Working paper no.

31

ASSESSMENT OF POTENTIAL GROWTH OF

MONTENEGRO`S ECONOMY USING THE PRODUCTION
FUNCTION AND THE KALMAN FILTER METHODS

Maja Ivanović Nina Vujanović

Podgorica, 2022
PUBLISHED BY: Central Bank of Montenegro
Bulevar Svetog Petra Cetinjskog 6
81000 Podgorica
Telephone: +382 20 664 997, 664 269
Fax: +382 20 664 576

WEBSITE: http://www.cbcg.me

PREPARED BY:
Maja Ivanović, Advisor to Vice-Governor,
Centre for Macroeconomic and Financial Research and Forecasting
Nina Vujanović, Advisor to Vice-Governor,
Centre for Macroeconomic and Financial Research and Forecasting

The views and opinions expressed in this working paper are those of the authors
and do not necessarily reflect the official views and policies of the Central Bank of
Montenegro.

TRANSLATED BY: Department for Financial Stability, Research and Statistics

DESIGNED BY: Nikola Nikolić

Users of this publication are requested to make reference to the source of information whenever they
use data from this study.
 Contents

1. Introduction..............................................................................................................5
2. The concept of potential output and measurement techniques.............................6
3. Production function.................................................................................................8
4. Kalman filter.............................................................................................................16
5. Concluding remarks and policy recommendations..............................................22
References....................................................................................................................24
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

1. Introduction

Assessing potential economic growth (output) is an important input for macroeconomic analysis.
Potential output is the level of gross domestic product (GDP) that an economy could generate if it
operated at its full capacity. Given that potential growth cannot be directly measured and that it
is an abstract concept, it should be estimated based on available data and using various statistical
and econometric methods. Potential output estimation tools include various time series filters, as
well as theory-based methods, e.g. the production function method. In the context of countries in
transition, these methodological approaches have certain limitations and, therefore, they should
be used in a complementary manner.

This working paper presents the results of the assessment of potential output and output gap for
Montenegro in the period from 2006 to 2023. The results were obtained using two methodologi-
cal approaches: the production function method and the Kalman filter method. The estimate of
the potential output using the production function are to be obtained by calculating the trend of
individual production factors. This approach allows us to identify the contribution of individual
productions factors and provides a good basis for identifying supply-side bottlenecks that ham-
per economic growth and progress. The production approach is based on a theoretical framework
although its aplication is somewhat more difficult due to the absence of certain time series. On
the other hand, the Kalman filter is an algorithmic method that is based on an algorithm. The ob-
tained results show a strong negative impact of the crisis caused by the coronavirus pandemic on
potential output, thus indicating that the negative gap will continue in the next two years, but the
real GDP will be very close to the potential GDP. The results are as expected becase usually after
major crises, developing economies face challenges to stimulate potential growth. Empirical find-
ings in the literature indicate that deviations of GDP from its potential remain present for years
following a crisis and/or a certain shock that caused the crisis. The results of the assessment in
this paper suggest that real GDP could catch up with potential GDP in 2023.

This paper provides a simple framework for thinking about potential economic growth by look-
ing at different channels through which potential economic performance can be influenced. Ad-
equate assessment of potential growth and the information on the size of the GDP gap are particu-

5
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

larly important for economic policy makers in the context of the crisis caused by the coronavirus
pandemic. However, it should be noted that the estimates are subject to some uncertainty given
that the pandemic is ongoing and certainly affects potential output.

The paper starts with the introduction after which the second part explains the concept of poten-
tial output in macroeconomic analysis and the methods for its calculation. The third part explains
the production function method used to provide the calculation of potential output and output
gap. The fourth part of the paper presetnts the methodology for calculating the potential output
using the Kalman filter and the results of this approach. The paper ends with the conclusions.

2. The concept of potential output and measurement

techniques

Potential output is the level of GDP that an economy could generate if it operated at its full capac-
ity. In other words, potential output is defined as the level of production that an economy could
sustain when the production factors at the current level of technology are fully deployed. Full
utilisation of production factors, capital and labour is considered the maximum production that
does not cause inflationary pressures (Okun, 1962), i.e. it has a sustainable employment rate. The
production gap is the difference between the actual level of production and the level of production
with full use of the production factor. Therefore, a positive gap in production can be considered a
measure of the excess aggregate demand over aggregate supply. As a result, a positive production
gap can also serve as an indicator of inflationary pressures in the economy. In contrast, negative
production gap values ​​signal excess production capacities and declining inflationary pressures.

The concept of potential output is of great importance for short-term and long-term macroeco-
nomic analysis. In the short run, the output gap indicator provides a mean value of short-term
transient impacts (Giorno et al., 1995), while in the long run, potential output indicates a sustain-
able position of economic growth that creates non-inflationary pressures (Rõõm, 2001).

By estimating potential output and the corresponding output gap, we can identify any increase in
underlying imbalances or structural positions in macroeconomics (Giorno et al., 1995). Moreover,
the assessment of actual performance in relation to the potential provides guidelines for calibra-
tion of both fiscal and monetary policies. If a country`s trend of potential growth lags behind the
neighbouring countries (comparable economies), that would require structural reforms to im-
prove the business environment in that country. Namely, it is assumed that structural reforms are
the driving force of economic growth (Berg et al. 1999, Kolodko 2000, Fischer et al. 2000).

Given that potential output and the output gap are not directly visible and measurable, and hav-
ing in mind the difficulties related to data in transition countries, this paper analyses the most

6
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

appropriate techniques for estimating potential output recommended by the empirical literature.
What we need to keep in mind is that in the context of transition economies, the output gap can
be subject to large and constant structural adjustments and changes, often reflected in high struc-
tural unemployment rates, imported inflation and various market failures that continuously af-
fect and usually limit potential production (Kastrati et al., 2017). Therefore, the output gap in
transition countries such as Montenegro may reflect not only changes in actual output but also in
the potential output.

Different methods are used to estimate a potential output and the corresponding output gap.
Potential output estimation tools include univariate statistical filtering methods (Hodrick and
Prescott, 1997; Baxter and King, 1999; and Christiano Fitzgerald, 2003); followed by the state
space models that include the Kalman filter technique, and theory-based methods such as multi-
variate models (Blagrave, 2015) and the production function (Cobb - Douglas) approach.

All methods used to estimate potential output rely on detrending techniques, i.e. decomposing or
filtering the observed macroeconomic variable (actual output) into unnoticed variables (compo-
nents): trend, cycle and corresponding inconspicuous parameters, which should be subsequently
assessed.

– long term trend component

– cyclic component (deviations from the long-term trend)

The filtering procedure requires the extraction of a number of unnoticed parameters that are usu-
ally included in the noise (e.g. trend and cycle variance). When using univariate techniques, the
task is to extract many unnoticed variables and parameters from the noise using only one visible
variable (real GDP), which can be rather an ambitious endeavour.

For instance, the Hodrick-Prescott (HP) filter is a two-sided symmetric moving average fil-
ter, which assumes that trend and cycle components are not correlated and that the trend is a
“smooth” process without any sudden changes over time.

Univariate methods have the advantage of being simple and easy to interpret, which makes them
widely used. However, some univariate models (such as the HP filter) suffer from a number of
shortcomings, which usually makes them less accurate compared to multivariate methods. For
example, they suffer from endpoint issues, which implies excessive sensitivity to the latest data.
In essence, the HP filter assumes that the initial and final variations are mostly trend changes and
not cyclical variations. Endpoint limitation leads to frequent revisions of historical estimates of
the output gap, with the greatest degree of uncertainty reserved for the last and most needed data
point required to make appropriate economic policies.

7
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

Also, as often mentioned, many transition economies tend to be subject to structural changes that
cannot be identified by univariate filtering techniques such as the HP filter (Kastrati et al., 2017).
Matching the mean trend to real data, with the cycle below the real effect as much as above, may
not be appropriate in the context of transition. Namely, other methods are needed that would en-
able the matching of a different mean value. Multivariate methods have a characteristic advantage
in the fact that they use theoretical bases to determine unnoticed indicators, i.e. potential output
and output gap. Using the production function, we can observe changes in the supply-side perfor-
mance based on observed developments in the labour volume, capital, and total factor productiv-
ity. For instance, if an increase in the rate of capital growth is accompanied by an increase in the
trend of total factor productivity, we can indicate a certain improvement in performance on the
supply side. On the other hand, if we notice an increase in the capital growth rate while the trend
of total factor productivity remains stagnant, we will conclude that the supply side is functioning
inefficiently. Hence, it is often concluded that the production function is a useful tool for macro-
economic analysis and evaluation of government structural policies.

To summarize, each of the presented methods has its benefits and drawbacks, and the choice of
the appropriate method for countries in transition is usually conditioned by data. When it comes
to small and open economies such as Montenegro, this is not an easy task because the available
time series are short and not all data necessary for a complete analysis are available.

The production function method makes it possible to identify various factors that contribute to
the growth of potential output. However, this method can be used with data from a rather limited
period of time, with quite sensitive assumptions, especially in terms of technological progress and
capital valuation.

3. Production function

The production function in Montenegro is estimated based on data from the period 2006 - 2020.
Total production is expressed in gross domestic product. The production function approach pro-
vides the output that is closely correlated with technological progress, employment and the level
of capital in the country. So the gross domestic product is obtained from the interdependence of
the three variables: (1) labour input ( ); (2) capital used ( ); and (3) efficiency used by these fac-
tors, that is, total factor productivity ( ).

The relations between these three variables can be shown as follows:

(1)

This equation is called the Cobb-Douglas production function. Taking the logarithm on both
sides of the equation, we get the following expression:

8
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

(2)

Assuming the Cobb-Douglas production function, it is known that the elasticity of 𝛼 labour in-
put to GDP is equal to the share of labour in equilibrium. Therefore, using equation (2), it is easy
to calculate the change in total factor productivity (𝑎t), or the growth rate At, from the observed
change in 𝑦𝑡, 𝑙𝑡, 𝑘𝑡, which is the growth rate 𝑌𝑡, 𝐿𝑡, 𝐾𝑡. The same formulation of the production
function can be applied to the equation for potential GDP as follows:

(3)

Similarly, as in equation (2), the logarithm of both sides of the equation gives the following:

(4)

If we subtract equation (4) from (2), because the output gap is defined as the difference between
real GDP (𝑌𝑡) and the potential GDP (𝑌𝑡∗), we get:

(5)

(6)

where is defined as TFP noise or the so-called Sol residue, i.e. output reduced by
the weighted sum of labour and capital inputs.

Using this method, we will first calculate the total factor productivity (At) as a residual of the cal-
culation formula for growth for the period from 2006 to 2020, while we will make assumptions
about the growth rates of capital and labour for the projected period from 2021 to 2023. The basic
assumption is that the weights of labour participation in production and capital are 65% and 35%,
respectively, while the annual rate of depreciation is 5%. Assumptions were obtained based on the
findings of research by D'Auria et al. (2010) and Bosworth and Collins (2003) who indicate that
the labour force share (𝛼) is set at the EU average of 65 percent. Since data on the value of capital
in Montenegro is not available, an assessment was made on the basis of previously conducted em-
pirical research. The capital assessment was done in line with the methodology of accumulation
of gross fixed capital formation in the base year minus the the capital value decrease over time.
This methodology, known as the fixed inventory method, requires the following data: the value
of share capital in the initial year, a continuous time series for the value of investments (i.e. gross
fixed capital formation), and the extent to which capital decreases in value over time. The process
of capital accumulation is shown in the following formula:

, (7)

9
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

with 𝐼𝑡 indicating the level of real investment in each period and 𝛿 depreciation of Tier1 capital in
the previous period. We use a rate of 5 percent for capital depreciation (delta).

We use employment to approximate the labour invested in the production function. In the chart
below we can see that we had positive employment rates during the years of economic growth, as
a result of increased labour demand. During the boom period, traditional labour-intensive activi-
ties such as construction, mining, and retail trade were on the rise. However, while employment
reacted positively to economic growth, the unemployment rate was still very high, even in peri-
ods of rapid economic growth. A high unemployment rate indicates that there is a high structural
component that the business cycle is not able to neutralize. Although the increase in economic
growth, especially during the investment boom recorded from 2017 to 2019, led to a decrease in
the unemployment rate, this rate was still high compared to the neighbouring countries and the
EU Member States.

Graph 1 – Growth rates of employment, working age population, labour force participa-
tion, and unemployment

Source: MONSTAT

It is the high unemployment, even during a period of rapid growth, which indicates that the econ-
omy is functioning below its potential. The limiting factor in reducing unemployment in Monte-
negro is the so-called structural unemployment which is conditioned by the mismatch of supply
and demand of professionals for certain job profiles. Investment cycle in the period 2016 - 2019
did condition the demand for certain professionals in the field of construction, tourism, agricul-
ture, and energy but restrictions on the supply side were still present.

Relatively low labour force participation rates as well as high structural unemployment on the
supply side indicate that a high percentage of the population does not benefit from an active la-

10
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

bour market. If we look at the year before the COVID-19 pandemic, 2019, less than half of women
of active age were in the labour force, while the youth unemployment rate was about 25%, which
was significantly higher than in the European countries. In the pandemic year, the indicators fur-
ther deterirated.

Unfavourable demographic situation additionally contributes to the rigidity of the labour mar-
ket in Montenegro. Low birth rates and increased life expectancy lead to rapid population aging.
Montenegro has a birth rate below simple reproduction of 1.7 and an increase in life expectancy.
As shown in the graph 1, the share of the population aged 15+ has not changed since 2012.

Due to the lack of data, no correction was made in the labour input to change the actual working
hours spent. Also, for the same reason, the labour component cannot be differentiated in terms
of quality resulting from a certain level of education. However, using the method presented in the
work of Rõõm (2001), we will point out the differences in productivity in different sectors of the
economy (primary, secondary, and tertiary sector). Namely, we calculate an index that describes
the total change in productivity, generated by the redistribution of labour. The index has higher
values ​​when people move to more productive sectors, assuming that real earnings can be used as
an approximation of productivity. Also, it is assumed that relative productivity in the sectors is
constant over time. The index is calculated using the following formula:

(8)

where h is an index that describes the redistribution of labour into more productive sectors, w –
real earnings, – average real earnings, L – number of employees, T – number of periods, indices
t and j describe the period, and i describes the sector.

Graph 2 – Labour participation in primary, secondary and tertiary economies

Source: Authors` calculations

11
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

Looking at the graph above, we can notice a stable prevailing share of the labour force in the ter-
tiary sector of the economy. The share of labour in the secondary economy has a downward trend,
while we also notice low employment in the primary sector, which was on a mild donwtrend in
recent years.

By applying the mentioned assumptions, we calculate the total factor productivity for the period
2006 – 2020, i.e. the third component of the production function and the results are presented in
the graph below.

Graph 3 – Contribution to the growth of the economy based on the production function,
2006-2020

Source: Authors` calculations

Growth averaged at 2.8% in the period 2006 - 2019 (or 1.6% with 2020 included). In the observed
period, capital participated with 2.66 pp annually, labour with 0.92 pp, and total factor produc-
tivity (hereinafter: TFP) accounted for 0.75 pp per year. If we were to include the year 2020 in the
observed period, capital would contribute 2.58 pp annually, labour 0.39 pp, while TFP would ac-
count for -1.42 pp per year.

A negative TFP contribution is unusual. What is particularly surprising is that the TFP is not only
negative in crisis times, such as the global financial crisis and its reflection on the crisis processes
in Europe in 2012. The low historical growth of TFP can be explained by a long process of eco-
nomic transformation and slow structural reforms, which caused the technological obsolescence
of a part of the capital and the slow development of organisational (managerial and functional)
capabilities of economic entities. In her working paper, Vujanović (2020) uses micro data (at the
company level) to measure the level of total factor productivity (technological efficiency) in the
services sector and the manufacturing sector of Montenegro in the period 2010-2019. The results

12
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

indicate that the manufacturing sector, which is considered the sector with the greatest potential
for technological growth, did not record an increase in total factor productivity, and that the TFP
trend was relatively stable from 2013 to 2019. Only industries that did not require high technolo-
gies experienced growth of technological efficiency. Also, during the ten-year period, total factor
productivity of the services sector, which contributes a significant 40% of GDP (MONSTAT, 2019),
remained stagnant. The referenced paper points out that there is a lot of room for improving the
total factor productivity in Montenegro and that digitalisation, which is important for the im-
provement of technologies – is not widespread in the country. There is room for improving total
factor productivity, especially in the field of high-tech manufacturing and service industries that
require a high level of expertise such as IT, telecommunications, scientific research and develop-
ment, and the like.

The next step is to estimate the potential GDP. Specifically, the average inputs of labour (lt*), capi-
tal (kt*) and TFP trend (at*) are estimated from the observed data, thus the potential GDP (yt*)
is given in equation (4).

The key input for calculating potential output and output gap is potential employment. Accord-
ing to the production method, the idea is to find a level of employment that does not accelerate
inflation. One approach is to approximate it using the natural Non-Accelerating Inflation Rate of
Unemoployment (NAIRU). In the production analysis, the labour market is concentrated on the
unemployment rate, i.e. the way in which the unemployment rate is developing depending on the
phase where we are in the business cycle. When the economy is at its “normal” or long-term level
of production (i.e. the potential GDP), there is an unemployment rate known as the “natural” un-
employment rate. This unemployment consists of frictional and structural unemployment, but
there is no cyclical unemployment associated with business cycles. The Okun`s law implies pro-
duction growth but also the structural unemployment rate due to slow adjustment of the labour
market. Namely, it is known that the labour market is always behind the overall business cycle,
i.e. that there is this persistence in explaining the labour market reaction.

Empirical works approximate potential employment using the unemployment level trend or long-
term unemployment rate. The basic assumption in the empirical literature is that technological
change causes a decrease in the skills of people who are not currently working, i.e. the longer a
person remains unemployed, the more their qualification decreases (Rõõm, 2001; Kawamoto et
al., 2017). Therefore, potential employment is calculated using the long-term unemployment rate.

Labour market projections are focused on the Okun-type unemployment rate, total labour re-
sources determined by a given working age population (population aged 15 and over) and the
participation rate of the working age population. The last two variables are given exogenously.
The data from the MONSTAT Labour Force Survey were used for the period 2006 - 2020. For the
projection period 2021 – 2023, projections of the UN population1 were used, as based on probabil-

1 UN 2019 Revision of World Population Prospects

13
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

istic projections of medium fertility and life expectancy at birth. These probabilistic projections of
mean fertility and life expectancy at birth were conducted using the Bayesian hierarchical model
(UN, 2019).

As for the unemployment rate, it depends on the NAIRU that we estimated by applying the HP
filter to the long-term unemployment rate and the unemployment rate projection for the period
2021–2023, in line with economic growth projections for the observed period. Given the time
period of observation and due to the lack of sufficiently long time series, in order to verify the
estimated NAIRU, we took a long-term average unemployment rate for Montenegro that approxi-
mates the unemployment rate in the steady state.

Finally, when we have the estimated unemployment rate, an estimate of the participation of the
working age population and an assumption of the expected available labour force, we can calcu-
late the potential employment.

The working age population is multiplied by the participation rate (as all those who are part of the
working age population are not part of the labour force), which is further multiplied by one minus
NAIRU (as not all those in the labour force are employed).

Potential employment = potential labour force*(1-(NAIRU/100))

Potential labour force = working age population * HP filter for participation of working age
population

It is assumed that capital is at its potential level.

Graph 4 – Capital trends from 2006 to 2023

Source: Authors` calculations

14
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

A strong investment activity in the period from 2016 to 2019 in transport, tourism, energy and
agriculture further raised the value of capital, which will affect the growth of potential rates of
economic activity. The investment in the first section of the highway gave a special impetus to
capital growth.

We will calculate the potential level of total factor productivity using the HP filter to find its trend
level. The methodology for estimating the growth of the TFP trend (changes compared to the pre-
vious year 𝑎𝑡∗) includes the following: first, the calculation of the year-on-year growth rate of the
real TFP (changes 𝑎t compared to the previous year), and then applying the HP filter (𝜆 = 1600)
on the growth rate of the real TFP.

An alternative approach is to assess technological progress using a simple trend or to approxi-

mate it through measures such as education, research and development costs. However, the lack
of available data limits us from using this approach.

After estimating potential GDP, we will calculate the output gap that represents the difference be-
tween the potential GDP (yt*) and the real GDP (yt).

Potential gross domestic product, estimated by production function, and realised gross domestic
product, are shown in Graph 5. The GDP series for the period 2021–2023 correspond to the fore-
casts of the Central Bank of Montenegro and the European Commission.

Graph 5 – Potential and real GDP (left) and output gap (right)

Source: Authors` calculations

The results of the production function indicate that Montenegro's GDP in the year of gaining inde-
pendence was below its potential. The rapid economic growth from 2006 to 2008, supported by a
strong inflow of foreign direct investments, caused a positive output gap, i.e. a positive difference
between the real and potential GDP. With the onset of the global economic crisis and the negative
GDP rate in 2009 (-5.8%), the production gap was declining, and in the period from 2010 to 2014

15
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

it entered the negative zone. From 2014 to 2019, an exponential growth of the production gap was
recorded. At that time, the positive GDP gap (i.e. the positive difference between real and potential
GDP) was at the highest level. However, with the onset of the COVID-19 epidemic, the production
gap abruptly narrowed.

Due to the simultaneous significant decline in real GDP in 2020 on one hand, and the growth of
potential GDP on the other hand, the highest level of negative output gap was recorded in 2020. In
addition to the positive projected GDP rates for the period 2021 - 2023, the negative GDP gap will
remain both in 2021 and 2022. The estimate indicates that reaching the level of potential produc-
tion or potential level of GDP can be expected in 2023.

We can conclude that the practical application of the production function method requires the
creation of certain assumptions such as constant returns on the labour to capital ratio, and they
are often stated as a significant limitation of this approach. These assumptions limit the elasticity
of production in relation to labour and capital to values ​​between zero and one, that is, their sum is
equal to one. Also, it is assumed that technological progress changes unhindered over time, which
may not be applicable in transition countries that are going through numerous structural chang-
es (Sramkova et al., 2010). Finally, the production function approach continues to use statistical
filters, such as the HP filter, to assess the trend in total factor productivity, which in turn carries
with it the shortcomings of the measures we previously highlighted.

4. Kalman filter

In this section, we measure potential output by an alternative method - the Kalman filter method,
which belongs to the state space model. Unlike the previous model, which is based exclusively
on economic theory, the Kalman filter is more an empirical-statistical method, but with a clear
theoretical approach in its setting (Oksanen, 2018). In a way, this method is more accessible for
estimating potential output because there is no need to approximate the missing series, such as
the capital level. However, unlike the production method, estimation of total factor productivity,
an important component for understanding technological advancement, is not possible by using
the Kalman filter method.

The Kalman filter generates a series of potential outputs by eliminating the noise in the series it fil-
ters in a way that minimizes the mean square errors of the estimated parameters (Kleeman, 1996).
For the purpose of filtering the data, an algorithm is created that can consist of two sets of equa-
tions that define the algorithm. The first set of equations (the so-called measurement equation)
expresses the GDP series as a function of potential GDP and the GDP gap. The second set of equa-
tions (the so-called state equations) define the process of data generation that ultimately forms the
potential GDP and the GDP gap. Given that the potential output and output gap is measured on
the basis of one known series (GDP) instead of multiple known series, this is an univariate model.

16
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

There are several ways to formulate the Kalman filter algorithm that generates two unknown ele-
ments. However, before formulating the algorithm itself, it is useful to analyse the nature of the
time series (of the GDP) to be broken down into potential output and the output gap. The nature
of the time series is assessed through the application of stationarity tests. If the tests indicate
that the GDP series is stationary, the values of its distribution (expected value and variance) do
not change, i.e. they are constant during the analysed period. The data series arbitrates around a
certain value and has no defined trend. Otherwise, the series is non-stationary and has a certain
defined trend.

The Augmented Dickey-Fuller (ADF) stationarity test indicates that the GDP series is a non-sta-
tionary series with a unit root2 (Table 1). The null hypothesis of the ADF test is that the series is
non-stationary, i.e. that it has a unit root. The impossibility of rejecting this hypothesis (p val-
ue> 0.05) confirms that the series is non-stationary. If the ADF test result rejects the null hy-
pothesis (p-values <0.05), the GDP series is stationed. The test indicates that the GDP series is
non-stationary (p value = 0.173). On the other hand, by differentiating the series, a stable station-
ary series is obtained (p-value = 0.00). Logically, the GDP series has a certain trend direction.

Table 1 – Stationary tests

ADF test applied on: P probability value Conclusion

Logarithm (GDP) 0.173 Series of GDP is non-stationary
Diferentiated logarithm (GDP) 0.00 Series of diferentiated GDP is stationary

This information is useful because the potential GDP series has the same features (stationarity
status) as the real GDP series. This means that the potential GDP series is also a non-stationary
series with a unit root.

After checking the stationarity of the series, the Kalman filter is generated by a series based on
the model setting (algorithm), and the initial parametres of the model based on which the filter-
ing of GDP data starts.

Model setting (algorithm)

When creating an algorithm, the equation of state should be formulated to define the potential
output as a series that is non-stationary and with a certain trend. Therefore, we define a given
equation of state as a autoregressive process of the first order with a constant.

Based on the above analysis, but also theory and practice, we opted for the following algorithm:

2 The series has a unit root if the series needs to be differentiated in order for it to be stationed. In other words, if the GDPt series (t-year) is non-stationary
and has a unit root, ∆ GDPt (GDPt - GDPt-1 ) is the stationary series.

17
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

The algorithm expressed by equations 9-11 is recursive and can be expressed in several ways (see
Rummel, 2015). The equation 9 (measurements) expresses GDP ( ) as a function of potential
GDP ( ) and the GDP gap ( ). Hence, this equation breaks down the known GDP series into
two unknown elements. The GDP series ( ) is logarithmic. Equation (10) is the first equation of
state that defines potential GDP as a autoregressive first-order process (random walk process with
constant ). and are random equation of state errors (10) and (11), respectively, which are
not mutually correlated. Also, these standard errors have a normal distribution with variance
and , respectively. The parameters , , are unknown parameters (regression coefficients).

Equation (11) expresses the GDP gap as a second-order autoregressive process. The recursiveness
of this process is the result of both the potential GDP ( ) and the GDP gap ( ) expressed as a
function of past values ( , respectively). This means that the Kalman filter filters
new values based on past values at time t, so that the GDP gap and potential GDP series are esti-
mated in a number of consecutive steps, and after a large number of iterations, a potential GDP
series is created. Thus, this series creation process is updated each time a new observation is en-
tered into the algorithm.

Equations (9) to (11), along with their parametres, define the Kalman filter that filters the values of
new series (potential output) based on unknown variables. In order for the algorithm of equations
to be able to generate two series, assumptions about the initial values of the parametres , , ,
and are required. Assumptions are given based on the theory (Kuttner, 1994), empirical
analysis (Rummel, 2015), but also as an arbitrary process (the trial and error method). Like Kas-
trati et al. (2018), the parametre is a parametre that we equate with the average value of GDP,
while the values of parametres and are taken from the theoretical framework (Kuttner,
1994) according to which the parametre of the first autoregressive ( ) is 1.4 and the second ( )
is -0.5, thus we meet the stationarity criterion according to which the sum of these two param-
eters is below 1. The other parametres were chosen by the trial and error method, but in a way
that maximizes the maximum probability function. The initial parametres are shown in Table 2.

Table 2 – Initial algorithm parametres

Parametres
Values 7.9 1.4 -0.5 -10.3 -10

Source: Author's calculations

18
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

The results of the model (filtering) may differ, depending on the initial parameters but also on the
algorithm settings, and this is the main drawback of the Kalman filter. Also, the output gap ob-
tained by this method is not necessarily correlated with the level of inflation (Giorno et al., 1995).
However, the potential benefits of this method are significant. The Kalman filter is suitable for fil-
tering the series that have missing values through adjustements of the algorithm (Koopman and
Ooms, 2002). In this sense, the Kalman filter can be used to predict future values and not just the
missing ones. Also, some additional non-measurable variables, such as the natural level of unem-
ployment, do not need to be used to estimate the potential GDP. In this particular case, there was no
need to approximate the level of capital, which was missing from the production function model.

The model results

For the purpose of estimating the series of the potential GDP, the series of GDP at current prices3
was used. Potential gross domestic product, estimated using algorithm 9-11, is presented in Graph
6 on the left-hand side, while estimation results are presented in the right-hand side of the Graph
6. The graph also shows the realised gross domestic product (at constant prices) during the ana-
lysed period. For the purposes of estimating the potential output, the available annual series of
GDP expressed at constant prices in the period 2000–2020 was used, as well as the GDP forecasts
of the Central Bank and the European Commission for the period 2021-2023. The Kalman filter re-
sults from 2006 onwards are presented for the purpose of comparison with the previous method.

Graph 6 – Potential output and real GDP (left-hand) and output gap (right-hand), the
Kalman filter method

Source: Authors' calculations

The Kalman filter results give the expected result. The level of GDP is above the potential in the
period from Montenegrin independence until 2010 when they are practically equal, which is the
result of a large decline in GDP in 2009 (-5.8%). Namely, the positive GDP gap grows until the out-

3 In the earlier Central Bank of Montenegro working paper (Ivanović et al., 2021), we used a series of GDP at constant prices.

19
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

break of the global financial crisis in 2008, and then declines. The high level of positive output gap
can be explained by the high investment cycle that preceded the financial crisis.

In the period from 2010 to 2016, the Montenegrin economy produced below its potential, which
indicates that the effects of the global economic crisis were long-term indeed. Due to the high
growth rate in the period 2017-2019, which was above 4%, Montenegro again entered the zone of
positive GDP gap. The explanation for the short-term return of the positive GDP gap can be found
in employment growth, growth in consumption supported by a good tourist season, as well as
more favourable credit conditions that encouraged investments. The construction and trade sec-
tors also recorded significant growth.

However, the pandemic brought about a quick shift from a slightly positive to a very negative GDP
gap. In 2020, both potential and real GDP suffered a sharp decline, with the level of real GDP fall-
ing significantly (-15.3%) due to a decline in the services sector, primarily tourism, which is an
experience shared by many countries which base their production on services. Montenegro re-
corded the most negative GDP gap in 2020 since its independence.

The results of the Kalman filter indicate that the gap is also significant in 2021, for the reason that
the growth of the potential GDP is higher than the growth of the real GDP. However, when inter-
preting the results of the potential growth assessment, it is necessary to keep in mind a certain
degree of uncertainty in the sense that the pandemic is still ongoing and that we use short-term
projections of the European Commission, which affects the accuracy of the assessment. A similar
interpretation is given in the ECB Economic Bulletin (2020). Namely, the authors of this bulletin
expect revisions of potential output for the euro area countries once the size and consequences
of the crisis have become clearer. The results indicate that in 2022 and 2023, the negative gap be-
tween potential and real GDP will be reduced, which means that in the coming period the econo-
my will recover more strongly and function closer to its maximum potential. Thus, just as in the
case of the global financial crisis, the crisis caused by the COVID-19 pandemic has a strong nega-
tive impact on both potential and real GDP, as well as on the negative GDP gap. The duration of the
negative GDP gap can be attributed to the high level of unemployment. Specifically, the number of
unemployed in December 2021 is higher by 20.8% compared to December 2020, while compared
to December 2019 it is higher by 52.6%.

GDP gap: a comparison of the two methods

A comparison of the output gap estimated by using the Kalman filter and the production method
indicates clear similarities, especially when observing its trend. Both series point to a positive
GDP gap in the period preceding the global financial crisis and just before the crisis caused by
the COVID-19 pandemic. However, the positive and negative output gaps are more pronounced
when estimated by the Kalman filter method, with the exception of the two-year period before the
pandemic. Certain smaller quantitative differences between the methods used are still expected,
bearing in mind that these are univariate and multivariate methods.

20
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

It is important to stress that both models indicate that the effects of the crisis, both the global fi-
nancial crisis and the COVID-19 crisis, are pronounced and long-lasting, which is the conclusion
given in other working papers (Fontanari et al., 2019). Same as Mc Morrow et al. (2021) we can
also conclude that the negative GDP gap is shorter after the crisis caused by the COVID-19 pan-
demic than after the global financial crisis. Mc Marrow et al. (2021) conclude that the potential
level of GDP will return to the pre-pandemic level faster and that long-term effects are possible
only if a part of the labour force withdraws from the labour market, and with lower investments
and capital obsolescence and/or bankruptcies in the coming period. Assuming that the coronavi-
rus pandemic will be fully controlled in the analysed period, our paper also indicates that poten-
tial GDP and real GDP will be almost equal by 2023, in the absence of the aforementioned negative
economic circumstances.

Graph 7 – Output gap: Production mathod and the Kalman filter method

Source: Authors' calculations

We can conclude that both methods give reliable and very logical results although each has its
own drawbacks. The disadvantages of the Kalman filter are its sensitivity to assumptions about
the initial parametres, and not so many links with the theoretical economic framework. On the
other hand, the production method is based on economic theory, which allows the estimation
of total factor productivity, which is important for understanding technological advancements.
Nevertheless, this method uses the Cobb-Douglas production function, which is based on the as-
sumptions of constant returns, technological progress that gradually changes over time, and ap-
plies the HP filter, which indicates the shortcomings of this methodological approach.

Given the different nature of these two methods, it is encouraging that they give very similar re-
sults which, points to high validity of their application in the economic policies of Montenegro.
Moreover, both methods support the conclusions of recent studies on different durations of the
negative GDP gap after the two economic crises in 2009 and 2020.

21
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

5. Concluding remarks and policy recommendations

The assessment of potential output and output gap has been an unavoidable subject of central
bank macroeconomic analyses. Estimates of these two indicators are primarily used as a measure
of the current state of the business cycle and as guidelines for determining fiscal, monetary and
macroprudential policies, so it is important to have relevant estimates of potential growth and
output gaps. However, as these are variables that cannot be measured directly, their estimates are
subject to some degree of uncertainty.

This working paper presents the results of estimating the potential growth and GDP gap for Mon-
tenegro in the period from 2006 to 2023 that have been obtained by using two different methodo-
logical approaches: the production function method and the Kalman filter method.

The results indicate that there are certain minor quantitative differences in the estimates depend-
ing on the applied method. With the Kalman filter method, the GDP gaps (both positive and
negative) are slightly larger than those obtained by the production method. It is not uncommon
for the application of different methodologies to assess the potential growth and GDP gap to re-
sult in different results of the analysed indicators (Jovičić, 2019). However, both methods point to
similar trends: a positive GDP gap before the global financial crisis and the crisis caused by the
COVID-19 pandemic, and a negative GDP gap after these negative macroeconomic shocks. Simi-
larly, both methods suggest that real GDP takes longer to catch up with the potential GDP after the
global financial crisis than after the COVID-19 crisis although the second data should be taken
with caution as it is based on GDP forecasts. Other recent publications came to similar conclusion.
Therefore, we can conclude that regardless of the different nature of these methods, a consistent
and very credible result has been obtained and that both methods are useful for economic policy
making.

The results of the production method analysis show that the contribution of factor productivity
is very small, and even negative in some periods. This low level of factor productivity indicates
the existence of structural issues in our economy that prevent more efficient use of the existing
resources. Earlier studies showed a very weak upward trend in total factor productivity in the
manufacturing and service industries of Montenegro, which indicates that the fourth technologi-
cal revolution has not gained momemntum (Vujanović, 2020). Therefore, it is often concluded that
the output gap in Montenegro, same as in other transition countries, largely reflects structural
changes and to a lesser extent the cyclical component (Kastrati et al., 2017). In this context, poli-
cies aimed at identifying and addressing these issues could have a very positive impact on long-
term economic growth.

In order to increase potential growth, it is necessary to implement structural reforms that must
aim at improving business conditions, product and labour markets, and strengthening the capac-
ity for innovations. It is in the empirical literature that increasing attention is being paid to struc-

22
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

tural reforms in helping to achieve stronger and more sustainable potential growth. Structural
reforms can affect the growth of production through different channels: productivity, capital ac-
cumulation and employment. The results of the IMF's 2015 paper suggest that sustainable and
well-coordinated structural reforms usually coincide with periods of improved productivity and
macroeconomic performance. They also confirm that a wide range of structural reforms, includ-
ing banking and financial sector reforms, labour market issues, infrastructure and legal system
reforms, and the business climate have impact on macroeconomic performance. For a transition
economy such as Montenegro, which still faces the issue of obsolete technologies, policies that
promote research and development, innovation and training of employees in necessary skills for
using latest technologies are important. Productivity can be increased through technological de-
velopment, especially in manufacturing industry, but also in the services sector which already
enjoys the benefits of digitalisation.

23
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

References

1. Baxter, M. and R. G. King (1999), “Measuring Business Cycles: Approximate Band-Pass fil-
ters for Economic Time Series”, The Review of Economics and Statistics, Vol. 81, No. 4, pp.
575- 593.
2. Berg, A., Borensztein, E., Sahay, R. and J. Zettelmeyer (1999), “ The Evolution of Output in
Transition Economies: Explaining the Differences”, IMF Working Paper No 73
3. Blagrave, P., Garcia-Saltos, R., Laxton, D. and Z. Fan (2015), “A Simple Multivariate Filter
for Estimating Potential Output”, IMF Working Paper, WP/15/79
4. Bosworth, B. P., and S.M. Collins (2003), “The Empirics of Growth: An Update,” Brookings
Papers on Economic Activity”, Vol. 34(2003–2), pp.113–206.
5. Christiano, L. J. and T. J. Fitzgerald (2003), “The band pass filter”, International Economic
Review, Vol. 44, No. 2, pp. 435-65.
6. D’Auria, F, C. Denis, K. Havik, K. McMorrow, C. Planas, R. Raciborski, W. Roger and A.
Rossi (2010),“The production function methodology for calculating potential growth rates
and output gaps”, European Commission Economic Papers No 420, July.
7. ECB (2020), The impact of COVID-19 on potential output in the euro area, ECB Economic
Bulletin, Issue 7/2020, available at: https://www.ecb.europa.eu/pub/economic-bulletin/ar-
ticles/2020/html/ecb.ebart202007_01~ef0a77a516.hr.html
8. Fischer, S. and R. Sahay (2000), “The Transition Economies after Ten Years”, IMF Working
Paper No 30
9. Fontanari, C., Palumbo, A., & Salvatori, C. (2020). Potential output in theory and practice:
a revision and update of Okun's original method. Structural Change and Economic Dynam-
ics, 54, 247-266.
10. Giorno, C, Richardson, P., Roseveare, D. and P. van den Noord (1995),“Estimating Poten-
tial Output, Output Gaps and Structural Budget Balances, ”, OECD Working Paper no. 152
11. Hodrick, R. J. and E. C. Prescott (1997), ‘Postwar US business cycles: an empirical investi-
gation”, Journal of Money, Credit and Banking, Vol. 29, No. 1, pp.1-16.
12. International Monetary Fund (2015): “Structural Reforms and Macroeconomic Perfor-
mance: Initial Considerations for the Fund”, November 2015, available at: http://www.imf.
org/external/np/pp/eng/2015/101315.pdf
13. Ivanović M., Vujanović, N., Kilibarda, M., Vlahović, A. (2021) Analiza uticaja pandemije
COVID 19 na ekonomiju i bankarski sistem Crne Gore, radna studija broj. 29, Centralna
banka Crne Gore.
14. Jovičić, G. (2017), “Procjena potencijalnog rasta i jaza BDP-a u Hrvatskoj”, Pregledi, P-38,
Hrvatska narodna banka, Zagreb
15. Kastrati, A. (2015). The relationship between output gap and excess liquidity in European
Transition Economies, Doctoral dissertation, Staffordshire University.

24
 ASSESSMENT OF POTENTIAL GROWTH OF MONTENEGRO`S ECONOMY
USING THE PRODUCTION FUNCTION AND THE KALMAN FILTER METHODS

16. Kastrati, A., Pugh, G. and V. Toci (2017), “Output Gap in Transition Economies Using Un-
observed Component Method: The Case of Czech Republic, Estonia and Kosovo”, Economic
Thought and Practice Vol. XXVI N. 2., pp. 477-500.
17. Kawamoto, T., Ozaki, T., Kato, N. and K. Maehashi (2017), “Methodology for Estimating
Output Gap and Potential Growth Rate: An Update ”, Bank of Japan, Research and Statis-
tics Department, Reports and Researh Paper, May 2017.
18. Kolodko, G. W. (2000), “Globalization and Catching-Up: From Recession to Growth in
Transition Economies”, IMF Working Paper No 100.
19. Koopman, S. J., & Ooms, M. (2002, August). Periodic unobserved component time se-
ries models: estimation and forecasting with applications. In 57th European Meeting of the
Econometric Society (ESEM2002) held in Venice.
20. Kuttner, K. N. (1994). Estimating potential output as a latent variable. Journal of business &
economic statistics, 12(3), 361-368.
21. Lindsay, K. (1996). Understanding and applying kalman filtering. In Proceedings of the Sec-
ond Workshop on Perceptive Systems, Curtin University of Technology, Perth Western Aus-
tralia (25–26 January 1996).
22. Mc Morrow, K., Blondeau, F., D’Auria, F., Döhring B., Hristov, A., Maier, C., Thum-Thysen,
A. (2021) Output gaps, potential output and the COVID-19 crisis: Policymaking under un-
certainty, accessed on https://voxeu.org/article/output-gaps-potential-output-and-covid-
19-crisis
23. Oksanen, H. (2018). New Output Gap Estimates for Assessing Fiscal Policy with Lessons for
Euro Area Reform (No. 7287). CESifo Working Paper.
24. Okun, A.M. (1962), “Potential GNP: Its Measurement and Significance”, in Proceedings of
the Business and Economic Statistics Section, pp. 98-104 (Washington: American Statistical
Association).
25. Rõõm, M. (2001), “Potential output estimates for Central and East European countries us-
ing production function method”, Working Papers of Eesti Pank. No 2, 2001.
26. Rummel, O. (2015). Estimating the output gap for Kenya: a practical guide to some states-
pace and Kalman filter trend-cycle decompositions. Centre for Central Banking Studies,
Bank of England.
27. Sramkova, L., Kobilicova, M. and A. Krajcir (2010), “Output gap and NAIRU estimates
within state-space framework: an application to Slovakia”, Financial Policy Institute, The
Ministry of Finance of the Slovak Republic, Economic Analysis 16.
28. Vujanović, N. (2020) Tehnološka efikasnost prerađivačkog i uslužnog sektora Crne Gore,
radna studija broj. 28, Centralna banka Crne Gore.

25