The fourth international workshop on empirical Stock-Flow consistent models will be in Paris, November 26-27, 2026.
See details here.
Background and Rationale
Stock-Flow Consistent (SFC) models have become increasingly ambitious as empirical and policy tools, but there is still no broadly shared approach to estimation and validation techniques. At the same time, the behavioural framework is unevenly balanced, as the focus is placed mostly on the demand side, while constraints arising from the supply side, as well as changes in the productive capacity, remain relatively underdeveloped. These gaps, from how parameters are estimated and identified, to how models are validated and their sensitivity assessed, run through the full empirical workflow and they interact with the behavioural specifications and time representations on which SFC models are built.
With regards to estimation techniques, most empirical SFC models rely on single-equation estimation in discrete time, often through ARDL, error-correction, or related partial-adjustment specifications, combined with calibrated parameters where data are limited or identification is weak. This practice has clear advantages, as it is transparent, computationally tractable and compatible with the necessarily modular construction of large models. However, it also requires careful treatment regarding simultaneity, endogeneity, cross-equation consistency and the propagation of uncertainty from separately estimated equations into the full model.
At the same time, alternative approaches are emerging, such as large-block or system-wide calibration, simulated-method-of-moments, Bayesian estimation and continuous-time estimation. Joint or system-wide approaches may better respect the fact that most variables in SFC models are co-determined within the accounting structure, rather than generated by isolated behavioural equations. However, they inherit the general difficulties of empirical SFC work (limited sample sizes, mixed-frequency data, weak identification, high dimensionality and discrepancies between observed data and accounting identities), while adding challenges of their own. These include specifying a likelihood or objective function for the full system, handling latent variables and accounting constraints explicitly, choosing priors or regularization in weakly identified directions and managing the computational burden of large models.
These methodological choices are closely tied to the behavioural specifications being estimated. Common specifications such as ECM, ARDL, or portfolio-allocation equations are not neutral vehicles for estimation. They shape the interpretation of estimated parameters, including adjustment speeds, long-run coefficients, lag structures and stock-flow relations. This issue is especially important when moving between discrete-time and continuous-time formulations, since parameters estimated at quarterly or annual frequency may not have the same interpretation as parameters defined in continuous time and may have other trade-offs. Additionally, it is crucial to consider not only how parameter estimates are obtained, but also how estimation methods, time representation and behavioural specifications interact. These questions motivate a focused discussion of how parameters are obtained in empirical SFC models and of what the community might reasonably treat as emerging practice.
Concerning validation, the empirical checks routinely used in SFC modelling such as reproduction of stylised facts, steady-state and accounting-consistency diagnostics and comparison of simulated against observed moments, are rarely articulated as a systematic part of the sensitivity-analysis workflow. Furthermore, there is no shared vocabulary, no minimum reporting standard for what an SFC model should disclose about the robustness of its results, while there is limited reusable infrastructure for performing sensitivity analysis at scale on models that are typically expensive to simulate and constrained by data availability.
Switching to the behavioural domain, empirical SFC models build on a demand-determined modelling framework, in line with the post-Keynesian tradition. Therefore, aggregated macroeconomic output is given as the sum of the components of aggregate demand (GDP from the expenditure side, including consumption, investment, etc). While supply constraints are not mandatory within SFC models, there has been increasing interest in formally incorporating supply-side constraints into the empirical SFC framework.
Supply side factors have currently been incorporated in SFC models in several ways, including, but not limited, to a) inflationary processes for one or more price variables in the model, b) constraint on the capital stock and the capacity utilisation rate, c) constraint on the workforce size and d) energy and material constraints.
These constraints can enter the model as behavioural drivers of economic variables, such as inflationary pressures reducing real wages and demand, which can be exacerbated by higher capacity utilisation rates increasing firms’ mark-ups over costs. On the other hand, constraints can also appear as hard limits where the model shifts to being supply determined at the constraint and one or more real variables is overridden to achieve this. This can involve rationing one component of demand, such as consumption (and therefore leading increases in saving), satisfying domestic demand through increasing imports in open economy models or allowing prices to shift such that while nominal variables are unconstrained, all real demand variables become constrained.
Closely related to the treatment of supply is the representation of the production structure itself. A growing number of empirical SFC models incorporate input-output data, yet most continue to rely on Leontief production structures with fixed technical coefficients. While this assumption is tractable and often defensible over short horizons, it sits uneasily with the long-run questions of structural change, technological progress and energy transition, that empirical SFC models are increasingly asked to address. This raises the question of whether, when and how technical coefficients should be allowed to vary over time. Doing so introduces its own conceptual, empirical and computational challenges, from preserving accounting coherence as input-output structures evolve.
Against this backdrop, the aim of this workshop is to bring together empirical SFC practitioners to take stock of current practices across these interconnected fronts and build a common understanding on emerging best practices. Specifically, the workshop seeks to map the approaches currently in use, to identify methodological and infrastructural gaps and identify a set of priorities that could guide the empirical SFC research programme in the coming years, potentially reported in a community white paper.