Electricity tariffs in emerging economies are often influenced by macroeconomic volatility and regulatory design, affecting both affordability and system stability. Understanding these interactions is crucial for anticipating price fluctuations and ensuring sustainable energy policy. This paper examines the influence of macroeconomic conditions on electricity tariff dynamics in Colombia by integrating econometric and machine learning approaches. Using monthly data from 2009 to 2024 and a set of 153 macroeconomic indicators condensed via principal component analysis (PCA), we assess the predictive performance of vector autoregressive (VAR), SARIMAX, and XGBoost models, as well as a hybrid VAR–XGBoost specification. Impulse-response analysis reveals that tariff components exhibit limited sensitivity to macroeconomic shocks, underscoring the buffering role of regulation and sector-specific drivers. However, forecasting exercises demonstrate that accuracy is highly component-specific: SARIMAX performs best for transmission and restrictions, and VAR dominates for distribution and losses, while the hybrid model outperforms for generation and commercialization. These findings highlight that although macroeconomic pass-through into tariffs is weak, hybrid approaches that combine structural econometric dynamics with nonlinear learning can deliver tangible forecasting gains. The study contributes to the literature on electricity pricing in emerging economies and offers practical insights for regulators and policymakers concerned with tariff predictability and energy affordability.

Digital finance complaints can reveal consumer vulnerabilities that are not visible in aggregate adoption or payment data; however, their relationship with market conditions remains poorly understood. Here, we audit 17,571,890 Consumer Financial Protection Bureau records and construct a harmonized monthly panel of 183,324 core digital finance complaints from January 2014 through July 2026. We combine these outcomes with 119 transformed FRED-MD series, principal component controls, and a statistical market stress innovation. Factor-augmented local projections show no statistically precise baseline response of aggregate core complaint intensity; however, the baseline aggregate coefficient is positive but imprecise and becomes statistically significant when two providers associated with an exceptional January 2025 concentration event are excluded. The more stable finding concerns composition: at 12 months, the cumulative log-odds response of access-and-account-restriction complaints is 0.219 ($p<0.001$), equivalent to 24.5% higher odds under the standard classification. This estimate depends partly on a combined “managing, opening, or closing” label; omitting these ambiguous cases reduces the coefficient to 0.080 ($p=0.152$). This represents a relative redistribution of reported complaint types. The 12-month result survives exclusion of virtual currency complaints, alternative HAC bandwidths and HC3 inference, window-robust event controls, and Holm adjustment across five categories and thirteen horizons. It attenuates and becomes imprecise under a deliberately narrow access definition that excludes ambiguous opening/closing labels. Predetermined high-stress states are not statistically different, and macro-augmented machine learning models do not beat a parsimonious autoregressive benchmark. The evidence is associational rather than structural, and indicates that market stress is more reliably associated with the composition of reported digital finance complaints than with their total volume.