TiRex vs Laplace

NX-AI · TiRex NXAI Community License zero-shot one-step change series

TiRex is a 35M-parameter, xLSTM-based zero-shot forecaster that emits quantile predictions directly. Loaded from NX-AI/TiRex on CPU. Commercial use is gated above €100M revenue; included here under research use with the required attribution.

Resources: GitHub · Model card · Paper (arXiv 2505.23719) · License

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-daily1328-0.851.0060.84
economic change-series, weekly2720-1.030.9500.82
economic change-series, monthly (annual cycle)2720-0.810.9370.80
M4 hourly, strongly seasonal414-0.880.9860.76
asset prices and returns, daily1392-0.581.0230.82
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

TiRex 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 0.06 vs Laplace (n=1339)weekly cycles: log-likelihood ratio 0.20 vs Laplace (n=2760)yearly cycles: log-likelihood ratio 0.18 vs Laplace (n=2760)soft waveforms: log-likelihood ratio 0.14 vs Laplace (n=180)hard waveforms: log-likelihood ratio 0.09 vs Laplace (n=234)price / returns: log-likelihood ratio 0.02 vs Laplace (n=1421)

Collaborative use

Two collaborative arms wrap TiRex's own predictive. @lap lets Laplace forecast the model's normal scores, which fixes coverage; &lap holds Laplace and that wrap in a long-only online portfolio, so the blend is never much worse than Laplace alone. The recalibration pulls raw coverage back toward the 0.90 target, and the portfolio collapses the loss rate against Laplace.

stratumraw cov@lap cov raw loss&lap loss
economic change-series, business-daily0.840.9259%32%
economic change-series, weekly0.820.9051%15%
economic change-series, monthly (annual cycle)0.800.9038%7%
M4 hourly, strongly seasonal0.760.8946%1%
asset prices and returns, daily0.820.9233%5%

Central-90% coverage (0.90 target) and the fraction of series where the arm loses to Laplace by a Diebold–Mariano test.

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.

Existing benchmarks

TiRex reports strongly on public forecasting leaderboards, which measure a different task than this study. Those rank multi-horizon accuracy on curated dataset collections; this study scores one-step-ahead density forecasts on economic change-series against Laplace. A model can lead one and trail the other, so the results here sit alongside, not against, the leaderboards below.

Architecture and methodology

TiRex is built on the xLSTM recurrent architecture rather than attention, at 35M parameters. It ingests the context and emits nine quantiles directly in one pass, and is trained for in-context zero-shot forecasting across both short and long horizons. Its small size is the notable part: it leads public leaderboards while being an order of magnitude smaller than the transformer foundation models.