Stratwise

ML trading strategies on your own exchange.
funds never leave. peer-reviewed at ACM ICAIF 2025.
tg:

[Stratwise @StratwiseAI]

Thank you @TradingProtocol for an interview 🤝

[Trading Strategy @TradingProtocol]

Listen to our Trading Strategy Podcast episode #16 with Stratwise (@StratwiseAI) and learn why a strategy built to survive first and earn second can turn market volatility into income without leverage.

In this episode, Nertila (@theblue_panda) from Trading Strategy discusses with Iaroslav (@Yava_web3) from Stratwise. Stratwise is an AI-driven crypto asset manager that runs non-custodial, spot-only strategies, including public vaults on Hyperliquid, using a machine-learning engine to trade volatility.

👉 View Stratwise vaults on Trading Strategy https://t.co/ap7YqyuHpX

👉 Listen to the podcast on Spotify https://t.co/isHzEDFT7R

👉 Listen to the podcast on YouTube https://t.co/fuIl7YfHpa

[Stratwise @StratwiseAI]

you missed us, admit it
new review below https://t.co/qVOmiZg9IU

[Stratwise @StratwiseAI]

true

[Stratwise @StratwiseAI]

@iinvestoradesam Hi! send us a DM.

[Stratwise @StratwiseAI]

Our preprint is now on arXiv: https://t.co/dFzZqLeqmw https://t.co/ASg9KzQsl3

[Stratwise @StratwiseAI]

@Yava_web3 🤝

[Stratwise @StratwiseAI]

relax

[Stratwise @StratwiseAI]

whatcha doin' ?

[Stratwise @StratwiseAI]

new week new article review https://t.co/q8fybmZEWo

[Stratwise @StratwiseAI]

are you watching?

[Stratwise @StratwiseAI]

stay strong

[Stratwise @StratwiseAI]

Pretraining provides useful, reusable structure, while source selection and adaptation determine how effectively that structure transfers to a target task. So, to answer your question: yes, pretrained time-series foundation models are useful forecasting tools and can outperform models trained on a single asset, but their advantage depends strongly on the relationship between the pretraining and target distributions. Our next study will investigate how proximity in probability space affects transfer performance, with potentially far-reaching implications for how pretraining datasets should be constructed.

[Stratwise @StratwiseAI]

Everyone should have an answer to this question https://t.co/Out3Zu21u1

[Stratwise @StratwiseAI]

i want more

[Stratwise @StratwiseAI]

hi

[Stratwise @StratwiseAI]

what's up guys

[Stratwise @StratwiseAI]

what's missing?

[Stratwise @StratwiseAI]

hiiiii what's the plan for the weekends?

[Stratwise @StratwiseAI]

let's meet up

[Stratwise @StratwiseAI]

So where is the martingale in our own research?

After two explainers on martingales in financial AI, a fair question: where is it in our work?

As a certificate. A supermartingale envelope bounds a crypto price over a whole holding window. That lets us size the collateral a spot–perpetual hedge needs before opening it - including the awkward moment when funding has been paid but the collateral transfer hasn't arrived yet.

As a loss. One condition, inverse wealth must be a supermartingale, plugs into Linear, MLP, GRU and Transformer policies and keeps wealth away from a capital floor

Enforced exactly under a known law, the condition becomes a guarantee; used only as a training penalty, it improves risk but does not guarantee it. We show where it helps - and, just as carefully, where it doesn't.

📄 Preprint coming soon on arXiv, with reproducible notebooks, independent audits and the negative results included. Stay tuned.