MONETARY POLICY REPORT | WINTER 2024 (appendix)
(authors: František Brázdik, Milan Frydrych, Tomáš Pokorný, Tomáš Šestořád, Jan Žáček)
The Czech National Bank employs a whole range of analytical and forecasting tools to prepare its forecasts. Central to these is the core projection model, which lies at the heart of the entire forecasting system. The g3+ model took over this role from the previous version g3 in mid-2019 (starting with Inflation Report III/2019).[1] However, analytical tools require constant refinement to reflect recent economic developments and forecasting experience. Innovations and improvements have therefore been made to g3+, as described in this appendix.[2] The appendix also presents a shadow forecast, which is intended to outline the forecast prepared using the updated version of the model, not to capture a specific risk of the baseline scenario. We plan to start using the updated g3+ model as the CNB’s new main forecasting tool in the spring Monetary Policy Report.
The updated g3+ core forecasting model
The enhancements made to the core model primarily reflect the extreme economic events seen in recent years, such as temporary sharp swings on the supply side and dramatic growth in energy prices and their role in firms’ production processes. We have revised certain linkages in the model to better capture the economic environment. Some of the previous expert judgement is now done via linkages built directly into the model. Other innovations include changes to the steady-state growth rates of certain macroeconomic variables. We also modify the way the external assumptions of the forecast are created. Among other things, we have started to monitor in detail the macroeconomic situation in Austria, another major European trading partner of the Czech Republic.
The Czech economy has been through a number of turbulent episodes over the last three years. To begin with, it faced the economic consequences of the Covid pandemic in 2020 and 2021. They were linked, among other things, with government measures that strongly affected the economic life of households and firms. The production capacity of the domestic and foreign economy was temporarily reduced, a factor that was difficult to capture in the g3+ structural model framework. Based on this experience, we have added short-term (sharp) supply swings to the decomposition of foreign economic activity into the output gap and potential output. This extension of the structure of potential output enables us to set more flexibly the long-term evolution of supply from the foreign economy using expert judgement (see Chart 1A). In addition, we now assess the position of the foreign economy in the business cycle directly within the core model, whereas the previous approach used methods outside it. This makes for a consistent estimate and provides a more complete structural narrative for the output gap in the foreign economy, as the estimate now incorporates, for example, inflation and the monetary policy stance abroad.
Chart 1 – New features of the updated g3+ model – foreign variables
1A Decomposition of effective EA6 GDP growth
y-o-y in %; contributions in pp
1B Weight on growth in the energy component of foreign PPI prices
%
Prices of energy commodities then surged in 2022. Producer and consumer price inflation began to soar. Foreign industrial producer prices (and hence prices of imports into the Czech Republic) were already broken down into an energy component and a core component in the g3+ model. Until the energy crisis broke out, however, our focus in terms of energy commodities was primarily on the price of oil, because the Czech and global economies had in the past faced a number of sharp oil price swings that had strongly affected producer prices of consumer goods, so it was more than desirable to capture their impacts precisely. We adjusted the energy component at the data level following the surge in prices of energy commodities (electricity and natural gas) in 2022. In the updated model, we nonetheless make a systematic change by also modifying the model structure itself. We expand the original definition of the energy component of foreign industrial producer prices to include other commodities, in particular electricity and gas.[3] We thus capture price movements on the energy market, which strongly affect companies’ costs. Moreover, we introduce a variable weight on the components of foreign PPI inflation into the model structure. This allows us to account for the temporary difference in the weight on energy price growth in the event of significant fluctuations in that growth relative to the core component (see Chart 1B).
In the domestic economy, we now consider the possibility of limited substitution between production factors (domestic intermediate goods, energy imports and other imports) for the production of goods for household consumption and exports.[4] This is particularly important in extreme situations such as the recent growth in energy prices, when other production inputs were substituted for expensive energy commodities, albeit to only a small extent. At the same time, we increase the imported energy intensity of domestic production.[5]
The changes to the core model also involve revisions to the validity of some relationships. Since 2008, the core model has used the concept of external demand,[6] a key factor for forecasting domestic exports. We have assumed up to now that the rate of growth in external demand for domestic exports in the long run can be expressed as roughly four times the rate of economic growth in the effective euro area. However, estimates of the relationship between demand for Czech exports and foreign economic activity over recent years indicate a steady decline in this elasticity. We therefore reduce the steady-state value of the multiple of external demand to three in the updated model. In addition, we make the coefficient time-variable. This allows it to diverge from its steady-state level (see Chart 2A) and fit the historical data more accurately. We also revise the effect of external demand on domestic investment activity in the short run. Besides the potential output of the foreign economy, which has long influenced investment demand in the Czech economy, we now incorporate the short-term effect of the effective euro area business cycle.
Along with making structural refinements to the model, we alter the steady-state growth rates of some macroeconomic variables too.[7] In the foreign block, the long-run steady-state level of energy-sector industrial producer price inflation is set to 2%. The steady-state year-on-year rate of growth in domestic imports and exports of goods and services is changed from 6.5% to 4.8%, in line with the revision of the trend in external demand for Czech exports as a multiple of effective euro area growth. This fits the observed data available for the pre-2020 – i.e. pre-pandemic – period.
Chart 2 – New features and forecasting performance of the updated g3+ model – domestic variables
2A Czech goods and services exports and external demand
y-o-y in %
2B Forecasting performance of the model for selected domestic variables
improvement in forecasting performance in % relative to current model
| 2001–2019 | 2001–2023 H1 | |
|---|---|---|
| CPI inflation | 3.2 | 4.2 |
| CZK/EUR rate | 0.4 | 5.0 |
| 3M PRIBOR | 14.7 | 20.8 |
| Imports | 17.3 | 8.7 |
| Exports | 24.0 | 25.9 |
Note: The figures in the table show the percentage difference in the root mean square errors (RMSE) of historical simulations between the current and updated model with full knowledge of the assumptions and with no expert judgement. The method for conducting such in-sample model simulations is described, for example, in Brázdik et al. (2022).
Besides making changes to the structure of the core g3+ model, we also modify the way the data input for the effective euro area economic outlook is used.[8] Up to now, we have prepared outlooks for consumer and producer prices and economic activity for five countries: Germany, Slovakia, France, Italy and Spain (the EA5). These countries (with specific weights)[9] then enter the aggregate effective euro area indicator. The remaining euro area countries are assumed to grow at the same pace on average as the EA5. Data for all the euro area countries in effective terms (the EA17) are used to describe the historical evolution.[10] There is therefore a methodological difference between the outlooks (EA5) and the historical evolution (EA17). However, our experience, underscored by events during the pandemic, indicates a need to change this practice, as the GDP growth rates of the EA5 and the EA17 can diverge considerably at times. We add Austria to the portfolio of countries we monitor closely and hence introduce an EA6 aggregate. We additionally unify the way we treat the data for forecasting and historical purposes, as we also use the effective EA6 indicator to describe the historical economic evolution of the Czech Republic’s major trading partners. This is because foreign industrial producer prices[11] and GDP turn out to follow very similar paths in effective terms in the EA6 and the EA17 (see Chart 3). By focusing on a narrower set of countries, we can also analyse developments in the economic area of greatest relevance to the Czech Republic in more detail.
In summary, the refinements made to the g3+ core forecasting model represent another step forward in forecasting practice at the CNB. The new and modified linkages in the foreign and domestic blocks help the model provide a truer description of the economic environment and enable it to capture the extraordinary economic phenomena of recent years in a structural – i.e. internally consistent, economically sound, objective and replicable – way. The richer model structure allows for more detailed analysis of economic developments. It meanwhile offers a consistent estimate of the position of the foreign economy in the business cycle and a truer representation of the role of energy prices in the production process. The changes to the model’s steady-state parameters fit the past long-run economic trends and also the expected future evolution of the Czech economy. The updated model delivers better forecasting performance. Table 2B shows that the forecasting performance of the model is substantially better not only for the period before the pandemic and the energy crisis, but also for the turbulent times of 2020–2023.
Shadow forecast
A shadow forecast was produced to demonstrate the effects of the changes made to the model and data. It is based on assumptions made about economic developments in the Czech Republic’s six largest trading partners in the euro area – the EA6 (see Chart 3). The changes made to the foreign outlook relative to that in the baseline scenario pertain to GDP growth, the identification of the position of the economy in the business cycle, industrial producer prices and their breakdown, and HICP inflation. The other foreign assumptions remain unchanged.
Chart 3 – Foreign assumptions of the shadow forecast
comparison of assumptions of baseline scenario with shadow forecast prepared using updated g3+ model
Effective euro area GDP
y-o-y changes in %
Effective euro area output gap
% of potential output
Effective euro area industrial producer prices – total
y-o-y changes in %
Effective euro area ind. producer prices – core component
y-o-y changes in %
The biggest differences are visible in the components of foreign producer prices, but their path as a whole is similar to that in the baseline scenario of the forecast. Core industrial producer price inflation is lower, while growth in energy producer prices is less negative than in the baseline scenario. This reflects the different compositions of the core and energy components of the PPI. Foreign GDP growth shows only slight differences compared to the baseline scenario, mostly in the past. This is linked with the switch to a narrower effective euro area aggregate, which involved an increase in the weights of countries that grew at slower rates (primarily Italy and Germany). Given the similar potential output growth estimates for the two aggregates, the lower GDP growth is reflected in a more negative output gap.
The shadow forecast drawn up using the updated g3+ model provides a similar view of the present state and future path of the domestic economy as the baseline scenario of the winter forecast but contains a few minor deviations (see Chart 4). Czech GDP growth is predicted to be rather lower in the shadow forecast. This is due to weaker domestic investment and export activity, reflecting a wider negative output gap in the effective euro area this year and the modified structural relationships in the updated model. For example, domestic investment activity now also takes the cyclical position of the foreign economy into account. This leads to lower investment growth than in the baseline scenario. The weaker domestic demand puts downward pressure on growth in consumer prices, so domestic inflation is a little lower in the shadow forecast than in the baseline scenario. Despite the rather weaker inflationary pressure, the decline in interest rates is similar as in the baseline scenario. The central bank thus returns inflation to the 2% target at the monetary policy horizon. The lower export activity caused by the more negative foreign output gap leads to a slight depreciation of the koruna this year, as opposed to a modest appreciation in the baseline scenario. Next year, the exchange rate strengthens at a similar pace and for the same reason as in the baseline scenario, i.e. mainly due to an improving trade balance of the Czech Republic.
Chart 4 – Shadow forecast of the main domestic variables
comparison of baseline scenario with shadow forecast prepared using updated g3+ model
3M PRIBOR
%
Nominal exchange rate
CZK/EUR
GDP
y-o-y changes in %
Inflation
y-o-y change in CPI in %
[1] The g3+ core forecasting model is described in detail in Andrle et al. (2009) and Brázdik et al. (2020), as well as in the Appendix to Inflation Report III/2019 Introducing the g3+ extended projection model and the cnBlog articles The CNB’s projection model gets a new plus and Model g3+ boduje (The g3+ model scores; in Czech only).
[2] In this appendix, we focus on the motivation for changing the model and on selected aspects of the change. Some minor – mostly technical – adjustments were made in addition to the modifications presented here. A more detailed account of the changes will be published in a forthcoming research paper.
[3] We use the Eurostat Main Industrial Groupings (MIG) classification.
[4] Up to now, the model has assumed that production factors are not substitutable and are used in fixed ratios.
[5] Here we use the OECD’s input-output tables.
[6] See Inflation Report II/2008.
[7] As well as changing selected steady-state growth parameters, we revised some parameters in the foreign block of the model. We backed these revisions by partially estimating the model on pre-pandemic data (covering 2000–2019).
[8] We have been using effective euro area indicators since 2006 (see Inflation Report III/2006). However, their definitions have changed over time.
[9] The weights of the individual countries in the calculation equal their shares in Czech exports.
[10] The EA17 aggregate lacks Luxembourg and Malta (due to limited data availability) and Croatia, which did not join the euro area until 2023 and is of negligible importance as far as the geographical structure of Czech exports is concerned.
[11] The differences are most visible at the level of the components of foreign industrial producer prices but are negligible in aggregate terms.