Modelos univariados de series de tiempo para predecir la inflación de corto plazo

Inflation forecasting plays a central role in the design of monetary policy, especially in institutions that progressively adopt a scheme of inflation targeting, as the Central Bank of Uruguay. This paper constitutes the first step in a research agenda in the field of inflation forecasting, where a battery of univariate time series models of the Consumer Price Index (CPI) and its components are assessed, based on its predictive power through different forecasting steps, focusing in the short term. Working with a sample that spans from 1997.03 to 2009.10 and a trimmed sample from 2003.01, direct and indirect forecasts for general CPI and an exclusion indicator are elaborated. Forecasting accuracy is then assessed in the period 2009.11-2010.07, comparing the results between them and two different benchmarks: a naïf model, constituted by the random walk; and the median of the inflation expectation survey published by the Central Bank. The main results show that in one step ahead predictions, the best model is the direct model with the trimmed sample (M2), prevailing over indirect projection and both benchmarks. This individual model is only outperformed by the linear combination of the M2 and the median of expectations. However, this choice is not available at the time of spreading the monthly predictions report. In the analysis by components, several models show a good performance, such as exclusion tradables and non-tradables, whereas the main sources of error of the indirect projection are the model of fruits and vegetables and the estimation by expert judgment of the regulated items. The models for the exclusion indicator, either direct or indirect, show a good performance. Finally, the quarterly forecasts issued from monthly univariate models 3, 2 or 1 steps forward show a better performance than the one step ahead prediction of the structural quarterly model. At the present state of work, it is suggested to use the M2 model for the CPI general level and its components.