Economic Time Series Decomposition Using Seasonal ARIMA in Macro-Level Planning Models
Year: 2026
Type: conference-paper
Abstract: Macroeconomic projections help inform fiscal and monetary policy decisions regarding government budget paths, central bank interest rate paths, and government debt issuance plans. Given the persistent underperformance across macroeconomic indicators between official forecasts and actual outcomes, the need to evaluate systematically these time series decomposition techniques for their ability to partition a series of macroeconomic indicators into their trended, seasonal, cyclical and irregular components with acceptable precision, for medium-term policy planning, becomes evident. This paper makes use of Seasonal Autoregressive Integrated Moving Average (SARIMA) modelling for a balanced set of six macroeconomic indicators such as growth rate of GDP, Consumer Price Index (CPI) inflation, unemployment rate, Industrial Production Index (IPI), trade balance, retail sales index, and the corresponding eight OECD economies (United States, United Kingdom, Germany, France, Japan, Australia, Canada, and South Korea) over a$\mathbf{3 0 0}$-month sample period (January 2000 to December 2024), resulting in 48 indicator-country time series. The dataset is divided into a training set (January 2000 to December 2019; 240 months), a validation set (January 2020 to December 2021; 24 months), and a test set (January 2022 to December 2024; 36 months); the validation period is the period of the structural breaks caused by the COVID-19 shock (2020-2021), which provides a stress test to the stability of the model in the face of sharp economic discontinuity. The Hyndman-Khandakar auto-ARIMA algorithm (AIC criterion; seasonal period:$\mathbf{m}=\mathbf{1 2}$months) is used to identify the SARIMA model per series and comparing them with 4 benchmarks: Exponential Smoothing (ETS), Holt-Winters Multiplicative (HW), Vector Autoregression (VAR) and Facebook Prophet. For the 48 series in the 36-month test set, SARIMA outperforms ETS (4.84%), HW (4.12%) and VAR (3.84%) with a mean MAPE of 3.24% that is within 0.24 percentage points of Prophet (3.48%). Overall, SARIMA performs best on the indicators of unemployment (MAPE: 1.84%) and CPI (2.48%), and worst on indicators of trade balance (MAPE: 4.84%) and retail sales (MAPE: 3.84%), showing that the two series are more volatile and less seasonal regular than labour market and price indicators. The 6.8-month reliable forecast horizon for SARIMA is greater than for ETS, which is 3.2-months: more than double the lead time for policy planners' fiscal and monetary decisions.
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W7225824919