skaters

FlowState vs Laplace

IBM · FlowState research checkpoint (arXiv:2508.05287) zero-shot one-step change series

FlowState is a state-space time-series model, loaded via IBM's granite-tsfm. Research checkpoint (research use only); run raw. The collaborative arms were built for the three strongest models first.

Resources: Model card · Paper (arXiv 2508.05287)

Live snapshot  Derived from the week-long round-robin study; counts grow as coverage deepens. Everything scores against Laplace on the identical series and windows.

Standalone

The model's own predictive, run zero-shot, scored per series against Laplace by a one-step-ahead (k=1) Diebold–Mariano test on the log-score differential.

stratumnwin / draw / loss vs Laplace med ΔLLCRPS ratio cov₀₀
economic change-series, business-daily1328-3.001.0380.70
economic change-series, weekly2720-2.921.0220.70
economic change-series, monthly (annual cycle)2720-2.770.9920.66
M4 hourly, strongly seasonal414-2.311.0560.68
asset prices and returns, daily1392-2.771.0550.65
beats Laplace draw loses to Laplace

Median per-series Δ log-likelihood in nats (negative is worse than Laplace); CRPS ratio to Laplace (above 1 is worse); raw central-90% coverage (0.90 target).

Star plot

FlowState standalone against Laplace, on the same six regime axes as the site's standalone radar. Each radius is the log-likelihood ratio, (wins + ½·ties) / n scaled so an even split with Laplace sits on the dashed 1.0 ring; outward beats Laplace more often, inward less. The M4-hourly set splits into soft and hard waveforms by corpus order, matching that radar.

economic (daily)weekly cyclesyearly cyclessoft waveformshard waveformsprice / returnsLaplace = 1.0economic (daily): log-likelihood ratio 0.01 vs Laplace (n=1339)weekly cycles: log-likelihood ratio 0.11 vs Laplace (n=2760)yearly cycles: log-likelihood ratio 0.02 vs Laplace (n=2760)soft waveforms: log-likelihood ratio 0.00 vs Laplace (n=180)hard waveforms: log-likelihood ratio 0.00 vs Laplace (n=234)price / returns: log-likelihood ratio 0.00 vs Laplace (n=1421)

Collaborative use

The recalibration (@lap) and portfolio (&lap) arms were built for the three strongest models first, so FlowState is scored standalone here. The sidecar wraps any per-step predictive, so these arms can be added without retraining. See the sidecar pattern.

Protocol

Fixed 128-length context, rolling one-step test window, no fitting, each model in its own environment. Strata split the cached FRED universe and the M4-hourly set by frequency and regime. Full method on the sidecar pattern page and in the methodology.

Architecture and methodology

FlowState pairs a state-space model (SSM) encoder with a functional-basis decoder. The decoder is continuous in time, which makes the model equivariant to the sampling rate and able to forecast at any resolution and horizon without retraining; it emits quantiles directly. It is an IBM research checkpoint loaded through granite-tsfm.