Monthly Inflation Forecasting in Madagascar Using Facebook Prophet: An Empirical Evaluation[Prévision de l'inflation mensuelle à Madagascar à l'aide de Facebook Prophet : une évaluation empirique]

This paper explores the application of the Facebook Prophet model to forecast monthly inflation in Madagascar using Consumer Price Index (CPI) data from February 2016 to October 2024. Facebook Prophet, developed by Taylor and Letham (2017), is an additive time series forecasting model designed to handle trend shifts, seasonal effects, and holiday impacts, while being accessible to non-expert users. The study aims to (i) model the evolution of monthly inflation in Madagascar and (ii) evaluate the predictive accuracy of Prophet through error metrics such as the Mean Absolute Percentage Error (MAPE) and the Mean Absolute Scaled Error (MASE). The logarithm of the CPI is used as the response variable, and time is the explanatory variable. The model achieved highly accurate results, with a MAPE of 0.03% and a MASE of 0.2278, indicating excellent forecasting performance compared to a naïve model. Diagnostic tests further confirm that the residuals are normally distributed, non-autocorrelated, and homoscedastic, supporting the model's robustness. Overall, the findings suggest that Facebook Prophet is a reliable and effective tool for forecasting inflation in the Malagasy context. The model's accessibility, performance, and resilience to data irregularities make it a promising solution for economic policymakers and analysts in developing countries.