Pronóstico del crecimiento trimestral de Costa Rica mediante modelos de frecuencia mixta
Fecha
Tipo
artículo original
Autores
Rodríguez Vargas, Adolfo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad de Costa Rica
Resumen
Descripción
Se evalúa la utilidad de modelos de frecuencia mixta para pronosticar la tasa de crecimiento trimestral del PIB real de Costa Rica: se estiman modelos bridge y MiDaS con diferentes longitudes de rezago usando información del IMAE y se calculan pronósticos (horizontes de 0-4 trimestres) que se comparan entre sí, con los de modelos ARIMA y con combinaciones de pronósticos. Combinar los pronósticos con mejor ajuste resulta útil especialmente para proyectar en tiempo real, mientras que los MiDaS muestran el mejor desempeño general: al incrementarse el horizonte su precisión disminuye levemente, su porcentaje de acierto de cambios en la tasa de variación del producto permanece estable y varios de ellos son insesgados. Los pronósticos de MiDaS simples con 9 y 12 rezagos resultaron insesgados para todos los horizontes y conjuntos de información evaluados, y son los que mostraron más diferencias significativas en capacidad de pronóstico con todos los demás modelos.
We assess the utility of mixed-frequency models to forecast the quarterly growth rate of Costa Rican real GDP: we estimate bridge and MiDaS models with several lag lengths using information of the IMAE and compute forecasts (horizons of 0-4 quarters) which are compared between themselves, with those of ARIMA models and with those resulting from forecast combinations. Combining the most accurate forecasts is most useful when forecasting in real time, whereas MiDaS forecasts are the best-performing overall: as the forecasting horizon increases, their precisionis affected relatively little; their success rates in predicting the direction of changes in the growth rate are stable, and several forecastsremain unbiased. In particular, forecasts computed from simple MiDaS with 9 and 12 lags are unbiased at all horizons and information sets assessed, and show the highest number of significant differences in forecasting ability in comparison with all other models.
We assess the utility of mixed-frequency models to forecast the quarterly growth rate of Costa Rican real GDP: we estimate bridge and MiDaS models with several lag lengths using information of the IMAE and compute forecasts (horizons of 0-4 quarters) which are compared between themselves, with those of ARIMA models and with those resulting from forecast combinations. Combining the most accurate forecasts is most useful when forecasting in real time, whereas MiDaS forecasts are the best-performing overall: as the forecasting horizon increases, their precisionis affected relatively little; their success rates in predicting the direction of changes in the growth rate are stable, and several forecastsremain unbiased. In particular, forecasts computed from simple MiDaS with 9 and 12 lags are unbiased at all horizons and information sets assessed, and show the highest number of significant differences in forecasting ability in comparison with all other models.
Palabras clave
DATOS DE FRECUENCIA MIXTA, MODELOS MIDAS, MODELOS BRIDGE, PRONÓSTICO EN TIEMPO REAL, MIXED -FREQUENCY DATA, MIDAS MODELS, BRIDGE MODELS, NOWCASTING