Regime Aware Inflation Forecasting Using Markov Switching and Hybrid Models

Accurate inflation forecasting is essential for designing effective macroeconomic policies, but the inflationary process is also recognized as regime-switching, with periods of stability and turmoil. Ignoring the heterogeneity of regimes in inflation forecasting may lead to biased results in forecast outcomes and policy implications. In this paper, we discuss a regime-informed framework for inflation forecasting that focuses on structural changes in the inflationary process. First, using a two-state Markov switching approach, we uncover hidden low-inflation and high-inflation regimes in a long historical sample of inflation data. We then go on to compare the accuracy of inflation forecasting of the conventional econometric model (ARIMA) and the machine learning model (Random Forest) in each of these regimes. The empirical results suggest that there is a high degree of regime dependence in the performance of the models, where although the models have equal performance in the low inflation regime, ARIMA outperforms the machine learning model in the high inflation regime. We propose a hybrid model based on these results. The hybrid model is more effective in terms of overall forecast robustness, where it decreases the chance of large forecast errors in high-inflation regimes and preserves forecast accuracy in low-inflation regimes. The above results highlight the importance of regime-dependent models in inflation forecasting and help policymakers gain useful insights into developing effective forecasting models in uncertain macroeconomic environments. It is important to note that the proposed model enhances forecast robustness by focusing on the isolation of volatile regimes rather than claiming superiority in inflationary crises.