Chronos vs Laplace

Amazon · Chronos Apache-2.0 zero-shot one-step change series

Chronos tokenizes a series into a fixed vocabulary and forecasts it autoregressively, like a language model over quantized values. Reused from the earlier foundation study; run zero-shot on the one-step change series.

Resources: GitHub · Model card · Paper (arXiv 2403.07815)

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-daily1328-1.121.0200.82
economic change-series, weekly2720-1.210.9920.82
economic change-series, monthly (annual cycle)2720-0.860.9770.79
M4 hourly, strongly seasonal414-0.521.0160.84
asset prices and returns, daily1392-0.871.0230.77
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

Chronos 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 0.04 vs Laplace (n=1339)weekly cycles: log-likelihood ratio 0.19 vs Laplace (n=2760)yearly cycles: log-likelihood ratio 0.18 vs Laplace (n=2760)soft waveforms: log-likelihood ratio 0.36 vs Laplace (n=180)hard waveforms: log-likelihood ratio 0.09 vs Laplace (n=234)price / returns: log-likelihood ratio 0.01 vs Laplace (n=1421)

Collaborative use

Two collaborative arms wrap Chronos'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-daily0.820.9267%31%
economic change-series, weekly0.820.9058%15%
economic change-series, monthly (annual cycle)0.790.9145%7%
M4 hourly, strongly seasonal0.840.9239%1%
asset prices and returns, daily0.770.9156%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.

Architecture and methodology

Chronos treats forecasting as language modelling. It scales and quantises the series into a fixed token vocabulary, trains a T5-style transformer on the token sequences with a cross-entropy loss, and forecasts by sampling tokens. This study runs the Chronos-Bolt variant (chronos-bolt-small), which patches the context and emits quantiles directly in a single forward pass rather than sampling autoregressively; it is far faster, but its log-likelihood is a quantile reconstruction, so read its CRPS as the primary signal.