MONTHLY AND QUARTERLY GDP ESTIMATES FOR INTERWAR BRITAIN James - - PowerPoint PPT Presentation

National Institute

• f Economic and

Social Research

MONTHLY AND QUARTERLY GDP ESTIMATES FOR INTERWAR BRITAIN

James Mitchell (NIESR) Solomos Solomou (Cambridge) Martin Weale (BOE and NIESR)

Monthly GDP estimates

• Derive monthly series of UK GDP for the

inter-war period from a set of monthly indicators that were constructed at the time

• Illustrate how the new data can contribute to
• ur understanding of the economic history of

the UK in the 1930s

• Use the series to draw comparisons between

recession profiles in the 1930s and the post- war period

Monthly indicators of economic activity

• Work by Burns and Mitchell (1946)

anticipated by The Economist

• The Economist collected monthly indicators

during the period 1924-1938 and published an aggregate indicator computed as the geometric mean of the indicators

• These data form the basis of our monthly

GDP indicator, together with data for quarterly industrial production

Need for monthly GDP estimates

• GDP is the most appropriate indicator of overall

economic activity

• Preferable to relate these new (or lost) The

Economist data directly to GDP

• Avoids subjective inference based on simply eye-

balling these data - or averaging them

• Unlike Rhodes (1937, JRSS) and Stock & Watson

(2002, JBES) we reduce these monthly data to an estimator of monthly GDP itself

• This involves exploiting available annual GDP data

for the interwar period (1924-1938)

• Monthly GDP data facilitate historical analysis

Monthly indicators of economic activity

• Averaging the indicator variables is not the only method of

aggregation

• Rhodes (1937, JRSS) – anticipating Stock & Watson (2002,

JBES) - suggested instead use of the first principal component of the series

• But it is not necessarily the aggregate that is most closely

correlated with GDP growth

• Some means is needed of selecting from the indicators a

composite which is closely linked to GDP rather than one which is simply a summary of the indicator data set

• And this must be consistent with the annual GDP data
• We derive monthly GDP series exploiting these new

monthly data from The Economist and quarterly industrial production

Interpolating GDP using monthly indicators

• Chow-Lin (1971, ReStat). Dynamics ad hoc
• Dynamic generalisations – accommodate dynamics

flexibly via AR polynomial

• Model data in logarithms

– Implies a nonlinear temporal aggregation constraint

• Regression based methods

– Mitchell et al. (2005, EJ)

• State-space methods

– Harvey & Pierse (1984, JASA) – Multivariate extensions: Moauro & Savio (2005, EcsJ)

• Dynamic factor model. Useful when N gets big(ish)

Interpolating monthly GDP using a dynamic factor model

• Based on Proietti and Moauro (2006, JRSSC)
• Work in log-levels rather than growth rates as in

Stock and Watson (1991) – Mariano & Murasawa (2003, JAE) - and can handle mixed frequency data

• Assumes a latent factor, the business cycle, drives

(co)variation in the observed mixed-frequency data

– Consistent with Burns and Mitchell’s (1946) characterisation of the business cycle as common movements in different economic indicators – Stone (1947, JRSS) and Stock & Watson (1991)

• Provides exact solution to nonlinearity of

aggregation constraint

Interpolating monthly GDP using a dynamic factor model

• Model monthly N-vector in levels rather than

first differences

• The common trend and the idiosyncratic

components are modelled as difference stationary processes

• (i=1,…,N) is composed of drift , an

individual AR plus the business cycle (common) AR component

– This flexibility means the model fits the data

* , , , 2 , , , * * * , , ,

, 1,..., ; 1,...,12 ( ) , (0, ) ( ) , ( , )

t m t m t m t m t m t m t m t m t m

t T m L NID L NID

η

µ φ µ η η σ ⎫ = + = = ⎪ ⎪ Δ = ∼ ⎬ ⎪ Δ = + ∼ ⎪ ⎭

*

η

y θ μ D μ β η η 0 Σ

, t m

µ

* , t m

μ

, it m

y Δ

, t m

y

i

β

1 * ,

( )

i it m

d L η

− 1 ,

( )

t m

L φ η

−

Estimation with constraints

• Identification
• Estimation by ML exploiting Kalman filter
• Partition data vector into
• where y1t,m represents the observed monthly

indicators from The Economist and y2t,m and y3t,m are monthly IP and GDP which are latent and the

• bjects we wish to estimate
• We observe annual GDP data y3t such that
• Similarly define a constraint given quarterly IP data

– Impose constraints by defining a cumulator variable

, 1 , 2 , 3 ,

( ' , , )'

t m t m t m t m

y y = y y

12 3 3 , 1 t t m m

y y

=

=∑

The Economist’s monthly data

Monthly GDP 1924 - 1938

• We present data at both market prices and

factor cost, but focus attention on the market price data

• Model subjected to diagnostic tests
• We could not accept the restriction that the

idiosyncratic components shared a common AR coefficient

• The factor loadings are all positive and are

mostly significantly different from zero at 5%

– The Economist’s series are coincident indicators of economic activity – Employment, cotton, IP and GDP most sensitive to the business cycle

Monthly GDP at 1938 Market Prices and Factor Cost

250 300 350 400 450 500 1924-1 1924-7 1925-1 1925-7 1926-1 1926-7 1927-1 1927-7 1928-1 1928-7 1929-1 1929-7 1930-1 1930-7 1931-1 1931-7 1932-1 1932-7 1933-1 1933-7 1934-1 1934-7 1935-1 1935-7 1936-1 1936-7 1937-1 1937-7 1938-1 1938-7 £mn 1938 prices Market prices Factor cost

Historical analysis of the Great Depression

• Burns and Mitchell (1946) dated peak at July 1929
• Our GDP data suggest Jan 1930
• They dated the recovery from Aug 1932
• We see no sustained recovery until well into 1933
• These apparently minor differences can have substantial

implications

– Say if we wish to address the role of particular policies (e.g. devaluation in Sept 1931; monetary expansion in April 1932) in generating recovery – The role of policy change in generating expectational changes – Sargent (1983) and Temin (1989) – Previously absence of British data prevented analysis – Our data indicate a “hesitant recovery path” consistent with the view that the policymakers found it hard to generate a favourable expectational change with a single policy move

The Profiles of Five UK Depressions

• 9%
• 8%
• 7%
• 6%
• 5%
• 4%
• 3%
• 2%
• 1%

0% 1% 2% 6 12 18 24 30 36 42 48 Months from Start of Recession GDP : Change from Peak 1930-1934 1973-1976 1979-1983 1990-1993 2008-

The postwar monthly GDP data are from NIESR; see Mitchell et al. (2005, EJ)

Conclusion

• Rather than use disparate indicators in an ad hoc manner to

draw conclusions about the profiles of business cycles, it is preferable to use these variables to construct high- frequency estimates of GDP itself

• The high frequency GDP data we provide are, we hope, of

use to economists and economic historians addressing a number of questions

• The global financial crisis of 2008 has resulted in renewed

interest in the homologies between the current events and the Great Depression

• But to-date (Eichengreen and O’Rourke, 2009) use has only

been made of lower frequency GDP or IP data