News for hackers. Stuff that matters.

dschaurecker

Re:OpenTSLM: Language models that understand time series on Wednesday October 01, 2025 @03:06PM (#45441878)
Very cool!
This is also why we make the strong argument of why this method is useful mainly for backtesting new parametrizations or forecasts. Actually trading live below the second-mark, currently does not seem to be very beneficial on the continuous intraday market (even in a sterile backtesting scenario).
Yes thank you, that would be at least slightly more refined than me just randomly writing companies which might be relevant! My email is dschaurecker(at)gmail.com, thanks again :)
Thank you all for the interest in our paper, it is cool to see that people are interested in the topic!
Our two follow-up papers are addressing exactly this (for Europe)! We are extending our high-frequency continuous intraday approach (CID) with a forecast-based day-ahead bidding stage, and subsequent CID forecast updates.

I'd also be quite interested in strategies for grid-scale BESS trading in the US' real-time markets. Do you know more about it, or could forward me to someone who would be willing to talk about it? ;)

Most real-world optimizations for flexible storage assets currently work across multiple markets, sometimes also with more sophisticated boundary conditions. What we show is that high-frequency trading on the continuous intraday market is relevant, especially when training for more optimal parametrized strategies.

It also seems like a sensible idea to publish details and theories about an idea, not necessarily a finished trading product though ;)

Are/where you already trading at a second to sub-second level on the continuous intraday markets? How did you backtest your strategies then, if so? Or is backtesting, e.g. for parametric extensions of the optimization, not yet quite relevant?