El comercio internacional de Guatemala a través de la ecuación de gravitación y métodos autorregresivos para el periodo de 2002 a 2023

Abstract

Esta tesis nació del interés de juntar ambas ramas que a primera vista son opuestas. Para esto se analizó el comercio internacional de Guatemala a través de un análogo de la ley de gravitación universal de Newton y, como estudio de caso, se pronosticó el comercio bilateral de Guatemala con Japón utilizando métodos autorregresivos que tuvieron sus inicios tratando de explicar fenómenos estocásticos. El alcance de la investigación es empírico. El modelo de gravitación cuenta con diferentes limita- ciones, por lo que sus resultados son exploratorios que puede servir de base para nuevas investigacio- nes. Adicionalmente, el pronóstico del comercio es efectivo a corto plazo y necesita una actualización constante de su serie de tiempo para extender su uso.

The objective was to model Guatemala’s international trade and forecast its bilateral trade with Japan. First, the gravity model of trade was applied to Guatemalan exports and imports from 2002 to 2023. The model comprises macroeconomic, governance, sociocultural, and geographic variables. The Poisson pseudo-maximum likelihood estimator (PPML), fixed effects panel analysis, and the Heckman sample selection estimator model were used. Trade was studied in three phases: with its neighboring countries (Mexico, Belize, El Salvador, and Honduras) and the United States; with the main trade blocs at the continental level (COMESA, TLCAN, MERCOSUR, ASEAN+3, EU, and ANZCERTA); and at the global level. PPML was found to be efficient in representing trade, and macroeconomic and sociocultural variables increased it, while governance and geographic variables decreased it. Because only the naive variant of the model was significant, coupled with selection bias, the results are a preliminary exploration of the effect that the variables had on Guatemalan international trade during 2002–2023. To improve these results, the structural variant composed of multilateral resistances is needed. These can be approximated by increasing the study period and adjusting exports and imports for trade shocks. To forecast bilateral trade with Japan, the autoregressive ARIMA and SARIMA models were applied to the main goods distributed by the agricultural, manufacturing, and extractive industries from 2002 to 2024. Only the results of the first two models were significant: coffee exports are modeled by ARIMA (0, 1, 3) (2, 1, 0)12, while clothing items exports are modeled by ARIMA (0, 1, 1) (2, 0, 0)12. Meanwhile, the time series for iron and steel imports from the extractive industry lacked a useful trend for making projections. Forecasts are representative in the short term; their uncertainty increases from 2027 onward. Constant updates on export data are recommended, or alternative estimation methods, such as neural networks, should be used to generate these forecasts.