Banca de QUALIFICAÇÃO: PIETROS ANDRE BALBINO DOS SANTOS

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE: PIETROS ANDRE BALBINO DOS SANTOS
DATA: 13/08/2026
HORA: 09:00
LOCAL: Virtual
TÍTULO:

Regional Climate Shocks and Market Uncertainty: A Distributional Analysis of Water Stress on Arabica Coffee Prices

PALAVRAS-CHAVES:

XGBoostLSS, GAMLSS, Distributional Regression, Price Volatility, Agrometeorology, Arabica Coffee.

PÁGINAS: 50
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Probabilidade e Estatística
SUBÁREA: Probabilidade e Estatística Aplicadas
RESUMO:

This study formally tests the popular belief that regional water stress drives up coffee prices, using Minas Gerais, Brazil’s leading Arabica coffee-producing state, as a case study within a globally determined price market. After confirming, via ADF/KPSS diagnostics, that the logreturn of the deflated CEPEA/ESALQ price series is stationary while the price level is not, we adopt the log-return as the primary target and employ a dual-framework distributional strategy: a parametric GAMLSS model (Sinh-Arcsinh) and a machine learning model (XGBoostLSS, Student-t), both estimating the conditional mean (µ) and volatility (σ) jointly. Automatic term selection in the GAMLSS framework removed all climate terms from every distributional parameter, a result consistent with the market efficiency hypothesis (H2). TreeSHAP analysis of the XGBoostLSS model showed that the mean absolute contribution of climate indices to σ was over twenty times larger than to µ (0.14 versus 0.006), revealing a threshold effect above which water deficit sharply amplifies volatility, a pattern that persisted after removing the most extreme observations. Re-specifying the GAMLSS volatility equation with a smooth water-deficit term, informed by this machine learning finding, yielded an individually significant coefficient (p = 0.007); however, a likelihood-ratio test did not justify the added complexity (p = 0.458). This indicates a qualitative convergence between frameworks not yet confirmed at conventional statistical thresholds, plausibly reflecting limited statistical power in the extremestress subsample (14% of observations). Out-of-fold calibration diagnostics confirmed adequate fit for the log-return model, while the secondary price-level specification exhibited autocorrelation, tree-extrapolation limitations, and calibration failure—reinforcing its treatment as a robustness comparison rather than a primary target. Overall, the evidence indicates that regional climate extremes in Minas Gerais do not move the average coffee price, but do amplify market uncertainty.

MEMBROS DA BANCA:
Externo ao Programa - VICTOR BUONO DA SILVA BAPTISTA - DEG/EENG (Suplente)
Presidente - PAULO HENRIQUE SALES GUIMARAES (Membro)
Interno - LUIZ OTAVIO DE OLIVEIRA PALA (Membro)
Externo ao Programa - FELIPE SCHWERZ - DEA/EENG (Membro)
Externo ao Programa - CAMILLA MARQUES BARROSO - DES/ICET (Membro)