# Nixtla blog > Technical articles, product announcements, and forecasting tutorials. Canonical HTML: https://www.nixtla.io/blog ## What the blog covers - TimeGPT product announcements and implementation guides - Forecasting and anomaly detection tutorials - Statistical, machine learning, and neural forecasting research - Benchmarking, evaluation, and production engineering - Customer and ecosystem integrations ## How to read articles Each article is available as canonical HTML and as complete Markdown. Use the Markdown URL for programmatic reading and the canonical HTML URL when citing the article. Article Markdown preserves headings, prose, code, and public image references. ## Article index 26 public articles, ordered by publication date from newest to oldest. ### [Categorical Exogenous Variables in NeuralForecast: Embeddings, Configuration and a Hands On Example](https://www.nixtla.io/blog/neuralforecast-exogenous-features) Learn how NeuralForecast turns categorical exogenous variables into learned embeddings, why that beats one hot or ordinal encoding, and how to configure it for your own models. - Published: 2026-08-21 - Author: Marco Peixeiro, Applied AI Scientist, Core maintainer of NeuralForecast - Categories: NeuralForecast - Tags: neuralforecast, time-series, forecasting, python, deep-learning, nixtla - Estimated reading time: 11 minutes - [Canonical HTML](https://www.nixtla.io/blog/neuralforecast-exogenous-features) - [Markdown](https://www.nixtla.io/blog/neuralforecast-exogenous-features.md) - [Header/preview image](https://www.nixtla.io/blog/images/neuralforecast-exogenous-features/editorial-forecast-closeup.webp) ### [Beyond Accuracy: What Makes a Good Forecast?](https://www.nixtla.io/blog/beyond-accuracy) Accuracy is essential, but a good enterprise forecast must also support real decisions, use the right information, earn trust, and operate reliably at scale. - Published: 2026-07-20 - Author: Nixtla Team, Nixtla - Categories: TimeGPT Forecasting - Tags: TimeGPT, Nixtla Engine, enterprise forecasting, forecast accuracy, forecast workflow - Estimated reading time: 3 minutes - [Canonical HTML](https://www.nixtla.io/blog/beyond-accuracy) - [Markdown](https://www.nixtla.io/blog/beyond-accuracy.md) - [Header/preview image](https://www.nixtla.io/blog/images/beyond-accuracy/editorial-forecast-closeup.webp) ### [TimeGPT Is Now on Microsoft Foundry: From Deploy to First Forecast](https://www.nixtla.io/blog/timegpt-azure-marketplace) TimeGPT is now available through Microsoft Foundry, letting Azure customers deploy it inside their own subscription, billed as standard Azure consumption, without a separate vendor review. - Published: 2026-07-15 - Author: Nixtla Team, Nixtla - Categories: Time Series Forecasting - Tags: Microsoft Foundry, Azure Marketplace, TimeGPT-2, enterprise forecasting, cloud deployment - Estimated reading time: 3 minutes - [Canonical HTML](https://www.nixtla.io/blog/timegpt-azure-marketplace) - [Markdown](https://www.nixtla.io/blog/timegpt-azure-marketplace.md) - [Header/preview image](https://www.nixtla.io/blog/images/timegpt-azure-marketplace/editorial-forecast-closeup.webp) ### [Still using Prophet or AutoARIMA for forecasting?](https://www.nixtla.io/blog/automl-vs-timegpt) We benchmarked TimeGPT-2.1 against Databricks AutoML on electricity, weather, and retail demand datasets, comparing accuracy, runtime, and per-series win rates. TimeGPT-2.1 improved MAE, RMSE, MAPE, and WAPE across all three benchmarks. - Published: 2026-05-28 - Author: Yibei Hu, Data Scientist - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT, Databricks, AutoML, benchmark, electricity forecasting, weather forecasting, retail forecasting - Estimated reading time: 9 minutes - [Canonical HTML](https://www.nixtla.io/blog/automl-vs-timegpt) - [Markdown](https://www.nixtla.io/blog/automl-vs-timegpt.md) - [Header/preview image](https://www.nixtla.io/blog/images/automl-benchmark/editorial-forecast-closeup.webp) ### [Forecasting 145,000 Time Series in 7 Minutes with TimeGPT](https://www.nixtla.io/blog/wiki-traffic-forecast) TimeGPT forecasts all 145,063 Wikipedia web traffic series in 6.6 minutes, beating a top Kaggle statistical baseline by 18.8% on SMAPE, with zero feature engineering. - Published: 2026-05-15 - Author: Nixtla Team, Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT, Wikipedia, zero-shot forecasting, SMAPE, Kaggle, Python - Estimated reading time: 6 minutes - [Canonical HTML](https://www.nixtla.io/blog/wiki-traffic-forecast) - [Markdown](https://www.nixtla.io/blog/wiki-traffic-forecast.md) - [Header/preview image](https://www.nixtla.io/blog/images/wiki-traffic-forecast/editorial-forecast-closeup.webp) ### [TimeGPT in Snowflake Just Got a Full Upgrade: Anomaly Detection, Explainability, and a One Command Install](https://www.nixtla.io/blog/timegpt-snowflake-integration) The Nixtla Snowflake integration is now part of the official nixtla package. Run forecasting, anomaly detection, SHAP based explainability, and evaluation, all from pure SQL, all inside Snowflake. - Published: 2026-02-23 - Author: Nixtla Team, Nixtla - Estimated reading time: 6 minutes - [Canonical HTML](https://www.nixtla.io/blog/timegpt-snowflake-integration) - [Markdown](https://www.nixtla.io/blog/timegpt-snowflake-integration.md) - [Header/preview image](https://www.nixtla.io/blog/images/timegpt-snowflake-integration/editorial-forecast-closeup.webp) ### [Nixtla Enterprise Expands with Leading Foundation Models, MCP, and Agentic Capabilities](https://www.nixtla.io/blog/genai-announcement) This release introduces three major capabilities that together expand Nixtla from a single-model offering into a full time series intelligence platform - Published: 2025-12-18 - Author: Han Wang, CTO - Nixtla - Categories: Artificial Intelligence - Tags: TimeGPT-2, Nixtla MCP, foundation models, enterprise forecasting, private preview - Estimated reading time: 3 minutes - [Canonical HTML](https://www.nixtla.io/blog/genai-announcement) - [Markdown](https://www.nixtla.io/blog/genai-announcement.md) - [Header/preview image](https://www.nixtla.io/blog/images/genai-announcement/editorial-forecast-closeup.webp) ### [TimeGPT 2.1: The Next Generation of Foundation Models for Time Series Forecasting](https://www.nixtla.io/blog/timegpt-2-1-announcement) Announcing the private preview of TimeGPT-2.1, the first multivariate model in the TimeGPT family. - Published: 2025-12-10 - Author: Cristian Challu, Co-Founder - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT-2, foundation models, enterprise forecasting, private preview - Estimated reading time: 2 minutes - [Canonical HTML](https://www.nixtla.io/blog/timegpt-2-1-announcement) - [Markdown](https://www.nixtla.io/blog/timegpt-2-1-announcement.md) - [Header/preview image](https://www.nixtla.io/blog/images/timegpt-2-1-announcement/editorial-forecast-closeup.webp) ### [Supercharge Your Sales Forecasts: A Complete Guide to Exogenous Variables in MLForecast](https://www.nixtla.io/blog/mlforecast-exogenous-variables) Learn how to incorporate external factors like prices, promotions, and calendar patterns into your time series forecasts using MLForecast's exogenous variables. - Published: 2025-12-05 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: MLForecast - Tags: mlforecast, time-series, forecasting, python, lightgbm, nixtla - Estimated reading time: 15 minutes - [Canonical HTML](https://www.nixtla.io/blog/mlforecast-exogenous-variables) - [Markdown](https://www.nixtla.io/blog/mlforecast-exogenous-variables.md) - [Header/preview image](https://www.nixtla.io/blog/images/mlforecast-exogenous-variables/editorial-forecast-closeup.webp) ### [Forecasting Championship Results using Time Series and Nixtla](https://www.nixtla.io/blog/championship-forecasting) Learn how to forecast championship standings using Nixtla's StatsForecast library. - Published: 2025-11-28 - Author: Piero Paialunga, Data Scientist - Categories: TimeGPT Forecasting - Tags: StatsForecast, forecasting, championship, sports analytics, time series, AutoARIMA - Estimated reading time: 34 minutes - [Canonical HTML](https://www.nixtla.io/blog/championship-forecasting) - [Markdown](https://www.nixtla.io/blog/championship-forecasting.md) - [Header/preview image](https://www.nixtla.io/blog/images/championship_forecasting/editorial-forecast-closeup.webp) ### [Automatic Model Selection with StatsForecast for Time Series Forecasting](https://www.nixtla.io/blog/statsforecast-automatic-model-selection) Stop testing statistical models manually. Use StatsForecast to automatically fit AutoARIMA, AutoETS, AutoCES, and AutoTheta models, then select the best performer for each series through cross-validation. - Published: 2025-11-20 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: StatsForecast, automatic model selection, AutoARIMA, AutoETS, cross-validation - Estimated reading time: 30 minutes - [Canonical HTML](https://www.nixtla.io/blog/statsforecast-automatic-model-selection) - [Markdown](https://www.nixtla.io/blog/statsforecast-automatic-model-selection.md) - [Header/preview image](https://www.nixtla.io/blog/images/statsforecast-automatic-model-selection/editorial-forecast-closeup.webp) ### [Damage Detection in Engineering Structures Using Nixtla](https://www.nixtla.io/blog/damagedetection) Learn how to detect cracks and structural damage using Nixtla's Anomaly Detection pipeline while accounting for temperature-induced variations in sensor data. - Published: 2025-10-15 - Author: Piero Paialunga, Data Scientist - Categories: TimeGPT Anomalies - Tags: TimeGPT, anomaly detection, structural health monitoring, damage detection, NDT, SHM, engineering structures, synthetic data, Python - Estimated reading time: 24 minutes - [Canonical HTML](https://www.nixtla.io/blog/damagedetection) - [Markdown](https://www.nixtla.io/blog/damagedetection.md) - [Header/preview image](https://www.nixtla.io/blog/images/damage_detection/editorial-forecast-closeup.webp) ### [TimeGPT 2: The Next Generation of Foundation Models for Time Series Forecasting](https://www.nixtla.io/blog/timegpt-2-announcement) Announcing the private preview of TimeGPT-2 Mini, TimeGPT-2, and TimeGPT-2 Pro, enterprise grade foundation models with up to 60% accuracy improvement, built for mission-critical time series forecasting. - Published: 2025-10-14 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT-2, foundation models, enterprise forecasting, private preview - Estimated reading time: 2 minutes - [Canonical HTML](https://www.nixtla.io/blog/timegpt-2-announcement) - [Markdown](https://www.nixtla.io/blog/timegpt-2-announcement.md) - [Header/preview image](https://www.nixtla.io/blog/images/timegpt-2-announcement/editorial-forecast-closeup.webp) ### [Anomaly Detection for Cloud Cost Monitoring with Nixtla](https://www.nixtla.io/blog/anomalyinproduction) Learn how to build a synthetic cloud cost dataset and use Nixtla's algorithms to detect spikes, drifts, and level shifts. This approach helps teams monitor performance and prevent unexpected billing surprises. - Published: 2025-09-26 - Author: Piero Paialunga, Data Scientist - Categories: Anomaly Detection - Tags: TimeGPT, anomaly detection, cloud cost monitoring, Python - Estimated reading time: 16 minutes - [Canonical HTML](https://www.nixtla.io/blog/anomalyinproduction) - [Markdown](https://www.nixtla.io/blog/anomalyinproduction.md) - [Header/preview image](https://www.nixtla.io/blog/images/anomaly_detection_monitoring/editorial-forecast-closeup.webp) ### [Anomaly Detection in Time Series with TimeGPT and Python](https://www.nixtla.io/blog/anomaly-detection) Discover how to use TimeGPT for scalable, accurate anomaly detection in Python. Includes real-world time series, exogenous variables, and adjustable confidence levels. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Anomaly Detection - Tags: TimeGPT, anomaly detection, Python, confidence intervals - Estimated reading time: 10 minutes - [Canonical HTML](https://www.nixtla.io/blog/anomaly-detection) - [Markdown](https://www.nixtla.io/blog/anomaly-detection.md) - [Header/preview image](https://www.nixtla.io/blog/images/anomaly_detection/editorial-forecast-closeup.webp) ### [Performance Evaluation of Anomaly Detection through Synthetic Anomalies](https://www.nixtla.io/blog/anomaly-detection-performance-evaluation) Discover how to find the minimum detectable anomaly in absence of a ground truth labelled dataset using synthetic anomalies. - Published: 2025-08-26 - Author: Piero Paialunga, Data Scientist - Categories: Anomaly Detection - Tags: TimeGPT, anomaly detection, synthetic anomalies, performance evaluation, minimum detectable anomaly - Estimated reading time: 14 minutes - [Canonical HTML](https://www.nixtla.io/blog/anomaly-detection-performance-evaluation) - [Markdown](https://www.nixtla.io/blog/anomaly-detection-performance-evaluation.md) - [Header/preview image](https://www.nixtla.io/blog/images/anomaly_detection_performance_evaluation/editorial-forecast-closeup.webp) ### [Automated Time Series Feature Engineering with MLforecast](https://www.nixtla.io/blog/automated-time-series-feature-engineering-with-mlforecast) Replace hours of custom feature engineering code with MLforecast's automated lag features, rolling statistics, and target transformations for faster, more reliable time series forecasting. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: MLforecast, automated feature engineering, lag features, target transformations - Estimated reading time: 17 minutes - [Canonical HTML](https://www.nixtla.io/blog/automated-time-series-feature-engineering-with-mlforecast) - [Markdown](https://www.nixtla.io/blog/automated-time-series-feature-engineering-with-mlforecast.md) - [Header/preview image](https://www.nixtla.io/blog/images/automated-time-series-feature-engineering-with-mlforecast/editorial-forecast-closeup.webp) ### [Effortless Accuracy Unlocking the Power of Baseline Forecasts](https://www.nixtla.io/blog/baseline-forecasts) Understand what are baseline forecasts, why they are important and learn to create them easily with Nixtla's statsforecast package. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: baseline forecasting, statsforecast, naive forecast, seasonal naive - Estimated reading time: 12 minutes - [Canonical HTML](https://www.nixtla.io/blog/baseline-forecasts) - [Markdown](https://www.nixtla.io/blog/baseline-forecasts.md) - [Header/preview image](https://www.nixtla.io/blog/images/baseline_forecasts/editorial-forecast-closeup.webp) ### [Eliminate Manual ARIMA Tuning Using StatsForecast AutoARIMA Automation](https://www.nixtla.io/blog/eliminate-manual-arima-tuning-using-statsforecast-autoarima-automation) Eliminate weeks of manual ARIMA parameter tuning with StatsForecast's AutoARIMA. Automatically select optimal model parameters for 50+ time series with confidence intervals in under 30 minutes. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: StatsForecast, AutoARIMA, automatic model selection, ARIMA parameters, prediction intervals - Estimated reading time: 14 minutes - [Canonical HTML](https://www.nixtla.io/blog/eliminate-manual-arima-tuning-using-statsforecast-autoarima-automation) - [Markdown](https://www.nixtla.io/blog/eliminate-manual-arima-tuning-using-statsforecast-autoarima-automation.md) - [Header/preview image](https://www.nixtla.io/blog/images/eliminate-manual-arima-tuning-using-statsforecast-autoarima-automation/editorial-forecast-closeup.webp) ### [Time Series Frequency Modelling with Fourier Transform and TimeGPT-1](https://www.nixtla.io/blog/frequency-ts-modelling) Discover how to decompose your time series in multiple components with Fourier Transform and model each component with TimeGPT-1. - Published: 2025-08-26 - Author: Piero Paialunga, Data Scientist - Categories: Time Series Forecasting - Tags: TimeGPT, Fourier Transform, frequency analysis, signal decomposition - Estimated reading time: 18 minutes - [Canonical HTML](https://www.nixtla.io/blog/frequency-ts-modelling) - [Markdown](https://www.nixtla.io/blog/frequency-ts-modelling.md) - [Header/preview image](https://www.nixtla.io/blog/images/fourier_modelling/editorial-forecast-closeup.webp) ### [Understanding Intermittent Demand](https://www.nixtla.io/blog/intermittent-demand) Learn how to forecast intermittent demand using Python and Nixtla's TimeGPT. This step-by-step guide covers handling sparse time series, fine-tuning, and using exogenous variables to improve accuracy. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT, intermittent demand, sparse data, fine-tuning - Estimated reading time: 16 minutes - [Canonical HTML](https://www.nixtla.io/blog/intermittent-demand) - [Markdown](https://www.nixtla.io/blog/intermittent-demand.md) - [Header/preview image](https://www.nixtla.io/blog/images/intermittent_demand/editorial-forecast-closeup.webp) ### [Long Term Mid Term and Short Term Forecasting with Polynomial Regression AutoARIMA and TimeGPT-1](https://www.nixtla.io/blog/multilevelforecasting) Learn how to match forecasting models to your time horizon for better accuracy. Compare polynomial regression for long-term trends, AutoARIMA for mid-term cycles, and TimeGPT-1 for short-term predictions using real currency exchange data. Includes code examples for multi-horizon forecasting strategies. - Published: 2025-08-26 - Author: Piero Paialunga, Data Scientist - Categories: Time Series Forecasting - Tags: TimeGPT, AutoARIMA, multi-horizon forecasting, polynomial regression - Estimated reading time: 15 minutes - [Canonical HTML](https://www.nixtla.io/blog/multilevelforecasting) - [Markdown](https://www.nixtla.io/blog/multilevelforecasting.md) - [Header/preview image](https://www.nixtla.io/blog/images/MultiLevelForecasting/editorial-forecast-closeup.webp) ### [Simple Anomaly Detection in Time Series via Optimal Baseline Subtraction (OBS)](https://www.nixtla.io/blog/optimal-baseline) Discover how to detect anomalies using Optimal Baseline Subtraction and enhance your forecasts with Nixtla’s TimeGPT on real-world weather data. - Published: 2025-08-26 - Author: Piero Paialunga, Data Scientist - Categories: Anomaly Detection - Tags: anomaly detection, optimal baseline subtraction, OBS, statsforecast - Estimated reading time: 18 minutes - [Canonical HTML](https://www.nixtla.io/blog/optimal-baseline) - [Markdown](https://www.nixtla.io/blog/optimal-baseline.md) - [Header/preview image](https://www.nixtla.io/blog/images/optimal_baseline/editorial-forecast-closeup.webp) ### [Savitzky Golay Filtering for Time Series Denoising](https://www.nixtla.io/blog/polynomial-filtering) Denoise your time series with polynomial smoothing using the Savitzky-Golay filter - Published: 2025-08-26 - Author: Piero Paialunga, Data Scientist - Estimated reading time: 14 minutes - [Canonical HTML](https://www.nixtla.io/blog/polynomial-filtering) - [Markdown](https://www.nixtla.io/blog/polynomial-filtering.md) - [Header/preview image](https://www.nixtla.io/blog/images/Polynomial_Filtering/editorial-forecast-closeup.webp) ### [Production-Ready Forecasting Pipeline with TimeGPT and Polars](https://www.nixtla.io/blog/production-ready-forecasting-pipeline-with-timegpt-and-polars) Learn how TimeGPT's native DataFrame compatibility lets you leverage Polars' blazing-fast performance for time series forecasting without data conversion overhead. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT, Polars, zero-shot forecasting, DataFrame libraries, scalable forecasting - Estimated reading time: 22 minutes - [Canonical HTML](https://www.nixtla.io/blog/production-ready-forecasting-pipeline-with-timegpt-and-polars) - [Markdown](https://www.nixtla.io/blog/production-ready-forecasting-pipeline-with-timegpt-and-polars.md) - [Header/preview image](https://www.nixtla.io/blog/images/production-ready-forecasting-pipeline-with-timegpt-and-polars/editorial-forecast-closeup.webp) ### [TimeGPT vs Snowflake - 50x Faster Forecasting with Better Accuracy](https://www.nixtla.io/blog/timegpt-in-snowflake) Discover SQL-native time series forecasting for Snowflake that's 50x faster than native tools. Nixtla provides state-of-the-art accuracy without Python, ML infrastructure, or complex setup. - Published: 2025-08-26 - Author: Khuyen Tran, Developer Advocate - Nixtla - Categories: Time Series Forecasting - Tags: TimeGPT, Snowflake, SQL forecasting, performance comparison - Estimated reading time: 11 minutes - [Canonical HTML](https://www.nixtla.io/blog/timegpt-in-snowflake) - [Markdown](https://www.nixtla.io/blog/timegpt-in-snowflake.md) - [Header/preview image](https://www.nixtla.io/blog/images/timegpt_in_snowflake/editorial-forecast-closeup.webp) ## Agent resources - [Site guide](https://www.nixtla.io/llms.txt) - [Complete reference](https://www.nixtla.io/llms-full.txt) - [Agent instructions](https://www.nixtla.io/agents.md)