Fábio Sandro dos Santos

We examine whether daily green bond exchange-traded fund (ETF) returns can be forecast more accurately than the zero-return benchmark. Using four ETFs with data through 17 April 2026, we compare linear, penalized, boosting, feedforward, recurrent, and combination forecasts in historical out-of-sample experiments. The contribution is empirical and evaluative, not a new algorithm. Combinations occasionally reduce squared error by less than 0.3%; the gains are statistically insignificant and heterogeneous across funds. An adjusted-price comparison with four conventional bond ETFs on identical dates finds no significant difference in combination-loss gains between the groups. Hyperparameter and dependence checks do not reverse these findings. Hypothetical close-to-close trading payoffs are sensitive to transaction costs, but are not executable backtests. Overall, the tested models provide limited incremental predictive information, without establishing market efficiency or a forecasting difficulty unique to green bonds.

High temperatures can significantly affect plant growth, development, and physiology. The identification of genotypes with desirable morphological and physiological responses under contrasting growing conditions is therefore relevant for ornamental pepper breeding. This study aimed to characterize the phenotypic responses of ten ornamental pepper ( Capsicum spp.) accessions grown under controlled (27.15 ± 1.16 °C; VPD = 2.1 ± 0.23 kPa) and high-temperature/high-VPD (39.03 ± 1.98 °C; VPD = 6.23 ± 0.85 kPa) greenhouse conditions using morphological, phenological, and physiological traits and self-organizing maps (SOMs). Differences in plant architecture and physiological responses were observed among accessions within each growth condition, while descriptive differences were also observed between conditions. In general, plant height and canopy diameter were lower under the high-temperature/high-VPD condition, while the magnitude of gas-exchange responses varied among accessions. SOM sensitivity analysis supported a 2 × 3 topology with six neurons for both conditions, based on the balance among training quality, topological preservation, and stability across 100 independent runs. Three superclasses were identified under each condition, although their composition and multivariate profiles differed between environments. CPCE 005 and CPCE 007 remained associated under both conditions and, under high-temperature/high-VPD, were characterized by compact plant architecture combined with higher relative values of gas-exchange traits. CPCE 014 showed a distinct multivariate profile under the contrasting conditions and formed an individual superclass. Overall, SOM analysis revealed a reorganization of the relationships among morphological and physiological traits between growing conditions and identified distinct accession-specific response patterns. These results highlight CPCE 005, CPCE 007, and CPCE 014 as promising accessions for further validation in breeding studies targeting ornamental peppers grown under high-temperature and high atmospheric evaporative demand.

Which liquid ETFs actually protect investors when U.S. equities crash? This paper answers that question using daily observations from 2012-03-02 to 2026-02-27 for SPY and eight cross-asset ETFs spanning gold, oil, broad commodities, the U.S. dollar, Treasuries, real estate, high-yield credit, and investment-grade credit. Pairwise DCC-GJR-GARCH models recover time-varying correlations, and nested downside regressions identify behavior in the lower 10%, 5%, and 1% tails of SPY returns. To prevent an incremental correlation decline from being misclassified as a safe haven, inference is based on HC3-robust Wald tests of the cumulative tail-state correlations. In the full sample, the U.S. dollar and U.S. Treasuries qualify as hedges, while the U.S. dollar and U.S. Treasuries reach statistically negative correlation in at least one downside state. Protection is sharply regime-dependent. During COVID-19, effective safe-haven behavior is limited to the U.S. dollar and U.S. Treasuries; during the Russia–Ukraine regime it is observed for the U.S. dollar; and in the tariff window the hedge set narrows to gold and U.S. Treasuries. Across specifications, dollar exposure appears most frequently, while sovereign duration produces the deepest negative dependence in the full sample, COVID-19, and the tariff window. Gold remains episodic, and cyclical ETFs generally retain positive equity dependence. The findings show that hedge and safe-haven labels are state-contingent properties of the cross-asset correlation structure rather than permanent attributes of individual ETFs.

This paper examines the multifractal organization of hourly wind-speed dynamics at nine automatic stations of the Brazilian National Institute of Meteorology in Paraíba, Northeast Brazil, over the common interval from 3 November 2017 to 17 November 2019. Missing records and values outside the inclusive 1.0–18.0 m/s analytical range are removed before station-level standardization, and multifractal detrended fluctuation analysis is applied to compacted retained-observation sequences. The stability-potential index (SPI) is defined as |α₀ − 0.5|, where α₀ is the singularity-spectrum peak and 0.5 is the uncorrelated benchmark. SPI therefore measures departure from random-like scaling and is interpreted as a temporal-organization indicator, not as a stand-alone measure of wind-power resource. Rolling windows of 720 retained observations, advanced by 168 observations, are assigned to seasons, El Niño, La Niña, neutral conditions, and tropical Atlantic-gradient regimes. The empirical design combines full-sample rankings, rolling spectra, two-way clustered ordinary least squares, Frisch–Newton bootstrap quantile regression, non-overlapping and coverage-threshold sensitivities, clock-span diagnostics, analytical-range and moment-order checks, 200 shuffled surrogates per station, and station-anchored inverse-distance-weighted maps. All full-sample α₀ estimates exceed 0.5, with SPI ranging from 0.1108 in Cabaceiras to 0.3937 in Areia. El Niño is associated with lower rolling SPI in both baseline and controlled mean regressions and in the lower and median quantiles; this sign persists after excluding low-coverage Camaratuba. La Niña is not significant at the mean but is negative at the 0.90 quantile. No north-warm Atlantic-gradient windows occur in the common sample, while the south-warm gradient is positive only near the median quantile. Four of nine stations have original raw spectrum widths above the shuffled 97.5% bound. Across stations, exact two-sided Spearman tests show no supported monotonic association between coverage, wind-distribution descriptors, or altitude and α₀, SPI, or W (n = 9; all p ≥ 0.194). The results document sample-specific spatial heterogeneity and climate-regime associations while showing that compacted-window duration, calm-wind screening, and the modest nine-station network require explicit sensitivity analysis and cautious generalization.

This paper presents a systematic review of the literature on the application of Artificial Intelligence (AI) techniques in the wind energy sector, aiming to identify trends, challenges, and opportunities in this evolving field. The increasing demand for renewable energy sources, combined with advancements in AI technologies, has fueled research efforts focused on optimizing the performance and efficiency of wind energy systems. In this context, various AI approaches, including artificial neural networks, machine learning algorithms, and optimization techniques, have been explored to enhance energy production forecasting. To ensure a comprehensive analysis that accurately reflects the current state-of-the-art, this study investigates a consolidated timeframe from 2015 to 2026 using the Methodi Ordinatio 2.0 methodology. By ranking a rigorously filtered portfolio of 400 experimental articles based on annualized citation density and Journal Impact Factor, this approach prevents temporal bias against recent advancements. The review highlights a critical architectural transition in the field: the historical predominance of Recurrent Neural Networks (RNNs/LSTMs) and the current paradigm shift toward Attention-based mechanisms and Transformers, which demonstrate superior capabilities in handling long-term dependencies in volatile wind speed time series. The results indicate that while advancements in AI offer significant benefits for wind energy production, challenges persist regarding model complexity, data quality, and the need for stronger collaboration between academia and industry to enable large-scale implementation. Furthermore, this study identifies specific emerging trends, emphasizing the adaptation of Foundation Models for time-series forecasting, the critical need for Explainable AI (XAI) to ensure operational trust, and the trade-off between computational cost and accuracy in edge computing for wind farms.

Soil remineralization with rock powders has re-emerged as a strategy to broaden gradual nutrient supply and reduce dependence on highly soluble fertilizers. Its agronomic and environmental usefulness, however, depends on kinetic models able to represent asymmetric trajectories and a realistic approach to release plateaus. Here, 210 observations arranged in 15 time series of Ca, Mg, [ ], Si, and [ ] release from basalt, biotite schist, and pegmatite were fitted with five nonlinear growth models (Logistic, Exponential, Gompertz, Richards, and Von Bertalanffy). Model performance was compared using [ ], RMSE, MAE, MAPE, AIC, and BIC. The Richards model yielded the highest [ ] in every series, with a mean of 0.9706, outperforming Gompertz (0.9412), Logistic (0.9287), Exponential (0.7862), and Von Bertalanffy (0.7619). The estimated parameters also revealed consistent contrasts among rocks and nutrients, especially for [ ], Si, and [ ], indicating that the additional shape flexibility becomes important when mineral dissolution reflects multiple overlapping processes. Overall, the results show that flexible sigmoidal models are useful for comparing remineralizers and supporting gradual nutrient-release strategies. All fit metrics and parameter estimates are reported in Appendix A.

This paper examines hourly wind speed dynamics at 12 INMET stations in Piauí, Brazil, asking whether the local multifractal organization of the wind field moves closer to or farther from a random benchmark across meteorological seasons. The observational base is reorganized into a rolling multifractal framework with windows of 720 h observations and steps of 168 h. For each full sample and rolling window, the singularity-spectrum peak is translated into a stochastic inefficiency index equal to its absolute distance from 0.5, so lower values indicate greater stochastic efficiency. The analysis combines this index with mean wind speed and with the empirical third moment of speed, [ ], proportional to mean wind power density under constant air density. Full-sample results identify PIRIPIRI, GILBUES, and TERESINA as closest to the random benchmark, whereas PARNAIBA is the most resource-rich station and the most structurally distant from that benchmark. Rolling results reveal marked seasonality: mean stochastic inefficiency is highest in autumn (0.2820) and lowest in winter (0.2488). Fixed-effects and quantile regressions show that winter and spring reduce stochastic inefficiency relative to summer, whereas autumn is statistically indistinguishable from the summer benchmark in most specifications. Overall, GILBUES, CANTO DO BURITI, OEIRAS, and CARACOL provide the most balanced combination of wind-resource magnitude and stochastic efficiency in the active 12-station subset.