Reading Time: 15 minutes

1. Introduction

Real gross domestic product (real GDP) is the broadest summary measure of economic activity. It captures the value of goods and services produced in an economy after adjusting for inflation, allowing comparisons over time that are not distorted by changes in prices. For a small, open, upper-middle-income economy like Serbia, the path of real GDP since 2000 tells a story of post-transition catch-up, exposure to global and regional crises, and gradual integration into European production and financial networks.

The analysis that follows draws on quarterly data for Serbia’s real GDP from 2000Q1 to 2025Q2, measured as chain-linked volumes with 2020 = 100, consistent with Eurostat’s national accounts framework for real activity. Chain-linked volume indices are designed to track real changes in output over time, rebasing the series to keep price structures up to date and ensure comparability across countries and across long time spans. The graphs used here are based on Eurostat source and have been processed using logarithmic transformations, differencing and STL (Seasonal-Trend decomposition using Loess) to isolate trend, seasonal and irregular components.

Over the full quarter-century, the broad picture is one of moderate but uneven convergence: real GDP moves from relatively low levels at the start of the 2000s towards a significantly higher plateau, interrupted by several pronounced downturns. The global financial crisis in 2008–2009, the euro-area turmoil and domestic fiscal consolidation in the early 2010s, the severe floods of 2014 and the COVID-19 shock in 2020 all leave distinct marks on the trajectory of output.

The most recent part of the sample, covering roughly 2022–2025, suggests that Serbia has moved into a phase of relatively steady, though not spectacular, expansion. Official releases point to real annual growth rates in the low-to-mid single digits over 2022–2024, with quarterly movements around one per cent in many recent quarters. Against this backdrop, the graphs show real GDP fluctuating above its pre-pandemic trend, with volatility that is more muted than during previous crises but still sensitive to swings in external demand, energy prices and domestic policy changes.

Seasonality is present but not dominant. Unlike highly tourism-specialised economies where summer peaks and winter troughs dominate the quarterly profile, Serbia’s GDP is driven more by industry, services and public expenditure; tourism accounts for a modest share of value added. The seasonal pattern is therefore noticeable but not overwhelming: some quarters, typically the final quarter of the year, tend to be stronger than others, yet these intra-year swings are small relative to the long-run rise in output and the major crisis-related shocks.

2. Description of the time series and components

2.1 Overall patterns in levels and adjusted series

Figure 1 presents the original and seasonally adjusted log of real GDP for Serbia over 2000Q1–2025Q2. Working in logarithms turns proportional changes into roughly constant step sizes, so that a one-per-cent increase has the same vertical height in 2001 and in 2024. This makes it easier to compare growth across phases where the level of GDP differs substantially.

The trajectory in Figure 1 can be read in several broad episodes. The early 2000s are characterised by post-transition normalisation and reconstruction, with real GDP rising fairly rapidly from a low base. The log-series slopes upward, signalling positive growth, though the path is far from perfectly smooth. This catch-up phase is abruptly interrupted around 2008–2009: both the non-adjusted and seasonally adjusted series show a sharp downward movement, corresponding to the global financial crisis and the associated collapse in external demand and capital inflows.

The subsequent decade is best described as a period of slower and more uneven progress. The trend continues to rise, but at a gentler gradient, and is punctuated by episodes of stagnation or mild contraction. The graphs around 2012–2014 suggest a stop-start pattern consistent with the euro-area sovereign-debt crisis, domestic fiscal consolidation and the impact of the severe floods that hit Serbia in 2014. These shocks appear as temporary dips or flat segments in the adjusted series, indicating interruptions to growth rather than a persistent reversal.

From the mid-2010s onwards, the seasonally adjusted series steepens again, reflecting a phase of stronger real growth supported by investment, export expansion and gradually improving macroeconomic stability. This recovery runs into the COVID-19 shock in 2020. In Figure 1, the pandemic shows up as one of the most pronounced short-run contractions in the sample, followed by a relatively rapid rebound over 2021–2022. National statistics confirm a deep drop in output in the second quarter of 2020, followed by strong year-on-year growth in late 2020 and 2021 as restrictions eased and global demand recovered.

By the final part of the sample, covering 2023–2025, the seasonally adjusted path of log real GDP is above its pre-COVID level and trending upwards, but the gradient is more moderate than in the early post-transition years. Short-term fluctuations remain visible, with occasional quarterly setbacks, reflecting the effects of high inflation, tighter monetary policy and elevated uncertainty linked to energy markets and geopolitics. Overall, Figure 1 conveys a story of long-run convergence interrupted by crises, with seasonal adjustment helping to strip out regular within-year fluctuations so that these cyclical and structural movements are easier to see.

2.2 Main features – trend, seasonal and irregular components

Figure 2 dissects the log-level series into three components using STL: a smooth trend-cycle, a recurring seasonal pattern and an irregular or residual term. The trend component captures the underlying direction of the economy; the seasonal component shows how each quarter typically deviates from that trend; and the irregular captures short-lived shocks and noise.

The trend line in Figure 2 confirms the qualitative reading from Figure 1. It rises steeply in the first decade, reflecting robust catch-up from the low base of the early 2000s. Around 2008–2009, the trend dips or flattens, reflecting the depth of the global downturn. In the subsequent years, the trend resumes its upward path but with visible pauses around 2012–2014, consistent with domestic and regional headwinds. After 2015, the trend becomes smoother and more firmly positive, indicating a more stable expansion phase. The COVID-19 period again stands out: the trend temporarily flattens or even declines, before regaining momentum as output recovers. By 2025Q2, the trend component lies significantly above its level in the mid-2000s, underscoring how much the economy has grown despite multiple crises.

The seasonal component is clearly present but relatively modest in amplitude. In the log-transformed series, seasonal swings appear as small, regular oscillations around zero, repeating with a period of four quarters. The pattern suggests that certain quarters, often the fourth quarter, and sometimes the second or third, tend to be slightly above trend, while others, typically the first quarter, fall slightly below. This is consistent with reporting schedules, budget cycles and weather-related factors that influence construction and agriculture. Importantly, the seasonal amplitude does not appear to increase dramatically over time, which supports the choice of working in logs: proportional seasonal effects remain fairly stable across the sample.

The irregular component captures what is left after removing trend and seasonality. For most of the period, it fluctuates in a relatively narrow band, indicating that short-run noise is present but not overwhelming. However, at several points the residual exhibits clear spikes or clusters of large movements, corresponding to major shocks. The quarters associated with the global financial crisis, the 2014 floods and the onset of the COVID-19 pandemic show up as pronounced outliers, where the observed GDP departs sharply from what would be expected given the usual trend and seasonal pattern. These episodes underscore how extraordinary events can temporarily dominate normal cyclical and seasonal dynamics.

2.3 Seasonal diagnostics and detailed patterns

Figures 3–8 provide a closer look at the seasonal and dynamic properties of Serbia’s real GDP, using both the log-level series and its first-differenced counterpart. Seasonal plots and subseries plots display the typical trajectory within a calendar year, while lag plots reveal how current values relate to past values.

In the seasonal plots of the log-level series (Figure 3), each year is broken down into its four quarters and overlaid, allowing the reader to see whether there is a consistent pattern such as “weak first quarter, strong fourth quarter” or “summer peaks”. For Serbia’s real GDP, the seasonal plot suggests a recurring but relatively mild pattern: output often dips slightly in the first quarter, then recovers through the middle of the year, with some strength towards year-end. The spread of lines across years is not extremely wide, which reinforces the impression that seasonality is present but limited in magnitude.

The seasonal subseries plot for the log level (Figure 4) flips the perspective, following each specific quarter, say, all Q1 observations, from 2000 to 2025. This view helps to detect gradual changes in the seasonal profile. For example, if Q3 values were systematically rising faster than the trend over time, one might infer intensifying summer tourism or export seasonality. In Serbia’s case, the subseries plots indicate that while all quarters have trended upwards over the long run, the relative ranking of quarters has not changed dramatically. Any shifts in seasonal strength appear incremental rather than abrupt.

Lag plots for the log-level series (Figure 5) chart each quarter’s value against its immediate predecessor. For a persistent macro series like GDP, one expects a strong positive relationship: high output in one quarter is usually followed by high output in the next, and the scatter of points in Figure 5 aligns broadly along an upward-sloping cloud. This visual impression is consistent with high autocorrelation, a hallmark of macroeconomic time series where shocks propagate gradually rather than being instantly absorbed.

The second set of diagnostics, Figures 6–8, applies the same tools to the first-differenced log series, which can be interpreted loosely as a measure of quarter-on-quarter growth in real GDP. Differencing removes the long-run trend, leaving a series centred around zero.

In the seasonal plots of the differenced series (Figure 6), the lines for different years are more tightly clustered around the horizontal axis, confirming that systematic seasonal effects have been largely filtered out. Remaining within-year patterns are weaker and more irregular, though in some cases particular quarters still show a tendency to deliver slightly higher or lower growth.

The subseries plots for the differenced series (Figure 7) highlight how the distribution of growth outcomes has evolved over time. Here, the crisis periods stand out clearly: during 2009 and 2020, several quarters for each season show deep negative bars, while the years of recovery display unusually strong positive values. In more tranquil periods, the subseries lines hover closer to zero, reflecting modest growth rates and fewer extreme movements.

Lag plots for the differenced series (Figure 8) are more diffuse than their level counterparts, as expected. Once trend and seasonality are removed, the quarter-to-quarter correlation weakens; shocks are less persistent and the series behaves more like a stationary process fluctuating around a stable mean. The cloud of points in Figure 10 still tilts upward, suggesting some short-run momentum, strong growth in one quarter often spills over into the next, but the relationship is far looser than in the level series.

2.4 Differenced series and decomposition

Figures 9 and 10 focus explicitly on the first- and seasonal-differenced log series, shedding light on the dynamics of growth rather than levels. Figure 9 presents the original and seasonally adjusted version of this growth-rate-like measure, while Figure 10 decomposes it into trend, seasonal and irregular components via STL.

In Graph 9, the differenced series oscillates around zero, with spikes corresponding to booms and busts. The most dramatic negative values cluster around the global financial crisis and the COVID-19 pandemic, where quarter-on-quarter changes are sharply negative. These episodes are followed by large positive spikes as the economy rebounds. Between crises, the series exhibits more moderate swings, with growth typically positive but occasionally dipping below zero during domestic slowdowns or external shocks.

Seasonal adjustment at this stage aims to strip out any remaining regular within-year patterns in growth rates. The adjusted series in Figure 9 is smoother than the raw differenced series, making it easier to distinguish genuine changes in momentum from seasonal noise. In the early 2000s, the adjusted growth series is volatile but predominantly positive, reflecting the high-growth, high-uncertainty environment of post-transition reconstruction. In the 2010s, the amplitude of fluctuations moderates somewhat, but the series still displays a sequence of small expansions and contractions. The COVID-19 period shows the largest single negative swing, followed by an exceptionally strong rebound, before growth settles back into a more moderate range in 2022–2025.

Figure 10 applies STL to the differenced series. In principle, once first differences has been taken, the trend component of the growth-rate series should hover close to zero: in the long run, growth rates may fluctuate around a relatively stable mean. The estimated trend in Figure 10 captures medium-term shifts in average growth. It is higher in the early 2000s, dips in the aftermath of the global financial crisis and during the early-2010s adjustment period, rises again in the years of stronger expansion before COVID-19, and then reflects the sharp negative and positive swings around the pandemic.

The irregular component in Figure 10 now carries most of the action, with spikes that mark crisis quarters and occasional outliers linked to idiosyncratic domestic developments. This decomposition reinforces the view that, once structural growth and seasonality are controlled for, much of the quarter-to-quarter variation in Serbia’s real GDP is driven by short-run shocks rather than by systematic seasonal or trend forces.

3. Economic outlook

The recent behaviour of Serbia’s real GDP must be read against a complex backdrop: the lingering effects of the pandemic, the inflation surge of 2022–2023, the European energy shock, and shifting geopolitical conditions in the Western Balkans and beyond. The most recent quarters in the dataset, up to 2025Q2, show output continuing to rise in seasonally adjusted terms, with positive quarter-on-quarter growth following the brief dip in early 2025 reported by national statistics.

Compared with earlier phases, the current expansion looks more measured but also more resilient. The trend component in both levels and growth rates suggests that the economy is no longer in the rapid catch-up phase of the early 2000s; instead, it is settling into a pattern of moderate, perhaps three-to-four-per-cent annual growth in real terms, interrupted by occasional quarterly setbacks. Seasonal patterns remain relatively stable, and there is no clear evidence of an abrupt structural change in intra-year dynamics. This is consistent with an economy whose sectoral mix, industry, trade, transport, public services, evolves gradually rather than undergoing sudden regime shifts.

From a regional perspective, Serbia’s GDP path resembles that of several other Western Balkan economies: a strong early-2000s rebound, a severe 2009 contraction, a slower and stop-start recovery in the early 2010s, and a sharp but temporary COVID-19 slump followed by a decent rebound. Where Serbia differs is in the relative importance of manufacturing, energy and public investment programmes, as opposed to the very strong tourism-driven seasonality seen in coastal economies. This is reflected in the moderate amplitude of the seasonal component and the absence of extremely large recurring peaks and troughs within each year.

The decomposition and diagnostics also highlight the role of policy and institutions. Periods of fiscal consolidation, structural reform and improved macroeconomic management appear to coincide with smoother trends and reduced volatility in the irregular component, while episodes of political or external turbulence are associated with clusters of large residual shocks. Looking ahead, the stability of the trend and the containment of irregular fluctuations will depend on the credibility of fiscal and monetary policy, the pace of structural reforms, and the progress of Serbia’s EU accession process, which frames many of the country’s legal and regulatory changes even though membership has not yet been achieved.

Taken together, the recent data suggest a cautiously positive narrative: Serbia has absorbed a sequence of substantial shocks yet continues to grow, with real GDP at a noticeably higher level than before the pandemic and with no obvious sign of a renewed structural slowdown in the trend. At the same time, the experience of 2009, 2014 and 2020, clearly visible in the irregular component, is a reminder that vulnerability to external and domestic shocks remains a central feature of the country’s macroeconomic landscape.

4. Methodological appendix

4.1 Graphical exploration

The analysis in this blog relies heavily on graphical tools applied to quarterly real GDP. Time-series graphs of the level and the log of GDP are used to identify long-run trends, cyclical movements and major breaks, such as the sharp drops during the global financial crisis and the COVID-19 pandemic. Decomposition plots separate the series into trend, seasonal and irregular components, making it easier to distinguish regular intra-year patterns from one-off shocks. Seasonal plots and subseries plots help reveal the typical shape of GDP within a year and how that pattern evolves over time, while lag plots provide visual evidence on persistence, whether strong quarters tend to follow strong quarters, or whether growth is more erratic.

These tools are powerful but must be interpreted with care. It is easy to mistake seasonal peaks and troughs for changes in underlying growth, or to read too much into single outliers that may reflect data revisions or one-off events. Graphical analysis is therefore best seen as a first step, guiding more formal modelling rather than replacing it.

4.2 Transformations: logs and differencing

Before applying seasonal adjustment, the series is transformed to stabilise its variability and make patterns clearer. When the variance of a time series increases with its level, a common feature of macroeconomic aggregates, taking logarithms is a standard remedy. If () denotes real GDP in quarter (), the transformed series is (). In this form, equal vertical distances correspond to roughly equal percentage changes, and seasonal swings become more comparable over time.

To focus on growth rates rather than levels and to remove persistence, first differences and seasonal differences are used. The first difference, (), captures quarter-on-quarter change, while the seasonal difference, (), with () for quarterly data, captures year-on-year change. Applied to the log series, these operations yield approximate percentage growth rates. Differencing removes much of the long-run trend and systematic seasonality, producing a series that fluctuates around a relatively stable mean and is better suited to studying short-run dynamics and shocks.

4.3 Seasonal adjustment and decomposition using STL

Seasonal adjustment and decomposition in this analysis are carried out using STL, Seasonal-Trend decomposition using Loess. The basic idea is that the observed series () can be represented as the sum of three components: (, where () is the slowly evolving trend-cycle, () is the regular seasonal pattern, and () is the irregular remainder capturing unsystematic noise and one-off shocks.

STL uses locally weighted regression (Loess) to estimate the trend and seasonal components in a flexible, non-parametric way. It proceeds iteratively, smoothing the series over long windows to extract a smooth trend, then over shorter windows to estimate the seasonal pattern, and updating these components until they converge. Robustness steps can be included to reduce the influence of extreme outliers on the estimated trend and seasonality. One of the strengths of STL, highlighted in both the original Cleveland et al. paper and modern forecasting textbooks, is its flexibility: the user can choose the degree of smoothness for trend and seasonal components and apply the method to a wide variety of series without specifying a full parametric model.

In the context of Serbia’s real GDP, STL is applied both to the log-level series and to its first- and seasonal-differenced version. For the level series, STL isolates the long-run growth path and the stable quarterly pattern, leaving a residual that highlights crisis-related shocks. For the differenced series, STL verifies that trend and seasonal components are much weaker, with most of the variation residing in the irregular term. The methodology is consistent with best practice in official statistics on seasonal adjustment, as codified in the European Statistical System’s guidelines and handbooks.

5. Conclusion

The quarter-century evolution of real GDP in Serbia, viewed through the lens of STL-based decomposition and seasonal diagnostics, reveals a nuanced story of convergence under constraint. The log-level series shows a clear upward trend from 2000Q1 to 2025Q2, confirming that the economy has expanded substantially in real terms despite multiple crises. The trend component extracted by STL highlights distinct phases: a rapid post-transition recovery, a sharp interruption during the global financial crisis, a prolonged period of slower and more volatile growth in the early 2010s, a firmer expansion in the latter part of the decade, and a deep but short-lived COVID-19 slump followed by recovery.

Seasonality is present but moderate. Quarterly fluctuations around the trend follow a relatively stable pattern, with some quarters systematically a little stronger or weaker than others, but without the dramatic seasonal swings observed in more tourism-dependent economies. Seasonal diagnostics and subseries plots confirm that this pattern has not changed radically over time; instead, intra-year variations remain secondary to the larger forces of trend and shocks.

The irregular components, especially in the level series, capture the economy’s exposure to extraordinary events. The crises of 2009, 2014 and 2020 stand out as quarters where real GDP deviates sharply from its usual path, reflecting global financial turmoil, natural disasters and pandemic-induced shutdowns. The decomposition of the differenced series shows that, once trend is removed, quarter-to-quarter movements are driven largely by such short-run shocks, with some persistence but no evidence of runaway instability.

Taken together, these patterns suggest an economy that has achieved meaningful real growth over the last twenty-five years, but one that remains vulnerable to external and domestic disturbances. The challenge for policymakers is to sustain the positive trend while reducing the size and impact of future irregular shocks. Continued progress in macroeconomic management, structural reform and institutional strengthening, alongside deeper integration into European production and financial networks, will be central to that effort. Applying the same toolkit of graphical analysis, transformations and STL decomposition to other key indicators, such as industrial production, wages, prices or tourism, would further illuminate the structure and resilience of Serbia’s economy and help inform that policy debate.

References

Cleveland, R. B., Cleveland, W. S., McRae, J. E., & Terpenning, I. (1990). STL: A seasonal-trend decomposition procedure based on Loess. Journal of Official Statistics, 6(1), 3–73. https://www.math.unm.edu/~lil/Stat581/STL.pdf

Eurostat. (2025). National accounts and GDP – Statistics explained. European Commission. https://ec.europa.eu/eurostat/statistics-explained/index.php/National_accounts_and_GDP

Eurostat. (2024). Gross domestic product (GDP) and main components (output, expenditure and income), quarterly – Chain linked volumes, index 2020 = 100 (namq_10_gdp). Retrieved from Eurostat database.

Hyndman, R. J., & Athanasopoulos, G. (2025). Forecasting: Principles and practice (3rd ed.). OTexts. https://otexts.com/fpp3