|
on Forecasting |
| By: | Polbin, Andrey; Shumilov, Andrei |
| Abstract: | Forecasting inflation is an important and challenging practical task. In particular, models with a large number of explanatory variables on relatively short samples can often overfit in-sample and, thus, forecast poorly. In this paper, we study the applicability of the model with Bayesian shrinkage of time-varying parameters based on hierarchical normal-gamma prior to forecasting inflation in Russia. Models of this type allow for possible nonlinearities in relationships between regressors and inflation and, at the same time, can deal with the problem of overfitting. Using monthly data for 2001-2022, we find that at short forecast horizons of 1-3 months Bayesian normal-gamma shrinkage TVP model with a large set of inflation predictors outperforms in forecasting accuracy, measured by mean absolute and squared errors, its linear counterpart, linear and Bayesian autoregression models without predictors, as well as naive models. At the horizon of six months, the autoregression model with Bayesian shrinkage exhibits the best forecast performance. As the forecast horizon rises (up to one year), statistical differences in the quality of forecasts of competing models of inflation in Russia decrease. |
| Keywords: | inflation; forecasting; time-varying parameter model; Bayesian shrinkage; normal-gamma prior |
| JEL: | C53 E37 |
| Date: | 2023 |
| URL: | http://d.repec.org/n?u=RePEc:pra:mprapa:118650&r=for |
| By: | Wilmer Martínez-Rivera; Eliana R. González-Molano; Edgar Caicedo-García |
| Abstract: | Based on monthly disaggregated Consumer Price Index (CPI) item series and macroeconomic series, we explore the advantages of forecast inflation from a disaggregated to an aggregated level by aggregating the forecasts. We compare the performance of this approach with the forecast obtained modeling aggregated inflation directly. For the aggregate level, we implement some of the techniques and models, helpful to work with many predictors, such as dimension reduction, shrinkage methods, and machine learning models. Also, we implement traditional time-series models. For the disaggregated data, we use its lags and a set of macroeconomic variables as explanatory variables. Direct and recursive forecast techniques are also explored. The sample period of the analysis is from 2011 to 2022, with forecasting and evaluation out of the sample from 2017. In addition, we evaluate the forecast accuracy during the COVID-19 period. We found a reduction in the forecast error from the disaggregate analysis over the aggregate one. **** RESUMEN: En este artículo se analiza la información mensual tanto agregada como desagregada del índice de precios al consumidor (IPC) en Colombia. Se explora las ventajas de pronosticar a nivel desagregado para luego agregar pronósticos y comparar con los pronósticos obtenidos al analizar la información agregada. El cálculo de pronósticos esta basado en el ajuste de modelos y técnicas que incluyen modelos de reducción de dimensión, modelos de selección de variables, modelos de Machine Learning así como modelos tradicionales de series de tiempo ARIMA. El periodo muestral de análisis es 2011 a 2022 cuyo cálculo de pronósticos fuera de muestra se da a partir de 2017 hasta 2022. |
| Keywords: | Inflación, datos desagregados, pronósticos agregados, Inflación, datos desagregados, pronósticos agregados |
| JEL: | C52 E17 E31 |
| Date: | 2023–10 |
| URL: | http://d.repec.org/n?u=RePEc:bdr:borrec:1251&r=for |
| By: | Leo Krippner |
| Abstract: | This article introduces the eigensystem autoregression (EAR) framework, which allows an AR model to be specified, estimated, and applied directly in terms of its eigenvalues and eigenvectors. An EAR estimation can therefore impose various constraints on AR dynamics that would not be possible within standard linear estimation. Examples are restricting eigenvalue magnitudes to control the rate of mean reversion, additionally imposing that eigenvalues be real and positive to avoid pronounced oscillatory behavior, and eliminating the possibility of explosive episodes in a time-varying AR. The EAR framework also produces closed-form AR forecasts and associated variances, and forecasts and data may be decomposed into components associated with the AR eigenvalues to provide additional diagnostics for assessing the model. |
| Keywords: | autoregression, lag polynomial, eigenvalues, eigenvectors, companion matrix |
| JEL: | C22 C53 C63 |
| Date: | 2023–10 |
| URL: | http://d.repec.org/n?u=RePEc:een:camaaa:2023-47&r=for |