Rethinking Time Series Models: It's Time to Get Real

Time series forecasting has been taken over by deep learning models. They're everywhere, showing up in everything from weather predictions to stock price analysis. But here's the catch: despite the flood of new architectures, we still don't really know what makes these models tick.

What's Missing in Time Series Models?

There's a problem lurking beneath the surface of all these models. we've a swarm of architectures, but benchmarking them has become a messy affair. It's like comparing apples to oranges. The nuances and design elements that truly drive performance are often glossed over, leading to conflicting results.

This oversight isn't just a minor glitch. The real issue is that critical design dimensions, which might seem like mere details, can actually have a massive impact. Imagine treating these nuances like side notes when they could shift the entire framework of the forecasting method.

Time to Rethink the Approach

Why should we care? Because without a clear understanding of what works and what doesn't, progress is stalled. We're spinning wheels without. Ask yourself: Can we afford to remain in this state of confusion?

Here's a bold take. Let's strip down the process to its core. Focus on the foundational aspects of the problem rather than being dazzled by new layers and flashy components. It's about understanding what's essential and not getting distracted by the fluff.

A New Way Forward

To shake things up, there's a proposal for a template called the 'auxiliary forecasting model card.' Think of it as a blueprint to help dissect and compare forecasting methods based on key design choices. This isn't just a theoretical exercise. It's a practical step towards clarity and advancement in the field.

The truth is, Latin America doesn't need AI missionaries. It needs better rails. And time series forecasting, better rails mean a clearer path to understanding what makes models truly effective. So, let's drop the pretense and get back to the basics. In Buenos Aires, stablecoins aren't speculation. They're survival. Let's treat time series forecasting with the same no-nonsense approach.