Independent researchBetween the bench and the market
When battery prices move together
Manuscript: Covariance-Aware Technology Intelligence for Battery Chemistry Portfolios
Independent research, with mentor guidance · March 2026 – present
Eight projects between the bench and the market
Bench experiments, devices, measurementsMarket prices, sourcing, portfolios
A battery chemistry is also a bet on a basket of raw materials. If the prices of a design’s materials tend to rise and fall together, its cost risk is not what you get by treating them one at a time, and a ranking that ignores this can put chemistries in the wrong order.
The question was how much that matters in practice. I linked Argonne’s BatPaC material inventories for 26 candidate chemistries to price histories from 2015 to 2024, attributed each design’s cost variance to how its material prices co-move, and checked which conclusions survived robust ranking, constrained optimization and rolling out-of-sample tests.
- InventoriesBatPaC material inventories for 26 candidate chemistries; 22 designs screened.
- Prices2015–2024 price histories for each material.
- CovarianceEach design’s cost variance attributed to how its material prices move together.
- TestsRobust and partial ranking, constrained optimization, rolling out-of-sample checks.
What the study found
- With the full covariance, 17 of the 22 screened designs changed rank and 19 of 231 pairwise comparisons reversed.
- Audits of coverage and scaling separate 187 comparable pairs from 44 that cannot be compared.
- The result measures exposure to material prices, not any company’s actual procurement loss. The rolling tests were mixed, so I don’t claim a covariance-aware ranking always lowers risk.
What I did
- Framed the question, built the data pipeline and the models, and wrote the manuscript as its sole author.
- A mentor advised me on the statistics, and I used AI tools to help organize and process online data.
Where it standsSubmitted to the International Journal of Technology Management in October 2026 and now in editorial screening.
Rank changed (17)Unchanged (5)
Reversed (19)Not reversed (212)
One dot per design (left) or per pair of designs (right). Counts are from the manuscript; the dots’ positions carry no meaning.