Time series forecasting in physical commodity trading
I'm a data engineer/scientist with a background in time-series modelling, currently exploring whether a time-series algorithm I have developed over the last few years has a genuine fit in physical commodity trading. What I've built is a time-series algorithm that tracks how relationships between variables change over time, rather than assuming they're fixed (basically state space representation). In practical terms that means it can estimate things like hedge ratios, cross-commodity correlations, or basis dynamics as they evolve. The properties that make it potentially interesting for this industry: it updates the model continuously as new data comes in, it produces a confidence estimate alongside the output (so you know when a relationship is stable vs. when it's breaking down), and it handles the kind of structural shifts and regime changes that tend to break simpler econometric approaches.
The use cases I've been exploring/researching as potentially relevant:
* Hedge ratio estimation and updating. (Particularly for cross-grade or cross-commodity hedges where the relationship isn't stable)
* Hedge effectiveness testing
* P&L attribution; decomposing daily moves into price, basis, curve shape, and residual in a way that is systematic and explainable
* Short-horizon basis and spread forecasting, conditional on the current state of the market
* Scenario modelling using exogenous variables; feed in supply/demand or macro assumptions, get a forecast of the assumed price reaction (for example)
I am primarily leaning on either just offering the algorithm through an API that can be integrated into existing workflows, or creating a dashboard SaaS with built-in scenario modeling and hedging calculations, etc. (but I assume most traders are already well saturated with both tools and screen real estate for tracking information, so another dashboard would just feel cluttered?)
The question I'm trying to get answers for, is whether any of this is even relevant to this field, or whether experienced people in the industry have solved it well enough that there's no real gap? As in, does time series forecasting even factor into the decision making in any part of a trading desk at a "significant" level, and can I assume that my algorithm would potentially bring value to some shops if I keep working on it?