Advances and Applications in Statistics
Volume 65, Issue 1, Pages 89 - 106
(November 2020) http://dx.doi.org/10.17654/AS065010089 |
|
FORECASTING MEXICAN PESO - US DOLLAR EXCHANGE RATE AFTER THE 2020 PANDEMIC WITH GARCH AND XGBOOST TIME SERIES MODELS
Alejandro Ruiz-Olivares, Martha Elva Ramírez-Guzmán and Sandy Yaredd Trujano-Ramos
|
Abstract: The relation peso-dollar is a big challenge for the Mexican economy. This is especially true after the intervention of the COVID-19 pandemic during 2020. The uncertainty derived from this pandemic encouraged us to forecast the future behavior of the exchange rate between peso and dollar. Therefore, this research presents an ARIMA-intervention-GARCH model and a machine learning method called extreme gradient boosting (XGBoost), for producing peso-dollar exchange rate forecast for 2021, from 2020 daily data. An ARIMA-intervention model identified depreciation of 0.92, 0.94 and 1.07 Mexican pesos per dollar during March 9th, 12th and 17th, respectively. However, appreciation of 1.15, 0.97 and 0.66 was detected from March 3rd to April 7th of 2020. The asymmetrical GJR-GARCH model with an asymmetrical Student-t distribution captured the volatility of the data with a persistence of 98% of probability. The significant leverage parameter of this model showed an asymmetrical volatility. The ARIMA, ARIMA-intervention-GARCH and XGBoost models were characterized by presenting high, medium and low forecasts for 2021. During the COVID-19 pandemic period, the ARIMA-intervention-GJR-GARCH model showed the lowest residuals. This model with ten thousand Monte Carlo simulations showed a positive trend in the peso-dollar exchange rate with a probability of 50, 40, 30 and 20% so that the dollar costs 25, 26, 27 and 28 or more, respectively, for April 2021. |
Keywords and phrases: ARIMA, intervention, GARCH, XGBoost, COVID-19.
|
|
Number of Downloads: 254 | Number of Views: 818 |
|