Big Analitics — Inteligencia Sectorial de Grado Institucional
Big Analitics — Institutional-Grade Sectoral Intelligence
Financial Technology (FinTech) representa una de las fuerzas mas transformadoras de las finanzas modernas. Desde pagos digitales y prestamos peer-to-peer hasta exchanges de criptomonedas y seguros integrados, las empresas FinTech estan redefiniendo como individuos y empresas interactuan con los servicios financieros.
El mercado global FinTech ha superado los $300 mil millones en valoracion y continua expandiendose a una tasa compuesta anual del 20%+, impulsado por la adopcion de smartphones, la regulacion de open banking y la digitalizacion creciente de los flujos financieros.
Este analisis cubre la era post-pandemia hasta Q4 2025, un periodo marcado por dramaticas subidas de tipos de interes, el colapso y recuperacion de los mercados crypto, mayor escrutinio regulatorio y el auge de los servicios financieros impulsados por IA.
El sector FinTech ha emergido de este periodo turbulento con una separacion mas clara entre modelos de negocio sostenibles y aquellos dependientes del capital barato.
Financial Technology (FinTech) represents one of the most transformative forces in modern finance. From digital payments and peer-to-peer lending to cryptocurrency exchanges and embedded insurance, FinTech companies are reshaping how individuals and businesses interact with financial services.
The global FinTech market has grown to over $300 billion in valuation and continues to expand at a 20%+ compound annual growth rate, driven by smartphone adoption, open banking regulation, and the increasing digitization of financial workflows.
This analysis covers the post-pandemic era through Q4 2025, a period marked by dramatic interest rate increases, the collapse and recovery of crypto markets, intensifying regulatory scrutiny, and the rise of AI-powered financial services.
The FinTech sector has emerged from this turbulent period with a clearer separation between sustainable business models and those dependent on cheap capital.
PayPal, Adyen, Block, Fiserv, Shift4, Nuvei, Flywire
MoneyLion, Dave, Chime, Revolut
Upstart, Pagaya, OppFi, Affirm
Coinbase, Lemonade
Wise, Remitly, Payoneer, dLocal, Global-e
Bill Holdings, Marqeta, Toast, WeBank
| Subsector | Empresas |
|---|---|
| AI Lending | Upstart, Pagaya Technologies |
| B2B Payments | Bill Holdings |
| BNPL | Affirm |
| Card Infrastructure | Marqeta |
| Cross-Border Payments | Payoneer |
| Crypto Exchange | Coinbase |
| Digital Bank China | WeBank |
| E-commerce Payments | Global-e Online |
| Emerging Market Payments | dLocal |
| Insurtech | Lemonade |
| Neobank FX | Revolut |
| Neobank Lending | MoneyLion |
| Neobank Micro | Dave Inc |
| Neobank Pure | Chime |
| Payments | PayPal |
| Payments Ecosystem | Block (Square) |
| Payments Infrastructure | Adyen, Fiserv |
| Payments Processing | Shift4 Payments |
| Payments Tech | Nuvei |
| Payments Vertical | Flywire |
| Remittances FX | Wise, Remitly |
| Subprime Lending | OppFi |
| Vertical FinTech | Toast |
Aplicamos 20 modelos analiticos que combinan estadistica clasica, machine learning y teoria de grafos para extraer inteligencia accionable de los datos financieros. A continuacion explicamos cada tecnica, que problema resuelve y que conclusiones hemos obtenido en el sector FinTech.
We apply 20 analytical models combining classical statistics, machine learning and graph theory to extract actionable intelligence from financial data. Below we explain each technique, what problem it solves, and what conclusions we have drawn for the FinTech sector.
Cada empresa se evalua en 14 metricas financieras clave (margenes, crecimiento, eficiencia, solvencia) y se posiciona en percentiles relativos al sector. No comparamos con el mercado general, sino con sus peers directos. Esto elimina el sesgo de comparar una FinTech con una petrolera.
Each company is evaluated across 14 key financial metrics (margins, growth, efficiency, solvency) and positioned in percentiles relative to the sector. We don't compare against the general market, but against direct peers. This eliminates the bias of comparing a FinTech with an oil company.
Conclusion: Adyen, PayPal y Fiserv lideran consistentemente en la mayoria de metricas. Las empresas de lending (Upstart, Affirm) quedan en los percentiles inferiores, especialmente en margenes y solvencia.
Conclusion: Adyen, PayPal and Fiserv consistently lead across most metrics. Lending companies (Upstart, Affirm) fall in the lower percentiles, especially in margins and solvency.
Que es: Combinamos tres algoritmos de clustering (K-Means, Agglomerative Clustering y DBSCAN) con reduccion de dimensionalidad PCA para agrupar empresas por similitud financiera. Usamos consensus clustering: solo asignamos una empresa a un grupo si al menos 2 de los 3 algoritmos coinciden.
What it is: We combine three clustering algorithms (K-Means, Agglomerative Clustering and DBSCAN) with PCA dimensionality reduction to group companies by financial similarity. We use consensus clustering: a company is only assigned to a group if at least 2 of 3 algorithms agree.
Para que sirve: Identifica arquetipos de negocio sin prejuicios humanos. En lugar de clasificar subjetivamente, dejamos que los datos revelen que empresas son realmente similares en su perfil financiero.
What it's for: Identifies business archetypes without human bias. Instead of classifying subjectively, we let the data reveal which companies are truly similar in their financial profile.
Conclusion: Emergen 3 arquetipos claros — Quality Compounders (PayPal, Fiserv, Wise: rentables y estables), High Growth / Pre-Profit (Affirm, Lemonade, MoneyLion: creciendo pero quemando caja), y Premium Quality (Adyen, Shift4: margenes excepcionalmente altos). Un inversor puede usar estos arquetipos para construir carteras con el perfil riesgo/retorno que busca.
Conclusion: 3 clear archetypes emerge — Quality Compounders (PayPal, Fiserv, Wise: profitable and stable), High Growth / Pre-Profit (Affirm, Lemonade, MoneyLion: growing but burning cash), and Premium Quality (Adyen, Shift4: exceptionally high margins). An investor can use these archetypes to build portfolios with their desired risk/return profile.
Que es: Regresion panel con efectos fijos por empresa (within-transformation) y regularizacion Ridge. Analiza datos longitudinales: 769 observaciones empresa-trimestre durante 7 anos. La transformacion within elimina las diferencias permanentes entre empresas, aislando que cambios internos impulsan los margenes.
What it is: Panel regression with entity fixed effects (within-transformation) and Ridge regularization. Analyzes longitudinal data: 769 company-quarter observations over 7 years. The within-transformation removes permanent differences between companies, isolating which internal changes drive margins.
Para que sirve: Responde a la pregunta critica: que factores realmente mueven los margenes operativos de una FinTech? No es una correlacion superficial — controla por el tamano, sector y momento temporal de cada empresa.
What it's for: Answers the critical question: what factors actually move a FinTech's operating margins? This is not a superficial correlation — it controls for each company's size, sector and time period.
Conclusion (R² = 0.84): La eficiencia operativa (ratio SGA/Revenue) es el driver #1 de los margenes. Mas importante que el crecimiento de ingresos o el nivel de deuda. Una FinTech que reduce su gasto en ventas y administracion del 60% al 50% de sus ingresos vera un impacto mayor en margenes que una que crece un 20% en revenue manteniendo su estructura de costes.
Conclusion (R² = 0.84): Operational efficiency (SGA/Revenue ratio) is the #1 driver of margins. More important than revenue growth or debt levels. A FinTech that reduces its sales and admin spend from 60% to 50% of revenue will see a bigger margin impact than one growing 20% in revenue while maintaining its cost structure.
Que es: Sistema de scoring de riesgo compuesto que evalua 4 dimensiones: riesgo financiero (solvencia, cobertura de deuda), riesgo operativo (volatilidad de margenes, concentracion de ingresos), riesgo de momentum (tendencia de deterioro en metricas clave) y riesgo anomalo (desviaciones detectadas por Isolation Forest).
What it is: Composite risk scoring system evaluating 4 dimensions: financial risk (solvency, debt coverage), operational risk (margin volatility, revenue concentration), momentum risk (deteriorating trends in key metrics) and anomaly risk (deviations detected by Isolation Forest).
Para que sirve: Ofrece una puntuacion de 0 a 100 por empresa que sintetiza multiples fuentes de riesgo en un solo indicador. Permite comparar rapidamente el perfil de riesgo entre empresas y detectar deterioros antes de que sean evidentes en los titulares.
What it's for: Provides a 0–100 score per company synthesizing multiple risk sources into a single indicator. Enables quick risk comparison across companies and detects deterioration before it becomes headline news.
Conclusion: Las empresas de pagos (Adyen, PayPal, Fiserv) presentan riesgo bajo y estable. Las de lending e insurtech (Upstart, Lemonade) muestran riesgo alto y volatil. Coinbase es un caso especial: alta volatilidad pero mejorando trimestre a trimestre.
Conclusion: Payments companies (Adyen, PayPal, Fiserv) show low, stable risk. Lending and insurtech companies (Upstart, Lemonade) show high, volatile risk. Coinbase is a special case: high volatility but improving quarter over quarter.
Que es: Algoritmo de machine learning no supervisado que detecta observaciones anormales en datos multidimensionales. Funciona construyendo arboles de decision aleatorios: los puntos que se aislan con menos divisiones son anomalos. Lo aplicamos en dos modos: temporal (detecta trimestres anomalos dentro de cada empresa) y cross-sectional (detecta empresas anomalas respecto al sector en cada trimestre).
What it is: Unsupervised machine learning algorithm that detects abnormal observations in multidimensional data. It works by building random decision trees: points isolated with fewer splits are anomalous. We apply it in two modes: temporal (detects anomalous quarters within each company) and cross-sectional (detects anomalous companies relative to the sector each quarter).
Para que sirve: Identifica senales de alerta temprana: un trimestre financieramente atipico puede ser un problema emergente o una oportunidad no reconocida. Tambien valida la calidad de los datos.
What it's for: Identifies early warning signals: a financially atypical quarter may be an emerging problem or an unrecognized opportunity. It also validates data quality.
Conclusion: Las mayores anomalias se concentran en Coinbase (trimestres de boom crypto vs. crash) y en Upstart (colapso del volumen de originacion de prestamos en 2022-2023). Estas anomalias coinciden con eventos de mercado conocidos, validando la fiabilidad del modelo.
Conclusion: The largest anomalies are concentrated in Coinbase (crypto boom vs. crash quarters) and Upstart (loan origination volume collapse in 2022–2023). These anomalies coincide with known market events, validating the model's reliability.
Que es: Cuatro modelos macro que miden como las condiciones macroeconomicas afectan al sector. Incluye: analisis de sensibilidad (correlacion de cada empresa con 11 variables FRED), drivers dependientes de regimen (que factores importan mas en expansion vs. recesion), factor model macroeconomico, y descomposicion de tendencias temporales.
What it is: Four macro models measuring how macroeconomic conditions affect the sector. Includes: sensitivity analysis (each company's correlation with 11 FRED variables), regime-dependent drivers (which factors matter more in expansion vs. recession), macroeconomic factor model, and temporal trend decomposition.
Para que sirve: Un inversor necesita saber si su cartera FinTech es vulnerable a una subida de tipos, a la inflacion o a una recesion. Estos modelos cuantifican esa exposicion con datos historicos reales, no con intuiciones.
What it's for: An investor needs to know whether their FinTech portfolio is vulnerable to rate hikes, inflation or recession. These models quantify that exposure with real historical data, not intuitions.
Conclusion: Resultado contraintuitivo — los tipos de interes altos benefician al sector FinTech en agregado. Razon: las empresas de pagos (que dominan el sector) obtienen rendimientos mas altos sobre los fondos de clientes que custodian (float income). Las perjudicadas son lending e insurtech, pero pesan menos en el sector. La inflacion alta y los spreads de credito amplios si afectan negativamente a todo el sector.
Conclusion: Counterintuitive result — high interest rates benefit the FinTech sector in aggregate. Reason: payments companies (which dominate the sector) earn higher yields on client funds they hold (float income). The losers are lending and insurtech, but they weigh less in the sector. High inflation and wide credit spreads do negatively affect the entire sector.
Que es: Construimos un grafo de correlacion donde cada empresa es un nodo y las aristas representan correlaciones significativas entre sus metricas de rentabilidad. Aplicamos medidas de centralidad (degree, betweenness, eigenvector) y deteccion de comunidades (algoritmo de Louvain) para identificar la estructura de interdependencias del sector.
What it is: We build a correlation graph where each company is a node and edges represent significant correlations between their profitability metrics. We apply centrality measures (degree, betweenness, eigenvector) and community detection (Louvain algorithm) to identify the sector's interdependency structure.
Para que sirve: Revela que empresas se mueven juntas y cuales son independientes — informacion critica para diversificacion. Si dos empresas estan en la misma comunidad de red, anadirlas ambas a tu cartera aporta menos diversificacion de la que parece.
What it's for: Reveals which companies move together and which are independent — critical information for diversification. If two companies are in the same network community, adding both to your portfolio provides less diversification than it appears.
Conclusion: Los procesadores de pagos (PayPal, Adyen, Fiserv, Block) forman un cluster muy conectado. Coinbase y Lemonade son nodos perifericos con baja correlacion con el resto — ofrecen diversificacion real pero con mayor volatilidad individual.
Conclusion: Payments processors (PayPal, Adyen, Fiserv, Block) form a tightly connected cluster. Coinbase and Lemonade are peripheral nodes with low correlation to the rest — they offer real diversification but with higher individual volatility.
Que es: Tecnica estadistica originada en medicina que adaptamos para medir el "time-to-profitability" de las FinTechs. Estima la probabilidad de que una empresa que aun no es rentable alcance la rentabilidad operativa en un periodo determinado, considerando censura por datos (empresas que aun no han alcanzado el evento).
What it is: A statistical technique originated in medicine that we adapt to measure the "time-to-profitability" of FinTechs. It estimates the probability that a company not yet profitable will reach operating profitability within a given period, accounting for data censoring (companies that haven't yet reached the event).
Para que sirve: Para un inversor en empresas growth/pre-profit, esta es la pregunta mas importante: cuanto tiempo necesita esta empresa para ser rentable? Y cual es la probabilidad de que lo logre?
What it's for: For a growth/pre-profit company investor, this is the most important question: how long does this company need to become profitable? And what is the probability it will get there?
Conclusion: Las empresas de pagos alcanzaron rentabilidad en una mediana de 8 trimestres. Las de lending necesitaron 14. Algunas (Lemonade, Marqeta) aun no han cruzado el umbral, lo que la curva de Kaplan-Meier refleja como observaciones censuradas — no fracasos, pero si retrasos significativos.
Conclusion: Payments companies reached profitability in a median of 8 quarters. Lending companies needed 14. Some (Lemonade, Marqeta) have not yet crossed the threshold, which the Kaplan-Meier curve reflects as censored observations — not failures, but significant delays.
Que es: A diferencia de la regresion clasica que modela la media, la regresion cuantilica modela diferentes puntos de la distribucion. Comparamos que factores impulsan los margenes en el percentil 10 (peores empresas) vs. el percentil 90 (mejores), revelando asimetrias ocultas.
What it is: Unlike classical regression that models the mean, quantile regression models different points of the distribution. We compare which factors drive margins at the 10th percentile (worst companies) vs. the 90th (best), revealing hidden asymmetries.
Conclusion: Para las peores empresas del sector, el driver principal es la estructura de costes. Para las mejores, es la capacidad de pricing y el crecimiento de revenue. Implicacion: las estrategias de turnaround deben centrarse en recortar costes, mientras que las empresas lideres deben proteger su poder de pricing.
Conclusion: For the worst companies in the sector, the main driver is cost structure. For the best, it's pricing power and revenue growth. Implication: turnaround strategies should focus on cutting costs, while leading companies should protect their pricing power.
Que es: Dos modelos de machine learning supervisado. El primero es un ensemble stacking (Random Forest + Gradient Boosting + Linear Regression) que intenta predecir margenes futuros. El segundo es un Gradient Boosting Regressor que captura relaciones no lineales entre features y rentabilidad.
What it is: Two supervised machine learning models. The first is an ensemble stacking model (Random Forest + Gradient Boosting + Linear Regression) that attempts to predict future margins. The second is a Gradient Boosting Regressor that captures non-linear relationships between features and profitability.
Transparencia: El modelo ensemble de prediccion tuvo resultados limitados (R² out-of-sample bajo) — documentamos honestamente por que: con solo 26 empresas y alta heterogeneidad sectorial, los modelos ML predictivos no tienen suficientes datos para generalizar. El modelo de drivers no lineales si aporta valor al revelar efectos umbral.
Transparency: The ensemble prediction model had limited results (low out-of-sample R²) — we honestly document why: with only 26 companies and high sector heterogeneity, predictive ML models don't have enough data to generalize. The non-linear drivers model does add value by revealing threshold effects.
Conclusion: Existe un efecto umbral en la eficiencia operativa: por debajo de un ratio SGA/Revenue del 45%, los margenes mejoran linealmente. Por encima, la relacion se aplana — la empresa esta en modo "gastando para crecer" y la eficiencia operativa deja de ser el factor limitante.
Conclusion: There is a threshold effect in operational efficiency: below an SGA/Revenue ratio of 45%, margins improve linearly. Above that, the relationship flattens — the company is in "spending to grow" mode and operational efficiency is no longer the limiting factor.
| Empresa | Subsector | Score | Cluster | |
|---|---|---|---|---|
| Adyen | Payments Infrastructure | 85 | Premium Quality | Ver |
| PayPal | Payments | 78 | Quality Compounders | Ver |
| Fiserv | Payments Infrastructure | 76 | Quality Compounders | Ver |
| Block (Square) | Payments Ecosystem | 72 | High Growth | Ver |
| Wise | Remittances FX | 71 | Quality Compounders | Ver |
| Toast | Vertical FinTech | 70 | High Growth | Ver |
| Shift4 Payments | Payments Processing | 69 | High Growth | Ver |
| SoFi / MoneyLion | Neobank | 65 | High Growth / Pre-Profit | Ver |
| Coinbase | Crypto Exchange | 62 | High Volatility | Ver |
| Affirm | BNPL | 58 | High Growth / Pre-Profit | Ver |
| Upstart | AI Lending | 55 | High Volatility | Ver |
| Lemonade | Insurtech | 48 | High Growth / Pre-Profit | Ver |
Mostrando 12 de 26 empresas. Score basado en modelo compuesto de 20 indicadores.
Showing 12 of 26 companies. Score based on composite model with 20 indicators.
Los informes estan disponibles en espanol e ingles. Selecciona el idioma de descarga.
Reports are available in both Spanish and English. Select your preferred language below.
Analisis orientado a negocio del rendimiento financiero del sector FinTech, dinamicas competitivas e implicaciones de inversion.
A business-oriented analysis of the FinTech sector's financial performance, competitive dynamics, and investment implications.
Executive Summary, Sector Scorecard, Archetypes, Profitability Drivers, Risk & Momentum, Cycle & Macro
Marzo 2026 | .docxExecutive Summary, Sector Scorecard, Archetypes, Profitability Drivers, Risk & Momentum, Cycle & Macro
March 2026 | .docx| Seccion | Descripcion | |
|---|---|---|
| Executive Summary | Numeros clave, estado del sector, top y bottom performers | Key numbers, sector status, top and bottom performers at a glance |
| Sector Introduction | Que es FinTech, tamano de mercado, subsectores, tendencias actuales | What is FinTech, market size, subsectors, current trends (AI, embedded finance, regulation) |
| Company Universe | Perfiles detallados de las 26 empresas | Detailed profiles of all 26 companies with business descriptions |
| Sector Scorecard | Rankings percentiles con 14 metricas clave y visualizacion semaforo | Percentile rankings across 14 key metrics with traffic-light visualization |
| Sector Archetypes | Tres clusters identificados via machine learning | Three company clusters identified via machine learning |
| Profitability Drivers | Factores que realmente impulsan los margenes FinTech | What factors actually drive FinTech margins, with quantile analysis |
| Cycle & Macro | Regimen actual de expansion/contraccion y sensibilidad a tipos, inflacion, VIX | Current expansion/contraction regime and sensitivity to interest rates, inflation, VIX |
| Risk & Momentum | Scores de riesgo (0-100) y seguimiento de momentum fundamental | Risk scores (0–100) for every company and fundamental momentum tracking |
| Temporal Evolution | Tendencias de revenue y margenes, survival analysis | Revenue and margin trends over time, survival analysis |
| Network Analysis | Grafo de red revelando que empresas se mueven juntas | Network graph revealing which companies move together |
| Conclusions | Insights sintetizados y perspectiva 2026-2027 | Synthesized insights and sector outlook for 2026–2027 |
Inmersion profunda en la metodologia, algoritmos y codigo detras del analisis — disenado para data scientists y analistas.
A deep dive into the methodology, algorithms, and code behind the analysis — designed for data scientists, analysts, and anyone who wants to understand how the models work.
Arquitectura, ETL pipeline, Feature Engineering, 20 modelos, validacion, stack tecnologico
Marzo 2026 | .docxArchitecture, ETL pipeline, Feature Engineering, 20 models, validation, tech stack
March 2026 | .docx| Seccion | Descripcion | |
|---|---|---|
| System Architecture | Pipeline de 6 fases, enfoque config-driven, estructura de directorios | 6-phase pipeline design, config-driven approach, directory structure |
| Data Ingestion | ETL pipeline para Excel S&P Capital IQ — parsing, mapping, FX, validacion | ETL pipeline for S&P Capital IQ Excel files — parsing, mapping, FX normalization, validation |
| Feature Engineering | 30 ratios con formulas, z-scores MAD, percentiles, 28 features avanzados | 30 financial ratios with formulas, MAD-based z-scores, percentile normalization, 28 advanced features |
| Clustering (4B) | PCA + K-Means + Agglomerative + DBSCAN con consensus clustering | PCA + K-Means + Agglomerative + DBSCAN with consensus clustering |
| Ridge Panel Regression (4C) | Within-transformation, efectos fijos, Ridge, cross-validation | Within-transformation for entity fixed effects, Ridge regularization, cross-validation |
| Ensemble Predictor (4H) | Stacking (RF + GBR + LR) — incluye analisis honesto de por que fallo | Stacking (RF + GBR + LR) — including honest analysis of why it failed |
| Macro Models (4J-4M) | Sensibilidad, drivers dependientes de regimen, factor model, descomposicion | Sensitivity analysis, regime-dependent drivers, macro factor model, trend decomposition |
| Network Analysis (4O) | Teoria de grafos — centralidad y deteccion de comunidades | Graph theory — centrality measures and community detection |
| Survival Analysis (4S) | Estimacion Kaplan-Meier del time-to-profitability | Kaplan-Meier estimation of time-to-profitability |
El analisis sigue un pipeline de 6 fases, completamente automatizado y reproducible. Cada fase transforma los datos del paso anterior, desde los archivos crudos de Excel hasta los informes finales.
Todo el pipeline esta parametrizado por un archivo de configuracion JSON. Para analizar un nuevo sector, basta con crear un nuevo config y ejecutar el mismo codigo.
The analysis follows a 6-phase pipeline, fully automated and reproducible. Each phase transforms data from the previous step, from raw Excel files to final reports.
The entire pipeline is parameterized by a JSON configuration file. To analyze a new sector, simply create a new config and run the same code.
Parseo de 78 archivos Excel de S&P Capital IQ (3 por empresa: Income Statement, Balance Sheet, Cash Flow). El pipeline detecta automaticamente el formato, mapea las columnas al esquema estandar, normaliza divisas (GBP, CNY → USD) y valida la integridad de los datos. El resultado es un dataset unificado en formato long con 127,927 observaciones.
Parsing of 78 Excel files from S&P Capital IQ (3 per company: Income Statement, Balance Sheet, Cash Flow). The pipeline automatically detects the format, maps columns to the standard schema, normalizes currencies (GBP, CNY → USD) and validates data integrity. The result is a unified long-format dataset with 127,927 observations.
Descarga de 11 indicadores macroeconomicos de la Federal Reserve (FRED): GDP growth, CPI, Fed Funds Rate, Unemployment, VIX, Credit Spreads, Consumer Confidence, M2 Money Supply, Industrial Production, 10Y Treasury Yield y Housing Starts. Cada variable se alinea al trimestre fiscal de cada empresa y se fusiona con el feature store.
Download of 11 macroeconomic indicators from the Federal Reserve (FRED): GDP growth, CPI, Fed Funds Rate, Unemployment, VIX, Credit Spreads, Consumer Confidence, M2 Money Supply, Industrial Production, 10Y Treasury Yield and Housing Starts. Each variable is aligned to each company's fiscal quarter and merged with the feature store.
Construccion del feature store con 242 variables por empresa-trimestre. Incluye: 30 ratios financieros (margenes, solvencia, eficiencia, liquidez), 7 metricas de crecimiento (YoY, QoQ), rankings percentiles intra-sector, z-scores robustos basados en MAD (Median Absolute Deviation), y 28 features avanzados (lags temporales, rolling means, interacciones entre variables). Cada feature tiene formula documentada y justificacion economica.
Construction of the feature store with 242 variables per company-quarter. Includes: 30 financial ratios (margins, solvency, efficiency, liquidity), 7 growth metrics (YoY, QoQ), intra-sector percentile rankings, robust z-scores based on MAD (Median Absolute Deviation), and 28 advanced features (temporal lags, rolling means, variable interactions). Each feature has a documented formula and economic justification.
Ejecucion de 20 modelos analiticos en tres categorias: 9 modelos base (scorecard, clustering, drivers, risk, anomaly, momentum, growth decomposition, margin analysis, peer ranking), 4 modelos macro (sensitivity, regime-dependent drivers, factor model, trend decomposition), y 7 modelos avanzados (network analysis, survival, quantile regression, rolling analytics, HHI concentration, ML ensemble, non-linear drivers). Detalle completo en la tab "Analisis".
Execution of 20 analytical models in three categories: 9 base models (scorecard, clustering, drivers, risk, anomaly, momentum, growth decomposition, margin analysis, peer ranking), 4 macro models (sensitivity, regime-dependent drivers, factor model, trend decomposition), and 7 advanced models (network analysis, survival, quantile regression, rolling analytics, HHI concentration, ML ensemble, non-linear drivers). Full detail in the "Analysis" tab.
Generacion automatica de 13 visualizaciones: heatmap de scorecard, scatter de clusters, ciclo sectorial, mapa de riesgo, momentum chart, growth decomposition, margin evolution, network graph, quantile comparison, rolling analytics, survival curves, HHI concentration y macro sensitivity.
Automatic generation of 13 visualizations: scorecard heatmap, cluster scatter, sector cycle, risk map, momentum chart, growth decomposition, margin evolution, network graph, quantile comparison, rolling analytics, survival curves, HHI concentration and macro sensitivity.
Produccion automatica de 4 informes en Word (.docx): Informe Economico (ES + EN) orientado a inversores y decision-makers, e Informe Tecnico (ES + EN) orientado a data scientists y analistas cuantitativos. Cada informe incluye las visualizaciones generadas en Phase 5.
Automatic production of 4 Word reports (.docx): Economic Report (ES + EN) oriented to investors and decision-makers, and Technical Report (ES + EN) oriented to data scientists and quantitative analysts. Each report includes the visualizations generated in Phase 5.
Language: Python 3.11
Data: pandas, numpy, openpyxl
ML: scikit-learn (clustering, ensemble, isolation forest)
Statistics: statsmodels (panel regression, quantile regression)
Graphs: networkx (network analysis, community detection)
Survival: lifelines (Kaplan-Meier estimation)
Last updated: March 2026 | Pipeline fully reproducible via master_runner.py