Methodology of Historical National Accounting
Charles later said that his contribution was to provide the data, not to test hypotheses (Thomas 2002). This was not only a matter of taste or temperament. It was a coherent methodological position that was shared, and argued for, by his mentors and models among the fathers of national accounting, particularly in Cambridge, and which reflected the primacy they gave to observation over theory.
They did not reject mathematical logic or statistical rigour. What they doubted were the deductive models of partial and general equilibrium originating with Walras, Samuelson, and Arrow-Debreu, which implied that markets necessarily allocated resources efficiently, that market outcomes were always for the best, and that they rewarded everyone justly. Simon Kuznets, their doyen, thought that the task of theory was to specify the variables to be measured. Theory was not immutable, but merely identified a set of empirical regularities, which needed to be revised in the light of new knowledge, and of changing social values. It was risky to accept data without understanding how they came into being. More than one model could be fitted to any set of data, and a good statistical fit could not by itself guarantee correctness. Charles met Kuznets several times at Harvard. Kuznets’s method, like that followed by Charles, was essentially inductive: ‘[F]rom measurement to estimation to classification to explanation to speculation’, though Charles might hesitate before that final step (Lundberg 1971: 460; see also Fogel 2000).
A similar scepticism was expressed by other leaders of national accounting. Richard Stone chose a dialogue from Crochet Castle (1831) by Thomas Love Peacock for his epigraph in the first volume of the Cambridge interwar national income series. Mr Mac Quedy praises the modern political economy, ‘the science of sciences’. He is mocked by THE REV. DR FOLLIOTT. ‘A hyperbarbarous technology, that no Athenian ear could have borne. Premises assumed without evidence, or in spite of it; and conclusions drawn from them so logically, that they must necessarily be erroneous’ (Stone and Rowe 1954: xxiv). Wassily Leontief, the interwar inventor of input-output analysis (a technique that partly overlaps with national accounting and is now integral to it), made similar points in two iconoclastic articles (Leontief 1937, 1971).
Kuznets, Stone, and Leontief were not mindless empiricists. All three of them eventually won the Nobel Memorial Prize in Economic Sciences.[1] Kuznets originated seminal cyclical theories of economic activity and inequality. Leontief described himself as a mathematician and theorist. He later wrote, ‘When I developed input-output analysis it was as a response to the weaknesses of classical-neoclassical supply-and-demand analysis ... I felt that general equilibrium theory does not see how to integrate the facts ... My feeling is that the fundamental theoretical understanding of economic fluctuations is as a dynamic process’ [i.e. not an equilibrium one] (Leontief quoted in Foley 1998: 117-119). Stone’s consumer expenditure research stimulated the development of standard econometric methods for dealing with serial correlation (Cochrane-Orcutt and Durbin-Watson) and also applied a simple pioneering neoclassical model of the individual consumer.
Governments had taken up national accounting in the 1930s because of the market failures of depression and war: it provided the means to design interventions and to monitor them in pursuit of prosperity and full employment; the data were published and available to all. Observation and deduction both have a role in science and are needed to discipline each other. The sceptical position was scientifically sound: a preference for observation over speculation as a way to the truth, and a reality check on theoretical constructs. Experience has borne it out: like other ventures of the 1930s (radar, electronic computation, nuclear physics) what began as an academic problem ended up as an elaborate state technology within less than a decade. Like those other technologies, it is enduring and pervasive. It works unobtrusively in the service of government, business, and scholarship. Its deductive rivals (especially computable general equilibrium) are less modest in their ambitions, but (with their assumption of rational behaviour) resemble belief systems as much as fail-safe techniques (which is not to underestimate the power of belief, and indeed its necessity). National accounting made it easier, and therefore necessary, for governments to provide those services that only governments can provide, or those they provide more efficiently (education, health, social insurance, central banking, infrastructure), and thus to satisfy the aspirations of voters. Indeed, national accounts may have gone beyond measuring the preferences of voters and into shaping them, by setting up a target for affluence and helping to monitor it. This success in devising a working model of the economy is one reason why governments have become so large and indispensable, why even conservatives have to be social democrats now.
Charles published his National Income towards the end of the heroic phase of historical national accounting. What next? Simon Kuznets was co-ordinating a project to compare and explain economic performance in seven different countries. Robin Matthews had signed up Charles to join him in the British volume in the late 1960s (together with John Odling-Smee). Imminent publication was announced in 1972, but it required 10 more years. The study moved beyond national accounting, to examine the determinants and scale of growth rates in different sub-periods. It drew on ‘growth accounting’, developed by Edward Denison and John Kendrick in the 1950s. This was based on the Solow neoclassical model of the same decade, in which economic growth was seen to arise out of increments of labour, capital, and knowledge. The flow of output in growing economies rose faster than the flow of economic inputs (labour, capital, natural resources). This large unexplained positive bonus was referred to as the ‘residual’, or more technically as ‘total factor productivity’ (TFP), that is, that part of growth not accounted for by growth in the inputs of capital or labour.
TFP represented ‘any contribution that may arise from increasing returns to scale and from the effects of technical progress and advances in knowledge, of shifts in resources between sectors, and of changes in the extent of obstacles to more efficient use of resources (e.g. restrictive practices on the part of management or trade unions). It will also reflect any errors in the measurement of inputs and output, and in the specification of the relationship between them’ (ibid.: 15). The study was a painstaking and immensely detailed breakdown of growth and TFP by sector and sub-period. It had a bearing on the perennial issue of performance and putative decline, both in international comparison, and in different periods of time. But it had taken too long, and its Keynesian assumptions were out of tune with the times. Although it stimulated a great deal of further research, of the seven projected country volumes only two others were ever published (France and Japan).[2]
The mood had already shifted in economics, from the empirically minded, social democratic, Keynesian consensus of the 1950s and 1960s, towards the efficient market and rational expectations models of the 1970s, which were associated with Chicago economics and its market-liberal rejection of the state. A similar movement also arose within the discipline of economic history. In the USA and Britain, it was previously concerned with the development of industries, technologies, firms, and social institutions such as labour unions and government regulation. In the 1960s, a new approach to the past emerged in the USA which came to be known as the ‘new economic history’ or ‘cliometrics’. It premised that individual rationality and market equilibrium provided a good explanatory framework for the economic past. It typically postulated a causal mechanism suggested by deductive economic theory and sought to measure it by means of a statistical test of the explanatory power of each of a cluster of quantitative variables on the ‘dependent variable’ to be explained. Identifying the relative importance of labour and capital as independent sources of economic growth also followed this procedure, but did not really count, since for Charles and his colleagues it was primarily an empirical investigation with no particular theoretical agenda.
By the mid-1970s, cliometrics had achieved some dazzling counter-intuitive findings, most notably Robert Fogel’s study which scaled down the contribution of railways to economic growth, and his joint study of slavery with Stanley Engerman (Time on the Cross), which showed that slavery was profitable. More relevant to Charles were Donald McCloskey and Lars Sandberg on the rationality of entrepreneurs, and McCloskey’s landmark 1970 macroeconomic article, ‘Did Victorian Britain Fail?’. Previous writers on British economic stagnation suggested that it had. McCloskey’s approach implied that the notion of failure was meaningless in neoclassical terms. From that angle, no slack was possible— capital and labour were assumed to pursue economic advantage to the hilt, and underperformance could only arise from resource deficiencies, not human failings. But however original, acute, and stimulating, these studies had none of the finality of national accounting: they depended on speculative theory, and their findings have remained a matter of controversy.
Charles hosted the first British cliometric conference at Clare College and was a natural candidate for the cliometric movement, perhaps even to lead it in the UK. His jointly authored British Economic Growth of 1982 incorporates the Solow neoclassical production-function growth model as the base of its TFP estimation. But the volume (like Kuznets, its general editor) explicitly eschewed ‘sophisticated statistical or econometric methods’, on the grounds that ‘A less formal approach to the data reduces the risk that all the emphasis is placed on a single explanation, which may in reality be false’ (Matthews et al. 1982: 10). In the late 1970s, time series analysis, which was a key clio- metric tool, turned out to depend for its validity on an implicit assumption (quite often inappropriate) of a stationary (i.e. stable) relation among its independent and dependent variables. In the next decade, new tools (testing for stationarity) emerged to deal with this problem. In the meantime, however, this issue had silently invalidated a good deal of prior cliometric work. The methods of national income analysis were not seriously affected.
An obvious progression from national accounting was to model the economy as a market equilibrium (‘computable general equilibrium’). This was the economist’s equivalent to the historian’s ‘seamless web’, the assumption that everything affected everything else. Unlike the scepticism of conventional history, it also embodied the neoclassical assumptions that markets were for the best. The procedure was to take some subset of variables from the economic universe, to specify their interrelation by means of simultaneous equations, to insert empirical parameters from the historical evidence, and to solve for a market equilibrium. It required strong assumptions, namely the standard economic ones that outcomes represented the equilibrium of market transactions, and that the relevant market relations were adequately captured. If these premises were accepted, and if the model was unique, then computable general equilibrium made it possible to estimate the respective impacts of changes in prices, technology, or economic policies, and to simulate the effect of alternative, counter-factual scenarios, while taking feedback effects into account. In its top-down encompassing explanatory ambitions, this approach was the most far-reaching application of the ‘new economic history’. It was a theory-intensive high-tech alternative to the data-centred, bottom-up and largely descriptive approach of the national income accountants.
An early exponent of computable general equilibrium in economic history was Jeffrey G. Williamson, at Wisconsin and later at Harvard. His first study appeared in 1974, and in 1986 he published Did British Capitalism Breed Inequality? This tested a famous Kuznets hypothesis, namely that in the long-run course of economic growth, inequality would first increase, before it began to decline once again. Crucial industrial skills would grow scarce in the early stages of industrialization, increasing earning differentials. Economic growth would stimulate investment in these skills, and their abundance in later stages would reduce the skill premium and hence inequality overall.
Charles accepted the book for review with few prior expectations, and his article eventually occupied thirty pages in The Journal of Economic History (Feinstein 1988). He was not familiar with computable general equilibrium, but no one had a more intimate knowledge of British historical statistics. He criticized both the data and the model. It was not that the model was inappropriate (though the reviewer’s tone might sometimes suggest it), but that models could not deliver with poor data. Implicitly this was Leontief’s view again, that description came before theory and was not inferior to it. Partly it arose from a difference of analytical temperament, a preference for measurable reality over abstract modelling.
But it also arose (I think) from another tension which the protagonists were loath to probe or even to admit, namely between those who regarded numbers as the servants of history, and those for whom history would vindicate economic theory and the primacy of markets. It would not be far-fetched (though not entirely true either) to associate cliometrics with the market-liberal doctrines with which it coincided. Charles hoped that the dust-up with Williamson left few lingering resentments, and that it encouraged respect for data integrity. It is still a focal point for a methodological divide that remains charged. It confirmed Charles in his convictions. Theories came and went, good data endured.
Charles chose instead to estimate capital formation backwards to the eighteenth century. This was an essential preliminary for complete national accounts and also impinged on the core preoccupation of economic history: fixed capital formation (machinery, buildings, infrastructure), or aggregate saving, its close analogue, had long been seen as the main driver of economic growth, in theories of growth from Adam Smith to Karl Marx, and on to W.W. Rostow. Kuznets had always suspected otherwise, and a key volume by Kendrick and Pech in 1961 established that it was not so crucial, and that human creativity (captured statistically within ‘total factor productivity’) was the decisive factor (Kendrick and Pech 1961). Instalments of the capital formation series came out during the 1970s and 1980s, and in their final form as the second half of Studies in Capital Formation in the United Kingdom, 1750-1920, which Charles edited with Sydney Pollard in 1988 (Feinstein and Pollard 1988). Despite his expertise, Charles took no part in the theoretical debates convulsing Cambridge on whether capital was a coherent and tractable category. His job, he said, was merely to measure its historical cost. But the acrimony of the debate unsettled him.
