Less theory, more reality: Semi-structural models have a lot to offer

Central banks today face the increasingly complex task of accurately predicting economic developments while clearly communicating the main scenarios and risks. Macroeconomic models are a key tool that enables them to fulfil this task. In addition to fully structural DSGE models, semi-structural models have gained increasing popularity in the last two decades.^[1] As the name suggests, their structure is more flexible and can better capture the changing economic environment. At the Czech National Bank, we are reflecting this trend by developing two new models of this type.

What are semi-structural models, and why do central banks develop them?

Semi-structural models combine elements of economic theory with empirical relationships estimated from data. Unlike DSGE models, not all of their equations are directly derived from the optimization behaviour of individual actors. Instead, they use theoretically motivated but data-flexible structures, which allow for faster adaptation of the model to new or changing economic phenomena. Therefore, central banks often use them as primary forecasting models or at least as tools that complement or shadow the main structural models – especially when making predictions and analysing alternative scenarios.

This flexibility has made semi-structural models popular across the world. According to Central Banking’s Economics Benchmarks 2024, more than 60% of central banks now use semi-structural models as their primary forecasting tool.

Figure 1 – Type of main prediction model of central banks

Note: The survey was conducted among 34 central banks; source: Central Banking Benchmarking Service/Economics 2024, CNB’s own processing.

Types of semi-structural models

Semi-structural models encompass a broad range of approaches that vary in their reliance on economic theory versus empirical data. At one end of the spectrum are models that are closely aligned with the structural approach, often referred to as gap models, which are based on theoretically derived equilibrium relationships. At the other end of the spectrum are models that are strongly econometric in nature, where the structure is primarily shaped by empirical estimation and dynamic data relationships.

Scheme 1 – The spectrum of semi-structural models in central banks and at the IMF

(Open the whole scheme in a new window.)

Note: The semi-structural models are arranged in columns from the most econometric to the most structural from left to right. Models that are used as primary forecasting models in central banks on their own are marked in red. Models that are used alongside other models at the primary level are marked in green, while secondary models are marked in blue. Source: CNB

Models such as the International Monetary Fund’s FINEX are closest to the structural approach. These models are based on clearly defined equilibrium relationships and explicitly separate long-term trends from short-term cyclical deviations (gaps). They are relatively compact and well suited to standard monetary policy analyses.

Models such as the US Federal Reserve’s FRB-US, the Bank of Canada’s LENS, and the European Central Bank’s ECB-BASE are moving towards greater empirical flexibility. While these models are still grounded in fundamental theoretical relationships, their dynamics are largely shaped by empirical estimation and allow for greater alignment with historical data.

At the more econometric end of the spectrum are models such as the Dutch central bank’s DELFI 2.0, the Polish central bank’s NECMOD, and the Bundesbank’s BbkM-DE. These models emphasize empirical estimates of intersectoral relationships and often employ an error correction mechanism (ECM) that links short-term dynamics with long-term trends. Their detailed structure allows for complex analyses, such as assessing the impact of the financial sector on the real economy, though this comes at the cost of increased complexity and more demanding interpretation.

The choice of a particular type of model or combination of approaches depends on the needs of the central bank. Some institutions prefer simpler, more communicative models for basic scenarios, while others favour more detailed models that allow for deeper analysis of structural changes and macro-financial linkages.

Advantages and disadvantages of semi-structural approaches

Different types of semi-structural models offer different advantages and trade-offs.

Models close to the structural concept provide high transparency, simpler interpretation, and easier communication of main scenarios. By emphasizing equilibrium relationships, these models help maintain the internal consistency of the economic narrative, which is key for understandable internal and external communication of predictions.

Models closer to the econometric approach, on the other hand, allow for greater flexibility, a better empirical fit to the data, and more detailed analysis of sectoral linkages and financial channels. This is especially valuable when the economy is undergoing significant changes that are not easily captured by traditional theoretical relationships.

However, each approach has its limits. More structural models can sometimes be less responsive to new shocks if these shocks fundamentally change existing trends. Econometric models, on the other hand, may lack a clear theoretical basis, which complicates the interpretation of predictions, especially when developing scenarios that incorporate the effects of monetary or fiscal policy.

As a result, an increasing number of central banks are combining different types of models in order to retain both analytical accuracy and the ability to communicate results clearly. This practice is confirmed by the results of a questionnaire survey conducted by the Czech National Bank among 23 central banks: most of them now develop their macroeconomic forecasts using two or more predictive models.

Table 1 – Number of predictive macroeconomic models and use of DSGE models in forecasting

Source: CNB questionnaire survey

The role of semi-structural models in different phases of the business cycle

The choice of model approach often depends on the nature of economic developments.

During periods of stable growth and standard conditions, more structural models tend to be more appropriate, supporting the systematic production of forecasts and scenarios with an emphasis on stabilizing key macroeconomic gaps.

By contrast, in times of heightened uncertainty or structural change, or when analysing the effects of new policies, models with a greater degree of empirical flexibility become more relevant. These models allow for quicker adaptation to new information and better capture of complex transmission mechanisms – for example, through the property market or changes in credit standards.

For this reason, most institutions do not rely on a single model but rather maintain a portfolio of tools to address a range of analytical requirements.

Our decision at the CNB: why we are combining multiple models

At the Czech National Bank, we have begun to develop two new semi-structural models aimed at expanding our modelling framework to better meet the diverse analytical needs of the monetary policy process. The objective is to achieve a more comprehensive representation of economic reality through varied modelling approaches, thereby reducing the risk of overlooking important factors and limiting the potential for monetary policy errors stemming from reliance on a single methodology.

This strategy builds on the experience of other central banks, which demonstrates that working with a portfolio of models make it possible to better capture economic uncertainties, respond more flexibly to new shocks, and enrich discussions of alternative scenarios. At the same time, we are responding to the key conclusions of last year’s external assessment of the CNB’s analytical and modelling framework for monetary policy.

The first model under development is a new policy model inspired by more structural approaches, such as the IMF’s FINEX. It will focus on key macroeconomic linkages and enhance the internal consistency of the baseline scenario, thereby facilitating clearer internal and external communication of the main transmission mechanisms.

The second model is a short-term forecasting tool oriented towards econometric analysis, enabling more agile use of current data. This model will incorporate a broader range of data sources and focus on rapidly identifying short-term trends, including sectoral heterogeneity and potential mismatches between different parts of the economy. Its design draws inspiration from the approaches of the German Bundesbank, the Dutch DNB, and the Polish NBP.

Combining these two approaches will enable us to test alternative scenarios more effectively, work more systematically with uncertainty, and improve the robustness and transparency of the results for the Bank Board and the public. At the same time, it will support richer discussion and internal scrutiny of the baseline forecast, which will continue to be produced using the g3+ structural DSGE model for the foreseeable future.

What’s next: working with multiple models

At the Czech National Bank, we are at the beginning of a comprehensive process to develop two new semi-structural macroeconomic models with the potential to expand our modelling framework. We are currently in the phase of designing and constructing these tools and the supporting infrastructure. This will be followed by shadowing the baseline forecast scenario and testing the new models in real time. Only after evaluating the results will the Bank Board decide on their possible “live” deployment and the future configuration of the forecasting framework.

Expanding the modelling framework to include additional semi-structural models brings not only new opportunities but also a number of important questions that must be addressed. How can multiple models be most effectively integrated into the monetary policy process? How can their outputs be combined in a consistent manner? And how can the results be communicated clearly and effectively to the Bank Board and the public?

Central banks across the globe are adopting different approaches – ranging from reliance on a single primary model to the systematic integration of multiple tools. Each strategy has its own strengths and drawbacks, and getting the configuration right is crucial to providing high-quality support for monetary policy. These issues are already under active consideration at the CNB and will be explored in greater depth in a forthcoming blog post.

^[1] Central banks and other national institutions also use other, mostly complementary, approaches that usually do not fulfil the role of core prediction model for monetary policy purposes. One example is the Bank of Canada’s CANVAS model, a representative of the class of simulation behavioural models (agent-based models, ABMs). Other approaches include overlapping generations (OLG) models, which are suitable for evaluating the long-term impacts of alternative scenarios, and stock-flow consistent models (SFC), which can faithfully model stocks and flows in the economy as captured in national accounts and input-output tables.