The CNB’s projection model gets a new plus

Forecasting future macroeconomic developments is key to deciding current monetary policy, as changes to monetary policy instruments always pass through to the economy with a time lag. In developed countries, macroeconomic forecasts are usually produced with the aid of core projection models. Central banks use these models to simulate the most likely scenario of future developments based on an assessment of the current position of the economy and on assumptions about the paths of foreign variables and other domestic variables that are independent of monetary policy (such as government fiscal policy). Alternative and sensitivity scenarios formulated using the models are sometimes coupled to the baseline scenario. The majority of advanced central banks employ projection models to create this fundamental input to monetary policy decision-making. The CNB is no exception.

“Gee-three” bows out…

The CNB used the g3 model as its main forecasting tool from 2008. Before that, it had used a semi-structural quarterly projection model called QPM.1 g3 was a structural dynamic stochastic general equilibrium model based on microeconomic foundations capturing the main characteristics of the Czech economy. Given the high openness of the economy, it emphasised foreign trade and the exchange rate linked to domestic monetary policy under inflation targeting. Forward-looking expectations and an endogenous monetary policy response played an important role.2

Immediately after its deployment 11 years ago, g3 successfully guided domestic monetary policy through the global economic crisis. It then assisted in the monetary policy debate about the exchange rate commitment and recently also contributed actively to the gradual return of Czech monetary policy to normal. It proved its usefulness and accuracy in practice, as evidenced by the following figure comparing the g3 forecasts with the actual historical path of consumer inflation. With the benefit of hindsight, one can conclude that the “gee-three” forecasts were very close to the observed data and did not deviate significantly from them, except between the end of 2013 and the start of 2017 (a period characterised by protracted and surprisingly large falls in foreign prices, which led to the extension of the CNB’s exchange rate commitment).

Consumer price index (year-on-year growth in %)

Consumer price index (year-on-year growth in %)
Source: CZSO data and CNB forecasts.
Note: The red lines represent all the quarterly g3 forecasts, starting with Inflation Report III/2008 and ending with Inflation Report II/2019, and the black line plots the observed data.

The good forecasting ability of g3 can be quantified using the deviations of its historical predictions of the main macroeconomic variables from the actual subsequently observed data.3 In the case of consumer inflation, the average deviation fluctuated around 0.8 pp one year ahead. This reliably kept the inflation forecasts at the monetary policy horizon within the tolerance band of ±1 pp around the CNB’s 2% inflation target. In the case of the exchange rate, the historical error of the one-year-ahead g3 projections was just under one koruna to the euro. For annual GDP growth and nominal wage growth it was in the range of 1.7–1.9 pp. In terms of the real economy, the error encompassed not only sizeable historical data revisions, but also the global financial and economic crisis, the size of which (unprecedented in the modern era) no one was capable of predicting reliably. From this perspective, too, the predictive power of g3 was very respectable, confirming its usefulness in practice. As a result, enhancing the forecasting ability of the core system was only one of the reasons why the CNB began work on a new model a few years ago.

…and g3+ takes the stage

g3 played the lead in a total of 44 forecasts over its lifetime and is bowing out to its successor g3+ starting with this year’s summer forecast. g3+ was successfully tested during its development and subsequently introduced to the public as an alternative scenario in Inflation Report II/2019. The new model contains a whole raft of extensions and thus adds a “plus” to the original “gee-three”, not only in the name, but also in the form of better forecasting properties and a more detailed and realistic description of the behaviour of the Czech economy, which has significantly changed and “grown up” over the last 10 years.

The Czech economy is a small open economy that is strongly affected by developments abroad. This finding may not be new, but it acted as a strong incentive to develop a more detailed description of the foreign part of the model. This segment, primarily describing developments in the effective euro area,4 has been extended to include new, mutually interconnected variables.5 This not only allows for consistent interpretation of developments abroad, but also opens up room to create various foreign scenarios using the core projection model. As regards foreign variables, the steady states for economic growth – which is slower following the global economic crisis – and foreign interest rates have also been revised. Interest rates are lower than before in the long run, because they reflect the ECB’s monetary policy over the last 10 years or so. The foreign block of the new model structured in this way allows observed developments in the euro area to be described more realistically and the outlooks for the euro area to be analysed consistently.

Several improvements have also been made to the part of g3+ describing the behaviour of the domestic economy. In particular, the structure of foreign trade is captured more accurately. The share of energy in imports to the Czech Republic is far larger than its share in Czech exports. The new breakdown of foreign trade into its energy and non-energy components therefore allows us to assess the competitiveness of Czech exporters on foreign markets more realistically. In addition, it is easier to identify the correct size of the contribution of energy import prices to the costs of domestic firms. Another major change in g3+ is that it includes households that spend their entire income on consumption each quarter. Such households account for a significant share of the Czech population, so this change will more credibly describe the actual consumption of households. The final, no less significant change is the incorporation of the specific path of the koruna exchange rate following the exit from the exchange rate commitment through an adjustment to the exchange rate equation.

The package of changes also contains a new and innovative approach to working with model expectations regarding future outlooks. This approach tries to find the right amount of information relevant to current monetary policy decisions and in particular to reflect the fact that forecasts of the more distant future are less reliable. Although there is currently no single tried and tested method for incorporating people’s expectations into the modelling framework, the approach used in g3+ is a step in a potentially promising direction.6

What does it mean?

The new g3+ model certainly changes neither the modelling approach, nor the analytical and forecasting modus operandi at the CNB. By switching to g3+, the team of model operators and developers in the CNB’s Monetary Department are trying to keep pace with current R&D in the field of macroeconomic modelling and with modern trends in the practical use of models to support monetary policy decision-making.

The advantages of the new model are that it is closer to the current reality of the Czech economy, has better forecasting properties, is more user-friendly and provides a more detailed economic story to the members of the CNB Bank Board and other users of the CNB’s forecasts.

g3+ has now gone live and is the CNB’s core projection model starting with this year’s summer forecast.7 We wish it a long and interesting life, the lowest possible errors and many successful forecasts!


1 An undoubted advantage of g3 over QPM was that it provided a more consistent theoretical framework for forecasting and monetary policy analysis.

2 Monetary policy endogeneity means that interest rates and the exchange rate respond in the forecast to the development of and expectations about all the other macroeconomic variables included in the model. The forecast thus also incorporates the expected behaviour of the central bank itself and indicates the interest rate and exchange rate path consistent with the overall outlook.

3 The deviations are measured using the root mean square error (RMSE).

4 The effective euro area indicators capture the effect of economic activity and inflation in the euro area on the Czech economy. The weights used in the calculation of the effective indicators equal the shares of the Czech Republic’s most important euro area trading partners in its total exports to the euro area.

5 The foreign part of the model now includes consumer prices in the effective euro area, the energy and core components of the producer price index for the effective euro area, the output trend and output gap, the euro-dollar exchange rate, the Brent crude oil price and the foreign equilibrium real interest rate. The impacts of changes in global oil prices and the euro-dollar exchange rate previously figured in the CNB’s forecasts only via expert adjustments to foreign and import prices and related domestic price categories (such as the PPI and administered prices).

6 A box in Inflation Report III/2019 and a CNB Working Paper currently in the pipeline will provide a more detailed description of g3+.

7 g3+ was used to prepare the baseline scenario of the CNB’s summer macroeconomic forecast, while the old core g3 model was used to create the alternative scenario. Rather than describing a specific risk of the baseline scenario, the alternative scenario is intended to outline a forecast from the perspective of the former core projection model (like in Inflation Report II/2019 except the other way around).