Forecasting time series using structural change information

This article presents a paper by A. Gantungalag and D. R. Osborne entitled "Time Series Forecasting Using Structural Change Information". Everyone has a need to predict what will happen in the near or distant future. Businesses need forecasts to allocate their budgets and evaluate future income and expenses, while government organizations need forecasts to develop policies and programs. Forecasting is the act of predicting what will happen in the future based on information and numerical data about past and present events. Despite the widespread use of forecasting, good forecasting is not easy.

According to the study, structural change is one of the main problems complicating macroeconomic projections, or time series forecasting. Persistent structural changes are common in economies, which means that any economic indicator changes its characteristics suddenly at a certain point in time and does not return to its old state. As structural changes occur, large differences between pre-and post-change periods occur, making pre-change information less relevant for future forecasting. Structural changes can be internal (systemic changes, policy effects, changes in measurement methods, etc.) or external (global pandemics, interstate wars, etc.).

Source: NSO Statistical Database. Note: The figure was drawn for a simplified paper of the NRCC Research Workshop.

This work by A. Gantungalag and D. R. Osborne proposed innovative methods to improve forecasting and compared the effectiveness of the methods with other commonly used forecasting methods based on simulated and real numerical data. According to the research, knowing when and how the structure of the time series has changed is important for choosing the appropriate method and improving the accuracy of the forecast. However, it is very difficult to accurately detect the period of structural change and determine the amount of change, due to many reasons.

They proposed a 2-step test to accurately detect structural changes. In addition, the researchers suggested using forecasting combinations based on the corresponding confidence interval instead of the point assessment of the period when structural changes occurred in forecasting.

The results of the study show that the confidence interval-based forecasting method proposed by the authors improves forecasting accuracy, regardless of the amount of structural change compared to other methods in this study. Furthermore, it has been shown that using a 2-step test to detect structural changes can reduce the prediction error of other commonly used methods.

This study by researchers A. Gantungalag and D. R. Osborn were published in the November 2022 issue of Empirical Economics magazine. You can read the original work here. Researcher Gantungala will discuss this research work at the NRCC research seminar on March 1, 2023, at 18:00, so if you want to get more detailed information from the researcher and exchange ideas, get the seminar information here.

# no pec # Calculate 3 + 4 3 + 4 # Calculate 6 + 12 # Calculate 3 + 4 3 + 4 # Calculate 6 + 12 6 + 12 test_output_contains("18", incorrect_msg = "Make sure to add `6 + 12` on a new line. Do not start the line with a `#`, otherwise your R code is not executed!") success_msg("Awesome! See how the console shows the result of the R code you submitted? Now that you're familiar with the interface, let's get down to R business!")