‘Let me tell you a story’: the politics of macroeconomic models

ABSTRACT

Various social science literatures suggest that the general character of macroeconomic models reflects their assumptive base. A more specialist literature in the Weintraub-Boumans-Morgan tradition shows how the mathematics of those models moulds together logical implications of particular starting assumptions, insights from generally accepted theoretical propositions, and professionalised common-sense about how the world works. I go one step further in arguing that a process of narrative moulding operates in tandem with this mathematical moulding. A naming strategy provides the mathematical properties of macroeconomic models with economic labels to create the feeling that they are something more than a merely mathematical structure. A storytelling strategy then informs policy-makers of where the solution to the system of equations positions the outer limits of both political desirability and political possibility. Future dedicated research programmes into the narrative dimensions of macroeconomic models can be expected to shed further light on how theory models can masquerade as policy models and substitute models as surrogate models.

KEYWORDS:

• Macroeconomic models
• mathematical moulding
• narrative moulding
• theory vs policy models
• surrogate vs substitute models

Introduction

In January 2017, Olivier Blanchard presented a challenge to macroeconomists that they are still to answer effectively. As the still recently retired Chief Economist of the International Monetary Fund, someone who had always had one foot in each of the theory and policy camps, he used a series of blog posts for the Peterson Institute for International Economics to argue that the two should now go their own way. Today’s dominant dynamic stochastic general equilibrium (DSGE) models, he wrote, retain relevance for purely abstract research, but provide little practical purchase beyond the hypothetical scenarios of their own self-made worlds. DSGE modellers were generally unsure how to respond. The fact that their models had successfully crossed over from the academic literature to central bank research departments had led them to assume that they offered bespoke policy advice.

In this paper, I ask how such assumptions have remained largely unchallenged for so long. After all, there has been widespread agreement for some time in the philosophy of science literature that macroeconomic models are not accurate representations of real economies. Moreover, very few economists would claim that the relations of which they write refer to actually existing conditions. Nonetheless, most DSGE modellers continue to draw inferences from their mathematical creations to the decisions actual policy-makers must make. To understand how this situation is maintained, I argue that macroeconomic models comprise of more than the system of equations defining their internal logic and the economic names allowing them to pass as something more than mathematical solutions to mathematical problems. They also have underlying plotlines that are intrinsic to the models themselves. This narrative content instructs the reader what the hypothetical policy-maker’s priorities should be and how actual policy-makers might be expected to enact parallel visions of the future.

Economists, economic philosophers and historians of economic thought have all displayed considerable interest in how macroeconomists make their arguments. Yet even if they have operated on a similar general terrain, they have only spoken around the specific issue that captures my attention. Milton Friedman’s (1953) account of economists’ methodological instrumentalism, for instance, suggests that whatever significance is contained within a model’s narrative content is external to the operation of the model itself. It is to be found in whether the model’s underlying theory produces sound predictions for policy-makers. Likewise with Philip Mirowski’s (1991) account of economists’ methodological naturalism, only this time the important factor is whether the model’s underlying theory mimics well-known scientific explanations from other fields. Perhaps it is unsurprising, though, that both Friedman and Mirowski externalise the narrative content of economic models, rendering it distinct from how models work in their own terms, because their primary concern is with the character of the ensuing argument and not with the specific operation of the models through which the argument is pursued.

Appreciation of narrative content as a constitutive aspect internal to economic models only really started when Marcel Boumans (1999) introduced the notion of mathematical moulding into the methodological literature. Roy Weintraub (1985) had already shown that economists told theoretical stories directly with their mathematics, but Boumans went further with his suggestion that the mathematics should itself be seen as a narrative instrument that linked theoretical assertions to commonly held understandings of wider macroeconomic realities. Mary Morgan (2012) went further still in insisting that the art of persuasion requires any such narrative instrument to work in prose form in addition to working as a study in formal mathematical logic. My paper is consciously positioned within the Weintraub-Boumans-Morgan lineage, but it adds the extra claim that the model’s internal plotlines must be understood as an encoded message about the most desirable state of the world beyond the model.

This evidently political manoeuvre becomes a crucial part of a macroeconomic model because it blurs the boundaries between what might act, in Blanchard’s terms, as theory models and policy models. Friedman’s account of methodological instrumentalism proceeds as if everything is ultimately a policy model, Mirowski’s of methodological naturalism as if everything is ultimately a theory model. Something beyond both is needed when theory models take on the appearance of policy models because their narrative content gives every impression of speaking directly to real-world economic phenomena. Theory models serve foremost as largely self-referential thought experiments, but the storylines in which they are embedded will usually be designed to persuade the reader that a real-world puzzle inspires the analytical endeavour. Operating within an explicitly political account of the Weintraub-Boumans-Morgan tradition makes it possible to understand how purely theoretical macroeconomic models can seem to speak directly to real-world problems, even if strictly speaking they do not. Continued political demands for particular stories might thereby help macroeconomic models survive even conspicuous real-world failures of application.

Two aspects of the narrative content are particularly important. The first impacts upon what the model can say about the world beyond its own mathematical structure via a naming strategy. The variables in the system of equations are given economic names, so even though its solutions are, strictly speaking, mathematical solutions to mathematical problems, they can be communicated to the reader through the standard macroeconomic language of ‘inflation’, ‘unemployment’, ‘output’ and ‘the price level’. In the case of DSGE models, this becomes more specifically the macroeconomic language of rationally induced output gaps, the expected dynamic path of aggregate prices, and a policy rule that gives the private sector complete clarity concerning the future price of money (Clarida et al. 1999, p. 1670). The second narrative element then comes into play, a storytelling strategy suggesting that the model under consideration is recognisable as a reflection of the real world. The story has the potential to collapse the cognitive space between the economically-named model components and their real-world equivalents, with the first-best solution drawn from the world within the model also being treated as the outcome actual policy-makers should be seeking to engineer. In this way, purely mathematical structures are brought to life as macroeconomic models, with the narrative content naturalising some policy options while foreclosing others.

This is a general argument that can underpin the study of any macroeconomic model that has historically had a path-shaping effect on economists’ professional consciousness (such as Tinbergen’s earliest national income accounting model, the Klein-Goldberger model of dynamic Keynesian theory, Phelps’s model of the vertical long-run Phillips curve, Dornbusch’s overshooting model, Kydland and Prescott’s time inconsistency model, etc.). However, space constraints require I focus substantively only on one model-type, today’s dominant DSGE models. The section ‘Economists’ concerns regarding DSGE modelling’ reviews the debate amongst economists about the future of DSGE models post-global financial crisis. Proponents tend to emphasise their success as theory models, sceptics their failure as policy models. The section ‘Philosophical insights into DSGE modelling’ shows that the source of the disagreement runs much deeper than surface appearances, all the way to the philosophical question of how purely mathematical structures can provide meaningful economic knowledge. In particular, theory and policy models have different criteria for what counts as a credible model. The section ‘The politics in the macromodelling process’ offers two illustrations of how the narrative content of macroeconomic models muddies the distinction between these criteria. They demonstrate that the choice of mathematical structure is inseparable from accompanying political preferences, even if those preferences are never rendered explicit. The storyboards are examined for two important forerunners of today’s dominant DSGE models: Robert Lucas’s (1972) early real business cycle theory and Thomas Sargent and Neil Wallace’s (1975) policy ineffectiveness proposition.

Economists’ concerns regarding DSGE modelling

‘If you have an interesting and a coherent story to tell’, Varadarajan Chari (2010, p. 32) informed a potentially hostile congressional hearing investigating alleged failures of macroeconomics during the global financial crisis, ‘you can do so within a DSGE model’. He left unsaid what such stories relate to and who their audience might be, but at least it was an acknowledgement that macroeconomic models only take the form they do because of the plotlines which invigorate their otherwise inanimate mathematics. The narrative content of DSGE models has a strong record facilitating wholly abstract thought experiments, with Blanchard (2017) suggesting that the primary purpose of theory models is that they ‘be used to think’. The pioneers of real business cycle theory enlisted rudimentary general equilibrium models to overturn prevailing macroeconomic theory (De Vroey 2016, p. 162). There was nothing within prior theoretical accounts to suggest that hypothetical policy-makers would not always be able to dampen market-based changes in demand. The revolutionary nature of real business cycle theory was that its animating narratives were used to argue, theoretically at least, that a very different model world was thinkable (Helgadóttir 2023, p. 253).

DSGE models have fulfilled another important function in making macroeconomic theory more accessible. Where once only the select few got to have their equations incorporated into the dominant model, now anyone with the requisite mathematical skills can introduce suitable theoretical qualifications (Helgadóttir 2022, p. 434). DSGE models have also allowed macroeconomists to understand a great deal more about how theory models with deep structural parameters perform compared to those that place more emphasis on tracking observed data (Inoue and Rossi 2011, p. 1195). They thus provide important insights into how anxious researchers should be about creating models that survive the Lucas critique (Hurtado 2014, p. 13). These are definite advances and justify the past half-century’s efforts in laying the foundations for the contemporary state of macroeconomic knowledge. However, they are all theoretical achievements, and none on its own means that theoretically-derived statements about the model world can easily be transposed into straightforward claims about policy. Boundary issues remain about how a theory model might be presented convincingly as the source of policy advice.

Blanchard (2017) states that the primary requirement of policy models is that they ‘should fit the main characteristics of the data’. Recognition that early real business cycle models fell short of such a standard prompted significant theoretical overhauls, with dynamics of perfect price adjustment being replaced by a New Keynesian emphasis on price stickiness (Goodfriend and King 1997, p. 232). This is the source of the modern synthesis on which most contemporary DSGE models are based. In themselves, though, these theoretical tweaks are insufficient to ensure that the policy stories told by DSGE modellers go beyond being theoretically thinkable to act as a plausible representation of the world beyond the model. Recent claims regarding a crisis of macroeconomics have focused on DSGE models’ inability to track the data during the meltdown of money markets in 2007–8, relegating the ensuing crisis to such an outlier position on a standard bell curve as to become a supposedly impossible historical event. Former Federal Reserve Chair Alan Greenspan (2013, p. 95) described the tails of the probability distribution in evidence at the time as not just fat but ‘morbidly obese’. Policy-makers were thus left to rely on older models to inform their emergency interventions (Krugman 2018, p. 156). Despite being less theoretically rigorous, they did at least explain how the crisis had become possible and what might be done to mitigate real-economy contagion from haemorrhaging asset prices.

Maybe it is only to be expected, though, that theory models underperform as policy models. Theory models are limited to enabling their users to understand how model-world dynamics must respond to changing variable values if they are to ensure that the system of equations continues to produce an equilibrium solution. By contrast, policy models enable their users to understand how real-world dynamics might respond to changes in actual fiscal and monetary policies, given what is already known about historical data. These are very different objectives that bear no necessary relation to one another. Yet as Francesco Sergi (2020, p. 170) has demonstrated, the global financial crisis did little to stop central banks using DSGE models for policy analysis. In the following decade, many countries adopted new policy models based on DSGE theory, including three-fifths of the G20 and all the G7. Some were first-time adopters, more published updates on existing policy models; some were used in isolation, more in conjunction with other types of policy model (Storm 2021, p. 78). But each is of a common form shared with their purely theoretical counterparts, making it increasingly difficult to distinguish policy models from theory models, despite their different objectives.

DSGE models are typically founded on a three-equation system: one describes the goods market equilibrium; one captures the evolution of aggregate prices; one refers to the decisions available to policy-makers (Walsh 2010, pp. 330–5). Each equation is presented formally, so there is complete mathematical clarity regarding permissible states in the model world. Boumans’s mathematical moulding allows macroeconomists to incorporate learned professional intuitions into their modelling endeavours, so that what emerges looks familiar to other practitioners. The first equation reproduces basic insights about the Hicks-Hansen IS curve; the second likewise for the Phillips curve relationship; the third produces a policy rule based on pre-commitment to price stability (McCallum and Nelson 1999, pp. 300–6). Alongside this mathematical moulding, I suggest, there exists an equally important process of narrative moulding. The storytelling strategy of the standard DSGE model relies heavily on that of the real business cycle theory on which it is founded, with the shared assumption of rational expectations placing policy-makers in a highly restricted environment. Agents’ demands for time-consistent counter-inflationary policies require central bankers to signal acceptance of social welfare losses the public finds suboptimal. The naming strategy for the mathematical components of the standard model service this particular plotline (see Clarida et al. 1999, pp. 1664–8). Equation 1 for goods market equilibrium is based on a specific mathematical definition of an inverse relationship between the output gap and the real interest rate, but where hypothetical agents have the foresight to always correctly second-guess policy-makers’ intentions. Equation 2 for the dynamic path of aggregate prices is based on an equally specific mathematical definition of a direct relationship between inflation and the output gap, again where rational agents know that path in advance. Because everyone is aware that everyone else knows the future that will unfold under various policy measures, equation 3 for the policy rule defines strict signalling of counter-inflationary intent as the optimal solution to the system of equations. More than a quarter of a century has now elapsed since Michael Woodford (1998, pp. 120–6) provided definitive mathematical derivations of these conditions from first principles of individual maximisation under rational expectations.

Is this, though, a theory model, a policy model, or a theory model masquerading as a policy model? Might the model’s proponents even answer differently depending on whether it has come under sustained scrutiny? Paul Pfleiderer argues that this is a common attribute of DSGE models, and he consequently calls them chameleons. He is concerned that macroeconomic models might be presented as providing clear-cut policy rationales for actual central bankers, even though they were formulated using assumptions that have no obvious real-world equivalents. A category confusion is thus implied regarding what can be said about relationships that only exist in the model world and those that exist beyond the model. But often, says Pfleiderer (2020, p. 84), such confusion is conveniently ignored until a model’s practical failures are highlighted: ‘A chameleon model is put forth as “saying something about the real world”, but when criticised is just a “building block” model’. Robert Sugden (2009, p. 4) makes a similar point when suggesting that his fellow economists almost always give readers reasons to expect that they will be learning something of real-world policy relevance, but then the real world is not mentioned once when the construction of the model is explained. Policy recommendations remain, despite the model’s unrealistic assumptions demonstrating that it is only a theory model. It is macroeconomic models’ narrative content that allows their users to act as if they can change character in such a way. They tell stories as if they are policy models for as long as nobody insists they should be scrutinised using real-world filters, but they are treated as theory models thereafter.

DSGE models’ inbuilt naming and storytelling strategies typically mask the point at which a theory model stops and a policy model starts. The narrative content creates the impression that the dynamics being studied belong to the world of everyday experience and not merely to the self-made model world. Consider Robert Lucas and Thomas Sargent’s 1979 paper, ‘After Keynesian Macroeconomics’, which acts as a crucial staging post in the development of today’s dominant DSGE models. Its title encapsulates its revolutionary spirit (De Vroey 2016, p. 151), and its storytelling strategy emphasises the failures of policy-makers to conquer inflation (Wren-Lewis 2016, pp. 23–4). But the ensuing model makes no use of then-recent data related to the inflationary episode of the 1970s. Moreover, its theoretical critique of existing policy-making practice sidesteps the inflation question almost completely, focusing instead on methodological claims about the need for microfounded identification restrictions. Countless economists have stressed the need to determine on an individual basis where the limits lie in what any macroeconomic model can reveal about the world beyond itself. Here, though, Lucas and Sargent’s (1979) storytelling recasts a theory model as something easily mistaken for a policy model. Following Weintraub (1985), they clearly had an important theoretical story to tell with their mathematics; following Boumans (1999), the mathematics was moulded with instinctive knowledge macroeconomists would have held at the time about the way the world works; following Morgan (2012), the storyboard embedded within the model contains prose features that bring it to life as a recognisably macroeconomic model. But to take my next step towards understanding the political content of Pfleiderer’s chameleons, it is first necessary to examine how such plotlines enable the elision of very different sets of philosophical claims.

Philosophical insights into DSGE modelling

As the previous section has illustrated, it is not always obvious what type of macroeconomic model is being discussed. The philosophy of science literature on model-based explanation helpfully distinguishes between surrogate and substitute models (Mäki 2009, p. 36). Surrogate models attempt to bring known facts about the real world into the world within the model, so that when elements of the two are given the same name it is recognisably the same thing being written about. This is what we would expect from Blanchard’s policy models. A direct resemblance relationship is implied, similar to the association between a picture and what it is a picture of (Gelfert 2011, p. 275). Substitute models, though, operate further removed from observational data, in the manner of Blanchard’s theory models. They are best viewed as ‘purposefully constructed artefacts’ (Knuuttila 2011, p. 263). At most the representational relationship in this latter instance involves the model ‘standing in for’ the real world, not ‘re-presenting’ it (Prendergast 2000, p. 7, 5).

As before with theory and policy models, a blurring effect can become apparent. Under the influence of politicised plotlines, substitute models reflecting hypothetical conditions often get treated as if they are surrogate models speaking directly to the real world. They can thus acquire properties that make them believable to opinion formers and legislators without the need to display data coherence. The most important feature of macroeconomic models in these moments is that their narrative content already has a self-selecting audience: they tell a story that people in positions of power want to hear. In the case of DSGE models, this typically takes the form of warnings that the state should not extend its reach too far into the economy. Such rebukes emerge from a substitute model, but they often get treated as if they were a feature of a surrogate model that is more closely related to real-world data.

This reveals an important tension. According to Margaret Morrison (1999, p. 64), substitute models are best conceived as autonomous investigative instruments used only for stimulating further conceptual activity into behavioural regularities. As mathematical formalisms they offer macroeconomists no direct access to understanding real-world dynamics that are noticeably less well ordered (Storm 2021, p. 79). They are fictions, whatever their creators’ success in choosing a mathematical structure that allows the economically-named components of the model to mimic how theoretical convention assumes they work in practice. But they are fictions that are usually readily accepted as windows onto the much messier process of managing actual economies. On what philosophical basis, then, might the users of substitute DSGE models claim inferential capacity to the real world, hence eroding the distance between surrogate and substitute models in a similar manner to the creation of grey-area zones between theory and policy models?

Sugden (2000, p. 23, 2009, p. 16) suggests that substitute models need only be based on ‘credible worlds’ to facilitate learning about the real world (see also Suárez 2004, p. 773). They might therefore still permit inductive inference about policy-makers’ most sensible course of action even if they serve no obvious representational function regarding the actual situations in which policy-makers find themselves. The criterion against which they must be judged is merely whether they capture one of the set of all thinkable worlds: ‘how the world could be’ (Sugden 2000, p. 24).

However, the narrative content of macroeconomic models creates fuzzy boundaries around the concept of a credible world. Political choices underpin the storytelling strategies through which macroeconomic models are presented, whereby what one believes is thinkable will be influenced by the state of the economy one hopes will be realised. It is therefore necessary to go further and add more substantive detail to how substitute macroeconomic models might pass Sugden’s credibility test. Uskali Mäki (2009, pp. 39–40) distinguishes between imaginability, conceivability, possibility and plausibility to capture four distinct types of credible world. They encompass a spectrum covering the grey-area zone between theory models and policy models; pure theory models sit at the imaginable end of the spectrum, pure policy models at the plausible end. The move from one to the next entails progressively stricter demands of data coherence, thereby embedding ever more characteristics of surrogate macroeconomic models into the construction of artefactual entities. It also changes what stories can be told by imposing different demands of realism at each stage.

To be imaginable requires only that the creative will of the modeller can call to mind the structure of the hypothetical economy, whereas to be conceivable requires the extra step that it is not disqualified by stylised facts about how the world works. The Arrow-Debreu model of general equilibrium is imaginable in the sense that they were able to write down a system of equations that demonstrates logically the right to think that such an equilibrium could exist, but the restriction it places on all markets forever adjusting instantaneously to equilibrium is hardly conceivable as a recognisable real-world feature. To be possible requires that the mechanism driving economically meaningful results in the model is also clearly in operation in the real world, whereas to be plausible requires that the real world responds to that mechanism in the same way as the model world. The appointment of policy-makers with supposedly the correct characteristics meant that conservative central banker models passed the test of possibility throughout their twenty-year heyday from the mid-1980s, but actual policy-makers’ failure to minimise the social loss function in the way the models predicted seems to have prevented the achievement of plausibility.

Surrogate models typically demand possibility as a minimum criterion of success, but substitute models might require conceivability at most. This is because surrogate models produce claims about how one causal variable affects another beyond the model when actual policy-makers are required to act, whereas substitute models produce claims about how variables interact within the theoretical construction of the model itself. It is therefore of potentially great significance which of these model-types the reader is asked to contemplate, especially when the language through which the model is committed to the page does not make it clear. Surrogate models will be described using standard macroeconomic naming practices of ‘inflation’, ‘unemployment’, ‘output’ and ‘the price level’. Substitute models will do the same, but the named features are this time more likely to have a formal mathematical character than an underlying empirical presence.

Policing the line between the two involves acknowledgement that solving a system of equations means only that one understands the logic of the initial assumptions which allowed each equation to be written down. The philosopher Jaakko Kuorikoski (2021, p. 190) calls for ‘a sharp distinction between this formal understanding and explanation proper’, whereby it is recognised that mathematical artefacts have epistemic value when used for abstract reasoning but stop short of entering ontologically into explanatory dependencies (Saatsi 2016, p. 1059). However, for the users of substitute macroeconomic models to engage in inferential induction to the real world, they must somehow bridge this gap between two entirely different types of truth claim. More likely, they will fall back on narrative content to mask the gap’s true nature.

The process of mathematical moulding will constrain what stories can be told, because it incorporates numerous theoretical assumptions about how the economy works in general (Morgan 2001, p. 366). For instance, Tinbergen’s earliest national income accounting model, the Klein-Goldberger model of dynamic Keynesian theory, Phelps’s model of the vertical long-run Phillips curve, Dornbusch’s overshooting model and Kydland and Prescott’s time inconsistency model will all have their own particular political lessons to impart. More generally, models of a competitive, price-taking, general equilibrium economy must assume away numerous goods market, labour market and money market rigidities that might otherwise be expected to impede the market-clearing solution. The parallel process of narrative moulding also has to tell a story of equilibrium produced through decentralised economic activity. Importantly, though, the story must appear to be about real-world situations that the reader can be persuaded relate to their own experiences. Yet for every increase in the formal nature of the mathematical structure of the model, there will be a corresponding narrowing of the range of stories the model will licence. DSGE macromodelling has certainly reduced how many substantive puzzles are of interest to the specialist research community (Wren-Lewis 2016, p. 27). However, this has not been through sustained observation generating settled empirical understanding of key macroeconomic variables, but by using a particular mathematical structure to disqualify all but a small number of stories that might be told about the model world.

It is the story that answers the question to which the model is aligned, not the structure of the model itself (Strassman and Polanyi 1995, p. 104). The decision that a macroeconomic question is worth considering arises in the first instance from impressions about what is important in the real world, and this is revealed through the content of the accompanying narrative rather than through the choice of model components. The final section provides two examples of the influence of the narrative plotline as an autonomous element of macroeconomic models. They are drawn from early interventions in real business cycle theory, pioneering studies that have shaped in crucial ways today’s dominant DSGE models. Real business cycle theory’s alternative storyboard – captured most dramatically in the title of Lucas’s (1980) paper, ‘The Death of Keynesian Economics’ – insists that theories of optimal control commit a category error in assuming that macroeconomic stabilisation is always in the gift of policy-makers. By operating in the Weintraub-Boumans-Morgan tradition, it is possible to see how a purely mathematical story is animated by forms of prose that make the mathematics economically interpretable to other specialists. Yet still something extra is needed to understand how these new substitute models attained such deep political resonance.

The politics in the macromodelling process

The relationship between model and narrative must be relatively straightforward if the narrative is to work in its own terms, but also sufficiently subtle for the model to be accepted for its economic insights rather than just for its normative position-taking. I outline two forms that early real business cycle theory’s storytelling strategies took in attempts to be sufficiently obvious so as not to be missed, but insufficiently overt so as not to draw too much attention. One I call a broadly ideological outlook, the other narrowly partisan. The former is brought into existence in substitute worlds that imagine an essential fallibility to macroeconomic policy-making (consistent with Mäki’s ‘possible’ credible world), the latter in substitute worlds that stress strict limits on governments’ capacity to enhance social welfare (consistent with Mäki’s ‘conceivable’ credible world). Both are theory models masquerading as something more than that.

The first appears in Robert Lucas’s rightly acclaimed 1972 paper, ‘Expectations and the Neutrality of Money’. Looking back at that paper today, it seems largely unremarkable, entirely typical of what has now come to pass as macroeconomic theory. In its own time, though, it was a bolt from the blue, simultaneously challenging both theoretical and methodological conventions. Macroeconomic theory had previously been a study of aggregates, trying to understand how different policy settings might influence the behaviour of the economy as a whole. Lucas (1972) refashioned the question triggering the modelling endeavour from ‘what happens to aggregate demand under different conditions?’ to ‘what counts as optimal policy settings once everyone is granted theoretically pristine behavioural characteristics?’

We see in these questions the transition from surrogate to substitute models. Lucas’s macroeconomic theory reflects basic microeconomic principles of agential rationality, with the assumption of rational expectations being inserted into a dynamic general equilibrium model. This was immediately accepted amongst his peer group as a step increase in clarity. Yet clarity regarding what precisely? The economic relationships thus depicted were self-evidently of Lucas’s model world rather than of the real world, but not that he says as much. His discussion moves from solving the equilibrium equations (of the model) to discussing the positive implications of the theory (within the model) and asserting the lessons for policy (again within the model). His naming strategy reminds the reader that they are engaged with a macroeconomic model, but it is his mathematics that provide these words with a specific storytelling interpretation. The famous 1972 paper seems to be a perfect illustration of Sugden’s (2000, pp. 27–8) lament that macroeconomists typically tease that they are talking about the real world in the language they use to establish the problem they are tackling, but never get round to actually doing so when constructing or working with the model.

Lucas (1981, p. 271) argues that the primary advantage of operating with a specifically substitute model is that each component can be explicitly specified in full; therefore, every statement the model facilitates about itself must necessarily be a statement of fact. They are mathematically true within the model, but only mathematically true, and still then only of the model. As Nancy Cartwright (2007, p. 226) suggests in a classic case of understatement, modern macroeconomic models post-Lucas typically ‘have a lot of structure’. That structure comes from assigning certain attributes to the behaviour of the agents in the model, so that the solution to the system of equations can be presented as confirmation of the model’s animating storyline. The list of assumptions that Lucas makes as he builds his model covers everything from limiting the amount of time everyone is allowed to live to asserting how their preferences will change as they move between the two states of being young and being old (Lucas 1972, pp. 104–11). The full list goes on for six-and-a-half pages of ever tighter restrictions on how his model-bound economic agents are required to behave. Yet when Lucas constructs his problem frame to say what his model implies about the real world, the relationship between the assumptions and the inferences is quietly forgotten for fear of casting doubt on the accompanying story.

This is not to say that he created an elaborate mathematical façade merely to make a narrowly partisan point. The story that brings his model to life is more subtle than that. It follows a series of steps. If (1), we allow every economic agent to hold rational expectations; if (2), we place two generations of those agents in separate island markets, where neither switching nor communication is allowed; and if (3), we construct everyone’s objective function on the proviso that they will seek rationally to maximise lifetime utility; then (4), it is possible to observe within the model dynamic fluctuations not dissimilar to the business cycle observed in the real world; as long as (5), economic agents find it difficult to distinguish in the moment between real economic expansions that enlarge output and monetary-induced expansions that inflate the general price level. Lucas’s model replicates the Phillips curve trade-off between unemployment and inflation in a dynamic general equilibrium setting, providing what had previously been a regularity witnessed in aggregate data with theoretical foundations of a microeconomic nature.

At this level, the model’s animating story is almost entirely descriptive within a clearly visible ‘if … then’ framework. However, it also possesses an all-important hortatory element only fractionally beneath the surface. These two narrative dimensions are conditional upon each other, and the model no longer works in the absence of either. Lucas’s path-breaking 1972 paper is broadly ideological in the terms I use here, because his microfounded Phillips curve urges caution upon his only ever hypothetical policy-makers, telling them that their ability to choose how to manage the unemployment-inflation trade-off is limited in the presence of rational economic agents. But it does not tell them exactly where those limits lie. Notice the switch when he says that real-life governments might therefore have to get used to doing less than they imagine is optimal from a social welfare perspective (the descriptive component of the story), even if how much less remains an open question on the basis of his model (the hortatory component).

The example I use of a more narrowly partisan politics of macromodelling appears in Thomas Sargent and Neil Wallace’s 1975 paper, ‘“Rational” Expectations, the Optimal Monetary Instrument, and the Optimal Money Supply Rule’. In its day it made a similar splash to Lucas’s 1972 paper, but it has aged less well. It takes Lucas’s microfoundations project as its starting point and still stands as a crucial contribution to the development of real business cycle theory based on the behaviour of rational individuals rather than economy-wide aggregates. Unlike Lucas’s model, though, which showed how the new theory could replicate in the model world something resembling the real-world attributes of the Phillips curve relationship, Sargent and Wallace’s more declarative conclusion that policy-makers have no ability to influence aggregate demand through monetary policy never made the leap from demonstration in their model world to demonstration of real-world similarity effects.

However, this was not for want of trying. The essential theme of Sargent and Wallace’s 1975 paper is that governments should vacate active stabilisation plans because even their best-intentioned interventions from a social welfare perspective are ultimately likely to do more harm than good. This conclusion split the macroeconomics profession during the 1970s. Some sought immediate demonstration of its empirical robustness because of the political message the policy ineffectiveness proposition carried about the obsolescence of optimal control policy (Barro 1977, p. 102). Yet even sympathetic interlocutors concluded fairly swiftly that true-in-the-model was unlikely ever to become ‘strictly true’, and that sound inferential induction to the real world was always likely to fail on the grounds that the whole argument was only ever ‘an approximation’ (Sheffrin 1996, p. 47, 42). Perhaps the necessary evidence was hiding in plain sight all along. After all, Sargent and Wallace (1975, p. 254) had admitted in the conclusion to their original paper: ‘Because of their ad hoc nature, neither the structure set out in section 1 nor the loss function of section 2 should be accepted as providing a suitable context within which to study macroeconomic policy’.

Yet this is by no means the message imparted by the storytelling strategy through which the paper’s main argument is delivered to its readers, when it is always actual policy-makers seemingly being told what they can and cannot do. Where there is a clear distinction between the descriptive and hortatory dimensions of the model’s animating story in Lucas’s paper, this is much less obviously so in Sargent and Wallace’s. Descriptively, their narrative concerns the necessity of restrictive rules consistent with a hands-off approach to policy-making, because the foresight that is attributed to rational economic agents means they cannot be tricked into believing the non-neutrality hypothesis: i.e. that the injection of more money into the economy can lead to a real expansion. Sargent and Wallace (1975, p. 249) allow for no exceptions in this regard: ‘There is no systematic rule that the authority can follow that permits it to affect the unexpected part of the price level’. Such a definitive disqualification of discretionary policy-making reflects the idea that rational economic agents will successfully pre-empt surprise changes to the money supply, but anything other than surprise changes will immediately be factored into behavioural choices and will therefore leave the real output level where it was. This conclusion partly reflects the assumptive base on which the model is built. As with Lucas’s 1972 paper, there is a significant amount of structure here, with seven-and-a-half pages being reserved for outlining all the assumptions. On this occasion, though, the mathematics also lends a crucial hand.

Policy-makers’ attempts to manage the level of aggregate demand are modelled as parameter interventions, but the mathematical expression of such an assumption leaves untouched the relative pattern of forces acting upon the mean of the price level. In turn, this displaces the mean successively to the right on a standard demand-and-supply diagram, such that policy can never impact overall levels of output (Frydman and Phelps 2013, p. 22). Lucas’s model captures elements of the Phillips curve relationship, whereby some of the new money introduced into the economy to stimulate additional activity has inflationary consequences, but some of it has the desired effect. Sargent and Wallace’s, by contrast, describes aggregate behaviour that is undetectable in any known data, where a monetary injection feeds only ever higher inflation consistent with a vertical long-run Phillips curve.

Lucas’s model is founded on conventional ‘if … then’ logic, where the hortatory dimension of the model’s animating story comes to the fore in the implications of that logic but without overpowering its descriptive dimension. Sargent and Wallace seem intent instead to let hortatory statements dominate. Lucas uses his model to explore the repercussions of assuming that all economic agents can act rationally with respect to both the present and the future. Sargent and Wallace appear to have asked what a rational expectations model would look like were it to establish the conclusion that active stabilisation policy is always likely to be self-defeating. Lucas’s credible world is ‘possible’ in Mäki’s terms as a stand-in for our own, because it is consistent with observations of the unemployment-inflation trade-off, but Sargent and Wallace’s is merely ‘conceivable’. These are two different forms of narrative moulding to accompany the models’ mathematical moulding.

Conclusion

The normative claims to which macroeconomists subscribe is not a reliable indicator of why they commit themselves to a particular class of model. Professional understandings of what counts as acceptable economic theory is a much better explanator. However, as is proved by the continued dominance of DSGE models even in the midst of widespread claims about their practical failures, at any moment macroeconomists typically have few core mathematical structures to choose from. They do have more options, though, when it comes to the stories that turn mathematical structures into specifically economic models. DSGE models are theory models that nonetheless get used to promote different policy recommendations, even if at this latter level it sometimes seems to be arguments over very small differences. It is the storytelling devices surrounding the models that explain such divergence. It is always a political act to summarise the findings of a macroeconomic model in a story containing specific policy lessons, but individual economists will be able to perform that act in different ways. Mathematical moulding always exists side-by-side with narrative moulding.

However, this positions the analysis on complex ontological terrain. The economic question of how reliable theory models are for policy analysis has its philosophical counterpart in how the construction of purely substitute worlds might be of epistemic value regarding real-world puzzles. Fortunately, the existing work of various economists, economic philosophers and historians of economic thought has made it possible to begin to understand the significance of political storytelling devices for hypothetical mathematical modelling. By building upon the prior research of Weintraub, Boumans and Morgan, I have shown that the narrative moulding of macroeconomic models often sustains important grey-area zones between theory and policy models and between surrogate and substitute models. I have provided examples of how two such interstitial spaces have been populated with political plotlines, where the storyboards serving as the historical backdrop for today’s DSGE models seem to have the characteristics of being broadly ideological (Lucas 1972) and narrowly partisan (Sargent and Wallace 1975). These, though, should not be mistaken for the outer edges of a spectrum containing all possibilities. What seems to one reader to be a broadly ideological political stance might look narrowly partisan to another. It might also carry altogether different political meaning for a third, or none at all for a fourth. Care must therefore be taken not to claim too much for these as explanatory categories.

Moreover, it is generally only ever the descriptive element of the underlying story that is rendered explicit, and usually only then in technical terms. The hortatory element typically receives no mention, being deemed to lie outside the boundaries of proper macroeconomics, even though this is where the embedded plotlines that are intrinsic to the success of the model are contained. The audience for macroeconomic models is therefore required to engage in a creative reading-in process if the full content of the underlying storyboard is to be pieced together, because this is not something they can rely on the models’ creators to do for them. There will consequently always be a degree of conjecture involved, as different readers are likely to depart from one another over how best to fill in the blanks of what is left unsaid. Further focus on the political content of macromodellers’ storytelling devices is therefore required if there is to be any chance of resolving the epistemic riddle of how purely substitute models can appear to pass definitive judgement on real-world policy options. This can also be expected to shed light on why certain theory models have historically retained their dominance despite evident failures in practice.

Acknowledgements

I would like to thank the participants at the original online workshop from which this special section has arisen for all their helpful comments, in particular Scott James for continuing to drive the project forward in a thoughtful, creative and typically selfless manner. I would also like to thank Andreas Kanaris Miyashiro and NPE's anonymous reviewers for pushing me to refine my argument still further about the ontology of hypothetical mathematical modelling in macroeconomics.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Economic and Social Research Council [grant number ES/K010697/1].

Notes on contributors

Matthew Watson

Matthew Watson is Professor of Political Economy in the Department of Politics and International Studies, University of Warwick. From 2013 to 2019, he was also a UK Economic and Social Research Council Professorial Fellow.