How does trend forecasting work?

We are able to forecast the future behavior of fashion trends up to 24 months in advance. We can do it by using a mix of historical data, advanced algorithms, and deep learning.

This is how it works:

1. We start with the past
   For each trend, we gather up to 8 years of historical data, including all associated social media posts and extra signals relevant to its evolution.

2. We run multiple models
   We apply five different statistical models, along with our proprietary deep learning model, to predict how the trend is likely to evolve over the next year.

3. We combine the results
   Finally, we use an ensembling technique (called Fforma) to combine the outputs of all models into a single, optimized forecast.