Methodology

Godley’s explicit mission from the 1950s to the early 1980s was to inform UK government policy-making. As stated on the inside cover of the 1983 book on macroeconomics, the aim was ‘to contribute a basis for the development of practical solutions to the problems of stagnation, unemployment and inflation’ (Godley and Cripps 1983). After moving to the DAE, Godley decided to continue the approach he had followed at the Treasury, seeking to provide a clear understanding of the current situation and prospects of the UK economy and an assessment of the likely consequences of alternative courses of government action (Macfarlane 2011). Applied economics as practised by the CEPG in the 1970s was not a tool for testing economic theory.

The Group’s work on macroeconomic policy was integrated by a computerized database and modelling system that provided systematic checks on whether past behaviour of the economy in each area was consistent with assumptions about ‘how things work’ as programmed into the computer and computed alternative scenarios for the future based on assumptions about special factors, ongoing trends, and policy responses. Godley checked all assumptions, results, implications, and interpretations with different members of the group. Each round of analysis was written up in an issue of the Cambridge Economic Policy Review (CEPR), printed and published by the DAE, and later the Gower Press as a contribution to current UK policy debate.

The Group’s approach may be familiar to applied economists of the older generation working in official institutions but is hardly documented in academic literature, although commentators have made reference to Godley poring over data and his ability to locate surprising figures.^[1] The main elements of the 1970s CEPG methodology were the following:^[2]

(i) study of available data, definitional (accounting) constraints on what it can show, how it has been put together, and likely sources of error;
(ii) s pecification of behavioural norms based on institutional foundations and specialist knowledge in each field;
(iii) examination of the available data as a test of behavioural assumptions and as a means of identification of special factors and longer-run trends;^[3]
(iv) identification of the main causes (trends, external factors, or policies) of current predicaments;
(v) simulation of outcomes of different policy responses; and
(vi) presentation of policy lessons learned from historical investigation and model simulations emphasizing the key trends and relationships involved.

This approach differed radically from the more abstract and scholastic methodologies espoused by academic economists who in most cases work alone or with one or two colleagues without the support of research teams. There is an inevitable clash between the two approaches as specific observation of particular countries and historical moments often yields results that conflict with more general propositions advanced by theorists, and the evidence examined by applied economists is typically far more diverse in its nature, structure, and accuracy than assumed by academic econometricians.

The CEPG’s work with national accounts and other official statistics was not a data-mining exercise in which regressions are run to determine ‘correct’ values of coefficients. Time series statistics, particularly at a highly aggregated level, could not provide a robust basis for specification of behavioural relationships, not only because the length of historical series available in the 1970s was often quite short but also because of features such as multicol- linearity, measurement error, and limited effective degrees of variation which made estimated coefficients highly dependent on the precise way in which behavioural assumptions were specified.

Godley’s view was that elasticities and other structural coefficients in the model’s behavioural equations need to be broadly correct but cannot be precisely so and therefore that policy assessments need to be robust with respect to exact parameter values. Elasticities and time paths of behavioural responses were specified a priori on the basis of institutional considerations such as an understanding of the trade-offs involved and implications for longer-run stability, informed by more detailed, specialist empirical investigation. Structural coefficients could in some cases be inferred from stock-flow dynamics with the average lag of responses being determined by the stock-flow ratio (e.g. acquisition of financial assets, adjustment of inventories, and imports). Coefficients were also derived from historical studies. For example, investigation of wage bargaining and the movement of average earnings needed to consider evidence on many different aspects of labour cost, including wage rates, wage ‘drift’, hours worked (standard, short-time, and overtime), and so on, by industry and occupation. Surveys and censuses as well as time series provided the raw material from which behavioural conclusions could be drawn.

‘Known’ values of structural coefficients were used to adjust historical time series for cyclical effects and determine ‘underlying trends’. This procedure was formalized in the computerized version of the CEPG model which updated estimates of trends and the size and pattern of departures from normal behaviour implied by given structural relationships, a procedure that differed from the ‘scientific’ method which requires models to be constructed on the basis of abstract assumptions rooted in an academic theoretical literature with coefficients and lag patterns estimated from time series data by econometric methods.^[4]

Systematic re-estimation of structural coefficients at each forecasting round, practised by macro-modellers in the Klein tradition, was criticized by the CEPG as a less-than-transparent practice that blurs the distinction between model and data, potentially invalidating prior estimates and policy advice in an unclear manner and limiting lessons that might otherwise be learned from newly released data.

For Godley, each single observation was potentially valuable as a source of information. If the model and data did not agree, either may be wrong, and his first course of action was to solicit the advice of the statisticians who prepared the data. A good example was an abnormal level of imports of manufactures in the mid-1970s due to the arrival of rigs to drill North Sea oil. Kinks of this kind were usually familiar to government statisticians who had a good idea why they happened. Depending on the case, the explanation of an unexpected figure might be (i) identifiable and measurable ‘special factors’ known to the statisticians, (ii) estimation error (again known or surmised by the statisticians), or (iii) incomplete or faulty behavioural assumptions requiring a rethink of some aspects of the behavioural model. Ultimately, significant errors in the sense of discrepancies between the model’s conditional (singleequation) prediction and actual values reported in official statistics had to be explained by specific historical circumstances. It was not good enough to classify errors as the outcome of unknown random shocks since they have important policy consequences. Only when there are plausible explanations for any large errors can we retain confidence in the model with which we interpret the past and examine current issues and policy choices.

Godley never trusted models as such and checked the output of various CEPG models very carefully, his standard challenge being: ‘Are you prepared to bet either way?’ It was not enough to be logically consistent. Researchers had to provide an assurance that, after considering all the evidence and as best as they could determine, their estimates were not biased in any particular direction.

Write-ups of a policy analysis could not be content with reporting simulation results. The main need was to explain why different outcomes may arise and why different policies will or will not have much impact on the outcome. Such explanations should be understandable in their own terms without reference to detailed model simulations. Thus, although model simulations are suggestive, helping analysts to rule out developments that may at first have seemed plausible to them and to assess the relative importance of different factors, their findings must be communicated in terms that not only are logically consistent but also make sense historically and institutionally. This requirement places a premium on simplicity of model structure. The more the behavioural detail, the more complex and in many cases less certain the inferences that can be drawn.

Godley’s own preferred method for understanding the implications of a model was to focus on one or two key equations and simulate arithmetic examples with pencil and paper until he could internalize the implied dynamics and find a way to convey the main point to others.^[5] This approach and his emphasis on accounting constraints and the need to ‘close’ the model dated back to his time at the Treasury where national accounting identities provided a framework that he could use to check the consistency of forecasts of behavioural variables provided by official committees.

[1] Mata (2012: 15—16). Cameron and Llewellyn (2010) wrote that Godley ‘would take a vast spreadsheetof numbers, study them for sometimes hours at a time and then pronounce: “That figure is wrong,” stabbing at it with an elegant oboist’s finger. He was invariably found to be right. How did he know? Theexplanation, via his econometrician colleague Hashem Pesaran, was that he had what amounted to a fullmacroeconomic model in his head, which, by some sort of subconscious process, he computed’.
[2] For a more complete exposition, see Cripps and Fetherston (1979).
[3] As Kaldor often remarked, the evidence could contradict but not confirm. In this sense, the door wasalways open for improved hypotheses.
[4] Mata (2012). The relegation of time series econometrics to a subsidiary role caused the CEPG’s modelto be regarded with increasing mistrust by academic econometricians as well as a priori theorists.
[5] Later, he used computer software (MODLER) in the same manner.