Zero-Shot and Ahead of the Pack: What Amazon Chronos Taught Us About the Future of Demand Forecasting

Foundation Model Amazon Chronos on Airline data 

We tested the Chronos Foundation model with just one simple time series data set that is available everywhere – the Airline Passenger traffic data from the last century.  
 
Chronos is a Time Series foundation model from Amazon.  
 
As you can see, Chronos-2 pretty much outperformed everything else including the Bolt and original. We also compared this to a traditional Holt Winters and Sarima expert selection from popular commercial software.  
 
Chronos-2 turned in a WMAPE of 4.16% and Normalized RMSE of 4.9%.  
 
Zero shot forecasting? Impressive. Although this is a highly forecastable series, the fact that it raced and outperformed everything else is indeed fantastic.  
 
Interesting to note that no training or supervision is needed to get this forecast result.  
 
Chronos is probably trained on this data ad nauseum – it has the Airline passenger traffic data etched in its core memory from time immemorial until now…. Perhaps it already knows the future. 
 
Why forecast when you already know the future? 
 
The real meat in the pudding will be known when we start using actual business data to test this – “sh*tty” data that is! 
 
Stay tuned for an update on how commoners like us ordinary business users can make use of Ai and several foundation models to make forecasting a bit simpler. 

Curious how AI-driven forecasting could fit into your demand planning process?  Contact the Valtitude team to explore what’s possible — and explore how our demand planning services can help your organization get ahead of the curve.