Data context and figures. This note interprets a set of monthly charts for Serbia’s retail trade (constant prices) and catering (constant prices). The figures include original vs seasonally adjusted series (in levels and logs), STL decompositions, a scatterplot matrix, first- and seasonally-differenced views, “trend vs seasonal contribution” comparisons, and a suite of seasonal diagnostics (seasonal plots, subseries plots, lag plots, and TSstudio boxplots).

1. Introduction

Retail trade and catering are two of the quickest-moving gauges of household demand and service activity. Retail captures the broad flow of goods to consumers, while catering reflects out-of-home consumption that is income-sensitive and seasonally patterned (weekends, holidays, tourism). Together they offer an early, high-frequency read on living standards, inflation pass-through to real activity, and the rhythm of services recovery after shocks.

Across the monthly span covered in the figures, both series display the familiar combination of a rising long-run envelope punctuated by cyclical slowdowns and sharp interruptions around global shocks. Seasonal adjustment is essential here: raw monthly paths are dominated by within-year patterns (holidays, tourist season, fiscal calendars). Comparing original and adjusted lines clarifies whether momentum is genuinely changing or merely reflecting the calendar.

2. Description of the time series and components

2.1 Overall patterns in levels and adjusted series

In levels, retail trade shows a steady upward drift punctuated by identifiable dips and rebounds. The most striking of these are the abrupt contraction and quick normalization around the pandemic shock and the subsequent period of strong catch-up as restrictions eased. Catering moves with higher volatility: it tends to slump more forcefully in downturns and surge when mobility and tourism recover, consistent with its discretionary nature.

The log-scale versions make two features clearer. First, growth appears more proportional over time: expansions look steadier once variance is stabilized by logs. Second, the seasonally adjusted paths strip out recurring month-of-year movements, revealing underlying momentum. In both series, the post-shock recovery is visible as a firm upward step in the adjusted line, followed by phases of moderation rather than uninterrupted acceleration, consistent with cooling global conditions and tighter financial settings after the initial rebound.

2.2 Trend, seasonal, and irregular components

The STL decompositions separate the observed series into a trend-cycle, a seasonal component, and an irregular remainder. For retail trade, the trend-cycle rises over the medium run, with two notable interruptions: a sudden trough (pandemic) and a subsequent robust upswing that flattens as conditions normalize. Seasonality is persistent and well-behaved: end-year and mid-year peaks recur with similar amplitude, suggesting stable shopping patterns, promotions, and holidays. The irregular component is usually modest, spiking only around large events.

Catering’s decomposition shows a more elastic trend-cycle, deeper troughs and steeper recoveries, reflecting sensitivity to mobility and tourism flows. Its seasonal amplitude is larger than retail’s: holiday peaks and tourist-season bulges are pronounced, while off-season months show systematic softness. Irregularities appear in short bursts (closures, reopening surges, event-driven demand), but the dominant picture is one of strong, recurring seasonality layered onto a cyclical trend.

2.3 Trend vs seasonal dominance

The comparative “trend vs seasonal contribution” view makes the balance of forces explicit. Retail trade appears more trend-dominated: long-run drivers (incomes, credit, employment) explain most of its variability once seasonality is netted out.

Catering tilts more toward seasonal dominance, with within-year cycles contributing a larger share of total movement, intuitive for a series tied to holidays, weather, and tourism calendars. Where the irregular share is elevated, one should suspect temporary policy measures, price shocks that compress real demand, or measurement noise; but in these figures, irregularity is episodic rather than chronic.

2.4 Seasonal diagnostics and detailed patterns

Seasonal plots and subseries charts distill the “typical year.” Retail’s profile is textbook: modest dips in the shoulder months and firm peaks around year-end, with a secondary uplift mid-year. The amplitude is stable over time, implying a robust seasonal calendar that planners can anticipate.

Catering’s seasonal signature is sharper: high-summer and end-year peaks stand out, while shoulder months are consistently soft, exactly the pattern one expects in a service tied to leisure travel and festivities.

Lag plots indicate positive month-to-month persistence, more so in retail than in catering where bursts around seasonal peaks produce wider scatter.

The TSstudio boxplots reinforce these readings: medians align with expected peak months, interquartile ranges widen in high-season for catering, and outliers congregate around shock periods.

2.5 High-frequency differenced views

The first- and seasonal-difference charts highlight changes rather than levels. For both series, the pandemic period shows extreme negative spikes followed by sharp positives as activity reopens, useful for pinpointing break dates and outliers.

Outside these episodes, the differenced paths oscillate around zero with occasional clusters of larger moves, suggesting that once trend and seasonality are removed, residual shocks are short-lived rather than persistent. The STL of the differenced series confirms that what remains is chiefly irregular noise with small residual seasonality, an internal consistency check for the earlier decompositions.

2.6 Cross-series co-movement

On logs, retail and catering co-move positively but not one-for-one. During expansions, catering often leads on the upside, consistent with income elasticity and tourism spillovers, while in downturns its contractions are steeper.

This asymmetry helps explain why catering’s seasonal and irregular shares are larger even when both series share the same macro backdrop.

3. Economic outlook

The latest stretch of seasonally adjusted data points to a normalization phase. After the strong post-reopening bounce, retail’s trend has cooled toward a steadier expansion path, while catering, though still exhibiting strong seasonal peaks, shows less dramatic month-to-month swings than in the immediate aftermath of restrictions. Interpreted against a European setting of disinflation, tighter real rates than in 2021–22, and still-elevated uncertainty, this pattern suggests resilient but more measured household demand.

Two risks matter for the near term. First, real income dynamics: if nominal wage growth continues to outpace inflation, retail should maintain steady gains even as interest-sensitive purchases rotate. Second, tourism and mobility: catering’s seasonal amplitude will remain high, but the level of the seasonal peaks will hinge on regional travel flows and domestic leisure budgets. Barring new shocks, the baseline is incremental improvement with familiar seasonal rhythms.

4. Methodological appendix

Graphical exploration. The analysis relies on visual tools to detect trend shifts, cyclical swings, seasonal regularities, volatility changes, and outliers. Care is taken not to confuse seasonality with trend or to over-read single observations.

Transformations. Logs are applied where variance grows with the level, making proportional changes comparable over time. In some views we use first differences ( ) and seasonal differences ( ) to focus on monthly and year-over-year changes and to remove persistent seasonality.

Seasonal adjustment and decomposition. STL (Seasonal-Trend decomposition using Loess) splits each series into trend-cycle, seasonal, and irregular parts. It iteratively smooths seasonal and trend windows and is robust to outliers. We also inspect diagnostics (seasonal plots, subseries, lag plots, TSstudio boxplots) to confirm that the extracted components match the observed calendar regularities.

5. Conclusion

The figures present a coherent narrative. Retail trade in Serbia is fundamentally trend-driven with stable, well-understood seasonality; catering is more seasonal and more sensitive to shocks, amplifying both slumps and recoveries. The post-pandemic era shows a decisive rebound followed by normalization. Seasonal diagnostics confirm persistent calendars rather than drifting seasonality, which supports the use of standard seasonal adjustment. Looking ahead, steady real-income gains would sustain retail, while catering’s peaks will hinge on tourism and leisure budgets. Extending the same toolkit to prices and wages—or to tourism nights and card transactions, would enrich the demand-side picture and help cross-validate these signals.

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.
• Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts.
• National Statistical Office of Serbia. (n.d.). Retail trade and catering statistics [Monthly series].
• Shumway, R. H., & Stoffer, D. S. (2017). Time series analysis and its applications (4th ed.). Springer.