Lesson 1Deployment considerations: retraining cadence, data pipelines, monitoring drift and backtesting forecastsPlan deployment a revenue forecastin systems, includin retrainin cadence, automated data pipelines, monitorin fi data an concept drift, an continuous backtestin to ensure stable performance in changin market conditions.
Designin robust data pipelinesSchedulin retrainin an updatesMonitorin data an concept driftOngoing backtestin an benchmarkinAlertin an rollback strategiesLesson 2Communicating forecasts to stakeholders: visualization of point forecasts and intervals, scenario analysis, transparency of assumptionsCommunicate revenue forecasts effective to stakeholders usin clear visualizations a point an interval predictions, scenario analysis, an transparent documentation a assumptions, limitations, an model risk considerations.
Visualizin point an interval forecastsScenario an what-if analysis designExplaining drivers an key featuresDocumentin assumptions an limitsTailorin messages to stakeholdersLesson 3Machine learning time series models: random forests/gradient boosting with lagged features, XGBoost/LightGBM, and sequence models (LSTM/GRU)Train machine learnin models fi time series revenue forecastin, includin tree ensembles wid lagged features an sequence models like LSTM an GRU, while handlin nonstationarity, seasonality, an product-level heterogeneity.
Random forests wid lagged featuresGradient boostin, XGBoost, LightGBMGlobal versus local forecastin modelsSequence models wid LSTM an GRUHandlin nonstationarity an scalinLesson 4Formulating forecasting objectives and evaluation horizons (e.g., next 3, 6, 12 months)Define forecastin goals fi core bank products by selectin prediction targets, horizons, an granularity, an align dem wid business decisions such as budgetin, pricin, liquidity plannin, an regulatory or risk reportin needs.
Choosin revenue targets an unitsSelectin forecast horizons an frequencyAlignin forecasts wid business decisionsGranularity by product, segment, an regionHandlin new products an short historiesLesson 5Identifying and sourcing time series data (public financial series, payment volumes, synthetic generation techniques)Learn how fi identify, assess, an source time series data fi bank revenue forecastin, includin internal product metrics, public financial series, an synthetic data dat safely augments scarce or noisy historical records.
Catalogin internal product revenue seriesUsin public macro an market data sourcesCollectin payment an transaction volume dataAssessin data quality, gaps, an revisionsSynthetic data generation fi stress scenariosLesson 6Training and hyperparameter tuning: grid/random search, Bayesian optimization, time-aware scoringOptimize model performance usin structured hyperparameter tunin strategies, includin grid an random search, Bayesian optimization, an time-aware scorin dat respects temporal orderin an focuses on business-critical horizons.
Definin search spaces an priorsGrid an random search trade-offsBayesian optimization workflowsTime-aware validation an scorinEarly stoppin an resource limitsLesson 7Model ensembling and reconciliation: simple model averaging, weighted ensembles, stacking for time seriesCombine multiple forecastin models fi bank revenue usin simple averages, weighted ensembles, an stackin, an apply hierarchical reconciliation to ensure coherent forecasts across products, branches, an organizational levels.
Simple an weighted model averaginStackin an meta-learners fi seriesDiversity an correlation among modelsHierarchical an grouped reconciliationEvaluatin ensemble stability over timeLesson 8Baseline time series methods: ARIMA, ETS, naive and seasonal naïve models, decomposition (trend/seasonality)Explore baseline time series models fi bank revenue, includin naive, seasonal naive, ARIMA, ETS, an decomposition, to establish reference performance an interpret trend an seasonality before usin complex machine learnin models.
Naive an seasonal naive benchmarksClassical decomposition a trend an seasonalityARIMA modelin fi bank revenue seriesExponential smoothin an ETS variantsComparin baselines across productsLesson 9Feature engineering for revenue: lags, rolling means/std, differencing, calendar effects, holiday indicators, cohort effects, marketing/campaign flagsEngineer predictive features fi bank revenue, includin lags, rollin statistics, differencin, calendar an holiday effects, cohort an lifecycle indicators, an marketin or campaign flags dat capture demand shifts an structural breaks.
Lag an lead features fi revenueRollin means, volatility, an ratiosCalendar, holiday, an payday effectsCohort an lifecycle based featuresMarketin an campaign impact flagsLesson 10Data splitting and cross-validation for time series: train/validation/test splits, expanding window CV, blocked CVDesign time-aware data splits an cross-validation schemes fi revenue forecastin, includin rollin an blocked approaches, to avoid leakage, mimic production use, an obtain reliable estimates a model performance over time.
Holdout train, validation, an test splitsRollin an expandin window validationBlocked cross-validation fi seasonalityPreventin temporal leakage in featuresBacktestin over multiple forecast originsLesson 11Evaluation metrics and error analysis: MAE, RMSE, MAPE, symmetric MAPE, prediction intervals and coverageEvaluate revenue forecasts usin metrics such as MAE, RMSE, MAPE, symmetric MAPE, an interval coverage, an perform detailed error analysis by segment, horizon, an regime to uncover biases an model weaknesses.
Scale-dependent error metricsPercentage an relative error metricsPrediction intervals an coverageHorizon an segment level diagnosticsRegime an event-driven error analysisLesson 12Macro and external regressors: using CPI, unemployment, interest rates, mobility, Google Trends; feature selection and lag alignmentIncorporate macroeconomic an external regressors into revenue models, such as CPI, unemployment, interest rates, mobility, an search trends, an learn techniques fi lag alignment, scalin, an feature selection to avoid overfitting.
Selectin relevant macro indicatorsAlignin lags between macro an revenueTransformin an scalin external dataFeature selection an regularizationStress an scenario overlays wid macros
