5 min read2 days ago
–
Press enter or click to view image in full size
Forecasting is often treated as a technological problem — throw data into a model, tweak a few knobs, and wait for predictions to appear. The reality is more nuanced. Forecasting is a structural problem. It’s about understanding how a system changes over time and what those changes reveal about the future.
And if you’ve spent any time with classical time-series models, you’ve encountered the cryptic but powerful trio that sits at the core of ARIMA: p, d, and q.
For many practitioners, PDQ is simply a set of parameters to tune. But for analysts who work deeply with data — economists, financial modelers, supply-chain analysts, machine-learning researchers — PDQ is more than configuration. It’s a…
5 min read2 days ago
–
Press enter or click to view image in full size
Forecasting is often treated as a technological problem — throw data into a model, tweak a few knobs, and wait for predictions to appear. The reality is more nuanced. Forecasting is a structural problem. It’s about understanding how a system changes over time and what those changes reveal about the future.
And if you’ve spent any time with classical time-series models, you’ve encountered the cryptic but powerful trio that sits at the core of ARIMA: p, d, and q.
For many practitioners, PDQ is simply a set of parameters to tune. But for analysts who work deeply with data — economists, financial modelers, supply-chain analysts, machine-learning researchers — PDQ is more than configuration. It’s a lens for understanding the underlying mechanics of a time series.
This article explores PDQ not as a formula, but as a framework — one that continues to matter in 2025, even in a landscape crowded with neural forecasting models.
Why PDQ Still Matters in a Neural-Forecasting World
With the rise of Transformers, N-Beats, DeepAR, and diffusion models, it’s fair to ask: Why talk about ARIMA at all?
Because ARIMA, anchored by PDQ, gives you three things modern deep learning often struggles with: