Forecasting using shapes - a new frontier

Transformer based Deep Learning architectures have shown immense promise in Generative AI use cases, such as text and image generation. 

Now, for the first time, ForegenAI is pioneering a new frontier use case of the same underlying breakthroughs to advance the science of Forecasting. One can call this new frontier as Shape-driven Forecasting.

Why Use Shapes for Forecasting? 

The modern kingpin of Forecasting is XgBoost (and variants) which is a Machine Learning model based on a gradient boosting framework. However, as practitioners know, these models are extremely limited both in versatility and scope. Most often, the forecasts generated by these models is reliable for one or two time periods at most. The usefulness of the forecasts decays extremely rapidly for longer term forecasts.

On the other hand, our breakthrough shape-driven forecasting recognizes that future evolves in patterns. Metrics (such as equity prices, sales volumes, customer demand etc.) may seem random in very short time frames, but evolve in specific shapes and patterns over the longer term. For example, expert traders know how prices affect Bollinger Bands, and how Bollinger Bands, in turn, constrain prices.

The challenge so far has been that there existed no Machine Learning or AI models that could discern shapes and patterns and make forecasts on shape evolution. However, the very recent AI breakthroughs leveraged by our our technology makes shape forecasting not only practical, but extremely compelling. Suffice to say that that what we have accomplished (see demo) was simply not possible just a few months ago. And, we are barely scratching the surface!

Since we rely on shape-driven forecasting, and not on numbers-driven forecasting, our forecasting is extremely reliable for much longer time frames, often weeks ahead.