseas emigration that followed. That situation has now changed and major contributions have recently been made by Geary and Stark (2004, 2015, 2018) in a series of papers on regional growth in the UK, and Andersson and Lennard (2019) who have constructed annual estimates for Ireland for the 1842–1913 period.

The various historical estimates of GDP for the UK and Ireland that exist often cover different geographical areas and different time periods. The challenge is to try and construct continuous times series for the three geographically-consistent areas discussed above using the best available compromise and balanced estimates incorporating the information from each of the three approaches to measuring GDP. The Bank of England’s Millennium dataset (Thomas and Dimsdale 2017) attempts to construct GDP measures for various different geographical areas on a consistent basis. The challenge for this volume is to derive a measure that provides sufficient continuity that business cycles comparisons can be made over time but that also encapsulates a wide variety of information. Our approach is discussed below.

4.2 Sources and Construction—Annual Data 1660–2018

To evaluate real GDP involves deflating disaggregated expenditure or industry output and input components at current prices using suitable price indices and then weighting the resulting volume components together to create aggregate real indices of output or expenditure. A real measure from the income side can also be constructed by deflating nominal domestic incomes with the GDP deflator from the expenditure side of the account.

Typically for quantity indices a Laspeyres index is used, where current price shares in a base period form the weights and these are used to produce an index calculated over a set number of years. The base period and weights are then shifted periodically to capture any changes in the structure of the economy. The resulting chains of data are then linked together at an appropriate point as an index number to form a “chained volume measure” or CVM. This volume measure can be left in index number form. But it is also sometimes referenced to a particular year’s GDP in current prices so that the current price volume measure takes the same value in that year. This is then a chained volume measure expressed in terms of that particular reference year’s prices. In older versions of the official accounts the weights were updated around every 5 years. In historical analysis with limited current price data, volume indices are often based on linking fewer and longer chains of historical data where the periods used for weights can be many years apart. Current best practice is to chain-link annually using weights based on the previous year.

4.2.1 1948 Onwards

All UK estimates for the last seventy years are on the basis of post-1922 UK borders (Great Britain + Northern Ireland) and are derived from official Office for National Statistics (ONS) estimates of GDP. The volume estimates in the National Accounts are generally annually chain-weighted estimates where the volume index is built up from the components of GDP weighted appropriately by nominal shares of each component in overall GDP.

The preferred measure of output is gross valued added or GDP valued at basic prices rather than market prices. Basic prices exclude taxes and subsidies on products like VAT and duties. This is to ensure that the weights used to construct volume estimates of GDP are not distorted by indirect taxes. For example, the volume measure of GDP at market prices will, from an output perspective, tend to overstate the weight placed on consumption relative to other expenditure components given the relatively high share of indirect taxes applied to consumer spending. Prior to the introduction of the ESA95 system of accounts GDP volumes were generally measured at factor cost rather than basic prices. Factor cost estimates also exclude taxes on production such as business rates and motor vehicle excise duty paid by companies. But output prices at factor cost are not generally available because such taxes cannot be applied on a per unit basis. This mattered less when estimates of output by industry generally used direct volume indicators of production which were then weighted together by industry estimates of nominal value added at factor cost to generate an aggregate volume measure in factor cost terms. But the move to measuring real industry value added by deflating gross output and inputs in value terms with appropriate price deflators means that basic prices are the most practical basis for constructing GDP volumes.

ONS currently maintain a full set of national accounts based on the ESA2010 system of accounts from 1997. The data are available at a quarterly frequency and annual totals in current price terms are balanced over time through the supply-use framework, with a statistical discrepancy introduced for years which have not been balanced. Volume estimates on the expenditure and output sides of the accounts are annually chain-linked, although prior to 1990 this is only possible for the expenditure side.

Prior to the introduction of ESA95 in 1998 ONS had largely been able to maintain a full set of aggregate and sector income accounts back to 1948 for annual data, 1955 for aggregate quarterly data and 1963 for sector income and financial accounts. These accounts had also been balanced in current price terms using the supply-use framework back to 1986. Following the introduction of ESA95 in 1998 the start date of the full set of accounts was brought forward to 1987 given the cost and difficulty of applying the initial ESA95 changes retrospectively. This has now been brought even further forward to 1997 following successive implementation of subsequent ESA95 and ESA2010 changes. To mitigate the impact of this ONS has tried to maintain a set of core national accounts series back to 1948 and 1955 covering the main expenditure and income aggregates and key sector series. One particular problem after 1998 was that a number of corruptions and errors developed in some of the components of the historic (pre-ESA95) data which ONS continued to make available. These were documented by Martin (2009). This led the ONS to decide that only a set of core series would be maintained back to 1948 (see Everett [2011] and Denley [2016]). When there are methodological revisions to the accounts in the future the ONS remain committed to apply them retrospectively to this set of core series.

Together this means that although a vast amount of official data exist on the national accounts in the post-World War II (WWII) period, a comprehensive set of accounts based on a consistent methodology does not exist and this causes a number of problems for researchers who require a set of aggregate-level data that are also consistent with those across sectors and industries. The most glaring gap is that there is not a set of real industrial output data that are consistent with the core measure of real GDP based on the expenditure and deflated income measures. This means it is not possible to accurately decompose movements in aggregate output and productivity into the contributions of different industries prior to the 1990s.

For the purposes of this volume we use the core chained volume index for GDP at basic prices from 1948 to 2018 but recognising the limitation that we cannot disaggregate this accurately into the contribution from different industries.

4.2.2 1920–1948

The UK data for this period are based on the balanced estimates of GDP at factor cost by Sefton and Weale (1995). They used Feinstein’s (1972) estimates of GDP, which he was able to construct using all three approaches. Instead of constructing a simple average “compromise measure” they used a least-squares approach to weight the three estimates according to their underlying reliability. They showed that the compromise measure would only be the least-squares estimate if each series were considered equally reliable. Chart 4.1 shows the differences between the compromise measure and balanced measure of Sefton and Weale between 1920 and 1948. The differences are not particularly large at the aggregate level with the largest occurring during WWII but at a disaggregated level balancing expenditure and income produced more plausible estimates of the household saving ratio in the early 1920s, so these are to be preferred.

Chart 4.1
figure 1

(Source Thomas and Dimsdale [2017] based on Feinstein [1972] and Sefton and Weale [1995])

Balanced versus compromise measures of real GDP growth 1920–1948

Full size image

The current price estimates for 1948 by Sefton and Weale are very similar to current ONS estimates for GDP at current prices in the same year, despite Sefton and Weale’s estimates being based on the pre-ESA95 system of national accounts. The series are linked using the ratio of the two series in 1948. The constant price estimates at 1938 prices are linked to the ONS chained volume measures in the same way.

4.2.3 1870–1920

There are a number of estimates for Great Britain, Ireland and UK GDP available over the C19th and early C20th period.

  • Output-based measures. Feinstein (1972) constructed estimates between 1855 and 1920 for both Great Britain and the UK. Since then Broadberry et al. (2015) have constructed current price and volume measures for Great Britain between 1700 and 1870. Andersson and Lennard (2019) have similarly constructed estimates for Ireland between 1842 and 1913 benchmarked to the decadal estimates of Irish GDP implied by the regional shares of Geary and Stark (2015) based on sectoral wage data.

  • Expenditure-based estimates. Feinstein constructed expenditure-based estimates of UK GDP from 1870 in his (1972) volume. Subsequently he pushed these back to 1830 using earlier work by Deane (1968), and his own work on investment and trade in Feinstein and Pollard (1988). These improved estimates were published in Mitchell (1988). Estimates are available on a constant and current price basis.

  • Income-based estimates. Feinstein (1972) constructed current price income estimates back to 1855. However he also subsequently made several improvements to the components on the income side, especially to the estimates of wage earners’ incomes (Feinstein 1990). Boyer and Hatton (2002) also improved the estimates of unemployment over the 1870–1913 period. These estimates are in the process of being brought together in an updated income measure by Solomou and Thomas (2019) but these have yet to be peer reviewed and accepted.

  • Compromise and balanced estimates. Feinstein himself provided compromise estimates based on an average of the three estimates he constructed in his (1972) volume. These were updated to include some revisions on the expenditure side arising from the work on gross capital formation and net trade in Feinstein and Pollard (1988). This “new” compromise estimate appeared in Mitchell (1988). Solomou and Weale (1991) provide balanced estimates based on Feinstein’s updated data for the 1870–1913 period using Feinstein’s subjective assessment about the reliability of each of the series. Solomou and Thomas (2019) have provided provisional estimates of a compromise measure based on Feinstein’s susbsequent revisions to the income measure.

Our approach in this volume, given the state of current knowledge and until more recent research can be brought together and synthesised, is to use the Solomou and Weale (1991) balanced estimates as the best current estimate for the 1870–1913 period, extended by Feinstein’s compromise measure to 1920. We recognise however this estimate is likely to be revised by ongoing work on historical UK accounts discussed above. Chart 4.2 shows how Solomou and Weale’s balanced estimates for the 1870 period compare with Feinstein’s earlier compromise estimates.

Chart 4.2
figure 2

(Source Thomas and Dimsdale [2017] based on Feinstein [1972], Mitchell [1988], and Solomou and Weale [1991])

Solomou and Weale’s (1991) balanced estimates

Full size image

4.2.4 1660–1870

As discussed in the previous section, for the period before 1870 it is now possible to construct measures for the UK as a whole back to 1830 using expenditure data and 1841 based on the income approach and the output-based data for Great Britain and Ireland. The input-output table of Horrell et al. (1994) for 1841 also provides a benchmark estimate that reconciles the measures from the different approaches.

However, there are many issues to resolve before such a UK-wide measure combining the three approaches can be used reliably for business cycle analysis. These are discussed in Solomou and Thomas (2019). We also know the period between 1841 and 1870 covers the period of the Irish famine and subsequent diaspora which reduced the population of Ireland by around a third and is likely to have significant and distinct effects on the pattern of growth in Ireland relative to that in Britain.

So for these reasons we use Broadberry et al.’s (2015) output-based estimates for British GDP from 1870 back to 1700, and extended this further backwards in time to 1660 using their estimates of English GDP. These represent estimates for British growth that from 1700 have been scrutinised by many scholars. In particular, they encapsulate the widely accepted “Crafts-Harley” viewFootnote 1 of growth during the industrial revolution that suggested only a modest improvement in growth unlike the earlier views of Hoffmann (1955) and Deane and Cole (1962). These estimates have also been used to analyse the shifts in the underlying trend rate of growth of GDP between 1270 and 1870 (Crafts and Mills 2017). So for comparability it is cleaner to use these same estimates for cyclical analysis. UK estimates that incorporate the information from the other measures and include the latest Irish GDP estimates can be found in Thomas and Dimsdale (2017).

When we undertake the narrative analysis in Chapter 6 we will however use some of the components of the expenditure side estimates to help explain what is driving the causes of the cycle. Chart 4.3 shows that movements in the UK expenditure-based measure over the period 1830–1870 do correspond broadly to the fluctuations in British output. The main differences occur in the early 1860s and this needs to be borne in mind when using the expenditure components to explain cycles over this period.

Chart 4.3
figure 3

(Source Thomas and Dimsdale [2017] based on Deane [1968] and Broadberry et al. [2015])

Expenditure and output measures compared

Full size image

4.3 Quarterly and Monthly GDP Estimates

Official estimates of quarterly GDP begin in 1955. It is possible to construct monthly estimates of GDP by using other partial indicators such as industrial production which can be used to interpolate the quarterly GDP series that go back to 1955 or the annual series stretching back to 1920 and earlier. The National Institute of Economic and Social Research (NIESR) currently constructs a monthly series for GDP which currently goes back to 1985 and they have also produced an earlier series that goes back to 1973.Footnote 2

Mitchell et al. (2012) carried out a similar exercise using quarterly industrial production and a set of monthly indicators from the Economist for the interwar period (1920–1938) benchmarking to Sefton and Weale’s (1995) balanced GDP data. This enables both monthly and quarterly GDP estimates for the interwar period from 1920 to 1938. In principle, this method could be extended to create a complete monthly series from 1913 to the present based on a comprehensive set of indicators.

The world war periods provide particular issues in providing high frequency estimates. For WWI (1972) Feinstein was able to make estimates on the expenditure and income side although no split between self-employed income and profits was possible. No indices of real output by industry were produced. For WWII estimates of GDP rely on those published in the Statistical Digest of the War which was republished with accompanying explanation by Howlett (1995). Again no GDP(O) and industry output estimates are available.

It is possible to interpolate the estimates of GDP with higher frequency information on the output of certain sectors during each war period. For example a quarterly munitions index was produced during WWII and was improved by Harrison (1990). This could be used to interpolate that part of national income and expenditure attributable to wartime procurement of goods and services on the expenditure side of the accounts. Monthly retail sales data also exist over the war period. The accumulation of sufficient indicators would allow the creation of a quarterly and possibly monthly GDP series back to WWI . Preliminary quarterly estimates for 1938–1955 based on a range of indicators are available in Chadha et al. (2019).

For the purposes of this volume we analyse the quarterly data of Mitchell et al. (2012) for 1920–1938 and ONS data for 1955–2018 pending future work on high frequency data for the war periods and their immediate aftermath.

4.4 Summary of GDP Data

For the data in the next chapter we now summarise the GDP data used. The different annual components are combined into an index with a base of 1948. The data represent growth rates of English GDP from 1660 to1700, Great Britain from 1700 to 1870, the UK including all of Ireland, 1870–1920 and the UK including Northern Ireland only from 1920. As discussed, the volume series for GDP is on a factor cost or basic price basis. The overall index of real GDP that we use is summarised in Table 4.1.

Table 4.1 Summary of real GDP used for cyclical analysis
Full size table

Taking the data at face value, the volatility of economic growth appears to have changed over time.

During the late C17th and early C18th GDP growth appears to have been particularly volatile (Chart 4.4). As discussed later, this may reflect the use of probate data in measuring agricultural output rather than the more reliable data from farm accounts. Growth became less volatile during the C18th and early C19th. During the mid-to-late C19th, the average growth rate of the economy picked up to around 2% per annum and there was less volatility in output (Table 4.2). But the volatility returned in the interwar period, during which there were two major recessions. The post-WWII period ushered in a golden age of prolonged periods of positive and relatively stable (annual) GDP growth, such as in the late 1950s/early 1960s, and between the early 1990s and the onset of the recent financial crisis. However growth in the ten years since the start of the financial crisis has been half of that observed in the previous two centuries. These estimates are used in the next section to derive a set of business cycle metrics for the UK since 1660.

Chart 4.4
figure 4

(Source See Table 4.1)

Annual GDP growth 1660–2018

Full size image
Table 4.2 Summary statistics on annual GDP growth
Full size table